Research On Enterprise Power Consumption Forecasting Model Based On Machine Learning Algorithms

In order to improve the quality and guarantee level of electric power services,in recent years,power load forecasting has been widely concerned by industry and academia.With the development of science and technology,machine learning has been applied to various fields.How to use machine learning technology combined with historical electricity data and current power consumption influencing factors to predict the power load with high precision has become a research hotspot.At present,the basic data of power load forecasting mainly comes from data acquisition and monitoring control system(SCADA),but there are some problems such as abnormal data and dimensional inconsistency,and data preprocessing is a necessary prerequisite to improve the quality of basic data and improve the accuracy of prediction.In this paper,combined with machine learning technology,an enterprise power anomaly detection module based on unsupervised learning is added to the data preprocessing stage according to the characteristics of lack of anomaly detection module,large amount of data,no sample label and high dimension of enterprise power prediction,Using isolated forest algorithm(iForest)and the 3?to detect the anomaly of enterprise electricity and comparing it with the K-means algorithm.The simulation results show that when the data scale is small,the detection rate of the two anomaly detection models is not much different,but when the data scale becomes larger,the anomaly detection rate of the module is significantly higher than that of the anomaly detection model based on K-means,and when the data dimension reaches 6174 dimension,its detection accuracy is 30.5% higher than that of the anomaly detection model based on K-means,which reaches 93.5%.Aiming at the limitation of using single model in most enterprises,this paper puts forward a prediction method of enterprise electricity consumption based on the model of Autoregressive Moving Average Model(ARMA)and Extreme Gradient Boosting(Xgboost),combining with the influence of temperature,holiday and other factors on the power consumption data of enterprises,and uses this model to compare with the forecast results of enterprise power prediction model based on ARMA,enterprise power prediction model based on Xgboost and enterprise electricity prediction model based on feedback neural network.The simulation results show that the prediction accuracy of the model reaches 97.02%,and the average increase of 2.76% compared with three models.In short,combined with data preprocessing,the combined prediction model proposed in this paper has the characteristics of short prediction time,strong pertinence and high prediction accuracy,it can effectively reveal the change law of enterprise electricity consumption based on the prediction results,and is suitable for the field of power system.