AI 中文总结
针对工业负荷参与电力市场化交互的瓶颈，论文开发了含统一建模、隐私保护参数识别、灵活性区域转换及协同优化框架的方法，显著提升了求解效率、降低了交互成本。
AI 中文摘要
工业负荷占中国用电量的60%以上，为平衡可变电力系统提供了大量灵活性，但复杂的生产约束、不完整的信息以及协调大型组合的计算负担限制了其市场参与度。本论文开发了工业负荷与电力系统市场化交互的建模与优化方法：首先，基于线性化状态任务网络（Linearized State Task Network）和连续资源任务网络的统一公式，可表示离散和连续工业过程以用于电力系统优化，在一个典型炼钢案例中，该方法将求解时间从24小时以上缩短至30分钟以内，同时保持建模精度；其次，一种隐私保护的识别方法结合过程知识与每小时智能电表数据来推断内部生产参数，使用21天的观测数据，其在水泥和钢粉生产上的误差为5.2%-8.5%，较传统机器学习基线方法的误差降低了一半以上；第三，一种数据驱动的方法将高维非凸灵活性区域转换为紧凑的线性表示，对于一个拥有超过10000个二进制变量的炼钢过程，所得模型仅需24-48个连续变量，误差为3.6%-10.3%；最后，一个协同优化框架结合降维投标与精确分解，可在毫秒级内为数万个资源分配电力，同时保持设备级可行性，在典型对比中，其相较于简化策略降低了40%的交互成本。这些方法共同提供了从工业过程建模、参数识别到灵活性聚合及电力系统交互的可行流程。
英文摘要
Industrial loads account for over 60% of China's electricity consumption and can provide substantial flexibility for renewable-dominated power systems. Their market participation remains limited by complex production constraints, unavailable equipment parameters, and the computational scale of resource coordination. This dissertation addresses these barriers from the perspective of a load aggregator. It reformulates the State Task Network and Resource Task Network as the Linearized State Task Network and continuous Resource Task Network. In a standard grid-interaction case, the reformulation reduces solution time from 24 hours to 30 minutes and supports coordinated optimization of 2,000 industrial users. Production Scheduling Identification combines process knowledge and cost-minimizing behavior with hourly smart-meter data; using 21 training days, it achieves load-model errors of 5.2% and 8.5% for cement and steel-powder cases, compared with 13.4%-19.2% for machine-learning baselines. Data-Driven Dimension Reduction yields errors of 3.6%-10.3% across three industrial cases and replaces 10,208 integer variables with 24-48 continuous variables in the steelmaking case. For market interaction, a joint bidding and power-disaggregation framework uses shadow prices to allocate real-time commands among tens of thousands of resources through millisecond-level arithmetic. In a representative comparison, it reduces interaction costs by 40% while retaining the solution quality of the joint optimization model. These contributions keep detailed industrial process models available for executable scheduling while providing aggregators with compact models for flexibility assessment, portfolio optimization, and electricity-market participation.
CommentsPhD thesis, Tsinghua University, June 2026, 175 pages
