Accurate construction duration prediction is critical for enhancing overall project efficiency in hydraulic engineering, as projects frequently encounter challenges such as overestimated timelines and delays during implementation. This study aims to develop an innovative prediction model by integrating diverse machine learning algorithms while considering multiple influencing factors, thereby improving the accuracy of construction duration prediction.
This study first employs the Random Forest (RF) algorithm to identify critical factors affecting construction timelines in water conservancy projects. Subsequently, a hybrid prediction model integrating Convolutional Neural Network (CNN) and Support Vector Regression (SVR) is proposed, with its hyperparameters optimized through the Northern Goshawk Optimization (NGO) algorithm. Comparative analysis with other prediction models using standard evaluation metrics confirms the enhanced accuracy and generalizability of this hybrid approach.
The research results demonstrate that RF-based feature selection provides an importance ranking of influencing factors, with total feature importance exceeding 97.43%, confirming the effectiveness of the selection. When comparing prediction results across different models in channel engineering, bridge-culvert engineering, and intake gate engineering, the NGO-CNN-SVR model achieved the smallest mean absolute error (MAE) and root mean square error (RMSE), along with the largest coefficient of determination (R2), yielding minimal prediction errors for hydraulic engineering construction duration. This study verifies that integrating feature selection, model fusion, and algorithm optimization can effectively enhance prediction accuracy.
With the rapid development of artificial intelligence, hydraulic engineering management should move toward digitalization and intelligence. This model enables rapid construction duration prediction in conventional server environments, adapts to the computing power conditions of small and medium-sized construction enterprises, and provides scientific theoretical support and practical application value for hydraulic engineering construction progress management.
Theoretically, the combined prediction model proposed in this study can better fit the nonlinear variation patterns of construction duration in hydraulic engineering and provides more reliable predictions. Practically, this model enables prediction of construction duration under varying factor conditions, allowing construction managers to promptly adjust project schedules while balancing quality and cost requirements, thereby enhancing the scientific rigor and systematic approach in construction management.
