Enhancing microgrid forecasting accuracy with a TCNN-TLS framework: A novel approach to mitigating uncertainty in renewable energy and load predictions

Accurate prediction of wind power, photovoltaic (PV) output and load demand in real-world scenarios is imperative to achieve optimal costing, ensure reliable operation and fortify the security of a microgrid system. This paper introduces a novel framework designed to improve the accuracy of short-term forecasting of the aforementioned energies and load profiles in an isolated microgrid. The employed settings contain a temporal convolutional neural network (TCNN) optimized with a pelican optimization algorithm (POA) to enhance its hyper-parameter selection. The proposed method incorporates advanced error modeling techniques using the T-location scale (TLS) distribution to review the prediction, making it more reliable. To evaluate the effectiveness, TCNN-TLS framework is compared with typical TCNN model with traditional methods like long short-term memory (LSTM) and artificial neural networks (ANN). The results demonstrate significant improvements after error adjustment, particularly in wind power and load forecasting, with notable reductions of 16.2% and 6.0% in RMSE and 17.4% and 5.7% in MAE, respectively. However, the improvement in PV output forecasting was limited, with slightly increasing RMSE from 232.91 to 285.08 after TLS adjustment.