结合可逆样本标准化的风速区间预测模型

To support the operation and management of long-span bridges, a study on the probabilistic prediction of wind speed at bridge sites with enhanced generalization ability was carried out. A wind speed interval prediction model based on gated recurrent unit neural network and quantile regression was established. The specific quantiles were used as the upper and lower limits of the interval for prediction to form the wind speed prediction interval. In order to improve the generalization ability of the model, reversible instance standardization was introduced to the wind speed interval prediction model. Reversible instance standardization performed standardization and inverse standardization of input and output for a single sample, unified the mean and standard deviation of each sample input of the model, and reduced the response of wind speed distribution deviation to the prediction accuracy of the model. Taking the non-typhoon data and typhoon data obtained from the on-site monitoring of a large-span cable- stayed bridge as an example, the effectiveness and generalization ability of the established wind speed interval prediction model were compared. The results show that there are remarkable differences between non-typhoon data and typhoon data, which validates that there is a distribution deviation phenomenon of wind speed. The gated recurrent unit neural network combined with quantile regression can effectively perform interval prediction. The model trained by non-typhoon data can perform interval prediction on non-typhoon data that meets the interval coverage requirements. The introduction of reversible sample normalization can effectively enhance the generalization ability of the model. The model trained with non-typhoon data can make interval predictions on typhoon data with maximum wind speeds far exceeding the non-typhoon data, and still maintain an interval coverage level of 0.9.