CN-120566456-B - Power network state identification method and system based on data feature analysis

CN120566456BCN 120566456 BCN120566456 BCN 120566456BCN-120566456-B

Abstract

The application provides a power network state identification method and a system based on data feature analysis, which are used for continuously improving the power network state identification precision through a data-driven optimization mechanism, providing full-flow intelligent decision support for power grid operation and maintenance and effectively promoting the mode transition of a power network from passive operation and maintenance to active prevention and control. The method comprises the steps of obtaining a multi-dimensional power data set of a target power network, carrying out dynamic feature extraction processing on the multi-dimensional power data set, generating a power network operation feature set, inputting the power network operation feature set into a power state identification model which completes debugging, carrying out joint analysis on voltage fluctuation features, current phase deviation features and equipment health features through the power state identification model, generating a power network state classification result, generating a network operation and maintenance strategy according to the power network state classification result, and sending the network operation and maintenance strategy to a power dispatching terminal to activate corresponding equipment maintenance operation or load adjustment operation.

Inventors

ZHOU JIANBO
CHAI YUAN
LIU LEI
YUAN CHAO
GAO KAIQIANG
HUO JIAHAO
GUAN TI
ZHANG KUN
BAI YINGWEI
LIANG DONG

Assignees

山东思极科技有限公司
国网山东省电力公司
中国电力科学研究院有限公司

Dates

Publication Date: 20260508
Application Date: 20250527

Claims (9)

1. A method for identifying a power network state based on data feature analysis, the method comprising: Acquiring a multi-dimensional power data set of a target power network, wherein the multi-dimensional power data set comprises real-time operation data, historical load data and equipment state monitoring data, and the equipment state monitoring data is acquired through sensors deployed in power network nodes; Carrying out dynamic feature extraction processing on the multidimensional electric power data set to generate an electric power network operation feature set, wherein the electric power network operation feature set comprises a voltage fluctuation feature, a current phase deviation feature and a device health degree feature; Inputting the power network operation characteristic set into a power state identification model which completes debugging, and carrying out joint analysis on the voltage fluctuation characteristic, the current phase deviation characteristic and the equipment health degree characteristic through the power state identification model to generate a power network state classification result, wherein the classification result comprises a normal state, an abnormal early warning state and a fault warning state; Generating a network operation and maintenance strategy according to the power network state classification result, and sending the network operation and maintenance strategy to a power dispatching terminal to activate corresponding equipment maintenance operation or load adjustment operation; the process of generating the network operation and maintenance strategy according to the power network state classification result comprises the following steps: If the classification result is an abnormal early warning state, extracting the fluctuation times exceeding a first threshold value in the voltage fluctuation characteristic, the monitoring duration of the phase difference continuously exceeding a preset angle in the current phase deviation characteristic and the reduction rate of the insulation resistance value in the equipment health characteristic; Generating a composite operation and maintenance instruction comprising local load transfer, equipment infrared detection and line inspection frequency improvement according to the mapping relation between the fluctuation times and the equipment overload risk, the relevance between the monitoring time length and the line impedance abnormality and the corresponding rule of the descending rate and equipment aging; And if the classification result is a fault alarm state, generating an emergency response strategy comprising fault equipment isolation, standby power supply switching and maintenance personnel scheduling according to the identified equipment nodes with the vibration amplitude exceeding a second threshold value in the equipment health degree characteristics and the line sections with abrupt phase difference changes in the current phase shift characteristics.
2. The method of claim 1, wherein the process of performing dynamic feature extraction processing on the multi-dimensional power data set to generate a power network operational feature set comprises: performing waveform decomposition processing on the real-time operation data to obtain a fundamental component and a harmonic component, and generating a voltage fluctuation characteristic based on an amplitude stability index of the fundamental component and an amplitude change rate of the harmonic component; performing time sequence alignment processing on the historical load data, extracting a load change gradient of each power network node in a preset time window, and generating a current phase shift characteristic by calculating the difference of the load change gradients of adjacent time windows; detecting abnormal signals of the equipment state monitoring data, obtaining equipment vibration amplitude, temperature change rate and insulation resistance value, and carrying out weighted fusion processing on the ratio of the vibration amplitude to a preset safety threshold value, a time integral value of the temperature change rate and the attenuation rate of the insulation resistance value to generate the equipment health degree characteristic; and determining the power network operation characteristic set according to the voltage fluctuation characteristic, the current phase offset characteristic and the equipment health characteristic.
3. The method of claim 1, wherein the power state identification model is commissioned by: acquiring a historical power data set, wherein the historical power data set comprises a plurality of multidimensional power data samples at historical time points and corresponding state labeling labels; Performing feature extraction processing on the multidimensional power data samples at each historical time point to generate historical voltage fluctuation features, historical current phase deviation features and historical equipment health features, and combining the historical voltage fluctuation features, the historical current phase deviation features and the historical equipment health features into a historical feature vector set; Acquiring an initial state classification model, wherein the initial state classification model comprises a feature weight distribution layer and a state decision layer, and the feature weight distribution layer is used for dynamically adjusting weight coefficients according to the contribution degrees of different features in the historical feature vector set; And inputting the historical feature vector set into the initial state classification model for iterative debugging, calculating the degree of difference between the output of the state decision layer and the state labeling label through a preset loss function, and reversely updating the parameters of the feature weight distribution layer until the degree of difference is lower than a preset threshold value.
4. The method of claim 3, wherein the process of backward updating the parameters of the feature weight distribution layer comprises: based on corresponding timestamp distribution of the historical voltage fluctuation feature in a historical fault case, extracting accumulated frequencies of voltage deviation from a baseline value at each historical time point, and generating a first weight coefficient of the historical voltage fluctuation feature according to a proportional relation between the accumulated frequencies and a preset frequency threshold; Performing time sequence matching on the initial weight coefficient and the phase shift duration time of the historical current phase shift characteristic in the historical load abrupt change event, extracting an abnormal interval section with the phase shift duration time exceeding a preset duration time, and dynamically adjusting the weight coefficient of the historical current phase shift characteristic according to the coverage rate of the abnormal interval section accounting for the historical monitoring total duration time to obtain a second weight coefficient; Coupling analysis is carried out on the dynamically adjusted weight coefficient and a stability index of the historical equipment health degree characteristic in a continuous monitoring period, the stability index is obtained by calculating a linear combination of equipment vibration amplitude range, temperature change rate variance and insulation resistance value attenuation in an adjacent monitoring period, and the weight coefficient of the historical equipment health degree characteristic is corrected based on a mapping relation between the stability index and equipment fault early warning level to obtain a third weight coefficient; And inputting the first weight coefficient, the second weight coefficient and the third weight coefficient into a gradient descent algorithm of the characteristic weight distribution layer, and updating the associated weight parameters of the historical voltage fluctuation characteristic, the historical current phase deviation characteristic and the historical equipment health characteristic through synchronous iteration until the classification probability distribution output by the state decision layer and the error rate of the state labeling label are converged within a preset range.
5. The method of claim 1, wherein the executing of the compound operation and maintenance instruction comprises: Generating a load distribution scheme comprising load reduction amount of the target power network node, load receiving priority of the adjacent power network node and maximum through flow threshold of a transmission line according to transferable load capacity of the target power network node and load bearing margin of the adjacent power network node in the local load transfer instruction; Starting a breaker control unit deployed on a connecting line between a target power network node and an adjacent power network node based on the position identifiers of the target power network node and the adjacent power network node in the load distribution scheme, synchronously activating a distributed power supply output adjustment module associated with the target power network node, and executing dynamic matching operation of the load reduction amount and the receiving priority defined in the load distribution scheme; Triggering a thermal imaging sensor group arranged on the surfaces of a wiring terminal, an insulating sleeve and a radiator of target equipment according to the position information of the target equipment in an equipment infrared detection instruction, and collecting equipment surface temperature distribution data of the target equipment in a steady-state operation stage after load transfer; Comparing the three-phase joint temperature difference gradient, the radiator axial temperature attenuation rate and the radial hot spot distribution characteristics of the insulating sleeve in the equipment surface temperature distribution data with the reference temperature parameters of corresponding parts in the historical temperature distribution map pixel by pixel to generate an infrared detection analysis result comprising potential overheat region coordinates, temperature difference alarm levels and heat dissipation anomaly types; extracting topological connection relation of a line section where the target equipment is located and electrified states of adjacent intervals according to the coordinates of the potential overheat area and the temperature difference alarm level in the infrared detection analysis result, and calculating a safe approaching path and detection residence time of the inspection robot to the potential overheat area in the line inspection frequency lifting instruction; Based on the safe approach path and the detection stay time, the inspection period parameters of the inspection robot are updated, and the travelling speed, the image acquisition frequency and the obstacle avoidance distance threshold value in the inspection path planning parameters are re-planned, so that the inspection robot performs multi-angle thermal imaging data compensation and insulation defect positioning operation on the potential overheat region in the adjusted inspection period.
6. The method of claim 1, wherein the process of collecting the device state monitoring data comprises: collecting equipment mechanical vibration waveform data through a vibration sensor arranged on the surface of an electric equipment shell, carrying out Fourier transform processing on the equipment mechanical vibration waveform data, and extracting a target vibration frequency component matched with the rotation frequency of an equipment bearing and a corresponding amplitude value; Inputting the target vibration frequency component and the corresponding amplitude value into a preset vibration characteristic analysis window, and generating a device vibration intensity index based on the range of the amplitude value and the fluctuation cycle number in the vibration characteristic analysis window; acquiring multi-position temperature sampling data of equipment through a distributed temperature sensor array embedded between an internal winding of the power equipment and a radiating fin, and performing space alignment processing on the multi-position temperature sampling data to generate an internal temperature distribution matrix of the equipment; Calculating a thermal stability coefficient of the equipment based on the temperature gradient difference value and the maximum temperature difference duration time of adjacent monitoring points in the temperature distribution matrix inside the equipment; Collecting leakage current waveform data in real time through an insulation monitoring device arranged in a device grounding loop, performing high-pass filtering treatment on the leakage current waveform data, and extracting harmonic component amplitude values outside a power frequency period; Generating an equipment insulation degradation trend index according to the accumulated growth rate and the mutation frequency of the harmonic component amplitude in the continuous monitoring period; and carrying out normalization fusion on the equipment vibration intensity index, the equipment thermal stability coefficient and the equipment insulation degradation trend index to generate an equipment state monitoring data set.
7. The method of claim 6, wherein the dynamic feature extraction process further comprises: performing three-layer wavelet packet decomposition on the mechanical vibration waveform data of the equipment, extracting an energy distribution spectrum of a third-layer wavelet node, generating mechanical vibration energy entropy based on the energy ratio of a high-frequency sub-band to a low-frequency sub-band in the energy distribution spectrum, matching the mechanical vibration energy entropy with an energy entropy threshold interval of bearing abrasion and component looseness in a historical fault case library, and outputting a mechanical component health index; Performing three-dimensional interpolation reconstruction on the equipment internal temperature distribution matrix to generate an equipment internal three-dimensional temperature field distribution model, extracting a high-temperature area occupation ratio exceeding a preset safety temperature in the three-dimensional temperature field distribution model, calculating an equipment heat dissipation efficiency attenuation coefficient according to time sequence correlation of the high-temperature area occupation ratio and the rotating speed of an equipment heat dissipation fan, and generating a thermodynamic stability index by combining a radiator surface temperature uniformity index; Inputting the equipment insulation degradation trend index into a pre-trained insulation life prediction model, outputting an equipment insulation residual life estimated value, and generating an insulation aging correction coefficient based on the difference between the insulation residual life estimated value and the equipment operation age; And linearly superposing the mechanical part health index, the thermal stability index and the insulation ageing correction coefficient according to preset weights to generate an equipment comprehensive health score, and mapping the equipment comprehensive health score to a preset health degree grade interval to generate the equipment health degree characteristic.
8. The method of claim 1, wherein the method further comprises: Continuously collecting operation state data of overhaul equipment after the power dispatching terminal executes equipment overhaul operation; Inputting the running state data into the power state identification model, and verifying the accuracy of the classification result through the power state identification model to obtain a verification result; And if the verification results of the continuous preset times are inconsistent with the expected state labels, activating a model parameter calibration flow, and recalculating the weight distribution proportion of the voltage fluctuation characteristic, the current phase offset characteristic and the equipment health degree characteristic.
9. A power network state identification system comprising a processor and a memory, wherein the memory stores a computer program which, when executed by the processor, causes the processor to perform the steps of the method of any of claims 1 to 8.

Description

Power network state identification method and system based on data feature analysis Technical Field The application belongs to the technical field of power data analysis, and particularly relates to a power network state identification method and system based on data feature analysis. Background Modern power networks are used as complex energy transmission systems, and accurate identification of the running state of the power networks is a core technology for guaranteeing the reliability of power supply. Traditional state monitoring mainly relies on basic operation parameters such as voltage, current and the like acquired by a SCADA system, and abnormality detection is realized by combining a threshold alarming mechanism. With the development of new energy large-scale grid connection and electric power Internet of things technologies, the existing monitoring system gradually exposes the problems of single data dimension, insufficient analysis granularity and the like. In recent years, the industry has explored the processing of power data by machine learning algorithms, but is limited to single time scale feature analysis, and is difficult to effectively capture complex state features in dynamic operation of a power grid. Therefore, the existing solution has obvious application bottlenecks that on one hand, a static characteristic analysis system is difficult to adapt to the dynamic operation requirement of a modern power grid and cannot send effective early warning in the early stage of system state evolution, and on the other hand, a discretized state judgment mechanism lacks collaborative analysis capability on multi-source data, so that the generated operation and maintenance strategy is insufficient in matching degree with actual working conditions. When dealing with complex power grid faults, the prior art often needs manual intervention to perform feature association analysis, emergency disposal timeliness is obviously affected, active defense capacity of a smart power grid is severely restricted by the technical short plates, and how to build a more intelligent power network state identification system becomes a technical problem to be overcome at present. Disclosure of Invention The application provides a power network state identification method and a system based on data feature analysis, which are used for continuously improving the power network state identification precision through a data-driven optimization mechanism, providing full-flow intelligent decision support for power grid operation and maintenance and effectively promoting the mode transition of a power network from passive operation and maintenance to active prevention and control. In a first aspect, an embodiment of the present application provides a method for identifying a power network state based on data feature analysis, which is applied to a power network state identification system, where the method includes obtaining a multi-dimensional power data set of a target power network, where the multi-dimensional power data set includes real-time operation data, historical load data, and device state monitoring data, where the device state monitoring data is collected by a sensor disposed in a power network node, performing dynamic feature extraction processing on the multi-dimensional power data set to generate a power network operation feature set, where the power network operation feature set includes a voltage fluctuation feature, a current phase offset feature, and a device health feature, inputting the power network operation feature set into a power state identification model that completes debugging, performing joint analysis on the voltage fluctuation feature, the current phase offset feature, and the device health feature by using the power state identification model, generating a power network state classification result, where the classification result includes a normal state, an abnormal state, and a fault alarm state, generating a network operation policy according to the power network operation policy, and sending the network operation policy to a power dispatching terminal to activate a corresponding operation or adjust a load maintenance policy. In a second aspect, an embodiment of the present application provides a power network state identification system, which includes a processor and a memory, wherein the memory stores a computer program, which when executed by the processor, causes the processor to perform the steps of the above method. In a third aspect, embodiments of the present application provide a computer readable storage medium comprising a computer program for causing a power network state identification system to perform the steps of the above method when the computer program is run on the power network state identification system. In the implementation of the application, the dynamic perception and accurate decision of the state of the power network are realized by combining the collaborative anal