Abstract
Reserves of polymetallic nodules is a great mineral resources, is widely distributed in the ocean floor, the distribution is complex and nonlinear effects by many factors, resulting in the distribution of polymetallic nodules in the seabed appeared to be random, and therefore, distribution of polymetallic nodules are non-linear aspects of the problem. In this paper, artificial neural networks in the most widely used BP network, the establishment of the geological factors that control the distribution of polymetallic nodules longitude, latitude, depth, slope and its distribution between the coverage prediction model. Number of hidden layer neuron identification, transfer function, training function, the choice of learning function, to promote empowerment, control of the network built by the pros and cons. Finalized the 4-5-1 network structure, hidden layer neurons in the transfer function using tansig function, transfer function of output layer neurons use purelin function, training function selection traindgx function, learning function, choose the default learngdm function, improve the promotion of ability to approach a "stop ahead" method. Final training into a more predictive value of the network, not exploration area is predicted.
Keywords: BP neural network;hidden layer;Regression Analysis;Generalization;Contour