Construction and Validation of a Supply Chain Demand Forecasting Model Based on Embedded AI

Abstract

Supply chain demand forecasting is an important basis for enterprise decision-making, which can not only help optimize resource allocation, but also enhance the market competitiveness of enterprises. To improve the accuracy of supply chain demand forecasting, a combined forecasting model considering univariate and multivariate variables is designed and embedded. On the univariate prediction model, the study considers optimizing the parameters of the backpropagation neural network through an improved whale optimization algorithm. On the multivariate prediction model, the study combines improved particle swarm optimization algorithm, convolutional neural network, and long short-term memory network. The findings showed that the accuracy, root mean square error, time consumption, and maximum memory usage of the univariate prediction model were 98.05%, 1.03%, 61ms, and 10.85%, respectively, which were significantly better than the comparison model. The maximum accuracy of the multivariate prediction model was 98.51%, the minimum was 96.02%, and the maximum root mean square error was 0.58. After embedded deployment, the maximum increase in time consumption of the combined prediction model was 47.76%, and the accuracy only decreased by 0.18%. The designed combination forecasting model has good performance and can provide model support for predicting supply chain demand.