摘要
针对小样本多源信息故障预测时存在的参数模型难以建立和预测结果不准确等问题,改进了BP神经网络训练算法,设计了初级概率预测器进行故障概率预测。对初级预测结果进行演化趋势要素计算分析,提出了修正函数对预测结果进行二次修正。利用某型发动机的工作参数数据和音频信息进行了算法验证。实验结果表明,基于演化计算修正的神经网络故障预测方法克服了特征参数较少和样本量不足时造成的预测结果不准确的问题,随着样本量的增加保持了较好的故障预测效果。
The fault prediction model is difficult to be established based on small sample parameters of multi-source information,and the predicted results are not accurate. A primary probability predictor using the BP neural network was proposed to predict failure probability. The prediction results were calculated and analyzed based on evolutionary trend element. The correction function was put forward to correct the prediction results. A certain type of engine working parameter data and audio information were used to verify this method. It shows that the problem of inaccurate prediction results caused by fewer characteristic parameters and insufficient sample size was maintained with the increase of sample size.
作者
李果
屈重年
刘旭焱
刘伟
杜军
LI Guo;QU Chongnian;LIU Xuyan;LIU Wei;DU Jun(College of Mechanical and Electrical Engineering,Nanyang Normal University,Nanyang 473061,Henan,China;Astronautics Engineering College,Air Force Engineering University,Xi’an 710038,China)
出处
《实验室研究与探索》
CAS
北大核心
2020年第8期9-12,17,共5页
Research and Exploration In Laboratory
基金
国家自然科学基金项目(61306007)
河南省科技攻关项目(172102310682,172102210095,182102310760)
南阳师范学院博士专项基金(2018ZX024)。
关键词
神经网络
演化计算
修正函数
故障预测
多源信息
neural network
evolutionary computation
correction function
failure prediction
multi-source information
