CN116169776A - Cloud edge cooperative artificial intelligent regulation and control method, system, medium and equipment for electric power system - Google Patents

Cloud edge cooperative artificial intelligent regulation and control method, system, medium and equipment for electric power system Download PDF

Info

Publication number
CN116169776A
CN116169776A CN202211089297.4A CN202211089297A CN116169776A CN 116169776 A CN116169776 A CN 116169776A CN 202211089297 A CN202211089297 A CN 202211089297A CN 116169776 A CN116169776 A CN 116169776A
Authority
CN
China
Prior art keywords
sub
value
matrix
action
power system
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
Application number
CN202211089297.4A
Other languages
Chinese (zh)
Inventor
李文臣
黄彦浩
何春江
仲悟之
褚晓杰
许成龙
许沛东
高天露
张俊
严剑峰
吕晨
刘新元
暴悦爽
郑惠萍
邹卫美
李勤新
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
State Grid Electric Power Research Institute Of Sepc
State Grid Corp of China SGCC
Wuhan University WHU
China Electric Power Research Institute Co Ltd CEPRI
Original Assignee
State Grid Electric Power Research Institute Of Sepc
State Grid Corp of China SGCC
Wuhan University WHU
China Electric Power Research Institute Co Ltd CEPRI
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by State Grid Electric Power Research Institute Of Sepc, State Grid Corp of China SGCC, Wuhan University WHU, China Electric Power Research Institute Co Ltd CEPRI filed Critical State Grid Electric Power Research Institute Of Sepc
Priority to CN202211089297.4A priority Critical patent/CN116169776A/en
Publication of CN116169776A publication Critical patent/CN116169776A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J13/00—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network
    • H02J13/00032—Systems characterised by the controlled or operated power network elements or equipment, the power network elements or equipment not otherwise provided for
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J13/00—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network
    • H02J13/00002—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network characterised by monitoring
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J13/00—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network
    • H02J13/00006—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network characterised by information or instructions transport means between the monitoring, controlling or managing units and monitored, controlled or operated power network element or electrical equipment
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/04—Circuit arrangements for ac mains or ac distribution networks for connecting networks of the same frequency but supplied from different sources
    • H02J3/06—Controlling transfer of power between connected networks; Controlling sharing of load between connected networks
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/38—Arrangements for parallely feeding a single network by two or more generators, converters or transformers
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
    • H02J2203/10—Power transmission or distribution systems management focussing at grid-level, e.g. load flow analysis, node profile computation, meshed network optimisation, active network management or spinning reserve management
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
    • H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Power Engineering (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Supply And Distribution Of Alternating Current (AREA)

Abstract

The invention discloses a cloud edge collaborative artificial intelligent regulation and control method, a system, a medium and equipment for an electric power system, wherein the method comprises the following steps: determining a predicted value corresponding to each sub-region based on observed data of all sub-regions in the power system; determining an execution action corresponding to each sub-region based on the predicted value; and executing the joint action based on the execution action corresponding to each sub-area so as to regulate and control the power system. According to the invention, operation is performed in each sub-area by using the corresponding edge platform, wide-area coordinated operation control of the power system is realized based on the cloud cooperation technology, effective decoupling and task sinking of the original centralized task can be realized, and a foundation is provided for maintaining safe and stable operation of a large-scale interconnected power grid under high-proportion new energy access.

Description

Cloud edge cooperative artificial intelligent regulation and control method, system, medium and equipment for electric power system
Technical Field
The invention relates to the technical field of crossing of artificial intelligence and an electric power system, and in particular relates to a cloud edge collaborative artificial intelligence regulation and control method and system of the electric power system.
Background
The national development of renewable energy source access is very important to relieve energy shortage and optimize energy structure. With the access of high-proportion new energy and high-proportion power electronic equipment, the power system presents a double-high characteristic, which brings randomness and fluctuation to the power system. It can be seen that the high-proportion new energy power grid has the characteristics of complex operation scene and many emergencies. The traditional centralized control operation means may not operate effectively due to the reasons that the adaptability of the stable control strategy is reduced, the communication requirement is expanded due to the fact that the diversity of controllable equipment causes a certain calculation burden and the operating scene is changed, and the like. In addition, the regional power grid scheduling capability is limited under the traditional scheduling scheme, the power generation plan is not matched with the new energy consumption requirement, and an effective multi-region scheduling coordination mechanism is lacked.
In recent years, deep reinforcement learning (Deep Reinforcement Learning, DRL) verifies the advantages of the deep reinforcement learning in planning and decision-making problems in the scenes of automatic driving, games and the like, can also be applied to solving the scheduling and control problems of a power system, and can extract knowledge from historical data to treat uncertainty of the power system. The DRL method is adopted by a plurality of students to solve the problems of load frequency control, emergency control of an electric power system, voltage control and the like, and good effects are obtained. However, the method is still based on centralized control, namely, information of a source side, a network side, a load side and the like is sent to a cloud center of a dispatching side, and a decision scheme is issued to each region after centralized optimization calculation of the cloud center. When the scale of the power system is increased, the problems of data delay and packet drop caused by communication blockage affect the real-time performance of tasks such as power grid running state evaluation, protection and the like, and the safety of the power system is endangered.
Therefore, a cloud edge collaborative artificial intelligent regulation and control method for the electric power system is needed.
Disclosure of Invention
The invention provides a cloud edge cooperative artificial intelligent regulation method, a system, a medium and equipment for an electric power system, which are used for solving the problem of how to try on intelligent regulation of the electric power system.
In order to solve the above problems, according to an aspect of the present invention, there is provided a cloud-edge collaborative artificial intelligence regulation method for an electric power system, the method comprising:
Determining a predicted value corresponding to each sub-region based on observed data of all sub-regions in the power system;
determining an execution action corresponding to each sub-region based on the predicted value;
and executing the joint action based on the execution action corresponding to each sub-area so as to regulate and control the power system.
Preferably, wherein the method further comprises:
extracting geographic position data based on the observed data of each node in the power system;
determining Euclidean distance between any two nodes based on geographic position data of each node, constructing an adjacency matrix W between any two nodes based on the Euclidean distance, and determining a degree matrix D and a first Laplace matrix L based on the adjacency matrix W;
determining a second Laplace matrix D based on the degree matrix D and the first Laplace matrix L -1/2 LD -1/2 Calculating feature vectors F corresponding to k minimum feature values of the second Laplace matrix respectively, normalizing the feature vectors F to obtain an n multiplied by k feature matrix F, clustering each row of data of the F as one sample by adopting a k-means algorithm to conduct sub-division so as to determine at least one sub-region.
Preferably, the determining the euclidean distance between any two nodes based on the geographic position data of each node, constructing an adjacency matrix W between any two nodes based on the euclidean distance, and determining the degree matrix D and the first laplace matrix L based on the adjacency matrix W, including:
L＝D-W，
wherein ,
Figure SMS_1
Figure SMS_2
Figure SMS_3
Figure SMS_4
wherein ,xi Geographic location data for an i-th node; x is x j Geographic location data for a j-th node; omega ij Is an element of the adjacency matrix W, i.e., adjacency matrix values between the i-th node and the j-th node; d, d ii An element of a degree matrix D; KNN (x) i )，KNN(x j ) Respectively x i and xj The nearest k points; sigma is a coefficient controlling the width of the neighborhood; exp (·) is an exponential operation of the natural constant e; n is the number of nodes.
Preferably, the determining the predicted value corresponding to each sub-region based on the observed data of all sub-regions in the power system includes:
for any one sub-region, observe data O t Inputting the predicted value into the current depth Q network model to output the predicted value;
the depth Q network model is provided with two neural network fitting target Q value vectors, wherein the target Q value vectors are respectively a target Q network and an evaluation Q network, the two networks have the same structure, 4 layers of full-connection layers are adopted, the number of neurons in the first layer of the hidden layer is the size of a state space, the number of neurons in the other layers of the hidden layer is decreased in proportion, the activation function is Relu, and the number of neurons in the last layer of the output layer is the size of an actionable space;
q value Q of current t state output t The calculation formula is as follows:
Figure SMS_5
/>
wherein γ represents a discount coefficient;
Figure SMS_6
A neural network prediction model representing a state at time t+1; θ - Parameters representing the target Q network; θ is an evaluation Q network parameter; a' represents the action of selecting the target Q network; argmax represents selecting the index with the largest value in the Q value vector, and obtaining the action corresponding to the index; o (O) t+1 Observation data at time t+1; r is (r) t+1 The prize value at time t+1 is indicated.
Preferably, the determining the execution action corresponding to each sub-region based on the predicted value includes:
and selecting an action corresponding to a larger value in the predicted values as an execution action corresponding to any one of the subareas.
Preferably, wherein the method further comprises:
calculating a prize value when performing the joint action;
and calculating a loss value based on the reward value and the predicted value corresponding to each sub-region, and updating model parameters of the depth Q network model by adopting a gradient descent algorithm.
Preferably, wherein the calculating the prize value when performing the joint action includes:
r t ＝r line,t +r gen,t ，
Figure SMS_7
wherein ,rt At time tA prize value; r is (r) line,t The load rate cost of the power transmission line; r is (r) gen,t The action cost of the controlled unit is; p is p l The load rate of the transmission line is the ratio of the current transmission power to the maximum transmission power allowed by the line for a long time; n (N) l Is the total number of transmission lines;
Figure SMS_8
penalty coefficients for overload and heavy-load lines, respectively; Δp g Is the active regulation of the controlled generator;
Figure SMS_9
the secondary and primary coefficients corresponding to the action cost function of the generator; n (N) g Is the adjustable group number of the system.
According to another aspect of the present invention, there is provided a cloud edge collaborative artificial intelligent regulation system for an electric power system, the system comprising:
the prediction value determining unit is used for determining a prediction value corresponding to each sub-region based on the observation data of all the sub-regions in the power system;
an execution action determining unit, configured to determine an execution action corresponding to each sub-area based on the predicted value;
and the regulation and control unit is used for executing the joint action based on the execution action corresponding to each sub-area so as to regulate and control the power system.
Preferably, wherein the system further comprises: a region dividing unit for:
extracting geographic position data based on the observed data of each node in the power system;
determining Euclidean distance between any two nodes based on geographic position data of each node, constructing an adjacency matrix W between any two nodes based on the Euclidean distance, and determining a degree matrix D and a first Laplace matrix L based on the adjacency matrix W;
Determining a second Laplace matrix D based on the degree matrix D and the first Laplace matrix L -1/2 LD -1/2 Calculating feature vectors f corresponding to the k minimum feature values of the second Laplace matrix respectively, and characterizing the feature vectorsAnd (3) normalizing the vector F to obtain an n multiplied by k feature matrix F, taking each line of data of the F as a sample, and carrying out sub-graph division by adopting k-means algorithm clustering to determine at least one sub-region.
Preferably, the area dividing unit determines a euclidean distance between any two nodes based on geographical position data of each node, constructs an adjacency matrix W between any two nodes based on the euclidean distance, and determines a degree matrix D and a first laplace matrix L based on the adjacency matrix W, and includes:
L＝D-W，
wherein ,
Figure SMS_10
Figure SMS_11
Figure SMS_12
Figure SMS_13
wherein ,xi Geographic location data for an i-th node; x is x j Geographic location data for a j-th node; omega ij Is an element of the adjacency matrix W, i.e., adjacency matrix values between the i-th node and the j-th node; d, d ii An element of a degree matrix D; KNN (x) i )，KNN(x j ) Respectively x i and xj The nearest k points; sigma is a coefficient controlling the width of the neighborhood; exp (·) is an exponential operation of the natural constant e; n is the number of nodes.
Preferably, the prediction value determining unit determines a prediction value corresponding to each sub-region based on observation data of all sub-regions in the power system, including:
For any one sub-region, observe data O t Input to the presentA deep Q network model to output a predicted value;
the depth Q network model is provided with two neural network fitting target Q value vectors, wherein the target Q value vectors are respectively a target Q network and an evaluation Q network, the two networks have the same structure, 4 layers of full-connection layers are adopted, the number of neurons in the first layer of the hidden layer is the size of a state space, the number of neurons in the other layers of the hidden layer is decreased in proportion, the activation function is Relu, and the number of neurons in the last layer of the output layer is the size of an actionable space;
q value Q of current t state output t The calculation formula is as follows:
Figure SMS_14
wherein γ represents a discount coefficient;
Figure SMS_15
a neural network prediction model representing a state at time t+1; θ - Parameters representing the target Q network; θ is an evaluation Q network parameter; a' represents the action of selecting the target Q network; argmax represents selecting the index with the largest value in the Q value vector, and obtaining the action corresponding to the index; o (O) t+1 Observation data at time t+1; r is (r) t+1 The prize value at time t+1 is indicated.
Preferably, the execution action determining unit is configured to determine, based on the predicted value, an execution action corresponding to each sub-region, and includes:
and selecting an action corresponding to a larger value in the predicted values as an execution action corresponding to any one of the subareas.
Preferably, wherein the system further comprises: an updating unit configured to:
calculating a prize value when performing the joint action;
and calculating a loss value based on the reward value and the predicted value corresponding to each sub-region, and updating model parameters of the depth Q network model by adopting a gradient descent algorithm.
Preferably, wherein the updating unit calculates the prize value when performing the joint action, includes:
r t ＝r line,t +r gen,t ，
Figure SMS_16
wherein ,rt A reward value at time t; r is (r) line,t The load rate cost of the power transmission line; r is (r) gen,t The action cost of the controlled unit is; p is p l The load rate of the transmission line is the ratio of the current transmission power to the maximum transmission power allowed by the line for a long time; n (N) l Is the total number of transmission lines;
Figure SMS_17
penalty coefficients for overload and heavy-load lines, respectively; Δp g Is the active regulation of the controlled generator;
Figure SMS_18
the secondary and primary coefficients corresponding to the action cost function of the generator; n (N) g Is the adjustable group number of the system.
Based on another aspect of the present invention, the present invention provides a computer readable storage medium having stored thereon a computer program which when executed by a processor implements the steps of any one of the cloud-edge collaborative artificial intelligence regulation methods of an electric power system.
Based on another aspect of the present invention, the present invention provides an electronic device, including:
The computer readable storage medium as described above; and
one or more processors configured to execute the programs in the computer-readable storage medium.
The invention provides a cloud edge collaborative artificial intelligent regulation and control method, a system, a medium and equipment for an electric power system, comprising the following steps: determining a predicted value corresponding to each sub-region based on observed data of all sub-regions in the power system; determining an execution action corresponding to each sub-region based on the predicted value; and executing the joint action based on the execution action corresponding to each sub-area so as to regulate and control the power system. According to the invention, operation is performed in each sub-area by using the corresponding edge platform, wide-area coordinated operation control of the power system is realized based on the cloud cooperation technology, effective decoupling and task sinking of the original centralized task can be realized, and a foundation is provided for maintaining safe and stable operation of a large-scale interconnected power grid under high-proportion new energy access.
Drawings
Exemplary embodiments of the present invention may be more completely understood in consideration of the following drawings:
FIG. 1 is a flow chart of a cloud edge collaborative artificial intelligence regulation method of an electric power system according to an embodiment of the invention;
FIG. 2 is a flow chart of implementing cloud-edge system regulation according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a power grid topology according to an embodiment of the present invention;
fig. 4 is a schematic diagram of cloud-edge collaboration of a power system according to an embodiment of the present invention;
FIG. 5 is a flow chart of cloud-edge cooperative regulation of a power system according to an embodiment of the invention;
fig. 6 is a schematic structural diagram of a cloud-edge collaborative artificial intelligent regulation system of an electric power system according to an embodiment of the present invention.
Detailed Description
The exemplary embodiments of the present invention will now be described with reference to the accompanying drawings, however, the present invention may be embodied in many different forms and is not limited to the examples described herein, which are provided to fully and completely disclose the present invention and fully convey the scope of the invention to those skilled in the art. The terminology used in the exemplary embodiments illustrated in the accompanying drawings is not intended to be limiting of the invention. In the drawings, like elements/components are referred to by like reference numerals.
Unless otherwise indicated, terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art. In addition, it will be understood that terms defined in commonly used dictionaries should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense.
Fig. 1 is a flowchart of a cloud-edge collaborative artificial intelligent regulation method 100 for an electric power system according to an embodiment of the invention. As shown in fig. 1, according to the cloud edge collaborative artificial intelligent regulation and control method for the electric power system, operation is performed on each sub-area by using a corresponding edge platform, wide-area coordinated operation control of the electric power system is realized based on a cloud collaborative technology, effective decoupling and task sinking of an original centralized task can be realized, and a foundation is provided for maintaining safe and stable operation of a large-scale interconnected power grid under high-proportion new energy access. According to the cloud edge collaborative artificial intelligent regulation and control method 100 for the power system, starting from the step 101, in the step 101, a predicted value corresponding to each sub-area is determined based on observed data of all sub-areas in the power system.
Preferably, wherein the method further comprises:
extracting geographic position data based on the observed data of each node in the power system;
determining Euclidean distance between any two nodes based on geographic position data of each node, constructing an adjacency matrix W between any two nodes based on the Euclidean distance, and determining a degree matrix D and a first Laplace matrix L based on the adjacency matrix W;
Determining a second Laplace matrix D based on the degree matrix D and the first Laplace matrix L -1/2 LD -1/2 Calculating feature vectors F corresponding to k minimum feature values of the second Laplace matrix respectively, normalizing the feature vectors F to obtain an n multiplied by k feature matrix F, clustering each row of data of the F as one sample by adopting a k-means algorithm to conduct sub-division so as to determine at least one sub-region.
Preferably, the determining the euclidean distance between any two nodes based on the geographic position data of each node, constructing an adjacency matrix W between any two nodes based on the euclidean distance, and determining the degree matrix D and the first laplace matrix L based on the adjacency matrix W, including:
L＝D-W，
wherein ,
Figure SMS_19
Figure SMS_20
Figure SMS_21
Figure SMS_22
wherein ,xi Geographic location data for an i-th node; x is x j Geographic location data for a j-th node; omega ij Is an element of the adjacency matrix W, i.e., adjacency matrix values between the i-th node and the j-th node; d, d ii An element of a degree matrix D; KNN (x) i )，KNN(x j ) Respectively x i and xj The nearest k points; sigma is a coefficient controlling the width of the neighborhood; exp (·) is an exponential operation of the natural constant e; n is the number of nodes.
Preferably, the determining the predicted value corresponding to each sub-region based on the observed data of all sub-regions in the power system includes:
For any one sub-region, observe data O t Inputting the predicted value into the current depth Q network model to output the predicted value;
the depth Q network model is provided with two neural network fitting target Q value vectors, wherein the target Q value vectors are respectively a target Q network and an evaluation Q network, the two networks have the same structure, 4 layers of full-connection layers are adopted, the number of neurons in the first layer of the hidden layer is the size of a state space, the number of neurons in the other layers of the hidden layer is decreased in proportion, the activation function is Relu, and the number of neurons in the last layer of the output layer is the size of an actionable space;
q value Q of current t state output t The calculation formula is as follows:
Figure SMS_23
wherein γ represents a discount coefficient;
Figure SMS_24
a neural network prediction model representing a state at time t+1; θ - Parameters representing the target Q network; θ is an evaluation Q network parameter; a' represents the action of selecting the target Q network; argmax represents selecting the index with the largest value in the Q value vector, and obtaining the action corresponding to the index; o (O) t+1 Observation data at time t+1; r is (r) t+1 The prize value at time t+1 is indicated.
In the invention, as shown in fig. 2, an electric power system node model is introduced to construct observation data, and a spectral clustering algorithm is adopted to obtain a region division result of high cohesion of node characteristics in each region of the system and low coupling of node characteristics between regions; then, according to the obtained partitioning result, deploying an edge computing platform in each region to interact with the cloud computing center; then, based on each edge computing platform, a DRL model is deployed; finally, the following: and core information and data sharing are transmitted through the cloud center and the edge platform. And according to the global observation data, each edge agent predicts the output action, updates the model parameters and realizes cloud edge cooperative regulation.
In the present invention, u is used t+l The method comprises the steps of representing the first group of observation data in the t group of observation data, namely the observation data at the t+l moment in a multi-node model of the power system; u (u) t+l ＝{data t+l,v,j ∣1≤v≤V,1≤j≤J},l∈[0,L-1]，data t+l,v,j The characteristic data of the jth type of the jth busbar node representing the ith step of observation data in the tth group of observation data, namely the characteristic data of the jth type of the jth power system node at the (t+l) th moment in the power system multi-node model, comprises the quantity of power transmission lines, the load factor, the overload and heavy load penalty coefficient of the node, the quantity of adjustable units, the active adjustment quantity, the action cost function and the like, V represents the quantity of the power system nodes, and J is the quantity of node characteristics.
Extracting geographical position data of each node from the observation data, calculating Euclidean distance between each node as electrical distance, and constructing an adjacent matrix W through a KNN algorithm, wherein the formula is as follows:
Figure SMS_25
Figure SMS_26
wherein ,xi For the position of the ith node, ω ij As elements of the adjacency matrix W, KNN (x i )，KNN(x j ) Respectively x i and xj The nearest k points; sigma is a coefficient controlling the width of the neighborhood; n is the number of nodes; exp (·) is an exponential operation of the natural constant e.
After obtaining the adjacency matrix, calculating a degree matrix D and a Laplace matrix L, wherein the method comprises the following steps:
Figure SMS_27
Figure SMS_28
L＝D-W，
wherein ,dii An element of a degree matrix D; n is the number of nodes.
In the present invention, a cut map is defined
Figure SMS_29
A k For the set of kth subgraphs, W is the weight between subgraphs. The point weights and high in the subgraphs, and the point weights and low cut graphs among the subgraphs are the optimal cut graphs.
Finally, a standardized Laplace matrix D is constructed based on D and L -1/2 LD -1/2 And calculating the feature vector F corresponding to each of the k minimum feature values of the standardized Laplace matrix, and normalizing the feature vector to obtain an n multiplied by k feature matrix F. Let F each line of data be a sample, and then collectAnd clustering by using a k-means algorithm to obtain a final sub-graph division result so as to finish the regional division of the power system.
In the invention, after k power system subareas are obtained by a spectral clustering algorithm, k regional dispatching centers are arranged in the subareas, and an edge computing platform with the capabilities of computing, storing, applying and the like is deployed. And the total dispatching center cloud platform performs distributed division on dispatching center tasks and sinks part of tasks to the edge computing platform. The power system has a "source-net-load" hierarchy. After each regional edge computing platform obtains the executed task, the edge side data is acquired. The source side data can be acquired from PMUs at the outlet ends of wind power plants, hydropower plants and the like; the network side data and the load side data are collected by an intelligent terminal and a merging unit in the power grid. The message is uploaded to the comprehensive automatic system of the transformer substation in a message mode through transmission modes such as optical fibers and wireless modes, the message is sent to an edge computing platform through Ethernet communication, and the edge computing platform is used for analyzing and storing the message.
Specifically, the cloud computing center collects observation state information O of all sub-areas t ＝[o 1,t ,o 2,t ,…,o i,t ,...,o n,t], wherein oi,t The local observation information of the intelligent agent in the ith area at the moment t is represented, and n is the number of subareas. Then, the cloud computing center collects global observation information O t Edge computing platforms issued to each sub-region, and intelligent agent input observables O of each edge computing platform t Outputting a predicted Q value through a deep Q network model, and selecting an action a corresponding to the maximum Q value i,t 。
Wherein, the observed data is specifically defined as:
O t ＝[o 1,t ,o 2,t ,…,o i,t ]，
o i,t ＝[u t ,u t+1 ,…,u t+L-1 ] T ，
wherein ,Ot Observation data at time t, t indicating the start time of observation data of the t th group, o i,t The t group observation data of the i region is L is a positive integer, and the window length of the observation data is。u t+l And the observation data of the first group in the observation data of the t group, namely the observation data of the t+l moment in the multi-node model of the power system, are shown.
In the invention, the intelligent agent adopts a deep Q network model (DQN) and is provided with two neural networks for fitting a target Q value, namely a target Q network and an evaluation Q network. The two networks have the same structure, 4 layers of full-connection layers are adopted, the number of neurons of the first layer of the hidden layer is the size of a state space, then the number is reduced proportionally, and the Relu is used as an activation function. The number of neurons of the last layer of output layer is the size of the actionable space, namely the output predicted value Q.
Q value Q of current t state output t The calculation formula is as follows:
Figure SMS_30
wherein γ represents a discount coefficient;
Figure SMS_31
a neural network prediction model representing a state at time t+1; θ - Parameters representing the target Q network; θ is an evaluation Q network parameter; a' represents the action of selecting the target Q network; argmax represents selecting the index with the largest value in the Q value vector, and obtaining the action corresponding to the index; o (O) t+1 Observation data at time t+1; r is (r) t+1 The prize value at time t+1 is indicated.
In step 102, an execution action corresponding to each sub-region is determined based on the predicted value.
In the present invention, for each sub-region, the edge computing platform determines to execute action a based on the larger value of the obtained predicted values i,t 。
In step 103, the joint action is executed based on the execution action corresponding to each sub-area, so as to perform power system regulation.
Preferably, the determining the execution action corresponding to each sub-region based on the predicted value includes:
and selecting an action corresponding to a larger value in the predicted values as an execution action corresponding to any one of the subareas.
In the present invention, each edge computing platform performs a joint action A t ＝[a 1,t ,…,a i,t ]Updating the observed state information of the system environment to O t+1 ＝[o 1,t+1 ,o 2,t+1 ,…,o i,t+1 ]Each edge computing platform stores observation information and local action tracks
Figure SMS_32
The edge computing platform can observe state information O according to the new system environment t+1 Each edge agent outputs a local action a i,t+1 Realizing regulation and control.
Preferably, wherein the method further comprises:
calculating a prize value when performing the joint action;
and calculating a loss value based on the reward value and the predicted value corresponding to each sub-region, and updating model parameters of the depth Q network model by adopting a gradient descent algorithm.
Preferably, wherein the calculating the prize value when performing the joint action includes:
r t ＝r line,t +r gen,t ，
Figure SMS_33
wherein ,rt A reward value at time t; r is (r) line,t The load rate cost of the power transmission line; r is (r) gen,t The action cost of the controlled unit is; p is p l The load rate of the transmission line is the ratio of the current transmission power to the maximum transmission power allowed by the line for a long time; n (N) l Is the total number of transmission lines;
Figure SMS_34
penalty coefficients for overload and heavy-load lines, respectively; Δp g Is the active regulation of the controlled generator; />
Figure SMS_35
Is the correspondence of the action cost function of the generatorSecondary and primary coefficients of (2); n (N) g Is the adjustable group number of the system.
In the present invention, the prize value r can also be obtained after the joint action is performed t The cloud computing center stores global states and action rewards tracks tau t C (O t ,A t ,r t ) In the playback buffer, the storage structure is a SumTree structure. Then, the cloud computing center randomly extracts the batch_size group data from the playback buffer, sends the batch_size group data to the edge computing platform, and each agent (one edge platform) calculates through the historical observation data
Figure SMS_36
According to historical prize value r t And calculating a loss value with the currently output predicted value, and updating network parameters by adopting a gradient descent method.
According to the invention, multiple agents are introduced into each edge computing platform, and the reinforcement learning process of each agent can be modeled into a Markov Decision Process (MDP), and the description is as follows:
a set of finite environmental states S, an action space A, a real value rewarding function R and a state transition probability
Figure SMS_37
State s t vS reflects all indexes of the running state of the system at the moment t, including photovoltaic, wind power, output values of a conventional unit, load requirements of each node, voltage, power angle, power and the like.
Action a t The E A represents actions which can be executed by the agent at the time t, including rescheduling actions of the generator and power grid topology changing actions. The size of the action space increases exponentially with the increase of the controllable devices, and a filter is adopted to extract the effective action set of the regional intelligent agent.
The probability of state transition P is SxA to [0,1 ]]Expressed in state s t Take action a down t After becoming a state
Figure SMS_38
Is a probability of (2).
Reward function r t E R, tableShow the agent taking action a t Obtaining the rewarding value of environmental feedback, which is the load rate r of the power transmission line line,t And the controlled unit action cost r gen,t And (3) summing. the prize value at time t is set as follows:
Figure SMS_39
wherein ,pl The load rate of the transmission line is defined as the ratio of the current transmission power to the maximum transmission power allowed by the line for a long time; n (N) l Is the total number of transmission lines;
Figure SMS_40
penalty coefficients for overload and heavy-load lines, respectively; Δp g Is the active regulation of the controlled generator; />
Figure SMS_41
The secondary and primary coefficients corresponding to the action cost function of the generator; n (N) g Is the adjustable group number of the system. All parameters are included in the observed data.
The example is developed and introduced based on a Grid2op platform of 2021L2RPN international power Grid dispatching competition by adopting a 36-node power Grid provided by a second track, as shown in fig. 2. The power grid is subjected to partition modeling by adopting a spectral clustering algorithm, edge computing platforms are arranged in the partitioned subareas, multiple agents are deployed, and meanwhile, a cloud computing center and the edge computing platforms are arranged to realize data sharing under cloud edge cooperation, so that training of the multiple agents is completed and the power grid is applied to regulation and control of a power system.
The following specifically exemplifies embodiments of the present invention
The example is developed and introduced based on a Grid2op platform of 2021L2RPN international power Grid dispatching competition by adopting a 36-node power Grid provided by a second track, as shown in fig. 3. The power grid is subjected to partition modeling by adopting a spectral clustering algorithm, edge computing platforms are arranged in the partitioned subareas, multiple agents are deployed, and meanwhile, a cloud computing center and the edge computing platforms are arranged to realize data sharing under cloud edge cooperation, so that training of the multiple agents is completed and the power grid is applied to regulation and control of a power system.
The specific embodiment of the invention relates to a cloud edge cooperative artificial intelligent regulation and control method for an electric power system, which specifically comprises the following steps:
step 1: as shown in step 1 of fig. 2, a power system node model is introduced, and a spectral clustering algorithm is adopted to model node partitions. The method comprises the steps of introducing a power system node model to construct observation data, and modeling node partitions by adopting a spectral clustering algorithm to obtain a region division result of high cohesion of node features in each region of the system and low coupling of node features between regions.
Step 2: as shown in step 2 of fig. 2, an edge computing platform is deployed in each sub-region to interact with the cloud computing center.
As shown in conjunction with the cloud edge collaborative architecture of fig. 4, the step 2 specifically includes the following steps:
step 2.1: after k power system subareas are obtained by a spectral clustering algorithm, setting k regional dispatching centers in the subareas, and deploying an edge computing platform with computing, storage, application and other capabilities. Step 2.2: and the total dispatching center cloud platform performs distributed division on dispatching center tasks and sinks part of tasks to the edge computing platform. Step 2.3: the power system has a "source-net-load" hierarchy. After each regional edge computing platform obtains the executed task, the edge side data is acquired. The source side data can be acquired from PMUs at the outlet ends of wind power plants, hydropower plants and the like; the network side data and the load side data are collected by an intelligent terminal and a merging unit in the power grid. The message is uploaded to the comprehensive automatic system of the transformer substation in a message mode through transmission modes such as optical fibers and wireless modes, the message is sent to an edge computing platform through Ethernet communication, and the edge computing platform is used for analyzing and storing the message.
Step 3: as shown in step 3 of fig. 2, a DRL model is deployed based on each edge computing platform.
The step 3 specifically comprises the following steps:
step 3.1: introducing multiple agents at respective edge computing platforms, the reinforcement learning process of each agent can be modeled as a Markov Decision Process (MDP), described as follows:
a group of finite environmental state S, action space A and real stateValue reward function R, state transition probability
Figure SMS_42
/>
State s t The E S reflects all indexes of the running state of the system at the moment t, including photovoltaic, wind power, output values of a conventional unit, load requirements of all nodes, voltage, power angle, power and the like.
Action a t The E A represents actions which can be executed by the agent at the time t, including rescheduling actions of the generator and power grid topology changing actions. The size of the action space increases exponentially with the increase of the controllable devices, and a filter is adopted to extract the effective action set of the regional intelligent agent.
The probability of state transition P is SxA to [0,1 ]]Expressed in state s t Take action a down t After becoming a state
Figure SMS_43
Is a probability of (2).
Reward function r t E R, representing the action a taken by the agent t Obtaining the rewarding value of environmental feedback, which is the load rate r of the power transmission line line,t And the controlled unit action cost r gen,t And (3) summing. the prize value at time t is set as follows:
Figure SMS_44
wherein ,pl The load rate of the transmission line is defined as the ratio of the current transmission power to the maximum transmission power allowed by the line for a long time; n (N) l Is the total number of transmission lines;
Figure SMS_45
penalty coefficients for overload and heavy-load lines, respectively; Δp g Is the active regulation of the controlled generator; />
Figure SMS_46
The secondary and primary coefficients corresponding to the action cost function of the generator; n (N) g Is the adjustable group number of the system. All parameters are included in the stepsIn the observation data of step 1.
Step 3.2: the intelligent agent adopts a deep Q network model (DQN), has two neural networks for fitting a target Q value, and is respectively a target Q network and an evaluation Q network. The two networks have the same structure, 4 layers of full-connection layers are adopted, the number of neurons of the first layer of the hidden layer is the size of a state space, then the number is reduced proportionally, and the Relu is used as an activation function. The number of neurons of the last layer of output layer is 1, namely, the predicted value is output.
The formula for calculating the predicted value is as follows:
q value Q of current t state output t The calculation formula is as follows:
Figure SMS_47
wherein γ represents a discount coefficient;
Figure SMS_48
a neural network prediction model representing a state at time t+1; θ - Parameters representing the target Q network; θ is an evaluation Q network parameter; a' represents the action of selecting the target Q network; argmax represents selecting the index with the largest value in the Q value vector, and obtaining the action corresponding to the index; o (O) t+1 Observation data at time t+1; r is (r) t+1 The prize value at time t+1 is indicated.
Step 4: as shown in step 4 of fig. 2, cloud edge cooperative regulation and control is realized through information interaction between a cloud center and an edge platform. And according to the global observation data, each edge agent predicts the output action, updates the model parameters and realizes cloud edge cooperative regulation.
The flow of the regulated algorithm program is shown in fig. 5, and is specifically as follows:
step 4.1: the cloud computing center collects the observation state information O of all subareas t ＝[o 1,t ,o 2,t ,…,o i,t], wherein oi,t And the local observation information of the ith area intelligent agent at the t moment is shown.
Step 4.2: the cloud computing center transmits the collected global observation information to the edge computing level of each sub-regionTable, intelligent body input observed quantity O of each edge computing platform t Outputting a predicted Q value through a deep Q network model, and selecting an action a corresponding to the maximum Q value i,t 。
Step 4.3: each edge computing platform performs joint action A t ＝[a 1,t ,…,a i,t ]Updating the observed state information of the system environment to O t+1 ＝[o 1,t+1 ,o 2,t+1 ,…,o i,t+1 ]Obtain the prize value r t 。
Step 4.4: each edge computing platform stores observation information and local action tracks
Figure SMS_49
Cloud computing center stores global state and action rewarding track tau t C (O t ,A t ,r t ) In the playback buffer, the storage structure is a SumTree structure. Taking TD-error as the priority weight of each group of data, namely calculating the difference of the output Q values of two networks of the intelligent agent depth Q network model of the edge calculation platform.
Step 4.5: the cloud computing center randomly extracts the batch_size group data from the playback buffer, and as the data priority is defined, samples with higher priority are easier to extract. The extracted data are sent to an edge computing platform, and each agent calculates Q through historical observation data t ＝Φ(O,a i,t ；θ i E ). According to historical prize value r t And calculating a loss value with the current output Q value, and updating network parameters by adopting a gradient descent method.
Step 4.6: based on the observed state information O of the new system environment t+1 Each edge agent outputs a local action a i,t+1 Realizing regulation and control.
Fig. 6 is a schematic structural diagram of a cloud-edge collaborative artificial intelligent regulation system 600 of a power system according to an embodiment of the present invention. As shown in fig. 6, a cloud edge collaborative artificial intelligent regulation system 600 of a power system according to an embodiment of the present invention includes: a predicted value determining unit 601, an execution action determining unit 602, and a regulating unit 603.
Preferably, the predicted value determining unit 601 is configured to determine a predicted value corresponding to each sub-area based on observation data of all sub-areas in the power system.
Preferably, the execution action determining unit 602 is configured to determine, based on the predicted value, an execution action corresponding to each sub-area.
Preferably, the regulating unit 603 is configured to execute a joint action based on the execution action corresponding to each sub-area, so as to regulate the electric power system.
Preferably, wherein the system further comprises: a region dividing unit for:
extracting geographic position data based on the observed data of each node in the power system;
determining Euclidean distance between any two nodes based on geographic position data of each node, constructing an adjacency matrix W between any two nodes based on the Euclidean distance, and determining a degree matrix D and a first Laplace matrix L based on the adjacency matrix W;
determining a second Laplace matrix D based on the degree matrix D and the first Laplace matrix L -1/2 LD -1/2 Calculating feature vectors F corresponding to k minimum feature values of the second Laplace matrix respectively, normalizing the feature vectors F to obtain an n multiplied by k feature matrix F, clustering each row of data of the F as one sample by adopting a k-means algorithm to conduct sub-division so as to determine at least one sub-region.
Preferably, the area dividing unit determines a euclidean distance between any two nodes based on geographical position data of each node, constructs an adjacency matrix W between any two nodes based on the euclidean distance, and determines a degree matrix D and a first laplace matrix L based on the adjacency matrix W, and includes:
L＝D-W，
wherein ,
Figure SMS_50
/>
Figure SMS_51
Figure SMS_52
Figure SMS_53
wherein ,xi Geographic location data for an i-th node; x is x j Geographic location data for a j-th node; omega ij Is an element of the adjacency matrix W, i.e., adjacency matrix values between the i-th node and the j-th node; d, d ii An element of a degree matrix D; KNN (x) i )，KNN(x j ) Respectively x i and xj The nearest k points; sigma is a coefficient controlling the width of the neighborhood; exp (·) is an exponential operation of the natural constant e; n is the number of nodes.
Preferably, the predicted value determining unit 601 determines a predicted value corresponding to each sub-area based on observation data of all sub-areas in the power system, including:
for any one sub-region, observe data O t Inputting the predicted value into the current depth Q network model to output the predicted value;
the depth Q network model is provided with two neural network fitting target Q value vectors, wherein the target Q value vectors are respectively a target Q network and an evaluation Q network, the two networks have the same structure, 4 layers of full-connection layers are adopted, the number of neurons in the first layer of the hidden layer is the size of a state space, the number of neurons in the other layers of the hidden layer is decreased in proportion, the activation function is Relu, and the number of neurons in the last layer of the output layer is the size of an actionable space;
q value Q of current t state output t The calculation formula is as follows:
Figure SMS_54
wherein γ represents a discount coefficient;
Figure SMS_55
Representation ofA neural network prediction model of a t+1 moment state; θ - Parameters representing the target Q network; θ is an evaluation Q network parameter; a' represents the action of selecting the target Q network; argmax represents selecting the index with the largest value in the Q value vector, and obtaining the action corresponding to the index; o (O) t+1 Observation data at time t+1; r is (r) t+1 The prize value at time t+1 is indicated.
Preferably, the performing action determining unit 602 is configured to determine, based on the predicted value, a performing action corresponding to each sub-area, including:
and selecting an action corresponding to a larger value in the predicted values as an execution action corresponding to any one of the subareas.
Preferably, wherein the system further comprises: an updating unit configured to:
calculating a prize value when performing the joint action;
and calculating a loss value based on the reward value and the predicted value corresponding to each sub-region, and updating model parameters of the depth Q network model by adopting a gradient descent algorithm.
Preferably, wherein the updating unit calculates the prize value when performing the joint action, includes:
r t ＝r line,t +r gen,t ，
Figure SMS_56
wherein ,rt A reward value at time t; r is (r) line,t The load rate cost of the power transmission line; r is (r) gen,t The action cost of the controlled unit is; p is p l The load rate of the transmission line is the ratio of the current transmission power to the maximum transmission power allowed by the line for a long time; n (N) l Is the total number of transmission lines;
Figure SMS_57
penalty coefficients for overload and heavy-load lines, respectively; Δp g Is the active regulation of the controlled generator;
Figure SMS_58
the secondary and primary coefficients corresponding to the action cost function of the generator; n (N) g Is the adjustable group number of the system.
The cloud-edge collaborative artificial intelligent regulation system 600 of the electric power system according to the embodiment of the present invention corresponds to the cloud-edge collaborative artificial intelligent regulation method 100 of the electric power system according to another embodiment of the present invention, and is not described herein.
Based on another aspect of the present invention, the present invention provides a computer readable storage medium having stored thereon a computer program which when executed by a processor implements the steps of any one of the cloud-edge collaborative artificial intelligence regulation methods of an electric power system.
Based on another aspect of the present invention, the present invention provides an electronic device, including:
the computer readable storage medium as described above; and
one or more processors configured to execute the programs in the computer-readable storage medium.
The invention has been described with reference to a few embodiments. However, as is well known to those skilled in the art, other embodiments than the above disclosed invention are equally possible within the scope of the invention, as defined by the appended patent claims.
Generally, all terms used in the claims are to be interpreted according to their ordinary meaning in the technical field, unless explicitly defined otherwise therein. All references to "a/an/the [ means, component, etc. ]" are to be interpreted openly as referring to at least one instance of said means, component, etc., unless explicitly stated otherwise. The steps of any method disclosed herein do not have to be performed in the exact order disclosed, unless explicitly stated.
It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention 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 invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. 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.
Finally, it should be noted that: the above embodiments are only for illustrating the technical aspects of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the above embodiments, it should be understood by those of ordinary skill in the art that: modifications and equivalents may be made to the specific embodiments of the invention without departing from the spirit and scope of the invention, which is intended to be covered by the claims.

Claims (16)

1. The cloud edge collaborative artificial intelligent regulation and control method for the electric power system is characterized by comprising the following steps of:
determining a predicted value corresponding to each sub-region based on observed data of all sub-regions in the power system;
determining an execution action corresponding to each sub-region based on the predicted value;
and executing the joint action based on the execution action corresponding to each sub-area so as to regulate and control the power system.
2. The method according to claim 1, wherein the method further comprises:
extracting geographic position data based on the observed data of each node in the power system;
determining Euclidean distance between any two nodes based on geographic position data of each node, constructing an adjacency matrix W between any two nodes based on the Euclidean distance, and determining a degree matrix D and a first Laplace matrix L based on the adjacency matrix W;
determining a second Laplace matrix D based on the degree matrix D and the first Laplace matrix L -1/2 LD -1/2 Calculating feature vectors F corresponding to k minimum feature values of the second Laplace matrix respectively, normalizing the feature vectors F to obtain an n multiplied by k feature matrix F, clustering each row of data of the F as one sample by adopting a k-means algorithm to conduct sub-division so as to determine at least one sub-region.
3. The method according to claim 2, wherein determining the euclidean distance between any two nodes based on the geographical location data of each node, constructing an adjacency matrix W between any two nodes based on the euclidean distance, and determining the degree matrix D and the first laplace matrix L based on the adjacency matrix W, comprises:
L＝D-W，
wherein ,
Figure FDA0003836402600000011
Figure FDA0003836402600000021
Figure FDA0003836402600000022
Figure FDA0003836402600000023
wherein ,xi Geographic location data for an i-th node; x is x j Geographic location data for a j-th node; omega ij Is an element of the adjacency matrix W, i.e., adjacency matrix values between the i-th node and the j-th node; d, d ii An element of a degree matrix D; KNN (x) i )，KNN(x j ) Respectively x i and xj The nearest k points; sigma is a coefficient controlling the width of the neighborhood; exp (·) is an exponential operation of the natural constant e; n is the number of nodes.
4. The method of claim 1, wherein determining the predicted value for each sub-region based on the observed data for all sub-regions in the power system comprises:
for any one sub-region, observe data O t Inputting the predicted value into the current depth Q network model to output the predicted value;
the depth Q network model is provided with two neural network fitting target Q value vectors, wherein the target Q value vectors are respectively a target Q network and an evaluation Q network, the two networks have the same structure, 4 layers of full-connection layers are adopted, the number of neurons in the first layer of the hidden layer is the size of a state space, the number of neurons in the other layers of the hidden layer is decreased in proportion, the activation function is Relu, and the number of neurons in the last layer of the output layer is the size of an actionable space;
Q value Q of current t state output t The calculation formula is as follows:
Figure FDA0003836402600000024
wherein γ represents a discount coefficient;
Figure FDA0003836402600000025
a neural network prediction model representing a state at time t+1; θ - Parameters representing the target Q network; θ is an evaluation Q network parameter; a' represents the action of selecting the target Q network; argmax represents selecting the index with the largest value in the Q value vector, and obtaining the action corresponding to the index; o (O) t+1 Observation data at time t+1; r is (r) t+1 The prize value at time t+1 is indicated.
5. The method of claim 1, wherein the determining, based on the predicted value, the corresponding execution action for each sub-region comprises:
and selecting an action corresponding to a larger value in the predicted values as an execution action corresponding to any one of the subareas.
6. The method according to claim 1, wherein the method further comprises:
calculating a prize value when performing the joint action;
and calculating a loss value based on the reward value and the predicted value corresponding to each sub-region, and updating model parameters of the depth Q network model by adopting a gradient descent algorithm.
7. The method of claim 6, wherein calculating the prize value when performing the joint action comprises:
r t ＝r line,t +r gen,t ，
Figure FDA0003836402600000031
wherein ,rt A reward value at time t; r is (r) line,t The load rate cost of the power transmission line; r is (r) gen,t The action cost of the controlled unit is; p is p l The load rate of the transmission line is the ratio of the current transmission power to the maximum transmission power allowed by the line for a long time; n (N) l Is the total number of transmission lines;
Figure FDA0003836402600000032
penalty coefficients for overload and heavy-load lines, respectively; Δp g Is the active regulation of the controlled generator;
Figure FDA0003836402600000033
the secondary and primary coefficients corresponding to the action cost function of the generator; n (N) g Is the adjustable group number of the system.
8. An electric power system cloud edge collaborative artificial intelligence regulation and control system, which is characterized in that the system comprises:
the prediction value determining unit is used for determining a prediction value corresponding to each sub-region based on the observation data of all the sub-regions in the power system;
an execution action determining unit, configured to determine an execution action corresponding to each sub-area based on the predicted value;
and the regulation and control unit is used for executing the joint action based on the execution action corresponding to each sub-area so as to regulate and control the power system.
9. The system of claim 8, wherein the system further comprises: a region dividing unit for:
extracting geographic position data based on the observed data of each node in the power system;
Determining Euclidean distance between any two nodes based on geographic position data of each node, constructing an adjacency matrix W between any two nodes based on the Euclidean distance, and determining a degree matrix D and a first Laplace matrix L based on the adjacency matrix W;
determining a second Laplace matrix D based on the degree matrix D and the first Laplace matrix L -12 LD -12 Calculating feature vectors F corresponding to k minimum feature values of the second Laplace matrix respectively, normalizing the feature vectors F to obtain an n multiplied by k feature matrix F, clustering each row of data of the F as one sample by adopting a k-means algorithm to conduct sub-division so as to determine at least one sub-region.
10. The system according to claim 9, wherein the area dividing unit that determines a euclidean distance between any two nodes based on the geographical position data of each node, constructs an adjacency matrix W between any two nodes based on the euclidean distance, and determines a degree matrix D and a first laplace matrix L based on the adjacency matrix W, comprises:
L＝D-W，
wherein ,
Figure FDA0003836402600000041
Figure FDA0003836402600000042
Figure FDA0003836402600000043
Figure FDA0003836402600000051
wherein ,xi Geographic location data for an i-th node; x is x j Geographic location data for a j-th node; omega ij Is the element of the adjacency matrix W, i.e. the ith node and Adjacency matrix values between the j-th nodes; d, d ii An element of a degree matrix D; KNN (x) i )，KNN(x j ) Respectively x i and xj The nearest k points; sigma is a coefficient controlling the width of the neighborhood; exp (·) is an exponential operation of the natural constant e; n is the number of nodes.
11. The system according to claim 8, wherein the predicted value determining unit determines the predicted value corresponding to each sub-region based on the observed data of all sub-regions in the power system, comprising:
for any one sub-region, observe data O t Inputting the predicted value into the current depth Q network model to output the predicted value;
the depth Q network model is provided with two neural network fitting target Q value vectors, wherein the target Q value vectors are respectively a target Q network and an evaluation Q network, the two networks have the same structure, 4 layers of full-connection layers are adopted, the number of neurons in the first layer of the hidden layer is the size of a state space, the number of neurons in the other layers of the hidden layer is decreased in proportion, the activation function is Relu, and the number of neurons in the last layer of the output layer is the size of an actionable space;
q value Q of current t state output t The calculation formula is as follows:
Figure FDA0003836402600000052
wherein γ represents a discount coefficient;
Figure FDA0003836402600000053
a neural network prediction model representing a state at time t+1; θ - Parameters representing the target Q network; θ is an evaluation Q network parameter; a' represents the action of selecting the target Q network; argmax represents selecting the index with the largest value in the Q value vector, and obtaining the action corresponding to the index; o (O) t+1 Observation data at time t+1; r is (r) t+1 The prize value at time t+1 is indicated.
12. The system according to claim 8, wherein the execution action determining unit configured to determine, based on the predicted value, an execution action corresponding to each sub-region includes:
and selecting an action corresponding to a larger value in the predicted values as an execution action corresponding to any one of the subareas.
13. The system of claim 8, wherein the system further comprises: an updating unit configured to:
calculating a prize value when performing the joint action;
and calculating a loss value based on the reward value and the predicted value corresponding to each sub-region, and updating model parameters of the depth Q network model by adopting a gradient descent algorithm.
14. The system of claim 13, wherein the updating unit, when performing the joint action, calculates the prize value, comprises:
r t ＝r line,t +r gen,t ，
Figure FDA0003836402600000061
wherein ,rt A reward value at time t; r is (r) line,t The load rate cost of the power transmission line; r is (r) gen,t The action cost of the controlled unit is; p is p l The load rate of the transmission line is the ratio of the current transmission power to the maximum transmission power allowed by the line for a long time; n (N) l Is the total number of transmission lines;
Figure FDA0003836402600000062
penalty coefficients for overload and heavy-load lines, respectively; Δp g Is the active regulation of the controlled generator;
Figure FDA0003836402600000063
the secondary and primary coefficients corresponding to the action cost function of the generator; n (N) g Is the adjustable group number of the system.
15. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the steps of the method according to any of claims 1-7.
16. An electronic device, comprising:
the computer readable storage medium recited in claim 15; and
one or more processors configured to execute the programs in the computer-readable storage medium.
CN202211089297.4A 2022-09-07 2022-09-07 Cloud edge cooperative artificial intelligent regulation and control method, system, medium and equipment for electric power system Pending CN116169776A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211089297.4A CN116169776A (en) 2022-09-07 2022-09-07 Cloud edge cooperative artificial intelligent regulation and control method, system, medium and equipment for electric power system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211089297.4A CN116169776A (en) 2022-09-07 2022-09-07 Cloud edge cooperative artificial intelligent regulation and control method, system, medium and equipment for electric power system

Publications (1)

Publication Number Publication Date
CN116169776A true CN116169776A (en) 2023-05-26

Family

ID=86418873

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211089297.4A Pending CN116169776A (en) 2022-09-07 2022-09-07 Cloud edge cooperative artificial intelligent regulation and control method, system, medium and equipment for electric power system

Country Status (1)

Country Link
CN (1) CN116169776A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116896512A (en) * 2023-09-08 2023-10-17 之江实验室 Cloud edge cooperative system evaluation method and device, storage medium and electronic equipment
CN117390536A (en) * 2023-12-11 2024-01-12 深圳市宝腾互联科技有限公司 Operation and maintenance management method and system based on artificial intelligence

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116896512A (en) * 2023-09-08 2023-10-17 之江实验室 Cloud edge cooperative system evaluation method and device, storage medium and electronic equipment
CN116896512B (en) * 2023-09-08 2024-01-09 之江实验室 Cloud edge cooperative system evaluation method and device, storage medium and electronic equipment
CN117390536A (en) * 2023-12-11 2024-01-12 深圳市宝腾互联科技有限公司 Operation and maintenance management method and system based on artificial intelligence
CN117390536B (en) * 2023-12-11 2024-04-02 深圳市宝腾互联科技有限公司 Operation and maintenance management method and system based on artificial intelligence

Similar Documents

Publication Publication Date Title
CN112615379B (en) Power grid multi-section power control method based on distributed multi-agent reinforcement learning
CN116169776A (en) Cloud edge cooperative artificial intelligent regulation and control method, system, medium and equipment for electric power system
Nie et al. Optimizing the post-disaster control of islanded microgrid: A multi-agent deep reinforcement learning approach
Mandal et al. Daily combined economic emission scheduling of hydrothermal systems with cascaded reservoirs using self organizing hierarchical particle swarm optimization technique
CN107045655A (en) Wolf pack clan strategy process based on the random consistent game of multiple agent and virtual generating clan
Hardiansyah et al. Solving economic load dispatch problem using particle swarm optimization technique
CN111414681B (en) Building evacuation simulation method and system based on shared deep reinforcement learning
CN105631528A (en) NSGA-II and approximate dynamic programming-based multi-objective dynamic optimal power flow solving method
Moghdani et al. An improved volleyball premier league algorithm based on sine cosine algorithm for global optimization problem
Zhou et al. Action set based policy optimization for safe power grid management
Al Zishan et al. Adaptive control of plug-in electric vehicle charging with reinforcement learning
CN115085202A (en) Power grid multi-region intelligent power collaborative optimization method, device, equipment and medium
Tang et al. Deep neural network-based hierarchical learning method for dispatch control of multi-regional power grid
Zhang et al. Deep reinforcement learning for load shedding against short-term voltage instability in large power systems
Senthil et al. Improved tabu search algorithm to economic emission dispatch with transmission line constraint
Serré et al. Reinforcement learning for Energies of the future and carbon neutrality: a Challenge Design
CN115689375A (en) Virtual power plant operation control method, device, equipment and medium
Hu et al. Energy management for microgrids using a reinforcement learning algorithm
Wang et al. Decentralized optimization algorithms for variable speed pumps operation based on local interaction game
Singh et al. Economic load dispatch using MRPSO with generator ramp rate limits constraint
Li et al. Multiagent deep meta reinforcement learning for sea computing-based energy management of interconnected grids considering renewable energy sources in sustainable cities
Lenin Integrated Algorithm for Decreasing Active Power Loss
Vlachos Rank-based ant colony algorithm for a thermal generator maintenance scheduling problem
Wooding et al. Control and management of active buildings
van der Sar et al. Multi-Agent Reinforcement Learning for Power Grid Topology Optimization

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