Integrated hybrid energy and time-of-use electricity tariffs for the resource-constrained project scheduling problem
将混合能源、分时电价和多班次调度整合到资源受限项目调度框架中，通过深度强化学习算法优化能源成本与碳排放，为项目管理者提供理论工具。
With growing environmental awareness and rising energy costs, projects increasingly face significant challenges in achieving energy efficiency and reducing carbon emissions. To address these issues, this paper integrates hybrid energy, time-of-use (TOU) tariffs and multi-shift scheduling into the resource-constrained project scheduling (RCPSP-HE-TOU-MS) framework. A mixed-integer model is formulated to reduce the energy cost and the carbon emissions. During project execution, the model optimizes the proportion of different energy sources and schedules high-energy-consumption tasks during low-electricity-price periods to achieve its objective. Considering the characteristics of the problem, the study introduces a preference-driven multi-objective deep reinforcement learning algorithm that combines exact optimization techniques with an improved multi-step prioritized experience replay dueling double deep Q-network to derive optimal scheduling strategies from training instances. By incorporating spatial pyramid pooling, cosine similarity evaluations and decomposing problems into scalar subproblems, the algorithm enhances adaptability, stability and training efficiency in multi-objective deep reinforcement learning. This model equips project managers with a robust theoretical tool to effectively manage energy consumption and to allocate energy resources rationally, thereby reducing project costs and carbon emissions.