Brief paper: Gradient dynamic optimization with Legendre chaos
Automatica (Journal of IFAC)
| Hi-index | 0.00 |
As the traditional ICP algorithm is liable to get local minimization problem, a chaotic BP neural network is presented in the ICP algorithm. In the algorithm, a searching area of real position was plotted centering on the indication of refer navigation system, then terrain altitude data was extracted from refer terrain map. These terrain data, along with corresponding position coordinates, were defined as several patterns and used to train BP network. The network can recognizes certain pattern class with measured water-depth data to determine vehicle's location. However, there are drawbacks of local minimization problem and slow rapidity of convergence in BP network, so improved ways were put forward. The improvement includes replacing common motivating function with chaotic motivating function for and determination of neural network's weights using chaotic search. The experimental results reveal that results of terrain matching can be improved, and matching failure caused by local convergence is overcome to a certain extent.