基于智能闭环建模的燃煤电厂新一代自适应控制与优化

The power grids in south east coast area of China have high level of super-high voltage electricity usage; encounter rapid increase of new energy sources; use coals from a large variation of coal mines. The coal fired electrical power plants in this area face many challenges: low load level operation; large load range operation; frequent load change and the need of tight frequency control; and coal type variations. This project studies adaptive model predictive control (MPC) and optimization of coal fired power plants based on intelligent closed-loop modeling. Linear parameter varying (LPV) model will be used to describe the nonlinear behavior caused by large operation range of the power plants; online identification will be used to cope with model variations caused by the coal type changes where identification test and model adaptation will be event triggered and not continuous, globally convergent estimation method and model matching method will be used. In MPC algorithm development, the property of the LPV model is utilized in prediction calculation and control optimization in order to achieve high precision and performance where multi-step dynamic prediction, closed-loop steady state target calculation are used. The adaptive LPV MPC outperforms linear MPC and have high performance for large load range and large coal type variations. In economical optimization, in order to avoid the high cost of traditional combustion optimization, it is proposed to use the real load as the objective function and use multi-variable system identification to calculate the gradients for hill-climbing optimization; moreover, the concept of optimal variance distribution is developed for the first time. For demonstrating the effectiveness and applicability of the project, industrial case study will be carried out for a ultra-super critical power generation unit in Zhejiang Province.

我国东南沿海地区电网具有特高压受电量大、新能源发电增速快、煤炭来源不稳定的特点，导致燃煤电厂负荷率低、调峰调频压力大、煤种切换频繁、环保形势严峻。本项目以智能闭环建模为核心，研究燃煤发电机组的自适应模型预测控制和运行优化。提出自适应闭环线性变参数（LPV）模型辨识的策略，使用触发式激励测试方法、全局收敛递推辨识算法、煤种变化识别与快速模型匹配方法，适应机组大范围变负荷引起的非线性特性和煤种变化引起的时变性。针对LPV模型的多模型组合结构特点，提出新一代自适应预测控制算法和闭环性能分析方法，采用多步非线性动态预测、闭环稳态目标值计算，突破现有线性预测控制的局限，实现机组深度调峰及煤种切换工况下的高性能控制。提出面向机组全局目标、基于系统辨识模型的多变量运行优化方法，克服传统热力试验成本高、无法在线优化的难题。将在浙江省某超超临界百万千瓦机组进行应用示范，以验证所提技术和方法的有效性。

经过4年的研究工作，本项目取得了丰富的成果，在电厂实现了部分成果的示范应用，并培养了人才。在闭环系统辨识建模方面，提出了多变量MPC闭环无激励辨识方法；对于辨识实验，发现白噪声测试信号优于低通信号; 提出了基于系统辨识的故障诊断方法。在模型预测控制方面，产生了多项成果，包括LPV模型的输出反馈鲁棒MPC的综合方法、积分对象的双层MPC控制、离线输出反馈鲁棒MPC控制、无偏跟踪MPC控制和3项QP的MPC控制。在运行优化方面，提出并验证了基于系统辨识的实时优化方法。在示范应用方面，在内蒙古盛乐电厂两台350MW超临界发电机组实现了AGC协调控制系统和汽温控制的MPC应用，取得很好的效果。本项目工作成果在著名学术期刊发表论文23篇，出版专著一部，申请发明专利5项，培养博士4名，博士后1名，硕士21名。主要参与者丁宝苍教授于2020年获得中国自动化学会自然科学二等奖。本项目在三个方面产生技术领先且实用性强的成果，应用领域除了电力工业，还包括其他流程工业。由于新冠疫情和项目申请工作进展缓慢，在浙江六横电厂1030MW发电机组的示范应用项目尚未结束，正在加速进行中。