Abnormal recognition of massive electric power data based on RF and CNN

Abstract

In electric power field, electric power data collection equipment collects electric power data every 15 minutes, 96-digit data will be generated for the same account every day. The data generated include highly available real-time electric power data, as well as abnormal data caused by problems such as equipment failures, changes in the enterprise industry, electric power theft and leakage, etc. When using the traditionally manual method to process the exception of electric power data, it cannot meet the needs of current massive data, high throughput, and low latency. In order to better distinguish abnormal data, indicate the cause of the abnormality, and provide support for subsequent abnormal data management on this basis, the paper has carried out the applied research of the combined anomaly analysis model based on random forest and convolutional neural network. Firstly, it analyses the characteristics of massive electric power data and the categories of abnormal causes in detail, and uses the LOF algorithm to perform abnormal data filtering on the massive electric power data. Then, it performs random forest and convolutional neural network model training on the massive electric power data, and uses an automated method to identify the cause of the abnormality for the massive electric power data. Finally, experiments were performed to verify the efficiency, low latency, and feasibility of using the combined RF and CNN model to process the classification of abnormal reasons. From the results, the combined model can well meet the needs of the classification and recognition of abnormal reason on massive electric power data, and it is an effective solution to solve the current problem of identifying abnormal reasons of electric power data.