基于粒子群优化SVM的LSTM电缆温度异常预测算法  

LSTM Abnormal Cable Temperature Prediction Algorithm Based on SVM with Particle Swarm Optimization

作　　者：关家华 郑楚韬 简淦杨 林晓璇 凌忠标 李果 黄燕生 

机构地区：[1]广东电网有限责任公司佛山供电局,广东佛山528000 [2]广州南网科研技术有限责任公司,广东广州510663 [3]南方电网科学研究院有限责任公司,广东广州510663

出　　处：《工业控制计算机》2024年第1期55-56,102,共3页Industrial Control Computer

基　　金：南方电网公司科技项目资助(GDKJXM20220063)。

摘　　要：为解决电缆温度异常预测问题,提出了一种电缆温度异常预测算法。首先,该算法采用支持向量机构建电缆温度异常判断的分类器,并采用粒子群算法进行自适应参数设置。其次,按照月份将电缆数据进行分类并得到12个数据集,同时将所有电缆数据作为1个数据集。接着,采用长短期记忆神经网络分别对这13个数据集进行训练,并通过训练预测精度加权将13个训练网络集成起来。最后,将电缆温度数据作为输入值,通过集成后的网络判断未来时刻电缆是否处于异常状态。实验结果表明,该文提出的预测算法效果良好,具备较高的应用价值。To address the issue of abnormal cable temperature prediction,this paper proposes an abnormal cable temperature prediction.Firstly,the algorithm employs support vector machine to construct a classifier for abnormal cable temperature judgment and uses particle swarm optimization for adaptive parameter settings.Secondly,the cable data is classified by month to obtain twelve datasets,and all cable data is used as one dataset.Subsequently,long short-term memory neural networks are used to train these thirteen datasets respectively,and the thirteen trained networks are integrated by training prediction accuracy weighting.Finally,the cable temperature data is used as input,and the integrated network is used to determine whether the cable is in an abnormal state in the future time.Experimental results show that the proposed prediction algorithm has good performance and high application value.

关 键 词：粒子群 分类器 长短期记忆人工神经网络 电缆温度 预测 

分 类 号：TM75[电气工程—电力系统及自动化] TP18[自动化与计算机技术—控制理论与控制工程]