The CTCN-LightGBM Joint Model for Industrial Balanced Loading Prediction

Abstract

Balanced industrial loading mainly relies on accurate multi-adjustment values, including the truck speed and chute flow. However, the existing models are weak in real-time loading prediction because the single-objective regression may ignore the correlation of multi-adjustment parameters. To solve the problem, we propose a joint model that fuses the composited-residual-block temporal convolutional network and the light gradient boosting machine (i.e., called CTCN-LightGBM). First, the instance selection deviations and abnormal supplement methods are used for data preprocessing and normalization. Second, we propose a side-road dimensionality reduction convolutional branch in the composited-residual-block temporal convolutional network to extract collaborative features effectively. Third, the feature re-enlargement method reconstructs extracted features with the original features to improve extraction accuracy. Fourth, the reconstructed feature matrix is utilized as the input of the light gradient boosting machine to predict multi-adjustment values parallelly. Finally, we compare the CTCN-LightGBM with other related models, and the experimental results show that our model can obtain superior effects for multi-adjustment value prediction.