Short-term auto parts demand forecasting based on EEMD—CNN—BiLSTM—Attention—combination model

Abstract

Demand forecasting of auto parts is an essential part of inventory control in the automotive supply chain. Due to non-stationarity, strong randomness, local mutation, and non-linearity in short-term auto parts demand data, and it is difficult to predict accurately. In this regard, this paper proposes a combination prediction model based on EEMD-CNN-BiLSTM-attention. First, the model uses the ensemble empirical mode decomposition method to decompose the original data into a series of eigenmode functions and a residual item to extract more feature information. And then uses the CNN-BiLSTM-attention model to analyze each mode separately. The components are predicted, and the prediction results are summed to obtain the final prediction result. The attention mechanism is introduced to automatically assign corresponding weights to the BiLSTM hidden layer states to distinguish the importance of different time load sequences, which can effectively reduce the loss of historical information and highlight the input of critical historical time points. Finally, the final auto parts demand prediction results are output through the fully connected layer. Then, we conduct an experimental analysis of the collected short-term demand data for auto parts. Finally, the experimental results show that the prediction model proposed in this paper has more minor errors, higher prediction accuracy, and the model prediction performance is better than the other nine comparison models, thus verifying the EEMD-CNN-BiLSTM-attention model for short-term parts demand forecasting effectiveness.