Development prediction of digital economy based on ARIMA and improved BP neural network

Digital economy is the core engine of China’s economic growth, and its scale prediction has strategic significance for policy making and industrial planning. In this study, the combined prediction model of ARIMA and improved BP neural network is constructed to systematically analyze the evolution law of digital economy’s proportion in GDP from 2005 to 2022. Aiming at the defects of traditional BP network, such as local optimal and slow convergence speed, the momentum factor (0.8) and Sigmoid activation function are introduced to optimize the network structure, and an improved 5-11-1 topology model is established. The empirical results show that the RMSE (18.241) and MAPE (15.24%) of the improved model are 39.7% and 20.6% lower than those of the traditional BP network during 2020-2022. The prediction accuracy of the combined model (weighted by inverse error variance) is 98.71%, which is 0.39 percentage points higher than that of the single ARIMA model. It can be seen from the research results that the proportion of digital economy will maintain an annual growth rate of 1.5-2.0 percentage points from 2023 to 2025, in which the combined model is predicted to account for 44.30% in 2025, which is closer to the actual development trend than the single neural network model (41.22%). From this, we can see the feasibility of BP neural network in the prediction of the development of digital economy.