Neural network based load prediction model for an ultra-supercritical turbine power unit

Abstract

Widespread implementation of the regional power grid centered automatic generation control (AGC) proposes higher demands on the unit load control precision, rate and response time of a large-scale ultra-supercritical power unit. To improve the unit load control quality with advanced intelligent control strategies, it is of great significance to establish an accurate load prediction model for the steam turbine unit. A 1000MW ultra-supercritical turbine power unit is taken as the object investigated in this work. By taking its regenerative cycle system, hot-side and cold-side steam parameters into consideration, a BP neural network with time-delay inputs and output time-delay feedbacks is adopted to establish a nonlinear dynamic load prediction model for the steam turbine unit. By optimizing the neural network model structure and the inputs/output time-delay orders through elaborate real-time simulation tests, the optimal model structure is determined, which is with higher load prediction accuracy, good generalization ability and fit for intelligent coordinated controller design to improve the unit load control.