数据-知识融合的水利工程建设安全风险灰色因子分解机预测模型  

Data and knowledge-driven Grey Factorization Machine prediction model for safety risk in water conservancy engineering construction

作　　者：张可[1,2] 张政 金伟 ZHANG Ke;ZHANG Zheng;JIN Wei(Business School,Hohai University,Nanjing211100,Jiangsu,China;Institute of Project Management,Hohai University,Nanjing211100,Jiangsu,China;Hangzhou Nanpai Engineering Construction Management Service Center,Hangzhou310000,Zhejiang,China)

机构地区：[1]河海大学商学院,江苏南京211100 [2]河海大学项目管理研究所,江苏南京211100 [3]杭州市南排工程建设管理服务中心,浙江杭州310000

出　　处：《水利水电技术（中英文）》2024年第1期134-143,共10页Water Resources and Hydropower Engineering

基　　金：国家社会科学基金项目(17BGL156);江苏省建设科技项目(521021012);河海大学中央高校基本科研业务费项目(B220207039)。

摘　　要：【目的】已有的数据驱动的水利工程建设安全风险预测方法对领域知识的挖掘和利用不足,预测结果的准确性和可解释性有待进一步提高。为了构建数据-知识融合的水利工程建设安全风险预测模型,【方法】将灰色聚类与因子分解机相结合,提出了一种融合领域知识的灰色因子分解机。首先,引入基于可能度函数的灰色聚类表征水利工程建设领域专家有关安全风险的先验知识。然后,将先验知识以参数的形式嵌入到因子分解机中,构建出数据-知识融合的灰色因子分解机。最后,基于随机梯度下降构造模型参数的求解算法,并结合实例对模型的有效性进行验证。【结果】实例应用结果显示,与传统因子分解机相比,灰色因子分解机的准确率、精确率、召回率和F_(1)值均得到了不同程度的提升。与支持向量机、深度因子分解机等其他基准模型相比,灰色因子分解机同样具有更好的预测性能。【结论】这表明,数据-知识融合驱动的灰色因子分解机模型能够更加准确地预测出安全风险,从而为水利工程建设安全风险管控提供更好的决策支持。[Objective]The existing data-driven safety risk prediction method for water conservancy engineering construction is insufficient in the mining and utilization of domain knowledge,and the accuracy and interpretability of the prediction result need to be further improved.In order to establish a data and knowledge-driven safety risk prediction model for water conservancy engineering construction,[Methods]a domain knowledge enhanced Grey Factorization Machine is proposed by combining Grey Clustering and Factorization Machine.Firstly,Grey Clustering based on Possibility Function is introduced to represent the prior knowledge of safety risks from the experts in the field of water conservancy engineering construction.Then,prior knowledge is incorporated into Factorization Machine model in the form of parameters to construct a data and knowledge-driven Grey Factoriza-tion Machine.Finally,a method for calculating model parameters is provided based on Random Gradient Descent,and the model is applied to a case to verify its effectiveness.[Results]The application result show that compared with traditional Factorization Machine,Grey Factorization Machine′s Accuracy,Precision,Recall and F_(1) Score are improved to varying degrees.Compared with Support Vector Machines,Deep Factorization Machine and other benchmark models,Grey Factorization Machine also has better predictive performance.[Conclusion]It indicates that the data and knowledge-driven Grey Factorization Machine can more accurately predict safety risks,and provide better decision-making support for safety risk management in water conservancy engi-neering construction.

关 键 词：因子分解机 风险交互 领域知识 可能度函数 灰色聚类 影响因素 

分 类 号：TV513[水利工程—水利水电工程]