Demand Forecast Of Inventory Based On Improved Algorithm Of BP Neural Networks

Along with the fast development of economy, research based on enterprise competition ability has been the front field of management science. Improving customers' satisfaction and lowing the inventory cost are two important means to improve enterprise competition ability. Therefore, inventory control is the key to improve enterprise competition ability. This paper applied BP neural networks technology to demand forecast of inventory. After preparation of sample data, improvement of BP algorithm and optimization of net structure, the net was trained. Then the modeling, forecasting and analyzing work with inventory was done to eliminate enterprises' high inventory and improve customer service.First, this paper pointed out the problem of traditional inventory control and then explained the related technology (Including Data Mining, Artificial Neural Networks and BP Neural Networks) and their application in inventory control, and made a general overview of the present domestic and international research on BP algorithm.Second, with the background application of CQDP hospital inventory control, this paper selected the main factors that affected the demand of the 17GY scalp-type Liuzhi aces in CQDP hospital, and adopted Data Mining technology to carry out data pretreatment with original data, including data clearing, data integration and data transformation. Then the sample data used for net training was obtained.Third, this paper introduced the principle of traditional BP neural networks. To overcome the disadvantage, an improved BP algorithm that adjusted the weight matrix W and the zoom coefficient T, the displacement parameter θ of tan-sigmoid transfer function in neutrons synchronously was presented. The information was stored in weight matrix and transfer function dispersedly.Finally, the original structure of BP neural networks was constructed. After trained repeatedly, the optimal number j of hidden layer node and the range of corresponding learning rate L were found according to the Mean Square Error (MSE) of training set and testing set. As a result, the optimal structure of net was obtained. After the formal training, the final inventory model was ascertained. The demand was forecasted. The forecast result was compared with two traditional methods' (regression analysis and classical EOQ Model). The conclusion was drawn after the sensitivity analysis.It is programmed that the computation program for the improved BP algorithm with C++ computer language, which can be used in different inventory items and is helpful to decision-making of inventory management.