城市道路汽车行驶工况构建方法

为了优化汽车行驶性能, 制定了反映中国实际道路行驶状况的测试工况, 以轻型汽车道路实测数据为数据源, 提出了城市道路汽车行驶工况构建方法; 数据采集覆盖主要时段和道路, 剔除了异常数据, 并引入多尺度小波变换对车速降噪; 利用3层小波分解过滤地面扰动的影响, 保留车速关键信息; 基于9种与行驶特性密切相关且具有代表性的特征参数建立汽车运动学片段特征体系; 分别利用主成分分析和自编码器对特征降维处理, 使用K-means++聚类算法确定运动学片段, 并引入Silhouette函数筛选聚类结果以替代人工选择, 确定聚类类别为2类; 以与相应聚类中心的距离为指标, 筛选出各类别中最能反映本类别特性的200个运动学片段, 作为候选运动学片段, 最终以基于最小性能值的评估方法确定代表性运动学片段, 完成了汽车行驶工况的构建, 分别得到主成分分析和自编码器2种降维处理对应的汽车行驶工况曲线。计算结果表明: 以主成分分析和自编码器2种处理方法为基础构建的汽车行驶工况对数据源均体现了较高的代表性与合理性, 基于主成分分析降维最终得到的数据与数据源的相对误差绝对值多数低于10%, 其中平均速度、平均行驶速度、怠速时间比、加速时间比、减速时间比、平均加速度、加速度标准差、平均减速度的相对误差分别为0.75%、5.50%、9.14%、9.80%、9.98%、8.45%、6.17%、7.73%, 仅速度标准差的相对误差较大, 为24.31%, 与自编码器方法得到的结果相比具有更强的综合代表性, 更适合用于汽车行驶工况的构建。

Vehicle driving cycle construction method of urban roads

In order to optimize vehicle driving performance, test cycles reflecting actual road driving cycles in China were constructed. Taking the measured road data of light-duty vehicles as the data source, the vehicle driving cycle construction method of urban road was proposed. Data collection covered the main period and roads, the abnormal data were eliminated, and the multi-scale wavelet transform was introduced to reduce the noise of vehicle speed. Three-layer wavelet decomposition was used to filter the impact of ground disturbance, and the key information of vehicle speed was retained. The car kinematics segment feature system based on 9 representative characteristic parameters closely related to the driving characteristics was established. Principal component analysis and auto-encoder were used to reduce the dimension of features. K-means++clustering algorithm was used to determine the kinematics segments and the Silhouette function was introduced to filter the clustering results to replace manual selection, and determine the number of clusters of 2 categories. The distance from the corresponding cluster center was used as an indicator, 200 kinematic segments in each category that can best reflect the characteristics of the category were selected as the candidate kinematic segments. Finally, the representative kinematic segments were determined based on the minimum performance value evaluation method, the construction of vehicle driving construction was finished, and the corresponding vehicle driving cycles curves based on principal component analysis and auto-encoder were obtained respectively. Calculation results show that vehicle driving cycle construction based on principal component analysis and auto-encoder is highly representative and reasonable according to the data source. The absolute valves of relative errors between the data based on principal component analysis and the data source are mostly less than 10%. The relative errors of average speed, average driving speed, idle time ratio, acceleration time ratio, deceleration time ratio, average acceleration, acceleration standard difference, and average deceleration are 0.75%, 5.50%, 9.14%, 9.80%, 9.98%, 8.45%, 6.17% and 7.73%, respectively. Only the relative error of velocity standard deviation reaches 24.31%. Therefore, the principal component analysis has stronger comprehensive advantages than the results obtained by the auto-encoder method, and it is more suitable for vehicle driving condition construction.