摘要
为了提高现有雷达故障的诊断效率和诊断精度,提出了一种基于增强学习神经网络的雷达故障诊断模型。首先,对基于神经网络的故障诊断模进行了构建和分析;然后,给出了采用马拉特(Mallat)塔式小波变换算法对故障输入数据进行特征提取的方法,将神经网络故障诊断模型的所有参数作为马尔科夫决策模型(MDP)的状态空间,采用增强学习中的行动者评论家算法来寻求最优参数,即采用评论家对当前状态进行评价,并通过行动者对当前状态进行不断改变。在上述讨论基础上,采用反向传播算法再次训练模型。仿真结果表明:文中方法具有较高的故障诊断精确度,相比其他方法具有故障诊断效率高的优点。
In order to improve the diagnosis efficiency and accuracy of the existing fault diagnosis methods, a fault diagnosis model based on reinforcement learning and neural network is proposed. Firstly, the fault diagnosis model based on neural network is con- structed and analyzed. The method for extracting the feature by using Mallat algorithm is given, the parameters of the MDP model of the neural network are trained by actor-critic algorithm in reinforcement learning, namely, the critic is used to evaluate the state and the actor is used for changing the state. The neural network is then retrained by the BP algorithm. For verifying the proposed method, the method based on neural network optimized by reinforcement learning is simulated and verified, the result shows that the proposed method has high diagnosis accuracy, meanwhile, compared with the other methods, it has the advantage of higher di- agnosis accuracy.
出处
《现代雷达》
CSCD
北大核心
2017年第12期15-19,共5页
Modern Radar
基金
四川省科技厅资金资助项目(2015GZ0279)
关键词
雷达
增强学习
故障诊断
神经网络
行动者评论家
radar
reinforcement learning
fault diagnosis
neural network
actor-critic
