Study And Application Of Demand Forecast In Auto Parts Company

Enterprises must take customer satisfaction and make profit to survive in the fierce competition market. Delivery was more and more focused in QCDS(Quality, Cost, Delivery, and Service) to satisfy the customer. As the traditional way to improve the delivery capacity by holding big inventory or investing a high productivity have huge risk, nowadays, enterprises are likely to build quick response system to cut inventory, and developed a serial of inventory control technology: ABC, EOQ, VMI, JIT, Kanban. These techniques solved the relative demand problem very well, but it’s more important and difficult to control the independent demand, enterprises must focus on demand forecast. M, an engine parts company, is in the downstream of the automobile industry, recognized that "zero inventory = the excess inventory is zero". After years of inventory control application and achieved certain results, M eyes on demand forecast and set topic targets, these topic has significance both in practical and analysis.Based on application of moving average method in company M, the author chose grey model and BP neural network to build demand forecast models and optimization them. The main effects of Different periods and weight, and sales order were applied on moving average to find the effect to prediction results. Two methods were tried to optimize the GM(1, 1) model by smoothing the original data and choosing different initial value. The factors which will impact to customer demand were recognized and selected and applied to the input variables to optimize BP neural network. A product M-H-VVT was chosen to study these 3 kinds of prediction models, a large number of test were applied based on 4 years’ sales volume of M-H-VVT, and the result shows that: moving average and GM(1, 1) model is not suitable to the product which demand with large fluctuations periodically, while BP neural network can get perfect result even the product with periodic, nonlinear, large fluctuations demand. In case of large fluctuations, the smaller n value can achieve higher prediction accuracy in moving average, and if the n value equals to half of the cycle period, the prediction accuracy is the worst; when the samples show increasing or decreasing trend, GM(1, 1) model prediction accuracy can be improved greatly; only the generalization ability been improved, BP neural network prediction model can be applied in practical.In the last, 3 months demand prediction for M-H-VVT products were practiced with BP neural network, two reached prediction goal, but the third month deviated too much, because of the customer developed a new user and increased the demand. Actually, application of prediction technology should being applied in a stability environment, when the environment changed, the prediction results should be revised based on the known information to improve the prediction accuracy.