Abstract: In order to realize effective identification and repair of fault data, a fault data identification algorithm based on outlier distance detection and an improved DE-LSTM data repair model were proposed to address the problem that fixed detectors were prone to abnormal and missing traffic data during the collection of dynamic traffic data. By utilizing the inherent continuity of time series data, effective identification of fault data was achieved through direct outlier localization and outlier distance detection. The differential evolution algorithm was used to optimize the number of hidden layer neurons and the initial learning rate of long short-term memory neural network, and the adaptive control strategy was introduced to improve the mutation factor and crossover factor in the traditional DE algorithm. The repair model of long short-term memory neural network based on the improved differential evolution algorithm was established and compared with the fixed threshold combined with traffic flow mechanism, LSTM neural network model and DE-LSTM repair model. The example verification results indicate that compared with the fixed threshold combined with traffic flow mechanism method, the outlier distance detection algorithm has a more efficient recognition rate, and the improved DE-LSTM model has good computational efficiency and repair performance.

Key words: traffic engineering; fixed detector; dynamic traffic data; fault data identification; data repair; optimization algorithm

摘要： 针对固定检测器在采集动态交通数据过程中易发生交通数据异常、数据缺失等问题，为实现故障数据有效识别及修复，提出了基于离群距离检测的故障数据识别算法及改进的DE-LSTM数据修复模型。利用时序数据的自身连续性，采用直接离群点定位和离群距离检测对故障数据进行有效识别。采用差分进化算法优化长短期记忆神经网络的隐含层神经元个数和初始学习率，并引入自适应控制策略改进传统DE算法中的变异因子、交叉因子，建立了基于改进差分进化算法优化长短期记忆神经网络的修复模型，并与固定阈值结合交通流机理、LSTM神经网络模型及DE-LSTM修复模型进行对比。实例验证结果表明：与固定阈值结合交通流机理法相比，离群距离检测算法识别率更为高效，改进的DE-LSTM模型具有良好的计算效率及修复性能。

关键词: 交通工程；固定检测器；动态交通数据；故障数据识别；数据修复；优化算法