Abstract： In view of the shortcomings of the simple genetic algorithm(SGA)，such as slow convergence and premature character，a new algorithm with fast convergence，good performance is proposed by improving the encoding，generation of initial population and genetic operators. And the new algorithm is tested and verified by experiments. An evolutionary neural network(ENN) model is constructed by using this algorithm to evolve the architecture and parameters of neural network. At last，the prediction on the thickness of the loosen zone around roadway is made with this ENN model，and the results show that the convergence and the forecasting capability are all good.
