Grid-Compatible Flexibility from Multi-Energy Systems via Cyclic-Terminal Economic MPC with Hybrid Thermal-Electrical Dynamics
作者: Azzam Abdul, Schwenkel Lukas, Scheurer Leon, Häbig Pascal, Hufendiek Kai
分类: eess.SY
发布日期: 2026-08-07
💡 一句话要点
提出统一经济模型预测控制框架以优化多能源系统的协调运行
🎯 匹配领域: 支柱一：机器人控制 (Robot Control)
关键词: 经济模型预测控制 多能源系统 热电协调 实时优化 智能电网
📋 核心要点
- 现有的电热基础设施控制方法在响应市场价格和实时计算速度上存在不足，难以满足动态需求。
- 本文提出了一种统一的经济模型预测控制框架，结合混合热电动态和时变经济信号，以优化多能源系统的协调运行。
- 实验结果表明，采用该框架后，闭环成本在多个时间周期内显著降低，且对终端权重的敏感性减弱，提升了系统的经济性。
📝 摘要（中文）
现有的电热基础设施需要能够快速响应市场价格的控制器。本文提出了一种统一的经济模型预测控制（EMPC）框架，用于协调集成的热电能源网络的运行。该方法基于循环终端EMPC，结合了混合热电动态、网络约束和时变经济信号，优化了联合热电单元、大规模热泵、热能存储、电池和电网交互。通过简化的区域供热网络和直流电力流网的降阶模型，确保了计算的可行性。该框架在一个校园规模的能源系统上进行了验证，结果显示预测视野与终端惩罚权重之间的联合调整显著影响了闭环成本。
🔬 方法详解
问题定义：本文旨在解决电热基础设施控制器在市场价格响应和实时计算速度上的不足，现有方法难以有效协调多种能源的运行。
核心思路：提出的统一经济模型预测控制框架通过结合混合热电动态、网络约束和时变经济信号，优化了热电单元和储能设备的协调运行。
技术框架：该框架包括多个模块：首先是混合动态模型的构建，其次是经济目标的定义，最后是基于降阶模型的实时优化计算。
关键创新：最重要的创新在于将循环终端EMPC与混合热电动态相结合，形成了一个单一的混合整数状态空间表示，显著提升了计算效率和系统协调性。
关键设计：在设计中，采用了降阶模型以提高计算效率，设置了适当的终端惩罚权重以优化闭环成本，并通过实验验证了不同权重对系统性能的影响。
🖼️ 关键图片
📊 实验亮点
实验结果显示，在校园规模的能源系统中，采用该框架后，闭环成本在多个周期内显著接近理论性能界限，且在不同的时间段内表现出良好的稳定性和经济性，验证了方法的有效性。
🎯 应用场景
该研究具有广泛的应用潜力，特别是在智能电网、可再生能源集成和城市能源管理等领域。通过优化多能源系统的协调运行，可以有效降低能源成本，提高系统的经济性和可靠性，推动可持续发展。
📄 摘要（原文）
Coupled electrical and thermal infrastructures need controllers that respond to market prices and still solve fast enough to run online. This paper presents a unified Economic Model Predictive Control (EMPC) framework for the coordinated operation of integrated thermal and electrical energy networks. Building on cyclic-terminal EMPC, the proposed approach incorporates hybrid thermal-electrical dynamics, network constraints, and time-varying economic signals within a single mixed-integer state-space representation, jointly optimizing combined heat and power units, large-scale heat pumps, thermal energy storage, batteries, and grid interactions under a convex economic stage cost. Computational tractability is ensured by reduced-order models of district heating networks and DC power flow grids. The framework is demonstrated on a campus-scale energy system under time-varying prices and demand profiles. A joint sweep of the prediction horizon against the terminal penalty weight shows that the two act as substitutes rather than as independent tuning knobs. Without terminal anchoring, the closed-loop cost approaches the periodic-reference average-performance bound only once the horizon spans several diurnal cycles. With a sufficiently large terminal weight, the bound is attained essentially tightly at every tested horizon, including the shortest one, so the horizon ceases to be a performance-critical parameter and becomes a purely computational one. The result reproduces on a second, independent price week. Beyond the weight at which the soft terminal constraint activates, closed-loop behavior is insensitive to the weight over a wide multi-decade plateau; below activation, cost and storage tracking both degrade markedly. A residual receding-horizon drift of the cost-neutral thermal-storage state is also documented and interpreted.