Application Research of Computer Machine Learning Algorithm in Anti-theft Analysis under Smart Grid

Abstract

Aiming at the low accuracy of traditional anti-stealing prediction methods, this paper proposes an anti-stealing prediction method based on power big data. The method constructs electricity stealing data samples according to abnormal rules, and introduces the constraint condition of line loss rate growth rate. This paper classifies the user's load data, proposes four indicators to measure the load curve, and obtains the characteristic variables. On the basis of classification, the extracted feature variables are dimensionally reduced, and local outlier factors are used to screen out users with abnormal electricity consumption. The method uses the maximum mutual information coefficient to measure the correlation between the management line loss and the specific behavior of users, and uses CFSFDP to locate abnormal electricity users, which has strong applicability and can detect many different types of electricity stealing behaviors. Finally, the algorithm is verified using the smart meter dataset, and the results demonstrate the good performance of the method.