铁路沿线风速短时智能预测模型及其大数据嵌入方法研究

Strong wind is one of the main natural disasters that cause the railway accidents. Strong wind along the railway makes the vibrating phenomenon of the running trains. It leads to the obvious increasing of the derailment and over-turning accidents, which endangers the safety of railways seriously. Since the wind speed along the railway shows strong random and nonlinear characteristics, it is still an unsolved scientific problem to describe the space-time changing law of the wind speed signals accurately and to realize the wind speed high-precision multi-step forecasting. This project will adopt the hybrid researching methods include the spot investigation, modeling calculation, embedded development and experimental evaluation to carry out the study on the intelligent railway short-time wind speed time series forecasting models and their big data embedding method. The main contents of the project include: ①Studying on the decomposition & reconstruction, reduction and clustering calculation of the non-stationary wind speed time series from the independent wind measurement stations along the railway, the changing law of the wind speed time series from the single point and the building of intelligent wind speed forecasting models for each independent wind measurement station; ②Studying on the spatial correlation identification of the wind speed time series from the associated wind measurement stations along the railway and the construction of association zones for these associated stations, the spatial correlation characteristics of the multi-point wind speed time series and the building of the spatial depth wind speed forecasting models for these associated wind measuring stations; ③Development of the software modules for the built wind speed forecasting models in the big data embedding environment based on the proposed forecasting algorithms. Evaluation of the built wind speed forecasting models and their related software computing modules. The research results will provide important theoretical and technical support for the development of the railway safety protecting framework under the strong wind environment.

强风是造成铁路行车事故的主要自然灾害之一。铁路沿线强风引发车体剧烈颤动，脱轨和翻车概率徒增，严重危及行车安全。铁路沿线风速呈现强随机非线性特性，对它实现时空演变规律准确描述和高精度多步预测仍是未解决的科学难题。本项目拟采用现场调研、建模计算、嵌入开发与实验评估相结合的手段，开展铁路沿线风速短时智能预测模型及其大数据嵌入方法研究，主要内容包括：①开展铁路沿线独立测风站非平稳风速时序的分解重构、降维和聚类计算，揭示铁路沿线单点风速时序的时间演变规律，建立各独立测风站的风速时间智能预测模型；②开展铁路沿线关联测风站的风速空间相关性辨识及关联单元构建，探明铁路沿线多点风速时序的空间关联特性，建立关联测风站的风速空间深度预测模型；③基于预测算法成果，开发大数据环境下风速预测模型嵌入软件模块。对预测模型和软件模块开展性能评估。研究结果为构建大风环境铁路行车安全保障体系提供理论基础和技术支撑。

铁路沿线风速预测是保障大风环境列车行车安全的重要措施之一。本项目聚焦铁路沿线风速高质量预测，取得如下成果：.（1）开展了铁路沿线风速特征预处理研究，使用二次分解算法精细划分了风速时序波动特征，筛选了风速显著特征，完成了风速聚类计算；.（2）开展了铁路沿线独立测风站风速混合智能预测研究，提出了融合多小波分析的多目标集成风速预测算法，实现了混合Boost提升算法与后处理滤波的风速大步长预测，建立了“修正-检验-描述”闭环的风速不确定性预测模型；.（3）开展了铁路沿线关联测风站风速混合预测研究，构建了融合“时域滤波-空间筛选-集成预测”多测风站混合风速预测模型，完成了WRF风速空间预测计算，并提出了面向WRF后处理的深度降尺度神经网络；.（4）开展了铁路沿线风速大数据预测研究，建立了风速强化学习与高分辨率大数据预测模型库，开发了风速Spark大数据预测软件模块DCPELM。.项目出版Elsevier、Springer等专著3部，其中1部获斯普林格·自然“中国新发展奖”；发表SCI论文46篇，其中ESI 1%高被引论文2篇；授权日本/俄罗斯/澳大利亚发明专利5项、中国发明专利4项；获法国巴黎国际发明展览会专利金奖2项、国家发明展览会专利金奖1项；培养博士/硕士研究生6名，博士后1名。