Distributed optimal power flow solving method and system based on self-supervised learning

Abstract

The invention discloses a distributed optimal power flow solving method and system based on self-supervised learning, and belongs to the field of power distribution network dispatching. According to the solution method, an alternating direction multiplier method (ADMM) is fused into a training process of a neural network, and a self-supervised learning algorithm is provided. According to the method, two neural networks are respectively trained to estimate the values of a global variable and a Lagrange multiplier, and mutual supervision of the two networks is realized based on iterative interaction of an ADMM algorithm. And then, estimating values of a global variable and a Lagrangian multiplier through the trained neural network, and obtaining a solution of the distributed optimal power flow based on the estimated values. According to the distributed optimal power flow acceleration solving method provided by the invention, the training of the neural network can be realized without a label, and the solving time of the distributed optimal power flow can be effectively shortened while certain accuracy is ensured.

Description

Technical Field

[0001] The present invention relates to a distributed optimal power flow solving technology based on self-supervised learning, belonging to the technical field of distribution network dispatching. Background Art

[0002] The Distributed Optimal Power Flow (DOPF) architecture divides large-scale distribution networks into multiple autonomous regions and employs a parallel computing framework to improve the efficiency of solving the optimal power flow problem. However, ensuring the consistency of variables within the distribution network's autonomous regions typically requires a large number of iterations, which significantly impacts the algorithm's convergence speed. Among the current mainstream alternating distributed algorithms, the Alternating Direction Method of Multipliers (ADMM) has garnered widespread attention due to its theoretical support for convex optimization and ease of engineering. Its decomposition and coordination mechanism balances optimization accuracy and privacy protection, but its requirement of hundreds of iterations still struggles to meet the requirements of minute-level real-time scheduling. Some existing methods utilize machine learning algorithms to accelerate the solution of DOPF, improving its efficiency to a certain extent. However, these methods primarily rely on supervised learning methods. Given the complex and ever-changing operational landscape of today's emerging distribution networks, machine learning algorithms require diverse datasets to address these complex and volatile scenarios. However, generating large amounts of labeled data is time-consuming, a significant limitation that has significantly hindered the development of machine learning applications in OPF solutions. Summary of the Invention

[0003] Purpose of the invention: In response to the shortcomings of the existing technology, the present invention provides a distributed optimal power flow solution method and system based on self-supervised learning, which can reduce the dependence on samples when using machine learning algorithms to assist in distributed optimal power flow solution.

[0004] Technical solution: In order to achieve the above invention objectives, the present invention adopts the following technical solution:

[0005] A distributed optimal power flow solution method based on self-supervised learning includes the following steps:

[0006] Based on the physical characteristics of the distribution network topology, the connectivity between nodes is evaluated using the electrical coupling strength. The distribution network is divided into several sub-regions using community mining theory. The optimal power flow problem of the distribution network is decomposed into optimal power flow sub-problems within each sub-region according to the divided sub-regions, and a distributed optimal power flow problem is established.

[0007] Based on the current values of global variables and Lagrange multipliers, the optimal power flow problem of each sub-region is solved separately to obtain the value of the boundary variable; based on the current values of the boundary variables of each sub-region and the value of the Lagrange multiplier, the parameters of the original network are updated;

[0008] Input the system load data into the original network to estimate the value of the global variable; update the target value of the Lagrange multiplier based on the current value of the global variable and the value of the Lagrange multiplier in the previous iteration, and update the parameters of the dual network based on this;

[0009] Input the system load data into the dual network and estimate the value of the Lagrange multiplier. Calculate the original residual and dual residual between the current iteration and the previous iteration to determine whether the algorithm has converged. If the algorithm does not converge, execute the next iteration, recalculate the optimal power flow in each sub-region, and update the original network and dual network in turn. If the algorithm converges, offline training is complete.

[0010] During the online solution process, the system load data is input into the original network and the dual network respectively, and the values of global variables and Lagrange multipliers are estimated respectively. The estimated values of global variables and Lagrange multipliers are used as the boundary variables of the optimal power flow in each sub-region, thereby completing the rapid solution of the distributed optimal power flow.

[0011] Furthermore, based on the physical characteristics of the distribution network topology, the connectivity between nodes is evaluated using the electrical coupling strength, and the community mining theory is used to divide the distribution network into several sub-areas, including:

[0012] Using the electrical coupling strength e between node i and node j ij To evaluate the connectivity between nodes, we use community mining theory and take modularity Q maximization as the goal to divide the distribution network into several sub-areas, which can be expressed as follows:

[0013]

[0014] Where E is the element e ij The symmetric matrix composed of i represents the sum of the elements in the i-th row of matrix E, C i Indicates the region to which node i belongs. When C i =C j When δ(C i C j )=1, otherwise, δ(C i C j )=0.

[0015] Furthermore, both the original network and the dual network use a deep neural network as the architecture, including an input layer, several hidden layers, an activation layer, and an output layer. The input data are the active load and reactive load u=[P Load ;Q Load ], the output data are the value of the global variable z and the value of the Lagrange multiplier λ respectively.

[0016] Furthermore, the loss function used to update the original network parameters is as follows:

[0017]

[0018] Where N is the set of all sub-areas of the distribution network, is the boundary variable of subregion n at the k+1th iteration, M n for The matching matrix with z is a 0-1 matrix; is the Lagrange multiplier of subregion n at the k+1th iteration, and ρ is the penalty factor of the alternating direction multiplier method.

[0019] Furthermore, the target value of the dual network is updated using the following formula:

[0020]

[0021] Where, is the target value of the Lagrange multiplier of subregion n at the k+1th iteration;

[0022] The loss function used to update the dual network parameters is as follows:

[0023]

[0024] Among them, λ is the output value of the dual network, λ *(k+1) is the target value of the Lagrange multiplier at the k+1th iteration, n λ is the dimension of λ.

[0025] Furthermore, the original residual and the dual residual Update using the following formula:

[0026]

[0027] The convergence condition of the algorithm is defined as follows:

[0028]

[0029] Among them, ε (k+1) is the residual error of the system, ε * is the tolerance threshold for the given residual.

[0030] A distributed optimal power flow solution system based on self-supervised learning, comprising:

[0031] The distribution network partitioning module is used to evaluate the connectivity between nodes based on the physical characteristics of the distribution network topology and the strength of electrical coupling. It also uses community mining theory to divide the distribution network into several sub-regions. Based on the divided sub-regions, the optimal power flow problem of the distribution network is decomposed into optimal power flow sub-problems within each sub-region, and a distributed optimal power flow problem is established.

[0032] The original network update module is used to solve the optimal power flow problem of each sub-region based on the current global variables and the values of the Lagrange multipliers to obtain the values of the boundary variables; based on the current values of the boundary variables of each sub-region and the values of the Lagrange multipliers, the parameters of the original network are updated;

[0033] The dual network update module is used to input the system load data into the original network and estimate the value of the global variable; based on the current value of the global variable and the value of the Lagrange multiplier in the previous iteration, the target value of the Lagrange multiplier is updated, and the parameters of the dual network are updated based on this;

[0034] The training control module is used to input the system load data into the dual network and estimate the value of the Lagrange multiplier. It calculates the original residual and the dual residual between the current iteration and the previous iteration to determine whether the algorithm has converged. If the algorithm does not converge, it executes the next iteration, recalculates the optimal power flow in each sub-region, and updates the original network and the dual network in turn. If the algorithm converges, the offline training is completed.

[0035] The online solution module is used to input the system load data into the original network and the dual network respectively during the online solution process, and estimate the values of global variables and Lagrange multipliers respectively; the estimated values of global variables and Lagrange multipliers are used as the boundary variables of the optimal power flow in each sub-region, thereby completing the rapid solution of the distributed optimal power flow.

[0036] The present invention also provides an electronic device comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and when the programs are executed by the processors, the distributed optimal power flow solution method based on self-supervised learning as described above is implemented.

[0037] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the distributed optimal power flow solution method based on self-supervised learning as described above is implemented.

[0038] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the distributed optimal power flow solving method based on self-supervised learning as described above.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] (1) A self-supervised learning mechanism is used to reduce data annotation costs. The original network and the dual network automatically extract features from system operation data (load data) through alternating learning. The original network estimates global variables based on load data, and the dual network updates parameters based on global variables and Lagrange multipliers, achieving collaborative optimization without external annotation.

[0041] (2) During the iteration process, by minimizing the primal and dual residuals, features related to the optimal power flow are automatically extracted from the data, and the interaction between sub-regions and the nonlinear characteristics of the system are learned. The trained model has strong adaptive generalization capabilities and can quickly adapt to different operating scenarios such as peak and valley loads and fault recovery, reducing the cost of remodeling and parameter adjustment.

[0042] (3) By demarcating power grid sub-regions through community mining, the global problem is decomposed into multiple sub-regions for parallel computation, significantly reducing computational complexity. Combining offline training with online rapid deployment, only one forward propagation is required in the online phase to estimate the optimal solution, significantly improving computational response speed and meeting the real-time optimization needs of large-scale distribution networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is a flow chart of the solution method of the present invention;

[0044] Figure 2 It is an improved IEEE123 node distribution network topology partition diagram in an embodiment of the present invention;

[0045] Figure 3 is a residual curve diagram during the training process of the original network and the dual network in an embodiment of the present invention;

[0046] Figure 4 This is an example of the voltage regulation result of the method proposed in the present invention. DETAILED DESCRIPTION

[0047] In order to have a clearer understanding of the features and advantages of the technical solution of the present invention, the composition and implementation of the specific solution are explained below with reference to the accompanying drawings.

[0048] Reference Figure 1 The present invention proposes a distributed optimal power flow solution method based on self-supervised learning, comprising the following steps:

[0049] Step 1: Based on the physical characteristics of the distribution network topology, the connectivity between nodes is evaluated using the electrical coupling strength. The distribution network is divided into several sub-regions using community mining theory. The optimal power flow problem of the distribution network is decomposed into optimal power flow sub-problems within each sub-region according to the divided sub-regions, and a distributed optimal power flow problem is established.

[0050] In this embodiment, the calculation example adopts the improved IEEE-123 node distribution network partition as follows Figure 2 As shown in the figure, a total of 12 photovoltaic generators and 12 static VAR compensators are configured in IEEE-123 nodes, and the distribution network in this example is divided into five sub-areas. It should be noted that the distribution network partitioning in this embodiment is based on the IEEE-123 node distribution network of this embodiment. When implementing the method of the present invention, the distribution network sub-area division can be adaptively adjusted according to the structure of the actual distribution network.

[0051] The present invention uses the maximization of modularity Q as the basis for the division of distribution network sub-areas, which is in the following form:

[0052]

[0053] Where, e ij is the coupling degree between node i and node j in the system, and E is the element e ij The symmetric matrix composed of i represents the sum of the elements in the i-th row of matrix E, C i Indicates the region to which node i belongs. When C i =C j When δ(C i C j )=1, otherwise, δ(C i C j ) = 0. In this embodiment, the electrical coupling strength is used to measure the coupling degree between node i and node j.

[0054] The optimal power flow problem of the distribution network is decomposed into optimal power flow sub-problems in each sub-area according to the divided sub-areas. The distributed optimal power flow achieves power system optimization by solving the optimal power flow sub-problems in each sub-area in parallel.

[0055] Based on the distributed optimal power flow problem established in step 1, offline training of the neural network is performed based on steps 2 to 7. Step 2 solves the optimal power flow of the sub-region at the current iteration based on the output value of the neural network, which is used for further updating of the subsequent neural network. Step 3 trains the original network based on the optimal power flow of the sub-region at the current iteration obtained in step 2 and the output of the dual network at the previous iteration. Step 4 updates the value of the global variable through the neural network updated in step 3, which is used to update the target value of the Lagrange multiplier in step 5. Step 5 calculates the target value of the Lagrange multiplier based on the optimal power flow of the sub-region at the current iteration obtained in step 2 and the global variables updated in step 4. Step 6 trains the dual network based on the target value of the Lagrange multiplier obtained in step 5. Step 7 calculates the residual value based on the global variables and Lagrange multipliers of the current iteration and the previous iteration to determine whether the training process is complete. If the residual value is less than the given value, the training is complete and the process proceeds to step 8. Otherwise, the process returns to step 2 and proceeds to the next iteration.

[0056] Steps 8 to 9 are the online solution process. In step 8, the system load value is input into the original network and the dual network respectively, the value of the global variable and the Lagrange multiplier is estimated, and it is used as the boundary condition for solving the distributed optimal power flow in step 9. Step 9 calculates the distributed optimal power flow solution based on the boundary conditions given in step 8.

[0057] The specific implementation method of each step is given below.

[0058] Step 2: Based on the current global variables and the values of the Lagrange multipliers, solve the optimal power flow problem for each sub-region separately. The solution formula is as follows:

[0059]

[0060] Among them, x n The variable of sub-region n is determined by the internal variable and boundary variables Composition, X n is the variable x n The feasible domain of z (k+1) is the global variable at the k+1th iteration, is the Lagrange multiplier of subregion n at the k+1th iteration, M n for and the matching matrix of z (0-1 matrix).

[0061] Step 3: Based on the values of the boundary variables and the Lagrange multipliers of each current sub-region, update the parameters of the original network;

[0062] Specifically, the original network uses a deep neural network as its architecture, including an input layer, several hidden layers, an activation layer, and an output layer. Its input data are the active load and reactive load u=[P Load ;Q Load ], the output data is the value of the global variable z. The following formula is the original network forward propagation formula:

[0063] h1=σ1(w1u+b1)

[0064] h i =σ i (w i h i-1 +b i ),i=2,3,...,j

[0065] z=σ j+1 (w j+1 h j +b j+1 )

[0066] Where j is the number of hidden layers in the original network, w i and b i are the weight matrix and bias vector of the i-th hidden layer respectively. σ i is the activation function of the i-th hidden layer, h i is the output of the i-th hidden layer.

[0067] Specifically, the loss function used to update the original network parameters at the k+1th iteration of the self-supervised learning algorithm is as follows:

[0068]

[0069] Where N is the set of all sub-areas of the distribution network, is the boundary variable of subregion n at the k+1th iteration. The superscript b corresponds to boundary, indicating the boundary variable. n for The matching matrix of z is a 0-1 matrix. is the Lagrange multiplier of subregion n at the k+1th iteration, and ρ is the penalty factor of ADMM.

[0070] Step 4: Input the system load data into the original network and estimate the value of the global variable z;

[0071] Step 5: Update the target value of the Lagrange multiplier according to the value of the current global variable and the value of the Lagrange multiplier in the previous iteration, and update the parameters of the dual network based on this;

[0072] Specifically, the dual network uses a deep neural network as its architecture, including an input layer, several hidden layers, an activation layer, and an output layer. Its input data are the active load and reactive load u=[P Load ;Q Load ], and the output data is the value of the Lagrange multiplier λ. The forward propagation formula of the dual network is similar to that of the original network, except that the output of the original network is a global variable, while the output of the dual network is a Lagrange multiplier. In practical applications, the original network and the dual network are two independent networks. The present invention achieves self-supervised learning through mutual supervision between the two networks.

[0073] Specifically, at the k+1th iteration of the self-supervised learning algorithm, the target value of the dual network is updated using the following formula:

[0074]

[0075] Where, is the target value of the Lagrange multiplier of subregion n at the k+1th iteration.

[0076] The loss function used to update the network parameters is as follows:

[0077]

[0078] Among them, λ is the output value of the dual network, λ *(k+1) is the target value of the Lagrange multiplier at the k+1th iteration, n λ is the dimension of λ.

[0079] Step 6: Input the system load data into the dual network and estimate the value of the Lagrange multiplier;

[0080] Step 7: Calculate the original residual and dual residual of this iteration and the previous iteration to determine whether the algorithm has converged. If the algorithm does not converge, return to step 2 and execute the next iteration to recalculate the optimal power flow of each sub-region and update the original network and dual network in turn. If the algorithm converges, the offline training is completed.

[0081] Specifically, at the k+1th iteration of the self-supervised learning algorithm, the original residual and the dual residual Update using the following formula:

[0082]

[0083] The convergence condition of the algorithm is defined as follows:

[0084]

[0085] Among them, ε(k+1) is the residual error of the system, ε * is the tolerance threshold for the given residual.

[0086] In this embodiment, both the original network and the dual network adopt a three-layer structure, the hidden layer size is 256, the error tolerance value ε* is 0.05, and the load fluctuation range of the input data set for training the original network and the dual network is [80%, 120%]. Figure 3 The residual curves during the training of the original network and the dual network in the embodiment of the present invention are shown. In this embodiment, the residual value of the proposed self-supervised learning algorithm converges to below the given ε* after 124 iterations.

[0087] Step 8: During the online solution process, the system load data is input into the original network and the dual network respectively, and the values of the global variables and the Lagrange multipliers are estimated respectively;

[0088] Step 9: Use the estimated global variables and Lagrange multipliers as the boundary variables of the optimal power flow in each sub-region, and then complete the rapid solution of the distributed optimal power flow.

[0089] The original and dual networks of the present invention automatically extract features from system operation data (load data) through alternating learning. The original network estimates global variables based on the load data, and the dual network updates parameters based on the global variables and Lagrange multipliers, achieving collaborative optimization without external annotation.

[0090] In order to further demonstrate the superiority of the method proposed in the present invention, Figure 4 The results of voltage regulation using the proposed method in this example are shown. As can be seen from the figure, the voltage distribution after regulation using the distributed optimal power flow algorithm described in this invention is relatively close to the optimal solution, with only significant deviations at nodes 57, 60, and 66. However, the overall voltage trend remains similar to the optimal solution, demonstrating the feasibility of the proposed method.

[0091] To further illustrate the superiority of the proposed method, Table 1 compares the time required to solve the distributed optimal power flow using the proposed method and the ADMM algorithm. As can be seen from the table, while the ADMM algorithm achieves higher accuracy, it takes significantly longer to solve. The proposed method, on the other hand, achieves faster online solution speeds while maintaining a certain level of accuracy.

[0092] Table 1 Time required for solving distributed optimal power flow using different methods

[0093] The proposed method ADMM algorithm IEEE-123 node system 1.26s 62.54s

[0094] Based on the same technical concept as the method embodiment, the present invention also provides a distributed optimal power flow solution system based on self-supervised learning, including:

[0095] The distribution network partitioning module is used to evaluate the connectivity between nodes based on the physical characteristics of the distribution network topology and the strength of electrical coupling. It also uses community mining theory to divide the distribution network into several sub-regions. Based on the divided sub-regions, the optimal power flow problem of the distribution network is decomposed into optimal power flow sub-problems within each sub-region, and a distributed optimal power flow problem is established.

[0096] The original network update module is used to solve the optimal power flow problem of each sub-region based on the current global variables and the values of the Lagrange multipliers to obtain the values of the boundary variables; based on the current values of the boundary variables of each sub-region and the values of the Lagrange multipliers, the parameters of the original network are updated;

[0097] The dual network update module is used to input the system load data into the original network and estimate the value of the global variable; based on the current value of the global variable and the value of the Lagrange multiplier in the previous iteration, the target value of the Lagrange multiplier is updated, and the parameters of the dual network are updated based on this;

[0098] The training control module is used to input the system load data into the dual network and estimate the value of the Lagrange multiplier. It calculates the original residual and the dual residual between the current iteration and the previous iteration to determine whether the algorithm has converged. If the algorithm does not converge, it executes the next iteration, recalculates the optimal power flow in each sub-region, and updates the original network and the dual network in turn. If the algorithm converges, the offline training is completed.

[0099] The online solution module is used to input the system load data into the original network and the dual network respectively during the online solution process, and estimate the values of global variables and Lagrange multipliers respectively; the estimated values of global variables and Lagrange multipliers are used as the boundary variables of the optimal power flow in each sub-region, thereby completing the rapid solution of the distributed optimal power flow.

[0100] It should be understood that the distributed optimal power flow solution system based on self-supervised learning in the embodiment of the present invention can implement all the technical solutions in the above method embodiment, and the functions of its various functional modules can be specifically implemented according to the methods in the above method embodiment. The specific implementation process can refer to the relevant description in the above embodiment, and will not be repeated here.

[0101] The present invention also provides an electronic device comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and when the programs are executed by the processors, the steps of the distributed optimal power flow solution method based on self-supervised learning as described above are implemented.

[0102] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the distributed optimal power flow solution method based on self-supervised learning as described above.

[0103] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the distributed optimal power flow solving method based on self-supervised learning as described above.

[0104] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, apparatus (systems), computer devices, or computer program products. Thus, 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 implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0105] The present invention is described with reference to flowcharts of methods according to embodiments of the present invention. It should be understood that each process in the flowcharts and combinations of processes in the flowcharts can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts. Figure 1 A device that specifies functions in a process or multiple processes.

[0106] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A function specified in a process or multiple processes.

[0107] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 The steps of a specified function in a process or multiple processes.

Claims

1. A distributed optimal power flow solution method based on self-supervised learning, characterized in that: The following steps are involved: Based on the physical characteristics of the distribution network topology, the connectivity between nodes is evaluated using the electrical coupling strength. The distribution network is divided into several sub-regions using community mining theory. The optimal power flow problem of the distribution network is decomposed into optimal power flow sub-problems within each sub-region according to the divided sub-regions, and a distributed optimal power flow problem is established. Based on the current values of global variables and Lagrange multipliers, the optimal power flow problem of each sub-region is solved separately to obtain the value of the boundary variable; based on the current values of the boundary variables of each sub-region and the value of the Lagrange multiplier, the parameters of the original network are updated; Input the system load data into the original network and estimate the value of the global variable; According to the value of the current global variable and the value of the Lagrange multiplier of the previous iteration, the target value of the Lagrange multiplier is updated, and the parameters of the dual network are updated based on this; Input the system load data into the dual network and estimate the value of the Lagrange multiplier; Calculate the original residual and dual residual of this iteration and the previous iteration to determine whether the algorithm has converged. If the algorithm does not converge, execute the next iteration, recalculate the optimal power flow of each sub-region, and update the original network and dual network in turn. If the algorithm converges, the offline training is completed. During the online solution process, the system load data is input into the original network and the dual network respectively, and the values of global variables and Lagrange multipliers are estimated respectively. The estimated values of global variables and Lagrange multipliers are used as the boundary variables of the optimal power flow in each sub-region, thereby completing the rapid solution of the distributed optimal power flow.

2. The method according to claim 1, characterized in that Based on the physical characteristics of the distribution network topology, the connectivity between nodes is evaluated using the electrical coupling strength, and the community mining theory is used to divide the distribution network into several sub-areas, including: Using the electrical coupling strength e between node i and node j ij To evaluate the connectivity between nodes, we use community mining theory and take modularity Q maximization as the goal to divide the distribution network into several sub-areas, which can be expressed as follows: Where E is the element e ij The symmetric matrix composed of i represents the sum of the elements in the i-th row of matrix E, C i Indicates the region to which node i belongs. When C i =C j When δ(C i C j )=1, otherwise, δ(C i C j )=0.

3. The method according to claim 1, characterized in that Both the original network and the dual network use a deep neural network as the architecture, including an input layer, several hidden layers, an activation layer, and an output layer. The input data are the active load and reactive load u=[P Load ;Q Load ], the output data are the value of the global variable z and the value of the Lagrange multiplier λ respectively.

4. The method according to claim 3, characterized in that The loss function used to update the original network parameters is as follows: Where N is the set of all sub-areas of the distribution network, is the boundary variable of subregion n at the k+1th iteration, M n for The matching matrix with z is a 0-1 matrix; is the Lagrange multiplier of subregion n at the k+1th iteration, and ρ is the penalty factor of the alternating direction multiplier method.

5. The method according to claim 4, characterized in that The objective value of the dual network is updated using the following formula: Where, is the target value of the Lagrange multiplier of subregion n at the k+1th iteration; The loss function used to update the dual network parameters is as follows: Among them, λ is the output value of the dual network, λ *(k+1) is the target value of the Lagrange multiplier at the k+1th iteration, n λ is the dimension of λ.

6. The method according to claim 5, characterized in that Raw residuals and the dual residual Update using the following formula: The convergence condition of the algorithm is defined as follows: Among them, ε (k+1) is the residual error of the system, ε * is the tolerance threshold for the given residual.

7. A distributed optimal power flow solution system based on self-supervised learning, characterized in that: include: The distribution network partitioning module is used to evaluate the connectivity between nodes based on the physical characteristics of the distribution network topology and the strength of electrical coupling. It also uses community mining theory to divide the distribution network into several sub-regions. Based on the divided sub-regions, the optimal power flow problem of the distribution network is decomposed into optimal power flow sub-problems within each sub-region, and a distributed optimal power flow problem is established. The original network update module is used to solve the optimal power flow problem of each sub-region based on the current global variables and the values of the Lagrange multipliers to obtain the values of the boundary variables; based on the current values of the boundary variables of each sub-region and the values of the Lagrange multipliers, the parameters of the original network are updated; The dual network update module is used to input the system load data into the original network and estimate the value of the global variable; According to the value of the current global variable and the value of the Lagrange multiplier of the previous iteration, the target value of the Lagrange multiplier is updated, and the parameters of the dual network are updated based on this; The training control module is used to input the system load data into the dual network and estimate the value of the Lagrange multiplier. It calculates the original residual and the dual residual between the current iteration and the previous iteration to determine whether the algorithm has converged. If the algorithm does not converge, it executes the next iteration, recalculates the optimal power flow in each sub-region, and updates the original network and the dual network in turn. If the algorithm converges, the offline training is completed. The online solution module is used to input the system load data into the original network and the dual network respectively during the online solution process, and estimate the values of global variables and Lagrange multipliers respectively; the estimated values of global variables and Lagrange multipliers are used as the boundary variables of the optimal power flow in each sub-region, thereby completing the rapid solution of the distributed optimal power flow.

8. An electronic device, characterized in that: include: one or more processors; Memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and when the programs are executed by the processors, the distributed optimal power flow solution method based on self-supervised learning as described in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the distributed optimal power flow solution method based on self-supervised learning according to any one of claims 1 to 6 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the distributed optimal power flow solution method based on self-supervised learning according to any one of claims 1 to 6 is implemented.

Images