An accurate prediction method for users' electricity consumption demand based on spatio-temporal data mining

In this paper, a precise prediction method for users' electricity consumption demand based on spatio-temporal data mining is proposed. Outlier detection is carried out through the box plot statistical verification method to effectively clean up data noise and provide a high-quality data basis for subsequent modeling. Based on the cleaned time series and non-time series data, a multi-dimensional evaluation index of users' electricity consumption characteristics is constructed, and a multi-dimensional factor electricity consumption similarity evaluation index is further established. The precise identification and classification of power users' electricity consumption patterns are achieved through factor analysis method. On this basis, combined with the ARIMA time series prediction model and the neural network spatial prediction model, a spatio-temporal fusion prediction framework is constructed to fully explore the spatio-temporal correlation characteristics of electricity consumption data. Ultimately, the prediction results of trend components, seasonal components and random components are integrated by using exponential multiplication to obtain high-precision predicted values of users' electricity consumption demand, thereby achieving accurate prediction of users' electricity consumption demand. The experimental results show that the proposed method is consistent with the actual electricity consumption. The predicted results are highly close to the actual values, with an extremely small error of 0.08%.