Abstract: In this paper, a radial flow magnetorheological (MR) valve was proposed, analyzing the working principle and magnetic circuit of the MR valve, deriving the mathematical relationship between the output pressure drop and the structural parameters of the MR valve based on MR fluid response basic mode, and establishing the calculation models for valve power consumption, MR fluid response and electromagnetic loop response. And a multi-physical field simulation analysis was carried out in COMSOL software to obtain the internal magnetic field characteristics and the dynamic process of magnetic fluid flow, generating magnetic field data based on the FEM model, and proposing an empirical model of feedforward neural network (FNN) to predict the magnetic flux density of MR valve. Adopting the design of experiment (DoE) technology for correlation analysis, two works were arranged, selecting the structural parameters to be optimized and developing an optimization platform for structural parameters. And then, the multi-objective optimization calculation was carried out based on the FNN model and multi-objective genetic algorithm (MOGA), and the optimal structural parameter set was obtained. Finally, four performance indexes of the initial and optimal MR valve with current change were simulated numerically. Taking the pressure drop and dynamic response time of the MR valve as the evaluation indicators, a test platform was setup to analyze the MR valve performance. The results show that when taking 2.0 A current, the pressure drop of the MR valve can increase up to nearly 2 times, the adjustable coefficient of pressure drop can increase by 24.40%, the corresponding rising response time efficiency can increase by 6.12%, and the falling response time efficiency can increase by 5.61%.