Power distribution network self-healing control method, equipment and medium

Abstract

The invention relates to the technical field of power distribution networks, in particular to a power distribution network self-healing control method and device and a medium, and the method comprises the steps: building a distributed structure model, obtaining a topological structure, collecting historical power distribution abnormal data, carrying out the data preprocessing and feature extraction, inputting the data into a first deep learning model, training the first deep model, and carrying out the self-healing control of the power distribution network; adjusting parameters to obtain a fault detection model, collecting real-time power distribution data, inputting the real-time power distribution data into the fault detection model, obtaining a power distribution abnormal position, inputting the power distribution abnormal position into a topological structure, obtaining a reconstruction power distribution scheme through calculation, obtaining an influence factor through calculation, collecting power distribution abnormal data and carrying out feature extraction. The second depth model is trained, parameters are adjusted, an oscillation judgment model is obtained, whether oscillation happens to the reconstructed power distribution scheme or not is judged, cost calculation is conducted on the reconstructed power distribution scheme without oscillation, a topological structure is recognized and updated, cost is reduced, and economic benefits are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power distribution network, and particularly relates to a power distribution network self-healing control method, device and medium. BACKGROUND

[0002] At present, distributed power supply is characterized by large quantity, wide coverage, low capacity, randomness and intermittence. The wide-range access of distributed power supply optimizes the balance of power distribution network, and the self-healing technology of smart power distribution network cannot be implemented without the action of power supply side or even distributed power supply.

[0003] The inverter-type distributed energy is a power supply that converts direct-current energy into alternating-current energy and performs distributed power supply. Since the inverter-type distributed energy has a higher degree of dependence on the power grid, the reconstruction of the power distribution network has a greater impact, which may cause a significant reduction in its output characteristics and is more likely to generate harmonics. Although the voltage and current after reconstruction are still within the safe range, it is more likely to cause oscillation. The oscillation not only reduces the service life and reliability of the equipment, but also may trigger a chain reaction, leading to a larger range of power grid failure.

[0004] With the expansion of the installed capacity of distributed power supply, the continuous increase of power grid load, and the improvement of the scale of smart power distribution network, the smart power distribution network faces a series of challenges in the operation process. Therefore, a self-healing control module is designed to automatically detect the fault location and make corrections when the power distribution network fails. SUMMARY

[0005] In view of the above existing problems, the present application is proposed.

[0006] Therefore, the present application provides a power distribution network self-healing control method, device and medium to solve the problem that the inverter-type distributed energy has a higher degree of dependence on the power grid, the reconstruction of the power distribution network has a greater impact, which may cause a significant reduction in its output characteristics and is more likely to generate harmonics. Although the voltage and current after reconstruction are still within the safe range, it is more likely to cause oscillation.

[0007] To solve the above technical problems, the present application provides the following technical solutions:

[0008] In a first aspect, the present application provides a power distribution network self-healing control method, comprising:

[0009] A distributed structure model is established, the topological structure is obtained, and after collecting historical power distribution abnormal data, data preprocessing and feature extraction are performed;

[0010] The processed historical power distribution abnormal data is input into a first deep learning model, the first deep model is trained through the distributed structure and the topological structure, the parameters are adjusted, and a fault detection model is obtained;

[0011] The real-time power distribution data is collected and input into the fault detection model to obtain the position of the power distribution anomaly, the position of the power distribution anomaly is input into the topology structure, and the reconstructed power distribution scheme is obtained by calculation;

[0012] The influence factor is calculated based on the reconstructed power distribution scheme, the power distribution anomaly data is collected and feature extraction is performed, the second deep model is trained using the feature data, the second cross-validation method is used to adjust the parameters, and the oscillation judgment model is obtained;

[0013] The oscillation judgment model is used to judge whether the reconstructed power distribution scheme has oscillation, the reconstructed power distribution scheme without oscillation is calculated for cost, and the topology structure is identified and updated.

[0014] As a preferred scheme of the power distribution network self-healing control method, the first deep model is trained through the distributed structure and the topology structure, and the parameters are adjusted, including:

[0015] The input data is calculated through each layer of the model to obtain the output result of the model;

[0016] The gradient is calculated according to the output result, and the model parameters are updated according to the gradient;

[0017] The first cross-validation method is used to adjust the parameters.

[0018] As a preferred scheme of the power distribution network self-healing control method, the real-time power distribution data is collected and input into the fault detection model to obtain the position of the power distribution anomaly, including:

[0019] The running state is judged according to the real-time power distribution data;

[0020] The power distribution data is continuously collected when the running state is normal, and the fault position is located when the running state is abnormal.

[0021] As a preferred scheme of the power distribution network self-healing control method, the reconstructed power distribution scheme is obtained by calculation, including:

[0022] The group intelligence algorithm is used to calculate the reconstructed power distribution scheme;

[0023] The power distribution network nodes are set to form a connected graph, the group intelligence algorithm parameters are initialized, the walking path is adjusted based on the connected graph, and the reconstructed power distribution scheme is output after the convergence condition is reached.

[0024] As a preferred scheme of the power distribution network self-healing control method, the influence factor is calculated based on the reconstructed power distribution scheme, including:

[0025] The influence factor I is calculated by the following formula:

[0026] I = a * AV i + b * Af i + g * AP i ,

[0027] Wherein, a, b, g are weight coefficients, a+b+g=1;AV i It is the voltage fluctuation caused by the i-th reconstruction of the distribution network scheme;Af i It is the frequency deviation caused by the i-th reconstruction of the distribution network scheme, AP i It is the power factor change caused by the i-th reconstruction of the distribution network scheme.

[0028] The beneficial effects of the preferred technical scheme are: by fusing the three key electrical quantities of voltage fluctuation, frequency deviation and power factor change, the potential weak link in the process of distribution network reconstruction is effectively captured. This single-valued index provides a quantifiable and comparable decision basis for subsequent scheme optimization, significantly improving the accuracy and response efficiency of self-healing control under complex working conditions.

[0029] As a preferred scheme of the distribution network self-healing control method described in the application, wherein: a distributed structure model is established, comprising:

[0030] The geographical position information of each node in the distribution network, the connection relationship between nodes, the impedance parameters of the line, the capacity and transformation ratio of the transformer, and the distribution network equipment parameters;

[0031] The distributed structure model is constructed by software, and each node in the distributed structure model is used as a monitoring point, and the fault position is quickly located by analyzing the data of each node.

[0032] The beneficial effects of the preferred technical scheme are: by integrating the node geographical information, the topological connection relationship and the accurate electrical parameters (such as line impedance, transformer capacity, etc.), a high-fidelity distributed structure model is constructed;Based on the model, the key equipment of the whole network is converted into parallel monitoring points, and the accurate mapping relationship between real-time data flow and the underlying physical network is used to have a significant identification advantage for the implicit ground fault in the multi-branch feeder loop network, and to provide a solid topological foundation for quickly isolating the fault area.

[0033] As a preferred scheme of the distribution network self-healing control method described in the application, wherein: the second cross-validation method is used to adjust the parameters, comprising:

[0034] The data set is divided into k subsets with similar size, and each time a subset is selected as a validation set, and the remaining k-1 subsets are used as training sets for model training and validation;

[0035] Repeat k times, each time select a different subset as the validation set, and finally take the average of the k validation results as the performance indicator of the model.

[0036] As a preferred scheme of the power distribution network self-healing control method, the first deep model is trained through a distributed structure and a topological structure, including:

[0037] The training process includes two stages of forward propagation and back propagation: in the forward propagation stage, the input data is calculated through each layer of the model to obtain the output result of the model;

[0038] In the back propagation stage, the loss function is calculated according to the output result of the model and the real label, the gradient of the loss function to the model parameters is calculated through the back propagation algorithm, the parameters of the model are updated according to the gradient, and the training process is repeated for multiple iterations.

[0039] The beneficial effects of the preferred technical scheme are: through the cross-validation mechanism of k cycles, the randomness deviation of single data division is effectively overcome, and the model parameter optimization process fully adapts to the complex distribution characteristics of the power distribution network oscillation data; combined with the forward-backward propagation mechanism under the distributed topological constraint, the electrical connection relationship and the device physical parameters are synchronously fused in the gradient update, so that the model learning process is simultaneously driven by the data law and the power grid physical law.

[0040] In a second aspect, the present application provides an electronic device, comprising:

[0041] a memory and a processor;

[0042] The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions, which realize the steps of the power distribution network self-healing control method.

[0043] In a third aspect, the present application provides a computer readable storage medium, which stores computer executable instructions, and the computer executable instructions are executed by the processor to realize the steps of the power distribution network self-healing control method.

[0044] Compared with the prior art, the application has the beneficial effects: the application realizes the whole-chain closed-loop control of fault positioning, reconstruction optimization and oscillation suppression through the deep coupling of the distributed structure model and the topological constraint: first, the distributed modeling based on geographic information and electrical parameters provides a physical basis for millisecond-level fault positioning; second, the influence factor calculation of multi-dimensional dynamic indicators such as voltage, frequency and power factor accurately quantifies the system impact of different reconstruction schemes; and then through the distributed training under the topological constraint and the k-round cross-validation mechanism, the deep learning model synchronously follows the data law and the physical law of the power grid, and finally forms the self-healing control closed loop of "fast fault isolation-reconstruction scheme optimization-oscillation active defense-cost dynamic optimization", which significantly improves the system resilience under high proportion of new energy access. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0046] Figure 1 The whole flowchart of a power distribution network self-healing control method according to an embodiment of the application. DETAILED DESCRIPTION

[0047] In order to make the above-mentioned purposes, features and advantages of the application more apparent and easy to understand, the specific embodiments of the application will be described in detail below with reference to the drawings in the specification. Obviously, the described embodiments are only a part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor should be within the protection scope of the application.

[0048] Embodiment 1, refer to Figure 1 According to an embodiment of the application, a power distribution network self-healing control method is provided, comprising:

[0049] S1: Establish a distributed structure model, obtain a topological structure, collect historical power distribution abnormal data, and then perform data preprocessing and feature extraction;

[0050] S2: input the processed historical power distribution abnormal data into a first deep learning model, train the first deep model through the distributed structure and the topological structure, adjust the parameters, and obtain a fault detection model;

[0051] S3: collect real-time power distribution data, input into the fault detection model, obtain the position of power distribution abnormality, input the power distribution abnormality position into the topological structure, and obtain a reconstruction power distribution scheme through calculation;

[0052] S4: Calculate the impact factor based on the reconstructed power distribution scheme, collect power distribution abnormal data and perform feature extraction, train the second deep model using feature data, adjust the parameters using the second cross-validation method, and obtain the oscillation judgment model;

[0053] S5: Use the oscillation judgment model to determine whether the reconstructed power distribution scheme has oscillation, calculate the cost of the reconstructed power distribution scheme that has not occurred, and update the topology structure.

[0054] It should be noted that through the deep integration of distributed node modeling and topology constraints, a closed-loop control chain of "perception-decision-defense" is constructed: the distributed model based on device parameters and geographic information realizes accurate positioning within 30 milliseconds of failure; the reconstruction scheme generated by the ant colony algorithm is quantitatively evaluated by voltage, frequency, and power factor dynamic impact factors, and the oscillation risk is predicted through the deep neural network under the topology constraint, and the final optimized scheme is completed through k rounds of cross-validation training. Cost optimization and topology update.

[0055] Embodiment 2, refer to Figure 1 For an embodiment of the present application, based on the above embodiment, a power distribution network self-healing control method is provided.

[0056] In the present application, in step S1, a distributed structure model is established, the topology structure is obtained, historical power distribution abnormal data is collected, data preprocessing and feature extraction are performed, including the following steps A1-A2:

[0057] A1: Establish a distributed structure model and obtain the topology structure;

[0058] A2: After collecting historical power distribution abnormal data, perform data preprocessing and feature extraction.

[0059] Specifically, in step A1, the distributed structure model is established, and the topology structure is obtained, which is embodied as follows:

[0060] The geographic location information of each node in the power distribution network, the connection relationship between nodes, the impedance parameters of the line, the capacity and transformation ratio of the transformer, and the device parameters of the power distribution network;

[0061] The distributed structure model is constructed by software, and each node in the distributed structure model can be used as a monitoring point to monitor the operation state of the power distribution network in real time. When a fault occurs, the fault location can be quickly located by analyzing the data of each node. The topology structure can show the connection relationship between each element in the power distribution network.

[0062] Specifically, in step A2, data preprocessing and feature extraction are performed, which are embodied as follows:

[0063] Obtain fault history data of the power distribution network and perform preprocessing; the fault history data includes current, voltage, power, temperature, etc. Feature extraction is performed on the preprocessed fault history data, key features are extracted, and the data dimension is reduced.

[0064] In the embodiments of the present application, the processed historical power distribution abnormal data in step S2 is input into the first deep learning model, the first deep model is trained through the distributed structure and the topological structure, the parameters are adjusted, and a fault detection model is obtained, including the following steps B1-B4:

[0065] B1: input the processed historical power distribution abnormal data into the first deep learning model, and train the first deep model through the distributed structure and the topological structure;

[0066] B2: adjust the parameters using the first cross-validation method to obtain the fault detection model.

[0067] Specifically, the first deep model is trained in step S2 using a deep learning model:

[0068] The feature-extracted fault history data is input into the designed deep learning model for training. During the training process, the model adjusts the parameters of the model to minimize the prediction error according to the features and labels (i.e. fault types or fault states) of the data; the training process usually includes two stages of forward propagation and back propagation: in the forward propagation stage, the input data is calculated through each layer of the model to obtain the output result of the model. In the deep learning model, the output result can be the probability distribution of the fault, the fault type label or the time of the fault occurrence, etc.; in the back propagation stage, the loss function (such as cross-entropy loss function, mean square error loss function, etc.) is calculated according to the output result of the model and the real label, and then the gradient of the loss function to the model parameters is calculated through the back propagation algorithm, and the parameters of the model are updated according to the gradient; the training process needs to be repeated for multiple iterations until the performance of the model reaches a satisfactory level.

[0069] In an optional embodiment, in the scenario where the collected historical power distribution abnormal data has time sequence dependence, the first deep model in step S2 can also use a time series model, convert the data into sliding window samples, capture long-term dependencies through a gating mechanism, and use a back propagation through time algorithm for training, combined with regularization techniques such as Dropout.

[0070] In another optional embodiment, in the scenario where the input historical power distribution anomaly data sample is small, the first deep model in step S2 can also adopt a Gaussian process regression model, based on a Bayesian framework, to regard the function as an implementation of a random process, and define the similarity between samples through a kernel function. The prediction result not only gives an estimated value of the fault time, but also provides a confidence interval.

[0071] In the embodiments of the present application, the first cross-validation method is used to adjust the parameters in step B2, and leave-one-out cross-validation is adopted, that is, only one sample is left out as a validation set each time, and the remaining samples are used as a training set for model training and validation. This process is repeated n times (n is the total number of samples in the data set), and different samples are left out as a validation set each time to effectively optimize the model parameters.

[0072] In an optional embodiment, when the collected historical power distribution anomaly data has a strict time sequence, the first cross-validation method used to adjust the parameters in step B2 can also be implemented through time series cross-validation, that is, the data is divided into consecutive blocks according to the time stamp, and it is ensured that the training set is always earlier than the validation set.

[0073] In another optional embodiment, when the parameters of the first deep model need to be adjusted at the same time, the first cross-validation method used to adjust the parameters in step B2 can also be implemented through nested cross-validation, that is, a double-layer cross-validation structure is adopted, the outer loop performs k-fold division to evaluate the generalization ability of the model, and the inner loop performs k-fold validation on each training set for parameter optimization.

[0074] It should be noted that by inputting the feature-extracted historical power distribution anomaly data into the deep learning model, combining distributed and topological structures for efficient training, using forward propagation to generate fault probability distribution or type label as key output, and then using back propagation to accurately calculate parameter gradients with loss functions such as cross-entropy or mean square error, dynamic optimization of model parameters is achieved. At the same time, the first cross-validation method is used to systematically optimize the parameters, and through multiple rounds of iterative training, the model's ability to capture power distribution fault features and prediction accuracy are significantly improved, ultimately constructing a highly adaptive fault detection model to provide reliable technical support for power grid anomaly identification.

[0075] In the embodiments of the present application, real-time power distribution data is collected in step S3, input into the fault detection model, and the location of the power distribution anomaly is obtained. The power distribution anomaly location is input into the topological structure, and the reconstructed power distribution scheme is obtained by calculation, including the following steps C1-C2:

[0076] C1: Collect real-time power distribution data, input into the fault detection model, and obtain the location of the power distribution anomaly;

[0077] C2: input the abnormal power distribution location into the topology structure, and calculate the reconfigured power distribution scheme through the swarm intelligence algorithm.

[0078] Specifically, the swarm intelligence algorithm used in step C2 to calculate the reconfigured power distribution scheme is the ant colony algorithm, which includes:

[0079] The power distribution network equipment is set as nodes, and a connected graph with connection relationships between the nodes is formed; taking negative charges, transformers, power sources and other equipment as nodes, the power distribution network can automatically form a network associated with each node. This connected graph constitutes the basic framework for the ant colony algorithm to search the solution space.

[0080] The key parameters of the ant colony are initialized, including the number of ants, the pheromone evaporation coefficient, and the heuristic function. The number of ants determines the parallelism of the algorithm in searching the solution space. The pheromone evaporation coefficient controls the decay rate of pheromone in the environment, thereby affecting the exploration and utilization capabilities of the algorithm. The heuristic function is usually designed based on the specific requirements of the problem and is used to guide the ants to consider the specific goals of the problem when selecting paths, such as minimizing the total loss of the power distribution network or improving power supply reliability.

[0081] Based on the connected graph, the ants adjust their walking paths according to the connection relationships of the current nodes, the pheromone concentration and the heuristic function.

[0082] Each ant is assigned a random starting node, and each ant selects the next node to visit based on the pheromone concentration and heuristic information (such as line impedance, load, etc.) at the current location. This process is usually achieved by calculating the transition probability of each possible path, and the size of the transition probability depends on the pheromone concentration and the heuristic function value on the path. Paths with high pheromone concentration and good heuristic function information are more likely to be selected.

[0083] Increase the pheromone concentration at the current location to encourage subsequent ants to also choose these paths. At the same time, in order to maintain the diversity of the algorithm, the pheromone needs to gradually evaporate in the environment, thereby reducing the attractiveness of old paths. In order to avoid the ants falling into local optimal solutions, the ants need to have a certain randomness when selecting paths.

[0084] Repeat the ant colony algorithm to calculate the reconfigured power distribution scheme until the preset number of iterations or convergence conditions are reached, and output M reconfigured power distribution network schemes. After reaching the convergence condition, we can select M optimal reconfigured power distribution network schemes from the multiple solutions generated by the algorithm for output.

[0085] In an alternative embodiment, in the scenario of small-scale power distribution network and the need for rapid reconstruction, the reconstruction power distribution scheme calculated by the swarm intelligence algorithm in step C2 can also be realized by the shortest path algorithm. The power distribution network is modeled as a weighted graph, with line impedance or power loss as edge weight value. The optimal power supply path is determined by shortest path search to obtain the reconstruction power distribution scheme.

[0086] In another alternative embodiment, in the scenario of large-scale power distribution network, the reconstruction power distribution scheme calculated by the swarm intelligence algorithm in step C2 can also be realized by genetic algorithm. The network topology is coded as a chromosome, and the optimal reconstruction power distribution scheme is obtained by selection, crossover and mutation operations.

[0087] In the embodiments of the present application, the influence factor is calculated based on the reconstruction power distribution scheme in step S4, the power distribution abnormal data is collected and the features are extracted, the second deep model is trained using the feature data, the parameters are adjusted using the second cross-validation method, and the oscillation judgment model is obtained, including the following steps D1-D3:

[0088] D1: calculating the influence factor based on the reconstruction power distribution scheme;

[0089] D2: collecting power distribution abnormal data and extracting features, training the second deep model using the feature data;

[0090] D3: adjusting the parameters using the second cross-validation method to obtain the oscillation judgment model.

[0091] Specifically, the influence factor is calculated based on the reconstruction power distribution scheme in step D1, and the specific calculation method is as follows:

[0092] I = a · AV i + β · Af i + γ · AP i ,

[0093] Wherein, a, β, γ are weight coefficients, a + β + γ = 1; AV i is the voltage fluctuation caused by the i-th reconstruction power distribution scheme; Af i is the frequency deviation caused by the i-th reconstruction power distribution scheme, AP i is the power factor change caused by the i-th reconstruction power distribution scheme.

[0094] Specifically, the second deep model is trained using the feature data in step S4, which is a deep neural network model:

[0095] The historical data of the power distribution network in different operating states are collected, including key parameters such as voltage, current, power factor and the like; the collected data are preprocessed, including data cleaning, denoising, normalization and the like, to ensure the accuracy and consistency of the data, the feature variables related to oscillation are selected from the preprocessed historical data, and the key features are extracted to obtain feature data; based on the physical characteristics and operating experience of the power distribution network, the feature variables related to oscillation are selected, and the features closely related to oscillation are further extracted by using a statistical method, the feature data are taken as input, and whether the power distribution network oscillates or not is taken as output label, and the neural network model is trained.

[0096] In an optional embodiment, in the scenario that the power distribution network topology and accurate parameters are a high-voltage direct-current transmission system, the second deep model in step S4 can also adopt an oscillation model based on physical mechanism, and the system damping ratio and oscillation frequency are calculated by eigenvalue analysis. The residual error between the real-time measurement data and the model prediction value is used for oscillation detection.

[0097] In another optional embodiment, when the type of oscillation to be identified in the power distribution network is unknown, the second deep model in step S4 can also adopt an unsupervised anomaly detection model, a similarity matrix of voltage and current phasors (based on cosine similarity) is constructed, and abnormal points are separated by Laplace feature mapping to detect oscillation (such samples are not included in the training data).

[0098] Specifically, the second cross-validation method is used to adjust the parameters in step S4, and k-fold cross-validation is used:

[0099] The data set is divided into k subsets of similar size, one subset is selected as the validation set each time, and the remaining k-1 subsets are used as the training set for model training and validation. Repeat k times, and select a different subset as the validation set each time. Finally, the average value of the k validation results is taken as the performance index of the model.

[0100] In an optional embodiment, in the scenario that the oscillation category to be identified by the power distribution network is extremely unbalanced, the second cross-validation method in step S4 can also be used to adjust the parameters, and stratified cross-validation can be used, the proportion of positive and negative samples in each fold is consistent with that in the whole set, to prevent the loss of rare categories in each fold and improve the detection rate of oscillation.

[0101] In another optional embodiment, when the input power distribution abnormal data is small sample data, the second cross-validation method in step S4 can also be used to adjust the parameters, and the bootstrap validation method can be used to generate the training set by sampling with replacement, and the samples not sampled are used as the validation set. Repeat 500 times to fully utilize the rare samples to adjust the parameters.

[0102] It should be noted that by calculating the comprehensive impact factor in the reconstructed power distribution scheme, combining the fine processing of historical power distribution abnormal data and the multi-dimensional feature extraction related to oscillation strength, a deep neural network model is constructed and trained; further, the model parameters are optimized by cross-validation technology, and finally a high-robustness oscillation judgment model is established, which significantly improves the recognition accuracy and generalization ability of the model for oscillation risk under complex operation state of the power distribution network, and provides more reliable data-driven decision support for safe and stable operation of the power grid.

[0103] In the embodiment of the application, whether the reconstructed power distribution scheme oscillates is judged by using the oscillation judgment model in step S5, the cost of the reconstructed power distribution scheme without oscillation is calculated, and the topological structure is identified and updated, including the following steps E1-E2:

[0104] E1: judging whether the reconstructed power distribution scheme oscillates by using the oscillation judgment model;

[0105] E2: calculating the cost of the reconstructed power distribution scheme without oscillation, and identifying and updating the topological structure.

[0106] Specifically, the cost calculation of the reconstructed power distribution scheme without oscillation in step E2 specifically embodies:

[0107] (1) equipment investment cost, including the cost of new or replaced equipment (such as transformers, lines, etc.);

[0108] (2) operation and maintenance cost, including the cost of daily operation and maintenance, repair and fault handling;

[0109] (3) energy consumption cost, considering the energy efficiency improvement or reduction of the reconstructed power distribution network;

[0110] (4) social and economic cost, such as indirect cost of power loss and user satisfaction decrease.

[0111] In summary, the application constructs a real-time monitoring network by a distributed structure model, accurately locates faults and generates multi-path reconstruction schemes by combining deep learning and cross-validation technology; a high-robustness oscillation judgment model is constructed by using comprehensive impact factor evaluation and adaptive cross-validation strategy, and multi-dimensional optimization analysis of equipment investment, operation and energy consumption, and social and economic cost is integrated, realizing the whole-chain closed loop from fault rapid response, oscillation early warning to economic decision, significantly improving the anti-disturbance ability and resource allocation efficiency of the power distribution network under complex working conditions, and providing systematic technical support for stable operation and cost control of the smart grid.

[0112] Embodiment 3, the above is a schematic scheme of a power distribution network self-healing control method.

[0113] The embodiment also provides an electronic device suitable for power distribution network self-healing control, comprising a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the power distribution network self-healing control method proposed in the above embodiment.

[0114] The embodiment also provides a storage medium having a computer program stored thereon, and the computer program is executed by a processor to realize the power distribution network self-healing control method proposed in the above embodiment.

[0115] The storage medium proposed in the embodiment and the power distribution network self-healing control method proposed in the above embodiment belong to the same inventive concept, and the technical details not described in the embodiment can be referred to the above embodiment, and the embodiment has the same beneficial effects as the above embodiment.

[0116] Through the above description of the embodiments, those skilled in the art can clearly understand that the present application can be realized by means of software and necessary universal hardware, and of course can also be realized by hardware. Based on such understanding, the technical solutions of the present application or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a floppy disk, a read-only memory (ROM), a random access memory (RAM), a FLASH memory, a hard disk, or an optical disc, and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods of various embodiments of the present application.

[0117] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application but not limit the present application, and although the present application 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 application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and all of them should be covered in the scope of the claims of the present application.

Claims

1. A self-healing control method for a power distribution network, characterized in that, include: A distributed structure model is established, the topology is obtained, and historical power distribution anomaly data is collected. Data preprocessing and feature extraction are then performed. The processed historical power distribution anomaly data is input into the first deep learning model. The first deep model is trained through distributed structure and topology structure, and the parameters are adjusted to obtain the fault detection model. Real-time power distribution data is collected and input into the fault detection model to obtain the location of power distribution anomalies. The location of the power distribution anomalies is then input into the topology, and a power distribution reconfiguration scheme is obtained through calculation. The influencing factors were calculated based on the reconfigured power distribution scheme. Power distribution anomaly data were collected and features were extracted. The feature data were used to train the second deep model. The parameters were adjusted using the second cross-validation method to obtain the oscillation judgment model. An oscillation judgment model is used to determine whether the reconfigured power distribution scheme has oscillations. The cost of the reconfigured power distribution scheme that has not oscillated is calculated, and the topology is identified and updated.

2. The self-healing control method for a power distribution network as described in claim 1, characterized in that, The process of training the first deep model using a distributed structure and topology, and adjusting the parameters, includes: The input data is processed through various layers of the model to obtain the model's output results; Calculate the gradient based on the output, and update the model parameters based on the gradient. The parameters were adjusted using the first cross-validation method.

3. The self-healing control method for a power distribution network as described in claim 2, characterized in that, The process of collecting real-time power distribution data and inputting it into the fault detection model to obtain the location of power distribution anomalies includes: Determine whether the operating status is normal based on real-time power distribution data; If the operation is normal, continue to collect power distribution data; if the operation is abnormal, locate the fault location.

4. The self-healing control method for a power distribution network as described in claim 3, characterized in that, The reconfigured power distribution scheme obtained through calculation includes: Swarm intelligence algorithms are used to calculate and reconfigure power distribution schemes. Set up distribution network nodes, form a connectivity graph, initialize swarm intelligence algorithm parameters, adjust the walking path based on the connectivity graph, and output the reconstructed distribution scheme after reaching the convergence condition.

5. The self-healing control method for a power distribution network as described in claim 4, characterized in that, The influencing factors calculated based on the reconfigured power distribution scheme include: Impact factor I is calculated using the formula: I=α·ΔV i +β·Δf i +γ·ΔP i , Among them, α, β, γ are weight coefficients, α+β+γ=1; ΔV i Let Δf be the voltage fluctuation caused by the i-th reconfiguration scheme of the distribution network; i Let ΔP be the frequency deviation caused by the i-th reconfiguration scheme of the distribution network. i Let be the change in power factor caused by the i-th power distribution network reconfiguration scheme.

6. The self-healing control method for a power distribution network as described in claim 5, characterized in that, The establishment of the distributed structure model includes: Geographical location information of each node in the distribution network, connection relationships between nodes, impedance parameters of lines, capacity and turns ratio of transformers, and parameters of distribution network equipment; A distributed structure model is constructed using software. By using a distributed structure, each node in the distributed structure model is used as a monitoring point. By analyzing the data of each node, the fault location can be quickly determined.

7. The self-healing control method for a power distribution network as described in claim 6, characterized in that, The parameter adjustment using the second cross-validation method includes: The dataset is divided into k subsets of similar size. Each time, one subset is selected as the validation set and k-1 subsets are selected as the training set for model training and validation. Repeat the process k times, selecting a different subset as the validation set each time, and finally take the average of the k validation results as the model's performance metric.

8. The self-healing control method for a power distribution network as described in claim 7, characterized in that, The training of the first deep model through a distributed structure and topology includes: Forward propagation phase and backward propagation phase; During the forward propagation phase, the input data is processed through each layer of the model to obtain the model's output. During the backpropagation phase, the loss function is calculated based on the model's output and the true labels. The gradient of the loss function with respect to the model parameters is calculated using the backpropagation algorithm. The model parameters are then updated based on the gradient. The training process is repeated iteratively multiple times.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

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

Images