Model parameter identification method and system based on improved deep reinforcement learning

Abstract

The invention discloses a model parameter identification method and system based on improved deep reinforcement learning, and the method comprises the steps: initializing to generate a plurality of random parameter vectors, and carrying out the iteration; in each iteration, the parameter vector with the highest novelty is selected for updating according to the novelty of the parameter vector; performing variation on the selected parameter vector to generate a plurality of variation vectors, and determining the updating direction of the parameter vector by the variation vectors; carrying out gradient estimation on the parameter vector of each variation, and calculating the fitness of the parameter vector; calculating an overall gradient by combining the energy gradient and the novelty gradient; and updating the parameter vector according to the total gradient obtained by calculation, and repeating the iteration process until the parameter vector with the minimum estimation residual error is obtained. According to the method, the uncertainty and environment change in the parameter identification process can be better processed, the local optimal solution is effectively avoided, the learning speed is increased, and the learning efficiency and the robustness of the strategy are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network dynamic simulation, and specifically to a model parameter identification method and system based on improved deep reinforcement learning. Background Art

[0002] With the transformation and upgrading of the global energy structure and the widespread application of new energy sources such as wind energy and solar energy, the reliability and stability research of new distribution networks with a high proportion of new energy integration has received widespread attention. In this context, the dynamic simulation of distribution networks has become an important tool to ensure the safety of power grid operation, optimize power grid structure and dispatching strategies. The accuracy of dynamic simulation models depends to a large extent on the accurate identification of model parameters, which is directly related to the reliability and practicality of simulation results.

[0003] Traditional model parameter identification methods mainly include physics-based methods and data-driven methods. Physical methods usually require detailed physical information of the system, but in practical applications, this information is often difficult to obtain or incomplete. As for data-driven methods, in traditional methods, the least squares method is widely used in parameter identification, and the parameters are estimated by minimizing the sum of squared errors between the model output and the actual output; the gradient descent method is suitable for problems with a large parameter space, and the parameters are updated iteratively to minimize the error. In intelligent optimization algorithms, genetic algorithms optimize the solution of the problem by simulating natural selection and genetic mechanisms; particle swarm optimization (PSO): by simulating the hunting behavior of bird flocks, it finds the optimal solution. Although the above methods simplify the complexity of parameter identification to a certain extent, they usually require a large amount of historical data and may perform poorly when dealing with high-dimensional data and nonlinear problems.

[0004] In recent years, deep learning technology has been widely studied due to its advantages in high-dimensional data processing and feature extraction. The successful application of deep learning in image recognition, speech processing and other fields has inspired its application in parameter identification of power system dynamic models. Summary of the invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] Therefore, the technical problem solved by the present invention is that under the existing methods, severe nonlinearity, high non-convexity and a large number of parameters pose a major challenge to the identification of parameters of the active distribution network dynamic simulation model. Deep learning models usually require a large amount of labeled data and a long training time, and the model has poor interpretability, which limits its application in power systems where safety is critical.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions: a model parameter identification method based on improved deep reinforcement learning, comprising: establishing a dynamic simulation model of a distribution network, and obtaining a parameter vector in the distribution network;

[0008] The parameter vector is used as the first iteration parameter of the improved deep reinforcement learning algorithm;

[0009] By improving the deep reinforcement learning algorithm, the parameter vector with the minimum estimated residual is obtained;

[0010] Improve the deep reinforcement learning algorithm, initialize and generate M random parameter vectors, and iterate. In each iteration, select the parameter vector with the highest novelty according to the novelty of the parameter vector Make updates;

[0011] For the selected parameter vector Perform mutation, generate multiple mutation vectors, and determine the parameter vector by the mutation vector Update direction, for each mutated parameter vector Perform gradient estimation, calculate its fitness, combine the gradient and novelty gradient, and calculate The overall gradient of

[0012] The parameter vector is updated according to the calculated overall gradient, and the iterative process is repeated until the parameter vector with the minimum estimated residual is obtained.

[0013] As a preferred solution of the model parameter identification method based on improved deep reinforcement learning described in the present invention, wherein: the data parameters include generator parameters, transformer parameters, transmission line parameters and load parameters;

[0014] The initialization includes initializing the acquired data parameters to generate M random parameter vectors and constructing a meta-group S, which is expressed as:

[0015]

[0016] Here, t represents the number of iterations.

[0017] As a preferred solution of the model parameter identification method based on improved deep reinforcement learning described in the present invention, wherein: the parameter vector with the highest novelty is selected Performing the update includes, at each iteration t, determining which parameter vector in the meta-population S is to be updated according to the novelty of the parameter vector;

[0018] Novelty is assessed by the Euclidean distance from one vector to another in the latest archive;

[0019] Based on the current parameter vector archive A, for each parameter vector in the set S The novelty evaluation is calculated as follows:

[0020]

[0021] Among them, 1≤k≤M, kVN stands for k nearest neighbor algorithm, which is used to select representative parameter vectors in A to evaluate The novelty of The most similar parameter vector; Represents the parameter vector novelty;

[0022] For each parameter vector in S, calculate the novelty score that determines the probability of being selected as an update for:

[0023]

[0024] Scoring by novelty Select the parameter vector with the highest novelty

[0025] As a preferred solution of the model parameter identification method based on improved deep reinforcement learning described in the present invention, wherein: the selected parameter vector The mutations include Apply Gaussian noise, the formula is:

[0026]

[0027] Where σ represents the fixed noise standard deviation, represents random noise and follows a normal distribution; N represents the N-dimensional identity matrix.

[0028] As a preferred solution of the model parameter identification method based on improved deep reinforcement learning described in the present invention, wherein: the calculation of its fitness includes calculating the power residual caused by the mismatch between the estimated parameters and the actual parameters

[0029]

[0030] Wherein, Y represents the power measurement vector, V represents the voltage measurement vector, and f represents the mathematical expression function;

[0031] Through Count down to get rewards The formula is:

[0032]

[0033] The performance gradient of is calculated by taking the sum of the perturbations of the sampled parameter vector weighted by the reward, as:

[0034]

[0035] Among them, α represents the learning rate, represents the random reward experienced in a complete iteration of multiple mutation vector interactions;

[0036] The novelty gradient of is calculated as:

[0037] in, Expressed as The novelty gradient.

[0038] As a preferred solution of the model parameter identification method based on improved deep reinforcement learning described in the present invention, wherein: the calculation The overall gradient includes balancing development and exploration by introducing time interval dynamic weights, updating The overall gradient calculation formula is:

[0039]

[0040] in, express The overall gradient of .

[0041] As a preferred solution of the model parameter identification method based on improved deep reinforcement learning described in the present invention, wherein: the updating parameter vector includes obtaining After that, The iterative update formula is:

[0042]

[0043] in, Represents the iterative update Will Add to archive A and update the existing parameter vector space;

[0044] Repeat the iterative process, calculate the estimated residuals, and obtain the minimum model parameter vector, the formula is:

[0045]

[0046] Where Y represents the active / reactive power measurement vector, θ represents the identified parameter vector, and V represents the voltage measurement vector; l represents the residual calculation; f(·) represents the mathematical expression of the model, which is used to map the parameter vector θ and the voltage measurement vector V to the power measurement vector Y; ∥·∥2 represents the 2-norm of l.

[0047] A model parameter identification system based on improved deep reinforcement learning, characterized in that: it includes:

[0048] Simulation model building module: building a dynamic simulation model of the distribution network;

[0049] Initialization module: Initialize and generate M random parameter vectors for iteration;

[0050] Parameter update module: It includes parameter vector perturbator, performance gradient estimator, novelty gradient estimator and parameter updater;

[0051] The parameter vector perturbator: In each iteration, to perform the evolution, the parameter vector θ identified in the previous iteration is used t and the random variance vector ε t , to generate multiple mutation parameter vectors θ` t , and θ` t and ε t The results are sent to the performance gradient estimator and the novelty gradient estimator to estimate the performance gradient and the novelty gradient respectively;

[0052] The performance gradient estimator: uses the θ produced by the parameter vector perturbator t and ε t , the performance gradient estimator determines θ t The direction in which the performance gradient should be moved Passed to the parameter updater;

[0053] The novelty gradient estimator: by using the generated θ` t and ε t Estimating novelty gradients to perform exploration and guide parameter identifiers to explore unvisited parameter spaces, is sent to the parameter updater;

[0054] The parameter updater: assigns dynamic weights to the performance gradient and the novelty gradient, and calculates θ t+1 And add it to the parameter vector archive, update the explored parameter space, and repeat the iterative process until the parameter vector with the minimum estimated residual is obtained.

[0055] Beneficial effects of the invention: The algorithm proposed in the invention follows the gradient in the parameter space, and moves in the direction of both showing new behaviors and obtaining high rewards. It can better handle the uncertainty and environmental changes in the parameter identification process, effectively avoid local optimal solutions, speed up the learning speed, and significantly improve the learning efficiency and robustness of the strategy, especially its ability to handle complex model structures with many parameters, and its robustness to the nonlinear characteristics of the model, while improving accuracy and computational efficiency. In the field of dynamic simulation of distribution networks, it can adapt to the rapid development of distributed new energy and changes in the distribution network form under the new situation. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the figures required for use in the description of the embodiments will be briefly introduced below. Obviously, the figures described below are only some embodiments of the present invention. For ordinary technicians in this field, other figures can be obtained based on these figures without paying creative work.

[0057] Figure 1 A flowchart of a model parameter identification method based on improved deep reinforcement learning provided for the first embodiment of the present invention. DETAILED DESCRIPTION

[0058] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the specification drawings. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0059] Example 1, reference Figure 1 , is an embodiment of the present invention, and provides a model parameter identification method based on improved deep reinforcement learning, comprising:

[0060] S1: Improve the deep reinforcement learning algorithm and iterate according to the acquired data parameters.

[0061] The data parameters include distribution network data parameters such as generator parameters, transformer parameters, transmission line parameters and load parameters.

[0062] Initialize and generate M random parameter vectors, which will be updated in each iteration, and only one vector will be probabilistically selected for update in each iteration.

[0063] Construct the meta-group S, expressed as:

[0064]

[0065] Here, t represents the number of iterations.

[0066] It should be noted that the purpose of constructing the meta-population is to enhance additional diversity. M and the remaining tuning parameters were determined using network search and cross-validation.

[0067] S2: Start iteration. In each iteration, according to the novelty of the parameter vector, select the parameter vector with the highest novelty. to update.

[0068] At each iteration t, which parameter vector in the meta-population S will be updated is determined based on its novelty.

[0069] Novelty is assessed by the Euclidean distance from one vector to another in the latest archive.

[0070] Based on the current parameter vector archive A, for each parameter vector in the set S The novelty evaluation is calculated as follows:

[0071]

[0072] Among them, 1≤k≤M, kVN stands for k nearest neighbor algorithm, which is used to select representative parameter vectors in A to evaluate The novelty of The most similar parameter vector; Represents the parameter vector The novelty.

[0073] It should be noted that, in theory, a smaller k can introduce higher distance variance, while a larger k means higher computational cost. The appropriate k value is determined through numerical experiments.

[0074] For each parameter vector in S, calculate the novelty score that determines the probability of being selected as an update for:

[0075]

[0076] Scoring by novelty Selecting parameter vectors with high novelty enables directed or guided exploration, so the parameter vector with the highest novelty is selected

[0077] S3: For the selected parameter vector Perform mutation, generate multiple mutation vectors, and determine the parameter vector by the mutation vector Update direction.

[0078] The mutation vector is distributed in In order to obtain N mutation vectors in the neighborhood of Apply Gaussian noise, the formula is:

[0079]

[0080] Where σ represents the fixed noise standard deviation, represents random noise and follows a normal distribution; N represents the N-dimensional identity matrix.

[0081] S4: parameter vector for each mutation Perform gradient estimation and calculate its fitness.

[0082] Calculate the power residual caused by the mismatch between the estimated parameters and the actual parameters

[0083]

[0084] Wherein, Y represents the power measurement vector, V represents the voltage measurement vector, and f represents the mathematical expression function;

[0085] Through Count down to get rewards The formula is:

[0086]

[0087] The performance gradient of is approximated by taking a sum of perturbations of the sampled parameter vector weighted by the reward, as:

[0088]

[0089] Among them, α represents the learning rate, represents the random reward experienced in a complete iteration of multiple mutation vector interactions; this means that the performance gradient depends on multiple mutation vectors, which can greatly avoid the high variance brought by a single mutation vector.

[0090] For the novelty gradient, first, calculate each perturbation vector The novelty, then, The novelty gradient of is approximately:

[0091] in, Expressed as The novelty gradient.

[0092] S5: Combine the gradient and novelty gradient to calculate The overall gradient of .

[0093] By using energy gradient and novelty gradient, the time interval dynamic weight ω is introduced t To balance development and exploration, therefore, it is necessary to update The overall gradient calculation formula is:

[0094]

[0095] in, express The overall gradient of .

[0096] S6: Update the parameter vector according to the calculated overall gradient, and repeat the iterative process until the parameter vector with the minimum estimated residual is obtained.

[0097] Get After that, The iterative update formula is:

[0098]

[0099] in, Represents the iterative update Will Add to archive A and update the existing parameter vector space;

[0100] It should be noted that the update mechanism combines the energy gradient and the novelty gradient. By dynamically adjusting the parameters, the model can achieve a balance between development and exploration and avoid falling into the local optimum. At the same time, the introduction of the learning rate further controls the update pace and ensures the stability and convergence of the training process.

[0101] Repeat the iterative process, calculate the estimated residuals, and obtain the minimum model parameter vector, the formula is:

[0102]

[0103] Where Y is the active / reactive power measurement vector, θ represents the identified parameter vector, and V is the voltage measurement vector; l represents the residual calculation, f(·) represents the mathematical expression of the model, which is used to map the parameter vector θ and the voltage measurement vector V to the power measurement vector Y; ∥·∥2 represents the 2-norm of l, also known as the Euclidean norm.

[0104] It should be noted that by calculating the residual of each iteration, it is ensured that the parameter vector finally obtained is the optimal parameter of the model, so that the model has the highest fit to the actual system.

[0105] The above embodiments also include a model parameter identification system based on improved deep reinforcement learning, specifically:

[0106] Simulation model building module: building a dynamic simulation model of the distribution network;

[0107] Initialization module: Initialize and generate M random parameter vectors for iteration;

[0108] Parameter update module: It includes parameter vector perturbator, performance gradient estimator, novelty gradient estimator and parameter updater;

[0109] Parameter vector perturbator: In each iteration, to perform the evolution, the parameter vector θ identified in the previous iteration is used t and the random variance vector ε t , to generate multiple mutation parameter vectors θ` t , and θ` t and ε t The results are sent to the performance gradient estimator and the novelty gradient estimator respectively to approximately estimate the performance gradient and the novelty gradient;

[0110] Performance Gradient Estimator: Uses the θ produced by the parameter vector perturbator t and ε t , the performance gradient estimator determines θ t The direction in which the performance gradient should be moved Passed to the parameter updater;

[0111] Novelty Gradient Estimator: By using the generated θ` t and ε t Estimating novelty gradients to perform exploration and guide parameter identifiers to explore unvisited parameter spaces, is sent to the parameter updater;

[0112] Parameter Updater: Assigns dynamic weights to the approximate performance gradient and novelty gradient, and computes θ t+1 And add it to the parameter vector archive, update the explored parameter space, and repeat the iterative process until the parameter vector with the minimum estimated residual is obtained.

[0113] The computer device may be a server. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data cluster data of the power monitoring system. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a model parameter identification method based on improved deep reinforcement learning is implemented.

[0114] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0115] Example 2 is an embodiment of the present invention, which provides a model parameter identification method based on improved deep reinforcement learning. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation / comparative experiments.

[0116] In order to verify the effectiveness of the model parameter identification method based on improved deep reinforcement learning, a simulation experiment was conducted. The experimental object is a typical power system model, including generators, transformers, transmission lines and loads. The purpose of the experiment is to compare the performance of the method of the present invention with the traditional method in terms of identification accuracy and convergence speed through parameter identification of the system.

[0117] First, the experimental environment is constructed. The experimental platform is modeled and simulated using MATLAB / Simulink. The system parameters include active power, reactive power, voltage and other measured values. The initial system parameters are obtained through actual measurements, and a certain random error is introduced to simulate the uncertainty in actual applications.

[0118] The experimental results are compared with the traditional method, which uses fixed weights and single gradient calculation without novelty gradient.

[0119] The final comparative experimental data is shown in Table 1.

[0120] Table 1 Simulation experiment data table

[0121]

[0122] By comparing the experimental data, it can be seen that the method of the present invention is significantly superior to the traditional method in terms of the convergence speed of the residual and the final residual value. After 10 iterations, the residual of the present invention has dropped by more than 50%, while the residual of the traditional method has only slightly decreased. For example, the residual of the present invention-object 1 dropped from 0.152 to 0.073, while the residual of the traditional method-object 1 only dropped from 0.150 to 0.142.

[0123] After further iterations of 20 and 30 times, the residual of the present invention continued to decrease significantly, while the residual of the traditional method decreased more slowly. For example, the residual of the present invention-object 1 decreased to 0.018 after 30 iterations, while the residual of the traditional method-object 1 only decreased to 0.127. Similarly, the data of other objects also showed similar trends.

[0124] These data fully demonstrate the efficiency and accuracy of the method of the present invention in optimizing power system parameters. By combining gradient and novelty gradient, the present invention can converge to the optimal parameter vector more quickly, effectively reducing the estimation residual. This effect is particularly important in the practical application of power systems, and can significantly improve the stability and operating efficiency of the system.

[0125] In addition, the dynamic weight balancing strategy of the present invention has shown excellent adaptability in experiments and can maintain stable optimization effects under different power system parameters. This further proves the innovation and practical value of the present invention in solving the problem of power system parameter identification. In contrast, the traditional method lacks a dynamic adjustment mechanism during the iteration process, resulting in slow convergence speed and high residual error, which makes it difficult to meet the high-precision requirements of practical applications.

[0126] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A model parameter identification method based on improved deep reinforcement learning, characterized in that: include: Establish a dynamic simulation model of the distribution network and obtain data parameters in the distribution network; Use data parameters as the first iteration parameters of the improved deep reinforcement learning algorithm; By improving the deep reinforcement learning algorithm, the parameter vector with the minimum estimated residual is obtained; Improve the deep reinforcement learning algorithm, initialize and generate M random parameter vectors according to the acquired data parameters, and iterate. In each iteration, select the parameter vector with the highest novelty according to the novelty of the parameter vector Make updates; For the selected parameter vector Perform mutation, generate multiple mutation vectors, and determine the parameter vector by the mutation vector Update direction, for each mutated parameter vector Perform gradient estimation, calculate its fitness, combine the gradient and novelty gradient, and calculate The overall gradient of The parameter vector is updated according to the calculated overall gradient, and the iterative process is repeated until the parameter vector with the minimum estimated residual is obtained.

2. The model parameter identification method based on improved deep reinforcement learning according to claim 1, characterized in that: The data parameters include generator parameters, transformer parameters, transmission line parameters and load parameters; The initialization includes initializing the acquired data parameters to generate M random parameter vectors and constructing a meta-group S, which is expressed as: Here, t represents the number of iterations.

3. The model parameter identification method based on improved deep reinforcement learning according to claim 2, characterized in that: The parameter vector with the highest novelty is selected Performing the update includes, at each iteration t, determining which parameter vector in the meta-population S is to be updated according to the novelty of the parameter vector; Novelty is assessed by the Euclidean distance from one vector to another in the latest archive; Based on the current parameter vector archive A, for each parameter vector in the set S The novelty evaluation is calculated as follows: Among them, 1≤k≤M, kVN stands for k nearest neighbor algorithm, which is used to select representative parameter vectors in A to evaluate The novelty of The most similar parameter vector; Represents the parameter vector novelty; For each parameter vector in S, calculate the novelty score that determines the probability of being selected as an update for: Scoring by novelty Select the parameter vector with the highest novelty 4. The model parameter identification method based on improved deep reinforcement learning according to claim 3, characterized in that: The selected parameter vector The mutations include Apply Gaussian noise, the formula is: Where σ represents the fixed noise standard deviation, represents random noise and follows a normal distribution; N represents the N-dimensional identity matrix.

5. The model parameter identification method based on improved deep reinforcement learning according to claim 4, characterized in that: The calculation of its fitness includes: Calculate the power residual caused by the mismatch between the estimated parameters and the actual parameters Wherein, Y represents the power measurement vector, V represents the voltage measurement vector, and f(·) represents the mathematical expression function; Through Count down to get rewards The formula is: The performance gradient of is calculated by taking the sum of the perturbations of the sampled parameter vector weighted by the reward, as: Among them, α represents the learning rate, represents the random reward experienced in a complete iteration of multiple mutation vector interactions; The novelty gradient of is calculated as: in, Expressed as The novelty gradient.

6. The model parameter identification method based on improved deep reinforcement learning according to claim 5, characterized in that: The calculation The overall gradient includes balancing development and exploration by introducing time interval dynamic weights, updating The overall gradient calculation formula is: in, express The overall gradient of .

7. The model parameter identification method based on improved deep reinforcement learning according to claim 6, characterized in that: The updating parameter vector includes obtaining After that, The iterative update formula is: in, Represents the iterative update Will Add to archive A and update the existing parameter vector space; Repeat the iterative process, calculate the estimated residuals, and obtain the minimum model parameter vector, the formula is: Where Y represents the active / reactive power measurement vector, θ represents the identified parameter vector, and V represents the voltage measurement vector; l represents the residual calculation; f(·) represents the mathematical expression of the model, which is used to map the parameter vector θ and the voltage measurement vector V to the power measurement vector Y; ∥·∥2 represents the 2-norm of l.

8. A model parameter identification system based on improved deep reinforcement learning using the method according to any one of claims 1 to 7, characterized in that: Simulation model building module: building a dynamic simulation model of the distribution network; Initialization module: Initialize and generate M random parameter vectors for iteration; Parameter update module: It includes parameter vector perturbator, performance gradient estimator, novelty gradient estimator and parameter updater; The parameter vector perturbator: In each iteration, to perform the evolution, the parameter vector θ identified in the previous iteration is used t and the random variance vector ε t , to generate multiple mutation parameter vectors θ` t , and θ` t and ε t The results are sent to the performance gradient estimator and the novelty gradient estimator to estimate the performance gradient and the novelty gradient respectively; The performance gradient estimator: uses the θ produced by the parameter vector perturbator t and ε t , the performance gradient estimator determines θ t The direction in which the performance gradient should be moved Passed to the parameter updater; The novelty gradient estimator: by using the generated θ` t and ε t Estimating novelty gradients to perform exploration and guide parameter identifiers to explore unvisited parameter spaces, is sent to the parameter updater; The parameter updater: assigns dynamic weights to the performance gradient and the novelty gradient, and calculates θ t+1 And add it to the parameter vector archive, update the explored parameter space, and repeat the iterative process until the parameter vector with the minimum estimated residual is obtained.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images