Hybrid Dynamic Bacterial Foraging Algorithm with a Long Short-Term Memory and Adaptive Neuro-Fuzzy System for Short-Term Load and Spinning Reserve Forecasting

Abstract

Accurate forecasting of short-term load and spinning reserve is essential for ensuring the secure operation of power systems, facilitating effective electricity generation and demand-side management. This paper introduces an innovative hybrid forecasting approach, integrating Long Short-Term Memory (LSTM) networks and Adaptive Neuro-Fuzzy system (ANF) models, optimized by a Dynamic Bacterial Foraging algorithm (DBFO). The LSTM model is best suited for detecting time-series patterns, but the ANF system contains fuzzy logic and ANN to be able to handle uncertainty and nonlinearity of data. The DBFO algorithm adjusts the hyperparameters of the two models by dynamically adjusting essential parameters according to changes in the environment. Extensive testing on actual power system data confirms that the proposed hybrid models perform better than conventional approaches, providing robust and reliable predictions for load and spinning reserve. Comparative studies with traditional machine learning tools and existing optimization algorithms also reinforce the superiority of the proposed methodology.