Based On The Structure Of The Neural Network Intelligent Control

Structural active control is more efficient in the control of seismic response, which makes active control have wide application prospect in the field of structural vibration control. But active control is hard to establish an accurate mathematics model and has the problem of time lag. Neural network need not establish accurate mathematics model, it sums up the relation implicit in the systematic input output through studying input output training sample data. Applying neural network to predict structural response may solve the problem of time lag in active control and offer basis for controlling decision. Now, general prediction neural network has BP neural network and RBF neural network mainly. BP neural network has shortcoming of slow learning speed and falling into the local minimum easily, and the study and remember of BP neural network is instable; compared with BP neural network, RBF neural network has fast learning, the strong dynamic simulation property, the better map property of input and output and global optimum property. It can avoid local minimum, can linearly adjustment weight value, and not back propagation error; at the same time it is realized by hardware relatively simply. It has been a good choice for real-time identification and controlling for various nonlinear systems. Predicting structural dynamic response quickly helps overcome time lag in structural control. The response of structure is predicted by RBF neural network, and compared with the results of BP neural network. The results of simulation suggest that RBF neural network is feasible and fast for prediction of structural dynamic response and the results are quite exact. The method can offer the comparatively accurate optimization performance index for the structural response in time, thus offering the fine assurance for realizing the online real-time control structural response. To improve further network's performance, namely, the prediction of structural response based on neural network has higher forecast precision for different earthquake excitation, we analyzed the training sample data of neural network: earthquake excitation is one important factor in predicting structural response, so when training sample is selected, the effect of earthquake excitation must be considered. Besides, the structural response is another important factor, so the structural response must be considered too. El-Centro wave and the calculating results of sequential optimal control algorithms are selected as training sample, and then the performance of network is tested with Kobe wave, North-Ridge wave and Tafl wave. The results of simulation suggest that RBF neural network may get the dynamic property of structure generally and the performance of the network is improved. The results of the prediction are more good in the effect of different earthquake excitation.