By analyzing a certain once through boiler’s FR/FW control system,a new control scheme based on growing and pruning dynamic recurrent fuzzy neural network (GAP-DRFNN)is proposed. This GAP-DRFNN can synthetically study main relative state parameters about FR/FW control,so as to calculate the optimal FR/FW by using least temperature deviation value of outlet of moisture separator as its training signal. As the data of current main relative state parameters input,GAP-DRFNN through structure learning can automatically increase and pruning neurons,and adjust the parameters and the recurrent weight of neural network dynamically based on stochastic gradient descent algorithm. The experimental results show the good performance for the system in variable conditions and this scheme’s celerity and precise on FR/FW control,it has better quality than traditional PID control method.