Research On Task Scheduling Algorithms Based On Q-Learning For Sensor Nodes

In dynamic Wireless Sensor Networks (WSNs) environments, how to effectively use constrained resources of sensor nodes while obtaining better application performance is a key issue. The task scheduling algorithms based on Q-learning for sensor nodes are effective ways to solve this problem. According to the learning type, the learning algorithms can be divided into the independent learning algorithms and the cooperative learning algorithms.In order to solve the problem of poor application performance in the existing independent learning algorithms at the low energy consumption scenarios, the independent Q-learning model is established and an independent task scheduling algorithm based on Q-learning and programming is proposed by defining some basic learning elements. This algorithm improves the application performance of sensor nodes by optimizing its learning policy. Specifically, the programming process based on the priority mechanism and the expired mechanism is established to make full use of empirical knowledge, which can strengthen the trend to schedule the task which is conducive to the learning and weaken the trend to schedule the task which is not conducive to the learning. The experimental results on NS3 show that this algorithm has the ability to schedule the tasks dynamically according to environmental changes; compared with other independent learning algorithms, it achieves better application performance at the low energy consumption scenarios.As a result of frequent exchange of cooperative information, the existing cooperative learning algorithms are not fit for the high application performance scenarios. In order to solve this problem, the cooperative Q-learning model is established and a cooperative task scheduling algorithm based on Q-learning and shared value function is proposed by defining some basic learning elements. Specifically, according to the change characteristic of value function, the algorithm designs the sending constraint and the expired constraint of state value to reduce the switching frequency of cooperative information while guaranteeing the cooperative learning effect. The experimental results on NS3 show that this algorithm can perform task scheduling dynamically according to environmental changes; compared with other cooperative learning algorithms, it can make sensor nodes complete its application functionality work normally with the cooperative learning algorithm, which can be apply to the high application performance scenarios.