Parallel LSTM and CNN for Demand Forecasting and Production Planning: A Classification Framework

Abstract

Abstract: Upon receiving order release updates from clients, the supplier of automotive components in this proposed alters its production plan for the next periods every week. Erroneous order releases can lead to significant costs, such as premium expedited shipping, production overtime, and excess inventory. For the purpose of studying order release variation, this setting is appropriate because the supply chain has adopted a JIT strategy with zero ideal inventory levels. For this reason, precise order releases are crucial for managing production volumes. Preprocessing, model training, and feature selection are the three primary components.  Imputation of missing values and data transformation are components of the data preparation phase. To identify the best features from historical data, a powerful GOA algorithm is used in feature selection.  For model training, we employed the Parallel LSTM-CNN framework to do this.  On the other hand, it makes LSTM and CNN obsolete. The data indicates that the success rate is 96.52%.