Hyperparameter Optimization of Deep Learning Model for Short-Term Electricity Demand Forecasting

• DOI: 10.46254/sa03.20220421
• Publication Date: 2023-04-20T19:14:35Z
• Authors: Félix Morales; Carlos Sauer; Hans Mersch; Diego Stalder; Miguel García-Torres
• Keywords: Hyperparameter; Hyperparameter Optimization

Short-term electricity demand forecasting represents a fundamental tool for decision-making by entities engaged in management since it allows the development of strategies to meet variations short periods.The accuracy predictive models is an important factor energy operations and scheduling generation sources at each instant.Intelligent based on Recurrent Neural Networks (RNN) require hyperparameter adjustment.These have several hyperparameters that substantially affect their performance.Our paper implements Long-Short Term Memory (LSTM) model four search methods adjust its hyperparameters.First, we select length historical window hidden state size LSTM cells optimization.Second, draw comparisons between grid search, random Bayesian scheme, genetic algorithm.The data set used training validation includes hourly consumption meteorological variables recorded Paraguay from 2015 2021.The proposed was evaluated through numerical experiments with classical error measures such as root mean square (RMSE), correlation, runtime, absolute percentage (MAPE).Our comparative study shows algorithm give optimal high test dataset.However, note may much more evaluations computational resources.