Regional sea level change prediction based on time-frequency analysis and intelligent algorithm

Abstract

The statistical forecasting model based on time series is one of the main means of sea level forecasting at present stage. However, the mechanism of sea level change is complex. The traditional method has some limitations for non-stationary nonlinear time series forecasting, and the prediction accuracy needs to be further improved. In this paper, we use the monthly mean tide level series from Zhapo Station (1959 ~ 2011), and combine the Ensemble Empirical Mode Decomposition(EEMD), Genetic Algorithm (GA) and Back Propagation (BP) Neural Network to propose a improved EEMD-GA-BP method for regional sea level change prediction. In this study, the EEMD method was used to decompose the original series and generate multiple intrinsic mode functions (IMF) according to different spectral characteristics of signals implied in the tide level series, to stabilize the time series, and improve signal to noise ratio. GA is used to optimize the weights and thresholds of BP Neural Network, due to the difficulty of determining the initial weight and threshold in BP Neural Network. Taking each IMF as the input factor of BP Neural Network, the future trend of each IMF is predicted respectively. Finally, the output of the IMF is reconstructed to obtain the predicted value of the original series. The results show that EEMD can effectively extract multi-time scale signals implicit in the series. BP Neural Network optimized by GA can well predict the future trend of sea level. Compared with the direct use of BP Neural Network algorithm, the use of EEMD for non-stationary non-linear time series smoothing, noise reduction and other processing can effectively improve the prediction accuracy. The use of GA optimize BP Neural Network can improve the accuracy. The EEMD-GA-BP algorithm provides a realistic meaning for the prediction of regional sea level change.