The resonance state and output characteristics of high-frequency magnetostrictive transducers are affected by various factors such as electromagnetic excitation conditions， heat generation and load conditions. In order to improve the output performance of high-frequency magnetostrictive transducers， this paper deeply studies the influence regularity with the changing of operating temperature and loads on the resonant frequency and optimal bias magnetic field of the transducer and designs a closed-loop control system. Firstly， according to the transducer impedance circle and output acceleration characteristics， the resonant frequency and the optimal bias magnetic field of the magnetostrictive transducer under different temperatures and loads are experimentally tested to find out their changing rules. Then， based on the test data， a BP neural network prediction model is established and optimized by genetic algorithm and particle swarm optimization algorithm to characterize the nonlinear relationship among temperature， load， resonance frequency and optimal bias current. Finally， the closed-loop control system of magnetostrictive transducer is built and the reference value of the closed-loop controller is adjusted in real time by using the prediction results of the proposed GA-BPNN prediction model to realize the automatic tracking of resonance frequency and optimal bias magnetic field. The experimental results prove that the control system is effective in optimizing the output characteristics of the transducer. The output acceleration amplitude of the transducer can be increased by an average of 25.65% under different working conditions.