Construction and Application of a Traffic Construction Investment Decision‐Making Model Based on the Coupling of Digital Twins and Multi‐Agent
Cheng Guangfeng, Luo Dong · Digital twins and applications. · 2025
ABSTRACT Aiming at the problems of insufficient quantification of dynamic risks and inefficient multi‐objective collaboration in transportation construction investment decisions, this paper proposes a digital twin‐driven multi‐objective dynamic optimisation framework. Firstly, based on the theory of system dynamics, a dynamic modelling mechanism is proposed to analyse the interaction effects of market demand, policy regulation, and risk factors. Combined with the bidirectional LSTM network, real‐time optimisation of dynamic weight parameters is achieved to enhance the model's adaptability. Secondly, a multi‐dimensional coupling mechanism is designed: integrating federated learning and the NSGA‐II algorithm, the collaborative optimisation logic of economic (Monte Carlo risk simulation), efficiency (multi‐agent resource scheduling) and environmental (life cycle ecological assessment) sub‐models is constructed to generate the Pareto frontier solution set to balance multi‐objective conflicts. Finally, a closed‐loop feedback mechanism is introduced to achieve real‐time synchronisation of macroeconomic data, industry dynamics, and project operation information, forming a theoretical closed loop of ‘data‐driven‐ simulation iteration‐decision optimisation’. At the technical level, core algorithms such as bidirectional LSTM cash flow prediction and Latin Hypercube Sampling (LHS) risk quantification are proposed. Validation results using Xi'an Metro Line 9 as a case study show that: the model reduces the Root Mean Square Error (RMSE) of Net Present Value (NPV) prediction from 1.85 billion yuan to 760 million yuan; it decreases the Mean Absolute Error (MAE) of passenger flow prediction by 80.1% (from 2.56 million passenger trips to 510,000 passenger trips); and it compresses the revenue volatility under extreme risk scenarios (passenger flow fluctuation of ± 10%) to ± 5.8%. The research indicates that through dynamic weight adjustment and multi‐objective collaborative optimisation, this framework provides a digital twin solution with both theoretical innovation and engineering practicality for investment decision‐making in complex transportation projects.