Trend-guided Small Hydropower System Power Prediction Based on Extreme Learning Machine

Abstract

A trend-guided extreme learning machine prediction model (TG-ELM) is proposed, which enriches the physical mechanism of the input and output of the traditional extreme learning machine (ELM), and solves the problems that a large amount of raw data in the traditional hydropower prediction model depends on a single data normalization process, which causes the prediction accuracy of the entire prediction model to decrease. The model first smoothed and repaired abnormal output data of the small hydropower group, and then extracted the trend of power change from the processed power curve, and used this trend as a new feature input for the extreme learning machine model to provide a correct and unique prediction trend for the prediction of the extreme learning machine. The model is applied to the power generation forecast of small hydropower groups in Guangxi, which greatly improves the accuracy of the power generation forecast and shortens the calculation time of the forecast. Based on the historical data of the power generation of the small hydropower group in Guangxi in August 2019 for simulation analysis, the root mean square error and average absolute error of the prediction are 4.16 WM and 6.11%, respectively. Compared with prediction methods such as backpropagation neural network and support vector machine, the proposed prediction model has distinct advantages and is more suitable for industrial applications.