Research On Prediction Of Safety Stocks Of Auto Parts Based On LSTM Model

According to the statistics of the China Association of Automobile Manufacturers,China's auto production and sales volume have been ranked first in the world for nine consecutive years.The continuous increase in the demand for automobiles has also driven the development of the auto parts market.As the great uncertainty in spare parts inventory preparation,and many auto manufacturers,4S stores,agents,etc.,lack a reasonable inventory solution for auto spare parts inventory.At the same time,it is not practical to fully realize the ideal "zero inventory" production and sales.Therefore,holding safety stocks has become one of the most effective strategies for resolving emergencies.This thesis on the automotive spare parts safety inventory settings,first analyzed several common inventory control methods,including ABC classification method,quantitative ordering method,mathematical statistics method.However,these traditional inventory quantification methods rely on existing practical experience and summary,which are inefficient and lack accuracy.With the development of artificial intelligence,people began to use BP neural networks to predict safety stocks and achieved remarkable results.However,the traditional BP neural network does not consider the time sequence of safety stock data and has certain defects.Thus,this thesis uses a Recurrent Neural Network model which based on Long-Short Term Memory(LSTM)to solve the problem of the forecasting on safety stocks of automotive parts.In this thesis,the three kinds of spare parts of the safety stock data and the data set downloaded on UCI website are used as experimental data,and then the model simulation experiment is carried out on Anconda3 software.Firstly,the LSTM model of safety stock forecasting is established.The appropriate input,output,activation function,loop body layer number,and optimization function of the model are selected through experiments.Meanwhile,because the network structure of LSTM,the initial connection weights and the choice of thresholds have a great impact on the network performance,and it cannot be accurately obtained.In view this problem,this thesis uses the bat algorithm to optimize the initialization of the weights and thresholds in LSTM neural network.In addition,the BA-LSTM algorithm is compared with the LSTM,BP,BA-BP and BA-SVR algorithm.The mean square error(MSE)is used to process and analyze the error between the prediction data and the original data.The experimental results show that compared with LSTM,BP,BA-BP and BA-SVR algorithm,the BA-LSTM algorithm has faster convergence speed and lower prediction error rate,which is more suitable for solving the problem of safety inventory forecasting of auto parts.Finally,the vehicle safety inventory forecasting system is designed and implemented,and the BA-LSTM algorithm is applied to the system to realize safety inventory forecasting.