A Quasi-moment-method Modelling of Energy Demand Forecasting

Abstract

This paper presents a novel approach to the modelling of electrical energy demand forecasting, based on the Quasi-Moment-Method (QMM). The technique, using historical energy consumption/demand data, essentially calibrates nominated ‘base’ models (in this case, nominal Harvey and Autoregressive models) to provide significantly better performing models. In addition to the novelty of the use of QMM, the paper identifies hitherto unreported singularities of the generic Harvey / logistic model, through which a simple, but remarkably pivotal modification is proposed, prior to the model’s use as base model in QMM calibration schemes. The treatment of the ‘Harvey singularities’ informed a similar and equally significant modification of the Autoregressive model utilized in the paper. For the purposes of validation and performance evaluation, computational results due to the QMM models are compared with corresponding results reported in three different journal publications, which utilized the Harvey and Autoregressive models in conventional regression schemes. And in terms of the usual model performance metrics (including Mean Absolute Percentage Error (MAPE) and Root Mean Square Percentage Error (RMSPE)), the results very clearly demonstrate the superiority of the QMM models for both energy demand prediction and forecasting. As representative examples, a QMM-calibrated Harvey model recorded an RMSE value of 495.45dB for total energy consumption prediction, as against 618.60dB obtained for the corresponding nominal Harvey model: and for the Autoregressive case, RMSE was obtained as 131.35dB for QMM model’s prediction of peak load demand, compared with the 173.40dB due to the nominal model.