Abstract:

In order to achieve assigning imaging tasks quickly and efficiently in the earth observation network, a novel component-based solution structure. Composed of task coordinated allocator, task scheduler, feature extractor and schedulability predictor is proposed. Based on the classic imaging satellite scheduling model, the features of imaging tasks are extracted, and the imaging task scheduling prediction problem is solved by using the BP neural network ensemble technique for variable hidden layer nodes. Simulation results demonstrate that the back propagation (BP) neural network ensemble used in this paper for a single imaging satellite can reach daily schedulability prediction accuracy more than 85%.

Key words: Earth observation network, Task schedulability, Predict, Neural network ensemble, BP neural network