Research On Reservoir-induced Seismic Risk Assessment Based On Neural Network And Genetic Algorithm

With the continuous development of water projects, which has brought many benefits to mankind, while the negative impact of water conservancy projects can't be ignored. Especially after a reservoir was constructed, sometimes the load of the reservoir was changed, the change of stress was produced, then Reservoir-induced Seismicity was caused. Despite the probability of Reservoir-induced Seismicity was small, the loss of human life and property which has been made by Reservoir-induced Seismicity was heavy.In order to reduce this loss, foreign scholars started to research it very early in the characteristics, the mechanism and the conditions of the Reservoir-induced Seismicity. These studies provided a solid foundation for the later prediction. From the beginning of the geological analogy method, genetic analysis method until the later method of probability and statistics, pattern recognition method, etc. are to some extent contribute on the Reservoir-induced Seismicity risk assessment, but because of Reservoir-induced Seismicity causes of the relationship between the conditions ,the causes of conditions and induce the relationship between the magnitude are very complex, only using the above approach can't describe their nonlinear relationship vividly.Because of the knowledge of conditions is limited on induced seismicity research, we can only seek to improve on the prediction method. Taking the recent rapid development of neural network algorithm into account, in particular the BP neural network, it has strong nonlinear processing capabilities that reflect the most essential part of artificial neural network. Using it to predict Reservoir-induced Seismicity is a feasible approach, but as BP neural network itself is easy to fall into local minimum point of problem. This paper combined with Genetic Algorithm which has strong global searching capabilities to optimize weights and thresholds of the Neural Network, and finally established the Genetic Algorithm/Neural Networks model for risk assessment of Reservoir-induced Seismicity on MATLAB.The application of induced seismic data on the Three Gorges Reservoir Area and its surrounding areas showed that the prediction performance is different greatly when induced seismic conditions of Reservoir-induced Seismicity use different coding schemes in BP Neural Network model which has been established. Knowing that the performance of the prediction is best, as the coding scheme is binary and has some intervals between the various factors, and increasing by sequence by comparison. However the strong randomness of BP Neural Network leads to the stability of prediction is not enough. On this basis, using Genetic Algorithm optimizes weights and thresholds of the BP Neural Network and then predicts it again. Results showed that it's better than the simple BP Neural Network model in terms of accuracy, stability, or efficiency.