Monthly wind distribution prediction based on nonparametric estimation and modified differential evolution optimization algorithm

To copy with global warming and increasing energy demand, wind power has been rapidly developed around the world in recent years. It is important to analyse the characteristics of wind speed distribution to improve the development and utilization of wind energy. Many studies focus on improving the estimation accuracy of wind speed distribution, but there are few studies on variation characteristics and predictability of it. Here, we present a novel horizontal-vertical-integration framework to predict wind speed distribution. To address the predictability problem of nonparametric estimation, we proposed a data sampling and mapping method. To indirectly optimize the learning rate of the hybrid neural network, an improved differential evolution optimization algorithm was designed. The effectiveness of the proposed methods was verified by using data of 9 wind stations and 6 comparison models. The results show that the absolute error of the proposed prediction framework is less than 0.0059, which is better than other comparison models.

Automated summary

To cope with global warming and increasing energy demand, wind power has been rapidly developed around the world. Analyzing the characteristics of wind speed distribution is crucial for improving the development and utilization of wind energy. While many studies have focused on improving the estimation accuracy of wind speed distribution, there is limited research on its variation characteristics and predictability. In this study, a novel horizontal-vertical-integration framework is proposed for predicting wind speed distribution. To address the predictability issue of nonparametric estimation, a data sampling and mapping method is proposed. Furthermore, an improved differential evolution optimization algorithm is designed to indirectly optimize the learning rate of the hybrid neural network. The effectiveness of the proposed methods is validated using data from 9 wind stations and 6 comparison models, and the results demonstrate that the absolute error of the proposed prediction framework is less than 0.0059, outperforming the other comparison models.