基于LSTM的阀冷系统入水温度及冷却裕度预测  

Prediction of converter valve cooling capacity and water inlet temperature based on LSTM

作　　者：廖毅[1] 罗炜[2] 蒋峰伟[1] 李亚锦[3] 于大洋[3] LIAO Yi;LUO Wei;JIANG Fengwei;LI Yajin;YU Dayang(Guangzhou Bureau,China Southern Power Grid EHV Power Transmission Co.,Ltd.,Guangzhou 510405,Guangdong,China;China Southern Power Grid Co.,Ltd.,Guangzhou 510623,Guangdong,China;School of Electrical Engineering,Shandong University,Jinan 250061,Shandong,China)

机构地区：[1]中国南方电网超高压输电公司广州局,广东广州510405 [2]中国南方电网公司,广东广州510623 [3]山东大学电气工程学院,山东济南250061

出　　处：《山东大学学报（工学版）》2021年第4期124-130,共7页Journal of Shandong University（Engineering Science）

摘　　要：为解决换流站阀冷系统状态缺乏智能预测手段和极端工况下冷却能力是否充裕难以评估的问题,提出基于长短期记忆网络(long short-term memory, LSTM)的换流阀冷却裕度预测方法。在阀冷系统冷却裕度指标量化评估的基础上,考虑多源影响因素,通过相关性强弱选择特征量并构建数据样本集,利用长短时记忆网络建立预测模型,并基于大量实际样本数据进行训练,对入水温度和冷却裕度发展趋势做出预测、提前预警,同时提供极端工况下冷却裕度的分析模型,为现场处理决策提供依据。通过穗东换流站的实例分析,验证了算法的有效性和可行性。In order to solve the problems of lacking of intelligent prediction means for the condition of converter valve cooling system and the difficulty of evaluating the cooling capacity under extreme conditions, a prediction method of cooling margin of converter valve based on LSTM(long short-term memory, LSTM)was proposed. On the basis of quantitative evaluation of cooling margin index of valve cooling system, the long-term and long-term memory network was used to establish the prediction model. Considering the multi-source influencing factors, the feature quantity was selected through the correlation strength and the data sample set was constructed, which was used to train the model. Training was carried out to predict the development trend of water inlet temperature and cooling margin.The analysis model of cooling margin under extreme conditions was provided, which provided the basis for on-site treatment decision. The effectiveness and feasibility of the algorithm were analyzed and verified by an example of converter station.

关 键 词：阀冷系统 长短时记忆网络算法 入水温度预测 冷却裕度 极端工况 

分 类 号：TM721.1[电气工程—电力系统及自动化]