基于混合约束自编码器的机动车工况智能构建方法

为构建更具代表性的机动车行驶工况,实测采集福州地区1辆机动车共20d的真实驾驶数据,选取14个特征参数表征运动学片段信息,运用主成分分析和K-means聚类划分运动学片段聚类,根据聚类中心的距离筛选备选片段并随机组合构建工况集合。提取11个特征参数计算构建工况的误差,选择集合中误差最小的工况作为构建工况,提出利用混合约束自编码器构建工况优化模型,并研究参数标定方法,最终将平均误差由2.97%缩小到2.39%。混合约束自编码器模型的分析验证结果表明,优化策略符合实际情况,可以有效避免随机选择带来的误差不确定性,验证了所提出行驶工况构建流程的合理性,并提升了工况预测的精确度,得到模型参数推荐值。对实现碳达峰目标下的机动车碳排放预测及排放控制具有重要的现实作用和意义。

Intelligent Construction Method of Vehicle Condition Based on Hybrid Constrained Autoencoder

In order to contract a representative vehicle driving cycle, real-world driving data recorded on 1 Hz of amotor vehicle in Fuzhou area for 20 days are collected, 14 characteristic parameters were selected based on themeasured driving data to represent the kinematic fragment information. The principal component analysis and K-meansclustering were applied for clustering the divided kinematic fragment, and the candidate fragments were chosenaccording to the distance of the clustering center and randomly combined to construct the condition set. Elevencharacteristic parameters were selected to calculate the error of the construction driving cycle. We choose the drivingstate with the most minor error in the set as the construction driving condition, and propose the optimization ofconstruction condition through hybrid constrained autoencoder, which reduces the average error from 2.97% to 2.39%.The hybrid constrained autoencoder model's analysis and validation show that the optimization strategy aligns with theactual situation and can effectively avoid the error uncertainty caused by the random selection. It verifies the rationalityand accuracy of the driving condition's construction process and gives model parameter recommendations value. It hassignificant practical effects and significance for achieving the carbon emission prediction and emission control ofvehicles under the carbon peak target.