基于BP神经网络及其优化算法的汽车车速预测

发布时间：2018-04-29 10:54

本文选题：车速预测 + 特征参数　；参考：《重庆大学》2014年硕士论文

【摘要】：汽车车速预测属于汽车节能与安全研究的重要组成部分之一，可用于汽车节能安全辅助驾驶、路线导航、自动变速器档位控制，以及混合动力汽车预测控制。因此汽车车速预测研究有着重要的理论基础和应用价值。本文主要完成以下研究工作： ①通过Kruskal-Wallis单因素方差分析和相关性分析，确定了BP神经网络车速预测模型的5个输入特征参数：平均车速、怠速时间比例、速度乘加速度方差、速度方差、正加速度均值。通过上述5个特征参数的箱线图分布及其统计数据，确定了以平均车速和怠速时间比例值为门限的6个BP神经网络车速预测子模型的分类界线，这6个分类界线确定了6种车速预测工况。 ②根据BP神经网络理论建立8输出节点和60输出节点的BP神经网络车速预测模型，每个车速预测模型根据车速预测工况包括6个子模型，子模型的输入部分均包括上述5个特征参数和历史邻近车速。通过对两个模型的训练和预测结果，结合汽车行驶工况统计分析，对BP神经网络车速预测模型的输入历史邻近车速长度和输出节点个数进行修正，建立了修正输入输出节点的BP神经网络车速预测模型，该模型在训练和预测上均取得了优于修正前的效果。 ③根据BP神经网络车速预测模型权阈值分布范围的差异，对遗传算法初始种群进行差异化初始化，利用遗传算法对BP神经网络车速预测模型的权阈值进行优化。遗传算法优化后的BP神经网络车速预测模型训练和预测效果均好于优化前。 ④根据粒子群算法适应度值的变化情况提出一种自适应的惯性权重调整公式，，对粒子群算法进行改进。差异化初始化粒子群算法初始种群，并在优化过程中分别设置权阈值和速度的搜索范围。粒子群算法优化后的BP神经网络车速预测模型训练和预测效果对比优化前有了提高。 ⑤根据遗传算法和粒子群算法对BP神经网络车速预测模型的优化结果，提出PSO-GA联合优化算法模型。其中在城市拥堵和高速公路(1,2,6车速预测)工况，采用遗传算法优化的BP神经网络车速预测子模型；在城市畅通和郊区(3,4,5车速预测)工况，采用粒子群算法优化的BP神经网络车速预测子模型。经过训练和预测，证明PSO-GA联合优化算法优化后的BP神经网络车速预测模型效果效果好于单独采用GA或PSO算法。
[Abstract]:Vehicle speed prediction is one of the important parts of vehicle energy saving and safety research. It can be used in automobile energy saving safety auxiliary driving, route navigation, automatic transmission gear control, and hybrid electric vehicle predictive control. Therefore, the research of vehicle speed prediction has important theoretical basis and application value. The main work of this thesis is as follows: 1 through Kruskal-Wallis single factor variance analysis and correlation analysis, five input characteristic parameters of BP neural network speed prediction model are determined: average speed, idle time ratio, velocity multiple acceleration variance and positive acceleration mean. Through the box diagram distribution of the above five characteristic parameters and its statistical data, the classification boundary of six BP neural network speed prediction submodels with the average speed and idle time ratio as the threshold is determined. These six classification boundaries determine six kinds of speed prediction conditions. (2) based on BP neural network theory, BP neural network speed prediction model with 8 output nodes and 60 output nodes is established. Each speed prediction model includes 6 submodels according to the speed prediction condition. The input part of the sub-model includes the above five characteristic parameters and the historical proximity speed. Based on the training and prediction results of the two models and combined with the statistical analysis of vehicle driving conditions, the input history and the number of output nodes of the BP neural network speed prediction model are modified. A BP neural network model for predicting the speed of the modified input and output nodes is established. The model is superior to the former in training and forecasting. 3According to the difference of the distribution range of weight threshold in the speed prediction model of BP neural network, the initial population of genetic algorithm is initialized differently, and the weight threshold of speed prediction model of BP neural network is optimized by genetic algorithm. The training and prediction effect of BP neural network speed prediction model after genetic algorithm optimization are better than those before optimization. 4 according to the variation of fitness of particle swarm optimization algorithm, an adaptive formula of inertial weight adjustment is put forward, and the particle swarm optimization algorithm is improved. The initial population of particle swarm optimization algorithm is initialized and the search range of weight threshold and speed are set in the optimization process. After particle swarm optimization, the training of BP neural network speed prediction model and the prediction effect are improved. 5 according to the optimization results of BP neural network speed prediction model based on genetic algorithm and particle swarm optimization algorithm, the PSO-GA joint optimization algorithm model is proposed. In the case of urban congestion and expressway speed prediction, the BP neural network speed prediction sub-model optimized by genetic algorithm is used, and the speed prediction is carried out in urban smooth flow and suburban area. The particle swarm optimization (PSO) algorithm is used to optimize the speed prediction submodel of BP neural network. After training and forecasting, it is proved that the effect of BP neural network speed prediction model optimized by PSO-GA combined optimization algorithm is better than that using GA or PSO algorithm alone.
【学位授予单位】：重庆大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP183;U491

【相似文献】