Musa A. Hameed¹ , Esam T. Yassen¹ , Wesam M. Jasim¹

1 Department of Computer Science, College of Computer Science and Information Technology University of Anbar, Anbar, IRAQ

The purpose of energy management systems in buildings is mainly to monitor energy consumption and adjust the device's operation to minimize energy bills. The objective of this work is to achieve cooperation between metaheuristics to normalization of data through two algorithms, these algorithms (Cat Swarm Optimization (CSO) and Gray Wolf Optimization (GWO) to find near optimal solution (prediction) utilize five ML algorithms. These algorithms, (Linear Regression (LR), Gradient Boosting Regression (GBoostR), Decision Tree Regression (DTR), Stochastic Gradient Descent Regression (SGDR), and Bayesian Ridge Regression (BRR)) and utilize deep learning (LSTM) to predict energy consumption and generation. After applying HL (metaheurstics and ML) to the prediction of energy consumption and evaluating the model utilizing four methods; these methods, the mean absolute error (MAE), mean squared error (MSE), root mean absolute error (RMAE), and root mean squared percentage error (RMSPE) in order to determine how well the models perform. The result of this work of heuristic learning (five methods), noticed that in the first method (HL1) utilized CSO with ML, the algorithm SGDR that uses MAE 0.0953296 to evaluate the model with CSO is better than four algorithms that are utilized in prediction of energy consumption and generation. In the second method (HL2) utilized GWO with ML, noticed that in the second method (HL2), the algorithm LR that uses MAE 0.0961145 to evaluate model with GWO is better than four algorithms that utilize in prediction of energy consumption and generation. The third method (HL3) utilized GWO with ML (Train 70%, test 30%), noticed LR that use MAE 0.1484461 to evaluate model with GWO is better than four algorithms. The four method (HL4) utilized (Train 90%, test 10%), noticed LR that utilized GWO that uses MAE 1.85332E-10 to evaluate model with GWO is better than four algorithms. Finally, methods (HL5), utilized CSO and GWO with LSTM, noticed GWO that use MAE 0.0021936 to evaluate model with GWO is better than CSO algorithms that utilize in normalization data that use to prediction of energy consumption and generation.

Keywords: Machine Learning (ML), Heuristic Learning (HL), Cat Swarm Optimization (CSO), Gray Wolf Optimization (GWO) and Long Short Term Memory (LSTM).