Genetic Algorithms And Artificial Neural Networks Apply In Forecast Of Dam Security Monitoring

Displacement series, seepage series and crack breadth series of dam safety monitoring forecast, because of many factors restricting them, often represent complex nonlinear characters in series. With more automatization of dam safety monitoring forecast, common forecast model often represents that fit effect of model is good and its forecast effect is bad, which does not meet the need of dam safety monitoring automatization. Founded on change speciality of series of dam safety monitoring forecast, artificial neural networks and nonlinear models of time series based on genetic algorithms are applied.In dam safety monitoring forecast, because of convergent speed of the Back-propagation algorithm(BP) being slow and the partial least extremum existing, disturbing accelerating back-propagation algorithm(DABP) is established. Because fit effect of threshold auto-regressive(TAR) model is sometimes better than its forecast effect, or forecast effect is bad, TAR is improved. Auto correlation coefficients figure is suggested to ascertain auto regressive items of bilinear time series model(BM). A scheme based on genetic algorithms is applied to deduce BM. Combined strongpoints of TAR model and BM, threshold bilinear time series model(TBM) is suggested. According to TBM comparing with combined forecast(CF) model based on neural network of variable weighting moduli, the internal ideal combination of TBM is more excellent than the external result combination of CF model. Examples show that these improvements are effected, and these models apply successfully in dam safety monitoring forecast.