基于近似动态规划的复杂工业过程数据驱动双网运行优化控制方法研究

Since accurately modeling complex industrial processes and operational index generation trajectories becomes more and more difficult, which pose great challenge on real applications of model-based operational optimization and control methods, then this project investigates an optimization and control method, which does not require knowledge of operational index trajectories and process dynamics. The characteristics existed in complex industrial processes including nonlinearity, exogenous disturbance, multi-time scales, transmission delay and data loss bring challenging problems to model-free operational optimization and control with double-loop networks, this project chooses a double-loop networks and double-layer hierarchical architecture, the multi-objective and multi-constraint operational optimization and control problem is formulated based on multiple dimensions and multi-time scales stochastic process theory; Utilizing the approximate dynamic programming (ADP) method, date-driven approximately optimal setpoint design algorithm is presented by integrating dynamic programming, reinforcement learning, H-infinity control, isomorphic mapping and function approximation methods. Followed by the analysis of the convergence of the proposed algorithm and stability condition; Further, a data-driven approximately optimal tracking controller design algorithm is also derived, and the convergence of the obtained algorithm and stability analysis are performed. Simulations and physical experiments are implemented for wastewater treatment process and grinding process to show the effectiveness of the proposed methods. This project not only proposes theoretical method for data-driven operational optimization and control of complex industrial processes with double networks, but also has a broad prospect in data-driven production, application and promotion of industrial processes.

鉴于复杂工业过程运行指标生成轨迹和被控过程精确建模困难给基于模型的运行优化控制实际应用带来了挑战，本项目研究不依赖工业过程动态模型和运行指标生成轨迹的数据驱动双网运行优化控制方法。复杂工业过程非线性、有界干扰、多时间尺度、传输时延和数据丢失等特征给不依赖模型的双网运行优化控制带来挑战性难题，本项目采用双网双层层级架构，基于多维多时间尺度随机过程理论，阐述多目标多约束复杂工业过程运行优化控制问题；基于近似动态规划，融合动态规划、增强学习、H∞控制、同构映射和函数近似等方法，提出数据驱动的近似最优设定值设计算法及收敛性、稳定性条件；进一步给出数据驱动的近似最优跟踪控制器设计算法及收敛性、稳定性条件。开展上述理论成果在污水处理过程和磨矿过程仿真和物理实验验证。本项目研究不仅为复杂工业过程数据驱动的双网运行优化控制提供理论方法，在工业过程数据驱动方式的生产应用与推广方面也具有广阔的前景。

鉴于传统的基于模型的运行优化控制方法在抗干扰和克服模型不精确的鲁棒性方面还有待于进一步研究，而现有的数据驱动的优化控制方法，还不能很好地解决复杂工业过程的非线性、多时间尺度难题；网络传输时延和数据丢失问题也亟待解决。项目组融合自适应动态规划、博弈、最优控制理论等技术方法，研究在被控系统动态不能精确建模甚至未知的情况下，具有鲁棒性、抗干扰性、网络时延和丢包补偿策略以及多时间尺度动态特性处理的具有创新性的自适应动态规划优化控制方法，提供完整的算法收敛性、系统稳定性和优化条件分析理论；同时应用解决实际复杂工业过程，尤其是磨矿过程的设定值优化、全厂级运行指标优化面临的系统模型不能精确建模、多时间尺度动态、外部干扰、网络通信有限等给工业过程运行优化带来的技术挑战和瓶颈问题；通过仿真和物理实验数据验证，实验结果统计分析，改进提出的理论、方法和技术，提高复杂工业过程能耗、产品产量和质量等性能指标。以闭环方式开展项目研究工作，得到了一套完整的针对复杂工业过程的数据驱动双网运行优化控制理论和方法，将数据驱动运行优化控制理论向更深层面推进。并且在复杂工业过程数据驱动方式的生产应用与推广方面也具有广阔的前景。.研究成果在国际顶级期刊IEEE Transactions on Cybernetics和国内权威期刊《控制理论与应用》等累计发表学术论文36篇，其中SCI收录20篇，EI收录15篇，1篇ESI高被引论文，在国内外权威学术会议发表学术论文10余篇，包括发表在国际权威会议IFAC World Congress 等的学术论文；授权或申请国家发明专利11项；授权软件著作权12项；荣获辽宁省自然科学三等奖1项、辽宁省自然科学学术成果奖学术论文类二等奖1项。研究成果得到国内外知名专家学者在国际顶级期刊Automatica和国内权威期刊《自动化学报》等肯定性评价和引用。超额完成了课题的原定指标。