Island division and load recovery method, device and equipment for power distribution network

Abstract

The invention provides an island division and load recovery method, device and equipment for a power distribution network. The method comprises the following steps: acquiring real-time state data of the power distribution network; preprocessing the real-time state data to obtain target state data; inputting the target state data into an island division model for processing to obtain an island division result; the island division model is obtained by training historical fault processing data; according to the island division result, utilizing a load recovery model to recover loads in sequence; the load recovery model comprises an evaluation network layer and a target network layer, the evaluation network layer outputs an action decision according to the island state data, the target network layer outputs action value evaluation according to the island state data, and the weight of the evaluation network layer is adjusted according to the action value evaluation. According to the invention, the efficiency and stability of power distribution network island division and load recovery can be improved.

Description

Technical Field

[0001] This invention relates to the field of power supply technology, and in particular to a method, apparatus and equipment for dividing power distribution network into islands and restoring loads. Background Technology

[0002] As a crucial link between the power system and users, the distribution network is prone to large-scale power outages after faults (such as line short circuits or distributed generation (DG) outages). Existing technologies have many shortcomings in distribution network islanding and load restoration. Traditional methods rely heavily on manual experience or simple rules, making them ill-suited to the complex and ever-changing operating scenarios of distribution networks. Optimization algorithms suffer from slow response times (minutes), low partitioning accuracy, low computational efficiency, and a tendency to get trapped in local optima. Load restoration often employs integer programming or heuristic algorithms, resulting in poor real-time performance (requiring tens of seconds to several minutes), failing to meet users' high demands for power supply reliability. Model-based methods are highly dependent on the accuracy of network topology and parameters; when faced with real-time changes in the power grid, model updates lag, leading to significant deviations in islanding results. These limitations constitute the bottlenecks of existing technologies, severely restricting the efficiency of power restoration and the improvement of power supply reliability after distribution network faults. Summary of the Invention

[0003] The technical problem to be solved by this invention is to provide a method, apparatus, and equipment for islanding and load restoration in power distribution networks. This can improve the efficiency and stability of islanding and load restoration in power distribution networks.

[0004] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: A method for islanding and load restoration in a power distribution network includes: Obtain real-time status data of the power distribution network; The real-time status data is preprocessed to obtain the target status data; The target state data is input into the island partitioning model for processing to obtain the island partitioning result; the island partitioning model is trained from historical fault handling data. Based on the islanding results, the load is restored sequentially using a load recovery model. The load recovery model includes an evaluation network layer and a target network layer. The evaluation network layer outputs action decisions based on the islanding status data, and the target network layer outputs action value assessments based on the islanding status data. The weights of the evaluation network layer are adjusted according to the action value assessments.

[0005] Optionally, the real-time state data is preprocessed to obtain target state data, including: The real-time status data is cleaned to obtain the first intermediate data; The first intermediate data is standardized to obtain the second intermediate data; Feature extraction is performed on the second intermediate data to obtain the target state data.

[0006] Optionally, the target state data is input into the island partitioning model for processing to obtain the island partitioning results, including: Based on the target status data, the distribution network structure is constructed to obtain the distribution network diagram; The network diagram is input into the attention layer of the island partitioning model for processing to obtain attention weights; The network node features are updated based on the attention weights to obtain the island partitioning results.

[0007] Optionally, the island partitioning model is trained using historical fault handling data, including: Based on the historical fault handling data, a historical distribution network diagram is obtained; The historical power distribution map is input into the attention layer of a preset learning model for processing to obtain attention training weights; Based on the attention training weights, the feature of the distribution network node is updated to obtain the island partitioning training result; Based on the island partitioning training results and historical fault handling results, the parameters of the preset learning model are adjusted to obtain the island partitioning model.

[0008] Optionally, based on the islanding results, the load is restored sequentially using a load recovery model, including: Based on the island partitioning results, the island state space is obtained; The isolated state space is input into the load recovery model evaluation network layer for processing to obtain action decisions; the weights of the evaluation network layer are obtained based on the action value evaluation results of the target network layer. Based on the action decision, the load is restored.

[0009] Optionally, the isolated state space is input into the load recovery model evaluation network layer for processing to obtain action decisions, including: The isolated state space is input into the first hidden layer of the load recovery model evaluation network layer and processed to obtain the first intermediate output. The second hidden layer of the first intermediate output input load recovery model evaluation network layer is processed to obtain the second intermediate output; The output layer of the second intermediate input-output load recovery model evaluation network layer is processed to obtain the action decision.

[0010] Optionally, the weights of the evaluation network layer are obtained based on the action value evaluation results of the target network layer, including: Based on the loss function between the output of the evaluation network layer and the output of the target network layer, the parameters of the evaluation network layer are adjusted to obtain the weights of the evaluation network layer.

[0011] Embodiments of the present invention also provide a power distribution network islanding and load restoration device, comprising: The acquisition module is used to acquire real-time status data of the power distribution network; The processing module is used to preprocess the real-time status data to obtain target status data; input the target status data into the island partitioning model for processing to obtain the island partitioning result; the island partitioning model is trained from historical fault handling data; according to the island partitioning result, the load is restored sequentially using the load recovery model; the load recovery model includes an evaluation network layer and a target network layer, the evaluation network layer outputs action decisions based on the island status data, the target network layer outputs action value assessments based on the island status data, and the weights of the evaluation network layer are adjusted according to the action value assessments.

[0012] Embodiments of the present invention also provide a computing device, including: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the power distribution network islanding and load restoration method of the present invention.

[0013] Embodiments of the present invention also provide a computer-readable storage medium storing a program that, when executed by a processor, implements the power distribution network islanding and load restoration method described in the present invention.

[0014] The above-described technical solution of the present invention has at least the following technical effects: The above-described power distribution network islanding and load restoration method of the present invention acquires real-time status data of the power distribution network; preprocesses the real-time status data to obtain target status data; inputs the target status data into an islanding model for processing to obtain islanding results; the islanding model is trained from historical fault handling data; and based on the islanding results, a load restoration model is used to restore loads sequentially; the load restoration model includes an evaluation network layer and a target network layer, the evaluation network layer outputs action decisions based on the islanding status data, the target network layer outputs action value assessments based on the islanding status data, and the weights of the evaluation network layer are adjusted according to the action value assessments. This improves the efficiency and stability of power distribution network islanding and load restoration. Attached Figure Description

[0015] Figure 1This is a flowchart illustrating the power distribution network islanding and load restoration method of the present invention; Figure 2 This is a schematic diagram of the power distribution network islanding and load restoration device of the present invention. Detailed Implementation

[0016] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0017] like Figure 1 As shown, an embodiment of the present invention proposes a method for islanding and load restoration in a power distribution network, comprising: Step S1: Obtain real-time status data of the power distribution network; Step S2: Preprocess the real-time status data to obtain the target status data; Step S3: Input the target state data into the island partitioning model for processing to obtain the island partitioning result; the island partitioning model is trained from historical fault handling data. Step S4: Based on the island partitioning results, the load is restored sequentially using the load recovery model. The load recovery model includes an evaluation network layer and a target network layer. The evaluation network layer outputs action decisions based on the island status data, and the target network layer outputs action value assessments based on the island status data. The weights of the evaluation network layer are adjusted according to the action value assessments.

[0018] In this embodiment, as Figure 1 As shown, in the method of islanding and load restoration in power distribution network, firstly, real-time status data of electrical equipment is acquired. Data such as node voltage, line current and distributed generation (DG) output are acquired through power management unit and remote terminal unit. Data such as line connection relationship, switch status and transformer capacity are acquired through acquisition and monitoring system. Data such as load power and user type are acquired through smart meter. Data such as fault location, type and duration are acquired through fault indicator and alarm system. Then, the real-time status data is preprocessed to obtain the target status data; Next, the target state data is input into the island partitioning model for processing to obtain the island partitioning result; the island partitioning model is trained from historical fault handling data. Finally, based on the island partitioning results, the load is restored sequentially using a load recovery model. The load recovery model includes an evaluation network layer and a target network layer. The evaluation network layer outputs action decisions based on the island status data, and the target network layer outputs action value assessments based on the island status data. The weights of the evaluation network layer are adjusted according to the action value assessments.

[0019] The present invention provides a method for islanding and load restoration in power distribution networks, which enables the power grid to output accurate islanding results in milliseconds after a fault, ensuring power balance, voltage and current within the island meet the constraints; load restoration is completed in seconds after islanding, prioritizing the restoration of high-importance loads (such as hospitals, industrial loads, etc.), thereby improving the load restoration rate; the solution of the present invention has generalizability and is adaptable to different fault types (line faults, DG outages) and load fluctuation scenarios.

[0020] In an optional embodiment of the present invention, step S2, preprocessing the real-time state data to obtain target state data, includes: Step S21: Clean the real-time status data to obtain the first intermediate data; Step S22: Standardize the first intermediate data to obtain the second intermediate data; Step S23: Extract features from the second intermediate data to obtain target state data.

[0021] In this embodiment, firstly, the real-time status data is cleaned to obtain first intermediate data; during the data cleaning process, data exceeding the data attribute threshold is deleted, and outliers are replaced with the sample median; missing values ​​are handled using interpolation, and the interpolation formula is:

[0022] in, For missing values, K is The K nearest neighbor samples, , For timestamps.

[0023] Then, the first intermediate data is standardized to obtain the second intermediate data; the dimensions are eliminated using a standardization formula, which is:

[0024] Where sy represents the original data, μ represents the sample mean, and σ represents the sample standard deviation. This is the standardized data.

[0025] Finally, feature extraction is performed on the second intermediate data to obtain the target state data. This mainly involves extracting topological features such as node degree and betweenness centrality, fault correlation features such as fault distance and DG support capacity, and load priority features.

[0026] In an optional embodiment of the present invention, step S3 involves inputting the target state data into an island partitioning model for processing to obtain island partitioning results, including: Step S31: Construct the distribution network structure based on the target state data to obtain the distribution network diagram; Step S32: Input the distribution network map into the attention layer of the island partitioning model for processing to obtain attention weights; Step S33: Update the node features of the distribution network according to the attention weights to obtain the island partitioning results.

[0027] In this embodiment, the graph structure characteristics of the distribution network topology are utilized to focus on critical fault nodes (such as those near the distribution gate generator (DG) or highly important loads) through an attention mechanism, achieving rapid end-to-end islanding. First, the distribution network structure is constructed based on the target state data to obtain the distribution network diagram. The structure of the distribution network diagram is: G = (V, E, X), where G is the distribution network diagram, V is the node set, E is the edge set, and X is the node feature matrix. The node set V includes buses, DGs, loads, and switches (total...). N The edge set E contains the lines connecting adjacent nodes, and the edge features are line impedance and current carrying capacity; the node feature matrix X contains the features of the target state data.

[0028] Then, the distribution network diagram is input into the attention layer of the island partitioning model for processing to obtain attention weights; the attention weights are used to represent the degree of impact of faults on different nodes; to obtain the attention weights, the node correlation degree is first obtained based on the node feature vectors, and the node correlation degree formula is:

[0029] in, This represents the degree of association between node i and node j. and represents the feature vectors of node i and node j respectively; W is the feature mapping weight matrix, which extracts key features related to island partitioning by performing a linear transformation on the input feature vectors. This indicates that the feature vectors of two nodes after transformation by the weight matrix are concatenated; 'a' is the weight vector in the attention mechanism, which is inner producted with the concatenated feature vector to highlight the importance of different feature dimensions; LeakyReLU is a linear unit activation function with leakage correction, used to introduce non-linearity, avoid the gradient vanishing problem during network training, and enable the model to learn more complex node relationships.

[0030] Then, based on the node correlation, the attention weight is obtained; the formula for the attention weight is:

[0031] in, Let i be the set of neighbors of node i. Let be the attention weight of node j on i. The higher the weight, the greater the influence of node j on i.

[0032] Finally, the node features of the distribution network are updated according to the attention weights to obtain the island partitioning results. The node feature update formula is:

[0033] in, Let i be the attention feature of node i. Let W be the ReLU activation function, and W be the feature map weight matrix. Let be the attention weight of node j for node i.

[0034] There are M predefined isolated islands. The "island affiliation label" of the corresponding node is calculated and output. The label value ranges from 0 to M-1, thereby clarifying the island affiliation of each node and obtaining the island division result.

[0035] In an optional embodiment of the present invention, step S3, wherein the island partitioning model is trained from historical fault handling data, includes: Step S34: Obtain the historical distribution network diagram based on the historical fault handling data; Step S35: Input the historical power distribution map into the attention layer of the preset learning model for processing to obtain attention training weights; Step S36: Update the network map node features according to the attention training weights to obtain the island partitioning training results; Based on the island partitioning training results and historical fault handling results, the parameters of the preset learning model are adjusted to obtain the island partitioning model.

[0036] In this embodiment, historical fault handling data is acquired, and labels are generated by combining historical fault scenarios with traditional optimization results. More than 1,000 historical fault scenarios are analyzed. The fault scenarios include faults with different fault locations and different DG outputs. The optimal island division result is obtained and used as a label set. Then, using the island partitioning model and method described above, historical fault handling data is input into the island partitioning model for processing to obtain the island partitioning training results. Based on the loss function between the island partitioning training results and the historical fault handling results, the parameters of the preset learning model are adjusted to obtain the island partitioning model. The loss function expression is:

[0037] Where L represents the loss function, N is the number of samples, and C is the number of classes. It is a sample i Category j Historical fault handling results It is a sample i Category j The training results are used to divide the islands into smaller parts. Training is stopped if the loss function does not decrease for five consecutive rounds to avoid overfitting.

[0038] In an optional embodiment of the present invention, step S4, based on the islanding results, utilizes a load recovery model to sequentially restore the load, including: Step S41: Based on the island division results, obtain the island state space; Step S42: The isolated state space is input into the load recovery model evaluation network layer for processing to obtain action decisions; the weights of the evaluation network layer are obtained based on the action value evaluation results of the target network layer. Step S43: Based on the action decision, restore the load.

[0039] In this embodiment, load recovery is performed by continuously learning the optimal recovery strategy, prioritizing the recovery of highly important loads. First, based on the islanding results, the island state space is obtained; the current state space within an island is represented as:

[0040] Where S is the island state space, Available output for distributed power sources The total power of the restored load, For voltage deviation, For current deviation, This represents the number of times the switch has been operated.

[0041] Then, the isolated state space is input into the load recovery model evaluation network layer for processing to obtain action decisions; the weights of the evaluation network layer are obtained based on the action value evaluation results of the target network layer. Based on the action decision, the load is restored. By processing the current islanded state space through the load restoration model, the action with the highest action value calculated by the evaluation network layer is obtained, executed, and the state is updated. This process is repeated until there is no load to restore or the upper limit of the restoration time is reached, and the final restoration result is output, such as restoring loads 1, 3, and 5, with a total restoration rate of 92%.

[0042] In an optional embodiment of the present invention, step S42 involves processing the isolated state space input load recovery model evaluation network layer to obtain action decisions, including: Step S421: Process the first hidden layer of the isolated state space input load recovery model evaluation network layer to obtain the first intermediate output; Step S422: Process the second hidden layer of the first intermediate output input load recovery model evaluation network layer to obtain the second intermediate output; Step S423: Process the output layer of the second intermediate input-output load recovery model evaluation network layer to obtain the action decision.

[0043] In this embodiment, the data from the isolated state space is passed to the first hidden layer of 128 neurons through a weight matrix and an activation function. In this layer, the data is weighted and summed, then processed by an activation function (such as the ReLU function) to introduce nonlinear factors and extract preliminary features from the state data. The calculation formula is as follows:

[0044] in, Here is the weight matrix of the first hidden layer. For bias vectors, For the isolated state space, This is the first intermediate output of the first hidden layer.

[0045] Then, the second hidden layer of the first intermediate output input load recovery model evaluation network layer is processed to obtain the second intermediate output; Output of the first hidden layer The data is then passed to the second hidden layer with 64 neurons, where it undergoes the same weighted summation and activation function processing to further extract higher-level features. The calculation formula is as follows:

[0046] in, This is the weight matrix of the second hidden layer. For bias vectors, This is the second intermediate output of the second hidden layer.

[0047] Finally, the output layer of the second intermediate input-output load recovery model evaluation network layer is processed to obtain the action decision.

[0048] The output of the second hidden layer The input is fed into an output layer of dimension n (where n is the number of actions), and the action value (Q-value) for each action is obtained through weighted summation. The calculation formula is as follows:

[0049] Where Q represents the action value. The weight matrix of the output layer. This is the bias vector.

[0050] Finally, the Q-value of each action output by the network layer is evaluated to guide the decision-making process, selecting the action with the largest Q-value to be executed.

[0051] In an optional embodiment of the present invention, step S42, in which the weights of the evaluation network layer are obtained based on the action value evaluation result of the target network layer, includes: Step S424: Adjust the parameters of the evaluation network layer according to the loss function between the output of the evaluation network layer and the output of the target network layer to obtain the weights of the evaluation network layer.

[0052] In this embodiment, the target network layer structure is consistent with the evaluation network layer. The parameters of the evaluation network layer are copied periodically (e.g., every 100 steps) to avoid training oscillations. Its input data is usually the current state information of the distribution network (such as multi-dimensional feature vectors such as node voltage, line power flow, and load power) and action information (such as switch state adjustment schemes). The output result is the value assessment of the current state-action pair, that is, the prediction of the long-term cumulative benefits that the action can bring in the future, which is used to guide the optimization strategy of the evaluation network layer.

[0053] Based on the loss function between the output of the evaluation network layer and the output of the target network layer, the parameters of the evaluation network layer are adjusted to obtain the weights of the evaluation network layer. The loss function is:

[0054] in, To evaluate network layer parameters, S represents the island state space. The discount factor is used to balance the importance of current rewards and future rewards; r is the reward. For the output of the target network layer, To evaluate the network layer output, Here are the parameters for the target network layer, and s represents the current state. Let 'a' be the next state and 'a' be the current action. For the next action.

[0055] The loss function is calculated through backpropagation. For evaluating network layer parameters gradient And use stochastic gradient descent to update the parameters, the update expression is:

[0056] in, The learning rate controls the step size for parameter updates. The loss function is minimized through iterative iteration. This allows the output of the evaluation network layer to more accurately approximate the output of the target network layer, thereby optimizing the performance of the evaluation network in the decision-making process.

[0057] The solution of this invention divides the power distribution network into islands using an islanding model and restores highly important loads sequentially and quickly using a load restoration model. This overcomes the bottlenecks of poor real-time performance and low accuracy of traditional methods, achieving the effect of "millisecond-level division and second-level restoration" after a distribution network fault.

[0058] like Figure 2 As shown, an embodiment of the present invention also provides a power distribution network islanding and load restoration device 20, comprising: Module 21 is used to acquire real-time status data of the power distribution network; Processing module 22 is used to preprocess the real-time status data to obtain target status data; input the target status data into the island partitioning model for processing to obtain island partitioning results; the island partitioning model is trained from historical fault handling data; according to the island partitioning results, the load is restored sequentially using a load recovery model; the load recovery model includes an evaluation network layer and a target network layer, the evaluation network layer outputs action decisions based on the island status data, the target network layer outputs action value assessments based on the island status data, and the weights of the evaluation network layer are adjusted according to the action value assessments.

[0059] Optionally, the real-time state data is preprocessed to obtain target state data, including: The real-time status data is cleaned to obtain the first intermediate data; The first intermediate data is standardized to obtain the second intermediate data; Feature extraction is performed on the second intermediate data to obtain the target state data.

[0060] Optionally, the target state data is input into the island partitioning model for processing to obtain the island partitioning results, including: Based on the target status data, the distribution network structure is constructed to obtain the distribution network diagram; The network diagram is input into the attention layer of the island partitioning model for processing to obtain attention weights; The network node features are updated based on the attention weights to obtain the island partitioning results.

[0061] Optionally, the island partitioning model is trained using historical fault handling data, including: Based on the historical fault handling data, a historical distribution network diagram is obtained; The historical power distribution map is input into the attention layer of a preset learning model for processing to obtain attention training weights; Based on the attention training weights, the feature of the distribution network node is updated to obtain the island partitioning training result; Based on the island partitioning training results and historical fault handling results, the parameters of the preset learning model are adjusted to obtain the island partitioning model.

[0062] Optionally, based on the islanding results, the load is restored sequentially using a load recovery model, including: Based on the island partitioning results, the island state space is obtained; The isolated state space is input into the load recovery model evaluation network layer for processing to obtain action decisions; the weights of the evaluation network layer are obtained based on the action value evaluation results of the target network layer. Based on the action decision, the load is restored.

[0063] Optionally, the isolated state space is input into the load recovery model evaluation network layer for processing to obtain action decisions, including: The isolated state space is input into the first hidden layer of the load recovery model evaluation network layer and processed to obtain the first intermediate output. The second hidden layer of the first intermediate output input load recovery model evaluation network layer is processed to obtain the second intermediate output; The output layer of the second intermediate input-output load recovery model evaluation network layer is processed to obtain the action decision.

[0064] Optionally, the weights of the evaluation network layer are obtained based on the action value evaluation results of the target network layer, including: Based on the loss function between the output of the evaluation network layer and the output of the target network layer, the parameters of the evaluation network layer are adjusted to obtain the weights of the evaluation network layer.

[0065] It should be noted that all implementation methods in the above method embodiments are applicable to the embodiments of this device and can achieve the same technical effect.

[0066] Embodiments of the present invention also provide a computing device, including: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the power distribution network islanding and load restoration method of the present invention. All implementations in the above method embodiments are applicable to the embodiments of this computing device and can achieve the same technical effects.

[0067] Embodiments of the present invention also provide a computer-readable storage medium storing a program that, when executed by a processor, implements the power distribution network islanding and load restoration method described in this invention. All implementations in the above method embodiments are applicable to the embodiments using this computer-readable storage medium and achieve the same technical effects.

[0068] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0069] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0070] In the embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0071] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0072] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0073] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0074] Furthermore, it should be noted that in the apparatus and method of the present invention, it is obvious that the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered equivalent solutions of the present invention. Moreover, the steps performing the above series of processes can naturally be executed in the order described, but are not necessarily required to be executed in chronological order; some steps can be executed in parallel or independently of each other. Those skilled in the art will understand that all or any step or component of the method and apparatus of the present invention can be implemented in any computing device (including processors, storage media, etc.) or network of computing devices, in hardware, firmware, software, or a combination thereof. This is something that those skilled in the art can achieve by using their basic programming skills after reading the description of the present invention.

[0075] Therefore, the object of the present invention can also be achieved by running a program or a set of programs on any computing device. The computing device can be a known general-purpose device. Therefore, the object of the present invention can also be achieved simply by providing a program product containing program code for implementing the method or apparatus. That is, such a program product also constitutes the present invention, and the storage medium storing such a program product also constitutes the present invention. Obviously, the storage medium can be any known storage medium or any storage medium developed in the future. It should also be noted that in the apparatus and method of the present invention, it is obvious that the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered equivalent to the present invention. Furthermore, the steps for performing the above series of processes can naturally be performed in the order described, but are not necessarily required to be performed in chronological order. Some steps can be performed in parallel or independently of each other.

[0076] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A power distribution network island division and load restoration method, characterized in that, The method comprises the following steps: acquiring real-time state data of a power distribution network; preprocessing the real-time state data to obtain target state data; inputting the target state data into an island division model for processing to obtain an island division result; the island division model is trained by historical fault handling data; according to the island division result, a load recovery model is used to sequentially recover the load; the load recovery model comprises an evaluation network layer and a target network layer, the evaluation network layer outputs an action decision according to island state data, the target network layer outputs an action value evaluation according to island state data, and the weight of the evaluation network layer is adjusted according to the action value evaluation. 2.The power distribution network islanding division and load restoration method of claim 1, wherein, The preprocessing of the real-time state data to obtain the target state data comprises the following steps: performing data cleaning on the real-time state data to obtain first intermediate data; performing standardization processing on the first intermediate data to obtain second intermediate data; performing feature extraction on the second intermediate data to obtain the target state data. 3.The power distribution network islanding division and load restoration method of claim 1, wherein, The processing of the target state data in the island division model to obtain the island division result comprises the following steps: constructing a distribution network structure according to the target state data to obtain a distribution network graph; inputting the distribution network graph into an attention layer of the island division model for processing to obtain an attention weight; updating a node feature of the distribution network graph according to the attention weight to obtain the island division result.

4. The power distribution network islanding and load restoration method of claim 1, wherein, The island division model is trained by historical fault handling data, which comprises the following steps: obtaining a historical distribution network graph according to the historical fault handling data; inputting the historical distribution network graph into an attention layer of a preset learning model for processing to obtain an attention training weight; updating a node feature of the distribution network graph according to the attention training weight to obtain an island division training result; adjusting parameters of the preset learning model according to the island division training result and a historical fault handling result to obtain the island division model.

5. The power distribution network islanding and load restoration method of claim 1, wherein, According to the island division result, a load recovery model is used to sequentially recover the load, which comprises the following steps: obtaining an island state space according to the island division result; inputting the island state space into an evaluation network layer of the load recovery model for processing to obtain an action decision; the weight of the evaluation network layer is obtained according to an action value evaluation result of a target network layer; recovering the load according to the action decision.

6. The power distribution network islanding and load restoration method of claim 5, wherein, The processing of the island state space in the evaluation network layer of the load recovery model to obtain the action decision comprises the following steps: inputting the island state space into a first hidden layer of the evaluation network layer of the load recovery model for processing to obtain a first intermediate output; inputting the first intermediate output into a second hidden layer of the evaluation network layer of the load recovery model for processing to obtain a second intermediate output; inputting the second intermediate output into an output layer of the evaluation network layer of the load recovery model for processing to obtain the action decision.

7. The power distribution network islanding and load restoration method of claim 5, wherein, The weight of the evaluation network layer is obtained according to the action value evaluation result of the target network layer, which comprises the following steps: adjusting parameters of the evaluation network layer according to a loss function between an output result of the evaluation network layer and an output result of the target network layer to obtain the weight of the evaluation network layer.

8. A power distribution network island division and load restoration device, characterized in that, The method comprises the following steps: An acquisition module is configured to acquire power distribution network real-time state data. A processing module is configured to preprocess the real-time state data to obtain target state data. The target state data is input into an island division model for processing to obtain an island division result, wherein the island division model is trained by historical fault processing data. According to the island division result, a load recovery model is used to sequentially recover the load, wherein the load recovery model comprises an evaluation network layer and a target network layer, the evaluation network layer outputs an action decision according to island state data, the target network layer outputs an action value evaluation according to island state data, and the weight of the evaluation network layer is adjusted according to the action value evaluation.

9. A computing device, comprising: Comprise: One or more processors; A storage device is configured to store one or more programs, when the one or more programs are executed by the one or more processors, so that the one or more processors implement the method as claimed in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a program, which is executed by the processor to implement the method as claimed in any one of claims 1 to 7.

Images