| 王晓玲,郭章潮,余佳,余红玲,刘长欣,吴斌平.地下厂房洞室群施工通风频率IHPO-XDF鲁棒预测模型[J].水利学报,2025,56(8):1072-1083 |
| 地下厂房洞室群施工通风频率IHPO-XDF鲁棒预测模型 |
| IHPO-XDF robust prediction model for ventilation frequency of underground powerhouse caverns |
| 投稿时间：2024-05-07 修订日期：2025-07-18 |
| DOI：10.13243/j.cnki.slxb.20240262 |
| 中文关键词: 地下厂房洞室群 通风频率鲁棒预测 深度森林模型 XGBoost 改进的猎人猎物算法 可解释性 |
| 英文关键词: underground powerhouse cavern group robust prediction of ventilation frequency deep forest model XGBoost improved hunter-prey algorithm interpretability |
| 基金项目:国家自然科学基金项目（52279137） |
|
| 摘要点击次数: 17 |
| 全文下载次数: 19 |
| 中文摘要: |
| 水电站地下厂房洞室群洞室布置纵横交错、通风死角多、风流组织素乱，确定合适的施工通风频率是保障通风安全的关键。但其施工过程中存在的电磁干扰和爆破振动常导致环境监测数据出现噪声与缺失现象，而现有基于机器学习的施工通风频率预测模型对异常值十分敏感，模型鲁棒性差。针对上述问题，选择深度森林（DF）模型作为通风频率预测的基础模型，并将其中的随机森林基学习器改进为极致梯度提升树（XGBoost），利用XGboost的梯度提升机制以及正则化策略增强模型的鲁棒性和泛化能力；此外，采用改进的猎人猎物优化（IHPO）算法对DF模型进行超参数优化，以弥补传统人工调参难以获得最优超参数的不足，从而构建出地下厂房洞室群施工通风频率IHPO-XDF鲁棒预测模型。进一步，基于Shapley加性解释（SHAP）对IHPO-XDF模型进行可解释性分析，挖掘影响施工通风频率预测结果的关键特征。案例研究表明，与XGBoost改进的DF模型、传统DF、梯度提升决策树（GBDT)和决策树（DT)4种模型相比，本文模型在预测精度方面分别提升3.48%、5.01%、13.13%和13.48%，且在异常值环境下预测精度降低幅度最小，表现出良好的鲁棒性。 |
| 英文摘要: |
| In the construction of hydropower underground power plant cavern complex, the arrangement of the caverns is complex, with many dead ends and disordered airflow patterns. Determining an appropriate construction ventilation frequency is crucial for ensuring safety. During the construction process of the underground power plant cavern complex, electromagnetic interference and blast vibration often result in noisy and missing data in environmental monitoring, and existing machine learning-based construction ventilation frequency prediction models are highly sensitive to outliers, with poor robustness. To address these issues, this paper proposes an IHPO-XDF robust prediction model for construction ventilation frequency in underground power plant cavern complexes. First, a deep forest (DF) network with strong universal learning ability and robustness is selected as the basis for ventilation frequency prediction. Second, the random forest base learner in the deep forest is improved to extreme gradient boosting tree (XGBoost), and the efficient gradient boosting mechanism and regularization strategy of XGboost are used to further enhance the robustness and generalization ability of the model. Furthermore, the Improved HunterPrey Optimization (IHPO) algorithm was adopted to optimize the hyperparameters of the DF model, which could compensate for the shortcomings of traditional manual parameter tuning and obtain the optimal hyperparameters, thereby mining and enhancing the inherent robustness of the model and revealing the complex nonlinear mapping relationships between the pollutant concentration, temperature, wind speed, and excavation volume and the construction ventilation frequency. Furthermore, the IHPO-XDF model was analyzed using the SHapley Additive exPlanations (SHAP) for interpretability, which could mine the key features that affected the prediction results of the construction ventilation frequency and enhance the credibility of the ventilation frequency prediction model. The case study showed that compared with the XGBoost-improved DF model, the traditional DF, the Gradient Boosting Decision Tree (GBDT), and the Decision Tree (DT) models, the proposed model achieved a respective improvement of $3.48 \%, 5.01 \%, 13.13 \%$, and $13.48 \%$ in prediction accuracy, and the IHPO-XDF model showed the smallest decrease in prediction accuracy in the abnormal value environment, at only $2.17 \%$, demonstrating good robustness. Meanwhile, the SHAP-based interpretability analysis of the prediction results indicated that the IHPOXDF algorithm had high credibility. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|