Neural network model of pipe network for state estimation based on modified genetic algorithm

Simple genetic algorithm (GA) has the shortcomings of low convergence rate and premature convergence, while BP neural network is prone to the local optimum and its structure is usually difficult to be determined. In this paper, a real-coded self-adaptive GA was introduced to optimize the weight and threshold so that binary-coded self-adaptive GA can find the best topological structure, and then a chaos genetic algorithm was proposed for global optimization of the weight and threshold. A macroscopic state model of pipe network was developed based on improved BP neural network. Case study shows that the new model has higher prediction accuracy.