地球观测网络成像任务可调度性预测方法

Prediction Method for Imaging Task Schedulability of Earth Observation Network

为了能够快速、合理地分配成像任务,充分发挥对地观测网络的观测效能,对成像任务可调度性预测问题进行了研究,提出一种由协同任务分配组件、任务调度组件、特征提取组件以及任务可调度性预测组件所构成的组件化求解架构。在成像卫星经典调度模型的基础上,提取成像任务特征,并采用变隐含层节点的反向传播(BP)神经网络集成技术求解成像任务可调度性问题。仿真结果表明,集成BP神经网络的平均预测准确度可以达到85%以上。

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%.