《火灾科学（中英文）》编辑部

基于残差建模的消防水射流落点强度预测深度学习模型研究

A Deep Learning Model for Predicting Impact Intensity of Fire Water Jets With Residual Modeling

投稿时间：2025-06-12 修订日期：2025-10-23

DOI：

基金项目:中国石化科技部项目（325045）液化烃罐区火灾事故消防力量部署决策模型及装备开发

作者	单位	邮编
侯晓静^*	中石化安全工程研究院有限公司	266100

中文关键词: 残差建模消防水射流落点强度双源数据约束驱动模型机器学习组合模型预测精度

英文关键词:Residual modeling Intensity of fire water jet landing point Dual-source data constraint-driven model Machine learning ensemble model Prediction accuracy

摘要点击次数: 160

全文下载次数: 0

中文摘要:

为提升复杂工况下消防水射流落点强度预测精度，提出双源数据约束驱动模型。该模型构建随机森林-机器学习基模型-多层感知机组合，学习实验推导模型预测值与实测值的残差误差，实现对传统公式系统偏差的非线性修正。通过残差建模平衡数据驱动拟合能力与实验推导的物理计算公式约束：随机森林筛选关键特征，支持向量机强化小样本泛化，多层感知机捕捉复杂非线性映射。实验表明，单源数据驱动模型在均方误差（MSE）、平均绝对误差（MAE）和拟合优度（r2）指标上显著优于实验推导模型，而双源数据约束驱动模型进一步优化性能，尤其在高落点强度场景中改善单源数据模型的低估问题，预测值波动更小且符合物理计算公式所表征的规律。该方法在保持计算效率的同时，实现预测精度与可解释性协同提升，为复杂水射流工况提供了优选预测方案。

英文摘要:

To improve the prediction accuracy of fire water jet impact intensity under complex working conditions, a physics-data driven hybrid model is proposed. The model constructs a combination of Random Forest, machine learning base model and Multi-Layer Perceptron (MLP) to learn the residual errors between the predicted values of pure physical models and measured values, and integrates physical constraints such as conservation laws of fluid mechanics to achieve non-linear correction for the systematic bias of traditional formulas. Residual modeling is used to balance the fitting capability of data-driven approaches and physical constraint mechanisms: Random Forest screens key features, Support Vector Machine (SVM) enhances small-sample generalization, and MLP captures complex non-linear mappings. Experiments show that pure data-driven models significantly outperform pure physical models in terms of the metrics of Mean Squared Error (MSE), Mean Absolute Error (MAE), and goodness of fit (r2), while the physics-data driven hybrid model further optimizes performance, especially improving the underestimation problem of pure data models in high-impact intensity scenarios. The predicted values have smaller fluctuations and conform to physical mechanisms. This method achieves synergistic improvement of prediction accuracy and physical interpretability while maintaining computational efficiency, providing a prediction scheme that balances the advantages of data-driven approaches and physical rationality for complex water jet working conditions.

关闭