10kV feeder line fault prediction method based on CNN and LightGBM

Abstract

The invention discloses a 10kV feeder line fault prediction method based on a CNN (Convolutional Neural Network) and LightGBM (LightGBM), and the method mainly comprises the steps: 1), obtaining the original data of a power distribution network, and carrying out the preprocessing of the original data of the power distribution network; 2) extracting features from the original data of the power distribution network, and constructing a feature set f {f1, f2... F15, f16, L}, wherein L is a label and represents whether the feeder line has a fault or not, the element f1, the element f2, the elementf3, the element f4, the element f5 and the element f6 are inherent attribute characteristics, the element f7, the element f8, the element f9, the element f10, the element f11, the element f12 and theelement f13 are statistical analysis characteristics, and the element f14, the element f15 and the element f16 are depth time sequence features extracted by a convolutional neural network CNN; 3) establishing a power distribution network fault prediction model; and 4) inputting the real-time data of the 10kV feeder line of the power distribution network into the power distribution network fault prediction model to obtain a feeder line fault prediction result. The 10kV feeder line fault prediction method has the advantages that the rapidity and timeliness are better, and the prediction result can provide auxiliary decision support for the operation and maintenance management personnel of the power distribution network on the premise of meeting the accuracy requirement.

Description

technical field [0001] The invention relates to the field of intelligent distribution networks, in particular to a CNN and LightGBM-based 10kV feeder fault prediction method. Background technique [0002] With the continuous improvement of industrialization, informatization and intelligence in modern society, electric energy, as an important part of secondary energy, not only plays a key role in social development, but also is irreplaceable in people's lives. The power system integrates five links of power generation, power transmission, power distribution, and power consumption. The network structure is complex and the operating environment is changeable. In recent years, the access of distributed power sources and charging piles has also brought great challenges to the power system. Any failure of any part of the power grid will affect users, ranging from economic losses to life-threatening threats. The distribution network is located at the end of the power system, close...

AI Technical Summary

Problems solved by technology

The risk assessment of the distribution network needs to establish risk indicators, which mostly rely on expert experience and fault-related static data, and the quantification rules often use AHP, DeFiel method and fuzzy evaluation method, etc., which leads to the objectivity and The real-time performance is poor; statistical analysis, correlation analysis and multidimensional analysis are generally used in the analysis of distribution network historical fault-related data. The structure of the distribution network is complex, and it is very difficult to predict the fault of the distribution network based on the mechanism modeling. The common mechanism modeling includes the temperature rise model of the power equipment and the oil and gas analysis model. Compared with the mechanism modeling, It is relatively easy to use intelligent algorithms based on data processing technology to predict faults in distribution networks, such as regression algorithms, clustering algorithms, support vector machines, and artificial neural networks. Variables and multivariate discrete time variables are modeled, lack of fusion processing of distribution network time series features and non-time series features, and lack of feature extraction method for distribution network time series variables, and with the distribution network data Accumulated year by year, some algorithms are no longer suitable for large-scale data scenarios

Images