The analysis of commodity demand predication in supply chain network based on particle swarm optimization algorithm

Abstract

The supply chain network model is constructed in this study based on comparison of traditional supply chain and the modern supply chain so as to solve the poor communication effect, uncirculated information, and unbalanced supply and demand in enterprises. After three algorithms and three commodity predication models are compared, a model combining with the network neural commodity demand predication method and the particle swarm optimization (PSO) algorithm is used to comprehensively evaluate the predication effect and algorithm performance by using the supply chain data of the enterprises, coming up with an optimal model. Results of the study show that: on national warehouses and regional warehouses, the difference between the predicted value and the actual value of autoregressive integrated (AR) mixture density networks (MDN) (AR-MDN) is 15%, the average outlier is between 450 and 150, the score of root mean square error (RMSE) and mean absolute percentage error (MAPE) is 117.342 and 2.334, respectively. It indicates that the fitting trend, prediction accuracy, and stability of the model are better than those of the autoregressive integrated moving average model (ARIMA) and multilayer perceptron-long short term memory (MLP-LSTM) model. Regarding determination of the stochastic requirements, the average optimal solution of the improved PSO (IPSO) is 0.45, indicating that performance of the algorithm is significantly stronger than that of the PSO algorithm and the artificial bee colony (ABC) algorithm; the comprehensive evaluation score of the combination model for the IPSO algorithm and the AR-MDN commodity prediction model is 67.41 with the optimal effect. The supply chain network model constructed in this study can provide enterprises with a good commodity demand predication method and improve their ability to respond to risks in the supply chain.