典型工况构建方法对新能源汽车能耗评价的影响分析

新能源汽车的能耗及其经济性与行驶工况高度关联。为了对新能源汽车的能耗进行合理评估,以西安市为例,分别应用聚类分析法、聚类马尔可夫分析法、短行程车速-加速度(Velocity-Acceleration,V-A)矩阵法和变步长V-A矩阵法构建城市客车运行工况,并进一步提出基于自组织映射(Self-organizing Maps,SOM)神经网络聚类的V-A矩阵法。对5种方法构建的工况进行对比和误差分析。在此基础上,基于一款插电式混合动力城市客车,应用全局优化方法——庞特里亚金最小值原理设计能量管理策略,分析车辆能耗和经济性以及5种工况的优缺点。研究结果表明：聚类分析法构建工况的平均特征值误差最大,计算量较大;变步长V-A矩阵法的平均特征值误差小于聚类法,计算量最小;短行程V-A矩阵法与变步长V-A矩阵法误差接近;聚类马尔可夫法的误差居中,计算量最大;基于SOM聚类的V-A矩阵法的平均特征值误差最小,能反映不同路段以及运行时间的差异,且能在聚类之后快速提取短行程的同时兼顾速度和加速度的分布;从能耗角度来看,基于SOM聚类的V-A矩阵法的能耗在5种方法中居中;聚类分析法构建的工况平均车速低于其他工况,但加减速频繁,能耗成本最高;聚类马尔可夫法由于对车速进行平滑处理,加减速频繁程度最小,能耗成本最低。

Impact Analysis of Typical Driving Cycle Construction Methods for Energy Consumption of New Energy Vehicles

The energy consumption and economy of new energy vehicles are highly correlated with the driving cycle. To reasonably evaluate the energy consumption of new energy vehicles, considering the city of Xi'an as an example, four approaches- cluster analysis, cluster Markov, micro-trip velocity-acceleration (V-A) matrix method, and variable step V-A matrix method, were used to construct urban bus driving cycles. Moreover, a new approach that jointly uses the self-organizing map (SOM) neural network clustering and V-A matrix method was proposed for driving cycle construction. A comparison and error analysis of the five methods were performed. Then, Pontryagin's minimum principle-based global optimization method was applied to design the energy management strategy for a plug-in hybrid urban bus. The energy consumption and economy as well as the characteristics of the five methods were calculated. The results show that: ① the average eigenvalue error of the cluster analysis method is the largest and the calculation amount is relatively large; the average eigenvalue error of the variable step V-A matrix method is smaller than that of the cluster analysis method, and the amount of calculation is the smallest; the error of the micro-trip V-A matrix method is close to that of the variable step V-A matrix method; the error of the clustering Markov method is in the middle, and its amount of calculation is the largest. The V-A matrix method-based SOM neural network clustering has the smallest error in the average eigenvalue error, which can reflect the difference between driving routes and times, and contains the characteristics of velocity and accelerates when fast extracting the micro-trip. ② From the perspective of energy consumption, the V-A matrix method-based SOM neural network clustering is at an intermediate level among the five approaches; the average velocity of the driving cycle constructed by the cluster analysis method is lower than that of the other four, but the acceleration and deceleration are frequent, and the cost of energy consumption is the highest; because the cluster Markov analysis smooths the velocity of the vehicle, the frequency of acceleration and deceleration is the lowest, and the energy consumption cost is the lowest.