基于 LSTM 模型的直接空冷机组多目标分布式最优控制研究

With the problems of high energy consumption, low cooling performance and frequent mechanical and electrical faults of axial-flow fans at the cold-end of air-cooled units, it is of great significance to study the optimal speed solution in each zone of the fans group under multiperformance objectives, so as to improve the energy-saving operation level of the cold-end of air-cooled units and reduce the energy consumption of thermal power plants. Taking 1×8 air-cooled units of 660MW unit as object, the aerodynamic field of fans group was numerically solved by using computational fluid dynamics under step and random input. Based on the dynamic characteristics of the fans group under step disturbance, the cluster analysis of the fans group was carried out, and the distribution scheme was formulated. A nonparametric dynamic response model of distributed driven fans under the double random disturbances of natural wind and speed command was constructed with the long short-term memory neural network. Considering the performance indicators such as inlet air flow of fans, motor energy consumption and mechanical and electrical loss, a non-dominated sorting genetic algorithm with the elite strategy was proposed, and the Perato optimal solution of fan speed command in each zone was calculated through fuzzy membership function. Under the random disturbance of natural wind, the simulation results show that the multi-objective distributed optimal control can significantly improve the operational economic performance of the cold-end, meanwhile reduce the mechanical and electrical losses as comparing to the existing fan speed centralized regulation strategy. The research can provide the theoretical basis and engineering references for the energy-saving and safe operation of direct air-cooled units.