Abstract:

Abstract Forecasting SMP is critical in power systems, allowing market participants and grid operators to make more informed decisions. SMP prediction faces nonlinearities, volatility, and intricate factor interactions. Meanwhile, several existing methodologies exhibit inaccuracies in their predictions. Furthermore, when used across multiple circumstances, single forecasting algorithms have lower accuracy.  This paper presents a novel forecasting model that combines Least Squares Support Vector Machines (LSSVM) and the Genetic Algorithm (GA) for (i) parameter optimization, and (ii) parameter optimization and input selection, for accurate SMP prediction. Furthermore, the performance of LSSVM-GA was observed through daily and weekly forecasts. GA optimizes the LSSVM parameters and forecast inputs concurrently to ensure the best possible performance. Historical data from the Single Buyer (SB) has been employed to train and evaluate this model. Correlation Analysis aids feature selection, boosting model generalization. Multiple forecast input combinations were examined to identify the most important forecasting features. The proposed daily forecast model exhibited a 3.54% performance improvement compared to the SB daily forecast model. Likewise, the proposed weekly forecast model outperformed the Single Buyer (SB) forecast by 1.19%. As per the results, the hybrid algorithm shows great potential as a viable option for generating precise forecasts of electricity prices.

Keywords: