A short-term load forecasting method for integrated regional energy systems based on multivariate phase space reconstruction

Abstract

To more accurately describe the load evolution law and its relationship with other variables, obtain the phase-space representation of the load data, and achieve accurate prediction of the short-term load of the DIES, a short-term load prediction method for the DIES based on multivariate phase-space reconstruction is proposed. Through correlation analysis of the short-term loads of DIES, the relationship between electric, cooling, and heating loads and meteorological characteristics is evaluated using the Pearson correlation coefficient to determine the input multivariate variables of the prediction model. The multivariate phase-space reconstruction technique is employed, and the delay time and embedding dimensions of the time series of the multivariate variables are determined by using the CC algorithm to optimize the phase-space reconstruction process, obtaining the phase-space representation of the electric, cooling, and heating loads and meteorological characteristics. The coupling characteristics of the electricity, cooling and heating loads and the meteorological characteristics over time are explored. Based on the Kalman filtering algorithm, a short-term load forecasting model for DIES is established, and the phase points reconstructed in phase space are used as the state vectors, which constitute the state-space description of the phase points. Kalman filtering theory is applied to realize the accurate forecasting of future short-term loads. The experimental results demonstrate that the method can clarify the correlation between electricity, cooling, and heating loads, and meteorological data through Pearson correlation analysis, and accordingly select an 8-dimensional multivariate time series for short-term load prediction. This method can accurately predict short-term changes in electricity, cooling, and heating loads with high prediction accuracy and stability. The root mean square error is 0.52 MW, the mean absolute error is 0.38 MW, the mean absolute percentage error is 2.1 %, and the p-values are all <0.01.