基于RBF神经网络的磁屏蔽性能计算方法

针对磁屏蔽性能理论计算存在的不足,提出了基于径向基函数(radial basis function, RBF)神经网络的磁屏蔽性能理论计算方法。首先,采用控制变量法对影响磁屏蔽性能的独立参数进行分离并建模;然后,利用RBF神经网络对非独立参数进行建模;最后,将训练好的RBF神经网络模块与分离出来的独立参数模型进行结合,得到磁屏蔽装置的磁屏蔽性能计算模型。通过对矩形磁屏蔽装置的磁屏蔽性能进行仿真计算,结果表明,提出的基于RBF神经网络的磁屏蔽性能理论计算方法与有限元数值法的最大相对误差为10.3%,95%的拟合结果与数值计算结果的相对误差在8%以内,与传统解析法相比,该方法更加接近数值法的求解精度,更适用于工程估算。

Calculation of magnetic shielding performance based on RBF neural network

In view of the shortcomings of theoretical calculation on magnetic shielding performance, a theoretical calculation method of magnetic shielding performance based on radial basis function (RBF) neural network is proposed. First of all, the control variable method is used to separate the independent parameters. And the independent parameters are modeled independently. Then, the finite element numerical method is used to obtain training data by Ansoft Maxwell software. And the training data is used to train RBF neural network to model the non independent parameters. Finally, the independent parameters model and the RBF neural network module are combined to obtain magnetic shielding factor calculation model of the magnetic shielding device. Taking the magnetic shielding factor calculation of the rectangular magnetic shielding device as an example, the simulation results show that the fitting results of the calculation model are consistent with those obtained by numerical calculation. The maximum relative error between the model calculation results and the numerical calculation results is 10.3%, and 95% of the relative error is less than 8%. Compared with the traditional analytic method, it is more close to the accuracy of the numerical method and is more suitable for engineering estimation.