基于小波神经网络的多传感器自适应融合算法

Adaptive fusion algorithm based on wavelet neural networks for multisensor measurement

针对多传感器融合系统的非线性和不确定性,将小波分析与神经网络相结合,提出一种基于小波神经网络的多传感器自适应融合算法.融合系统包括扩展卡尔曼滤波器、小波神经网络、融合知识库以及航迹融合算法.该算法以分布式融合结构为基础,利用环境信息理论和测量方差归一化方法构建小波神经网络,并且通过数值样本训练小波神经网络,使其在融合过程中实时估计各传感器的信任度,再由融合知识库根据各传感器信任度来选择适合的航迹融合算法,最终得到全局状态估计.实验结果表明,提出的融合算法可以根据环境变化在线自适应融合来自多传感器的测量值,对不确定信息具有很好的融合能力.

To solve the problems of nonlinear and uncertain fusion systems, an adaptive fusion algorithm based on wavelet neural networks（WNNs） for multisensor measurement was proposed. The fusion system consisted of extended Kalman filters （ EKFs）, WNNs, knowledge base （KB） and track-to-track fusion algorithms. Based on the distributed fusion method, sensor precision values, sensor states and the local estimation errors were transferred from sensors to WNNs to deduce the relevant sensor confidence degrees in the real-time process of data fusion. In order to obtain the sensor confidence degrees, contextual information theory and normalized variable method were introduced to WNNs and the experimental data were implemented to train WNNs. According to the rules about the sensor confidence degrees, KB made decisions to select suitable track-to-track fusion algorithms. Simulation results show that the algorithm can effectively adjust the system to adapt contextual changes and has strong fusion capability in resisting uncertain information.