A method to establish a neural network based on the whole knowledge

Aiming at problems in establishing a traditional neural network, this paper presents an algorithm called WKBNN (Whole Knowledge-Based Neural Network) which combines induction and analysis, two methods of machine learning to distill the information from the existing data, and then finds some rules to compute the initial parameters of the neural network. This method not only endows those crabbed parameters actual meaning, but also has more prominent advantages than the method of the traditional neural network. According to the simulation of an instance, the paper has compared WKBNN with classical BP algorithm, and then found that WKBNN has shortened the calculating time and improved calculating accuracy greatly. Obviously, it is a beneficial attempt to extent application of the neural network.