An advanced multistage multi-step tidal current speed and direction prediction model

Non-stationarity and non-linearity of the tidal current speed (TCS) and tidal current direction (TCD) time series are among the main barriers for enhancing the TCS and TCD prediction accuracy. In this regard, this paper proposes an improved complete ensemble empirical mode decomposition adaptive noise (ICEEMDAN) which is employed to decompose the non-stationary TCS and TCD time series into several components (modes) with unique characteristics. Then, to capture the nonlinear pattern of TCS and TCD in different modes, several prediction engines based on least squares support vector machine (LSSVM) are developed. To modify the prediction error which occurs in predicting different components, a prediction modification stage based on a combination of extreme learning machines (ELMs) is utilized to reconstruct the final prediction values. The proposed TCS and TCD prediction model, named ICEEMDAN-LSSVM-ELM, has been evaluated using the data recorded from Shark river entrance, NJ. Performance of the proposed prediction model is compared with various well-developed benchmark models.