Study on Diagnosis of Daily Electricity Consumption in Hotel Facilities Based on Amalgamation of Gaussian Process Regression and Teaching Learning Based Optimizer

Abstract

The increase in electricity consumption has negative effects on the environment, including greenhouse gas emissions, global warming, and rapid climate change. In an attempt to address this global problem, this research aims at developing machine learning models for the prediction of electricity consumption in hotels. The developed model steps on coupling Gaussian process regression and teaching learning based optimizer (GPR-TLBO) to forecast daily electricity consumption in hotel facilities. In this context, teaching learning based optimizer is implemented to circumvent the shortcomings of manual tuning-based intelligent models through improving the search abilities of Gaussian process regression by adaptively tuning its structure. This involves optimizing the type of kernel function and its design parameters. The developed model is assessed through its comparison against eight widely used machine learning models. These models involve: regression trees (RT), support vector machines (SVM), Elman neural network (ERNN), generalized regression neural network (GRNN), back propagation artificial neural network (BPANN), cascade forward neural network (CFNN), adaptive neuron fuzzy inference system tuned by particle swarm optimizer (ANFIS-PSO) and adaptive neuron fuzzy inference system tuned by genetic algorithm (ANFIS-GA). The comparison is carried with regards to six performance assessment metrics, namely mean absolute error (MAE), mean absolute percentage error (MAPE), root-mean squared error (RMSE), relative absolute error (RAE), root relative squared error (RRSE) and geometric reliability index (GRI). Average ranking algorithm is exploited to create a holistic evaluation of the nine machine learning models based on their reported accuracies. The levels of dependencies between the machine learning models are studied based on Spearman’s rank correlation coefficient. Results illustrated that the developed GPR-TLBO model significantly outperformed the reminder of the machine learning models yielding MAE, MAPE, RMSE, RAE, RRSE and GRI of 223.98, 8.04%, 263.37, 0.57, 0.54 and 1.1, respectively. In addition, it was interpreted that the developed GPR-TLBO model provided a better overall performance gaining average and standard deviation of rankings that are equal to one and zero, respectively. The Spearman’s correlation analysis demonstrated that higher levels of dependencies are exhibited between room degree day and daily electricity consumption.