基于AMSD-WTSSA-DELM模型的铁路沿线短期风速预测方法  

Short-term wind speed prediction method along the railroad based on AMSD-WTSSA-DELM model

作　　者：尼比江·艾力 张林鍹 李奕超[3] 景雨啸 高金山 王渊 谢明浩 罗晓龙 AILI Nibijiang;ZHANG Linxuan;LI Yichao;JING Yuxiao;GAO Jinshan;WANG Yuan;XIE Minghao;LUO Xiaolong(School of Electrical Engineering,Xinjiang University,Urumqi 830047,China;National Engineering Research Center for Computer Integrated Manufacturing Systems,Tsinghua University,Beijing 100084,China;Scientific Research Institute of China Railway Urumqi Group Corporation,Urumqi 830063,China;Hami Power Supply Department,China Railway Urumqi Group Corporation,Hami 835000,China;Power Supply Department,China Railway Urumqi Group Corporation,Urumqi 830011,China)

机构地区：[1]新疆大学电气工程学院,新疆乌鲁木齐830047 [2]清华大学国家计算机集成制造系统工程技术研究中心,北京100084 [3]中国铁路乌鲁木齐集团有限公司科学研究所,新疆乌鲁木齐830063 [4]中国铁路乌鲁木齐集团有限公司哈密供电段,新疆哈密835000 [5]中国铁路乌鲁木齐集团有限公司供电部,新疆乌鲁木齐830011

出　　处：《铁道科学与工程学报》2025年第2期543-556,共14页Journal of Railway Science and Engineering

基　　金：中国国家铁路集团有限公司青年专项课题(Q2023T002);新疆维吾尔自治区自然科学基金资助项目(2022D01C431)。

摘　　要：我国西北地区铁路沿线风速较强且存在非平稳性和波动性,导致风速预测精确度不高、模型泛化性差。基于此,提出一种基于AMSD-WTSSA-DELM的组合预测模型。首先,利用高度非平稳的原始风速序列、分量的长期相关表现、分量所包含的潜在模式及趋势和周期性等内在信息,进行每步分解处理,分别建立分解条件以及自适应更新阈值;为避免过度分解加入自适应重构方法,分解至无高复杂度分量为止,从而实现适应性较强的自适应多步分解。其次,提出WTSSA算法,即通过在麻雀搜索算法(SSA)中融入混沌映射、自适应权重和自适应t分布扰动策略,提升SSA全局搜索和局部探索能力,加快收敛速度,并通过测试函数验证WTSSA算法的卓越性。然后针对AMSD输出的各分量,分别建立由WTSSA优化权重和偏置的深度极限学习机(DELM)模型。最后汇总所有分量的预测数据,合成最终的预测输出。实验结果表明:模型在2组实际铁路沿线风速数据预测性能上提升效果明显,以第1组实验数据为例,本文方法与DELM相比,平均绝对误差(E_(mae))和均方根误差(E_(rmse))分别降低90.32%和82.25%,决定系数(R^(2))提升43.00%。综上所述,研究成果有效克服了风速的非线性特征导致的时迟问题,具有高泛化性能,能够预测短期风速变化,从而帮助铁路系统做出更有效的安全决策,为列车安全运行提供有力的技术支撑。To solve the problems of low prediction accuracy and poor generalization in wind speed forecasting along Northwestern China’s railways caused by strong non-stationarity and stochastic fluctuations,this study proposed a hybrid AMSD-WTSSA-DELM prediction framework.First,the original wind speed series with high non-stationarity,the long-term correlation performance of the components,the underlying patterns,trends and periodicity contained in the components were used to decompose each step,and the decomposition conditions and adaptive update thresholds were established.In order to avoid excessive decomposition,the adaptive refactoring method was added to decompose until there are no high-complexity components,so as to achieve adaptive multi-step decomposition with strong adaptability.Furthermore,the WTSSA algorithm was introduced by integrating chaotic mapping,adaptive weighting and the adaptive t-distribution perturbation strategies are integrated into SSA,which improved the global search and local exploration capabilities of the original SSA,accelerated the convergence speed,and verified the excellence of the WTSSA algorithm through test functions.Then,for each component of AMSD output,a Deep Extreme Learning Machine(DELM)model with WTSSA optimized weights and biases was established.Finally,the forecast data for all components was summarized to synthesize the final forecast output.The experimental results show that the proposed model has a significant improvement effect on the prediction performance of wind speed data along two groups of the actual railway,and the first set of experimental data as an example,the mean absolute error(E_(mae))and root mean square error(Ermse)of DELM reduced by 90.32% and 82.25%,respectively,and the coefficient of determination(R^(2))increased by 43.00%.In summary,the prediction model proposed in this paper effectively overcomes the time-lag problem caused by the nonlinear characteristics of wind speed,which has high generalization performance and can predict short-term wind speed ch

关 键 词：短期风速预测 自适应多步分解 深度极限学习机 改进麻雀搜索算法 铁路沿线风速 

分 类 号：U298.12[交通运输工程—交通运输规划与管理]