A three-stage multi-view feature fusion method for electricity stealing classification and prediction

Abstract

The invention discloses a three-stage multi-perspective feature fusion method for classification and prediction of electricity consumption behavior. First, the customer electricity consumption data to be analyzed is used as a test set, and the daily electricity consumption, the current day's electricity meter reading, and the previous day's electricity meter The missing data in the readings are filled with "‑1" and "0" respectively to form two preprocessed data; secondly, for each preprocessed data, features are extracted from different perspectives, and the features extracted from all perspectives are merged, Use multiple machine learning algorithms for classification prediction to obtain the electricity stealing probability of customers in the training set and test set; finally, use the linear model and the tree model to predict the output of the second stage, and then calculate the average , to obtain the electricity stealing probability to be predicted finally. Based on the integrated learning method of the existing pile model, the present invention increases the diversity of data, the diversity of models and over-fitting processing, so as to realize more accurate prediction of the electricity stealing probability of customers.

Description

technical field [0001] The invention relates to a machine learning method for classifying and predicting customers' electricity consumption behavior, in particular to a three-stage multi-view feature fusion method for classifying and predicting electricity theft. Background technique [0002] The development of social economy has made the social electricity consumption increase year by year. Driven by interests, the phenomenon of abnormal electricity consumption by customers, that is, electricity theft, is becoming more and more serious. Electricity theft by customers not only caused significant economic losses to power supply companies, but also seriously affected the normal order of power supply and consumption. According to the statistics of the State Grid Corporation of China, in recent years, the loss caused by theft of electricity by customers has reached tens of millions of yuan. In recent years, the method of customer electricity theft has also developed from brutal...

AI Technical Summary

Problems solved by technology

[0003] When analyzing the electricity consumption behavior of a large number of customers, due to the huge number of customers and the serious lack of historical electricity consumption data, the existing machine learning methods are faced with many problems such as missing value processing, feature extraction, feature selection, and model fusion. This challenge not only requires high computing resources, but also requires a lot of time to combine and select features of hundreds of dimensions and thousands of dimensions.

At the same time, it is difficult for a single classification algorithm to obtain a better prediction result of the probability of customer electricity theft. Therefore, research on methods that can better adapt to data loss, reduce the feature selection process, and improve prediction accuracy has strong social needs and great economic benefits. value

Images