基于DeepAR的短期风速概率预测

Short-term Wind Speed Probabilistic Prediction Model Using DeepAR

为对铁路沿线风速提前进行预判,保障桥梁施工及高速铁路列车运行时的安全,提出基于深度自回归模型(DeepAR)的短期风速预测方法。采用平潭海峡公铁两用大桥和西堠门大桥实测风速进行验证,并以包括小波包分解下的卷积神经网络和循环神经网络混合模型(WPD-CNNLSTM-CNN)在内的4种模型作为点预测对比模型,以SimpleFeed-Forward、ARIMA、Random Walk模型进行置信度为50%与95%的区间预测作为对比模型。研究结果表明:无论是点预测还是区间预测,DeepAR模型都能够在具有随机性、间歇性的短期风速序列中提取到特征信号并进行精度较高的预测,且相比于其他模型具有更好的准确性与泛化能力,可满足实际工程短期风速预测需求。

In order to predict the wind speed along the railway in advance and ensure the safety of bridge construction and high-speed train operation,a short-term wind speed probabilistic prediction method based on the DeepAR was proposed.The method was verified by the measured wind speed data of the Pingtan Straits Highway-Railway Bridge and the Xihoumen Bridge.Four models including the hybrid model using wavelet packet decomposition,convolutional neural network and convolutional long short term memory network(WPD-CNNLSTM-CNN)were used as the comparison model for point prediction.The Simplefeed-forward model,auto-regressive integrated moving average(ARIMA)model and Random Walk model were used to carry out interval prediction with confidence of 50%and 95%as the comparison model for interval prediction.The experimental results show that in both point prediction and interval prediction,the DeepAR prediction model can extract characteristic signals from random and intermittent short-term wind speed sequences and make predictions with high accuracy.Compared with other models,the DeepAR model,with better accuracy and generalization ability,can meet the needs of short-term wind speed prediction in practical construction projects.