Research on wind speed behavior prediction method based on multi-feature and multi-scale integrated learning

Wind power is just in place for a major improvement to the power grid in security and economy. More accurate forecast methods are emerged to benefit grid operators and their customers. A wind speed behavior prediction method based on multi-feature and multi-scale integrated learning (MFMS) is developed in this paper. The behavior characteristics are proposed to solve the problem of the temporal and spatial feature extraction distinctly of wind speed. Multi-scale feature fusion and spatial pyramid pooling algorithms are proposed to reduce the loss of micro-space-scale and short-time-scale information. Meanwhile, environmental features and background features are applied to consider the influence of the overall wind farm environment on the wind speed of local wind turbines. The wind speed data of sixteen wind turbines in a wind farm at Zhangjiakou in North China are used to verify the method. Results show that MAPE, RMSE and R2 of 4-h ultra-short-term wind speed prediction are 6.164%, 0.275 and 0.966, respectively, which are improved by 1.786%-6.757%, 0.051-0.216 and 0.014-0.074 compared with other methods. The method proposed in this paper can be applied to wind speed prediction and provide a guideline for wind farm power prediction.