改进Q-Learning输电线路超声驱鸟设备参数优化研究

Research on Parameter Optimization of Improved Q-Learning Ultrasonic Bird Repellent Equipment for Transmission Lines

超声波驱鸟是一种解决输电设备鸟害的重要手段,但现场使用超声波驱鸟器工作模式较单一,易产生鸟类适应问题。提出了一种改进Q-Learning输电线路超声驱鸟设备参数优化方法,针对涉鸟故障历史数据量少以及鸟类的适应性问题,将强化学习算法应用于输电线路超声驱鸟设备参数优化;针对传统强化学习算法在设备终端应用中存在收敛慢、耗时长的缺点,提出一种基于动态学习率的改进Q-Learning算法,对不同频段超声波的权重进行自适应优化。实验结果显示,改进Q-Learning算法最优参数的迭代收敛速度大幅提高,优化后驱鸟设备的驱鸟成功率达到了76%,优于传统强化学习算法模式,较好地解决了鸟类适应性问题。

Ultrasonic bird repellent is an important method to solve the problem of bird damage in power transmission equipment,but the sole mode of operation that ultrasonic bird repellent was used in the field caused problems of the adaptability of birds.This paper presented an improved parameter optimization method for ultrasonic bird repellent equipment of Q-Learning transmission line,and the reinforcement learning algorithm is applied to the parameter optimization of ultrasonic bird drive equipment of transmission lines in order to solve the problem of little historical data of birds-related faults and the adaptability of birds.In view of the shortcomings of traditional reinforcement learning algorithms in device terminal applications,which have slow convergence and long time-consuming,an improved Q-Learning algo-rithm based on dynamic learning rate was proposed,which adaptively optimized the weights of ultrasound in different frequency bands.The experimental results showed that the iterative convergence speed of the optimal parameters of the improved Q-Learning algorithm was great-ly improved,and the success rate of bird repellent equipment after optimization was 76%,which is better than the traditional reinforcement learning algorithm mode,and can better solve the adaptability problem of birds.