Advanced Materials & Sustainable Manufacturing 2024 , 1, 10003
8 of 14
Figure 4. Correlation analysis results of T oa and H oa . The prediction results of the prediction model for the first twenty data sets under the four operating conditions are shown in Figure 5. It is noted that the predicted values of random forest and gradient boosting regression are closer to the actual values, and the prediction of linear regression is less effective. In principle, linear regression is suitable for situations where there are few variables and the relationships between the variables are not particularly complex. However, papermaking is a complex process involving mass and heat transfer, and the relationship between variables is complex. In most cases, the relationship between parameters and parameters is non-linear, so linear regression is less effective in making regression predictions for non-linear data. Both Random Forest and Gradient Boosting Decision Tree are integrated learning algorithms, they both consist of multiple decision trees, and the final result needs to be decided by all decision trees together. However, Random Forest and Gradient Boosting Decision Tree differ in their ideas. Random Forest adopts the idea of Bagging in machine learning, in which, Bagging draws samples from the training set to train weak classifiers by uniform sampling with put-back, and the training sets of each classifier are independent of each other. The training sets of decision trees are independent of each other, and the trees of Random Forest can be generated in parallel with each other. The gradient boosting decision tree uses the Boosting idea, where the training sets of each classifier are not independent of each other. The composed trees need to be generated serially in order, and the training sets of each weak classifier are sampled from the results of the previous weak classifier.
Made with FlippingBook interactive PDF creator