摘要
在北斗定位系统/惯性导航系统(BDS/INS)中,由于非线性噪声、精确构建运载体模型困难等原因,经卡尔曼滤波,列车的速度曲线会有部分信息损失,导致定位信息不准确。因此,提出了基于小波神经网络辅助的卡尔曼滤波法。神经网络的输入是卡尔曼滤波前的信号,其输出是滤波后的信号。考虑影响小波神经网络精度的因子,对权值、阈值和学习率进行了优化。在此基础上,通过加入对数函数对学习率进行改进,利用遗传算法(GA)和粒子群优化(PSO)算法组合优化权值和阈值。实例仿真表明,经过优化的小波神经网络提高了收敛速度、定位精度,同时在保持原速度波形的整体趋势前提下,降低了因滤波导致的信号误差。
In the process of Beidou navigation satellite inertial navigation（ BDS/INS）,because it is difficult to accurately construct the model of conveyance carrier and the noise is nonlinear,etc.,after Kalman filtering,some of the information of the speed curve of train may be lost,and inaccurate location information is obtained. Therefore,a wavelet neural network aided method for Kalman filtering is proposed. With this method,the input of the neural network is the signal before Kalman filtering,and the output is filtered signal. In addition,the factors that affect the accuracy of the wavelet neural network are considered; the weights,the threshold and the learning rate are optimized. On this basis,the learning rate is improved by adding logarithmic function,and the weights and thresholds are optimized by combining the genetic algorithm（ GA） and the particle swarm optimization（ PSO）algorithm. Instance example simulation shows that after wavelet neural network optimization,not only the convergence speed is improved; the positioning accuracy is also effectively enhanced; under the premise of maintaining the original speed waveform,the signal error caused by filtering is reduced,thus the purpose of the experiment is achieved.
出处
《自动化仪表》
CAS
2018年第1期74-78,共5页
Process Automation Instrumentation
关键词
小波神经网络
粒子群优化算法
卡尔曼滤波
遗传算法
BDS/INS定位
学习率
阈值
Wavelet neural network （ WNN ）
Particle swarm optimization algorithm （ PSO ）
Kalman filtering
Genetic algorithm
BDS/INS positioning
Learning rate
Threshold
