A distributed power supply intelligent power compensation method, system and device

Abstract

The application belongs to the technical field of power systems, and particularly relates to a distributed power supply intelligent power compensation method, system and equipment. The method converts the extracted power grid environment features into a power grid weighted graph based on a network embedding strategy, and converts the obtained power compensation request of the distributed power supply into a power compensation request weighted graph. The power grid weighted graph and the power compensation request weighted graph are used to train a strategy network through a deep reinforcement learning method to obtain a distributed power supply node selection scheme. Then, a shortest path algorithm is used to optimize the link between the distributed power supplies to obtain a link selection scheme, and the power compensation of the distributed power supply is completed. The power compensation problem of the distributed power supply is modeled as a classical network embedding problem, and a deep reinforcement learning method is used to fully capture the nonlinear relationship between the power grid energy features, so that the intelligent power compensation of the distributed power supply in the power grid environment is finally realized, and the response speed and power grid resource utilization rate can be improved.

Description

Technical Field

[0001] This invention belongs to the field of power system technology, specifically relating to a method, system, and equipment for intelligent power replenishment of distributed power sources. Background Technology

[0002] To meet the ever-increasing demand for electricity, developing new power architectures and new power systems has become an inevitable choice. Distributed power sources are typically deployed on the user side, which can reduce losses caused by long-distance power transmission and improve energy efficiency. At the same time, they can alleviate the imbalance between power supply and demand to some extent during peak electricity demand periods. Furthermore, in the event of natural disasters or grid failures, distributed power sources can serve as backup power sources, enhancing the system's resilience.

[0003] While distributed generation (DG) can play a vital role in resource allocation and resilience, its flexibility, heterogeneity, and uneven distribution present significant challenges to the crucial process of power replenishment. Flexibility necessitates rapid response to changing demand, increasing the complexity of power replenishment control algorithms. DG encompasses various types, such as photovoltaics, wind power, and biomass energy, each with specific performance characteristics and power replenishment requirements, posing new challenges to grid resource orchestration. Traditional methods often suffer from excessive computational complexity and incomplete consideration of grid characteristics, resulting in excessively long decision-making times in complex grid environments. These methods fail to meet the demands of rapid response and high resource utilization for large-scale distributed generation power replenishment in the future, leading to increased costs for operators and unmet service quality for users. Summary of the Invention

[0004] The purpose of this invention is to address the aforementioned problems in the existing technology by providing a distributed power supply intelligent power replenishment method, system, and device that can improve power replenishment response speed and resource utilization.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows:

[0006] In a first aspect, the present invention proposes a method for intelligent power replenishment of distributed power sources, the method comprising:

[0007] S1. Extract power grid environment features and obtain power replenishment requests from distributed power sources. Based on the network embedding strategy, convert the extracted power grid environment features into a power grid weighted graph and convert the obtained power replenishment requests from distributed power sources into a power replenishment request weighted graph.

[0008] S2. Construct a policy network and train it using a deep reinforcement learning method with a weighted graph of the power grid and a weighted graph of the power replenishment request to obtain the selection scheme of the power replenishment node for the distributed power source.

[0009] S3. Based on the selection scheme of the power replenishment node for distributed power sources, the shortest path algorithm is used to optimize the links between distributed power sources to obtain the link selection scheme.

[0010] S4. Based on the distributed power source, complete the power replenishment scheme for the power replenishment node selection scheme and the link selection scheme.

[0011] In S1, the expression for the weighted graph of the power grid is G. E ={N E ,L E AN E ,AL E}; where N E AN represents the set of power replenishment nodes in a power system. A power replenishment node is a node in a power system that has the capability to replenish power for distributed generation sources. E This represents the set of attributes of the power replenishment node, including the total power p of the i-th power replenishment node. i The power W of the i-th power replenishment node i The number of other power supply nodes connected to the i-th power supply node in the power system. L E Represents the set of transmission links between power replenishment nodes; AL E This represents the set of attributes of the power transmission link between nodes, including the maximum carrying power Z of the i-th power transmission link. i ;

[0012] The expression for the weighted graph of the power replenishment request is G. D ={N D ,L D ,AN D ,AL D}, where N D Indicates the set of distributed power sources in the power replenishment request; AN D This represents the set of resource demands of distributed power sources, which includes the required replenishment power e of the k-th distributed power source. k The power demand for replenishment of the k-th distributed power source is v. k L D It is a set of distributed power source links; AL D This represents the set of demand for links between distributed power sources, including the power occupancy m of the k-th link between distributed power sources. k .

[0013] S2 includes: S21, constructing the state space; the specific construction steps include: extracting the feature composition vector of each power replenishment node.

[0014] Let represent the remaining power of the i-th power replenishment node, and its calculation formula is:

[0015]

[0016] In the above formula, This represents the i-th power replenishment node in the power system; This represents the k-th distributed power source in the power replenishment request; This represents the number of distributed power sources deployed in the i-th power replenishment node of the power system;

[0017] Let represent the current carrying capacity of the i-th transmission link, and its calculation formula is:

[0018]

[0019] In the above formula, This represents the i-th transmission link in the power system; This represents the k-th distributed power source link; This represents the number of distributed generation links deployed in the i-th transmission link of the power system;

[0020] The average distance between the i-th power supply node and other power supply nodes is represented by the following formula:

[0021]

[0022] In the above formula, Represents the i-th power replenishment node With the y-th power replenishment node The distance between them; |N E | Represents the total number of power replenishment nodes in the power system;

[0023] Then, the vectors v of all the electrified nodes are aggregated into the following feature matrix M. E To form a state space:

[0024]

[0025] S22. Construct a policy network; the policy network includes an input layer, a convolutional layer, a probability layer, a filtering layer, and an output layer; the input layer is used to receive the feature matrix M from the state space. E The convolutional layer is used based on the feature matrix M E Extract the available resource vector of the i-th power replenishment node.

[0026]

[0027] In the above formula, ReLU is the activation function; w is the weight; and b is the bias.

[0028] The probability layer is used to calculate the probability that the k-th distributed power source is deployed in each power replenishment node of the power system; and the probability that the k-th distributed power source is deployed in the i-th power replenishment node of the power system. The calculation formula is:

[0029]

[0030] The filtering layer is used to remove power supply nodes that do not meet the following resource constraints:

[0031] Current power constraints:

[0032] Power demand constraints:

[0033] Maximum power carrying capacity constraint:

[0034] The output layer is used to select the power supply node with the highest probability from the power supply nodes that have passed through the filtering layer as the power supply node of the k-th distributed power source using a greedy strategy.

[0035] S23. Calculate the loss function and update the parameter gradient of the policy network through gradient backpropagation until the policy network converges, and output the selection scheme of the distributed power source for the supplementary power node.

[0036] In S23, the loss function Loss is:

[0037]

[0038] In the above formula, Γ k This represents the true label vector of the k-th distributed power source. If the k-th distributed power source is deployed in the ith power replenishment node, then the ith dimension of the true label vector is set to 1, and the other dimensions are set to 0.

[0039] The parameter gradient update formula for the policy network is:

[0040] g = lr·Reward·g s ;

[0041]

[0042]

[0043]

[0044] In the above formula, g represents the updated parameter gradient; lr represents the learning rate; Reward represents the reward function; g s Reve(G) represents the gradient of the parameters before the update, obtained by taking the derivative of the loss function Loss; DCost(G) is the revenue earned from providing supplementary power services to the power system; D ,t) represents the cost incurred by the power system in providing supplementary power services; This represents the actual number of hops between two distributed power sources in a power node; |L D | indicates the number of links between distributed power sources.

[0045] Secondly, this invention proposes a distributed power intelligent power replenishment system, the distributed power intelligent power replenishment system comprising:

[0046] The data acquisition module is used to extract power grid environmental features and obtain the power replenishment requests of distributed power sources. Based on the network embedding strategy, the extracted power grid environmental features are converted into a power grid weighted graph and the obtained power replenishment requests of distributed power sources are converted into a power replenishment request weighted graph.

[0047] The power replenishment node selection module is used to construct a policy network. The grid environment characteristics and the power replenishment requests of distributed generation are input into the policy network. The policy network is trained using a deep reinforcement learning method to obtain the power replenishment node selection scheme for distributed generation.

[0048] The link selection module is used to select links based on the distributed power source's selection scheme for the replenishment node, using the shortest path algorithm to obtain the link selection scheme;

[0049] The power replenishment module is used to complete the power replenishment of distributed power sources based on the power replenishment node selection scheme and link selection scheme of distributed power sources.

[0050] The expression for the weighted graph of the power grid is G. E ={N E ,L E ,AN E ,AL E}; where N E AN represents the set of power replenishment nodes in a power system. A power replenishment node is a node in a power system that has the capability to replenish power for distributed generation sources. E This represents the set of attributes of the power replenishment node, including the total power p of the i-th power replenishment node. i The power supply w of the i-th power supply node i The number of other power supply nodes connected to the i-th power supply node in the power system. L E Represents the set of transmission links between power replenishment nodes; AL E This represents the set of attributes of the inter-node transmission link, including the maximum carrying power z of the i-th transmission link. i ;

[0051] The expression for the weighted graph of the power replenishment request is: G D={N D ,L D ,AN D ,AL D}; where N D Indicates the set of distributed power sources in the power replenishment request; AN D This represents the set of resource demands of distributed power sources, which includes the required replenishment power e of the k-th distributed power source. k The power demand for replenishment of the k-th distributed power source is v. k L D It is a set of distributed power source links; AL D This represents the set of demand for links between distributed power sources, including the power occupancy m of the k-th link between distributed power sources. k .

[0052] The power replenishment node selection module includes a state space construction module, a policy network construction module, and a network training and output module; the state space construction module is used to first extract the features of each power replenishment node to form the following vector:

[0053]

[0054] Let represent the remaining power of the i-th power replenishment node, and its calculation formula is:

[0055]

[0056] In the above formula, This represents the i-th power replenishment node in the power system; This represents the k-th distributed power source in the power replenishment request; This represents the number of distributed power sources deployed in the i-th power replenishment node of the power system;

[0057] Let represent the current carrying capacity of the i-th transmission link, and its calculation formula is:

[0058]

[0059] In the above formula, This represents the i-th transmission link in the power system; This represents the k-th distributed power source link; This represents the number of distributed generation links deployed in the i-th transmission link of the power system;

[0060] The average distance between the i-th power supply node and other power supply nodes is represented by the following formula:

[0061]

[0062] In the above formula, Represents the i-th power replenishment node With the y-th power replenishment node The distance between them; |N E | Represents the total number of power replenishment nodes in the power system;

[0063] Then, the vectors v of all the electrified nodes are aggregated into the following feature matrix M. E To form a state space:

[0064]

[0065] The policy network construction module is used to construct the policy network;

[0066] The policy network includes an input layer, a convolutional layer, a probability layer, a filtering layer, and an output layer.

[0067] The input layer is used to receive the feature matrix M from the state space. E ;

[0068] The convolutional layer is used based on the feature matrix M E Extract the available resource vector of the i-th power replenishment node.

[0069] In the above formula, ReLU is the activation function; w is the weight; and b is the bias.

[0070] The probability layer is used to calculate the probability that the k-th distributed power source is deployed in each power replenishment node of the power system; and the probability that the k-th distributed power source is deployed in the i-th power replenishment node of the power system. The calculation formula is:

[0071]

[0072] The filtering layer is used to remove power supply nodes that do not meet the following resource constraints:

[0073] Current power constraints:

[0074]

[0075] Power demand constraints:

[0076]

[0077] Maximum power carrying capacity constraint:

[0078]

[0079] The output layer is used to select the power supply node with the highest probability from the power supply nodes that have passed through the filtering layer as the power supply node of the k-th distributed power source using a greedy strategy.

[0080] The network training and output module is used to calculate the loss function and update the parameter gradient of the policy network through gradient backpropagation until the policy network converges.

[0081] Finally, the distributed power supply scheme for selecting replenishment nodes is output.

[0082] The loss function Loss is:

[0083]

[0084] In the above formula, Γ k This represents the true label vector of the k-th distributed power source;

[0085] If the k-th distributed power source is deployed in the ith replenishment node, then the ith dimension of the true label vector is set to 1, and the other dimensions are set to 0.

[0086] The parameter gradient update formula for the policy network is:

[0087] g = lr·RReward·g s ;

[0088]

[0089]

[0090]

[0091] In the above formula, g represents the updated parameter gradient; lr represents the learning rate; Reward represents the reward function; g s Reve(G) represents the gradient of the parameters before the update, obtained by taking the derivative of the loss function Loss; D Cost(G) is the revenue earned from providing supplementary power services to the power system; D ,t) represents the cost incurred by the power system in providing supplementary power services; This represents the actual number of hops between two distributed power sources in a power node; |L D | indicates the number of links between distributed power sources.

[0092] Thirdly, the present invention proposes a distributed power intelligent power replenishment device, the distributed power intelligent power replenishment device including a memory and a processor; the memory is used to store computer program code and transmit the computer program code to the processor; the processor is used to execute the aforementioned distributed power intelligent power replenishment method according to the instructions in the computer program code.

[0093] Fourthly, the present invention proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the aforementioned intelligent power replenishment method for distributed power sources.

[0094] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0095] 1. The present invention discloses an intelligent power replenishment method for distributed power sources. First, it extracts grid environment features and obtains power replenishment requests from distributed power sources. Based on a network embedding strategy, it converts the extracted grid environment features into a weighted grid graph and the obtained power replenishment requests from distributed power sources into a weighted graph of power replenishment requests. Then, it constructs a strategy network and trains the strategy network using the grid weighted graph and the power replenishment request weighted graph through deep reinforcement learning to obtain a power replenishment node selection scheme for the distributed power sources. Next, based on the power replenishment node selection scheme, it optimizes the links between distributed power sources using a shortest path algorithm to obtain a link selection scheme. Finally, it completes the power replenishment of the distributed power sources based on the power replenishment node selection scheme and the link selection scheme. This design models the distributed power replenishment problem as a classic network embedding problem and utilizes deep reinforcement learning to fully capture the nonlinear relationships between grid energy features, ultimately achieving intelligent power replenishment of distributed power sources in the grid environment. This allows the distributed power sources to self-learn and optimize their power replenishment strategies under constantly changing power demand and high distributed power source flexibility, thereby improving response speed and grid resource utilization, and enhancing the overall operating efficiency and reliability of the power system. Therefore, this invention can improve response speed and grid resource utilization, and is suitable for application in large-scale distributed power supply replenishment scenarios.

[0096] 2. The intelligent power replenishment method for distributed power sources described in this invention uses the revenue-expenditure ratio as an indicator to design a reward function. This function simultaneously considers the revenue and expenditure costs of distributed power source power replenishment operations and reflects the utilization efficiency of power system resources. The policy network determines its subsequent deployment actions based on the reward value, thereby improving the success rate of distributed power source power replenishment requests and further enhancing the utilization rate of power grid resources. Therefore, this invention can improve the success rate of distributed power source power replenishment requests and further enhance the utilization rate of power grid resources. Attached Figure Description

[0097] Figure 1 This is a flowchart of the intelligent power replenishment method for distributed power sources described in this invention.

[0098] Figure 2 This is a structural block diagram of the distributed power intelligent replenishment system described in this invention.

[0099] Figure 3This is a structural block diagram of the distributed power intelligent power replenishment device described in this invention. Detailed Implementation

[0100] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings.

[0101] Example 1:

[0102] See Figure 1 A method for intelligent power replenishment of distributed power sources is carried out in the following steps:

[0103] S1. Extract power grid environmental features and obtain power replenishment requests from distributed generation sources. Based on a network embedding strategy, convert the extracted power grid environmental features into a weighted power grid graph and the obtained power replenishment requests from distributed generation sources into a weighted power replenishment request graph. The expression for the resulting weighted power grid graph is G. E ={N E ,L E ,AN E ,AL E}; where N E AN represents the set of power replenishment nodes in a power system. A power replenishment node is a node in a power system that has the capability to replenish power for distributed generation sources. E This represents the set of attributes of the power replenishment node, including the total power p of the i-th power replenishment node. i The power supply w of the i-th power supply node i The number of other power supply nodes connected to the i-th power supply node in the power system. L E Represents the set of transmission links between power replenishment nodes; AL E This represents the set of attributes of the power transmission link between nodes, including the maximum carrying power Z of the i-th power transmission link. i The expression for obtaining the weighted graph of the power replenishment request is G. D ={N D ,L D AN D ,AL D}, where N D Indicates the set of distributed power sources in the power replenishment request; AN D This represents the set of resource demands of distributed power sources, which includes the required replenishment power e of the k-th distributed power source. k The power demand for replenishment of the k-th distributed power source is v. k L D It is a set of distributed power source links; AL D This represents the set of demand for links between distributed power sources, including the power occupancy m of the k-th link between distributed power sources. k ;

[0104] S2. Construct a policy network. Train the policy network using a deep reinforcement learning method with the weighted graph of the power grid and the weighted graph of the power replenishment request to obtain the selection scheme of the power replenishment node for the distributed generation. The specific steps are as follows:

[0105] S21. Construct the state space; to ensure the policy network fully learns the features of the power grid nodes and to fully consider training feasibility, extract the features of the supplementary power nodes to form the following vector:

[0106] in Let represent the remaining power of the i-th power replenishment node, and its calculation formula is:

[0107]

[0108] In the above formula, This represents the i-th power replenishment node in the power system; This represents the k-th distributed power source in the power replenishment request; This represents the number of distributed power sources deployed in the i-th power replenishment node of the power system;

[0109] Let represent the current carrying capacity of the i-th transmission link, and its calculation formula is:

[0110]

[0111] In the above formula, This represents the i-th transmission link in the power system; This represents the k-th distributed power source link; This represents the number of distributed generation links deployed in the i-th transmission link of the power system;

[0112] The average distance between the i-th power supply node and other power supply nodes is represented by the following formula:

[0113]

[0114] In the above formula, Represents the i-th power replenishment node With the y-th power replenishment node The distance between them; |N E | Represents the total number of power replenishment nodes in the power system;

[0115] The vectors v of all the electrified nodes are aggregated into the following feature matrix M. E To form a state space:

[0116]

[0117] S22. Construct a policy network; the policy network includes an input layer, a convolutional layer, a probability layer, a filtering layer, and an output layer; the input layer is used to receive the feature matrix M from the state space. E The convolutional layer is used based on the feature matrix M E Extract the available resource vector of the i-th power replenishment node.

[0118]

[0119] In the above formula, ReLU is the activation function; w is the weight; and b is the bias.

[0120] The probability layer is used to calculate the probability of the k-th distributed power source being deployed at each supplementary power node in the power system using the softmax function; the probability of the k-th distributed power source being deployed at the i-th supplementary power node in the power system. The calculation formula is:

[0121]

[0122] The filtering layer is used to remove power supply nodes that do not meet the following resource constraints:

[0123] Current power constraints:

[0124]

[0125] Power demand constraints:

[0126]

[0127] Maximum power capacity:

[0128]

[0129] The output layer is used to select the power supply node with the highest probability from the power supply nodes that have passed through the filtering layer as the power supply node of the k-th distributed power source using a greedy strategy.

[0130] S23. Calculate the loss function and update the parameter gradients of the policy network through gradient backpropagation until the policy network converges; the loss function Loss is:

[0131]

[0132] In the above formula, Γ k This represents the true label vector of the k-th distributed power source;

[0133] If the k-th distributed power source is deployed in the ith replenishment node, then the ith dimension of the true label vector is set to 1, and the other dimensions are set to 0.

[0134] The parameter gradient update formula for the policy network is:

[0135] g = lr·Reward·g s ;

[0136]

[0137]

[0138]

[0139] In the above formula, g represents the updated parameter gradient; lr represents the learning rate; Reward represents the reward function. When the reward function value increases, it indicates that the quality of the current power replenishment strategy is high; when the reward function value decreases, it indicates that the current optimization direction deviates from the correct direction. s Reve(G) represents the gradient of the parameters before the update, obtained by taking the derivative of the loss function Loss; D Cost(G) is the revenue earned from providing supplementary power services to the power system; D ,t) represents the cost incurred by the power system in providing supplementary power services; This represents the actual number of hops between two distributed power sources in a power node; |L D | indicates the number of links between distributed power sources;

[0140] S24. Use the trained policy network to output the distributed power source's selection scheme for the power replenishment node.

[0141] S3. After all distributed power sources have been deployed, the shortest path algorithm is used to optimize the links between distributed power sources to obtain the link selection scheme with the minimum total length.

[0142] S4. Based on the distributed power source, complete the power replenishment scheme for the power replenishment node selection scheme and the link selection scheme.

[0143] Example 2:

[0144] See Figure 2 A distributed power generation intelligent power replenishment system includes a data acquisition module, a power replenishment node selection module, and a link selection module. The data acquisition module extracts grid environment features and obtains power replenishment requests from distributed power sources. Based on a network embedding strategy, it converts the extracted grid environment features into a grid weighted graph and the obtained distributed power source power replenishment requests into a power replenishment request weighted graph. The expression for the grid weighted graph is G. E ={N E ,L E ,AN E ,ALE}; where N E AN represents the set of power replenishment nodes in a power system. A power replenishment node is a node in a power system that has the capability to replenish power for distributed generation sources. E This represents the set of attributes of the power replenishment node, including the total power p of the i-th power replenishment node. i The power supply w of the i-th power supply node i The number of other power supply nodes connected to the i-th power supply node in the power system. L E Represents the set of transmission links between power replenishment nodes; AL E This represents the set of attributes of the inter-node transmission link, including the maximum carrying power z of the i-th transmission link. i The expression for the weighted graph of the power replenishment request is G. D ={N D ,L D ,AN D ,AL D}, where N D Indicates the set of distributed power sources in the power replenishment request; AN D This represents the set of resource demands of distributed power sources, which includes the required replenishment power e of the k-th distributed power source. k The power demand for replenishment of the k-th distributed power source is v. k L D It is a set of distributed power source links; AL D This represents the set of demand for links between distributed power sources, including the power occupancy m of the k-th link between distributed power sources. k The power replenishment node selection module is used to input the grid environment characteristics and the power replenishment request of the distributed generation into a pre-built policy network, and to train the policy network using a deep reinforcement learning method to obtain the power replenishment node selection scheme for the distributed generation. The power replenishment node selection module includes a state space construction module, a policy network construction module, and a network training and output module. The state space construction module is used to first extract the power replenishment node features to form the following vector:

[0145] in Let represent the remaining power of the i-th power replenishment node, and its calculation formula is:

[0146]

[0147] In the above formula, This represents the i-th power replenishment node in the power system; This represents the k-th distributed power source in the power replenishment request; This represents the number of distributed power sources deployed in the i-th power replenishment node of the power system;

[0148] Let represent the current carrying capacity of the i-th transmission link, and its calculation formula is:

[0149]

[0150] In the above formula, This represents the i-th transmission link in the power system; This represents the k-th distributed power source link; This represents the number of distributed generation links deployed in the i-th transmission link of the power system;

[0151] The average distance between the i-th power supply node and other power supply nodes is represented by the following formula:

[0152]

[0153] In the above formula, Represents the i-th power replenishment node With the y-th power replenishment node The distance between them; |N E | Represents the total number of power replenishment nodes in the power system;

[0154] Then, the vectors v of all the electrified nodes are aggregated into the following feature matrix M. E To form a state space:

[0155]

[0156] The policy network construction module is used to construct a policy network; the policy network includes an input layer, a convolutional layer, a probability layer, a filtering layer, and an output layer; the input layer is used to construct a feature matrix M based on the state space as follows. E :

[0157]

[0158] The convolutional layer is used to extract the available resource vector of the i-th power supplement node based on the feature matrix:

[0159] Let represent the available resource vector of the i-th power replenishment node; ReLU is the activation function; w is the weight; b is the bias; the probability layer is used to solve for the probability that the k-th distributed source is deployed in each power replenishment node of the power system; the probability that the k-th distributed source is deployed in the i-th power replenishment node of the power system. The calculation formula is:

[0160]

[0161] The filtering layer is used to remove power supply nodes that do not meet the following resource constraints:

[0162] Current power constraints: Power demand constraints: Maximum power carrying capacity constraint:

[0163] The output layer uses a greedy strategy to select the power supply node with the highest probability from the power supply nodes passing through the filtering layer as the power supply node of the k-th distributed power source; the network training and output module is used to calculate the loss function and update the parameter gradient of the policy network through gradient backpropagation until the policy network converges; the loss function Loss is: In the above formula, Γ k This represents the true label vector of the k-th distributed power source; if the k-th distributed power source is deployed in the ith power replenishment node, then the ith dimension of the true label vector is set to 1, and the other dimensions are set to 0.

[0164] The parameter gradient update formula for the policy network is:

[0165] g = lr·RReward·g s ;

[0166]

[0167]

[0168]

[0169] In the above formula, g represents the updated parameter gradient; lr represents the learning rate; Reward represents the reward function; g s Reve(G) represents the gradient of the parameters before the update, obtained by taking the derivative of the loss function Loss; D Cost(G) is the revenue earned from providing supplementary power services to the power system; D ,t) represents the cost incurred by the power system in providing supplementary power services; This represents the actual number of hops between two distributed power sources in a power node; |L D | indicates the number of links between distributed power sources;

[0170] The trained policy network is used to output a distributed power source selection scheme for the power replenishment node.

[0171] The link selection module is used to select links between distributed power sources based on the distributed power source's selection scheme for replenishment nodes, using the shortest path algorithm to obtain the link selection scheme; the replenishment module is used to complete the replenishment of distributed power sources based on the distributed power source's selection scheme for replenishment nodes and the link selection scheme.

[0172] Example 3:

[0173] See Figure 3 A distributed power intelligent power replenishment device includes a memory and a processor; the memory is used to store computer program code and transmit the computer program code to the processor; the processor is used to execute the distributed power intelligent power replenishment method described in Embodiment 1 according to the instructions in the computer program code.

[0174] Example 4:

[0175] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the distributed power intelligent replenishment method described in Embodiment 1.

[0176] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program goods. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program goods embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0177] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program goods according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can 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, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0178] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0179] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0180] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for intelligent power replenishment of distributed power sources, characterized in that: The intelligent power replenishment method for distributed power sources includes: S1. Extract power grid environment features and obtain power replenishment requests from distributed power sources. Based on the network embedding strategy, convert the extracted power grid environment features into a power grid weighted graph and convert the obtained power replenishment requests from distributed power sources into a power replenishment request weighted graph. S2. Construct a policy network and train it using a deep reinforcement learning method with a weighted graph of the power grid and a weighted graph of the power replenishment request to obtain the selection scheme of the power replenishment node for the distributed power source. S3. Based on the selection scheme of the power replenishment node for distributed power sources, the shortest path algorithm is used to optimize the links between distributed power sources to obtain the link selection scheme. S4. Based on the distributed power source's selection scheme for power replenishment nodes and link selection scheme, complete the distributed power source power replenishment; S2 includes: S21. Construct the state space; specific steps include: extracting the feature vector of each power supply node. ,in Indicates the first The remaining power of each replenishment node; Indicates the first The current power carrying capacity of each transmission link; Indicates the first The average distance between each power supply node and other power supply nodes; then the vectors of all power supply nodes are... Aggregate into feature matrix To form a state space; S22. Construct a policy network; the policy network includes an input layer, a convolutional layer, a probability layer, a filtering layer, and an output layer; The input layer is used to receive the feature matrix from the state space. ; The convolutional layer is used based on the feature matrix. Extracting the first Available resource vector of each power replenishment node ; The probability layer is used to solve for the first... The probability of a distributed power source being deployed at various power replenishment nodes in a power system; The filtering layer is used to remove power replenishment nodes that do not meet the following resource constraints: current power constraint, power demand constraint, and maximum carrying power constraint. The output layer is used to select the power supply node with the highest probability from the power supply nodes that have passed through the filtering layer as the first power supply node. A distributed power supply replenishment node; S23. Calculate the loss function and update the parameter gradient of the policy network through gradient backpropagation until the policy network converges, and output the selection scheme of the distributed power source for the supplementary power node.

2. The intelligent power replenishment method for distributed power sources according to claim 1, characterized in that: In S1, the expression for the weighted graph of the power grid is: ;in This represents the set of power replenishment nodes in a power system. Power replenishment nodes are nodes in a power system that have the ability to replenish power for distributed generation. Represents the set of attributes of the power supply node, which includes the first... Total power of each power replenishment node , No. The power supply of each power supply node In the power system and the first Number of other power supply nodes connected to each power supply node ; This represents the set of power transmission links between the replenishment nodes; The set of attributes representing the power transmission link between nodes, including the first... Maximum carrying capacity of each transmission link ; The expression for the weighted graph of the power replenishment request is: ,in This represents the set of distributed power sources in the power replenishment request; This represents the set of resource requirements for distributed power sources, which includes the first... The required replenishment power for each distributed power source , No. The power replenishment requirement of a distributed power source ; It is a collection of links between distributed power sources; This represents the set of requirements for links between distributed power sources, which includes the first... Power utilization of distributed power supply links .

3. The intelligent power replenishment method for distributed power sources according to claim 2, characterized in that: The The calculation formula is: ； In the above formula, Indicates the first in the power system One power replenishment node; Indicating the request for power replenishment One distributed power source; Indicates the first deployment in the power system The number of distributed power sources in each power replenishment node; The The calculation formula is: ； In the above formula, Indicates the first in the power system One power transmission link; Indicates the first One distributed power source link; Indicates the first deployment in the power system The number of links between distributed power sources in a power transmission link; The The calculation formula is: ； In the above formula, Indicates the first One power supply node With the One power supply node The distance between them; This indicates the total number of power replenishment nodes in the power system; The expression for the state space is: ； The convolutional layer is used to extract the first... according to the following formula. Available resource vector of each power replenishment node : ； In the above formula, For activation functions; As weight; For bias; The probability layer is used to calculate the first probability according to the following formula. The first distributed power source deployed in the power system The probability of a power replenishment node : ； The current power constraint is: ; The power demand constraint is: ; The maximum power carrying capacity constraint is: .

4. The intelligent power replenishment method for distributed power sources according to claim 3, characterized in that: In S23, the loss function for: ； In the above formula, Indicates the first The true label vector of the distributed power source, if the ... The distributed power supply was deployed in the [number]th [location]. In the nth supplementary power node, the nth true label vector Set one dimension to 1, and the rest of the dimensions to 0; The parameter gradient update formula for the policy network is: ； ； ； ； In the above formula, This represents the gradient of the updated parameters; Indicates the learning rate; Represents the reward function; This represents the gradient of the parameters before the update, obtained by applying the loss function. Obtained by differentiation; Revenue earned from providing supplemental power services to the power system; This represents the cost incurred by the power system in providing supplemental power services; This represents the actual number of hops between two distributed power sources in a power node; This indicates the number of links between distributed power sources.

5. A distributed power intelligent power replenishment system, characterized in that: The distributed power intelligent power replenishment system includes: The data acquisition module is used to extract power grid environmental features and obtain the power replenishment requests of distributed power sources. Based on the network embedding strategy, the extracted power grid environmental features are converted into a power grid weighted graph and the obtained power replenishment requests of distributed power sources are converted into a power replenishment request weighted graph. The power replenishment node selection module is used to construct a policy network. The grid environment characteristics and the power replenishment requests of distributed generation are input into the policy network. The policy network is trained using a deep reinforcement learning method to obtain the power replenishment node selection scheme for distributed generation. The link selection module is used to select links based on the distributed power source's selection scheme for the replenishment node, using the shortest path algorithm to obtain the link selection scheme; The power replenishment module is used to complete the power replenishment of distributed power sources based on the power replenishment node selection scheme and link selection scheme of distributed power sources; The power replenishment node selection module includes a state space construction module, a policy network construction module, and a network training and output module. The state space construction module is used to first extract the feature composition vector of each power replenishment node. ,in Indicates the first The remaining power of each replenishment node; Indicates the first The current power carrying capacity of each transmission link; Indicates the first The average distance between each power supply node and other power supply nodes; then the vectors of all power supply nodes are... Aggregate into feature matrix To form a state space; The policy network construction module is used to construct a policy network; the policy network includes an input layer, a convolutional layer, a probability layer, a filtering layer, and an output layer; The input layer is used to receive the feature matrix from the state space. ; The convolutional layer is used based on the feature matrix. Extracting the first Available resource vector of each power replenishment node ; The probability layer is used to solve for the first... The probability of a distributed power source being deployed at various power replenishment nodes in a power system; The filtering layer is used to remove power replenishment nodes that do not meet the following resource constraints: current power constraint, power demand constraint, and maximum carrying power constraint. The output layer is used to select the power supply node with the highest probability from the power supply nodes that have passed through the filtering layer as the first power supply node. A distributed power supply replenishment node; The network training and output module is used to calculate the loss function and update the parameter gradient of the policy network through gradient backpropagation until the policy network converges, and outputs the selection scheme of the distributed power source for the supplementary power node.

6. A distributed power intelligent replenishment system according to claim 5, characterized in that: The expression for the weighted graph of the power grid is: ;in This represents the set of power replenishment nodes in a power system. Power replenishment nodes are nodes in a power system that have the ability to replenish power for distributed generation. Represents the set of attributes of the power supply node, which includes the first... Total power of each power replenishment node , No. The power supply of each power supply node In the power system and the first Number of other power supply nodes connected to each power supply node ; This represents the set of power transmission links between the replenishment nodes; The set of attributes representing the power transmission link between nodes, including the first... Maximum carrying capacity of each transmission link ; The expression for the weighted graph of the power replenishment request is: ,in This represents the set of distributed power sources in the power replenishment request; This represents the set of resource requirements for distributed power sources, which includes the first... The required replenishment power for each distributed power source , No. The power replenishment requirement of a distributed power source ; It is a collection of links between distributed power sources; This represents the set of requirements for links between distributed power sources, which includes the first... Power utilization of distributed power supply links .

7. A distributed power intelligent replenishment system according to claim 6, characterized in that: The The calculation formula is: ； In the above formula, Indicates the first in the power system One power replenishment node; Indicating the request for power replenishment One distributed power source; Indicates the first deployment in the power system The number of distributed power sources in each power replenishment node; The The calculation formula is: ； In the above formula, Indicates the first in the power system One power transmission link; Indicates the first One distributed power source link; Indicates the first deployment in the power system The number of links between distributed power sources in a power transmission link; The The calculation formula is: ； In the above formula, Indicates the first One power supply node With the One power supply node The distance between them; This indicates the total number of power replenishment nodes in the power system; The expression for the state space is: ； The convolutional layer is used to extract the first... according to the following formula. Available resource vector of each power replenishment node : ； In the above formula, For activation functions; As weight; For bias; The probability layer is used to calculate the first probability according to the following formula. The first distributed power source deployed in the power system The probability of a power replenishment node : ； The current power constraint is: ; The power demand constraint is: ; The maximum power carrying capacity constraint is: .

8. A distributed power intelligent replenishment system according to claim 7, characterized in that: The loss function for: ; In the above formula, Indicates the first The true label vector of the distributed power source, if the ... The distributed power supply was deployed in the [number]th [location]. In the nth supplementary power node, the nth true label vector Set one dimension to 1, and the rest of the dimensions to 0; The parameter gradient update formula for the policy network is: ； ； ； ； In the above formula, This represents the gradient of the updated parameters; Indicates the learning rate; Represents the reward function; This represents the gradient of the parameters before the update, obtained by applying the loss function. Obtained by differentiation; Revenue earned from providing supplemental power services to the power system; This represents the cost incurred by the power system in providing supplemental power services; This represents the actual number of hops between two distributed power sources in a power node; This indicates the number of links between distributed power sources.

9. A distributed power supply intelligent power replenishment device, characterized in that: The distributed power intelligent power replenishment device includes a memory and a processor; the memory is used to store computer program code and transmit the computer program code to the processor; the processor is used to execute the distributed power intelligent power replenishment method as described in claims 1-4 according to the instructions in the computer program code.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the distributed power intelligent replenishment method as described in claims 1-4.

Images