Tourism Emergency Events Prediction Based On Neural Network

The BP algorithm of neural network has been widely used in pattern recognition, image recognition, and data mining. BP algorithm has strong learning ability, and can fit the solution space by learning patterns lineraly. And it has become the effective tool for system modeling, emulation and prediction. This thesis based on the traditional Neural Network adopts many adjusted intelligence swarm algorithms to optimize the neural network training process, and improves the accuracy of prediction. The major works in this paper includes:(1) Adopt genetic algorithm to optimize the neural network initial parameters: adopt the advantages of genetic algorithm in the field of non-derivative optimization, and the searching ability in complex, nonlinear and non-differentiable function to optimize the neural network initial parameters, and then use the advantages of BP algorithm in local searching to optimize the solution. Through the experiments, we have found that this method can find the better solution.(2) Adopt genetic algorithm to adjust the topology of neural network: by adding efficiency factor to every gene, we can represent every gene's contribution to the whole chromosome's fitness, and remove the genes whose efficiency values are below the threshold when decoding chromosome. Through this method, we can optimize and simplify the topology of neural network.(3) Adopt particle swarm optimization (PSO) algorithm to optimize the initial parameters of neural network: in order to avoid PSO algorithm plunging in local optimal value, this thesis proposes adjusted tactics to renew inertia weight and the method of detecting prematurity. Meanwhile, it proposes the multi-PSO algorithm, and through experiments we find these methods can ensure the PSO algorithm and find better optimal value. By optimizing the initial parameters of neural network, it can accelerate and refine the BP algorithm training and prediction.This thesis focuses on the prediction of tourism emergency events, adopts neural network algorithm combined with many improved method and intelligence swarm algorithms to build the prediction model from the collected tourism emergency events. By experiments, we find that through these improved methods it has better results in training complexity and prediction accuracy.