The Research Of Demand Forecasting Model Based On Improved Gray System, BP Neural Network And Support Vector Machine

In the operation process of no matter raw materials production or procurement of the finished product, there need to take up a lot of money, inventory and production equipment, so to do very precise demand forecast for the capital of the enterprise operation, inventory management and production equipment utilization will be very important practical significance. Many companies are facing the inventory has been occupied by large, low utilization of funds, these question became to the bottleneck of restricting corporate profit growth, demand forecast inaccuracy is one of the very important reason, so the excavation of the demand forecast potential is very necessary. The practice has proved that a scientific and reasonable demand forecasts can play a great role in promoting the development of enterprises.Based on the theory of gray theory, BP neural network and support vector machine, contributed a combination forecasting model based on improved gray-BP neural network-support vector machine, the model has a high precision prediction. The main context is described as follows:(1) Build improved gray prediction model and BP neural network prediction model. First built an improved gray prediction model, and do a numerical example for the improved gray prediction model; through a comprehensive analysis of the factors affecting demand for cars, considering the GDP, oil consumption, population, per capita consumption level and the highway mileage of the impact on demand for cars, build the BP neural network model, and do a numerical example on auto demand forecast.(2) Build a support vector machine prediction model and its parameter optimization. In this paper, demand forecasting belong to a nonlinear regression problem, build a non-linear support vector machine for regression prediction model, select the Gaussian radial basis as its nuclear function and particle swarm optimization algorithm on the impact of support vector machine model prediction accuracy parameter has been optimized to improve the prediction accuracy of support vector machine model;(3) Build a linear combination forecasting model based on improved gray theory-BP neural network-support vector machine. The model is based on the squared error and minimum as the objective function, the model can calculate the proportion of the three forecasting methods in combination forecasting model. And2007-2011forecast data and actual data comparing the results of the inspection. The results show that the combination forecasting model accurate than the average forecast error of any single prediction model.