基于WRF数值模式的SDAE-SVR风速预测模型研究

风速预测是风力预报中的核心与基础, 采用天气研究和预报(Weather Research and Forecasting, WRF)模式进行风力预报往往存在风速预测误差较大的问题. 为了提高风速预测精度, 提出了一种基于深度学习和支持向量回归(Support Vector Regression, SVR)相结合的风速预测模型. 该模型以WRF模式预报输出的多种气象变量为基础, 结合气象自动观测站传感器的实测风速, 引入堆栈降噪自动编码(Stacked De-noising Auto-Encoder, SDAE)深度网络来学习样本数据中隐含的深度特征, 然后将该深度网络最后一层输出的深度特征置入回归器SVR中, 利用SVR良好的回归预测性能对WRF模式预报的未来1h风速进行预测订正. 结果表明: 所建立的SDAE-SVR风速预测模型具有较高的风速预测精度, 在对典型日的WRF模式预报未来1h风速的预测订正中, 其平均百分比误差与均方根误差仅为8.28％与0.8 066 m·

Wind speed forecasting model of SDAE-SVR based on WRF numerical pattern

Wind speed forecasting is essential for wind forecasting. Using the Weather Research and Forecast (WRF) pattern in wind speed forecasting constantly comes up with some problematic issues with it. To tackle the problem in order to improve the accuracy of wind speed prediction, a forecasting model combining deep learning with Support Vector Regression (SVR) is proposed in this paper. The input data for the model is collected based on the various meteorological variables generated from the WRF pattern, and are combined with the real wind speed collected from the sensor of automatic meteorological observation. Stacked De-noising Auto-Encoder (SDAE) deep network is constructed to learn the intrinsic data characteristic, then the last layer of the deep network output characteristics is fed into SVR, and the good regression forecasting performance of the SVR is used to correct the future 1h wind speed forecasting of the WRF pattern. The experimental results show that the established SDAE-SVR wind speed forecasting model performs with higher accuracy in correcting the future 1h wind speed prediction of WRF pattern for typical days, and the mean average percentage error as well as root-mean-square error are limited within the range of 8.28％ and 0.8 066 m·