四川轻化工大学学报：自然科学版

针对现有的冷缝检测手段存在检测效率较低、对数据解析要求高、波形信息不能进一步深入挖掘等问题，提出了基于核主成分分析（Kernel Principal Component Analysis,KPCA）、改进海鸥优化算法（ImproveSeagulloptimizationalgorithm,ISOA）、核极限学习机（Kernel Extremel Learing Machine,KELM）相结合的冷缝检测分类方法（KPCA-ISOA-KELM）。首先，采用冲击弹性波法在隧道疑似冷缝区域中采集数据构成数据集，并利用KPCA对数据进行降维操作；然后，利用ISOA对KELM的参数进行优化；最后，利用含最优参数的KELM实现对检测数据分类。通过与KPCA-SOA-KELM、KPCA-FOA-KELM、KPCA-CNN-LSTM进行对比验证，结果表明，KPCA-ISOA-KELM模型对现场检测具有一定的指导意义，并且相较于其他3种模型，该模型能达到96.00%的准确率和0.9598的加权F1-score。

The cold joint detection classification method, named KPCA-ISOA-KELM, is proposed to address several existing issues in current cold joint detection techniques, such as low detection efficiency, high data parsing requirements, and limited exploitation of waveform information, which combines Kernel Principal Component Analysis(KPCA), Improved Seagull Optimization Algorithm(ISOA), and Kernel Extreme Learning Machine(KELM). Firstly, data sets are collected using the impact elastic wave method in suspected cold joint areas within tunnels, and KPCA is utilized to perform dimensionality reduction on the data. Then, ISOA is employed to optimize the parameters of KELM. Finally, the optimized KELM classifier is utilized to classify the detection data. Comparative validation is conducted against KPCA-SOA-KELM, KPCA-FOA-KELM, and KPCA-CNN-LSTM models. The results indicate that the KPCA-ISOA-KELM model provides valuable guidance for on-site detection and achieves a higher accuracy rate of 96.00% and a weighted F1-score of 0.9598 compared to the other three models.