Taking an ultra supercritical 1000 MW unit as the object

the factors influencing the unit load were analyzed in detail.By reasonably selecting the model input parameters and applying a BP neural network with n-order time-delayed inputs and m-order time-delayed output feedback

the unit load prediction model which takes the regenerative cycle characteristics into account was established.Moreover

various perturbation experiments were carried out to compare the load prediction results by neural network with different structures.The results show that

the model with 2-order time delay for the inputs and 2-order time delay for output feedback can predict the unit load accurately under different perturbations

and it can meet the engineering application need for improving the unit load response speed with the controllable variables in the regenerative cycle system under automatic generation control(AGC)condition.