Table 5 Comparative study of various Deep Learning and Machine Learning models for Predicting Dynamic behaviors and Thermal Conductivity, including our proposed model.
From: Deep regression analysis for enhanced thermal control in photovoltaic energy systems
Deep learning model | Machine learning model | System analyzed | Performance | Reference |
|---|---|---|---|---|
ANN | – | Latent heat thermal energy storage system | ANN outperforms numerical models in predicting heat stored in the finned tube via phase change material, with an average absolute mean relative error of 5.58. | |
ANN | – | Centralized PCM storage system | A trained ANN accurately predicts storage system exhaust atmospheric temperature, showing a high correlation with numerical outcomes. | |
ANN with Fuzzy Inference System (ANFIS) | SVM | Thermal energy storage performance of a solar collector with PCM | The SVM model demonstrates superior performance over ANN and ANFIS models based on the current dataset. | |
ANN | Random forest | Prediction of building energy consumption | ANN slightly outperforms Random Forest, with an RMSE of 4.97 compared to RF’s 6.10. | |
ANN | – | Thermal energy storage system with PCM | ANN models accurately predict heat absorption and emission during charging and discharging, demonstrating confidence levels of 95% and low uncertainty at 5%. | |
ANN | – | Dynamic behavior of building envelope with PCMs | An artificial neural network model successfully forecasts heat flux, with an average model error of 0.34 W/m2. | |
- | LR, Decision Tree, kNN | Prediction of building energy consumption | Linear Regression (LR) and Support Vector Regression (SVR) models yield the best results among all models tested. | |
ANN | Gaussian, SVR, Linear Regression | Space heating and cooling load prediction for residential building | Among the models tested, the Gaussian radial basis function kernel SVR model exhibited the best performance, achieving a 4% adjusted mean absolute error and root mean-square error. | |
- | RFR, XGBR | Incorporating PCMs into cementitious composites | The gradient boosting model yields the highest R-SQUARE of 0.977, RMSE of 2.419, and MAE of 1.752 in predicting compressive strength. | |
CNN FNN | – | Cooling effectiveness of PV solar panels using thermal imaging videos | The CNN model outperformed the FNN model, with an RMSE of 2.3%, MAE of 1.2%, and R-square of 0.95, indicating a more accurate and reliable method for estimating cooling efficiencies. | [Our proposed model] |
