Research On Gyro Error Compensation And Stability Control Of Platform Based On Neural Network

The gyro-stabilization platform is the core of photoelectric tracking systems.It can isolate the impact of the carrier movement,wind resistance torque,internal friction and other factors on the photoelectric system to keep the visual axis of the probe on the platform stable in the inertial space.Thus,it is widely used in military as well as civilian areas.In this dissertation,a two-axis gyro-stabilized platform is used for the study.It is mainly studied from two aspects deeply.One aspect is the signal error compensation for the fiber optic gyro and the other aspect is the control performance improvement of the stability servo ring.First of all,this dissertation summarizes the research status of gyro-stabilized platform at home and abroad,introduces the faced with development problem of the gyro-stabilized platform,methods of fiber optic gyro error compensation,and the commonly used algorithms for the photoelectric platform stability control.As for the neural network,the dissertation summarizes its learning algorithms and its stability research status.Furthermore,the overall design scheme is given based on the design indicators,the devices selection is completed and the mathematical model of the uniaxial servo system is established.Following the research,the factors affecting the stability of the photoelectric platform and the kinematic principle of external disturbance resistance are analyzed,the relationship between platform stability and isolation degree is deduced to provide a theoretical basis for the design of servo control system.Secondly,in order to analyze the effect of fiber optic gyro random error on the stability of platform,this dissertation introduces the working principle of fiber optic gyro,the sources of noise,and the principle of ALLAN variance.During the process of random error modeling of fiber optic gyro,as to the problem of the simulation difficulty for non-stationary classification noise,the dissertation utilizes a simulation method of making white noise go through a multi-stage series filter.And the accuracy of the modeling method is verified accurate by a variety of power spectral density methods.Compared with the existing methods of using orthogonal wavelet to simulate the type of noise,the method used is relatively simple and convenient.And through the simulation analysis of the effect of gyro random error on the angular rate and angular position of the platform system,it shows that the gyro error compensation is important to improve the stability of the platform.Then,in order to improve the accuracy of gyro output,a method based on wavelet filtering and making use of single neuron approximation property is proposed.Meanwhile,a single multiplication recursive neuron model is proposed and a linearly decreasing inertia weight firefly differential evolution algorithm is also proposed to train the new neuron model.This novel evolution algorithm can effectively overcome the shortcomings of the BP algorithm,such as easily falling into the local minimum,the slow convergence speed and the poor robustness.All the characters lay a solid foundation for the rapid convergence of single neuron network.By comparing and analyzing the gyro error property change before and after compensation through the ALLAN variance,the gyro error is compared and analyzed,the results show that the proposed method can effectively compensate the gyro error and weaken the impact of random error on the stability of the platform.Then,a PID controller is designed for the stabilization servo system of the photoelectric platform,a series correction controller is designed based on the frequency domain method,and an adaptive compound controller is also designed combining the advantages of single neuron PID controller and recursive adaline neural network.Through the control performance analysis of the three controllers on the aspects of the dynamic response and the stability precision,the experimental results show that the adaptive composite controller based on neural network is superior for the reason of combination of the advantages of the traditional control method and neural network.Finally,as a further discussion of the control method for photoelectric platform based on neural network,a backstepping robust controller based on recursive RBF neural network is designed for DC torque motor system.The stability of the controller is proved by mathematical proof and a simulation example is also given.This advantage that this method does not need system identification makes it outstanding in the stability control of the photoelectric platform.At the same time,for the problem that photoelectric platform system has high requirement for the real-time but usually the neural network has the delay,and is easily influenced by the pulse and the reaction diffusion,that easily causes the divergence of neural network,this dissertation presents a simple method to determine the global stability of neural networks and gives a rigorous theoretical proof.This has made a positive theoretical contribution to the practical application of neural network in the related fields such as stability control of photoelectric platform.