中国电机工程学报2017，Vol.37Issue(20)：5852-5861,10.DOI:10.13334/j.0258-8013.pcsee.171391
内嵌专业知识和经验的机器学习方法探索(二):引导学习的应用与实践
Study of a Novel Machine Learning Method Embedding Expertise (Part Ⅱ):Applications and Practices of Guiding Learning
摘要
Abstract
This paper investigated how to apply guiding learning to electrical engineering research and development (R&D),by taking the example of the comprehensive evaluation and diagnosis of power distribution network's health index.Firstly,the background of distribution network's health index study was briefly introduced.Secondly,a guiding learning algorithm with embedded knowledge & experience was proposed.The algorithm adopted Softmax Regression as the benchmark learning model,then the knowledge function was established to interpret the professional knowledge & experience into the learning goals,and the Symbiotic Organisms Search for non-convex non-continuous learning objective function optimization was applied to obtain the optimal learning parameters.Thirdly,under four typical scenarios of training samples (e.g.whether they were labeled,whether there existed noise samples,or whether they were balanced sample sets),performances of the guiding leaming algorithm were tested and compared with the traditional Softmax Regression.Results show that guiding learning possesses high robustness and security;it can be applied to open complex learning tasks,such as power system under stochastic environment;Guiding learning is a promising method towards safety oriented artificial intelligence (safe AI).Finally,the development trend of machine learning software/platform was discussed,and a "unit" design framework for machine learning platform/system was presented for power system applications.关键词
引导学习/知识/经验/机器学习/人工智能/智能电网/健康诊断Key words
guiding learning/knowledge/experience/machine learning/artificial intelligence/smart grid/health diagnostic分类
信息技术与安全科学引用本文复制引用
尚宇炜,马钊,彭晨阳,盛万兴,苏剑,刘伟..内嵌专业知识和经验的机器学习方法探索(二):引导学习的应用与实践[J].中国电机工程学报,2017,37(20):5852-5861,10.基金项目
国家电网公司科技项目(EPRIPDKJ(2014)2863).Project Supported by Technology Project of SGCC (EPRIPDKJ(2014)2863). （EPRIPDKJ(2014）
