CN115833101B

CN115833101B - Power scheduling method, device, electronic equipment and storage medium

Info

Publication number: CN115833101B
Application number: CN202211558177.4A
Authority: CN
Inventors: 刘源杰; 解鑫; 李飞; 袁晓敏; 石成功; 胡比洋
Original assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Current assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Priority date: 2022-12-06
Filing date: 2022-12-06
Publication date: 2023-11-14
Anticipated expiration: 2042-12-06
Also published as: CN115833101A

Abstract

The disclosure provides a power dispatching method, a device, electronic equipment and a storage medium, and relates to the technical fields of deep learning, power dispatching, smart cities and the like. The specific implementation scheme is as follows: respectively inputting the running state information of the target power grid into a plurality of scheduling strategy models to obtain a first action sequence output by any scheduling strategy model; wherein, the plurality of scheduling strategy models adopt different optimization targets; generating a second action sequence according to the first action sequences output by the scheduling strategy models; and performing power scheduling on the plurality of generator sets according to the power parameter adjustment actions of the plurality of generator sets in the second action sequence. Therefore, by setting different optimization targets for different scheduling strategy models and combining the outputs of the scheduling strategy models adopting the different optimization targets, a second action sequence combining the advantages of multiple targets can be obtained, the robustness of the second action sequence is improved, and the probability of occurrence of operation breakdown of a target power grid is reduced.

Description

Power scheduling method, device, electronic equipment and storage medium

Technical Field

The present disclosure relates to the field of AI (Artificial Intelligence ), and in particular, to the technical fields of deep learning, power scheduling, smart cities, and the like, and more particularly, to a power scheduling method, apparatus, electronic device, and storage medium.

Background

The power dispatching is an effective management means for ensuring safe and stable operation of the power grid, external reliable power supply and orderly execution of various power production works. In the power dispatching, various power generation types are involved, such as thermal power generation, new energy (such as wind energy, light energy and water energy) power generation, and how to dispatch the power to each power generation set is very important under the condition that the power generation sets of various power generation types participate in power generation.

Disclosure of Invention

The disclosure provides a power scheduling method, a power scheduling device, an electronic device and a storage medium.

According to an aspect of the present disclosure, there is provided a power scheduling method including:

acquiring operation state information of a target power grid, wherein the operation state information comprises topology state information and load state information of the target power grid;

respectively inputting the running state information into a plurality of scheduling strategy models to obtain a first action sequence output by any scheduling strategy model; the first action sequence comprises power parameter adjustment actions of a plurality of generator sets in the target power grid, and the plurality of scheduling strategy models are trained by adopting loss functions corresponding to different optimization targets;

Generating a second action sequence according to the first action sequences output by the scheduling strategy models;

and performing power scheduling on the plurality of generator sets according to the power parameter adjustment actions of the plurality of generator sets in the second action sequence.

According to another aspect of the present disclosure, there is provided a power scheduling apparatus including:

the system comprises an acquisition module, a control module and a control module, wherein the acquisition module is used for acquiring the running state information of a target power grid, and the running state information comprises the topology state information and the load state information of the target power grid;

the input module is used for respectively inputting the running state information into a plurality of scheduling strategy models to obtain a first action sequence output by any scheduling strategy model; the first action sequence comprises power parameter adjustment actions of a plurality of generator sets in the target power grid, and the plurality of scheduling strategy models are trained by adopting loss functions corresponding to different optimization targets;

the generating module is used for generating a second action sequence according to the first action sequences output by the scheduling strategy models;

and the scheduling module is used for performing power scheduling on the plurality of generator sets according to the power parameter adjustment actions of the plurality of generator sets in the second action sequence.

According to still another aspect of the present disclosure, there is provided an electronic apparatus including:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the power scheduling method set forth in the above aspect of the disclosure.

According to yet another aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium of computer instructions for causing the computer to perform the power scheduling method set forth in the above aspect of the present disclosure.

According to a further aspect of the present disclosure, there is provided a computer program product comprising a computer program which, when executed by a processor, implements the power scheduling method set forth in the above aspect of the present disclosure.

It should be understood that the description in this section is not intended to identify key or critical features of the embodiments of the disclosure, nor is it intended to be used to limit the scope of the disclosure. Other features of the present disclosure will become apparent from the following specification.

Drawings

The drawings are for a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:

fig. 1 is a flowchart of a power scheduling method according to an embodiment of the disclosure;

fig. 2 is a flow chart of a power scheduling method according to a second embodiment of the disclosure;

fig. 3 is a schematic diagram of a multi-objective learning principle provided by an embodiment of the present disclosure;

fig. 4 is a flow chart of a power scheduling method according to a third embodiment of the disclosure;

fig. 5 is a flow chart of a power scheduling method according to a fourth embodiment of the disclosure;

fig. 6 is a flow chart of a power scheduling method according to a fifth embodiment of the disclosure;

fig. 7 is a flow chart of a power scheduling method according to a sixth embodiment of the disclosure;

fig. 8 is a flow chart of a power scheduling method according to a seventh embodiment of the disclosure;

fig. 9 is a schematic diagram of a power scheduling principle provided by an embodiment of the disclosure;

FIG. 10 is a schematic diagram of a cross-section out-of-limit correction flow provided by an embodiment of the disclosure;

fig. 11 is a schematic structural diagram of a power dispatching device according to an embodiment eight of the present disclosure;

FIG. 12 illustrates a schematic block diagram of an example electronic device that may be used to implement embodiments of the present disclosure.

Detailed Description

Exemplary embodiments of the present disclosure are described below in conjunction with the accompanying drawings, which include various details of the embodiments of the present disclosure to facilitate understanding, and should be considered as merely exemplary. Accordingly, one of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

The main goal of power scheduling is to: safe, high quality and economical.

Wherein, safety means: the safety, stability and normal operation of the power grid are ensured, and the safe and reliable power supply to the power consumer is ensured.

The quality refers to: the electric energy quality is ensured, and the frequency and the voltage are kept to be qualified.

Economy means: and the power generation tasks are reasonably arranged according to the principles of fairness and fairness, so that the optimal utilization of power generation resources is realized.

The types of power generation involved in power scheduling mainly include: thermal power generation and new energy power generation (wind power generation, photovoltaic power generation and hydroelectric power generation), and under the condition that various power generation types of power generation sets participate in power generation, the power generation tasks and plans of the power generation sets need to be reasonably specified. The method comprises the steps of reasonably planning and arranging generator sets of various power generation types according to a predicted load curve and an economic dispatching principle, distributing power generation tasks of each power plant, and providing the processing condition of each generator set in 96 time nodes (namely, 1 hour comprises 4 time nodes, namely, 15 minutes comprises one time node) per day.

The power dispatching of each generator set is reasonably and efficiently carried out, and the problems of section out-of-limit, new energy power consumption, operation cost, supply and demand balance and the like are required to be considered at the same time, wherein the difficulty is mainly expressed in that: the power grid body quantity, the new energy output, the cost control and the safe operation.

Wherein, the electric wire netting volume: because the coverage range of the large power grid is wider, the structure of the large power grid is complex, the influence factors are more, and the cooperative treatment among all the generator sets is emphasized.

New energy output: because the new energy is accessed into the power grid, the uncertainty of the output of the new energy causes the change condition of the power grid tide to be changeable. The high-proportion new energy is connected into and applied to high-proportion power electronic equipment, so that the inertia of the system is continuously reduced, the fault resistance is reduced, the fault form is more complex, the risk of cascade reaction is increased, and higher requirements are put on the accuracy and timeliness of a regulation strategy.

Cost control: the unit price difference between the thermal generator set and the new energy generator set is larger, and the running cost is directly related to the output control of the generator sets with different power generation types.

Safe operation: grid dispatching is a long-term decision process and needs to be operated safely and efficiently for a long time as much as possible.

And (3) carrying out power dispatching on each generator set, for example, giving out a corresponding generator set output adjustment strategy according to the running condition of the power grid at each moment, so that the power generation cost of the power system can be minimized and the new energy consumption can be promoted on the premise of ensuring the running safety and stability of the power system.

In the related art, aiming at the operation output plan formulation in power grid dispatching, the main solution mainly comprises the following three types:

first, SCUC (Security Constrained Unit Commitment, safety restraint assembly). The model is used for making a startup and shutdown plan of the generator set for a plurality of time periods by taking social benefit maximization and the like as optimization targets under the condition of meeting the safety constraint of the power system. The objective function of the long-period genset combination SCUC calculated by the electric energy market before date is as follows:

wherein N represents the total number of the generator sets; t represents the total number of time periods to be considered, e.g., having a total of 5 time periods, then t=5; p (P) _i,t Representing the output of the generator set i in a period t; c (C) _i,t (P _i,t ) Indicating the operating cost of genset i during period t,representing the start-up cost of genset i during period t, wherein C _i,t (P _i,t ) Is a multi-section linear function related to each output section and corresponding energy price declared by the generator set A number; />Indicating the no-load cost of the generator set i in the period t; m represents a network power flow constraint relaxation penalty factor; />Forward flow relaxation variable, < > representing line l>A reverse power flow relaxation variable representing line l, NL representing the total number of lines; />Forward flow relaxation variable representing section s, < +.>The reverse power flow relaxation variable of the section s is represented, and NS represents the total number of sections.

From the above formula (1), the objective nature of SCUC is: on the premise of safe and stable operation of the power grid, the cost of the running, starting and idle generator set is minimized. The safe and stable SCUC constraint conditions which meet the clear SCUC constraint conditions of the electric power market are mainly included in the following items shown in the table 1.

TABLE 1

Second, SCED (Security Constrained Economic Dispatch, safety-constrained economic dispatch). The model is to pursue that the running cost of all started generator sets (commanded units) is minimum in a certain research period, and the output is the output force of each generator set. Since the SCED economic model does not contain a 0/1 binary integer variable reflecting the on/off state of the generator set, the optimization objective function of the SCED is as follows:

wherein, the explanation of each parameter in the formula (2) is the same as the formula (1).

Wherein, the constraint condition of SCED is the same as SCUC.

Third, a power scheduling method based on reinforcement learning. Reinforcement learning is used to solve the power scheduling problem, and in fact, the power scheduling problem is modeled as a model that reinforcement learning can solve. The reinforcement learning mainly relates to core variables such as states, environments, actions and the like, and the corresponding variables in the power grid dispatching task are as follows:

1. state: the current running state of the power grid, wherein the core elements comprise: topological states of the power grid, load conditions of the power grid and the like;

2. environmental Environment: the simulator of the power grid can be provided by a national power grid, and has the functions of receiving the output data of the generator set, calculating the current power flow condition of the power grid, calculating a corresponding rewarding report value according to constraint, returning the rewarding report value to the Agent, and guiding the learning of the Agent;

3. action: the output result of the model is the output condition of each generator set of the power grid;

4. awarding Reward: the variable is the core of AI power dispatching, and Reward's design needs to convert the objective function and constraint condition of traditional power dispatching into rewarding function, guiding Agent to learn.

The power dispatching method based on reinforcement learning mainly takes the power grid state and the like as input, and utilizes the learning capacity of a neural network to automatically generate the output condition of each generator set.

Of the above solutions, the first two solutions obtain an output arrangement by means of constraint solving, and the third solution learns the output arrangement by means of artificial intelligence, which has at least the following problems:

(1) The former two schemes only consider target optimization in the current state, and often have the problems of complex calculation and unfriendly long-term safe operation of the power grid.

(2) The third scheme is easy to ignore various hard constraints of the power grid, and is easy to cause the problem of power grid operation breakdown.

(3) Large power grids often cover a wide range in actual geographic positions, the operation of the large power grids is very complex in influencing factors, and the traditional scheme is difficult to process in a targeted manner under different conditions.

In view of at least one of the problems presented above, the present disclosure proposes a power scheduling method, apparatus, electronic device, and storage medium.

The power scheduling method, apparatus, electronic device, and storage medium of the embodiments of the present disclosure are described below with reference to the accompanying drawings. Before describing embodiments of the present disclosure in detail, for ease of understanding, general technical words are first introduced:

Output: in a power plant or a power plant, the electrical power generated by a generator set is called the output, which is the electrical energy per unit time.

And (3) tide: voltage, power at steady state operation of the grid. In power engineering, "power flow" also refers to the distribution of voltage (including amplitude and phase angle), active power, reactive power, etc. throughout the grid.

Cross section out-of-limit: the current and power of several lines from a certain tie-source center to the load center exceed the stability limit.

Section: under a certain ground state tide, a set of power transmission lines with the same active power flow direction and similar electric distances is called a power transmission section.

Section power flow (or called power flow section): in a larger power grid, in a certain mode, a bundle of channels formed by a plurality of lines or transformers is limited in power flow transmission from a plurality of nodes with similar electric distances to another region or nodes with similar electric distances, the total amount of power flow transmission of any element in the channels is not changed significantly, and all elements are related to each other in power flow transmission.

Fig. 1 is a flowchart of a power scheduling method according to an embodiment of the disclosure.

The power scheduling method is configured in a power scheduling device for example, and the power scheduling device can be applied to any electronic equipment so that the electronic equipment can execute a power scheduling function.

The electronic device may be any device with computing capability, for example, a PC (Personal Computer ), a mobile terminal, a server, and the like, and the mobile terminal may be, for example, a vehicle-mounted device, a mobile phone, a tablet computer, a personal digital assistant, a wearable device, and other hardware devices with various operating systems, touch screens, and/or display screens.

As shown in fig. 1, the power scheduling method may include the steps of:

step 101, acquiring operation state information of a target power grid, wherein the operation state information comprises topology state information and load state information of the target power grid.

In the embodiment of the present disclosure, the target power grid may be a power grid of a certain area, or the target power grid may also be a national power grid, or the like, which is not limited in this disclosure.

In the embodiment of the disclosure, the operation state information may at least include topology state information of the target power grid and load state information of the target power grid, and in addition, the operation state information may further include power parameter information (such as current output, voltage, etc.) of each generator set in the target power grid, load information of each line, and so on.

In the embodiment of the disclosure, the operation state information of the target power grid may be obtained, where the operation state information may be operation state information corresponding to the current time or the current time node of the target power grid.

For example, the operational status information may be provided to the relevant staff of the target grid, or the operational status information may be queried from a designated interface, etc.

102, respectively inputting the running state information into a plurality of scheduling strategy models to obtain a first action sequence output by any scheduling strategy model; the first action sequence comprises power parameter adjustment actions of a plurality of generator sets in a target power grid, and a plurality of scheduling strategy models are trained by adopting loss functions corresponding to different optimization targets.

In an embodiment of the disclosure, a power generating set of at least one power generation type (such as thermal power generation, new energy (such as wind energy, light energy, water energy) power generation, etc.) may be included in the target power grid, for example, at least one thermal power generating set, at least one wind power generating set, at least one photovoltaic power generating set, at least one hydro power generating set, etc. may be included in the target power grid.

In the embodiment of the disclosure, the optimization targets may include operation cost optimization (for example, the minimum operation cost of all the turned-on generator sets is taken as an optimization target), security optimization (for example, the maximum social benefit is taken as an optimization target under the condition of meeting the security constraint of the power system), new energy consumption optimization (for example, the complete consumption of new energy is taken as an optimization target), section control optimization (for example, all sections in the target power grid are taken as an optimization target without exceeding limits), and the like.

In an embodiment of the disclosure, the power parameter adjustment action is used for indicating a power parameter adjustment amount of the generator set, wherein the power parameter may include an output, a voltage, and the like. Taking the power parameter as the output as an example, the power parameter adjustment action can be used for indicating the output increment which needs to be adjusted by the generator set, such as +2, -2, +3, -3, and the like. For example, the current output of the generator set a is 50, the power parameter adjustment action of the generator set a in the first action sequence indicates that the output increment of the generator set a to be adjusted is +2, and the output to which the generator set a needs to be adjusted is 52.

In the embodiment of the disclosure, different scheduling policy models may be trained by using loss functions corresponding to different optimization targets.

For example, assume a total of three optimization objectives, namely: the cross section control optimization, the new energy optimization and the running cost optimization can be provided with 3 scheduling strategy models, namely a scheduling strategy model 1, a scheduling strategy model 2 and a scheduling strategy model 3, wherein the optimization targets adopted by the scheduling strategy model 1 are cross section control optimization, the optimization targets adopted by the scheduling strategy model 2 are new energy optimization, and the optimization targets adopted by the scheduling strategy model 3 are running cost optimization.

For another example, assume that there are three optimization objectives in total, namely: the cross section control optimization, the new energy optimization and the running cost optimization can be provided with 4 scheduling strategy models, namely a scheduling strategy model 1, a scheduling strategy model 2, a scheduling strategy model 3 and a scheduling strategy model 4, wherein the optimization targets adopted by the scheduling strategy model 1 are cross section control optimization, the optimization targets adopted by the scheduling strategy model 2 are new energy optimization, the optimization targets adopted by the scheduling strategy model 3 are running cost optimization, and the scheduling strategy model 4 adopts all the optimization targets (namely, the optimization targets adopted by the scheduling strategy model 4 can simultaneously comprise cross section control optimization, new energy optimization and running cost optimization).

In the embodiment of the disclosure, the running state information may be respectively input into a plurality of scheduling policy models to obtain a first action sequence output by any one scheduling policy model, where each first action sequence may include power parameter adjustment actions of a plurality of generator sets in a target power grid.

Step 103, generating a second action sequence according to the first action sequences output by the scheduling policy models.

In the embodiment of the disclosure, the second action sequence may be generated according to the first action sequences output by the scheduling policy models. For example, the first action sequences output by the scheduling policy models may be fused to obtain the second action sequence.

And step 104, performing power scheduling on the plurality of generator sets according to the power parameter adjustment actions of the plurality of generator sets in the second action sequence.

In the embodiment of the disclosure, the power scheduling may be performed on the plurality of generator sets according to the power parameter adjustment actions of the plurality of generator sets in the second action sequence.

As an example, take the second action sequence as { a } ₁₁ ,a ₁₂ ,…,a _1M Example, where M is the occurrence in the target gridThe number or the number of the motor groups, and M is a positive integer, can be according to a ₁₁ Power scheduling of the generator set 1 according to a ₁₂ Power dispatching of genset 2, …, according to a _1M And carrying out power dispatching on the generator set M.

According to the power dispatching method, the running state information of the target power grid is respectively input into a plurality of dispatching strategy models, so that a first action sequence output by any dispatching strategy model is obtained; the power parameter adjustment actions of the multiple generator sets in the target power grid are included in any first action sequence, and the multiple scheduling strategy models are obtained through training by adopting loss functions corresponding to different optimization targets; generating a second action sequence according to the first action sequences output by the scheduling strategy models; and performing power scheduling on the plurality of generator sets according to the power parameter adjustment actions of the plurality of generator sets in the second action sequence. Therefore, by setting different optimization targets for different scheduling strategy models and combining the outputs of the scheduling strategy models adopting different optimization targets, a second action sequence combining the advantages of multiple targets can be obtained, the robustness of the second action sequence is improved, and accordingly actions are adjusted according to power parameters with higher robustness in the second action sequence, power scheduling is performed on each generator set, and the probability of occurrence of operation breakdown of a target power grid can be reduced.

It should be noted that, in the technical solution of the present disclosure, the related processes of collecting, storing, using, processing, transmitting, providing, disclosing, etc. of the personal information of the user are all performed on the premise of proving the consent of the user, and all conform to the rules of the related laws and regulations, and do not violate the popular regulations of the public order.

In order to clearly illustrate how the first action sequence is generated according to the first action sequences output by the scheduling policy models in the above embodiments of the present disclosure, the present disclosure also proposes a power scheduling method.

Fig. 2 is a flowchart of a power scheduling method according to a second embodiment of the disclosure.

As shown in fig. 2, the power scheduling method may include the steps of:

step 201, obtaining operation state information of a target power grid, wherein the operation state information comprises topology state information and load state information of the target power grid.

Step 202, the running state information is respectively input into a plurality of scheduling strategy models to obtain a first action sequence output by any scheduling strategy model, wherein the plurality of scheduling strategy models comprise a first scheduling strategy model and at least one second scheduling strategy model, the first scheduling strategy model is trained by adopting a loss function corresponding to a plurality of optimization targets, and the plurality of optimization targets comprise optimization targets adopted by the at least one second scheduling strategy model.

The first action sequence comprises power parameter adjustment actions of a plurality of generator sets in a target power grid, and a plurality of scheduling strategy models are trained by adopting loss functions corresponding to different optimization targets.

It should be noted that, the explanation of the steps 201 to 202 may be referred to the related descriptions in any embodiment of the disclosure, and will not be repeated herein.

In the embodiment of the disclosure, the plurality of scheduling policy models may include a first scheduling policy model and at least one second scheduling policy model, where each second scheduling policy model may be obtained by training a loss function corresponding to an optimization objective, and different second scheduling policy models use different optimization objectives.

The first scheduling policy model may be obtained by training a loss function corresponding to a plurality of optimization targets, for example, the target loss function may be determined according to the loss function corresponding to the plurality of optimization targets, and the first scheduling policy model may be trained according to the target loss function. And, the plurality of optimization objectives employed by the first scheduling policy model includes the optimization objective employed by each of the second scheduling policy models.

For example, assume a total of three optimization objectives, namely: the section control optimization, the new energy optimization and the running cost optimization can be provided with 3 second scheduling policy models, namely a second scheduling policy model 1, a second scheduling policy model 2 and a second scheduling policy model 3, wherein the optimization targets adopted by the second scheduling policy model 1 are section control optimization, the optimization targets adopted by the second scheduling policy model 2 are new energy optimization, the optimization targets adopted by the second scheduling policy model 3 are running cost optimization, and the first scheduling policy model can adopt all the optimization targets, namely the optimization targets adopted by the first scheduling policy model can simultaneously comprise section control optimization, new energy optimization and running cost optimization.

Step 203, determining a similarity between the first action sequence output by the at least one second scheduling policy model and the first action sequence output by the first scheduling policy model.

In the embodiment of the present disclosure, the similarity between the first action sequence output by any one of the second scheduling policy models and the first action sequence output by the first scheduling policy model may be determined based on a similarity calculation algorithm, so as to obtain the similarity of the second scheduling policy models.

Step 204, determining the weight of the at least one second scheduling policy model according to the similarity of the at least one second scheduling policy model.

In the embodiment of the present disclosure, for any one second scheduling policy model, the weight of the second scheduling policy model may be determined according to the similarity of the second scheduling policy model, where the weight and the similarity are in a positive correlation (i.e., in a positive relationship), that is, the greater the similarity, the higher the weight, and vice versa, the smaller the similarity, and the lower the weight.

In one possible implementation manner of the embodiment of the present disclosure, the similarity of each second scheduling policy model may be normalized, so as to obtain the weight of each second scheduling policy model.

As an example, the target coefficient may be determined according to a sum of the similarities of the second scheduling policy models, where the target coefficient and the sum of the similarities of the second scheduling policy models are in positive correlation. For example, the sum of the similarities of the second scheduling policy models may be used as the target coefficient. Therefore, the ratio of the similarity of each second scheduling policy model to the target coefficient can be used as the weight of each second scheduling policy model.

For example, the number of the second scheduling policy models is marked as q, q is a positive integer, and the similarity of the kth second scheduling policy model is Sim _k The weight of the kth second scheduling policy model is:

therefore, the weight of each second scheduling policy model is determined by normalizing the similarity of each second scheduling policy model, so that the rationality and the accuracy of weight determination can be improved.

Step 205, weighting the first action sequence output by the at least one second scheduling policy model according to the weight of the at least one second scheduling policy model, so as to obtain a second action sequence.

In the embodiment of the present disclosure, the first action sequences output by the second scheduling policy models may be weighted or fused according to the weights of the second scheduling policy models, so as to obtain second action sequences.

As an example, the first action sequence and the second action sequence may include M element values, where M is the number of generator sets in the target power grid, and the element values at the same position in the first action sequence output by each second scheduling policy model may be weighted according to the weight of each second scheduling policy model, so as to obtain the second action sequence.

For example, assuming that m=3 and q=3, the weight of the second scheduling policy model 1 is 0.2, the weight of the second scheduling policy model 2 is 0.3, the weight of the second scheduling policy model 3 is 0.5, the first action sequence output by the second scheduling policy model 1 is { +2, +1, -3}, the first action sequence output by the second scheduling policy model 1 is { +1, +2, -2}, the first action sequence output by the second scheduling policy model 3 is { +2, -1, +1}, the first element value in the second action sequence may be: 0.2 (+2) +0.3 (+1) +0.5 (+2) = +1.7, the second element value in the second sequence of actions may be: 0.2 (+1) +0.3 (+2) +0.5 (-1) = +0.3, the third element value in the second sequence of actions may be: 0.2 (-3) +0.3 (-2) +0.5 (+ 1) = -0.7, i.e. the second sequence of actions may be { +1.7, +0.3, -0.7}.

And step 206, performing power scheduling on the plurality of generator sets according to the power parameter adjustment actions of the plurality of generator sets in the second action sequence.

The explanation of step 206 may be referred to the relevant descriptions in any embodiment of the disclosure, and will not be repeated here.

It should be noted that, in actual operation of the power grid, under the common influence of multiple targets (safety, new energy consumption, cost control, etc.), there may be output schemes focusing on different optimization targets, and the reinforcement learning method in the related art only can perform optimization learning for a single target, which easily leads to operation breakdown of the power grid.

In the application, a multi-objective learning method is introduced, and the advantages of multiple objectives can be combined by setting different optimization objectives for different scheduling strategy models and fusing the outputs of the scheduling strategy models in a linear combination mode, so as to obtain more robust power parameter adjustment actions and reduce the probability of occurrence of operation breakdown of a target power grid.

As an example, there are three targets in total, respectively: as shown in fig. 3, the cross-section control, the new energy and the running cost are exemplified, and the similarity between the action sequence 1 output by the second scheduling policy model 1, the action sequence 2 output by the second scheduling policy model 2, the action sequence 3 output by the second scheduling policy model 3 and the action sequence 4 output by the first scheduling policy model may be calculated, and based on the similarity, the action sequence 1 output by the second scheduling policy model 1, the action sequence 2 output by the second scheduling policy model 2, and the action sequence 3 output by the second scheduling policy model 3 may be fused to obtain a final action sequence (referred to as a second action sequence in the present disclosure).

According to the power scheduling method, as the optimization targets adopted by the first scheduling policy model comprise all the optimization targets adopted by the second scheduling policy model, the weight of any second scheduling policy model is determined according to the similarity between the first action sequence output by any second scheduling policy model and the first action sequence output by the first scheduling policy model, so that the rationality and reliability of weight calculation can be improved, and the action sequences output by the scheduling policy models adopting different optimization targets are linearly fused based on the weight of the reliability, so that the reliability of the fused second action sequence can be improved.

In order to clearly illustrate how to perform power scheduling on a plurality of power generating sets according to power parameter adjustment actions of the plurality of power generating sets in the second action sequence in any embodiment of the disclosure, the disclosure further provides a power scheduling method.

Fig. 4 is a flow chart of a power scheduling method according to a third embodiment of the disclosure.

As shown in fig. 4, the power scheduling method may include the steps of:

step 401, obtaining operation state information of a target power grid, wherein the operation state information comprises topology state information and load state information of the target power grid.

Step 402, the running state information is respectively input into a plurality of scheduling policy models to obtain a first action sequence output by any scheduling policy model.

Step 403, generating a second action sequence according to the first action sequences output by the scheduling policy models.

The explanation of steps 401 to 403 may be referred to the relevant description in any embodiment of the present disclosure, and will not be repeated here.

Step 404, a sensitivity matrix is obtained.

In this embodiment of the present disclosure, the sensitivity matrix may be a matrix of m×n, where M is the number of generator sets in the target power grid, N is the number of sections in the target power grid, and an ith row and a jth column element in the sensitivity matrix are used to indicate that the ith generator set adjusts a unit power parameter to cause a power flow variation of the jth section, i is a positive integer not greater than M, j is a positive integer not greater than N, and M and N are both positive integers.

In the embodiment of the present disclosure, the sensitivity matrix may be calculated in advance, or may be calculated in real time, which is not limited in this disclosure.

As a possible implementation, the manner of pre-calculating the sensitivity matrix may include the following steps:

1. and acquiring the actual power flow of each section in the target power grid at a first time node, wherein the first time node is positioned before the current time node.

2. And adjusting the power parameter of the ith generating set at the first time node, such as reducing or increasing the output of the ith generating set by a set value.

3. And determining the predicted power flow corresponding to each section after the power parameter adjustment of the ith generating set according to the power parameter of the ith generating set before adjustment, the power parameter of the ith generating set after adjustment and the actual power flow of each section at the first time node.

For example, the simulator may be used to predict the predicted power flow corresponding to each section after the power parameter adjustment of the ith generator set according to the power parameter of the ith generator set before the adjustment, the power parameter of the ith generator set after the adjustment, the actual power flow of each section at the first time node, the historical operation information of the ith generator set, and the historical operation information of each section.

4. And determining each element of the ith row in the sensitivity matrix according to the predicted power flow and the actual power flow corresponding to each section, and according to the power parameter of the ith generator set before adjustment and the power parameter of the ith generator set after adjustment.

For example, taking the power parameter as the output, assume that the difference between the output of the ith generator set after adjustment and the output of the ith generator set before adjustment is Δ ₁ The difference between the predicted power flow and the actual power flow corresponding to the jth section is delta ₂ Then the first sensitivity matrixThe value of the j-th element of line i is: delta ₂ /Δ ₁ 。

As another possible implementation manner, the manner of calculating the sensitivity matrix in real time may include the following steps:

1. and acquiring the actual power flow of each section in the target power grid at the current time node.

2. And adjusting the current parameter of the ith generating set at the current time node, such as reducing or increasing the output of the ith generating set by a set value.

3. And determining the predicted power flow corresponding to each section after the power parameter adjustment of the ith generating set according to the power parameter of the ith generating set before adjustment, the power parameter of the ith generating set after adjustment and the actual power flow of each section at the current time node.

For example, the simulator may be used to predict the predicted power flow corresponding to each section after the power parameter adjustment of the ith generator set according to the power parameter of the ith generator set before the adjustment, the power parameter of the ith generator set after the adjustment, the actual power flow of each section at the current time node, the historical operation information of the ith generator set, and the historical operation information of each section.

Therefore, the predicted power flow of each section is predicted according to the variation of the power parameter actually regulated by the generator set and the actual power flow of each section, so that the sensitivity matrix is calculated according to the variation, the actual power flow and the predicted power flow, and the accuracy and the reliability of a calculation result can be improved.

Step 405, determining a first predicted power flow of the at least one section according to the actual power flow of the at least one section, the sensitivity matrix and the second sequence of actions.

The actual power flow may be the power flow of the section at the current time node (or the current moment).

In the embodiment of the present disclosure, the predicted power flow corresponding to the next time node (or the next time) of the current time node for each section may be determined according to the actual power flow, the sensitivity matrix and the second action sequence of each section, which is denoted as the first predicted power flow in the present disclosure.

As an example, the second motion sequence includes M element values, the second motion sequence may be taken as a vector of 1*M, the sensitivity matrix is a matrix of m×n, and the second motion sequence may be multiplied by the sensitivity matrix to obtain a vector of 1*N, where the vector of 1*N includes N power flow increments of the cross section. For any section, the actual power flow of the section at the current time node and the corresponding power flow increment can be added, and the first predicted power flow of the section at the next time node can be obtained.

Step 406, determining a first section from the at least one section based on the first predicted power flow of the at least one section.

In the embodiment of the disclosure, the first section may be determined from the sections according to the first predicted power flow of the sections.

As a possible implementation manner, the section with the largest first predicted power flow may be taken as the first section.

As another possible implementation manner, a section with the maximum power flow out of limit may be determined according to the first predicted power flow of each section, and the section is used as the first section.

As an example, a candidate cross-section may be determined from the cross-sections according to a first predicted power flow of each cross-section, where the first predicted power flow of the candidate cross-section is not in a set first power flow interval (or referred to as a power flow value range) corresponding to the candidate cross-section, for example, the first predicted power flow of the candidate cross-section is smaller than a minimum power flow threshold (i.e., a lower limit) of the first power flow interval corresponding to the candidate cross-section, or the first predicted power flow of the candidate cross-section is larger than a maximum power flow threshold (i.e., an upper limit) of the first power flow interval corresponding to the candidate cross-section, that is, the candidate cross-section is a power flow out-of-limit cross-section. For any one candidate section, a deviation between the first predicted power flow of the candidate section and the corresponding first power flow interval may be determined, where the deviation may be a difference between the first predicted power flow and an upper limit of the first power flow value, or the deviation may be a difference between the first predicted power flow and a lower limit of the first power flow value, so that in the present disclosure, the first section may be determined from the candidate sections according to the deviation of each candidate section.

For example, the candidate cross-section with the largest deviation may be the first cross-section, i.e., the candidate cross-section with the largest power flow threshold.

Alternatively, a candidate cross section having a deviation larger than the set deviation threshold may be used as the first cross section.

Therefore, the first section can be determined in different modes, and the flexibility and applicability of the method can be improved. In addition, according to the size of the power flow out of limit of each section, a first section needing power flow adjustment is determined from each section, so that the second action sequence is adjusted according to the first predicted power flow of the first section, the rationality of the adjusted action sequence can be improved, and the probability of the power flow out of limit of the section is reduced.

Step 407, adjusting the second motion sequence according to the first predicted power flow of the first section to obtain a third motion sequence.

In the embodiment of the disclosure, the second action sequence may be adjusted according to the first predicted power flow of the first section, so as to obtain the third action sequence. For example, the second action sequence may be adjusted based on an equivalent reverse matching method, and the equivalent reverse matching method will be described in detail later, which will not be described herein.

Step 408, performing power scheduling on the multiple generator sets according to the power parameter adjustment actions of the multiple generator sets in the third action sequence.

In the embodiment of the disclosure, the power scheduling may be performed on the plurality of generator sets according to the power parameter adjustment actions of the plurality of generator sets in the third action sequence.

As an example, the third action sequence is { a } ₂₁ ,a ₂₂ ,…,a _2M By way of example, can be based on a ₂₁ To the generator set 1Power scheduling according to a ₂₂ Power dispatching of genset 2, …, according to a _2M And carrying out power dispatching on the generator set M.

Taking the power parameter as the output, the third action sequence is { +1, +2, -2, …, -1} for example, if the output of the generator set 1 at the current time node is 48, then the generator set 1 is subjected to power dispatching, that is, the generator set 1 is controlled to adjust the output of the next time node to 49, if the output of the generator set 2 at the current time node is 56, then the generator set 2 is subjected to power dispatching, that is, the generator set 2 is controlled to adjust the output of the next time node to 58, if the output of the generator set 3 at the current time node is 52, then the generator set 3 is subjected to power dispatching, that is, the generator set 3 is controlled to adjust the output of the next time node to 50, …, and if the output of the generator set M at the current time node is 62, then the generator set M is controlled to adjust the output of the next time node to 61.

According to the power scheduling method, the predicted power flow of each section in the target power grid can be predicted according to the second action sequence, so that the first section with the power flow out of limit is determined according to the predicted power flow of each section, the second action sequence is adjusted based on the predicted power flow of the first section, the adjusted action sequence is used for carrying out power scheduling on each generator set in the target power grid, the number of the sections with the power flow out of limit can be reduced, and power grid scheduling is safely carried out.

In order to clearly illustrate how the second action sequence is adjusted according to the first predicted power flow of the first section in any embodiment of the disclosure to obtain the third action sequence, the disclosure further provides a power scheduling method.

Fig. 5 is a flowchart of a power scheduling method according to a fourth embodiment of the present disclosure.

As shown in fig. 5, the power scheduling method may include the steps of:

step 501, obtaining operation state information of a target power grid, wherein the operation state information comprises topology state information and load state information of the target power grid.

Step 502, the running state information is respectively input into a plurality of scheduling policy models to obtain a first action sequence output by any scheduling policy model.

Step 503, generating a second action sequence according to the first action sequences output by the scheduling policy models.

Step 504, a sensitivity matrix is obtained.

The sensitivity matrix is a matrix of M x N, M is the number of generator sets in a target power grid, N is the number of sections in the target power grid, and an ith row and a jth column of elements in the sensitivity matrix are used for indicating the power flow variation of the jth section caused by the adjustment of unit power parameters by the ith generator set, i is a positive integer not greater than M, and j is a positive integer not greater than N; m and N are positive integers.

Step 505, determining a first predicted power flow of the at least one section according to the actual power flow of the at least one section, the sensitivity matrix and the second action sequence.

Step 506, determining a first section from the at least one section based on the first predicted power flow of the at least one section.

The explanation of steps 501 to 506 may be referred to the relevant description in any embodiment of the present disclosure, and will not be repeated here.

Step 507, executing at least one cycle according to the first predicted power flow of the first section to update the second action sequence.

In an embodiment of the present disclosure, at least one cyclic process may be performed according to the first predicted power flow of the first section to perform the updating of the second action sequence.

As one possible implementation, the first round loop process may include the steps of:

1. the column in which the first section is located in the sensitivity matrix (e.g., the j-th column), denoted as the first target column in this disclosure, is determined.

2. And determining a first generator set corresponding to the largest element and a second generator set corresponding to the smallest element in the first target column of the sensitivity matrix.

For example, a first generator set corresponding to an element with a positive value and a maximum value and/or a second generator set corresponding to an element with a negative value and a minimum value may be determined from a first target column in the sensitivity matrix.

3. And adjusting the power parameter adjustment action of the first generator set and/or the power parameter adjustment action of the second generator set in the second action sequence to obtain a second action sequence updated in the first round of circulation process.

For example, the power parameter adjustment action of the first generator set and/or the power parameter adjustment action of the second generator set in the second action sequence may be adjusted based on an equal-amount reverse matching method, so as to obtain a second action sequence updated in the first-round circulation process.

As an example, the power parameter adjustment action of the first generator set in the second action sequence may be added by an increment value Δ, and the power parameter adjustment action of the second generator set in the second action sequence may be subtracted by an increment value Δ, to obtain a second action sequence updated during the first cycle.

Therefore, aiming at the first section with the out-of-limit trend, the first generator set with the largest positive influence and the second generator set with the largest negative influence are determined from the generator sets, and the power parameter adjustment action of the first generator set and/or the power parameter adjustment action of the second generator set in the second action sequence are adjusted, so that the stability of the power flow adjustment of the section can be realized, and the updated second action sequence is more reasonable and reliable.

As one possible implementation, the non-first round robin procedure may include the steps of:

1. And determining a third predicted power flow of at least one section according to the second action sequence, the sensitivity matrix and the actual power flow of at least one section, which are updated in the previous cycle.

As an example, the second motion sequence updated in the previous cycle includes M element values, the second motion sequence updated in the previous cycle may be taken as a vector 1*M, the sensitivity matrix is a matrix of m×n, and the second motion sequence updated in the previous cycle may be multiplied by the sensitivity matrix to obtain a vector 1*N, where the vector 1*N includes N sections of power flow increment. For any section, the actual power flow of the section at the current time node and the corresponding power flow increment can be added, and then the third predicted power flow of the section at the next time node can be obtained.

2. Judging whether a second section exists in each section, wherein the third predicted power flow of the second section is not in a set second power flow interval (or called a power flow value range) corresponding to the second section, namely the second section is a power flow out-of-limit section. If no second section exists in the sections, the cycle may be ended.

3. If there are second sections in each section, a third section may be determined from each second section based on a third predicted power flow for each second section.

As a possible implementation manner, the second section with the largest third predicted power flow may be used as the third section.

As another possible implementation manner, the second section with the largest power flow out of limit may be determined according to the third predicted power flow of each second section, and the second section is used as the third section.

As an example, for any one of the second sections, a deviation between a third predicted power flow of that second section and a corresponding second power flow interval may be determined, so that in the present disclosure, the third section may be determined from each second section according to the deviation of each second section.

For example, the second cross section with the largest deviation may be the third cross section, i.e., the second cross section with the largest power flow out of limit.

Alternatively, the second cross section, in which the deviation is larger than the set deviation threshold value, may be used as the third cross section.

Thus, the determination of the third section in different ways can be achieved, and the flexibility and applicability of the method can be improved. In addition, according to the size of the power flow out of limit of each second section, a third section needing power flow adjustment is determined from each second section, so that the second action sequence updated in the previous cycle is adjusted according to the third predicted power flow of the third section, the rationality of the adjusted action sequence can be improved, the probability of the power flow out of limit is reduced, and the safety of the adjusted action sequence is improved.

4. And determining a corresponding second target column of the third section in the sensitivity matrix.

5. And determining a third generator set corresponding to the largest element and a fourth generator set corresponding to the smallest element in the second target column of the sensitivity matrix.

For example, a third generator set corresponding to an element with a positive value and a maximum value and/or a fourth generator set corresponding to an element with a negative value and a minimum value may be determined from the second target column in the sensitivity matrix.

6. And adjusting the power parameter adjustment action of the third generator set and/or the power parameter adjustment action of the fourth generator set in the second action sequence updated in the previous round of circulation process to obtain a second action sequence updated in the current round of circulation process.

For example, the power parameter adjustment action of the third generator set and/or the power parameter adjustment action of the fourth generator set in the second action sequence updated in the previous cycle may be adjusted based on an equal-amount reverse matching method, so as to obtain the second action sequence updated in the present cycle.

As an example, the power parameter adjustment action of the third generator set in the second action sequence updated during the previous cycle may be added with an increment value Δ, and the power parameter adjustment action of the fourth generator set in the second action sequence updated during the previous cycle may be subtracted with an increment value Δ, so as to obtain the second action sequence updated during the present cycle.

Thus, when there is no cross section with a power flow out of limit in each cross section, the cycle is ended, and a safe operation sequence can be obtained. In addition, in any one cycle process, aiming at the section with the power flow out of limit, a third generator set with the largest positive influence and a fourth generator set with the largest negative influence are determined from the generator sets, and the power parameter adjustment action of the third generator set and/or the power parameter adjustment action of the fourth generator set in the second action sequence updated in the previous cycle process are adjusted, so that the stability of the power flow adjustment of the section can be realized, and the updated second action sequence is more reasonable and reliable.

Step 508, the second action sequence updated in the last cycle is used as the third action sequence.

In the embodiment of the disclosure, the second action sequence updated in the last cycle process may be used as the third action sequence.

Step 509, performing power scheduling on the multiple generator sets according to the power parameter adjustment actions of the multiple generator sets in the third action sequence.

The explanation of step 509 may be referred to the relevant descriptions in any embodiment of the disclosure, and will not be repeated here.

The power scheduling method of the embodiment of the disclosure can update the second action sequence at least once, so as to improve the rationality and reliability of the updated second action sequence.

Fig. 6 is a flowchart of a power scheduling method according to a fifth embodiment of the present disclosure.

As shown in fig. 6, the power scheduling method may include the steps of:

step 601, obtaining operation state information of a target power grid, wherein the operation state information comprises topology state information and load state information of the target power grid.

Step 602, the running state information is respectively input into a plurality of scheduling policy models to obtain a first action sequence output by any scheduling policy model.

Step 603, generating a second action sequence according to the first action sequences output by the scheduling policy models.

In step 604, a sensitivity matrix is obtained.

The sensitivity matrix is a matrix of M x N, M is the number of generator sets in a target power grid, N is the number of sections in the target power grid, and the ith row and the jth column elements in the sensitivity matrix are used for indicating the power flow variable quantity of the jth section caused by the adjustment of unit power parameters by the ith generator set, i is a positive integer not greater than M, j is a positive integer not greater than N, and M and N are both positive integers.

Step 605, determining a first predicted power flow of the at least one section according to the actual power flow of the at least one section, the sensitivity matrix and the second action sequence.

Step 606, determining a first section from the at least one section based on the first predicted power flow for the at least one section.

The explanation of steps 601 to 606 may be referred to the relevant description in any embodiment of the present disclosure, and will not be repeated here.

Step 607, executing the iteration process for the set number of times for the second action sequence according to the first predicted power flow of the first section.

In the process of cyclically updating the power parameter adjustment operation of each generator set in the second operation sequence, a section where there is always a trend out of limit may occur, and the cycle process cannot be ended at this time.

In the embodiment of the disclosure, the iterative process of the set number of times may be executed on the second action sequence according to the first predicted power flow of the first section.

As one possible implementation method, the first round iteration process may include the following steps:

1. a corresponding target column (e.g., the j-th column) of the first cross-section in the sensitivity matrix is determined, and is denoted as the first target column in this disclosure.

As one possible implementation, the non-first round iterative process may include the steps of:

1. and determining a fourth predicted power flow of at least one section according to the second action sequence, the sensitivity matrix and the actual power flow of at least one section, which are updated in the iteration process of the previous round of the iteration process.

As an example, the second action sequence updated in the previous iteration process includes M element values, the second action sequence updated in the previous iteration process may be taken as a vector 1*M, the sensitivity matrix is a matrix of m×n, the second action sequence updated in the previous iteration process may be multiplied by the sensitivity matrix to obtain a vector 1*N, and the vector 1*N includes trend increments of N sections. For any section, the actual power flow of the section at the current time node and the corresponding power flow increment can be added to obtain the fourth predicted power flow of the section at the next time node.

2. And determining a fourth section from the sections according to the fourth predicted power flow of the sections.

As a possible implementation manner, the section with the largest fourth predicted power flow may be referred to as a fourth section.

As another possible implementation manner, a section with the maximum power flow out of limit may be determined according to the fourth predicted power flow of each section, and the section is used as the fourth section.

As an example, a target section may be determined from the sections according to a fourth predicted power flow of each section, where the fourth predicted power flow of the target section is not in a set power flow interval (or referred to as a power flow value range) corresponding to the target section, that is, the target section is a power flow out-of-limit section. For any one target section, the deviation between the fourth predicted power flow of the target section and the corresponding set power flow interval can be determined, so that in the present disclosure, the fourth section can be determined from the target sections according to the deviation of the target sections.

For example, the target cross section with the largest deviation may be the fourth cross section, i.e., the target cross section with the largest power flow out of limit.

Alternatively, the target cross section having a deviation larger than the set deviation threshold may be used as the fourth cross section.

Thus, the determination of the fourth section in different ways can be achieved, and the flexibility and applicability of the method can be improved.

3. And determining a corresponding third target column of the fourth section in the sensitivity matrix.

4. And determining a fifth generator set corresponding to the largest element and a sixth generator set corresponding to the smallest element in the third target column of the sensitivity matrix.

For example, from the third target column in the sensitivity matrix, a fifth generator set corresponding to the element with the positive value and the maximum value may be determined, and/or a sixth generator set corresponding to the element with the negative value and the minimum value may be determined.

6. And adjusting the power parameter adjustment action of the fifth generator set and/or the power parameter adjustment action of the sixth generator set in the second action sequence updated in the previous iteration process to obtain a second action sequence updated in the present iteration process.

For example, the power parameter adjustment action of the fifth generator set and/or the power parameter adjustment action of the sixth generator set in the second action sequence updated in the previous iteration process may be adjusted based on an equivalent reverse matching method, so as to obtain the second action sequence updated in the present iteration process.

As an example, the power parameter adjustment action of the fifth generator set in the second action sequence updated in the previous iteration process may be added with an increment value Δ, and the power parameter adjustment action of the sixth generator set in the second action sequence updated in the previous iteration process may be subtracted with an increment value Δ, so as to obtain the second action sequence updated in the present iteration process.

Therefore, in any iteration process, aiming at the section with the power flow out of limit, a fifth generating set with the largest positive influence and a sixth generating set with the largest negative influence are determined from the generating sets, and the power parameter adjustment action of the fifth generating set and/or the power parameter adjustment action of the sixth generating set in the second action sequence updated in the previous round of circulation process are adjusted, so that the stability of the power flow adjustment of the section can be realized, and the updated second action sequence is more reasonable and reliable.

Step 608, the second action sequence obtained by updating the last iteration process is taken as a third action sequence.

In the embodiment of the disclosure, the second action sequence obtained by updating the last iteration process can be used as a third action sequence.

Step 609, performing power scheduling on the multiple generator sets according to the power parameter adjustment actions of the multiple generator sets in the third action sequence.

The explanation of step 609 may be referred to the relevant description in any embodiment of the present disclosure, and will not be repeated here.

The power scheduling method of the embodiment of the disclosure can update the second action sequence at least once, so as to improve the rationality, reliability and safety of the updated second action sequence.

Fig. 7 is a flowchart of a power scheduling method according to a sixth embodiment of the disclosure.

As shown in fig. 7, the power scheduling method may include the steps of:

step 701, obtaining operation state information of a target power grid, wherein the operation state information comprises topology state information and load state information of the target power grid.

Step 702, the running state information is respectively input into a plurality of scheduling policy models, so as to obtain a first action sequence output by any scheduling policy model.

Step 703, generating a second action sequence according to the first action sequences output by the scheduling policy models.

In step 704, a sensitivity matrix is obtained.

Step 705, determining a first predicted power flow of the at least one section based on the actual power flow of the at least one section, the sensitivity matrix and the second sequence of actions.

Step 706, determining a first section from the at least one section based on a first predicted power flow for the at least one section.

The explanation of steps 701 to 706 may be referred to the relevant descriptions in any embodiment of the disclosure, and are not repeated here.

Step 707, according to the first predicted power flow of the first section, the power parameter adjustment actions of the generator set matched with the set adjustment strategy in the second action sequence are adjusted, so as to obtain a third action sequence.

In the embodiment of the present disclosure, the adjustment policy may be preset, for example, the adjustment policy may include, but is not limited to: robust policies, aggressive policies, compromise policies, etc.

As an example, taking the adjustment policy as an example including a robust policy, an aggressive policy, and a compromise policy, the explanation of each adjustment policy may be as shown in table 2:

TABLE 2

The value of the generator set that can output is the difference between the current output and the rated output of the generator set, for example, the current output of the generator set is 98, and the rated output is 150, and the value of the output is 52.

In the embodiment of the present disclosure, when the adjustment strategy is a robust strategy, the power parameter adjustment actions of the first generator set, the second generator set, the third generator set, the fourth generator set, the fifth generator set, and the sixth generator set in the above embodiment in the second action sequence may be allowed to be updated, so as to obtain the third action sequence.

When the regulation strategy is a aggressive strategy, only the electric power parameter regulation actions of the first generator set, the second generator set, the third generator set, the fourth generator set, the fifth generator set and the thermal generator set in the sixth generator set in the second action sequence are allowed to be updated so as to obtain a third action sequence.

When the adjustment strategy is a medium-size adjustment strategy, the adjustment actions of the electric power parameters of the first generator set, the second generator set, the third generator set, the fourth generator set, the fifth generator set and the thermal generator set and the hydroelectric generator set in the second action sequence can be allowed to be updated, so that the third action sequence is obtained.

Step 708, performing power scheduling on the plurality of generator sets according to the power parameter adjustment actions of the plurality of generator sets in the third action sequence.

The explanation of step 708 may be referred to the relevant descriptions in any embodiment of the disclosure, and will not be repeated here.

According to the power scheduling method, the power parameter adjustment actions of the generator set in the second action sequence can be adjusted based on different adjustment strategies, and the flexibility and applicability of the method can be improved.

In order to clearly illustrate how to perform power scheduling on a plurality of power generating sets according to power parameter adjustment actions of the plurality of power generating sets in the third action sequence in any embodiment of the disclosure, the disclosure further provides a power scheduling method.

Fig. 8 is a flowchart of a power scheduling method according to a seventh embodiment of the disclosure.

As shown in fig. 8, the power scheduling method may include the steps of:

step 801, obtaining operation state information of a target power grid, wherein the operation state information comprises topology state information and load state information of the target power grid.

Step 802, the running state information is respectively input into a plurality of scheduling policy models to obtain a first action sequence output by any scheduling policy model.

Step 803, generating a second action sequence according to the first action sequences output by the scheduling policy models.

In step 804, a sensitivity matrix is obtained.

Step 805, determining a first predicted power flow for the at least one section based on the actual power flow for the at least one section, the sensitivity matrix, and the second sequence of actions.

Step 806, determining a first section from the at least one section based on the first predicted power flow of the at least one section.

Step 807, according to the first predicted trend of the first section, adjusting the power parameter adjustment action of the generator set matched with the set adjustment strategy in the second action sequence to obtain a third action sequence; the number of the adjustment strategies is multiple, and the number of the third action sequences is multiple.

The explanation of steps 801 to 807 can be referred to the relevant description in any embodiment of the present disclosure, and will not be repeated here.

At step 808, scores are obtained for a plurality of third action sequences.

In embodiments of the present disclosure, each third action sequence may be scored to obtain a score for the third action sequence.

As one possible implementation, the third action sequences may be respectively scored based on a right-tail function of the standard normal distribution (also referred to as a complementary cumulative distribution function of the standard normal distribution, a Q function, a state-cost function) to obtain scores of the third action sequences.

As an example, any third action sequence may be scored based on the Q function using a Q network (Q-Learning) algorithm, resulting in a score for the third action sequence.

Where the input of the Q function may be state and action and the output scalar.

For example, the operating state information of the target grid and the third action sequence may be input into a Q network, and the rewards, values or scores obtained by taking the third action sequence under the operating state information may be determined by the Q network based on the Q function.

As another possible implementation manner, the power flow of each section may be predicted based on the sensitivity matrix and the third action sequence, and the score of the third action sequence may be determined according to the power flow of each section.

As an example, for any one of the third action sequences, the fifth predicted power flow of each section at the next time node may be determined according to the actual power flow of each section at the current time node, the third action sequence and the sensitivity matrix, and the implementation principle is similar to the determination manner of the fourth predicted power flow, the third predicted power flow or the first predicted power flow, which is not described herein. Further, a fifth cross section may be determined from the cross sections based on a fifth predicted power flow of each cross section, wherein the fifth predicted power flow of the fifth cross section is not within a third power flow interval (or referred to as a power flow value range) set corresponding to the fifth cross section, that is, the fifth cross section is a power flow out-of-limit cross section. Thus, in the present disclosure, the score of the third action sequence may be determined based on the number of fifth profiles.

The score of the third action sequence and the number of the fifth sections are in a negative correlation (namely in a negative correlation), namely, the smaller the number of the fifth sections is, the higher the score of the third action sequence is, whereas the higher the number of the fifth sections is, the lower the score of the third action sequence is.

As still another possible implementation manner, the power parameter adjustment operation of the generator set with the power generation type (such as thermal power generation and hydroelectric power generation) in the third action sequence may be adjusted or disturbed, and the trend of each section may be predicted according to the adjusted or disturbed third action sequence and the sensitivity matrix, so that the score of the third action sequence may be determined according to the trend of each section.

As an example, for any one of the third operation sequences, the power parameter adjustment operation of the generator set of the set generation type in the third operation sequence may be adjusted according to the rated power parameter corresponding to the generator set of the set generation type, so as to obtain an adjusted third operation sequence.

For example, the power parameter adjustment operation of the generator set of the generation type in the third operation sequence may be adjusted based on the neighborhood disturbance performed by the simulated annealing method.

For example, taking the set generation type as hydroelectric generation and the electric power parameter as the output as an example, assuming that the current output of the hydroelectric generating set a is 140, the rated output is 150, and assuming that the output adjustment amount corresponding to the hydroelectric generating set a in the third operation sequence is +8, the output of the hydroelectric generating set a needs to be adjusted to 148 at the next time node, and since the difference between 150 and 148 is 2, the output of the hydroelectric generating set a at the next time node can be adjusted to 152, that is, the output adjustment amount of the hydroelectric generating set a in the third operation sequence is updated to 12.

As another example, for any one of the third action sequences, the adjustment (for example, the mirror symmetry adjustment) of the electric power parameter of the generator set with the set generating type in the third action sequence may be directly performed, so as to obtain an adjusted third action sequence.

For example, assuming that the thermal generator set B in the third operation sequence has an output adjustment amount of +2, the output adjustment amount of the thermal generator set B in the third operation sequence may be updated to-2.

Therefore, in the present disclosure, the sixth predicted power flow of each section at the next time node may be determined according to the adjusted third action sequence, the sensitivity matrix, and the actual power flow of each section at the current time node, and the implementation principle is similar to the determination manner of the fifth predicted power flow, the fourth predicted power flow, the third predicted power flow, or the first predicted power flow, which is not described herein. And determining a sixth section from the sections according to the sixth predicted power flow of each section, wherein the sixth predicted power flow of the sixth section is not in a fourth power flow interval (or called a power flow value range) set corresponding to the sixth section, that is, the sixth section is a power flow out-of-limit section, and further the score of the third action sequence can be determined according to the number of the sixth sections.

The score of the third action sequence and the number of the sixth sections are in a negative correlation (namely in a negative correlation), namely, the smaller the number of the sixth sections is, the higher the score of the third action sequence is, whereas the higher the number of the sixth sections is, the lower the score of the third action sequence is.

Thus, scoring of a plurality of third action sequences in different ways can be achieved, and flexibility and applicability of the method can be improved.

Step 809, determining a target action sequence from the plurality of third action sequences based on the scores of the plurality of third action sequences.

In the embodiments of the present disclosure, the target action sequence may be determined from the plurality of third action sequences according to scores of the plurality of third action sequences. For example, the third action sequence with the highest score may be used as the target action sequence.

And step 810, performing power scheduling on the multiple generator sets according to the power parameter adjustment actions of the multiple generator sets in the target action sequence.

In the embodiment of the disclosure, the power scheduling of the multiple generator sets can be performed according to the power parameter adjustment actions of the multiple generator sets in the target action sequence.

As an example, take the target action sequence as { a } ₃₁ ,a ₃₂ ,…,a _3M By way of example, can be based on a ₃₁ Power scheduling of the generator set 1 according to a ₃₂ Power dispatching of genset 2, …, according to a _3M And carrying out power dispatching on the generator set M.

According to the power scheduling method, a mode of scoring the action sequences obtained based on different adjustment strategies can be achieved, the score of each action sequence is obtained, and therefore the action sequence with the higher score is selected, power scheduling is conducted on the generator set in the target power grid, the number of cross sections with the out-of-limit trend can be reduced, and efficient and safe power grid scheduling is achieved.

In any embodiment of the disclosure, taking the electric power parameter as an output to perform an example, aiming at the cross-section power flow out-of-limit problem, the new energy consumption problem, the running cost problem, the supply and demand balance problem and the like in the power grid dispatching, in the disclosure, modeling can be performed by adopting the electric power expert experience knowledge and the reinforcement learning algorithm of the self-adaptive environment to help the safe, efficient and economic running of the power grid. The initial motion generated by the reinforcement learning model in the AI field can be utilized, and after a post-processing module constructed by expert experience, the better motion output is finally obtained by a neighborhood searching method such as simulated annealing. The implementation principle can be as shown in fig. 9.

The cross section out-of-limit correction flow may be as shown in fig. 10:

1. calculating a sensitivity matrix of the section tide: giving the section tide at the current moment or the current time node, carrying out output adjustment on the generator set, and calculating the sensitivity matrix of the section tide based on the section tide variation.

a. Acquiring section power flow of the current time node, namely actual power flow of each section;

b. carrying out output adjustment on the ith generating set;

c. predicting the predicted power flow of each section after the output of the ith generating set is adjusted based on the simulator;

d. Calculating the sensitivity of the section power flow change brought by the unit output of the ith generating set according to the difference value between the predicted power flow and the actual power flow corresponding to each section, namely calculating each element of the ith row in a sensitivity matrix;

e. by the method, the sensitivities of all the generator sets are calculated, namely, element values corresponding to each row in the sensitivity matrix are calculated.

2. After the actual power flow of the sections is obtained, the predicted or predicted power flow of the sections can be compared with the predicted maximum power flow and the predicted minimum power flow of each section in advance, and the section exceeding the maximum power flow and the section lower than the minimum power flow are used as out-of-limit sections so as to correct the out-of-limit sections.

a. Predicting the predicted power flow of each section corresponding to the next time node according to the actual power flow of each section, the action sequence output by the model and the sensitivity matrix;

b. determining whether cross sections are out of limit according to the predicted power flow of each cross section, and if so, selecting the cross section with the largest out of limit;

c. the method comprises the steps of determining a column of a section with the largest out-of-limit in a sensitivity matrix, determining a generator set A corresponding to an element with the largest value in the column, namely selecting the generator set A with the largest positive sensitivity, and determining a generator set B corresponding to an element with the smallest value in the column, namely selecting the generator set B with the smallest negative sensitivity.

d. Adopting an equivalent reverse matching method to carry out section adjustment, namely carrying out balance adjustment on the output adjustment actions of the generator set A and the generator set B in the action sequence output by the model;

e. selecting the generator set for multiple times until the adjustment is finished;

f. and c, predicting the predicted power flow of each section corresponding to the next time node based on the updated action sequence and the actual power flow of each section, judging whether an out-of-limit section exists or not, and if yes, re-executing the steps c to subsequent steps.

Wherein, different adjustment strategies can be set for different power grid environments.

For example, by analyzing the types of various generator sets and the conditions of key generator sets, it can be obtained that the wind generator set and the photovoltaic generator set can be adjusted to a predicted output value (namely, rated output), and the hydro generator set can exceed the predicted output value (namely, rated output). The cost control can be achieved, the cost of the thermal generator set is relatively high, and the cost of the new energy generator set is relatively low. Thus, for changes in operating conditions, an adaptive adjustment may be made by constructing three output action treatments (i.e., adjustment strategies) as shown in table 2.

In the case where the adjustment policy is plural, the plurality of operation sequences obtained by the adjustment may be plural, and a more preferable operation sequence may be searched for from the plurality of operation sequences.

Multiple possible sequences of actions can be generated in the same grid state thanks to the taking of multiple action adjustment strategies. The multiple feasible action sequences can be scored to obtain scores of the multiple feasible action sequences, and a reasonable or optimal action sequence is determined from the multiple feasible action sequences according to the scores of the multiple feasible action sequences.

As an example, the superiority or inferiority of an action may be assessed by a Q-network in a reinforcement learning model, i.e., in the present disclosure, the Q-network may be employed to score each sequence of actions based on a Q-function.

As another example, the total prize for the sequence of actions may be calculated based on the sequence of actions and the sensitivity matrix, the profile flow may be predicted, the number of profiles out of limit based on the flow, and the remaining prizes combined, such that the score for the sequence of actions may be determined based on the total prize for the sequence of actions.

As yet another example, neighborhood perturbation may be performed by simulated annealing methods to find a more optimal solution when choosing an optimal solution from a plurality of possible action sequences. For example, mirror symmetry of the new energy source can be predicted for one or more hydroelectric generating sets in the action sequence at random, that is, mirror symmetry adjustment can be performed on the output adjustment action of the hydroelectric generating sets in the action sequence according to the rated output corresponding to the hydroelectric generating sets. For another example, mirror symmetry of the current output value of the observed state can be randomly performed on one or more thermal generator sets in the action sequence, i.e. the output number adjustment actions of the thermal generator sets in the action sequence are subjected to mirror symmetry adjustment.

Through neighborhood disturbance and combination of a scoring module, a more reasonable action sequence can be obtained.

In conclusion, the multi-objective learning and multi-adjustment strategy post-processing method is adopted, so that the safety of an output action sequence is improved and the operation cost is reduced when the model is used for coping with different power grid operation states; in addition, the number of cross sections that are out of limit can be reduced. In addition, the method is designed for the problems of cross section out-of-limit, supply and demand balance and the like, and in order to cope with the change of the power grid state in different environments, various post-treatment schemes with self-adapting adjustment strategies are provided, and a power generation scheme with better effect is obtained by combining a simulated annealing algorithm. The method can be applied to the problem of the power output arrangement formulation of multiple units in the large power grid dispatching of the generator units with multiple power generation types, the historical data is learned through deep learning, expert experience is assisted, and a simulated annealing search algorithm is combined, so that the power grid dispatching is performed efficiently and safely.

Corresponding to the power scheduling method provided by the embodiments of fig. 1 to 8, the present disclosure also provides a power scheduling apparatus, and since the power scheduling apparatus provided by the embodiments of the present disclosure corresponds to the power scheduling method provided by the embodiments of fig. 1 to 8, the implementation of the power scheduling method is also applicable to the power scheduling apparatus provided by the embodiments of the present disclosure, and will not be described in detail in the embodiments of the present disclosure.

Fig. 11 is a schematic structural diagram of a power dispatching device according to an embodiment of the present disclosure.

As shown in fig. 11, the power scheduling apparatus 1100 may include: an acquisition module 1101, an input module 1102, a generation module 1103 and a scheduling module 1104.

The obtaining module 1101 is configured to obtain operation state information of the target power grid, where the operation state information includes topology state information and load state information of the target power grid.

The input module 1102 is configured to input the running state information into a plurality of scheduling policy models respectively, so as to obtain a first action sequence output by any scheduling policy model; the first action sequence comprises power parameter adjustment actions of a plurality of generator sets in a target power grid, and a plurality of scheduling strategy models are trained by adopting loss functions corresponding to different optimization targets.

The generating module 1103 is configured to generate a second action sequence according to the first action sequences output by the plurality of scheduling policy models.

The scheduling module 1104 is configured to perform power scheduling on the plurality of generator sets according to the power parameter adjustment actions of the plurality of generator sets in the second action sequence.

In one possible implementation manner of the embodiment of the present disclosure, the plurality of scheduling policy models includes a first scheduling policy model and at least one second scheduling policy model, the first scheduling policy model is trained by using a loss function corresponding to a plurality of optimization targets, and the plurality of optimization targets includes an optimization target adopted by the at least one second scheduling policy model; a generating module 1103, configured to: determining the similarity between the first action sequence output by the at least one second scheduling policy model and the first action sequence output by the first scheduling policy model; determining the weight of at least one second scheduling policy model according to the similarity of the at least one second scheduling policy model; and weighting the first action sequence output by the at least one second scheduling policy model according to the weight of the at least one second scheduling policy model to obtain a second action sequence.

In one possible implementation of the embodiment of the disclosure, the generating module 1103 is configured to: determining a target coefficient according to the sum of the similarity of at least one second scheduling policy model; and determining the weight of the second scheduling policy model according to the ratio of the corresponding similarity to the target coefficient aiming at any second scheduling policy model.

In one possible implementation of the embodiments of the present disclosure, the scheduling module 1104 may include:

the system comprises an acquisition unit, a sensitivity matrix, a power supply unit and a power supply unit, wherein the sensitivity matrix is a matrix of M, N is the number of generator sets in a target power grid, N is the number of sections in the target power grid, the ith row and the jth column of elements in the sensitivity matrix are used for indicating the power flow variation of the jth section caused by the adjustment of unit power parameters by the ith generator set, i is a positive integer not greater than M, j is a positive integer not greater than N, and M and N are both positive integers.

And the first determining unit is used for determining a first predicted power flow of the at least one section according to the actual power flow of the at least one section, the sensitivity matrix and the second action sequence.

And the second determining unit is used for determining the first section from the at least one section according to the first predicted power flow of the at least one section.

And the adjusting unit is used for adjusting the second action sequence according to the first predicted power flow of the first section so as to obtain a third action sequence.

And the scheduling unit is used for performing power scheduling on the plurality of generator sets according to the power parameter adjustment actions of the plurality of generator sets in the third action sequence.

In a possible implementation manner of the embodiment of the present disclosure, the second determining unit is configured to: determining a candidate section from the at least one section according to the first predicted power flow of the at least one section, wherein the first predicted power flow of the candidate section is not in a set first power flow interval corresponding to the candidate section; for any candidate section, determining the deviation between the first predicted power flow of the candidate section and the corresponding first power flow interval; a first cross-section is determined from the at least one candidate cross-section based on the deviation of the at least one candidate cross-section.

In one possible implementation manner of the embodiment of the present disclosure, the obtaining unit is configured to: acquiring the actual power flow of at least one section in a target power grid; aiming at an ith generating set in a target power grid, carrying out power parameter adjustment on the ith generating set; determining a second predicted power flow corresponding to at least one section after the power parameter adjustment of the ith generating set according to the power parameter of the ith generating set before adjustment, the power parameter of the ith generating set after adjustment and the actual power flow of the at least one section; and determining each element of the ith row in the sensitivity matrix according to the second predicted power flow and the actual power flow corresponding to the at least one section, and according to the power parameter of the ith generator set before adjustment and the power parameter of the ith generator set after adjustment.

In one possible implementation manner of the embodiment of the present disclosure, the adjusting unit is configured to: executing at least one cyclic process according to the first predicted trend of the first section so as to update the second action sequence; and taking the second action sequence updated in the last cycle process as a third action sequence.

In one possible implementation of an embodiment of the present disclosure, the first round of looping process includes: determining a first target column corresponding to the first section in the sensitivity matrix; determining a first generator set corresponding to a maximum element and a second generator set corresponding to a minimum element in a first target column of the sensitivity matrix; and adjusting the power parameter adjustment action of the first generator set and/or the power parameter adjustment action of the second generator set in the second action sequence to obtain a second action sequence updated in the first round of circulation process.

In one possible implementation of an embodiment of the present disclosure, the non-first round robin procedure includes: determining a third predicted power flow of at least one section according to the second action sequence, the sensitivity matrix and the actual power flow of at least one section, which are updated in the previous cycle; ending the cycle if there is no second section in at least one section; the third predicted power flow of the second section is not in a set second power flow interval corresponding to the second section; determining a third section from the at least one second section according to a third predicted power flow of the at least one second section when the second section exists in the at least one section, and determining a second target column corresponding to the third section in the sensitivity matrix; determining a third generator set corresponding to the largest element and a fourth generator set corresponding to the smallest element in a second target column of the sensitivity matrix; and adjusting the power parameter adjustment action of the third generator set and/or the power parameter adjustment action of the fourth generator set in the second action sequence updated in the previous round of circulation process to obtain a second action sequence updated in the current round of circulation process.

In one possible implementation manner of the embodiment of the present disclosure, the adjusting unit is configured to: executing an iteration process for a set number of times on the second action sequence according to the first predicted power flow of the first section; and updating the second action sequence obtained in the last iteration process to be used as a third action sequence.

In one possible implementation of an embodiment of the present disclosure, any one iteration process includes: determining a fourth predicted power flow of at least one section according to the second action sequence, the sensitivity matrix and the actual power flow of the at least one section, which are updated in the previous iteration process of the present iteration process; determining a fourth section from the at least one section based on a fourth predicted power flow of the at least one section; determining a third target column corresponding to the fourth section in the sensitivity matrix; determining a fifth generator set corresponding to the largest element and a sixth generator set corresponding to the smallest element in a third target column of the sensitivity matrix; and adjusting the power parameter adjustment action of the fifth generator set and/or the power parameter adjustment action of the sixth generator set in the second action sequence updated in the previous iteration process to obtain a second action sequence updated in the current iteration process.

In one possible implementation manner of the embodiment of the present disclosure, the adjusting unit is configured to: and adjusting the power parameter adjustment action of the generator set matched with the set adjustment strategy in the second action sequence according to the first predicted trend of the first section so as to obtain a third action sequence.

In a possible implementation manner of the embodiment of the present disclosure, the adjustment policy is multiple, the third action sequence is multiple, and the scheduling unit is configured to: obtaining scores for a plurality of third action sequences; determining a target action sequence from the plurality of third action sequences according to the scores of the plurality of third action sequences; and performing power scheduling on the multiple generator sets according to the power parameter adjustment actions of the multiple generator sets in the target action sequence.

In a possible implementation manner of the embodiment of the present disclosure, a scheduling unit is configured to: scoring the plurality of third action sequences by adopting a right tail function of standard normal distribution so as to obtain scores of the plurality of third action sequences; or, for any third action sequence, determining a fifth predicted power flow of at least one section according to the actual power flow of the at least one section, the third action sequence and the sensitivity matrix; determining a fifth section from the at least one section according to a fifth predicted power flow of the at least one section, wherein the fifth predicted power flow of the fifth section is not in a set third power flow interval corresponding to the fifth section; determining a score for the third action sequence based on the number of fifth sections; or, for any third action sequence, according to the rated power parameter corresponding to the set generating set, adjusting the power parameter adjustment action of the generating set with the generating type in the third action sequence to obtain an adjusted third action sequence; determining a sixth predicted power flow of the at least one section according to the adjusted third action sequence, the sensitivity matrix and the actual power flow of the at least one section; determining a sixth section from the at least one section according to a sixth predicted power flow of the at least one section, wherein the sixth predicted power flow of the sixth section is not in a set fourth power flow interval corresponding to the sixth section; and determining the score of the third action sequence according to the number of the sixth sections.

According to the power dispatching device, the operation state information of the target power grid is respectively input into a plurality of dispatching strategy models, so that a first action sequence output by any dispatching strategy model is obtained; the power parameter adjustment actions of the multiple generator sets in the target power grid are included in any first action sequence, and the multiple scheduling strategy models are obtained through training by adopting loss functions corresponding to different optimization targets; generating a second action sequence according to the first action sequences output by the scheduling strategy models; and performing power scheduling on the plurality of generator sets according to the power parameter adjustment actions of the plurality of generator sets in the second action sequence. Therefore, by setting different optimization targets for different scheduling strategy models and combining the outputs of the scheduling strategy models adopting different optimization targets, a second action sequence combining the advantages of multiple targets can be obtained, the robustness of the second action sequence is improved, and accordingly actions are adjusted according to power parameters with higher robustness in the second action sequence, power scheduling is performed on each generator set, and the probability of occurrence of operation breakdown of a target power grid can be reduced.

To achieve the above embodiments, the present disclosure also provides an electronic device that may include at least one processor; and a memory communicatively coupled to the at least one processor; the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the power scheduling method according to any one of the above embodiments of the present disclosure.

To implement the above embodiments, the present disclosure also provides a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the power scheduling method set forth in any one of the above embodiments of the present disclosure.

To achieve the above embodiments, the present disclosure also provides a computer program product comprising a computer program which, when executed by a processor, implements the power scheduling method set forth in any of the above embodiments of the present disclosure.

According to embodiments of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

FIG. 12 illustrates a schematic block diagram of an example electronic device that may be used to implement embodiments of the present disclosure. The electronic device may include the server and the client in the above embodiments. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the disclosure described and/or claimed herein.

As shown in fig. 12, the electronic apparatus 1200 includes a computing unit 1201 that can perform various appropriate actions and processes according to a computer program stored in a ROM (Read-Only Memory) 1202 or a computer program loaded from a storage unit 1208 into a RAM (Random Access Memory ) 1203. In the RAM 1203, various programs and data required for the operation of the electronic device 1200 may also be stored. The computing unit 1201, the ROM 1202, and the RAM 1203 are connected to each other via a bus 1204. An I/O (Input/Output) interface 1205 is also connected to bus 1204.

Various components in the electronic device 1200 are connected to the I/O interface 1205, including: an input unit 1206 such as a keyboard, mouse, etc.; an output unit 1207 such as various types of displays, speakers, and the like; a storage unit 1208 such as a magnetic disk, an optical disk, or the like; and a communication unit 1209, such as a network card, modem, wireless communication transceiver, etc. The communication unit 1209 allows the electronic device 1200 to exchange information/data with other devices through a computer network, such as the internet, and/or various telecommunications networks.

The computing unit 1201 may be a variety of general and/or special purpose processing components having processing and computing capabilities. Some examples of computing unit 1201 include, but are not limited to, a CPU (Central Processing Unit ), GPU (Graphic Processing Units, graphics processing unit), various dedicated AI (Artificial Intelligence ) computing chips, various computing units running machine learning model algorithms, DSPs (Digital Signal Processor, digital signal processors), and any suitable processors, controllers, microcontrollers, and the like. The computing unit 1201 performs the various methods and processes described above, such as the power scheduling methods described above. For example, in some embodiments, the power scheduling methods described above may be implemented as a computer software program tangibly embodied on a machine-readable medium, such as storage unit 1208. In some embodiments, some or all of the computer program may be loaded and/or installed onto the electronic device 1200 via the ROM 1202 and/or the communication unit 12012. When the computer program is loaded into the RAM 1203 and executed by the computing unit 1201, one or more steps of the power scheduling method described above may be performed. Alternatively, in other embodiments, the computing unit 1201 may be configured to perform the power scheduling methods described above by any other suitable means (e.g., by means of firmware).

Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuit System, FPGA (Field Programmable Gate Array ), ASIC (Application-Specific Integrated Circuit, application-specific integrated circuit), ASSP (Application Specific Standard Product, special-purpose standard product), SOC (System On Chip ), CPLD (Complex Programmable Logic Device, complex programmable logic device), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs, the one or more computer programs may be executed and/or interpreted on a programmable system including at least one programmable processor, which may be a special purpose or general-purpose programmable processor, that may receive data and instructions from, and transmit data and instructions to, a storage system, at least one input device, and at least one output device.

Program code for carrying out methods of the present disclosure may be written in any combination of one or more programming languages. These program code may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus such that the program code, when executed by the processor or controller, causes the functions/operations specified in the flowchart and/or block diagram to be implemented. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, RAM, ROM, EPROM (Electrically Programmable Read-Only-Memory, erasable programmable read-Only Memory) or flash Memory, an optical fiber, a CD-ROM (Compact Disc Read-Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display ) monitor) for displaying information to a user; and a keyboard and pointing device (e.g., a mouse or trackball) by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic input, speech input, or tactile input.

The systems and techniques described here can be implemented in a computing system that includes a background component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such background, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: LAN (Local Area Network ), WAN (Wide Area Network, wide area network), internet and blockchain networks.

The computer system may include a client and a server. The client and server are typically remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, also called a cloud computing server or a cloud host, and is a host product in a cloud computing service system, so that the defects of high management difficulty and weak service expansibility in the traditional physical hosts and VPS service (Virtual Private Server, virtual special servers) are overcome. The server may also be a server of a distributed system or a server that incorporates a blockchain.

It should be noted that, artificial intelligence is a subject of studying a certain thought process and intelligent behavior (such as learning, reasoning, thinking, planning, etc.) of a computer to simulate a person, and has a technology at both hardware and software level. Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing, and the like; the artificial intelligence software technology mainly comprises a computer vision technology, a voice recognition technology, a natural language processing technology, a machine learning/deep learning technology, a big data processing technology, a knowledge graph technology and the like.

According to the technical scheme of the embodiment of the disclosure, the operation state information of the target power grid is respectively input into a plurality of scheduling strategy models to obtain a first action sequence output by any scheduling strategy model; the power parameter adjustment actions of the multiple generator sets in the target power grid are included in any first action sequence, and the multiple scheduling strategy models are obtained through training by adopting loss functions corresponding to different optimization targets; generating a second action sequence according to the first action sequences output by the scheduling strategy models; and performing power scheduling on the plurality of generator sets according to the power parameter adjustment actions of the plurality of generator sets in the second action sequence. Therefore, by setting different optimization targets for different scheduling strategy models and combining the outputs of the scheduling strategy models adopting different optimization targets, a second action sequence combining the advantages of multiple targets can be obtained, the robustness of the second action sequence is improved, and accordingly actions are adjusted according to power parameters with higher robustness in the second action sequence, power scheduling is performed on each generator set, and the probability of occurrence of operation breakdown of a target power grid can be reduced.

It should be appreciated that various forms of the flows shown above may be used to reorder, add, or delete steps. For example, the steps recited in the present disclosure may be performed in parallel, sequentially, or in a different order, provided that the desired results of the technical solutions presented in the present disclosure are achieved, and are not limited herein.

The above detailed description should not be taken as limiting the scope of the present disclosure. It will be apparent to those skilled in the art that various modifications, combinations, sub-combinations and alternatives are possible, depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present disclosure are intended to be included within the scope of the present disclosure.

Claims

1. A power scheduling method, the method comprising:

respectively inputting the running state information into a first scheduling strategy model and at least one second scheduling strategy model to obtain a first action sequence output by any scheduling strategy model; any one of the first action sequences comprises electric power parameter adjustment actions of a plurality of generator sets in the target power grid, and a plurality of scheduling strategy models are trained by adopting loss functions corresponding to different optimization targets;

Determining the similarity between a first action sequence output by the at least one second scheduling policy model and a first action sequence output by the first scheduling policy model;

determining a weight of the at least one second scheduling policy model according to the similarity, wherein the weight is positively correlated with the similarity;

weighting the first action sequence output by the at least one second scheduling strategy model according to the weight to obtain a second action sequence;

2. The method of claim 1, wherein the first scheduling policy model is trained using a loss function corresponding to a plurality of optimization objectives, and the plurality of optimization objectives includes an optimization objective employed by the at least one second scheduling policy model.

3. The method of claim 2, wherein the determining weights for the at least one second scheduling policy model based on the similarities comprises:

determining a target coefficient according to the sum of the similarity of the at least one second scheduling policy model;

And determining the weight of the second scheduling policy model according to the ratio of the corresponding similarity to the target coefficient aiming at any second scheduling policy model.

4. The method of claim 1, wherein the power scheduling the plurality of gensets according to the power parameter adjustment actions of the plurality of gensets in the second sequence of actions comprises:

acquiring a sensitivity matrix, wherein the sensitivity matrix is a matrix of M, M is the number of generator sets in the target power grid, N is the number of sections in the target power grid, and an ith row and a jth column of elements in the sensitivity matrix are used for indicating the power flow variation of the jth section caused by the adjustment of a unit power parameter by the ith generator set, i is a positive integer not greater than M, j is a positive integer not greater than N, and M and N are both positive integers;

determining a first predicted power flow of at least one section according to the actual power flow of the at least one section, the sensitivity matrix and the second action sequence;

determining a first section from the at least one section according to a first predicted power flow of the at least one section;

according to the first predicted power flow of the first section, the second action sequence is adjusted to obtain a third action sequence;

And performing power dispatching on the plurality of generator sets according to the power parameter adjustment actions of the plurality of generator sets in the third action sequence.

5. The method of claim 4, wherein the determining a first section from the at least one section based on the first predicted power flow for the at least one section comprises:

determining a candidate section from the at least one section according to the first predicted power flow of the at least one section, wherein the first predicted power flow of the candidate section is not in a set first power flow interval corresponding to the candidate section;

determining the deviation between a first predicted power flow of any candidate section and a corresponding first power flow interval;

the first section is determined from at least one candidate section based on a deviation of the at least one candidate section.

6. The method of claim 4, wherein the acquiring a sensitivity matrix comprises:

acquiring the actual power flow of the at least one section in the target power grid;

aiming at an ith generating set in the target power grid, carrying out power parameter adjustment on the ith generating set;

Determining a second predicted power flow corresponding to the at least one section after the power parameter adjustment of the ith generator set according to the power parameter of the ith generator set before adjustment, the power parameter of the ith generator set after adjustment and the actual power flow of the at least one section;

and determining each element of the ith row in the sensitivity matrix according to the second predicted power flow and the actual power flow corresponding to the at least one section, and according to the power parameter of the ith generator set before adjustment and the power parameter of the ith generator set after adjustment.

7. The method of claim 4, wherein said adjusting the second sequence of actions to obtain a third sequence of actions based on the first predicted power flow of the first section comprises:

executing at least one cycle according to the first predicted power flow of the first section so as to update the second action sequence;

and taking the second action sequence updated in the last circulation process as the third action sequence.

8. The method of claim 7, wherein the first round of the cyclic process comprises:

Determining a first target column corresponding to the first section in the sensitivity matrix;

determining a first generator set corresponding to a maximum element and a second generator set corresponding to a minimum element in the first target column of the sensitivity matrix;

and adjusting the power parameter adjustment action of the first generator set and/or the power parameter adjustment action of the second generator set in the second action sequence to obtain a second action sequence updated in the first-round circulation process.

9. The method of claim 8, wherein the non-first round of the cyclic process comprises:

determining a third predicted power flow of the at least one section according to a second action sequence obtained by updating in the previous round of circulation, the sensitivity matrix and the actual power flow of the at least one section;

ending the cycling process if there is no second section in the at least one section; the third predicted power flow of the second section is not in a second power flow interval corresponding to the second section;

determining a third section from at least one second section according to a third predicted power flow of the second section when the second section exists in the at least one section, and determining a second target column corresponding to the third section in the sensitivity matrix;

Determining a third generator set corresponding to the largest element and a fourth generator set corresponding to the smallest element in the second target column of the sensitivity matrix;

and adjusting the power parameter adjustment action of the third generator set and/or the power parameter adjustment action of the fourth generator set in the second action sequence updated in the previous round of circulation process to obtain a second action sequence updated in the current round of circulation process.

10. The method of claim 4, wherein said adjusting the second sequence of actions to obtain a third sequence of actions based on the first predicted power flow of the first section comprises:

executing an iteration process for the second action sequence for a set number of times according to the first predicted power flow of the first section;

and updating the last iteration process to obtain a second action sequence serving as the third action sequence.

11. The method of claim 10, wherein any one of the iterative processes comprises:

determining a fourth predicted power flow of the at least one section according to a second action sequence, the sensitivity matrix and the actual power flow of the at least one section, which are updated in the previous iteration process of the present iteration process;

Determining a fourth section from the at least one section according to a fourth predicted power flow of the at least one section;

determining a third target column corresponding to the fourth section in the sensitivity matrix;

determining a fifth generator set corresponding to the largest element and a sixth generator set corresponding to the smallest element in the third target column of the sensitivity matrix;

and adjusting the power parameter adjustment action of the fifth generator set and/or the power parameter adjustment action of the sixth generator set in the second action sequence updated in the previous iteration process to obtain the second action sequence updated in the present iteration process.

12. The method according to any one of claims 4-11, wherein said adjusting the second sequence of actions according to the first predicted power flow of the first section to obtain a third sequence of actions comprises:

and adjusting the power parameter adjustment action of the generator set matched with the set adjustment strategy in the second action sequence according to the first predicted power flow of the first section so as to obtain the third action sequence.

13. The method of claim 12, wherein the adjustment strategy is a plurality and the third action sequence is a plurality,

The power scheduling of the plurality of generator sets according to the power parameter adjustment actions of the plurality of generator sets in the third action sequence includes:

obtaining scores for a plurality of the third action sequences;

determining a target action sequence from the plurality of third action sequences according to the scores of the plurality of third action sequences;

and performing power scheduling on the plurality of generator sets according to the power parameter adjustment actions of the plurality of generator sets in the target action sequence.

14. The method of claim 13, wherein the obtaining scores for a plurality of the third action sequences comprises:

scoring the third action sequences by adopting a right tail function of standard normal distribution so as to obtain scores of the third action sequences;

or,

determining a fifth predicted power flow of the at least one section according to the actual power flow of the at least one section, the third action sequence and the sensitivity matrix for any one of the third action sequences;

determining a fifth section from the at least one section according to a fifth predicted power flow of the at least one section, wherein the fifth predicted power flow of the fifth section is not in a set third power flow interval corresponding to the fifth section;

Determining a score for the third action sequence based on the number of fifth sections;

or,

aiming at any one of the third action sequences, adjusting the power parameter adjustment action of the generator set with the set generation type in the third action sequence according to the rated power parameter corresponding to the generator set with the set generation type so as to obtain an adjusted third action sequence;

determining a sixth predicted power flow of the at least one section according to the adjusted third sequence of actions, the sensitivity matrix and the actual power flow of the at least one section;

determining a sixth section from the at least one section according to a sixth predicted power flow of the at least one section, wherein the sixth predicted power flow of the sixth section is not in a set fourth power flow interval corresponding to the sixth section;

and determining the score of the third action sequence according to the number of the sixth sections.

15. A power scheduling apparatus, the apparatus comprising:

The input module is used for inputting the running state information into the first scheduling strategy model and at least one second scheduling strategy model respectively so as to obtain a first action sequence output by any scheduling strategy model; any one of the first action sequences comprises electric power parameter adjustment actions of a plurality of generator sets in the target power grid, and a plurality of scheduling strategy models are trained by adopting loss functions corresponding to different optimization targets;

the generation module is used for determining the similarity between the first action sequence output by the at least one second scheduling strategy model and the first action sequence output by the first scheduling strategy model; determining a weight of the at least one second scheduling policy model according to the similarity, wherein the weight is positively correlated with the similarity; weighting the first action sequence output by the at least one second scheduling strategy model according to the weight to obtain a second action sequence;

16. The apparatus of claim 15, wherein the first scheduling policy model is trained using a loss function corresponding to a plurality of optimization objectives, and the plurality of optimization objectives comprises an optimization objective employed by the at least one second scheduling policy model.

17. The apparatus of claim 15, wherein the scheduling module comprises:

the system comprises an acquisition unit, a sensitivity matrix, a power supply unit and a control unit, wherein the sensitivity matrix is a matrix of M, M is the number of generator sets in the target power grid, N is the number of sections in the target power grid, the ith row and the jth column of elements in the sensitivity matrix are used for indicating the power flow variation of the jth section caused by the adjustment of unit power parameters of the ith generator set, i is a positive integer not greater than M, j is a positive integer not greater than N, and M and N are both positive integers;

a first determining unit, configured to determine a first predicted power flow of at least one section according to an actual power flow of the at least one section, the sensitivity matrix, and the second action sequence;

a second determining unit configured to determine a first section from the at least one section according to a first predicted power flow of the at least one section;

the adjusting unit is used for adjusting the second action sequence according to the first predicted power flow of the first section so as to obtain a third action sequence;

18. An electronic device, comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the power scheduling method of any one of claims 1-14.

19. A non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the power scheduling method of any one of claims 1-14.

20. A computer program product comprising a computer program which, when executed by a processor, implements the steps of the power scheduling method according to any one of claims 1-14.