Fusing multi-source quality statistical data for construction risk assessment and warning based on deep learning

Abstract

In the context where accidents and fatalities in the construction industry remain persistently high, the assessment and early warning of construction risks become critically imperative. However, existing research has not fully leveraged the multi-source construction quality statistical data available, leading to a hindrance in formulating data-driven effective risk warning strategies. Considering the urgent need for optimized risk identification and assessment methods, this study introduced a novel risk assessment and warning strategy for comprehensive processing of multi-source construction practical quality monitoring statistical data based on cloud model-optimized Dempster-Shafer (D-S) evidence theory. This method employed 4M1E for monitoring data categorization and integrated a deep learning prediction model based on Long Short-Term Memory (LSTM) for forecasting potential quality monitoring data. Subsequently, the combination of the cloud model and d-S evidence theory facilitated the transformation of LSTM-predicted multi-source quality data between quantitative and qualitative, enabling comprehensive risk assessment and warning considerations. The findings revealed that the prediction results and the actual data were both at the highest level, indicating a high likelihood of a collapse incident. Lastly, the developed BIM-integrated platform offered a dynamic approach to risk assessment and real-time control, serving as a potential reference for warning risks in construction. This study is a pioneering study to assess and predict risk by cohesively integrating various methods and realizing a practical BIM platform. This study not only provides new measures to mitigate risks in construction but also advocates for the wider integration of multi-source quality statistical data into risk management.