基于Kmeans-EMD与IWOA-Elman的碾压速度异常值检测与修正

碾压速度是评价压实质量的重要指标，但在监控过程中，碾压速度易受施工环境、定位漂移等干扰而出现异常检测值，影响压实质量的评价精度，但目前还缺乏对碾压速度异常值检测与修正的相关方法研究。为保障碾压速度的数据质量，结合碾压速度的时序变化特征，利用Kmeans算法初步定性检测异常值，弱化异常值对经验模态分解(EMD)结果的影响，并基于EMD实现对异常值的精细定量检测，提高异常值检测的精度；进而利用经混沌种群初始化、非线性收敛因子、自适应惯性权重与鲶鱼效应-黄金正弦改进的鲸鱼优化算法(IWOA)优化Elman神经网络，并构建碾压速度异常值修正模型，实现对碾压速度异常值的修正。将本文方法应用于西南某大型水电工程，结果表明：Kmeans算法与EMD的联合作用相比箱线图法可更高精度地检测碾压速度中的异常值；IWOA-Elman神经网络预测值与真实值的相关系数达到0.907 75，相比常规模型不仅可以更好地确保数据的完整性与可靠性，还可以为压实质量的高精度评价奠定良好的数据基础。

Outlier detection and correction for rolling speed based on Kmeans-EMD and IWOA-Elman

Rolling speed is an important indicator for evaluating compaction quality, but in the monitoring process, the rolling speed is easily disturbed by construction environment, positioning drift and other disturbances, resulting in outliers which will affect the accuracy of compaction quality evaluation. However, methods for detecting and correcting outliers of rolling speed are not reported in current research. To ensure the data quality of the rolling speed, based on the characteristics of the time series of rolling speed, we adopted Kmeans algorithm to detect outliers preliminarily and weaken the influence of outliers on the results of empirical mode decomposition (EMD). According to the results of EMD, the fine quantitative detection of outliers was achieved, which in turn improved the accuracy of outlier detection. Furthermore, whale optimization algorithm (WOA) improved by chaos population initialization, nonlinear convergence factors, adaptive inertial weights and catfish effect-golden sine was used to optimize the Elman neural network and then a correction model was established for the correction of outliers of rolling speed. According to the application of the proposed method to a large hydropower project in Southwest China, the combined effect of Kmeans algorithm and EMD can detect outliers of rolling speed with higher accuracy than the box plot method; the correlation coefficient between the predicted value of the IWOA-Elman neural network and the actual value can reach up to 0.907 75, compared to conventional models, the IWOA-Elman neural network can not only ensure better data integrity and reliability, but also lay a good data foundation for the high-precision evaluation of compaction quality.