Deep learning LSTM and random forest ML-aided design tools for energy cooling capacity modelling

Abstract

Cooling capacity (Qe) is a critical metric in cooling system design across commercial, residential, and industrial sectors. It represents the system’s ability to remove heat and maintain a steady temperature, contributing to energy efficiency. This study employed two novel machine-learning approaches to model cooling system efficiency: long short-term memory (LSTM) and Random Forest (RF). At the outset correlation, fuzzy sensitivity analyses of dependent and independent variables were conducted. Based on the fuzzy sensitivity analysis results, three different modeling schemes (C1, C2, and C3) were developed, and the predictive models focused on data validation and calibration. Furthermore, the model’s performance is evaluated using metrics such as correlation coefficient (R), coefficient of determination (R²), mean square error (MSE), and root mean square error (RMSE). The results reveal that LSTM models demonstrated exceptional performance during training, achieving R values of 0.997 and R² values of 0.998, indicating a strong fit to the dataset. However, R values dropped to 0.796 during the testing phase, with corresponding R² values of 0.799. In contrast, the RF model exhibited superior generalization on the testing data, with an R-value of 0.933 and an R² value of 0.966. Although the MSE and RMSE values were slightly higher for the RF models than the LSTM models. The overall performance demonstrated the robustness of RF in predicting cooling system efficiency, with better generalization and lower error rates during testing. This indicates that the RF-M1 model achieved the best overall results due to its superior testing performance metrics.