Research Of Safety Stock Prediction Model Based On Improved BP Algorithm

As a buffer, safety stock is a kind of additional inventory which held by enterprise.It's a very important link for business supply chain market. Safety stock determined by the amount of inventory changes in demand, delivery lead times change, the length of delay in delivery and inventory costs and shortage of storage costs and other factors, has a complex non-linear features, which makes its prediction very difficult.In recent years, artificial neural network develops rapidly.It has a good non-linear which can simulate the human brain activity and has a self-organizing learning, large-scale parallel processing, fault tolerance and the ability to adapt to the external environment. So it has been the preferred prediction method for many projects. But the artificial neural network has some drawbacks itself, such as: slow convergence or no convergence, local minimum problem, random network selection and so on.According to the above problems, this paper uses VSBPNN (Variable-step & Self-adjustment Hidden Nodes Back-Propagation Neural Network) safety stock prediction on based on previous studies, compared to the standard BP algorithm and traditional economic methods, it achieves a better result. The paper main includes the following four areas:(1) Through the research and analysis base of subsidiary dye warehouse of a business Group in Zhejiang Province, this paper summarizes influence factors of the raw materials inventory and its demand for a product and obtains the product and its raw materials two years of storage and a library of data in 2008 and 2009. And pretreatment of the raw data, including removing invalid data, adding blank data points, according to forecast demand for training set and prediction set allocation, integrated and normalized training set data and so on.(2) According to the defects and shortcomings of standard BP algorithm, this paper proposes VSBPNN algorithm and gives ideas and formulas derivation of the improved algorithm. First, calculate the weighted matrix and the iterative process step on the basis of the objective function and constraint functions, and then analyze the output of the adjacent nodes in the hidden layer correlation, remove and merge the hidden layer nodes associated with the larger, so that it can better adapt to the needs of the network and improve the effectiveness of network nodes learning to improve prediction accuracy.(3) Based on the idea of improved algorithm to determine the initial network structure prediction model of raw materials and its products'safety stock. Combining references theory and repeated training method to determine the number of hidden neurons and learning rate of the size of the network, optimization of network structure, to determine the appropriate values and parameters of the final Predictive model after several experiments. Using the data that be put in and out storage training model repeatedly, reducing the uncertainty caused by interference, model mismatch, parameter changes to adjust the network in time so that the appropriate forecasting model based on actual data.(4) Compared with the simulation results getting the optimal solution and then analyzing the performance index of the solution through compared with the standard BP algorithm and other improved algorithms and the traditional economic forecasting methods. After this work get the evaluation of safety stock of prediction model based on improved BP neural network. Finally, summary of the strengths and weaknesses for this paper's work, pointing out that the next step work plan and prediction results can be used as a basis for business decisions.