The Research Of Autonomous Learning Back-Propagation Algorithm

Neural network evolved from biological neurons, neurons connected to a united network,the most widely used neural network is the back-propagation neural network, it can adjusting the interconnected nodes in the network, finally achieve the purpose of information processing, So BP neural network can solve the network classification problems. In such a rapid development of education career, who can master the new trend of developing information and learn scientific knowledge, he can grasp the future. Therefore, it is a new challenge that make autonomous learning and comparatively mature neural network be a field.Many factors influenced the nonlinear mapping and generalization ability, made the network easily fall into the local extremum value and convergent speed, these factors includes random generated connection weights of each layer and thresholds in the network, lacks the strict theoretical basis to the network structure and the number of hidden layer nodes, without consider the sample datas to the iterative learning effects. According to the above questions of BP and its basic principle, combined with the features and the basic model of autonomous learning. This paper puts forward the global optimal value for autonomous learning algorithm, constructs autonomous learning neural network model, and more importantly, improved the activation function, and put forward the output layer neurons learning error function, increase tightness variables for the change of weight values, increase learning rates dynamic change, Thus the improved BP algorithm of autonomous learning Back-Propagation algorithm can solve most tasks.This paper do many improvements which influence the functions of BP neural network, including:network structure, hidden neuron numbers, weights and thresholds of network, total error, activation function and learning rates, etc.This paper constructs autonomous learning neural network model and puts forward autonomous learning algorithm of neural network, make two groups simulation experiment about the number of national college students enrollment and graduate ratios, compare the actual values and the predict values of enrollment numbers and employment ratios. Two groups of experiment results indicate:for the standard BP algorithm, increase the momentum of BP algorithm, particle swarm algorithm and autonomous learning back propagation, autonomous learning back propagation has better performances in convergence speed, forecasting precision, generalization ability, etc.