计及源荷不确定性的光/氢/储建筑能源系统高效热电协同智能调度

Intelligent Scheduling for Efficient Thermal-Electric Coordination of a Photovoltaic/Hydrogen/Storage Building Energy System Considering Source-Load Uncertainties

光伏耦合燃料电池热电联产技术在低碳建筑能源系统中具有广泛的应用前景。然而,该系统的热电协同调度面临多能流、强耦合及源荷不确定性的问题。为解决该问题,本文建立了新型建筑能源系统的多能耦合模型,构造了以日内综合能耗最优为目标、以电热供需平衡及储能容量边界为约束条件的优化问题。通过在多组随机光伏出力及电热负荷下训练神经网络,本文提出一种改进的深度强化学习(深度确定性策略梯度, DDPG)算法,实现电/热储能设备可充放区间的快速评价及热电协同调度的快速决策。仿真结果表明,本文提出的改进DDPG算法在典型日场景下的训练收敛速度显著提升,并使热电协同调度综合能耗降低10.36%。基于60组具有10%-30%不确定区间的随机源荷场景仿真结果表明,相比于传统DDPG、基于规则的方法和动态规划方法,本文提出的改进DDPG算法可实现近似理论极限的调度结果,提高了系统对不确定性的鲁棒性和适应性。

The cogeneration technology of photovoltaic(PV) coupled fuel cells holds significant potential for widespread application in low-carbon building energy systems. However, the thermalelectric coordinated scheduling of this system faces challenges related to multi-energy flows, strong coupling, and source-load uncertainties. To address these issues, this paper establishes a novel multi-energy coupling model for the system and formulates an optimization problem with the objective of minimizing intraday comprehensive cost, subject to constraints on thermal-electric balances and device storage boundaries. Through training under various sets of random PV and thermalelectric loads, this paper proposes an improved deep reinforcement learning algorithm, specifically deep deterministic policy gradient(DDPG), enabling rapid evaluation of charge/discharge intervals for storage devices and facilitating swift decision-making for scheduling. Simulation results demonstrate that the improved DDPG significantly improves the training convergence speed under a typical winter day scenario, reducing the overall scheduling cost by 10.36 %. Besides, simulation results under 60 uncertain scenarios, with uncertainty intervals ranging from 10% to 30%, indicate that, compared to DDPG, rule-based method, and dynamic programming, the improved DDPG can achieve approximately theoretically optimal results, enhancing robustness and adaptability to uncertainty.