论文部分内容阅读
为了提高短期交通流预测精度,寻求最优交通流分组策略,通过对短期历史交通流量数据的分析,运用遗传算法优化支持向量回归机的惩罚参数、核函数参数和不敏感损失函数3个参数,构建了GA-SVR模型。首先对采集的数据采用算术平均值进行了降噪处理,然后根据交通数据特征分为连续5个星期五时间、相邻前5个工作日和当天3种时间周期序列,通过不同时间周期序列确定了最优的训练样本集。最后结合采集的数据进行了验证,并且与传统SVR模型进行了精度对比。结果表明:GA-SVR模型预测精度优于传统SVR模型,且基于当天数据构建的训练样本集总体预测精度最高。
In order to improve the prediction accuracy of short-term traffic flow and find the optimal traffic flow grouping strategy, the author analyzes the short-term historical traffic flow data, and uses genetic algorithm to optimize the three parameters of penalty parameter, kernel function parameter and insensitive loss function of support vector regression machine, GA-SVR model is constructed. First of all, the collected data are denoised by the arithmetic mean. Then, the data are divided into five consecutive five-day-time, the first five-day-adjacent and the same-day time-series according to the traffic data characteristics, Optimal training sample set. Finally, the collected data were verified and compared with the traditional SVR model. The results show that the prediction accuracy of GA-SVR model is better than that of traditional SVR model, and the prediction accuracy of training sample set based on the current data is the highest.