Short term electricity load forecasting using hybrid prophet-LSTM model optimized by BPNN

Abstract

Electrical load forecasting plays a vital role in the operation and planning of power plants for the utility companies and policy makers to design stable and reliable energy infrastructure. Load forecasting is categorized in long-term, mid-term and short-term. Among them, short term load forecasting that monitors weekly, daily, hourly and even sub-hourly operations is gaining a lot of attention which saves time and cost while satisfying consumers’ needs without interruption. Different models such as conventional, Artificial Intelligence (AI) and hybrid models have been developed to investigate short-term load forecasting. However, these models suffers various issues such as low speed convergence (conventional), high complexity (AI) and so on. Consequently, this work proposes a hybrid method using Prophet and Long Short Term Memory (LSTM) models to overcome the above limitations in an effort to predict accurate load. The Prophet model utilize linear as well as non-linear data to predict original load data but still some of the residuals are left which are regarded as non-linear data. Here, these residuals (non-linear data) are trained by employing LSTM, and finally both the forecasted data from Prophet and LSTM are trained by Back Propagation Neural Network (BPNN) to further enhance prediction accuracy. Elia Grid real time quarter hour based electrical load data from 2014 to 2021 has been utilized to verify working performance of proposed hybrid technique by computing Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and Mean Average Error (MAE). Results substantiate that proposed hybrid models outperforms the standalone models of Autoregressive Integrated Moving average (ARIMA), LSTM and Prophet model on the basis of reduced errors with least computation time.