CN-121980462-A - New energy-containing distribution network OCS system abnormal data detection method

CN121980462ACN 121980462 ACN121980462 ACN 121980462ACN-121980462-A

Abstract

The invention discloses a new energy distribution network-containing OCS system abnormal data detection method, which comprises the steps of firstly analyzing abnormal data types of the new energy distribution network-containing OCS system according to a distribution network OCS system frame, selecting typical data types, secondly adopting a WOA-PSO to jointly optimize a hidden layer structure and weight parameters of an ELM-AE model so as to improve feature extraction and model training effects, then carrying out unsupervised feature extraction on the multi-source operation data through the optimized ELM-AE model based on the distribution network OCS system multi-source operation data, calculating reconstruction errors, finally designing dynamic threshold judgment detection data based on the reconstruction errors, selecting an abnormal detection evaluation index, and evaluating the accuracy of the ELM-AE model abnormal detection. The method is superior to the traditional detection model in the aspects of identification precision, instantaneity and robustness, and can effectively support real-time anomaly monitoring and operation safety guarantee of the OCS system containing the new energy distribution network in a high-uncertainty environment.

Inventors

YANG JIAYU
ZHENG CHANYAN
LIN XIUJING
XU ZIJIE
CAI LIMIN
LIN HANTIAN

Assignees

广东电网有限责任公司潮州供电局

Dates

Publication Date: 20260505
Application Date: 20260126

Claims (7)

1. A method for detecting abnormal data of an OCS system containing new energy distribution network is characterized in that, Analyzing abnormal data types of the OCS system containing the new energy distribution network, and selecting typical data types; Adopting a WOA-PSO algorithm to jointly optimize a hidden layer structure and weight parameters of an ELM-AE model; Based on multisource operation data of the distribution network OCS system, carrying out unsupervised feature extraction on the multisource operation data through an optimized ELM-AE model, and calculating a reconstruction error; based on the reconstruction error, dynamic threshold judgment detection data are designed, and an abnormality detection evaluation index is selected to evaluate the accuracy of the abnormality detection of the ELM-AE model.
2. The method for detecting abnormal data of OCS system with new energy distribution network according to claim 1, wherein the selected typical data types include: photovoltaic inverter power, node voltage amplitude, energy storage SOC, feeder current, PMU phase angle difference, wind power, breaker status, and other data.
3. The method for detecting abnormal data of the OCS system with the new energy distribution network according to claim 1, wherein the hidden layer structure and the weight parameters of the ELM-AE model are jointly optimized by using a WOA-PSO algorithm, and the process is as follows: Step a, initializing a WOA-PSO algorithm population and population parameters, calculating a WOA-PSO algorithm population fitness value, and randomly initializing a particle position and a particle speed according to a particle swarm algorithm; step b, updating whale surrounding coefficient vector And Selecting a random number P; c, judging whether P is smaller than 0.5, if so, continuing to judge, if not, adopting spiral bubble attack to update the individual position of whale, and executing step e; Step d, judging If not, updating the individual position of whale by searching the hunting object, and executing the step e; And e, calculating an ELM-AE model hidden layer structure and weight parameters, calculating individual fitness, judging whether conditions are met, outputting a parameter optimization result if the conditions are met, and otherwise, executing the step b.
4. The method for detecting abnormal data of the new energy-containing distribution network OCS system according to claim 1, wherein the reconstruction error is calculated by performing unsupervised feature extraction on the multi-source operation data through an optimized ELM-AE model based on the multi-source operation data of the distribution network OCS system: For each sample set Assuming the hidden layer has L neurons, an activation function is used The output of the ELM-AE model is expressed as: ; In the formula, Representing the reconstruction result output by the ELM-AE model, wherein L is the number of neurons of the hidden layer, Is the first The input weight vector of the individual hidden layer neurons, Is the first A hidden layer neuron bias; Is the first A weight vector of the hidden layer neurons connected to the output layer; Calculating the response of each sample in the hidden layer by using a matrix form and adopting an activation function g (.) to obtain a hidden layer output matrix: ; In the formula, For the hidden layer output matrix, X is the target output matrix, To input the connection weights of the layer to the hidden layer, Is a bias vector; Outputting hidden layer As an input, the output target is still the original input X, and the weight of the output layer is solved by minimizing the square reconstruction error of the output layer and adopting Moore-Penrose pseudo-inverse method analysis: ; In the formula, In order to output the layer weight matrix, A pseudo-inverse matrix of the output matrix for the hidden layer; The self-encoder extracts the effective characteristics of the input data by compressing and reconstructing the input data, and the self-encoder is utilized to complete the reconstruction of the input: ; then error is reconstructed The calculated expression of (2) is: 。
5. The method for detecting abnormal data of the new energy distribution network-containing OCS system according to claim 4, wherein the process of detecting and solving abnormal type data based on the optimized ELM-AE model is as follows: acquiring OCS system sample data according to sensor data transmission, and dividing training sets and test sets of different abnormal types of data; Initializing a WOA-PSO algorithm population and population parameters, designing an individual coding structure, expressing a group of ELM-AE model parameters by each individual, and calculating a WOA-PSO algorithm population fitness value; Solving ELM-AE model parameters according to whale swarm algorithm, and outputting parameter optimization results; And inputting the test set data into an ELM-AE model for training according to the parameter optimization result, extracting hidden layer characteristics and completing reconstruction, and calculating the reconstruction error of each type of data.
6. The method for detecting abnormal data of the OCS system with the new energy distribution network according to claim 1, wherein the method for designing dynamic threshold judgment detection data based on the reconstruction error is characterized by adopting the following modes: Dynamic threshold The calculation mode based on normal samples and statistical distribution is adopted: ; In the formula, As the mean value of the sample, Is the standard deviation of the two-dimensional image, Is a sensitivity parameter, and is compared with a dynamic threshold according to the obtained reconstruction error result if If it is determined that the detected data is abnormal, if And judging that the detection data is normal.
7. The method for detecting abnormal data of OCS system including new energy distribution network according to claim 6, wherein the selecting abnormal detection evaluation index comprises accuracy rate Recall rate of 、 Value and detection rate False detection rate ; Selecting an anomaly detection evaluation index to evaluate the accuracy of the ELM-AE model on anomaly detection, and adopting the following calculation mode: ; ; ; ; ; In the above-mentioned method, the step of, In order to detect the correct number of abnormal samples, For the correct number of tests in a normal sample, The number of errors is detected for the normal data, Is the number of false detections in the abnormal sample.

Description

New energy-containing distribution network OCS system abnormal data detection method Technical Field The invention relates to the technical field of abnormal data detection of power distribution networks, in particular to a method for detecting abnormal data of an OCS (online distribution System) system containing new energy. Background At present, many results are achieved in the aspect of detection of abnormal data of a power distribution network. The traditional abnormal data detection method mainly focuses on statistical methods, state evaluation and various artificial intelligent algorithms, such as K-means algorithm, support Vector Machine (SVM), neural network, long-term short-term memory network (LSTM) and the like, which are widely applied to abnormal data detection. Aiming at the problems of time delay and experience dependence in power dispatching monitoring data, a basic isolated forest detection model is built, an abnormal state and buffer capacity dynamic update model is combined, a logarithmic interval isolation strategy is designed based on a mahalanobis distance by considering data distribution difference, a detection model is built through multi-subtree integration, a distributed density peak value clustering algorithm is provided according to a clustering center selection principle and difference statistics, a load abnormality detection method for optimizing a fuzzy C-means clustering model based on an improved particle swarm algorithm is provided, a K-means clustering center initialization mechanism is improved from data characteristics, and the performance of the fuzzy C-means clustering model in power abnormality detection is enhanced. The method realizes the anomaly detection of the power data to a certain extent, but when facing millions to tens of millions of single sample data, the model training and testing depend on a large number of samples, and the calculation cost is high. Meanwhile, the existing method is mostly aimed at a single data type, and is difficult to effectively cope with the abnormal detection task of the multi-component heterogeneous data. An OCS system abnormal data detection method containing new energy distribution network is designed to solve the problems. Disclosure of Invention The invention aims to solve the defect that the prior art lacks an accurate and effective detection means for the multi-element heterogeneous data characteristics of the OCS system containing the new energy distribution network, and provides the method for detecting the abnormal data of the OCS system containing the new energy distribution network, which realizes the real-time abnormal monitoring and operation safety guarantee of the multi-element heterogeneous data in a high-uncertainty environment. In order to achieve the above purpose, the present invention adopts the following technical scheme: An OCS system abnormal data detection method for a new energy-containing distribution network comprises the following steps: Analyzing abnormal data types of the OCS system containing the new energy distribution network, and selecting typical data types; Adopting a WOA-PSO algorithm to jointly optimize a hidden layer structure and weight parameters of an ELM-AE model; Based on multisource operation data of the distribution network OCS system, carrying out unsupervised feature extraction on the multisource operation data through an optimized ELM-AE model, and calculating a reconstruction error; based on the reconstruction error, dynamic threshold judgment detection data are designed, and an abnormality detection evaluation index is selected to evaluate the accuracy of the abnormality detection of the ELM-AE model. Further preferably, the selected typical data type includes: photovoltaic inverter power, node voltage amplitude, energy storage SOC, feeder current, PMU phase angle difference, wind power, breaker status, and other data. Further preferably, the process of jointly optimizing the hidden layer structure and the weight parameters of the ELM-AE model by adopting the WOA-PSO algorithm is as follows: Step a, initializing a WOA-PSO algorithm population and population parameters, calculating a WOA-PSO algorithm population fitness value, and randomly initializing a particle position and a particle speed according to a particle swarm algorithm; step b, updating whale surrounding coefficient vector AndSelecting a random number P; c, judging whether P is smaller than 0.5, if so, continuing to judge, if not, adopting spiral bubble attack to update the individual position of whale, and executing step e; Step d, judging If not, updating the individual position of whale by searching the hunting object, and executing the step e; And e, calculating an ELM-AE model hidden layer structure and weight parameters, calculating individual fitness, judging whether conditions are met, outputting a parameter optimization result if the conditions are met, and otherwise, executing the step b. Further preferably