Research on Data-Driven Demand Forecasting and Service Optimisation Model for Electricity Users’ Behaviours

Abstract

This paper proposes a user electricity data mining method based on deep learning and improved locust optimization algorithm, and at the same time adopts the Pearson correlation coefficient method to reduce its dimension to improve the data mining effect of linear weighted KFCM algorithm. In order to deal with the electricity demand of massive electricity customers, the user electricity demand forecasting model is constructed based on the Extreme Learning Machine ELM algorithm by combining the relationship between short-term loads and factors of electricity customers. Construct the service optimization model with the maximization of benefit index as the objective function, and use the BAS algorithm to solve the optimal solution in order to achieve the effect of user service optimization. Determine the experimental platform and model parameters, and carry out an example analysis of demand forecasting and service optimization for electricity users.C class users have a small electricity load except for breakfast and dinner, and the maximum time period of the electricity load is from 18:00 to 20:00 hours. Combined with MAPE, the ELM model improves 4.57% than SVR, 21.9% than LSTM, and 34.37% than ARIMA, which indicates that the ELM model is more effective and higher in demand forecasting for electricity users. In addition, the optimal solution of the effect of the BAS algorithm is 69 yuan, 102 yuan and 49 yuan higher than that of the GA algorithm in terms of dividend transmission benefit, energy saving and emission reduction benefit, and electricity right trading benefit, respectively, and the optimal solution based on the BAS algorithm is closer to the actual benefit value, which fully proves the effectiveness of the service optimization model based on the BAS algorithm.

Keywords: electricity user behavior; ELM; BAS; electricity demand forecasting; service optimization models