Probabilistic Cross-wind Speed Prediction For Train Driving On Railroad Bridge

Strong wind has a great influence on the safety of railway.It is one of the key natural factors that need to be paid attention to.Especially to the bridges in mountainous area,the terrain is special and the wind field is complex,causing that the railway traffic is more significantly affected by the strong wind.In recent years,the high-speed railway construction in the central and western regions of China has developed rapidly,and the proportion of the bridge in the whole railway line is further increased.It is of great significance for the safety of the whole line and the comfort of the train to carry out the study of the short-term wind speed prediction at the railway bridge position.At present,the theory of wind speed prediction is mainly divided into physical methods,such as climate and other factors,and statistical methods focusing on short-term forecasting.The short-term wind speed prediction mainly includes linear time prediction model,nonlinear intelligent model and mixed model.Most of these methods are applied to wind power generation and wind power forecasting,the related research in rail transit is less.In this thesis,a hybrid prediction model based on the variational mode decomposition and kernel density estimation is proposed to predict the short time wind speed at the bridge position along the railway line,in order to achieve the high precision prediction of wind speed and the probability prediction of wind speed.The specific research work is as follows:1)In this thesis,two commonly used data decomposition methods,empirical wavelet decomposition and discrete wavelet transform,are compared with variational mode decomposition.By constructing complex signals containing high,middle and low frequency,the decomposition performance of the three methods is tested,and the correlation coefficients are compared with the original sequences.The advantages and disadvantages of the decomposition methods in the selection of decomposition layers,modal aliasing,false components and decomposition precision are analyzed.The results show that variational mode decomposition possesses excellent data decomposition ability and can accurately decompose non-stationary and nonlinear complex signals.2)In this thesis,we select a set of measured storm flow data in the mountain area.On the basis of the variational mode decomposition,the multiple kernel density estimation and the conditional kernel density estimation are applied to the decomposed wind velocity subsequence,and the deterministic prediction is realized with the kernel regression model,and then the multivariable numerical characteristics and interval estimation methods are combined to realize the wind speed probabilistic prediction.As final result: high precision predictive value of wind speed and 95% probability level wind speed range are obtained.3)In this thesis,five common indicators are used to compare and analyze the deterministic prediction results of the hybrid method proposed in this thesis and the single KDE method.At the same time,the probabilistic prediction results of the hybrid model are evaluated by using the predictive interval coverage probability(PICP),the average coverage error(ACE),the predicted interval normalized mean width(PINAW),the coverage width criterion(CWC)and the definition quality(WS).The results show that the deterministic prediction accuracy of the hybrid model is higher than that of a single model,the results of probabilistic prediction conforms to the actual situation,so the hybrid method has a higher engineering reference value.