Developing a predictive method based on optimized M5Rules–GA predicting heating load of an energy-efficient building system

Abstract

The main objective of this study is to examine the feasibility of several novel machine learning models and compare their network performance with the hybrid evolutionary based algorithm. In this regard, the best fit from the above machine learning-based solutions (i.e., known as M5Rules) were combined with the genetic algorithm (GA). These techniques were used to estimate the amount of heating load (HL) mitigation from an EEB (energy efficiency buildings) system. Then, the mentioned methods are utilized to identify a relationship between the input and output parameters of the EEB system. The amount of HL was taken as the essential output of the EEB system, while the input parameters were relative compactness, surface area, wall area, roof area, overall height, orientation, glazing area, and glazing area distribution. The predicted results for datasets from each of the above-mentioned models were evaluated according to several known statistical indices such as correlation coefficient (R2), mean absolute error (MAE), root mean squared error (RMSE), relative absolute error (RAE), and root relative squared error (RRSE) as well as novel ranking systems of colour intensity rating and total ranking method. The M5Rules has been proposed as the best predictive network in this study and combined with the GA optimization algorithm. The results of the M5Rules–GA network indicated the R2, MAE, RMSE, RAE, and RRSE for the training and testing datasets were (0.9992, 0.0406, 0.0617, 6.2156, and 6.0189) and (0.9984, 0.0401, 0.0548, 6.4058, and 6.1785), respectively. Comparing to another non-hybrid proposed model with high accuracy (i.e., MLP Regressor with the R2 equal to 0.9876 and 0.9903 for the training and testing datasets, respectively), the results revealed that the M5Rules-GA network model could accomplish better performance.