基于双层神经网络模型参数辨识的变结构多模型自主导航方法

Autonomous navigation method based on model parameters identification of TVSMM

在卫星拒止环境下,井下采煤机定位方式通常采用惯导/里程计的组合导航方式。但在采煤机开采过程中产生的复杂振动会引起传感器时变和非线性的器件误差,导致传统卡尔曼滤波精度下降。针对这一问题,提出了一种基于双层神经网络模型参数辨识的改进变结构多模型导航算法(TVSMM)。设计了一种基于支持向量机(SVM)和极限学习机(ELM)相结合的双层神经网络对系统噪声变化特性进行训练和辨识,实现快速、准确地辨识导航系统模型参数。仿真结果表明,通过双层神经网络模型参数辨识相较于单层神经网络的模型参数辨识精度提高12%。基于TVSMM的井下惯导/里程计组合导航算法可有效抑制振动噪声带来的航向角发散,航向精度提升42%,水平定位精度提升43%,对井下复杂环境中的自主导航定位具有较好的参考应用价值。

In the GPS-denied environment,the positioning method of shearer usually adopts the integrated navigation method of INS/odometer.However,the complex vibration generated during the mining process of the shearer will cause time-varying and non-linear errors of the sensors,resulting in a decrease in the accuracy of the traditional Kalman filter.In response to this question,an improved two-layer network model parameters identification based variable structure multi-model(TVSMM)navigation algorithm is proposed.A two-layer neural network based on support vector machine and extreme learning machine is designed to train and identify the noise variation characteristics of the system.The model parameters of navigation system can be identified quickly and accurately.The simulation results show that the model parameters identification accuracy by double-layer neural network is 12%higher than that by single-layer neural network.The underground INS/odometer integrated navigation algorithm based on TVSMM can effectively suppress the heading angle divergence caused by vibration,improve the heading accuracy by 42%,and improve the horizontal positioning accuracy by 43%,which has good reference and application value for autonomous navigation and positioning in complex underground environment.