Advanced Materials & Sustainable Manufacturing 2024 , 1, 10003
9 of 14
Figure 5. Prediction comparison of different models under four working conditions.
Figure 6 shows the evaluation of the effectiveness of the exhaust fan temperature and humidity prediction model, which is a comparison of the area of a triangle consisting of three angular evaluation indicators (R 2 , MAE and MAPE, the calculation details refer to Eqs. (10) – (12)), with MAE and MAPE as the inverse and their actual values as an increasing trend from the center to the apex of the radar plot. As shown in Table 5, it means that the larger the area of the triangle in the figure, the better the performance of the model in all dimensions. It can be seen that in most cases the triangles of Random Forest and Gradient Boosting Regression have a high overlap and the area is larger than that of Linear Regression. ܴ ሺ݁ ǡሻൌ σ ሺ݁ െ݁ ҧ ሻሺ െҧ ሻ ୀଵ ඥσ ሺ݁ െ݁ ҧ ሻ ଶ ήσ ሺ െҧ ሻ ଶ ୀଵ ୀଵ (10) ൌ ͳ݊ ȁ݁ െ ȁ ୀଵ (11) ൌ σ ԝ ୀଵ ȁሺ݁ െ ሻോ݁ ȁ݊ ή ͳͲͲΨ (12) where e i is the real targets, whereas p i is the predicted output of the model. According to the analysis above, it can be concluded that the prediction errors of both Random Forest and Gradient Boosting Regression are relatively small. From the values of the evaluation indices, the predicted values of random forest are better than those of gradient boosting regression, with better agreement with the actual value curve and a closer trend. It shows that the random forest model has better robustness and generalisation ability, and also shows more stable effect under different working conditions. So, it can be regarded that the random forest can be applied to complex drying conditions to provide reference for index prediction. It is also shown that the model is real-time and the prediction results can be used for online process monitoring.
Made with FlippingBook interactive PDF creator