Study And Application Of Neural Network Based Mathematical Model For Ultra-Supercritical Power Unit

Since an ultra-supercritical(USC) power unit is with higher efficiency and favorable to energy saving and environmental protection, it has increasingly become the main unit in China's power industry. From the control point of view, an USC boiler turbine unit is a multi-input multi-output object with nonlinear and strong coupling characteristics.An USC power unit in operation must maintain the supply-demand balance both inside and outside the unit itself to meet the power grid load demand quickly and ensure the security and stability of the running unit depending on the coordinated control system(CCS). The widespread implementation of the regional power grid centered automatic generation control(AGC) proposes higher demand on the capability of deep peaking-load regulation for a large-scale USC power unit. It is of great significance to improve the coordinated control quality with advanced intelligent model based control strategies and methods.In this paper, a 1000 MW ultra-supercritical boiler turbine power unit is taken as the object investigated and its coordinated control modes and coordinated control logic are analysed. In deep understanding of artificial neural network theory and nonlinear system neural network modeling methods, neural network based prediction models for the load and main steam pressure are established by considering its regenerative cycle system. It is shown by online verification that the model has high accuracy and good dynamic characteristics. On this basis, by combining the "condensate throttling" technology proposed in recent years, which accelerates the unit load response speed by rational use of energy storage in the unit's regenerative system, a new integrated coordinate system intelligent predictive optimal control method is proposed. The method utilizes the load and the main steam pressure prediction models respectively to optimize the deaerator water level control valve opening, the steam turbine control valve opening and the total fuel demand to improve the coordinated control effect. Real-time control algorithm is developed with MATLAB platform, and detailed coordination optimal control simulation tests are carried out by real-time two-way data communication with a full-scope simulator of the given 1000 MW USC power unit. It is shown that this method is of favorable engineering practicability, and it can effectively improve the dynamic load response speed and load control accuracy, and reduce the control deviation of the main steam pressure at the same time.