Application of Grey Neural Network Combination Model to Forecast Fixed Asset Investment in Beijing of China


  • Chunjiang Yu


Fixed asset investment has consistently been a rather important indicator of the vitality of the economy. Due to its non-linearity and randomness, this paper focuses on establishing improved grey neural network models for better prediction results to help government achieve a better economic decision-making process. Firstly, Simpson formula is utilized to improve the background value of GMC (1, n) model to increase the precision of prediction. Secondly, two kinds of grey neural network models are constructed –residual- modified grey neural network model and parallel grey neural network model both with trapezoid formula and Simpson formula respectively. The two systems share the same input training data for GMC (1, n) model and BP neural network model, while the output training data for the former one is the difference between the GMC (1, n) predicted output and the actual value and the latter one is the actual value. The newly proposed model i.e. parallel grey neural network model connects GMC (1, n) and BP neural network simply by giving different weight to each model based on their precision of prediction. And the results show that the parallel grey neural network model with Simpson formula has the best prediction accuracy of fixed asset investment with both root mean squared error and mean absolute percentage error less than 3% which are far less than that of classic residual-modified grey neural network model. This satisfying result enables this effective model to be utilized to help with government in the economic policy-making process.