A short-term load demand forecasting: Levenberg–Marquardt (LM), Bayesian regularization (BR), and scaled conjugate gradient (SCG) optimization algorithm analysis

Electrical load forecasting is of the utmost significance in the power business since it may serve as a reference for downstream operations such as power grid dispatch, resulting in substantial financial advantages. Presently, the urban energy sector incorporates the functioning of multiple load demand clusters, which contributes to the reduction of utility grid strain. However, the integrated functioning of numerous users results in dynamic load needs, necessitating accurate electric load forecasting for monitoring operations. Load forecasting is a multifaceted task that necessitates the use of approaches beyond statistical methods. The concept of cluster load demand forecasting is a novel application that is seldom addressed in the literature. This article examines different machine learning techniques, such as linear regression, support vector machines, Gaussian process regression, and artificial neural networks (ANN), to determine the most efficient approach for predicting short-term load demand in cluster loads. The effectiveness of these solutions is evaluated by quantifying several aspects, such as error metrics and computing time. This discovery demonstrates that the artificial neural network (ANN) produces highly accurate forecasting outcomes. Furthermore, three separate optimization strategies are utilized to choose the most effective ANN training algorithm: Bayesian regularization, Levenberg–Marquardt, and scaled conjugate gradient. We assess the efficacy of optimization methods by running training, testing, validation, and error analysis. The results indicate that the ANN models based on the BR and LM optimization algorithms yield the most accurate electrical load demand forecasting results.