Due to the high-dimensional circuit simulation, efficient performance modelling of today's analogue circuits is a critical but difficult task. To overcome this problem, we offer a unique performance modelling approach based on the Radial Basis Function in this study. The basic concept is to use data gathered at an early stage to help with efficient performance modelling at a later level. This is accomplished by using Radial Function inference to statistically estimate the performance association between early and late phases. This study thoroughly considers four implementation challenges, including previous mapping, lacking prior information, fast solver, and prior and hyper-parameter selection, in order to make the suggested method feasible. The radial basis function network uses a radial function as its transfer function. One of the advantages of RBFNNs is that they take less time to learn than current neural network architectures, making them useful in situations where speed is important. The brain computer was designed using an analogue-digital hybrid design, which is efficient and accurate. There are three contributions to the proposed RBFNN-GP. To minimize the error in a dataset, a local generalization error bound can be calculated by adding the sample mean and variance. To increase the generalization capacity of the brain, the self-organizing structure approach is established. Additionally, the study's results were based on a circuit's sensitivity, which means that the impact of the study's findings may vary depending on the circuit. The convergence of the proposal comes in third. In theory, RBFNN performs well when it comes to adjusting a structure's parameters. Two numerical simulations and a real-world application show that it is better than some well-known methods. The results of the comparison confirmed that RBFNN is effective. © 2023 Elsevier B.V., All rights reserved.