Comparative Study of Optimization Technique-Based Deep Learning Approach for Short Term Load Forecast

An efficient operation based on the precise demand forecast is necessary for an industry to function economically with considerable savings and emission reductions. The deep learning (DL) techniques produce better predictions where their performance leans on their parameter values, set either manually or automatically. This paper briefly presents details of four metaheuristic algorithms used to obtain the parameter values. It also analyzes their ability to determine the best collection of parameter values for the bilateral long short-term memory (BiLSTM)-based DL technique. These are (i) particle swarm optimization (PSO), (ii) JAYA optimization, (iii) sine cosine optimization (SCA), and (iv) gray wolf optimization. The work was assessed using pharmaceutical industrial demand based on EnergyPlus software for one-week prediction and analyzed using statistical measures MAE, MAPE, and RMSE. The results show that the gray wolf optimizer optimized BiLSTM technique produced better prediction by 59.17, 44.29, 33.27, and 51.32% in MAPE over BiLSTM, PSO-BiLSTM, JAYA-BiLSTM, and SCA-BiLSTM, respectively.