Evaluation of Harris Hawks Optimization Algorithm in ANN Training Applied to Electricity Consumption Prediction in Malaysia

Electricity consumption has been increasing over the past century and is expected to continue increasing soon. Predicting electricity consumption has thus been a topic of interest. Artificial Neural Networks (ANNs) are reliable methods to be used as prediction models. However, as standalone models, they often have low accuracy. Hence optimizers including swarm intelligence algorithms are used for improving the models. In this paper, a recently introduced swarm intelligence algorithm, namely Harris Hawks optimization (HHO) has been employed for optimizing an ANN for the prediction of electricity consumption. Specifically, HHO is used as a training algorithm to get the best connection weights and biases of the ANN. The experimental results prove the effectiveness of HHO in optimizing the prediction model. In addition to HHO, two more recent and high-performing swarm intelligence algorithms, namely Dragonfly Algorithm (DA) and Whale optimization Algorithm (WOA) are also employed for optimizing the ANN. The performance of the prediction models optimized by HHO, DA, and WOA are evaluated by comparison of the root mean square error of the ANNs during the training and testing, the percentage decrease in RMSE from the initial to the final iterations during the training and the time taken for the algorithm to converge.