Unbiased Grey Model Combined with Artificial Intelligent Prediction Models

We propose the modified residual method based on the unbiased Grey Model (GM(1,1)) combined with the Auto-Regressive Integrated Moving Average (ARIMA) and backpropagation neural network (BPNN). The model adopts BPNN when describing the nonlinear characteristics of the curve, and the GM-ARIMA model, which is widely used in linear time series prediction. Specifically, the BPNN model is used to simulate the curve and extract the nonlinear characteristics. Then, the residuals generated by the nonlinear model contain only linear features. Next, the GM-ARIMA model is applied to fit the residuals to extract the linear features. A Hybrid prediction model based on BPNN and GM-ARIMA models has the predicted values of the BPNN model combined with the predicted values of the residuals that are fitted by the GM-ARIMA model. In this research, the model enhances the local search ability of the genetic algorithm and avoids the neural network training falling into local minima. The results show that the improved GM-ARIMA and GM-BPNN model have strong searchability, improve accuracy, and shorten the training time. Finally, the application of the model and the prediction results are described by example tests.