基于价值函数和策略函数的站用直流电源系统电源故障预判算法
A Power Failure Prediction Algorithm for Station DC Power Supply System Based on Value Function and Strategy Function
许明政 1李仲强 1蔡志强 1刘嘉豪 1徐强1
作者信息
- 1. 国网安徽省电力有限公司超高压分公司,安徽 合肥 230000
- 折叠
摘要
在站用直流电源系统电源故障预测过程中,主要依托于单一的价值函数实现故障预判,无法应对策略微弱变化引起的维度爆炸等问题,使得故障预判结果的G-mean值较低.因此,提出结合价值函数和策略函数的站用直流电源系统电源故障预判算法.针对电源历史运行参数进行归一化处理,经过灰色分析处理筛选出符合要求的参数,组成故障预判分析数据集.以描述数据序列方差变化为目标,构建改进价值函数.利用改进的价值函数和策略函数组成强化学习智能体,设计包含多层感知机网络的电源故障预判模型.经过数据训练调整模型处于最优状态,获取高质量的直流电源系统电源故障预判结果.实验结果表明:所提算法得出的电源故障预判结果G-mean值为0.98,保证了故障预判结果的准确性.
Abstract
In the process of predicting power supply faults in station DC power supply systems,fault prediction mainly relies on a single value function,which cannot cope with dimensional explosions caused by weak changes in strategies,resul-ting in low G-mean values in fault prediction results.Therefore,a power failure prediction algorithm for station DC power supply systems is proposed that combines the value function and strategy function.Normalize the historical operating param-eters of the power supply,select the parameters that meet the requirements through grey analysis,and form a fault predic-tion analysis dataset.Construct a combination value function to describe the variance changes in a data sequence.The rein-forcement learning agent is composed of improved value function and strategy function,and a power failure prediction model including multi-layer perceptron network is designed.After data training and adjusting the model to be in the optimal state,high-quality DC power system power fault prediction results are obtained.The experimental results show that the G-mean value of the proposed method for power supply fault prediction is 0.98 ensuring the accuracy of the fault prediction results.
关键词
价值函数/策略函数/多层感知机MLP/站用直流电源系统/电源故障/故障预判Key words
value function/strategy function/multilayer perceptron MLP/station DC power supply system/power fail-ure/fault prediction引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据