Research On Structural Intelligent Control Algorithms Based On Gate Recurrent Unit Neural Network

For active control systems such as structural active,semi-active and intelligent control,the control algorithm is the key to affect the control effect.The traditional control algorithm requires the accurate model of the controlled object as the basis for estimating the control force,but it is difficult to realize for the structure with complex form and large volume.Therefore,more research and development talents begin to pay attention to the intelligent control algorithm that does not depend on the accurate model.Among them,the artificial neural network control algorithm stands out with its strong self-learning and adaptive ability,and is widely used in the field of vibration control.However,due to the increasing volume and complexity of the structure,the above-mentioned shallow learning methods such as artificial neural network expose some problems,such as insufficient expression ability of highly nonlinear functions,easy over fitting,and limited generalization ability when dealing with complex analysis and regression problems.In conclusion,it is necessary to study a new neural network more suitable for vibration control of civil engineering structures.Deep learning is a machine learning strategy that simulates the mechanism of human brain for data analysis and interpretation.Compared with shallow learning,deep learning can realize the approximation of complex nonlinear functions through multi hidden layer by layer learning,complete the feature extraction of external input data,and accurately predict and identify logarithmic data.Therefore,this paper considers the combination of deep learning theory and structural vibration control theory to propose an intelligent control algorithm based on gate recurrent unit network(GRU)and apply it to the field of structural vibration control.Taking the high-rise building structure as the main research object,relevant basic theoretical research,controller design research and structural simulation analysis have been carried out successively.The main research contents include:1)The GRU network prediction model framework for building structure response prediction is built through the simulation software MATLAB,and the network training and parameter optimization are carried out.The results show that GRU can be used to predict the response of building structures,and the predicted value fits well with the calculated value,and the prediction effect is good;2)Taking the three-story benchmark steel frame structure as the main research object,the proposed GRU network prediction model is applied to the design and research of intelligent control algorithm,and the effectiveness of this method is verified by simulation with Simulink system;At the same time,its control effect,generalization performance and robustness are compared with long short-term memory network(LSTM)and back propagation neural network(BP)control algorithms respectively.It is found that its overall performance is the best and has high application prospects;3)Taking the 20-storey benchmark steel frame structure as the main research object,the convolutional neural network(CNN)with strong feature learning ability is combined with GRU control algorithm,and the feasibility and excellence of CNN-GRU controller are verified through super parameter optimization,visual analysis and comparison of various controllers,which solves the problems of large data input dimension,strong nonlinearity and large calculation workload of high-rise building structure,thus,the operation speed of high-rise building controller is improved,and the accuracy and stability of control are guaranteed at the same time;4)Based on the research of CNN-GRU control algorithm above,aiming at the sensor failure problem of high-rise building structural control system,based on the idea of active fault-tolerant control and fault detection method,a multi-channel CNN-GRU intelligent active fault-tolerant control system for sensor fault detection and separation was established.Meanwhile,in order to verify the effectiveness of the proposed method,a 20-layer Benchmark steel frame structure is simulated and analyzed.The results show that: When some sensors fail,the above strategy can actively identify and eliminate the fault information of the sensor,and switch to the controller designed offline in advance to consider the failure of the sensor,ensuring the security and stability of the control system.