Modeling of the charging characteristic of linear-type superconducting power supply using granular-based radial basis function neural networks

Since superconducting coils cause the current decay due to connection resistance and intrinsic characteristic in the persistent current mode, various current compensations should be required to maintain stable property in the superconducting magnet system. As an achievable solution, we fabricated a new prototype power supply, i.e., linear-type superconducting power supply (LTSPS) and we carried out the operating characteristics of LTSPS in the small scale magnets. In order to apply the LTPSP for real scale magnet, charging characteristics for various magnet scales should be expected. In the paper, based on the experimental results, we design a modeling of charging characteristic of LTSPS using a granular-based radial basis function neural networks (RBFNNs) realized with aid of the granular techniques. In contrast with the plethora of existing approaches, here we consider a development strategy in which a topology of the network is predominantly based upon a collection of information granules formed on a basis of available experimental data. The underlying design tool guiding the development of the granular-based RBFNN is based on two categories of methods such as K-means clustering and the fuzzy C-means (FCM) clustering method. As shown in the experimental studies, the granular-based RBFNN possesses essential properties of universal approximation. The model of charging characteristic of LTSPS constructed is concerned with a limited dataset (where the limitations are associated with high costs of acquiring data) as well as strong nonlinear characteristics of the underlying phenomenon. For modeling and evaluation of the performance of the SPS neural network, in advance we also run experiments for other publicly available data such as a well-known NASA software project data as well as synthetic nonlinear data.