Research And Application Of Inventory Forecasting Model Based On The Improved BP Neural Network

Manufacturing enterprises usually enlarge the market share to deal with the increasing market competition. The enterprise inventory cost will increase as the increasingly inventory demand of raw materials, components, semi-finished and finished products. In order to improve customer satisfaction and reduce the inventory cost, inventory control has become a key factor to enhance the competitiveness of enterprises. Therefore, demand forecasting of inventory has been paid more and more attention.There are lots of traditional demand forecasting methods of inventory, while in the influence of "multiple factors" such as the company scale expands, the individual needs and multiple market demands and customized demand, it is difficult for the traditional methods to meet the forecasting demand as the high prediction error and low prediction accuracy. Therefore, it is important for enterprise to reasonably control inventory cost and find a better demand forecasting method to solve the inventory problems those trouble lot of enterprises.Based on a large number literatures about inventory forecast, with the appearance and development of the BP neural network, its unique study, summary and nonlinear characteristics has gradually been excavated and fully applied in the field of the forecast. At present, the classification method combined with the BP neural network is research focus on the study of BP neural network. This paper takes the inventory demand of intelligent logistics delivery tank of M company as the research background, two aspects of improvement and optimization to inventory forecast is put forward based on the related academic study.Firstly, the inventory requirements influence factors of logistics delivery tank in M company is analyzed from multiple angles on the basis of previous studies. In order to improve the customer satisfaction, the input variables of BP neural network are filtered through the principal component analysis(PCA) to customer requirements. That is, the input variables of BP neural network are filtered based on the comprehensive factor analysis method that combined the House of Quality(HOQ) with PCA.Secondly, about the number of node in hidden layers selection problems of BP neural network, lots of experts and scholars determine it according to the empirical formula, which is too empirical. To solve thus problem, this paper come up with Binary Segmentation to optimize it, the best number of number of node in hidden layers could be determined more quickly and accurately. According to experience formula, the general area of node number in hidden layers is obtained. The Binary Segmentation method is used to narrow the interval range, and the corresponding mean and square error of output value by BP neural network for different node number in hidden layers is compared. The node number which has the minimum mean and square error is the best number of node in hidden layers.The purpose of this paper is to build an inventory demand forecasting method for company M, this is inventory demand forecasting model of BP neural network. the final forecast output is obtained through the experimental training and simulating prediction of the forecasting models of BP neural network both before and after improved by MATLAB 10.0. The study shows that the output error based on inventory demand forecasting model of improved BP neural network is more smaller, and the prediction accuracy is higher.