Input wind speed forecasting for wind turbines based on spatio-temporal correlation

Abstract

The scale and power of wind turbines are constantly increasing, resulting in more obvious wake effect in wind farms. Each wind turbine is affected differently by the wake, which increases the prediction deviation of input wind speed. Although Lidar can obtain accurate wind speed in advance, it may only be installed on part of the wind turbine in the wind farm due to economic costs. Accordingly, a method for predicting input wind speed of downstream wind turbines in real time with data from upstream wind turbines is proposed. Based on the wake effect, the method groups upstream and downstream wind turbines together to build a prediction model and improves prediction accuracy by mapping mechanical anemometers to Lidar. Firstly, matched upstream and downstream wind turbines are calculated by the WFSim model and the correlation between wind turbines. Then a Boost-CNN-GRU model based on an iterative approach over sliding windows is constructed to predict the input wind speed of the downstream turbine. The experimental results show that the prediction accuracy of the proposed method meets the engineering requirements with MAPE below 3%, MAE below 0.3, and RMSE below 0.4.