Adaptive control of resource flow to optimize construction work and cash flow via online deep reinforcement learning

Due to complexity and dynamics of construction work, resource, cash flows, poor management them usually leads time cost overruns, bankruptcy, even project failure. Existing approaches in failed achieve optimal control resource flow a dynamic environment with uncertainty. Therefore, this paper introducess model method adaptive the flows optimize work projects. First, mathematical based on partially observable Markov decision process is established formulate complex interactions as well uncertainty variability diverse influence factors. Meanwhile, efficiently find solutions, deep reinforcement learning (DRL) introduced realize continuous labor material thereby optimizing flows. To assist training DRL, simulator discrete event simulation also developed mimic features external environments project. Experiments simulated scenarios illustrate that our outperforms vanilla empirical genetic algorithm, possesses remarkable capability projects environments, hybrid agent DRL best result. This contributes optimization coupled may serve step stone for adopting technology management.