【摘要】 阈值范围的合理设定,对AFC(自动售检票)系统实时进站客流量数据的有效性检测至关重要。采用改进的小数据量法计算历史进站客流数据时间序列的Lyapunov指数,验证该序列的混沌特性;利用C_C方法确定混沌时间序列的时间延迟和最佳嵌入维数,对原时间序列进行相空间重构,确定模型训练测试样本集;采用大范围网格搜索方法优化混沌支持向量机回归模型参数,利用优化后的模型预测各时段的进站客流量;利用训练样本集中各时段进站客流量预测残差序列的分布特性,确定在某一置信度下各时段进站客流量预测残差的置信区间,从而确定未来时段的进站客流量异常检测的阈值上限和阈值下限。试验结果表明,该方法有效地增强了系统对于AFC系统实时进站客流量数据有效性检测的能力。更多还原

【Abstract】 The reasonable setting of threshold range is very important for the effective data of real-time passenger flow entrance detected by AFC system. In this paper,an improved small data method is used to calculate the Lyapunov exponent of historical time series data,and test the chaotic properties of the series.Then,C_C method is used to determine the time delay and the best embedding dimension of the chaotic time series,reconstruct the original time series by phase space reconstruction and determine the model training test sample set. A large range of grid search is employed to optimize the chaos-support vector regression model parameters,and the optimized model is used to forecast the passenger flow in different periods. The collected passenger volume at different time intervals in the training set is used to predict the residuals distribution,and determine the residues confidence interval of passenger volume prediction residues in each time period. This step is done at a certain confidence level so as to determine the threshold values of entrance passenger flow anomaly detection. The experimental result proves that this method can effectively enhance the system ability to detect the real-time entrance passenger flow data of the AFC system.更多还原