On the performance evaluation of 3PL using subtractive clustering based RBFNN and UDM

Nowadays, the application of neural network technology in the evaluation of third party logistics (3PL) is very limited in China. The reason lies in the difficulty to find high quality training samples for neural network self-learning. This paper adopts uniform design method (UDM) to design representative, uniform and large-scale samples. And then use those samples to train the subtractive clustering based radial basis function neural network (SC-RBFNN) which is applied to carry out the 3PL evaluation. The result of the experiment shows that the generalization ability of the subtractive clustering algorithm based RBFNN combined with UDM is far better than that of traditional RBFNN. This method not only has the ability of determining the number of clusters and their values automatically and realizes non-linear approaching, but also conquers the performance limitations of traditional RBFNN. Moreover it avoids the subjectivity and uncertainty of traditional evaluation.