CN116780509A - Power grid random scene generation method integrating discrete probability and CGAN - Google Patents
Power grid random scene generation method integrating discrete probability and CGAN Download PDFInfo
- Publication number
- CN116780509A CN116780509A CN202310711243.5A CN202310711243A CN116780509A CN 116780509 A CN116780509 A CN 116780509A CN 202310711243 A CN202310711243 A CN 202310711243A CN 116780509 A CN116780509 A CN 116780509A
- Authority
- CN
- China
- Prior art keywords
- node
- power
- input
- equivalent injection
- injection power
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000002347 injection Methods 0.000 claims abstract description 107
- 239000007924 injection Substances 0.000 claims abstract description 107
- 238000009826 distribution Methods 0.000 claims abstract description 37
- 238000012549 training Methods 0.000 claims abstract description 27
- 238000011160 research Methods 0.000 claims abstract description 15
- 230000008485 antagonism Effects 0.000 claims abstract 2
- 230000006870 function Effects 0.000 claims description 31
- 238000005315 distribution function Methods 0.000 claims description 19
- 238000005070 sampling Methods 0.000 claims description 13
- 238000010276 construction Methods 0.000 claims description 11
- 239000011159 matrix material Substances 0.000 claims description 10
- 238000005457 optimization Methods 0.000 claims description 8
- 238000004458 analytical method Methods 0.000 claims description 6
- 230000006399 behavior Effects 0.000 claims description 6
- 210000002569 neuron Anatomy 0.000 claims description 6
- 238000000638 solvent extraction Methods 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 5
- 238000007619 statistical method Methods 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 4
- 238000013528 artificial neural network Methods 0.000 claims description 4
- 238000004422 calculation algorithm Methods 0.000 claims description 4
- 230000004927 fusion Effects 0.000 abstract description 5
- 230000002708 enhancing effect Effects 0.000 abstract 1
- 238000004590 computer program Methods 0.000 description 7
- 238000012545 processing Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 6
- 238000007726 management method Methods 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000010248 power generation Methods 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000003012 network analysis Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000035699 permeability Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application discloses a power grid random scene generation method for discrete probability and condition generation countermeasure network fusion, which comprises the steps of firstly, constructing a discrete joint probability model of equivalent injection power time sequence fluctuation and prediction random error of each source load node. Secondly, constructing and training conditions to generate an antagonism network (CGAN) model, and training a CGAN generator and a discriminator until convergence. And finally, based on the trained CGAN, taking the equivalent injection power prediction information of each source load node in the research time window as a condition value input by the CGAN, generating random noise based on the discrete joint probability distribution of each source load node, generating the equivalent injection power of each source load node in the research time window through the CGAN, and constructing a power grid random scene corresponding to the research time window. The application provides a random scene generation method for generating a power grid future period based on discrete probability model and condition generation and anti-learning fusion, which is used for enhancing the pertinence and adaptability of scene generation and laying a foundation for formulating a robust scheduling strategy adapting to multiple scenes.
Description
Technical Field
The application belongs to the field of power system automation, and particularly relates to an intelligent construction method for a power grid operation scene fused by discrete probability and CGAN.
Background
The high permeability renewable energy grid connection brings great challenges to the power grid, and the current power grid operation, scheduling and planning strategies need to be correspondingly adjusted to adapt to the uncertainty of the renewable energy. The new energy power generation represented by wind energy and photovoltaic is affected by the randomness and fluctuation of primary energy, and the power generation output has randomness and fluctuation in response.
In the existing research, the scene analysis method can effectively characterize the uncertainty of renewable energy source output through a group of discrete scenes, so that the uncertainty optimization problem is converted into a multi-scene deterministic optimization problem or a random optimization problem. However, due to the characteristics of complexity and unknowing of the uncertainty of the renewable energy sources, particularly when a high-proportion new energy source is accessed into a power grid in a large scale, the problems of complex and various power grid operation modes, difficulty in prediction, difficulty in planning and arrangement and the like are caused, and the problems of difficulty in covering a historical scene, low scene generation efficiency and the like exist in scene construction.
In recent years, with rapid development of artificial intelligence technology and continuous improvement of computer efficiency, historical data is analyzed through data driving to obtain an inherent statistical rule, and an uncertainty model of renewable energy sources is built by using a black box model. The artificial intelligence method has the advantages of strong processing capacity for high-dimensional nonlinear data, no need of complex physical model, high online calculation speed and the like, becomes a research hotspot in the field of power grid safety analysis, and has good application effects in the fields of power system mode analysis, transient stability, operation rules, network analysis optimization, scheduling operation and the like.
Disclosure of Invention
The application aims to overcome the defects in the prior art, provides a power grid random scene generation method integrating discrete probability and CGAN, and solves the problems of complex and various power grid operation modes, difficult planning and arrangement, low scene generation efficiency and the like.
In order to achieve the above purpose, the application is realized by adopting the following technical scheme:
in a first aspect, the application provides a method for generating a random scene of a power grid based on node equivalent injection power discrete probability model and condition generation countermeasure learning fusion, comprising the following steps:
step 1: the method comprises the steps of obtaining historical data of new energy sources and loads of source load nodes, including a predicted power historical data set and an actual power historical data set, forming a predicted power historical data set and an actual power historical data set of node equivalent injection power, respectively counting time sequence fluctuation and predicted random errors of the equivalent injection power of each source load node of a power grid based on the historical data, and constructing a discrete joint probability model of the time sequence fluctuation and the predicted random errors of the equivalent injection power of each source load node;
step 2: constructing a condition generation countermeasure network (CGAN) model, comprising a generator and a discriminator neural network architecture, taking the historical predicted power of the equivalent injection power of each source load node as a condition value, randomly sampling based on the discrete joint probability distribution of the equivalent injection power to form noise, and training a CGAN generator and a discriminator until convergence to obtain a trained CGAN generator network model and a trained discriminator network model;
step 3: obtaining power prediction data of new energy and load of a source load node corresponding to a research time window, and forming a prediction data set of node equivalent injection power; based on the trained CGAN, the prediction information of the equivalent injection power of each source load node corresponding to the research time window is used as a condition value of CGAN input, random noise is generated based on the discrete joint probability distribution of each source load node, the equivalent injection power of each source load node of the research time window is generated through the CGAN, and a power grid random scene corresponding to the research time window is constructed.
Further, in step 1, a predicted power history data set and an actual power history data set of new energy and load of a source load node are obtained, a predicted power history data set and an actual power history data set of node equivalent injection power are formed, and a discrete probability joint distribution function of the node equivalent injection power is constructed based on time sequence fluctuation and a predicted random error of the node equivalent injection power, and the method comprises the following steps:
and obtaining a predicted power historical data set and an actual power historical data set of the new energy and the load of the source load node from a new energy prediction and load prediction database of the grid EMS energy management system to form a predicted power historical data set and an actual power historical data set of the node equivalent injection power.
Wherein P is a，i (t) is the equivalent actual power of node i at time t,wind power at time t for node i, +.>For the photovoltaic output of node i at time t, < >>The load power of the node i at the time t is obtained; p (P) f，i (t) is the equivalent predicted power of node i at time t,>wind power prediction power for node i at time t, < >>Predicted power for node i at time t, +.>Predicting power for the load of the node i at the time t;
counting and analyzing time sequence fluctuation and prediction random errors of the equivalent injection power of each source load node, and constructing a time sequence fluctuation model and a discrete probability model of the prediction errors of the equivalent injection power of each source load node;
and combining a time sequence fluctuation model of the node equivalent injection power based on historical data statistical analysis and a discrete probability model of the prediction error to obtain a discrete probability joint distribution function.
Further, in step 1, combining a time sequence fluctuation model of the node equivalent injection power based on statistical analysis of historical data with a discrete probability model of a prediction error to obtain a discrete probability joint distribution function, including:
statistics and analysis of historical time sequence fluctuation and prediction random error of equivalent injection power of each source load node, and construction of equivalent injection power t of each source load node i Time (t) i ∈[t 1 ，t 2 ]) Time sequence fluctuation model of (a)And a discrete probability model that predicts random errors;
binding t i And calculating the discrete probability joint distribution condition of the injection power of each node at the moment to obtain the discrete joint probability distribution of the injection power of the node at the moment.
Further, combine t i The discrete probability model of the historical time sequence fluctuation and the prediction random error of the equivalent injection power of each source load node at the moment calculates the discrete joint probability distribution of the equivalent injection power of each node at the moment, and the method comprises the following steps:
the time sequence fluctuation characteristic of the equivalent injection power of the node i at the time t can be expressed as follows:
ΔP i (t)＝P a，i (t)-P a，i (t-1) (3)
the prediction error of the equivalent injection power of the node i at this time can be expressed as:
ε i (t)＝P a，i (t)-p f，i (t) (4)
based on formulas (3) and (4), the discrete probability model construction steps of the nodes are as follows:
as can be seen from the formula (3), the volatility is the equivalent injection power at time t+1 minus the equivalent injection power at time t; taking a discretization formula factor C, partitioning the fluctuation value of the node equivalent injection power in the variation range of the fluctuation value, respectively counting the occurrence frequency of the fluctuation value of each discretization interval, taking the frequency as the estimated value of the probability that the fluctuation value falls into the discretization interval, and calculating the expected value of all the fluctuation values falling into each interval;
use (2 XN) w ) Matrix G of w Representing discrete probability distribution of time sequence fluctuation of node equivalent injection power in historical data, G w The 1 st behavior fluctuation interval of (2) and the probability of the 2 nd behavior falling into the fluctuation interval are G w Can be expressed as:
in N w As a time-series waveThe number of states of the mobility discrete probability distribution satisfies:
delta in w，max The maximum value of the equivalent injection power fluctuation of the node in the historical data;representation pair->The value of (2) is rounded down;
counting equivalent actual power of k times of node i in historical data to enable delta to be calculated w (t) is the fluctuation value at t hours, G w The elements in (a) can be calculated by the following formula:
note that: k-1 fluctuation values can be obtained in k time periods; f in i (x) As an indication function, which is defined as
According to the above formula (6) (7) (8) (9), the (k-1) fluctuation values of the k period nodes i can be converted into N-containing values w Discrete probability distribution of individual state values and corresponding probabilities thereof, namely formula (5), maintains probability characteristics of time sequence fluctuation while merging states;
dividing the prediction error probability based on the delta t of the prediction time from the current time; calculating the error of the corresponding time period according to a formula (4), partitioning the error in the variation range of the error, respectively counting the occurrence frequency of the error value of each discretization interval as the estimated value of the probability that the error value falls into the discretization interval, and calculating the expected value of all the error values falling into each interval so as to obtain the discrete probability distribution of the error;
counting the equivalent injection power of the k time period nodes i in the historical data to obtain k error values which are converted into N-containing values w Discrete probability distributions of individual state values and their corresponding probabilities;
for discrete random variables, the joint distribution probability function is P (x=x & y=y), i.e.:
P(Y＝y|X＝x)P(X＝x)＝P(X＝x|Y＝y)P(Y＝y) (10)
joint distribution function:
using equations (10) and (11), a joint distribution function of the volatility and the random error can be obtained.
The fluctuation change range of the equivalent injection power of each source load node at each moment can be obtained through the obtained discrete joint distribution function, and the possible random change trend of each source load node based on the historical actual data can be obtained by combining the power prediction information of the corresponding node.
Further, in step 2, a condition generating countermeasure network (CGAN) model is constructed, including a generator and a arbiter neural network architecture; taking the historical predicted power of the equivalent injection power of each source load node as a condition value, carrying out random sampling based on the discrete joint probability distribution of the equivalent injection power to form noise, and training a CGAN generator and a discriminator until convergence to obtain a trained CGAN generator network model and a trained discriminator network model, wherein the method comprises the following steps:
the construction condition generates an countermeasure network model CGAN, an input layer and a hidden layer of the generator respectively comprise a full-connection layer, a BN layer and an activation function, 256 neurons and 512 neurons respectively, and an output layer comprises the full-connection layer and a Tanh function. The input layer of the discriminator comprises a full connection layer, a BN layer and an activation function, 256 neurons are contained, and the output layer comprises the full connection layer and a Sigmond function.
Based on a predicted power historical data set of the equivalent injection power of the source load node of the power grid, performing time sequence transverse splicing on the predicted information of the equivalent injection power of each source load node to serve as a condition value C of CGAN input, performing random number generation based on discrete joint probability distribution obtained by each node, performing corresponding time sequence transverse splicing to form a noise Z input sequence, performing longitudinal splicing on the condition value C and the noise Z to serve as input of a generator, performing longitudinal splicing on the equivalent injection power of a historical scene node to serve as input of a real sample X, performing longitudinal splicing on the real sample X and the noise Z to serve as input of the generator, and training a condition generation countermeasure network model CGAN;
taking gradient punishment sampling points, calculating a loss function of the generator and the discriminator, and respectively updating the network weights of the generator and the discriminator by adopting an adam optimization algorithm;
if the training is not finished, returning to the next round of training, and when the loss functions of the generator and the discriminator are converged, finishing the training.
Further, training is ended when the loss functions of both the generator and the arbiter converge.
Further, performing time sequence transverse splicing on the equivalent injection power prediction information of each source load node to serve as a condition value C input by a CGAN, performing random number generation based on discrete joint probability distribution obtained by each node, performing corresponding time sequence transverse splicing to form a noise Z input sequence, performing longitudinal splicing on the condition value C and the noise Z to serve as input of a generator, performing longitudinal splicing on the equivalent injection power of a historical scene node to serve as input of a real sample X, performing longitudinal splicing on the real sample X and the noise Z to serve as input of the generator, and training a condition generation countermeasure network model CGAN, wherein the method comprises the following steps:
the equivalent injection power predicted value of each source load node at each moment is matched with noise, and a research time window [ t ] is set 1 ，t 2 ]M times, Z i，j And C i，j The random sampling noise input and the conditional input at the j-th moment of the node i are respectively represented, and then the noise input sequence set of the node i is Z i ＝{Z i，1 ，Z i，2 ，...，Z i，m-1 ，Z i，m The conditional input sequence set is C i ＝{C i，1 ，C i，2 ，...，C i，m-1 ，C i，m The noise input length is consistent with the condition input length, and the noise input z and the condition input c are longitudinally spliced into a matrix, so that the noise and the prediction information form an up-down corresponding relation at each moment;
the real sample input sequence set of the historical data node i is X i ＝{X i，1 ，X i，2 ，...，X i，m-1 ，X i，m The input length of the real sample of the historical data is consistent with the condition input length, the input of the real sample X of the historical data and the condition input c are longitudinally spliced into a matrix, and the real sample of the historical data and the prediction information form an up-down corresponding relation at each moment;
generating a sample input sequence set as X' i ＝{X′ i，1 ，X′ i，2 ，...，X′ i，m-1 ，X′ i，m The input length of the generated sample is consistent with the condition input length, and the generated sample X' input and the condition input c are longitudinally spliced into a matrix, so that the generated sample and the prediction information form an up-down corresponding relation at each moment;
inputting the data subjected to longitudinal splicing of the noise Z and the predicted value C into a generator G, and outputting a generated sample X'; the predicted value C and the real sample X are longitudinally spliced and then input into a discriminator D, and the predicted value C and the generated sample X' are longitudinally spliced and then input into the discriminator D, and the discriminator D outputs the discriminating value of the real sample and the generated sample.
Further, the method further comprises:
and evaluating the generated scene set from the aspects of volatility, time correlation and effectiveness 3, and screening the scene set.
Compared with the prior art, the application has the beneficial effects that:
1. the application discloses an intelligent construction method for a power grid operation scene considering multiple uncertainties, and belongs to the field of power system automation. Aiming at the problems that the current high-proportion new energy is accessed into a power grid in a large scale, so that the power grid operation mode is complex and various, the planning and arrangement are difficult, and the like, the application provides a power grid random scene generation method for generating the challenge learning fusion based on a discrete probability model and conditions, so that the pertinence and scene adaptability of sample generation are enhanced.
2. The specific functions of the generator of the application are: historical data is input, corresponding scene expansion scenes are output, and the problem of too few scenes is solved to a great extent.
Drawings
Fig. 1 is a flow chart of the present application.
Detailed Description
The application is further described below with reference to the accompanying drawings. The following examples are only for more clearly illustrating the technical aspects of the present application, and are not intended to limit the scope of the present application.
Embodiment one:
aiming at the problems that the current high-proportion new energy is accessed into a power grid in a large scale, so that the operation modes of the power grid are complex and various, the planning and the arrangement are difficult, and the like, the embodiment provides a power grid random scene generation method based on power injection node discrete probability model and condition generation countermeasure learning fusion, which comprises the following steps:
step 1: historical data of grid source load node injection power is obtained from a grid EMS energy management system, fluctuation characteristics and prediction random errors of the node injection power are counted and analyzed, and a time sequence fluctuation model and a prediction error discrete probability model of the grid source load node injection power based on historical data statistic analysis are calculated respectively.
Step 1.1: time period information [ t ] corresponding to scene to be generated aiming at power grid 1 ，t 2 ]Obtaining power grid bus node source load power prediction data from new energy prediction and load prediction software of a power grid EMS energy management system, forming node injection power prediction information, carrying the node injection power prediction information into the discrete probability model mentioned in the step 1, and respectively calculating t i Time (t) i ∈[t 1 ，t 2 ]) A time sequence fluctuation discrete probability model of node injection power and a prediction error random discrete probability model.
Step 1.2: binding t i And calculating a discrete probability joint distribution function of the injection power of the moment node to obtain the discrete probability distribution of the random predicted value.
Step 2: building a condition generation countermeasure network CGAN model, dividing the power grid busbar node source load power prediction data set mentioned in the step 1.1 into a training set and a test set, and dividing [ t ] in the training set 1 ，t 2 ]The method comprises the steps that time period node injection power prediction information is used as a condition value C input by a CGAN, random number generation is carried out based on a discrete probability joint distribution function obtained by 1.2, a noise Z input sequence is formed, the condition value C and the noise Z are longitudinally spliced to be used as input of a generator, grid source load injection node historical data is used as a real sample X input by the CGAN, and the condition value C and the real sample X are longitudinally spliced to be used as input of a discriminator;
and taking gradient punishment sampling points, calculating the loss functions of the generator and the discriminator, and respectively updating the network weights of the generator and the discriminator by adopting an adam optimization algorithm. Training is ended when the loss function images of both the generator and the arbiter converge. If the training is not finished, returning to the next round of training.
Step 3: after training, a generator model in the CGAN is extracted, the node injection power prediction information of the test set in the corresponding period is selected to serve as a condition value C input by the CGAN, longitudinal splicing is carried out on the condition value C and the discrete probability distribution noise Z meeting the random prediction value, spliced data are input into a trained generator G, and a scene data set based on the prediction value can be generated.
Specifically, the technical scheme adopted by the application is as follows:
firstly, constructing a discrete random probability model of a new energy node and a load node based on historical data statistical analysis, wherein the time sequence fluctuation discrete random probability model can be expressed as:
the prediction error discrete random probability model can be expressed as:
wherein the method comprises the steps ofA power prediction value representing the power injection node i in period t +.>Representing the actual power of the power injection node i during period t.
Based on formulas (1) and (2), the new energy node discrete random probability model construction steps are as follows:
as can be seen from the formula (1), the fluctuation is the injection power at time t+1 minus the injection power at time t. Taking a discretization formula factor C, partitioning the fluctuation value of the node injection power in the variation range of the fluctuation value, respectively counting the occurrence frequency of the fluctuation value of each discretization interval, taking the frequency as the estimated value of the probability that the fluctuation value falls into the discretization interval, and calculating the expected value of all the fluctuation values falling into each interval.
Using 1 (2 XN) w ) Matrix G of w Characterizing the discrete probability distribution of the volatility of the busbar load over the investigation period, G w The 1 st behavior fluctuation value of (2), namely the state value, the probability corresponding to the 2 nd behavior fluctuation value, G w Can be expressed as:
in N w The state number of the discrete probability distribution is as follows:
delta in w，max To study the maximum value of the fluctuation in the period;representation pair->C is a discretized formula factor;
counting equivalent actual power of k times of node i in historical data to enable delta to be calculated w (t) is the fluctuation value at t hours, G w The elements in (a) can be calculated by the following formula:
note that: k time nodes can find k-1 fluctuation values. F in i (x) As an indication function, which is defined as
According to the above formulas (4), (5), (6) and (7), k-1 fluctuation values of the bus load of the k time node can be converted into N-containing values w The discrete probability distribution of the state values and their corresponding probabilities, equation (3), preserves the probability characteristics of the volatility while merging the states.
Dividing the prediction error probability based on deltat of the prediction time from the current time, solving the error of the corresponding time period according to a formula (2), partitioning the error in the variation range of the error, respectively counting the occurrence frequency of error values of each discretization interval as the estimated value of the probability that the error value falls into the discretization interval, and calculating the expected value of all the error values falling into each interval so as to obtain the discrete probability distribution of the error;
counting the equivalent injection power of the k time period nodes i in the historical data to obtain k error values which are converted into N-containing values w Discrete probability distributions of individual state values and their corresponding probabilities.
For discrete random variables, the joint distribution probability function is P (x=x & y=y), i.e.:
P(Y＝y|X＝x)P(X＝x)＝P(X＝x|Y＝y)P(Y＝y) (8)
from an observation of the formula, it is a probability function that X and Y occur simultaneously, and its (joint) distribution function is then the accumulation of the probability functions. Similarly, the joint distribution function:
by using the formulas (8) and (9), the joint distribution function of the fluctuation and the random error can be obtained.
And the change range of each time node on the basis of the historical data can be obtained through the obtained joint distribution function. The historical data is put into the probability model, and as a result, random predictors which account for fluctuations and errors and vary around the historical data can be obtained.
Secondly, taking the [ t_1, t_2] period node injection power prediction information in the training set as a condition value C of CGAN input, generating random numbers based on a discrete probability joint distribution function acquired by 1.2 to form a noise Z input sequence, longitudinally splicing the condition value C and the noise Z to be used as the input of a generator, taking the grid source load injection node historical data as a real sample X of CGAN input, and longitudinally splicing the condition value C and the real sample X to be used as the input of a discriminator.
Assuming m time nodes in a set prediction time interval, Z i，j And C i，j The random sampling noise input and the conditional input at the j-th moment of the node i are respectively represented, and then the noise input sequence set of the node i is Z i ＝{Z i，1 ，Z i，2 ，...，Z i，m-1 ，Z i，m The conditional input sequence set is C i ＝{C i，1 ，C i，2 ，...，C i，m-1 ，C i，m And matching a noise to the predicted value at each time. NoiseThe input length is consistent with the condition input length, the noise input z and the condition input c are longitudinally spliced into a matrix, and the noise and the prediction information form an up-down corresponding relation at each moment. Similarly, the same is true for the input data processing of the real sample X of the historical section data and the predicted data condition value C, and the input data processing of the predicted data condition value C and the generated sample X'.
And inputting the data subjected to longitudinal splicing of the noise Z and the predicted value C into a generator G, and outputting a generated sample X'. The predicted value C and the real sample X are longitudinally spliced and then input into a discriminator D, and the predicted value C and the generated sample X' are longitudinally spliced and then input into the discriminator D, and the discriminator D outputs the discriminating value of the real sample and the generated sample.
Thirdly, selecting gradient punishment sampling points, calculating a loss function of the generator and the discriminator, and respectively updating network weights of the generator and the discriminator by adopting an adam optimization algorithm. If the training is not finished, returning to the next round of training.
Finally, when the loss functions of the discriminator and the generator are converged, the training is finished, a generator model is extracted from the condition generation countermeasure network CGAN, the power grid source load injection node prediction data in the test set is selected as the condition C to be longitudinally spliced with the random noise Z, the spliced data is input into the trained generator G, a scene data set based on the prediction value can be generated,
in general, firstly, the node injection power fluctuation and the prediction random error are statistically analyzed based on the historical data of the grid source load injection node, and a discrete random probability model of the power injection node is constructed. Then, a counternetwork (Conditional Generative Adversarial Nets, CGAN) model is generated by using a pyrach building condition, random sampling is carried out based on a discrete random probability model of the injection node, the sampling result is used as noise input by the CGAN, and historical actual operation scene data is used as a true value input by the CGAN. Finally, the generated scene sets are evaluated in terms of volatility, time correlation and effectiveness 3, respectively. The simulation result verifies the effectiveness of the intelligent construction method of the power grid operation scene.
It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The foregoing is merely a preferred embodiment of the present application, and it should be noted that modifications and variations could be made by those skilled in the art without departing from the technical principles of the present application, and such modifications and variations should also be regarded as being within the scope of the application.
Claims (7)
1. A power grid random scene generation method based on a discrete probability model and condition generation of source load node equivalent injection power is characterized by comprising the following steps:
the method comprises the steps of obtaining historical data of new energy sources and loads of source load nodes, including a predicted power historical data set and an actual power historical data set, forming a predicted power historical data set and an actual power historical data set of node equivalent injection power, respectively counting time sequence fluctuation and predicted random errors of the equivalent injection power of each source load node of a power grid based on the historical data, and constructing a discrete joint probability model of the time sequence fluctuation and the predicted random errors of the equivalent injection power of each source load node;
the construction condition generates an antagonism network model CGAN, which comprises a generator and a discriminant neural network architecture, takes the historical predicted power of the equivalent injection power of each source load node as a condition value, carries out random sampling based on the discrete joint probability distribution of the equivalent injection power to form noise, and trains the CGAN generator and the discriminant until convergence to obtain a trained generator network model and a trained discriminant network model of the CGAN;
obtaining power prediction data of new energy and load of a source load node corresponding to a research time window, and forming a prediction data set of node equivalent injection power; based on the trained CGAN, the prediction information of the equivalent injection power of each source load node corresponding to the research time window is used as a condition value of CGAN input, random noise is generated based on the discrete joint probability distribution of each source load node, the equivalent injection power of each source load node of the research time window is generated through the CGAN, and a power grid random scene corresponding to the research time window is constructed.
2. The power grid random scene generating method according to claim 1, wherein obtaining a predicted power history data set and an actual power history data set of new energy sources and loads of source load nodes, forming a predicted power history data set and an actual power history data set of node equivalent injection power, and constructing a discrete probability joint distribution function of the node equivalent injection power based on time sequence fluctuation of the node equivalent injection power and a predicted random error, comprises:
obtaining a predicted power historical data set and an actual power historical data set of new energy and load of a source load node from a new energy prediction and load prediction database of an EMS energy management system of the power grid, and forming a predicted power historical data set and an actual power historical data set of node equivalent injection power;
wherein P is a，i (t) is the equivalent actual power of node i at time t,for the wind power of the node i at the time t,for the photovoltaic output of node i at time t, < >>The load power of the node i at the time t is obtained; p (P) f，i (t) is the equivalent predicted power of node i at time t,>wind power prediction power for node i at time t, < >>Predicted power for node i at time t, +.>Predicting power for the load of the node i at the time t;
counting and analyzing time sequence fluctuation and prediction random errors of the equivalent injection power of each source load node, and constructing a time sequence fluctuation model and a discrete probability model of the prediction errors of the equivalent injection power of each source load node;
and combining a time sequence fluctuation model of the node equivalent injection power based on historical data statistical analysis and a discrete probability model of the prediction error to obtain a discrete probability joint distribution function.
3. The method for generating the random scene of the power grid according to claim 2, wherein combining the time sequence fluctuation model of the node equivalent injection power based on the statistical analysis of the historical data with the discrete probability model of the prediction error to obtain the discrete probability joint distribution function comprises:
statistics and analysis of historical time sequence fluctuation and prediction random error of equivalent injection power of each source load node, and construction of equivalent injection power t of each source load node i Time (t) i ∈[t 1 ，t 2 ]) A time sequence fluctuation model of (1) and a discrete probability model for predicting random errors;
binding t i And calculating the discrete probability joint distribution condition of the injection power of each node at the moment to obtain the discrete joint probability distribution of the injection power of the node at the moment.
4. A method of generating a grid random scene as recited in claim 3, in combination with t i The discrete probability model of the historical time sequence fluctuation and the prediction random error of the equivalent injection power of each source load node at the moment calculates the discrete joint probability distribution of the equivalent injection power of each node at the moment, and the method comprises the following steps:
the time sequence fluctuation characteristic of the equivalent injection power of the node i at the time t can be expressed as follows:
ΔP i (t)＝P a，i (t)-P a，i (t-1) (3)
the prediction error of the equivalent injection power of the node i at this time can be expressed as:
ε i (t)＝P a，i (t)-p f，i (t) (4)
based on formulas (3) and (4), the discrete probability model construction steps of the nodes are as follows:
as can be seen from the formula (3), the volatility is the equivalent injection power at time t+1 minus the equivalent injection power at time t; taking a discretization formula factor C, partitioning the fluctuation value of the node equivalent injection power in the variation range of the fluctuation value, respectively counting the occurrence frequency of the fluctuation value of each discretization interval, taking the frequency as the estimated value of the probability that the fluctuation value falls into the discretization interval, and calculating the expected value of all the fluctuation values falling into each interval;
use (2 XN) w ) Matrix G of G Representing discrete probability distribution of time sequence fluctuation of node equivalent injection power in historical data, G w The 1 st behavior fluctuation interval of (2) and the probability of the 2 nd behavior falling into the fluctuation interval are G w Can be expressed as:
in N w The state number of the discrete probability distribution of the time sequence fluctuation is as follows:
delta in w，max The maximum value of the equivalent injection power fluctuation of the node in the historical data;representation pair->The value of (2) is rounded down;
counting equivalent actual power of k times of node i in historical data to enable delta to be calculated w (t) is the fluctuation value at t hours, G w The elements in (a) can be calculated by the following formula:
note that: k-1 fluctuation values can be obtained in k time periods; f in i (x) As an indication function, which is defined as
According to the above formula (6) (7) (8) (9), the (k-1) fluctuation values of the k period nodes i can be converted into N-containing values w Discrete probability distribution of individual state values and corresponding probabilities thereof, namely formula (5), maintains probability characteristics of time sequence fluctuation while merging states;
dividing the prediction error probability based on the delta t of the prediction time from the current time; calculating the error of the corresponding time period according to a formula (4), partitioning the error in the variation range of the error, respectively counting the occurrence frequency of the error value of each discretization interval as the estimated value of the probability that the error value falls into the discretization interval, and calculating the expected value of all the error values falling into each interval so as to obtain the discrete probability distribution of the error;
counting the equivalent injection power of the k time period nodes i in the historical data to obtain k error values which are converted into N-containing values w Discrete probability distributions of individual state values and their corresponding probabilities;
for discrete random variables, the joint distribution probability function is P (x=x & y=y), i.e.:
P(Y＝y|X＝x)P(X＝x)＝P(X＝x|Y＝y)P(Y＝y) (10)
joint distribution function:
using equations (10) and (11), a joint distribution function of the volatility and the random error can be obtained.
The fluctuation change range of the equivalent injection power of each source load node at each moment can be obtained through the obtained discrete joint distribution function, and the possible random change trend of each source load node based on the historical actual data can be obtained by combining the power prediction information of the corresponding node.
5. The grid random scene generation method according to claim 1, wherein the building condition generation countermeasure network model CGAN includes a generator and a arbiter neural network architecture; taking the historical predicted power of the equivalent injection power of each source load node as a condition value, carrying out random sampling based on the discrete joint probability distribution of the equivalent injection power to form noise, and training a CGAN generator and a discriminator until convergence to obtain a trained CGAN generator network model and a trained discriminator network model, wherein the method comprises the following steps:
setting up conditions to generate an countermeasure network model CGAN, wherein an input layer and a hidden layer of the generator respectively comprise a full-connection layer, a BN layer and an activation function, 256 neurons and 512 neurons are respectively contained, and an output layer comprises the full-connection layer and a Tanh function; the input layer of the discriminator comprises a full-connection layer, a BN layer and an activation function, 256 neurons are contained, and the output layer comprises the full-connection layer and a Sigmond function;
based on a predicted power historical data set of the equivalent injection power of the source load node of the power grid, performing time sequence transverse splicing on the predicted information of the equivalent injection power of each source load node to serve as a condition value C of CGAN input, performing random number generation based on discrete joint probability distribution obtained by each node, performing corresponding time sequence transverse splicing to form a noise Z input sequence, performing longitudinal splicing on the condition value C and the noise Z to serve as input of a generator, performing longitudinal splicing on the equivalent injection power of a historical scene node to serve as input of a real sample X, performing longitudinal splicing on the real sample X and the noise Z to serve as input of the generator, and training a condition generation countermeasure network model CGAN;
taking gradient punishment sampling points, calculating a loss function of the generator and the discriminator, and respectively updating the network weights of the generator and the discriminator by adopting an adam optimization algorithm;
if the training is not finished, returning to the next round of training, and when the loss functions of the generator and the discriminator are converged, finishing the training.
6. The grid random scene generation method according to claim 5, wherein training is ended when the loss functions of the generator and the arbiter are converged.
7. The method for generating a random scene of a power grid according to claim 5, wherein the method comprises the steps of performing time sequence transverse splicing of the prediction information of the equivalent injection power of each source load node as a condition value C of CGAN input, performing random number generation based on discrete joint probability distribution obtained by each node, performing corresponding time sequence transverse splicing to form a noise Z input sequence, performing longitudinal splicing of the condition value C and the noise Z as input of a generator, performing longitudinal splicing of the equivalent injection power of a historical scene node as an actual sample X, performing longitudinal splicing of the actual sample X and the noise Z as input of the generator, and training a condition generation countermeasure network model CGAN, and comprises the following steps:
the equivalent injection power predicted value of each source load node at each moment is matched with noise, and a research time window [ t ] is set 1 ，t 2 ]M times, Z i，j And C i，j The random sampling noise input and the conditional input at the j-th moment of the node i are respectively represented, and then the noise input sequence set of the node i is Z i ＝{Z i，1 ，Z i，2 ，...，Z i，m-1 ，Z i，m The conditional input sequence set is C i ＝{C i，1 ，C i，2 ，...，C i，m-1 ，C i，m Noise inputThe length is consistent with the condition input length, the noise input z and the condition input c are longitudinally spliced into a matrix, and the noise and the prediction information form an up-down corresponding relation at each moment;
the real sample input sequence set of the historical data node i is X i ＝{X i，1 ，X i，2 ，...，X i，m-1 ，X i，m The input length of the real sample of the historical data is consistent with the condition input length, the input of the real sample X of the historical data and the condition input c are longitudinally spliced into a matrix, and the real sample of the historical data and the prediction information form an up-down corresponding relation at each moment;
generating a sample input sequence set as X' i ＝{X′ i，1 ，X′ i，2 ，...，X′ i，m-1 ，X′ i，m The input length of the generated sample is consistent with the condition input length, and the generated sample X' input and the condition input c are longitudinally spliced into a matrix, so that the generated sample and the prediction information form an up-down corresponding relation at each moment;
inputting the data subjected to longitudinal splicing of the noise Z and the predicted value C into a generator G, and outputting a generated sample X'; the predicted value C and the real sample X are longitudinally spliced and then input into a discriminator D, and the predicted value C and the generated sample X' are longitudinally spliced and then input into the discriminator D, and the discriminator D outputs the discriminating value of the real sample and the generated sample.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310711243.5A CN116780509A (en) | 2023-06-15 | 2023-06-15 | Power grid random scene generation method integrating discrete probability and CGAN |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310711243.5A CN116780509A (en) | 2023-06-15 | 2023-06-15 | Power grid random scene generation method integrating discrete probability and CGAN |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116780509A true CN116780509A (en) | 2023-09-19 |
Family
ID=88009297
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310711243.5A Pending CN116780509A (en) | 2023-06-15 | 2023-06-15 | Power grid random scene generation method integrating discrete probability and CGAN |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116780509A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117060402A (en) * | 2023-10-09 | 2023-11-14 | 山东浪潮数字能源科技有限公司 | Energy internet platform architecture method based on distributed smart grid |
-
2023
- 2023-06-15 CN CN202310711243.5A patent/CN116780509A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117060402A (en) * | 2023-10-09 | 2023-11-14 | 山东浪潮数字能源科技有限公司 | Energy internet platform architecture method based on distributed smart grid |
| CN117060402B (en) * | 2023-10-09 | 2024-01-09 | 山东浪潮数字能源科技有限公司 | Energy internet platform architecture method based on distributed smart grid |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Yan et al. | Unsupervised learning for fault detection and diagnosis of air handling units | |
| Moeini et al. | Fuzzy rule-based model for hydropower reservoirs operation | |
| Wang et al. | A hybrid optimization-based recurrent neural network for real-time data prediction | |
| CN113505458A (en) | Cascading failure key trigger branch prediction method, system, equipment and storage medium | |
| CN116780509A (en) | Power grid random scene generation method integrating discrete probability and CGAN | |
| Li et al. | A novel reliability estimation method of multi-state system based on structure learning algorithm | |
| Dash et al. | Exact model averaging with naive Bayesian classifiers | |
| Sun et al. | Small sample reliability assessment with online time-series data based on a worm wasserstein generative adversarial network learning method | |
| Cao et al. | Fast and explainable warm-start point learning for AC Optimal Power Flow using decision tree | |
| CN117318018A (en) | Short-term prediction method and system for wind power output | |
| Georgopoulou et al. | Two-level, two-objective evolutionary algorithms for solving unit commitment problems | |
| CN112100784A (en) | Power distribution network planning method based on clustering net rack topology and reliability estimation association model | |
| CN112183872A (en) | Blast furnace gas generation amount prediction method combining generation of countermeasure network and neural network | |
| CN115936926A (en) | SMOTE-GBDT-based unbalanced electricity stealing data classification method and device, computer equipment and storage medium | |
| CN113128130B (en) | Real-time monitoring method and device for judging stability of direct-current power distribution system | |
| CN112600208B (en) | Multi-energy flow power flow calculation method and system for regional distributed energy system | |
| Naybour et al. | Efficient risk based optimization of large system models using a reduced petri net methodology | |
| Hu et al. | A short-term wind speed forecasting model based on improved QPSO optimizing LSSVM | |
| Gu et al. | Machine learning enhanced narmax model for dst index forecasting | |
| Flores et al. | Wind speed forecasting using a hybrid neural-evolutive approach | |
| Liu et al. | A fusion prognostic approach based on multi-kernel relevance vector machine and Bayesian model averaging | |
| CN113705067B (en) | Microgrid optimization operation strategy generation method, system, equipment and storage medium | |
| CN112531741B (en) | Decision method and device for flexible load to participate in frequency modulation | |
| Mirshekali et al. | Reinforcement Learning-Based Prediction of Alarm Significance in Marginally Operating Electrical Grids | |
| CN117557047A (en) | Power distribution equipment operation and maintenance optimization method and system based on deep reinforcement learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |