| Economic investment prediction is an important issue related to the nation's economy and the people's livelihood, especially industrial investment is concerned widely. The character of the multifactor time series of economic investment prediction determines that it is a nonlinear prediction, so that it is difficult to predict using traditional methods. However, the Radial Basis Function (RBF) Neural Network has excellent nonlinear character, especially for highly nonlinear proceeding. By the introduction of RBF neural network theory, which is applied widely in other fields, the paper establishes a new economic investment prediction model, opening up a new development space for economic investment.In the paper, we make an overview on the theories of investment and economic prediction and the structure, property and training algorithm of the RBF neural network. For the historical data, which are data of investment value and gross production value of Heilongjiang province from 1999A.D.to 2003A.D, we establish models, do the simulation, and analyze simulated results .The main innovation points are as follows:1. It decides the main elements of the problem by grey relationship analysis to all relations among all elements of the investment. Taking the historical data in many years as samples, the grey relationship degrees among all elements is analyzed.2. The thesis establishes a model using RBF neural network, do simulated training. The relative errors of the forecast results are all no more than 3%. And it compares the results with results from the model of BP neural network algorithm. Simulated experimental results show that the RBF neural network has a faster training speed, higher accuracy and better generalization ability than those of the BP neural network.3. Using enumerative method for fine-tuning to investment distribution in a small range, this paper determines the greatest gross production value. Taking the two smaller investment values in the primary, secondary, and tertiary industries as the center, we make adjustments in a certain range.4. It applies to make a reasonable adjustment to the investment distribution by the thought of the multi-grid, and determine the optimal investment distribution. It can effectively eliminate the error of frequency weight, and enhance iterative convergent ratios to construct multiple layers grids (5%, 10%, 15%, 20%) , whose sizes are from small to big, and iterate in each layer grids. And multi-grid method can also greatly reduce workload.Based on established model of investment prediction and optimized structure, we predict respectively the investment distribution and gross product of 2004 .The result is very consistent with the law of development of economic systems. The model reflects the nonlinear relationship among elements of the investment, it has a great prediction effect, and it provides a new effective method for the economic investment prediction. |
