An Intelligent Quality Evaluation Method for TRD Wall Construction Based on Real-Time Monitoring and Probabilistic Prediction Models

<p dir="ltr">Conventional quality assessment methods for Trench-cutting Re-mixing Deep (TRD) walls suffer from delayed feedback and limited sampling coverage. Existing research predominantly focuses on deterministic prediction, without adequately capturing the uncertainties introduced by fluctuating construction conditions and sparse sampling, which restricts real-time quality control and comprehensive performance evaluation. To address these limitations, this study proposes an intelligent assessment framework based on multi-source data fusion and probabilistic learning. A real-time monitoring system and a three-dimensional geological model are first established to continuously gather construction energy parameters, stabilizer characteristics, and stratigraphic information. Building on these data, an MIC-BO-NGBoost model is developed by integrating the Maximal Information Coefficient (MIC), Bayesian Optimization (BO), and Natural Gradient Boosting (NGBoost). The model provides probabilistic predictions, generating both probability distributions and confidence intervals for compressive strength and permeability coefficient. These outputs enable dynamic and full-domain classification of TRD wall quality. Case study results verify that the MIC-BO-NGBoost model achieves excellent predictive performance (R² = 0.953), demonstrating that the proposed method supports intelligent, robust, and comprehensive quality evaluation of TRD wall construction.</p>