Sales Demand Prediction Model of Gated Recurrent Unit Neural Network Based on Improved Shape Distance Loss Function

Abstract

Under the background of diversification and refinement of chemical products, product demand prediction is playing a guiding role in production planning. In this paper, a new sales demand prediction model based on improved shape distance Loss function of Gated Recurrent Unit Neural Network (ISD_GRUNN) is proposed for the long-term prediction of the sales quantity of chemical products. The improved shape distance is determined by the change trend, amplitude and distance between the two points. Compared with MSE which only considers the difference between the corresponding time point sequence values as the loss function, the change trend and range of the time series will be taken into account in the improved shape distance as the loss function. The experimental results show that the improved shape distance as the loss function can be better used for sales long-term prediction.