基于季节ARIMA和指数平滑模型的我国全社会总用电量的预测

Prediction of Total Electricity Consumption in China Based on Seasonal ARIMA and Exponential Smoothing Model

摘要: 本文基于2010年1月份~2019年12月份我国社会总用电量季度数据，采用Rstudio软件进行分析，对数据进行预处理，通过比较AIC信息准则拟合最优的ARIMA模型，以2010年1月份~2018年12月份的数据作为训练集，2019年的数据作为测试集，对该序列进行1阶12步差分后，序列变的平稳，因此可采用季节ARIMA模型进行预测；由于该序列具有趋势性和季节性的特征，因此采用Holt-Winters三参数指数平滑模型，应用两种模型分别对2019年的数据进行预测。通过测试集和预测值计算误差，根据平均误差最小原则选择最优的预测模型。最终的平均误差结果显示Holt-Winters三参数指数平滑模型的平均误差值为0.0232087，远小于季节ARIMA模型的0.0315013，因此选用Holt-Winters三参数指数平滑模型作为我国全社会总用电量的预测模型。

Abstract: Based on the quarterly data of China’s total social electricity consumption from January 2010 to December 2019, this paper uses Rstudio software to analyze and preprocess the data, fits the optimal ARIMA model by comparing AIC information criteria, takes the data from January 2010 to December 2018 as the training set and the data from 2019 as the test set, and makes a first-order 12 step difference for the sequence. The series becomes stable, so seasonal ARIMA model can be used for prediction; because the series has the characteristics of trend and seasonality, Holt winters three parameter exponential smoothing model is used to predict the data in 2019. The error is calculated through the test set and prediction value, and the optimal prediction model is selected according to the principle of minimum average error. The final average error results show that the average error of Holt winters three parameter exponential smoothing model is 0.0232087, which is far less than 0.0315013 of seasonal ARIMA model. Therefore, Holt winters three parameter exponential smoothing model is selected as the prediction model of total power consumption in China.