Dynamic Cost Estimation and Optimization Strategy in Engineering Cost Combining Reinforcement Learning

Abstract Accurate cost estimation and optimization are crucial in engineering project management, as budget overruns and resource misallocations often lead to financial and operational inefficiencies. Traditional cost estimation methods, including regression models and heuristic approaches, struggle to adapt to the complex and dynamic nature of engineering projects. We proposes a reinforcement learning (RL)-based dynamic cost estimation and optimization strategy that continuously refines cost predictions and budget allocations. The proposed framework integrates a deep learning-based cost estimation model with an RL-driven optimization strategy, enabling adaptive learning from historical and ongoing project data. A multi-objective optimization framework is incorporated to balance cost, project quality, and timeline constraints using Pareto-front analysis. The RL agent learns optimal cost allocation policies through iterative interactions with the environment, improving decision-making efficiency. Experimental evaluations demonstrate that the RL-based model outperforms conventional machine learning approaches, achieving lower mean absolute error and root mean square error in cost estimation. Additionally, the RL-driven optimization strategy results in an average cost reduction of approximately 7% across different project categories. The integration of multi-objective reinforcement learning further enhances cost efficiency while maintaining project feasibility. These findings validate the proposed approach as an effective solution for improving cost estimation accuracy and optimization in engineering project management.