中文摘要：

为了处理车辆轴温可能出现的跳变、缺失、噪声等异常数据，有效降低误报率，提出了基于动态时间规整算法的车辆轴温状态监测方法，将轴温历史监测数据和历史统计数据进行指数平滑预处理，在训练阶段将数据反复迭代得到不同轴温模式的参考样本，计算了实时轴温和参考样本各数据帧之间的欧氏距离，得到帧匹配距离矩阵，运用动态规划和回溯的思想，求出累积距离矩阵和动态规整路径，将动态规整距离作为2个时间序列相似度的量化指标，找出最小动态规整距离对应的轴温模式，从而得到状态监测结果。仿真结果表明：在MATLAB仿真中，输入1000个时长为50～300min的轴温测试样本，其最大响应时间小于0．4S，共出现29次错误匹配，误报率低于3％。通过对测试样本和参考样本的各数据帧进行指数平滑处理，有效消除车辆轴温出现跳变的干扰，虽然跳变值和跳变点数量不同，但相对动态规整距离无变化，对状态监测结果无影响。可见，车辆轴温状态监测方法能够满足车辆轴温状态监测的实时性和准确度要求，减少了误报率。

英文摘要：

In order to handle with the abnormal data of vehicle axle temperature, such as jump, deletion and noise, a monitoring method of vehicle axle temperature based on dynamic time warping method was put forward to reduce the false alarm rate. The historical monitoring data and historical statistical data were preprocessed by using exponential smoothing method. At the training stage, the data were iterated to get the reference samples of different axle temperature modes. The frame matching distance matrix was obtained by computing Euclidean distances of data frames between real-time axle temperatures and reference samples. With the idea of dynamic programming and backtracking, the cumulative distance matrix and dynamic time warping path were calculated. The dynamic time warping distance was taken as the quantitative similarity index of two time series to the corresponding axle temperature mode for the minimum dynamic time warping distance, thus the axle temperature condition was achieved. Simulation result shows that when 1 000 test samples of axle temperature with the time ranges of 50 min to 300 min are inputted in MATLAB, the maximum response time is less than 0.4 s, there are 29 false matches, and the false alarm rate is below 3~. The jump interferences of axle temperature are effectively eliminated by processing the data using exponential smoothing method. The values and numbers of axle temperature jumps are different, but the relative dynamic time warping distances are invariable. Obviously, the method can meet the real-time and accuracy requirements of vehicle axle temperature monitoring and reduces the false alarm rate. 3 tabs, 6 figs, 21 refs.