基于多粒度LSTM网络模型的换流站设备分析研究

Analysis and Research of Equipment in Converter Station Based on Multi-granularity LSTM Network Model

换流站的设备状态分析对日常工作的展开具有重要意义。提出了一种基于多粒度长短期记忆(long short-term memory, LSTM)网络模型的时间序列预测方法,从采样粒度、时粒度和天粒度对设备的各个监测参量展开分析。对典型的设备数据趋势进行分类,针对不同类别的数据进行数据预处理,包括离群点处理、缺失值处理和标准化处理;根据相关系数大小进行特征选择,并通过滑动窗口法构建样本集;对每类数据设计多粒度LSTM网络模型进行预测,得到不同时间粒度的状态预测值。实验结果表明,多粒度LSTM网络模型的预测误差小于ARIMA和Xgboost模型,该模型具有较好的预测效果。

The equipment state analysis of converter station is of great significance to the development of daily work. A time series prediction method based on multi-granularity long short-term memory(LSTM) network model is proposed in this paper. The monitoring parameters of the equipment are analyzed from sampling granularity, hour granularity and day granularity. The typical data trends of the equipment are classified, and the data of different types are preprocessed, including outlier processing, missing value processing and standardization processing. Feature selection is performed according to the correlation coefficient, and the sample set is constructed by the method of sliding window. Multi-granularity LSTM network model is designed for each type of data to obtain the state prediction values at different time granularities. The experimental results show that the prediction error of multi-granularity LSTM network model is smaller than that of ARIMA and Xgboost models, and it has a good prediction effect.