Deep random forest prediction algorithm based on LSTM and GRU improvement and application research

Most of the existing deep learning prediction algorithms are based on deep neural networks, which are suitable for continuous process modelling, but are not ideal for discrete processes such as processing of grain and oil products. Deep random forest prediction algorithms are able to model predictions for discrete processes; however, it lacks a backpropagation process compared to deep neural network-based prediction algorithms and cannot fully extract information from the utilized data. Aiming at this problem, this paper proposes to improve the prediction algorithm of deep random forest using LSTM and GRU, and apply it to the study of prediction of pollutants in grain and oil processing. Firstly, the LSTM and deep random forest were used to predict pollutants in grain and oil processing processes respectively, and the characteristics of the two methods were compared; on this basis, the LSTM and GRU models are introduced into the deep random forest model to fully extract the data information using the back-propagation process, and the deep random forest improvement models a and b are proposed. An example validation of the prediction method for contaminants in grain and oil processing shows that the prediction accuracy of the proposed algorithm is much better. The algorithm proposed in this paper is an important improvement to the deep random forest algorithm and is of great importance for the study of prediction of discrete processes.