摘要
电网中的无功功率、三相电网不平衡等因素使得电能质量问题日趋严重,必须采取有力的监测措施去改善和控制电网中电能质量。监控模块中的预警预测算法的优化是亟须解决的重要问题。本文选取上海市某220kV变电站电能质量数据,以电压偏差数据为例,根据其时间序列特征提出一种基于ARIMA-LSTM组合模型的电压偏差预测方法。利用ARIMA模型对时间序列数据拟合,将原始序列分解为两条序列,即预测值序列和误差值序列。LSTM模型对误差值序列进行拟合优化,并与ARIMA模型所得预测值序列叠加得到最终预测结果。实验对比分析了单一模型ARIMA与组合模型ARIMA-LSTM在误差值序列优化后的精确度。实验表明,组合模型将预测的误差值进一步优化后,预测效果优于单一模型。实验证明了该方法是有效可行的,值得优先采用。
Factors such as reactive power and three-phase unbalance in the power grid make power quality problems more and more serious.Therefore,effective monitoring measures must be taken to improve and control the power quality in the power grid.The optimization of the early warning and prediction algorithm in the monitoring module is an important problem that needs to be solved urgently.This paper selects the power quality data of a 220kV substation in Shanghai,takes the voltage deviation data as an example,and proposes a voltage deviation prediction method based on the ARIMA-LSTM combined model according to its time series characteristics.The ARIMA model is used to fit the time series data,and the original series is decomposed into two series,namely the predicted value series and the error value series.The LSTM model fits and optimizes the error value sequence,and superimposes it with the predicted value sequence obtained by the ARIMA model to obtain the final prediction result.The experiment compares and analyzes the accuracy of the single model ARIMA and the combined model ARIMA-LSTM after the error value sequence is optimized.Experiments show that after the combined model further optimizes the predicted error value,the prediction effect is better than that of the single model.Experiments have proved that this method is effective and feasible,and it is worthy of priority.
作者
李孟特
于晟华
王森
曹戈
戴雨聪
LI Mengte;YU Shenghua;WANG Sen;CAO Ge;DAI Yucong(State Grid Shanghai Pudong Electric Power Supply Company,Shanghai 200122,China.)
出处
《电力大数据》
2022年第5期28-35,共8页
Power Systems and Big Data
基金
国家自然科学基金资助项目(批准号:51777112)。
关键词
差分整合移动平均自回归模型
长短期记忆神经网络
时间序列
组合模型
电压偏差
auto regressive integrated moving average model
long-short term memory neural network
time series
combined model
voltage deviation
