Power distribution network topology identification method and device based on intelligent circuit breaker
By using decision trees and deep neural network models based on smart circuit breakers to filter key feature data, the problems of fast, accurate, and low-cost topology identification of distribution networks are solved, and efficient topology identification in dynamic environments is achieved.
Patent Information
- Application Number
- CN202510822266.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-19
- Publication Date
- 2025-11-07
AI Technical Summary
Existing distribution network topology identification technologies struggle to achieve fast, accurate, and low-cost identification, especially in dynamically changing distribution networks where it is difficult to track load switching and tie switch actions in real time.
A method based on intelligent circuit breakers is adopted, which uses decision trees to filter key feature data and deep neural network models for topology identification. Combined with gradient boosting algorithm and deep learning technology, efficient analysis of electrical data is achieved.
It enables rapid and accurate identification of distribution network topology without requiring a large number of monitoring nodes and grid-injected signals, improving identification efficiency and accuracy, and adapting to various operating conditions.
Smart Images
Figure CN120914972A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of power distribution network topology identification, in particular to a power distribution network topology identification method and device based on intelligent circuit breakers. BACKGROUND
[0002] In the power distribution network, topology identification is a basic ability with wide demand. Accurate topology identification can provide important support for state analysis, power flow calculation, fault diagnosis, fault isolation, and network self-healing of the power distribution network. However, at present, due to the high complexity and high dynamics of the power distribution network, the rapid and accurate identification of the power distribution network topology has always been a difficult problem.
[0003] The current power distribution network topology identification technology can be at least divided into the following four types of schemes according to the implementation principle:
[0004] (1) Measurement data analysis scheme based on smart meters
[0005] This technology relies on smart meters to build a data acquisition network at the user side, and through high-frequency acquisition of voltage effective value, current phase angle, active / reactive power and other electrical quantity data of user nodes, the core algorithm adopts Pearson correlation coefficient method, mutual information entropy analysis, dynamic time warping (DTW) algorithm to calculate the space-time correlation of different node electrical quantities. For example, when the voltage fluctuation curves of two nodes show strong correlation, it can be determined that they are in the same power supply branch. Some improved schemes will introduce a synchronous phasor measurement unit (PMU) to realize accurate phase difference measurement through microsecond-level timestamp alignment, and further deduce the topology connection order. This scheme has strict requirements for the deployment density and data quality of the meter. In areas where the meter is not fully covered, topology inference will appear chain breakage error due to data missing. When there is communication delay or data packet loss, the voltage phase data collected asynchronously will produce cumulative error, causing misjudgment of the upstream and downstream relationship.
[0006] (2) Active detection scheme based on impedance spectrum feature detection
[0007] In the substation or distribution cabinet, a dedicated signal generator (such as a sweep impedance analyzer) is deployed to inject a test signal of a specific frequency into the line, and the signal propagation attenuation characteristics, reflection coefficient and impedance change in the transmission line are measured through multi-point arranged sensors. The line impedance matrix model is established by using the transmission line theory, and the line topology structure is inversely calculated by combining the frequency domain response curve matching algorithm. The advanced version will use time domain reflectometry technology to accurately calculate the line length and branch point position by analyzing the time delay of the reflected wave at the end of the conductor. This method needs to interrupt the power supply to install the test equipment, which leads to a sharp increase in implementation cost, and the measurement process is easily affected by the stray capacitance of the line, parallel compensation devices and other parasitic parameters. Especially in the distribution network containing power electronic equipment (such as photovoltaic inverters and SVG), the high-frequency test signal will produce serious harmonic interference, causing the impedance measurement value to deviate from the true value by more than 20%. In addition, this method can only obtain a static topology snapshot and cannot track dynamic changes such as load switching and tie switch action in real time, making it difficult to meet the needs of intelligent power grid topology updating at the minute level.
[0008] (3) Channel feature analysis scheme based on power line carrier communication
[0009] The channel impulse response (CIR), signal-to-noise ratio (SNR), multipath time delay spread and other physical layer parameters are synchronously collected during communication by using the deployed power line carrier communication (PLC) system. The mapping relationship between channel features and topology structure is established by machine learning model. Some systems will build a channel fingerprint database and use K-nearest neighbor (KNN) algorithm for real-time matching and retrieval. Due to the strong uncertainty of the transmission characteristics of carrier signals in complex distribution networks, the impedance mutation at the branch point of the line will cause the signal to produce multipath effect, resulting in a sharp fluctuation of time delay spread parameters, and the branch containing nonlinear load will cause signal harmonic distortion, so that the mapping relationship between channel impulse response features and topology is distorted, and therefore the reliability of this method is defective.
[0010] (4) Load mode mining scheme based on deep learning
[0011] This scheme converts topology identification into a time series classification problem, and constructs a graph convolution network (GCN) or long short-term memory network (LSTM) model by long-term collection of node load curves (usually containing tens of dimensions of characteristics such as daily load rate, power factor, harmonic distortion rate, etc.). During the training process, the network automatically learns the power propagation characteristics under different topologies, such as the lagging influence mode of upstream node power fluctuation on downstream nodes. Some studies introduce a transfer learning framework, pre-train the model using simulation-generated topology-load data sets, and then fine-tune the parameters using a small amount of real data. Since the dimension of the electrical data analyzed for the distribution network is often too high, the efficiency of data-driven modeling and reasoning is generally very low. For example, using SVN to analyze the topology of multiple collection points and high-dimensional data often cannot complete the task within a tolerable time. In addition, load curves are easily disturbed by user power consumption behavior (such as random charging of electric vehicles), and these noise characteristics unrelated to topology can mislead the model attention mechanism, resulting in false alarms or missed detection. SUMMARY
[0012] The technical problem to be solved by the present application is to provide a power distribution network topology identification method and device based on intelligent circuit breakers, which can realize non-invasive, low-cost, fast and accurate identification of the topology of the power distribution network.
[0013] The technical solution adopted by the present application to solve the technical problem is to provide a power distribution network topology identification method based on intelligent circuit breakers, comprising the following steps:
[0014] Constructing a decision tree based on each electrical data collected by the intelligent circuit breaker in the power distribution network to be tested, and determining key feature data from the electrical data collected by the intelligent circuit breaker based on a gradient boosting decision tree method;
[0015] Obtaining the key feature data of each intelligent circuit breaker in the power distribution network to be tested;
[0016] Inputting the key feature data of each intelligent circuit breaker in the power distribution network to be tested into a topology identification model to obtain the network topology of the power distribution network to be tested.
[0017] The key feature data is determined from the data collected by the intelligent circuit breaker based on the gradient boosting decision tree method, specifically comprising:
[0018] Arranging each sample of the decision tree in descending order according to the gradient absolute value;
[0019] In the sorted result, the first s1x n samples are selected to form a large gradient sample subset W1, and for the remaining samples, s2x n samples are randomly selected to form a small gradient sample subset W2, wherein n is the total number of samples, and 0
[0020] merge the large gradient sample subset W1 and the small gradient sample subset W2, and construct a new decision tree based on the merged sample set, and calculate the information gain of each node in the new decision tree;
[0021] Repeat the above steps until the termination condition is reached, and select the samples corresponding to the K nodes with the maximum information gain as the key feature data.
[0022] The calculation method of the information gain is: Wherein, I j (f) is the information gain, x i represents the ith sample, x i,j ≤f represents that the split feature j of the sample x i is less than or equal to the split point f, x i,j >f represents that the split feature j of the sample x i is greater than the split point f, g i is the negative gradient of the loss function of the ith sample after each iteration, r(f) represents the number of split features j of the sample x i less than or equal to the split point f, q(f) represents the number of split features j of the sample x i greater than the split point f.
[0023] The topology identification model is trained based on a deep neural network model, and the deep neural network model has five hidden layers, and the number of neurons in each hidden layer is 30, 25, 60, 100 and 50 respectively.
[0024] The topology identification model is trained in the following way:
[0025] Collect the historical key feature data of each intelligent circuit breaker in the power distribution network and the network topology structure;
[0026] Preprocess the historical key feature data and network topology results to form complete key feature profile data and network topology data at each intelligent circuit breaker at each time point;
[0027] Label the complete key feature profile data at each intelligent circuit breaker at each time point to form a labeled sample set, and divide the labeled sample set into a training set and a validation set;
[0028] Train the deep neural network model using the training set, and after training, use the validation set for verification, and use the verified model as the topology identification model.
[0029] The preprocessing of the historical key feature data and network topology results includes missing value completion processing and error value deletion processing.
[0030] The complete key feature profile data at each intelligent circuit breaker at each time point is labeled, specifically: the network topology data is divided into m categories, and the category of the network topology structure at the time corresponding to each key feature profile data is taken as the label of the key feature profile data.
[0031] The technical scheme adopted by the present application to solve its technical problems is to provide a power distribution network topology identification device based on intelligent circuit breakers, comprising:
[0032] A determination module is configured to construct a decision tree based on each electrical data collected by the intelligent circuit breakers in the power distribution network to be tested, and determine key feature data from the electrical data collected by the intelligent circuit breakers using a decision tree method based on gradient boosting.
[0033] An acquisition module is configured to acquire key feature data of each intelligent circuit breaker in the power distribution network to be tested.
[0034] An identification module is configured to input the key feature data of each intelligent circuit breaker in the power distribution network to be tested into a topology identification model to obtain the network topology of the power distribution network to be tested.
[0035] The technical scheme adopted by the present application to solve its technical problems is to provide an electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the power distribution network topology identification method based on intelligent circuit breakers.
[0036] The technical scheme adopted by the present application to solve its technical problems is to provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program is executable by a processor to implement the steps of the power distribution network topology identification method based on intelligent circuit breakers.
[0037] Advantages
[0038] Due to the adoption of the above technical scheme, the present application has the following advantages and positive effects compared with the prior art: the present application considers effectively reducing the dimensionality of high-dimensional data while ensuring accuracy, and filters out a large number of sample components with low importance in power distribution network topology identification through intelligent algorithms, thereby improving the efficiency and accuracy of identification. The present application does not need to deploy a large number of monitoring nodes in the network and does not need to inject power or signals into the power grid, but only uses a small number of low-voltage intelligent circuit breakers deployed on some branch nodes. Through in-depth analysis of electrical parameters of the power distribution network, the topology under various general working conditions can be determined. BRIEF DESCRIPTION OF DRAWINGS
[0039] Figure 1 is a flowchart of the power distribution network topology identification method based on intelligent circuit breakers according to the first embodiment of the present application. DETAILED DESCRIPTION
[0040] The application will be further described below in connection with specific embodiments. It should be understood that these embodiments are only used to illustrate the application and not used to limit the scope of the application. Furthermore, it should be understood that after reading the content of the application, those skilled in the art can make various modifications or changes to the application, and these equivalent forms also fall within the scope defined by the appended claims.
[0041] The first embodiment of the application relates to a power distribution network topology identification method based on intelligent circuit breakers, which includes two parts of model training and model identification, as shown in Figure 1 , specifically comprising:
[0042] Step 1, based on the electrical data collected by the intelligent circuit breakers in the power distribution network to be tested, a topology identification model is constructed, and based on the electrical data collected by the intelligent circuit breakers in the power distribution network to be tested, a decision tree is constructed, and the decision tree method based on gradient boosting is used to determine the key feature data from the electrical data collected by the intelligent circuit breakers.
[0043] This step specifically includes:
[0044] Data collection:
[0045] There are several intelligent circuit breakers in the power distribution network, and these intelligent circuit breakers only need to be deployed in part of the branch nodes in the power distribution network, that is, a large number of nodes in the power distribution network can be existing ordinary circuit breakers and do not need to be replaced. Each intelligent circuit breaker D i collects local electrical parameters, including: voltage amplitude V i , injected active power P i , injected reactive power Q i , etc. After collection, these data are sent to a collection point E together with the current opening and closing state Z i of the intelligent circuit breaker. The collection point E can be an edge side intelligent fusion terminal or a cloud platform, etc. At the same time, the network topology structure at that time is recorded at the collection point E.
[0046] Data preprocessing:
[0047] The data collected at the collection point E are preprocessed to complete missing values and exclude error values, etc., to form complete electrical profile data and network topology data at each intelligent circuit breaker at each time point.
[0048] Data labeling:
[0049] The complete electrical profile data at each intelligent circuit breaker at each time point is labeled, that is, the recorded network topology is divided into m different categories, and the network topology category at the time corresponding to each electrical profile data is taken as the label of the electrical profile data. After marking, the labeled sample set C is formed, and the labeled sample set C is further divided into a training set and a validation set.
[0050] Since the dimension of the distribution network electrical data is often high, the efficiency of data-driven modeling and reasoning is low. To solve this problem, the topological recognition features are screened, that is, the various electrical data collected by each intelligent circuit breaker is analyzed and judged, and the key feature data important for topological recognition is screened out.
[0051] The gradient boosting-based decision tree method is used to calculate the importance of each electrical data in the embodiment, so as to screen out the key feature data, which is specifically as follows:
[0052] On the training set X={x1, x2, x3,..., x n}, the gradient boosting-based decision tree method is used, and the negative gradient corresponding to each round of iteration loss function is denoted as G={g1, g2, g3,..., g n}. The decision tree is constructed based on the training set X={x1, x2, x3,..., x n}, wherein x i represents the i-th sample of the decision tree, that is, the i-th electrical data.
[0053] In the embodiment, the large gradient samples are sampled at a sampling rate of s1, and the small gradient samples are sampled at a sampling rate of s2, and 0
[0054] (1) Set the training termination condition, such as reaching the accuracy threshold or the maximum number of iterations.
[0055] (2) Arrange each sample in the decision tree in descending order according to the gradient absolute value; in the embodiment, the gradient can be understood as an index of the influence degree of the data sample on the model prediction result, and the sample with large gradient absolute value means that it has more influence on the training direction adjustment of the model.
[0056] (3) In the sorted result, the first s1x n samples are selected to form a large gradient sample subset W1;
[0057] (4) For the remaining samples, s2x n samples are randomly selected to form a small gradient sample subset W2;
[0058] (5) Merge the large gradient sample subset W1 and the small gradient sample subset W2, that is, W newW1∪W2, and based on the merged sample set W new A new decision tree is constructed, and the information gain of each node in the new decision tree is calculated.
[0059] In the calculation of the information gain of each node, in order to balance the influence of large gradient samples and small gradient samples in the decision tree splitting process, the weight of small gradient samples is set to (1-s1) / s2, and the information gain of the split point f of the split feature j is:
[0060]
[0061] Where x i,j ≤f indicates that the split feature j of the sample x i is less than or equal to the split point f, and x i,j >f indicates that the split feature j of the sample x i is greater than the split point f. r(f) indicates the number of split features j of the sample x i less than or equal to the split point f, and q(f) indicates the number of split features j of the sample x i greater than the split point f, i.e.:
[0062] (6) Repeat steps (2)-(5) above until the training end condition set in step (1) is reached.
[0063] The sum of the information gain of all nodes of a feature as a split feature represents the importance of the feature to the result. The samples corresponding to the K nodes with the largest information gain are selected as key feature data, so that K key feature data can be obtained, denoted as B={B1, B2, B3,..., B K}.
[0064] Topology identification model training:
[0065] In training the topology identification model, first set the training termination condition. The embodiment can set the loss function value reaching a threshold value φ L or the number of iterations reaching φ t as the termination condition.
[0066] Deep neural network model is adopted for training: a typical model architecture adopted by the embodiment, deep neural network model, has 5 hidden layers, and the number of neurons in each hidden layer is 30, 25, 60, 100 and 50 respectively.
[0067] In the training, the input of the deep neural network model is: the components of each sample in the training set projected on each important feature dimension output by the gradient boosting based decision tree method. The adopted supervision signal is: the label in the current data, that is, the category of the current network topology. The training set is used for training, and after the training termination condition is reached, the training is ended, and the validation set is used for verification, and if the verification is passed, the topology identification model Q is obtained.
[0068] Step 2, obtaining the key feature data of each intelligent circuit breaker in the to-be-tested power distribution network.
[0069] In this step, the key feature data B = {B1, B2, B3,..., B K} obtained by the gradient boosting based decision tree method is used as the input of the deep neural network model Q. ij} are collected by each intelligent circuit breaker. Then the collected key feature data are collected to the collection point E. The collection point E can be an intelligent circuit breaker, an edge side intelligent fusion terminal, or a cloud platform, etc.
[0070] Step 3, inputting the key feature data of each intelligent circuit breaker in the to-be-tested power distribution network into the topology identification model to obtain the network topology of the to-be-tested power distribution network.
[0071] In this step, the key feature data F = {x ij |i∈[1,K],j∈[1,n]} are input into the topology identification model Q, and the probability H = [h i ] 1＜i＜m corresponding to the topology structure of the current to-be-tested power distribution network is calculated. The maximum value h j in these probabilities is compared with the threshold value φ1. If h j > φ1, it is inferred that the current power distribution network topology is the working condition topology corresponding to the label j. When the calculated probabilities corresponding to all general working condition topologies of the to-be-tested power distribution network are all less than the threshold value φ2, it is inferred that the topology of the to-be-tested power distribution network is a general working condition topology that has not been seen in the training process, at this time, the "unknown new topology" identifier is output. At the same time, the current input data is recorded.
[0072] In actual application, the training of B = {B1, B2, B3,..., B K} obtained by the gradient boosting based decision tree method and the topology identification model Q can be iterated continuously. In actual topology identification and monitoring, the important feature components and the deep neural network can be continuously iterated and updated through the continuous electrical data in the actual power distribution network, so that they can have faster and better sensitivity and judgment accuracy to the new environment and changes.
[0073] In particular, for the actual sample of "unknown new topology" given by the topology identification model Q, the exact network topology at the time is further determined, then the set of general working condition topologies is expanded, and this topology is added, and the electrical data sample at the time is labeled, and then the labeled sample is added to the sample set C.
[0074] It can be found that the application considers effectively reducing the dimension of high-dimensional data while ensuring accuracy, and filters out a large number of sample components with low importance in distribution network topology identification through intelligent algorithms, thereby improving the efficiency and accuracy of identification. The application does not need to deploy a large number of monitoring nodes in the network and does not need to inject power or signals into the power grid, but only uses a small number of low-voltage intelligent circuit breakers deployed on some branch nodes. Through in-depth analysis of electrical parameters of the distribution network, the topology under various general working conditions can be determined.
[0075] The second embodiment of the application relates to a distribution network topology identification device based on an intelligent circuit breaker, comprising:
[0076] A determination module is configured to construct a decision tree based on each electrical data collected by the intelligent circuit breaker in the distribution network to be tested, and determine key feature data from the electrical data collected by the intelligent circuit breaker by using a decision tree method based on gradient boosting;
[0077] An acquisition module is configured to acquire the key feature data of each intelligent circuit breaker in the distribution network to be tested;
[0078] An identification module is configured to input the key feature data of each intelligent circuit breaker in the distribution network to be tested into a topology identification model to obtain the network topology of the distribution network to be tested.
[0079] The determination module comprises:
[0080] A sorting unit is configured to sort each sample of the decision tree in descending order according to the absolute value of the gradient;
[0081] A selection unit is configured to select the first s1x n samples in the sorted result to form a large gradient sample subset W1, and randomly select s2x n samples from the remaining samples to form a small gradient sample subset W2, wherein n is the total number of samples, and 0
[0082] A merging and calculating unit is configured to merge the large gradient sample subset W1 and the small gradient sample subset W2, construct a new decision tree based on the merged sample set, and calculate the information gain of each node in the new decision tree;
[0083] The repeated selection unit is configured to repeat the operations of the above units until a termination condition is reached, and select samples corresponding to K node pairs with the largest information gain as the key feature data.
[0084] The merging calculation unit calculates the information gain of each node in the following manner: wherein, I j (f) is the information gain, x i represents the ith sample, x i,j ≤f represents that the split feature j of the sample x i is less than or equal to the split point f, x i,j >f represents that the split feature j of the sample x i is greater than the split point f, g i is the negative gradient of the loss function of the ith sample after each iteration, r(f) represents the number of split features j of the sample x i ≤f, and q(f) represents the number of split features j of the sample x i >f.
[0085] The topology identification model is trained based on a deep neural network model, and the deep neural network model has five hidden layers, and the number of neurons in each hidden layer is 30, 25, 60, 100 and 50, respectively.
[0086] The topology identification model is trained in the following manner:
[0087] The historical key feature data of each intelligent circuit breaker in the power distribution network and the network topology structure are collected.
[0088] The historical key feature data and the network topology result are preprocessed to form complete key feature profile data and network topology data at each intelligent circuit breaker at each time point.
[0089] The complete key feature profile data at each intelligent circuit breaker at each time point are labeled to form a labeled sample set, and the labeled sample set is divided into a training set and a validation set.
[0090] The deep neural network model is trained using the training set, and after the training is completed, the validation set is used for verification, and the model that passes the verification is used as the topology identification model.
[0091] The preprocessing of the historical key feature data and the network topology result includes missing value completion processing and error value deletion processing.
[0092] The complete key feature profile data at each intelligent circuit breaker at each time point is labeled, specifically: the network topology data is divided into m categories, and the category of the network topology structure at the time corresponding to each key feature profile data is taken as the label of the key feature profile data.
[0093] The third embodiment of the present application relates to an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the intelligent circuit breaker-based power distribution network topology identification method of the first embodiment when executing the computer program.
[0094] The fourth embodiment of the present application relates to a computer-readable storage medium, which stores a computer program, and the computer program implements the steps of the intelligent circuit breaker-based power distribution network topology identification method of the first embodiment when executed by a processor.
[0095] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage, etc.) containing computer-usable program code.
[0096] The present application is described with reference to flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks. Figure 1 The means for implementing the functions specified in one or more flows and / or blocks.
[0097] These computer program instructions can also be stored in a computer-readable storage medium, which can guide the computer or other programmable data processing devices to work in a specific manner, so that the instructions stored in the computer-readable storage medium produce a product including an instruction method, which implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks. Figure 1 The means for implementing the functions specified in one or more flows and / or blocks.
[0098] These computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are generated to realize the computer-implemented processes in the computer or other programmable devices, and the instructions executed in the computer or other programmable devices provide the steps for implementing the functions specified in the flowchart Figure 1 one flow or multiple flows and / or the functions specified in the block Figure 1 one flow or multiple flows and / or the functions specified in the block
[0099] The above description is merely a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.
Claims
1. A method for identifying topology of power distribution network based on intelligent circuit breaker, characterized in that, The method comprises the following steps: Based on the electrical data collected by the intelligent circuit breaker in the to-be-tested power distribution network, a decision tree is constructed, and a decision tree method based on gradient boosting is used to determine the key feature data from the electrical data collected by the intelligent circuit breaker; Obtain the key feature data of each intelligent circuit breaker in the to-be-tested power distribution network; Input the key feature data of each intelligent circuit breaker in the to-be-tested power distribution network into a topology identification model to obtain the network topology of the to-be-tested power distribution network.
2. The smart circuit breaker based power distribution network topology identification method of claim 1, wherein, The decision tree method based on gradient boosting is used to determine the key feature data from the data collected by the intelligent circuit breaker, specifically comprising: Arrange the samples of the decision tree in descending order according to the gradient absolute value; In the sorted result, select the first s1×n samples to form a large gradient sample subset W1, and randomly select s2×n samples from the remaining samples to form a small gradient sample subset W2, wherein n is the total number of samples, and 0 Merge the large gradient sample subset W1 and the small gradient sample subset W2, and construct a new decision tree based on the merged sample set, and calculate the information gain of each node in the new decision tree; Repeat the above steps until the termination condition is reached, and select the samples corresponding to the K nodes with the largest information gain as the key feature data.
3. The smart circuit breaker based power distribution network topology identification method as claimed in claim 2, wherein, The calculation method of the information gain is: where I j (f) is the information gain, x i represents the i-th sample, x i,j ≤f represents that the split feature j of the sample x i is less than or equal to the split point f, x i,j >f represents that the split feature j of the sample x i is greater than the split point f, g i is the negative gradient of the loss function of the i-th sample after each iteration, r(f) represents the number of split features j of the sample x i less than or equal to the split point f, and q(f) represents the number of split features j of the sample x i greater than the split point f.
4. The smart circuit breaker based power distribution network topology identification method as claimed in claim 1, wherein, The topology identification model is trained based on a deep neural network model, and the deep neural network model has five hidden layers, and the number of neurons in each hidden layer is 30, 25, 60, 100 and 50.
5. The smart circuit breaker based power distribution network topology identification method as claimed in claim 1, wherein, The topology identification model is trained by the following method: Collect the historical key feature data of each intelligent circuit breaker in the power distribution network and the network topology structure; Preprocess the historical key feature data and network topology results to form complete key feature profile data and network topology data of each intelligent circuit breaker at each time point; Label the complete key feature profile data of each intelligent circuit breaker at each time point to form a labeled sample set, and divide the labeled sample set into a training set and a validation set; Train the deep neural network model using the training set, and verify the model after training using the validation set, and use the verified model as the topology identification model.
6. The smart circuit breaker based power distribution network topology identification method as claimed in claim 5, wherein, The preprocessing of the historical key feature data and network topology results includes missing value completion processing and error value deletion processing.
7. The smart circuit breaker based power distribution network topology identification method as claimed in claim 5, wherein, The labeling of the complete key feature profile data of each intelligent circuit breaker at each time point comprises: dividing the network topology data into m categories, and taking the category of the network topology structure at the time corresponding to each key feature profile data as the label of the key feature profile data.
8. A smart circuit breaker based power distribution network topology identification apparatus, characterized in that, Comprise: The determination module is configured to construct a decision tree based on the electrical data collected by the intelligent circuit breaker in the to-be-tested power distribution network, and to determine the key feature data from the electrical data collected by the intelligent circuit breaker using a decision tree method based on gradient boosting; The acquisition module is configured to obtain the key feature data of each intelligent circuit breaker in the to-be-tested power distribution network; An identification module is configured to input the key feature data of each intelligent circuit breaker in the to-be-tested power distribution network into a topology identification model to obtain a network topology of the to-be-tested power distribution network.
9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the intelligent circuit breaker-based power distribution network topology identification method according to any one of claims 1-7.
10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the intelligent circuit breaker-based power distribution network topology identification method according to any one of claims 1-7.