【作者】 杨丽；

【导师】 赵伟；

【作者基本信息】 南京航空航天大学 ， 制导与控制， 2010， 硕士

【摘要】 惯性导航系统(Inertial Navigation System,INS)与全球定位系统(Global Position System,GPS)相结合是提高导航精度和可靠性的重要途径之一。论文针对GPS/INS系统中GPS信号应用不可靠的问题,对惯性组合导航系统建立误差预测模型,以修正GPS信号无效期间INS的导航参数,从而提高INS独立工作时的精度。为了提高模型的训练速度、泛化性能和防止陷入局部极小的情况,选用先进的小波神经网络(Wavelet Neural Network,WNN)并引入了遗传算法(Genetic Algorithm,GA)进行误差预测模型的建立。同时,在建模前,对惯性导航系统中占主体地位的惯性元件建立准确的数学模型并对其误差进行有效地补偿或削弱。论文以小波变换的尺度特性为基础,主要针对惯性导航系统中随机误差建立了相应的模型,以提高系统的导航精度,研究了基于小波多尺度理论的组合导航系统的滤波算法;建立GPS/INS组合导航系统的小波神经网络滤波算法,以提高INS独立工作时的精度。研究使用小波多分辨率分析(Wavelet multi-resolution analysis,WMRA)处理陀螺信号,对载体最大范围内角运动的陀螺测量信号进行频谱分析和频带能量统计,确定陀螺信号的有用频带,并利用小波包动态阈值降噪算法针对陀螺各频带特征进行自适应滤波降噪。针对惯性器件研究利用小波理论分析陀螺的漂移特性,提取光纤陀螺的随机漂移;在此基础上,研究基于自适应线性神经网络建立随机漂移的预测模型,利用小波神经网络非线性预测算法对陀螺漂移实时预测。将平稳小波变换(Stationa Wavelete Transformation,SWT)应用到GPS/INS组合导航中,利用小波阈值滤波对惯导信号多分辨分析后各层的小波系数进行适当处理,有效抑制了惯导的误差,提高了组合导航的精度。研究利用小波变换对GPS、INS的位置、速度信息进行动态滤波的算法,对信号的低频发展趋势以小波系数的形式加以提取进而得到INS相对于GPS的误差。针对GPS/INS组合导航系统在实际应用中遇到的问题,提出小波神经网络的自适应滤波算法,旨在利用滤波器得到的INS误差模型修正GPS信号无效期间INS的导航参数,从而提高INS独立工作时的精度。该模型采用小波神经网络避免了其它网络存在的局部最小化的缺陷,并将小波分析和遗传算法引入其中,对GPS失效情况下的INS数据预测和建模,有效提高了导航精度。在GPS有效时,该算法采用基于遗传算法的小波神经网络对GPS/INS信号建立INS位置、姿态、速度误差预测模型。在GPS失效时,利用已建立的预测模型预测INS位置、姿态、速度误差来修复INS数据,实现机体在复杂机动下的实时精确导航定位。更多还原

【Abstract】 The integrated INS/GPS navigation is one of the most common used integration navigation approach aims to improve the navigation accuracy and reliability. This paper mainly focuses on the condition when GPS signal is invalid in the integrated INS/GPS navigation system. In order to keep the navigation accuracy when the INS works independently, the mathematical error prediction model of the navigation system is established and then the parameters of the INS can be amended. The genetic algorithm combined with the advanced wavelet neural network is used in the modeling on the navigation errors in the purpose of speeding the training process, enhancing universality and avoiding the model into a local minimum situation. Meanwhile, the inertial components are modeled accurately before it working.To improve the navigation accuracy, the inertial navigation system random error is modeled according to the wavelet transform scale features, and the integrated navigation system filtering algorithm based on wavelet multi-scale theory is studied. In order to keep the navigation accuracy when the INS works independently, the wavelet neural network filtering algorithm is introduced into the INS/GPS integrated navigation system.The method of using wavelet multi-resolution to analysis and process gyro signals is researched. The spectrum analysis and band energy statistics for the gyro measured signal are realized to determine the useful band of the signal, and then the gyro signals are adaptively filtered by the wavelet packet noise reduction algorithm with dynamic threshold in different band features. The wavelet theory is studied to analyze the drift characteristics of the inertial devices, and the random drift noise of the fiber optic gyro is dynamic calibrated using this method. Further more, the random drift error of the gyro is modeled based on the adaptive linear neural networks theory and real time forecasted by the wavelet neural network nonlinear prediction algorithm. The use of wavelet transform for integrated INS/GPS navigation system and the appropriate treatment to the wavelet coefficients of the INS signal layers by the wavelet threshold multi-resolution analysis effectively inhibit the effects of the INS error and improve the accuracy of integrated navigation.The dynamic filter algorithm with wavelet transform focus on the position and speed signals of INS/GPS systems is studied which could develop the trend of low-frequency signals in the form of wavelet extraction system and then get the error of INS relative to the GPS. And for the problem of integrated INS/GPS system encountered in the practical application, the wavelet neural network adaptive filter algorithm is put forward, which aims to compensate the INS navigation parameters by the calculated INS error model, thereby enhancing the independent INS work accuracy. This INS error model is based on the wavelet neural network which avoid the defect of local minimize problem compared to other kinds of network model method. And the wavelet and genetic algorithm are introduced into the system to analyze and predict the INS parameters in the GPS lost lock condition, effectively improve the navigation accuracy. When GPS is effective, the INS position, attitude and speed errors are modeled upon the INS/GPS message using the genetic algorithm based on the wavelet neural network. And when the GPS is useless, INS navigation parameters are modified by the established error prediction model. Thus the real-time precision navigation and positioning for the maneuvering aircraft comes true.更多还原