Prediction of household power consumption using grasshopper optimization algorithm and artificial neural networks

Hybrid intelligent systems are technologies that are created by combining different artificial intelligence methods or techniques. In these systems, artificial neural networks (ANN), fuzzy logic and evolutionary algorithms are generally used together, and it is possible to obtain more successful results with these systems in experimental analysis. Especially by integrating swarm-based optimization algorithms with ANN, many problems such as classification, estimation, and curve estimation can be easily solved. Grasshopper optimization algorithm (GOA) is one of the dominant meta-heuristic optimization algorithms that have been widely used in recent years, which is created by combining with ANN. In this study, a hybrid intelligent system (GOA-ANN) based on GOA and ANN is proposed. In the study, a multi-layer feed-forward ANN architecture is trained with GOA and the optimal weights of the network are determined. With the developed GOA-ANN hybrid intelligent system, firstly the non-linear XOR problem is solved, and then the electrical power consumption for the household is estimated on a publicly-available dataset. In the experimental studies conducted for the prediction of household power consumption, RMSE, MAPE and R2 error measurement results calculated with the GOA-ANN hybrid intelligent system proposed in the thesis study, RMSE, MAPE and R2 error measurement results of Particle Swarm Optimization based ANN (PSO-ANN) and Backpropagation Algorithm based ANN (BPA-ANN) hybrid intelligent systems are compared. In the GOA-ANN system, the average RMSE, MAPE and R2 error criteria in the household power consumption estimation for June 2007 are calculated as 2.7712, 0.9447, 0.8345 respectively. In addition, in the house power consumption estimation using the GOA-ANN hybrid intelligent system, the average RMSE, MAPE and R2 criteria for May 2008 are 2.0856, 1.1418, 0.9248 respectively. The average RMSE, MAPE and R2 criteria for June 2009 are 3.705, 1.5597, 0.9249 respectively. Moreover, the average RMSE, MAPE and R2 criteria for January 2010 are calculated as 3.1996, 1.6493 and 0.8782 respectively. The findings in the experimental analyzes carried out in the study show that the performance of the GOA-ANN hybrid intelligent system proposed in the thesis study is higher than the PSO-ANN and BPA-ANN methods for the RMSE and MAPE measurement criteria.