基于统计推断的汽车运行工况试验采样时间研究

发布时间：2018-06-04 18:35

本文选题：汽车运行工况 + 采样时间　；参考：《吉林大学》2016年硕士论文

【摘要】：由于缺乏相应的理论支持,一直以来,在进行汽车运行工况试验时,试验人员很难确定何时终止采样。为保证试验数据的完备性,试验人员往往会延长采样时间,从而导致大量时间被浪费,延长了汽车开发周期。为解决这一问题,本文利用统计推断方法,从城市道路每天每车行驶次数分布和区域每天每车每平方公里行驶次数分布两个角度,提出了采样时间确定方法。首先,如果要在道路尺度上进行分析,就必须有对应城市的电子地图。本文提出一种面向交通网络分析的道路电子地图生成方法。该方法利用覆盖城市区域的网格与汽车轨迹线的交点存储轨迹的位置信息,大大减小了储存空间和道路生成时的处理时间。同一网格线下,同一道路的所有轨迹线与网格线的交点的平均值就可以代表道路中心线与网格的交点。按照汽车的行驶顺序将交点依次连接就得到了电子地图。汽车运行工况试验的关键在于采样获得的城市道路行驶次数分布要与真实的城市道路流量分布相一致,所以还需要一种快速的地图匹配算法。本文提出一种基于网格的道路匹配算法,相比较于其他方法,此方法更关注找到匹配道路而不是匹配在道路的哪个位置。最终本文得到了汽车在每条道路上的行驶次数和城市交通网络,进而获得道路流量的抽样分布。大量的理论和实证研究发现,城市交通网络是复杂网络。道路的流量分布具有幂律分布特性,也就是说如果采样得到的城市道路行驶次数分布能够真实的反映道路流量分布幂律特性,采样就可以终止。但是,在对城市道路流量分布进行估计之前,需要保证每条道路行驶次数估计的准确性。本文采用中心极限定理对每条道路行驶次数的准确性进行量化估计,当精度满足试验之前确定的阈值时,就可以进行城市道路行驶次数分布的参数估计。由于在双对数坐标下,幂律分布为一条直线,所以文中选用普通最小二乘法来估计分布参数。然而在实际应用中发现,分布参数的置信区间随时间波动太大,难以用于终止时间的判断,所以最终以分布参数的稳定性作为判断采样终止的依据。通过两次模拟采样,得到了长春的合理采样时间为50天左右。由于在道路尺度上需要分析的状态多,分析步骤繁琐,需要进行电子地图生成和道路匹配等原因,造成了估计的采样时间偏长。为解决上述问题,本文还提出一种在区域尺度上确定采样时间的方法。进行区域尺度分析的第一步就是考虑汽车运行工况试验分析在区域尺度的可行性。在文中,通过K核算法和最优分割理论获得了长春的道路K核等级。对同一等级下道路的速度加速度联合概率分布(VA分布)进行统计分析发现,它们之间具有较高的相似性,且同一等级的道路经常聚集在一个区域。这一现象表明,同一K核等级下的道路可以放在一起分析。也就是说,在区域尺度上进行汽车运行工况采样时间的分析是合理的。基于长春100台出租车一个月的试验数据和北京2340台出租车7天的试验数据,本文分别得到了长春和北京的区域行驶次数分布。这两个分布将会作为判断采样分布质量的标准分布。采用K-S检验,发现Nakagami分布和指数分布可以对城市的区域行驶次数分布进行描述。接着,通过统计推断理论估计出了长春和北京采样时间,其中对于Nakagami分布和指数分布的参数估计使用的是最大似然估计法。经计算,长春需要26天就能完成采样,北京则需要95天左右。对区域尺度的采样时间确定方法进行分析后发现,影响采样精度(采样终止条件)的主要因素为每个采样区域行驶次数的准确性。通过分析试验数据发现,采样得到的城市区域行驶次数的变异系数和区域行驶次数抽样分布与真实分布的相似性度量ab?具有线性关系。本文通过公式推导,从理论上证明了这一线性关系。这说明在进行汽车运行工况采样的时候,可以根据计算得到的城市区域行驶次数的变异系数判断是否终止采样。
[Abstract]:Because of the lack of theoretical support, it is difficult for the experimenters to determine when to terminate the sampling during the test of the vehicle operating conditions. In order to ensure the completeness of the test data, the experimenters often extend the sampling time, resulting in a large amount of time wasted and the extension of the vehicle development cycle. This paper uses this problem to solve this problem. The method of statistical inference, from the two angles of the distribution of each car per vehicle per day and the distribution of each square kilometer per square kilometer per day in the city, proposed the method of determining the sampling time. First, if we want to analyze the road scale, we must have the corresponding City Electronic Map. This paper presents a traffic network analysis. The method of generating the road map. This method uses the intersection of the grid covering the city area and the car track line to store the location information of the track, which greatly reduces the storage space and the processing time of the road generation. Under the same grid, the average value of the intersection of all the track lines and the grid lines on the same road can represent the center of the road. The intersection of line and grid. The key of the vehicle running test is that the distribution of urban road travel times is in accordance with the true urban road flow distribution, so a fast map matching algorithm is needed. Compared with other methods, the road matching algorithm in grid is more concerned about finding the way to match the road instead of matching the road. Finally, this paper obtains the number of cars on each road and the urban traffic network, and then obtains the sampling distribution of the road traffic. A large number of theoretical and empirical studies find that the city is a city. Traffic network is a complex network. The flow distribution of road has power law distribution, that is to say, if the distribution of urban road travel times can truly reflect the power law characteristic of road flow distribution, sampling can be terminated. But before estimating the urban road flow distribution, every road must be guaranteed. The accuracy of the number of times is quantified by the central limit theorem. When the precision meets the threshold determined before the test, the parameter estimation of the number of urban road travel times can be estimated. Because the power law distribution is a straight line in the double logarithmic coordinates, the article selects the general. In the practical application, it is found that the confidence interval of the distribution parameters fluctuates too much with time, and it is difficult to use the judgment of the termination time. Therefore, the stability of the distribution parameters is used as the basis for judging the termination of the sampling. The reasonable sampling time of Changchun is 50 days by the two simulated sampling. In order to solve the above problems, a method to determine the sampling time on the regional scale is also proposed. The first step of the regional scale analysis is to take the examination of the regional scale. The feasibility of vehicle operation test analysis at regional scale is considered. In this paper, the road K nuclear grade in Changchun is obtained by K kernel algorithm and optimal segmentation theory. The statistical analysis of the joint probability distribution of velocity and acceleration (VA distribution) on the same grade road shows that they have high similarity and the same grade of road between them. It is often gathered in one area. This phenomenon indicates that the road under the same K nuclear grade can be analyzed together. That is, it is reasonable to analyze the sampling time of the vehicle operating conditions at the regional scale. Based on the one month test data of 100 taxis in Changchun and the test data of 2340 taxis in Beijing for 7 days, this paper respectively The distribution of regional travel times in Changchun and Beijing will be obtained. These two distributions will be used as the standard distribution of sampling distribution quality. Using K-S test, it is found that Nakagami distribution and exponential distribution can describe the distribution of regional travel times in cities. Then, the sampling time of Changchun and Beijing is estimated by statistical inference theory. The maximum likelihood estimation method is used for the parameter estimation of the Nakagami distribution and the exponential distribution. It is calculated that Changchun takes 26 days to complete the sampling and Beijing takes about 95 days. After analyzing the sampling time determination method of the regional scale, it is found that the main factors affecting the sampling accuracy (sampling termination condition) are each sampling area. Through the analysis of the test data, it is found that the variation coefficient of the number of urban travel times and the sampling distribution of the regional travel times are linear with the similarity measure of the true distribution of the urban area. This paper has proved this linear relationship theoretically by derivation of the formula. This shows that the running condition of the car is carried out in the operation of the car. At the time of sampling, we can decide whether to stop sampling according to the coefficient of variation of the number of times traveled in the urban area.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：U467

【相似文献】