Distribution network risk assessment method based on distribution network elements and neural network

By clustering and neural network training of distribution network elements and combining internal and external data for risk assessment, the problem of difficulty in risk assessment of distribution network is solved, a more comprehensive and reasonable risk assessment is achieved, and the power supply reliability and Electrical energy quality.

CN115034608BActive Publication Date: 2024-11-08GUANGZHOU POWER SUPPLY BUREAU GUANGDONG POWER GRID CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202210644697.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-06-09
Publication Date
2024-11-08
Estimated Expiration
2042-06-09

AI Technical Summary

Technical Problem

The existing technology is difficult to effectively evaluate the risks of the distribution network, especially due to its complex network structure and multiple influencing factors, which increases the difficulty of risk assessment. A single-sided assessment cannot fully describe the operating trend of the distribution network.

Method used

Using a risk assessment method based on distribution network elements and neural networks, the distribution network elements are clustered, key elements are obtained, and the risk assessment model is trained using a ensemble neural network and optimization algorithm to dynamically evaluate the data in the evaluation area, combined with internal Comprehensive analysis with external data.

Benefits of technology

A more comprehensive and reasonable assessment of the distribution network risks has been achieved, which reduces the difficulty of risk prediction, improves power supply reliability and power quality, reduces power outage losses caused by failures, and establishes a more reasonable risk assessment system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115034608B_ABST
    Figure CN115034608B_ABST
Patent Text Reader

Abstract

The present invention discloses a distribution network risk assessment method based on distribution network elements and neural networks, comprising the following steps: S1, clustering distribution network elements to obtain key distribution network elements, wherein the distribution network elements include internal data and external data in the assessment area; S2, training the ensemble neural network through the key distribution network elements to obtain a risk assessment model, wherein the ensemble neural network is trained using an optimization algorithm; S3, dynamically assessing distribution network risks through the risk assessment model, inputting the prediction sample set into the risk assessment model, and the risk assessment model outputs the prediction result to assess the distribution network risk. The present invention clusters according to different distribution network elements, which can greatly reduce the difficulty of the machine model for risk prediction, integrates multiple machine learning to dynamically assess distribution network risks, and effectively analyzes a large amount of data information, which can more comprehensively and reasonably assess risks, avoid or reduce power outage losses caused by distribution network failures, and improve power supply reliability and power quality.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of electric power technology, and in particular to a distribution network risk assessment method based on distribution network elements and neural networks. Background Art

[0002] As an important intermediate link between the transmission network and users, the operation status of the distribution network directly affects the reliability and quality of electricity. With the continuous development of social economy, the construction, operation and management of the distribution network have begun to enter the "fine" stage. As one of the key links in the construction of a strong and intelligent power grid, the whole society pays more attention to it. As the scale of the distribution network increases, its network structure becomes more and more complex, so the difficulty of collecting distribution network operation data increases day by day, and there are many influencing factors. It is difficult to describe the operation status of the distribution network from a single aspect. It is necessary to collect information on multiple indicators, which increases the difficulty of risk assessment of the distribution network. Summary of the invention

[0003] The purpose of the present invention is to solve the problem that it is difficult to evaluate the risk of distribution network in the prior art, and to provide a distribution network risk assessment method based on distribution network elements and neural networks, which can effectively analyze a large amount of data in the assessment area and can evaluate the risk of distribution network more comprehensively and reasonably.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0005] A distribution network risk assessment method based on distribution network elements and neural network includes the following steps:

[0006] S1. Clustering distribution network elements to obtain key elements of the distribution network, wherein the distribution network elements include internal data and external data within the evaluation area;

[0007] S2. Training the ensemble neural network by using key elements of the distribution network to obtain a risk assessment model, wherein the ensemble neural network is trained by using an optimization algorithm;

[0008] S3. Dynamically evaluate the distribution network risk through the risk assessment model, input the predicted sample set into the risk assessment model, and the risk assessment model outputs the prediction result to evaluate the distribution network risk.

[0009] Furthermore, the internal data includes data of the voltage system, patrol APP, marketing and distribution, asset integration system, electric energy platform, metering automation system, marketing integration system, marketing and operation monitoring system, energy management system, power grid operation management system, power distribution and utilization system, and rapid power restoration system.

[0010] Furthermore, the external data includes data on weather, population, geography, economy, and time and solar terms.

[0011] Further, in step S1, the method of clustering the distribution network elements is as follows: first, a preset number of clusters of the distribution network elements is preset, the clustering error of the preset number of clusters is obtained, and the optimal number of clusters of the distribution network elements is determined by the preset clustering error; according to the optimal number of clusters of the distribution network elements, the distribution network elements are clustered to obtain multiple key elements of the distribution network.

[0012] Furthermore, according to the optimal number of clusters of distribution network elements, the distribution network elements are clustered as follows: the data of the distribution network elements are input, a point is randomly selected as the initial cluster center point, the probability of becoming the next cluster center is calculated according to the distance between the training data and the cluster center point, the next cluster center is selected by the roulette method, and it is determined whether the number of cluster centers reaches the optimal number of clusters. If not, clustering continues. If the optimal number of clusters is reached, each distribution network element is assigned to the point with the closest Euclidean distance to the cluster center, and the cluster centers of each group are recalculated according to the mean. If convergence is achieved, the clustering results are output. If not convergence is achieved, it is reallocated to the point with the closest Euclidean distance to the cluster center, and the cluster centers of each group are recalculated according to the mean until convergence.

[0013] Further, the ensemble neural network includes neural networks, support vector machines, extreme learning machines and / or other machine learning.

[0014] Further, in step S2, the method of using the optimization algorithm to train the ensemble neural network is as follows: setting the population size, number of particle dimensions, maximum number of iterations, horizontal crossover probability and vertical crossover probability parameters of the optimization algorithm; calculating the fitness value of each optimized particle to obtain the individual optimal value and the global optimal value of the optimized particle, and then comparing the individual optimal value and the global optimal value of the optimized particle, taking the one with better fitness value as the current optimal position to adjust the parameters of the ensemble neural network.

[0015] Furthermore, when the ensemble neural network is trained using an optimization algorithm, a mapping is established between the optimization particles and the weights and thresholds of the risk assessment model, that is, the weights and thresholds of the risk assessment model are encoded into real number vectors to represent individuals in the population, and the weights and thresholds constitute the initial population of the optimization algorithm.

[0016] Furthermore, when the optimization algorithm is used to train the ensemble neural network, the optimization objects include the correction amount for the connection weights between the hidden layer and the output layer, the thresholds of each neuron in the output layer, the connection weights between the input layer and the hidden layer, and the thresholds of each neuron in the hidden layer.

[0017] Further, in step S2, in order to optimize the weights of different machine learning results in the ensemble neural network, virtual prediction is used to calculate the relative error entropy values ​​and coefficient of variation of different machine learning models, including the following steps:

[0018] S21. Calculate the proportion p of the relative error of the jth machine learning model at the tth moment jt , p jt The calculation formula is:

[0019]

[0020]

[0021] In the formula, e jt is the relative prediction error of the jth machine learning prediction method at time t; y t is the actual value at time t; y jt is the predicted value of the jth machine learning prediction method at time t; j = 1, 2, ..., k; k is the number of machine learning models;

[0022] S22. Calculate the entropy value H of the prediction relative error of the j-th machine learning model j , H j The calculation formula is:

[0023]

[0024] In the formula, n represents the total learning time, p jt is the proportion of the relative error of the jth machine learning model at time t;

[0025] S23. Calculate the coefficient of variation D of the relative error of the prediction of the jth machine learning model j , D j The calculation formula is:

[0026] D j =1-H j

[0027] In the formula, H j is the entropy value of the prediction relative error of the jth machine learning model;

[0028] S24. Calculate the weight coefficient W of the j-th machine learning model j , W j The calculation formula is:

[0029]

[0030] Where D j is the coefficient of variation of the prediction relative error of the jth machine learning model, k is the number of machine learning models,

[0031] Compared with the prior art, the present invention clusters the assessment areas according to different distribution network element indicators, which can greatly reduce the difficulty of the machine model for risk prediction in the assessment area. It combines multiple machine learning risk assessment models to dynamically assess the operating risk of the low-voltage part of the assessment area, effectively analyzes the large amount of data information generated during the operation and failure of the assessment area, correctly selects risk indicators, and establishes a more comprehensive and reasonable risk assessment system to assess risks, avoid or reduce power outages caused by distribution network failures, and improve power supply reliability and power quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is an overall flow chart of an embodiment of a distribution network risk assessment method based on distribution network elements and neural networks of the present invention.

[0033] Figure 2 It is a clustering flow chart of an embodiment of a distribution network risk assessment method based on distribution network elements and neural networks of the present invention.

[0034] Figure 3 The risk assessment model flow chart of the embodiment of the distribution network risk assessment method based on distribution network elements and neural network of the present invention. DETAILED DESCRIPTION

[0035] The following is a further description of a distribution network risk assessment method based on distribution network elements and neural networks according to the present invention in conjunction with the accompanying drawings and specific embodiments.

[0036] See also Figure 1 The present invention discloses a distribution network risk assessment method based on distribution network elements and neural network, which specifically includes the following steps:

[0037] S1. Clustering distribution network elements to obtain key distribution network elements, wherein the distribution network elements include internal data and external data in the evaluation area.

[0038] S2. The ensemble neural network is trained by using key elements of the distribution network to obtain a risk assessment model, wherein the ensemble neural network is trained by using an optimization algorithm.

[0039] S3. Dynamically evaluate the distribution network risk through the risk assessment model, input the predicted sample set into the risk assessment model, and the risk assessment model outputs the prediction result to evaluate the distribution network risk.

[0040] The internal data includes data from business systems such as voltage system, patrol APP, marketing and distribution (power supply reliability), marketing and distribution (rapid power restoration), asset integration system, electric energy platform, metering automation system, marketing integration system, marketing and operation monitoring system, EMS (energy management system), power grid operation management system, power distribution and utilization system, and rapid power restoration system. The external data includes data on meteorology, population, region, economy, time and solar terms, etc.

[0041] Clustering the assessment area according to different distribution network element indicators can greatly reduce the difficulty of the machine model for risk prediction in the assessment area. Since the risk assessment of the assessment area involves many correlation factors and fault indicator factors, how to accurately cluster this high-dimensional data is a difficult problem. In view of the characteristics of risk prediction in the assessment area, the present invention first clusters the distribution network elements to obtain the key elements of the distribution network. The specific implementation steps are as follows:

[0042] The data of the distribution network elements are clustered. First, the preset number of clusters of the distribution network elements is preset to J, where J=8-12, and J is preferably 10. The clustering error of the preset number of clusters is obtained, and the optimal number of clusters of the distribution network elements is determined according to the preset clustering error.

[0043] The clustering error is expressed as:

[0044]

[0045] Among them, Ci is the i-th cluster, p is the sample point in Ci, m i is the mean of all samples in Ci, SSE is the clustering error of all samples, which represents the quality of clustering effect, and k∈J represents the number of clusters in the preset number of clusters.

[0046] When k is less than the actual number of clusters, the increase in J will greatly increase the degree of aggregation of each cluster, so the decrease in SSE will be large. When J reaches the optimal number of clusters, the return on the degree of aggregation obtained by further increasing J will rapidly decrease, so the decrease in SSE will drop sharply, and then tend to be flat as the value of J continues to increase. Therefore, it can be judged that when J=k is the optimal number of clusters.

[0047] According to the optimal number of clusters, the distribution network elements are clustered. The specific steps are as follows: Figure 2 As shown in the figure, the data of distribution network elements are input, and a point is randomly selected as the initial cluster center point. The probability of becoming the next cluster center is calculated based on the distance between the training data and the cluster center point. The roulette method is used to select the next cluster center. It is determined whether the number of cluster centers reaches the optimal number of clusters. If not, clustering continues. If the optimal number of clusters is reached, each distribution network element is assigned to the point with the closest Euclidean distance to the cluster center, and the cluster centers of each group are recalculated based on the mean. If convergence is achieved, the clustering results are output. If not convergence, it is reallocated to the point with the closest Euclidean distance to the cluster center, and the cluster centers of each group are recalculated based on the mean until convergence.

[0048] After determining the optimal number of clusters of distribution network elements, the distribution network elements are clustered to obtain multiple key elements of the distribution network, including the impact of medium-voltage typhoons, comprehensive governance mechanisms in the assessment area, low-voltage tripping, low-voltage faults, weather and time. The key elements of the distribution network are learned through an ensemble neural network, the weight coefficients of the key elements of the distribution network are determined, and a risk assessment model is established. The risk of the distribution network is assessed through the output of the prediction results of the risk assessment model.

[0049] The prediction effect of overload risk in the assessment area of ​​a single machine learning prediction method varies greatly under different environments, and it lacks dynamic adaptive prediction capabilities. Figure 3 As shown, in this embodiment, the ensemble neural network includes neural network, support vector machine, extreme learning machine and other machine learning, and other machine learning can be decision tree, random forest, Bayesian learning, etc. The combination of neural network, support vector machine, extreme learning machine and other machine learning can comprehensively play the advantages of different machine models, can provide more superior generalization performance, and point out the direction for adaptive prediction technology. Ensemble machine learning generates a group of individual learners (or weak learners, which are usually generated from training data by existing learning algorithms, such as neural networks, support vector machines, etc.), and then combines them using a certain strategy to form a risk assessment model.

[0050] When learning neural networks, support vector machines, extreme learning machines and other machines, targeted training and hyperparameter optimization are required for different algorithms. When typical machine learning algorithms such as neural networks, support vector machines, and extreme learning machines are trained on a large scale, the algorithms are prone to local optimality and premature maturity, and training stagnation problems occur, thereby affecting prediction accuracy and generalization and promotion capabilities. The present invention uses an optimization algorithm to train the ensemble neural network. When using the optimization algorithm for training, the local optimality and premature maturity problems can be effectively avoided. Compared with genetic algorithms, particle swarm algorithms, etc., the optimization algorithm has obvious advantages in solving the dimensionality curse problem and global convergence performance.

[0051] The method of using the optimization algorithm to train the ensemble neural network is as follows: setting the population size, number of particle dimensions, maximum number of iterations, horizontal crossover probability and vertical crossover probability parameters of the optimization algorithm; calculating the fitness value of each optimized particle to obtain the individual optimal value and the global optimal value of the optimized particle; then comparing the individual optimal value and the global optimal value of the optimized particle, taking the one with better fitness value as the current optimal position to adjust the parameters of the ensemble neural network.

[0052] When training the ensemble neural network using an optimization algorithm, a mapping is established between the optimized particles and the weights and thresholds of the risk assessment model, that is, the weights and thresholds of the risk assessment model are encoded into real number vectors to represent the individuals in the population, and a population of these real number vectors is randomly generated. The optimization objects include: the correction amount for the connection weights between the hidden layer and the output layer, the thresholds of each neuron in the output layer, the connection weights between the input layer and the hidden layer, and the thresholds of each neuron in the hidden layer. The weights and thresholds constitute the initial population of the optimization algorithm.

[0053] Specifically, in this embodiment, an optimization algorithm is used to train the ensemble neural network, including:

[0054] (1) Initialize the population

[0055] Suppose X is a randomly generated matrix with D columns and M rows, which represents a D-dimensional population composed of M individuals.

[0056] (2) Horizontal crossover operation

[0057] All the particles in the population are randomly combined in pairs without repetition, a total of M / 2 pairs of combinations. For each pair of combinations, the particles are horizontally crossed according to the following formula:

[0058]

[0059] In the formula, r 1 and r 2 are random values uniformly distributed on [0, 1]; c 1 and c 2 are random values uniformly distributed on [-1, 1]; X(i, d) and X(j, d) are the d-th dimensions of the parent particles X(i) and X(j) respectively; MS hc (i, d) and MS hc (j, d) are the d-th dimensions of the offspring generated by the horizontal crossover of X(i) and X(j) respectively, called the moderate solution.

[0060] The horizontal crossover results are saved in the moderate solution matrix MS hc . Calculate the fitness value of the particles and compare it with the fitness values of their parent particles X(i) and X(j). The one with the smaller fitness value is retained in X and participates in the next iteration.

[0061] Normalize each dimension of the particles obtained by horizontal crossover, and then randomly pair all the dimensions of the particles in pairs without repetition, a total of D / 2 pairs. For any pair of dimensions, generate a random number rand. If rand < Pv (Pv is usually taken as 0.2 - 0.8), then perform a vertical crossover operation on this pair of dimensions.

[0062] (3) Vertical crossover operation

[0063] Different from horizontal crossover, vertical crossover is an operation performed by all particles between different dimensions. Assume that vertical crossover is performed on the d 1 and d 2 dimensional, then they are cross-linked vertically to produce the moderate solution MS vc (i,d) is:

[0064] MS vc (i,d)=r·X(i,d 1 )+(1-r)·X(i,d 2 )

[0065] i∈N(1,M),d 1 ,d 2 ∈N(1,D)

[0066] In the formula, X(i,d 1 ) represents the first dimension of the parent particle X(i), X(i,d 2 ) represents the second dimension of the parent particle X(i), r is a random number uniformly distributed on [0,1]; MS vc (i,d) is the dth particle of particle i 1 Peacekeeping 2 The offspring produced by vertical crossover.

[0067] The vertical crossover results are stored in the mean solution matrix MS vc In the process, after the vertical crossover result is denormalized, the fitness value of the particle in the mean solution matrix is ​​calculated and compared with the fitness value of its parent particle. The particle with good fitness value is saved in X for the next iteration.

[0068] The offspring and parent generations generated by horizontal and vertical crossovers compete with each other to continuously generate new populations; if the new fitness value is better than the current individual optimal value, the current individual optimal value is replaced by the new fitness value; if the updated individual optimal value is better than the current global optimal solution, the current global optimal solution is replaced by the individual optimal value to complete the optimization of the risk assessment model.

[0069] In the framework of a collection of multiple hybrid machine learning models, there are differences in the risk prediction results of the assessment area obtained based on different machine learning algorithms. How to optimize the weights of different machine learning results is the key to the successful application of the ensemble machine learning method. The present invention uses virtual prediction to calculate the relative error entropy value and coefficient of variation of different machine learning models for different prediction scales, thereby solving the problem that the weight coefficient of the ensemble machine learning prediction of the risk prediction of the assessment area is difficult to determine and the dynamic adaptive ability is poor. The main implementation steps are as follows:

[0070] S21. Calculate the proportion p of the relative error of the jth machine learning model at the tth momentjt , p jt The calculation formula is:

[0071]

[0072]

[0073] In the formula, e jt is the relative prediction error of the jth machine learning prediction method at time t; y t is the actual value at time t; y jt is the predicted value of the jth machine learning prediction method at time t; j=1, 2, ..., k; k is the number of machine learning models; in the present invention, k=4 is set.

[0074] S22. Calculate the entropy value H of the prediction relative error of the j-th machine learning model j , H j The calculation formula is:

[0075]

[0076] In the formula, n represents the total learning time, p jt is the proportion of the relative error of the jth machine learning model at time t;

[0077] S23. Calculate the coefficient of variation D of the relative error of the prediction of the jth machine learning model j , D j The calculation formula is:

[0078] D j =1-H j

[0079] In the formula, H j is the entropy value of the prediction relative error of the jth machine learning model;

[0080] S24. Calculate the weight coefficient W of the j-th machine learning model j , W j The calculation formula is:

[0081]

[0082] Where D j is the coefficient of variation of the prediction relative error of the jth machine learning model, k is the number of machine learning models,

[0083] In summary, the present invention uses the above-mentioned risk assessment model that combines multiple machine learning to dynamically assess the operating risk of the low-voltage part of the assessment area, effectively analyzes a large amount of data information generated during the operation and failure of the assessment area, correctly selects risk indicators, and establishes a more comprehensive and reasonable risk assessment system to assess risks, avoid or reduce power outage losses caused by distribution network failures, and improve power supply reliability and power quality. It can be targeted to overhaul and adjust the equipment within the assessment area, improve the normal operation of power equipment, and improve the power supply reliability of the assessment area. Improve the current status of comprehensive governance work in the assessment area, improve the governance efficiency of the heavy risk assessment area, and conduct in-depth mining and analysis of the influencing factors of the operation risk of the distribution assessment area by conducting data collaborative application research based on key elements of the distribution network, predict and mine the risk assessment area, and promptly report and warn, helping professional departments to improve the level of customer electricity service, and provide support for peak summer, holiday power supply, and daily active services and active maintenance.

[0084] The above description is a detailed description of the preferred feasible embodiments of the present invention, but the embodiments are not intended to limit the scope of the patent application of the present invention. All equivalent changes or modified changes completed under the technical spirit disclosed by the present invention should fall within the patent scope covered by the present invention.

Claims

1. A distribution network risk assessment method based on distribution network elements and neural network, It is characterized in that The following steps are involved: S1. Clustering distribution network elements to obtain key elements of the distribution network, wherein the distribution network elements include internal data and external data within the evaluation area; S2. Training the ensemble neural network by using key elements of the distribution network to obtain a risk assessment model, wherein the ensemble neural network is trained by using an optimization algorithm; S3. Dynamically evaluate the distribution network risk through the risk assessment model, input the prediction sample set into the risk assessment model, and the risk assessment model outputs the prediction result to evaluate the distribution network risk; In step S1, the distribution network elements are clustered as follows; First, a preset number of clusters of the distribution network elements is preset, a clustering error of the preset number of clusters is obtained, and an optimal number of clusters of the distribution network elements is determined by the preset clustering error; the distribution network elements are clustered according to the optimal number of clusters of the distribution network elements to obtain a plurality of key elements of the distribution network; According to the optimal number of clusters of distribution network elements, the distribution network elements are clustered as follows: input the data of the distribution network elements, randomly select a point as the initial cluster center point, calculate the probability of becoming the next cluster center according to the distance between the training data and the cluster center point, select the next cluster center by the roulette method, and judge whether the number of cluster centers reaches the optimal number of clusters. If not, continue clustering. If the optimal number of clusters is reached, assign each distribution network element to the point with the closest Euclidean distance to the cluster center, recalculate the cluster centers of each group according to the mean, and output the clustering results if converged. If not, reassign it to the point with the closest Euclidean distance to the cluster center, and recalculate the cluster centers of each group according to the mean until convergence; The ensemble neural network includes a neural network, a support vector machine, an extreme learning machine and / or other machine learning; In step S2, the method of using the optimization algorithm to train the ensemble neural network is as follows: setting the population size, number of particle dimensions, maximum number of iterations, horizontal crossover probability and vertical crossover probability parameters of the optimization algorithm; calculating the fitness value of each optimized particle to obtain the individual optimal value and the global optimal value of the optimized particle; then comparing the individual optimal value and the global optimal value of the optimized particle, taking the one with better fitness value as the current optimal position to adjust the parameters of the ensemble neural network; When the ensemble neural network is trained using an optimization algorithm, a mapping between the optimization particles and the weights and thresholds of the risk assessment model is established, that is, the weights and thresholds of the risk assessment model are encoded into real vectors to represent individuals in the population, and the weights and thresholds constitute the initial population of the optimization algorithm; When the optimization algorithm is used to train the ensemble neural network, the optimization objects include the correction amount to the connection weight between the hidden layer and the output layer, the threshold of each neuron in the output layer, the connection weight between the input layer and the hidden layer, and the threshold of each neuron in the hidden layer; In step S2, in order to optimize the weights of different machine learning results in the ensemble neural network, virtual prediction is used to calculate the relative error entropy values ​​and coefficient of variation of different machine learning models, including the following steps: S21. Calculate the proportion p of the relative error of the jth machine learning model at the tth moment jt , p jt The calculation formula is: In the formula, e jt is the relative prediction error of the jth machine learning prediction method at time t; y t is the actual value at time t; y jt is the predicted value of the jth machine learning prediction method at time t; k is the number of machine learning models; S22. Calculate the entropy value H of the prediction relative error of the j-th machine learning model j , H j The calculation formula is: In the formula, n represents the total learning time, p jt is the proportion of the relative error of the jth machine learning model at time t; S23. Calculate the coefficient of variation D of the relative error of the prediction of the jth machine learning model j , D j The calculation formula is: D j =1-H j In the formula, H j is the entropy value of the prediction relative error of the jth machine learning model; S24. Calculate the weight coefficient W of the j-th machine learning model j , W j The calculation formula is: Where D j is the coefficient of variation of the prediction relative error of the jth machine learning model, k is the number of machine learning models, 2. The distribution network risk assessment method based on distribution network elements and neural network as claimed in claim 1, It is characterized in that The internal data includes data from the voltage system, patrol APP, marketing and distribution, asset integration system, electric energy platform, metering automation system, marketing integration system, marketing and operation monitoring system, energy management system, power grid operation management system, power distribution and utilization system, and rapid power restoration system.

3. The distribution network risk assessment method based on distribution network elements and neural network as claimed in claim 1, It is characterized in that The external data include data on weather, population, geography, economy, and time and solar terms.

Citation Information

Patent Citations

  • Method and apparatus for evaluating risk of heavy overload in distribution network station area

    CN109117974A

  • Power utilization user payment risk rating method based on convolutional neural network

    CN110298574A

  • Distribution network weak link identification method based on user importance and equipment operation state

    CN110378610A