在日益多元化、集成化的热力系统中，由于多能流耦合和强非线性等因素的制约，传统以物质流为核心的分析方法以及线性简化策略均已无法有效处理变工况条件下全局优化过程中准确性、时效性和鲁棒性间此消彼长的矛盾，限制了总体能源利用效率的进一步提升。本文引入热阻、热动势和热量源等标准元件构建热力系统的热量流模型，进而结合回路通流阻力分析与工质物性分析，提出了规范化的系统整体模型构建框架，从而说明了描述工质焓流的桑基图仅反映局部能量守恒关系，无法替代热量流模型刻画热量传递与转换的规律性。同时，阐释了充灌量通过密度约束和热力学状态约束影响系统性能，推导得滑压曲线为充灌量一定而引起的表观约束。针对热量流模型分离出线性拓扑约束和显式非线性元件约束，相应提出了普适的分层-分治求解算法：通过直接代入求解显式非线性元件约束，通过矩阵运算求解线性系统拓扑约束，从而显著减少需迭代求解的非线性隐式约束的数量。相比于传统局部拟线性模型的整体迭代求解及其线性系数的摄动更新，有效缩减了非线性空间的收敛维度。对比结果表明，既有工程运行经验及固定端差等线性表征简化均易偏离系统真实工况，最大计算误差达8.6%。以燃气-蒸汽联合循环发电系统为对象，进行了规范化建模框架和分层-分治求解算法的应用研究，并从准确性、迭代嵌套层数、时间复杂度、初值个数及初值偏差极限等角度同商业软件、传统分析策略进行定量对比分析，验证了其在满足计算鲁棒性需求下，可在合理的计算时长内解决大规模热力系统的运行优化问题，并在此基础上澄清了热量流模型与分层-分治算法的理论依据。针对实际燃气-蒸汽联合循环冷、热、汽、电多能联供系统，结合深度学习、知识库等信息驱动技术，开发并搭建机组实时优化平台，通过数据采集、元件特征参数辨识、运行优化及结果发布等模块，辅助运行人员通过DCS进行优化工况调整与控制。试验结果表明，发电标准煤耗降低0.415 g/kWh，阐明了实时优化平台提升机组变工况效率的有效性。
Thermodynamic systems are becoming more and more diversified and integrated, and their constraints are nonlinear and highly coupled. Therefore, traditional analysis approaches based on the fluid flow perspective together with linear simplification strategies cannot effectively deal with the contradiction between the accuracy, rapidity and robustness of global optimization, which limits further improvement of system efficiency.Introducing standard devices such as thermal resistances, thermo-motive forces and heat sources, the heat current model of thermodynamic systems is built in this thesis. Furthermore, combining with flow resistance analysis in closed loops and physical properties of fluids, a standardized modeling framework for thermodynamic systems is proposed. On this basis, it is revealed that the Sankey diagram, which describes the enthalpy flow carried by the working fluid, can only reflect the local energy conservation relationship but cannot replace the heat current model to describe the global heat transfer and conversion laws. Moreover, it is clarified that the fluid charge affects the system performance through density constraint and thermodynamic state constraint, and the sliding pressure curve is deduced as the apparent constraint caused by a certain fluid charge.Linear topology constraints and explicit nonlinear component constraints can be separated in heat current models, and a universal hierarchical and categorized algorithm is accordingly proposed. The algorithm solves explicit nonlinear component constraints by direct substitution, deals linear topology constraints with matrix calculation, and hence implicit nonlinear constraints are minimized. Compared with the global iterative solution of traditional local quasi-linear model and the update of its linear coefficients by perturbation method, the convergence dimension of the nonlinear space is effectively reduced. Comparison results show that existing engineering operation experiences and linear simplification such as fixed terminal temperature difference can easily deviate from the actual condition under off-design conditions, with a maximum deviation of 8.6%.The proposed standardized modeling framework and hierarchical and categorized algorithm are applied on a gas-steam combined cycle power generation system. The quantitative comparison is performed with commercial software and traditional analysis strategy from the aspects of accuracy, number of nested iteration levels, time complexity, number of initial values required, and deviation limits of initial values. It is verified that the proposed method can meet the requirements of computational robustness and could solve the global operation optimization for large-scale thermodynamic systems in a reasonable calculation time. Based on this comparison, the theoretical basis of heat current model and hierarchical and categorized algorithm is clarified.For an actual gas-steam combined cycle cogeneration system of cooling, heating, steam and power, combined with deep learning, knowledge base and other information-driven technologies, a real-time optimization platform involving data acquisition, characteristic parameter identification, operational optimization and result release modules is developed to assist operators to adjust and control the system through DCS for its performance optimization. Experimental results show that the standard coal consumption for power generation can be reduced by 0.415 g/kWh, which illustrates the effectiveness of the optimization platform to improve the off-design efficiency of the unit.
