基于最优加权组合模型的煤炭消费预测分析

Coal consumption forecasting based on optimum weighted composition model

为了研究最优的煤炭消费预测模型,为我国能源结构优化提供依据,基于差分自回归移动平均(ARIMA)、灰色预测(GM)和人工神经网络(ANN)模型构建了8个组合预测模型,对我国煤炭消费量进行预测分析,应用评价指标R、MAE、MAPE和RMSE对预测模型精度进行比较,筛选出最优组合模型并预测分析未来10年我国煤炭消费趋势。研究结果表明:(1)最优加权组合模型均方根误差、平均绝对误差、平均相对误差等参数均较小,预测效果明显优于单项和简单组合预测模型;(2)构建了权重为(0.73,0.09,0.18)的我国煤炭消费预测最优加权组合模型ARIMA-GM-ANN。(3)将煤炭消费增长趋势分为"缓慢上升期"、"急速增长期"、"下降期"和"平稳期"四个阶段,2013年煤炭消费量达峰,约43.14亿t,2020年以后,煤炭消费量稳定在35.5亿t左右。

To provide the basis for optimization of China’s energy structure,coal consumption forecasting model is studied.Based on ARIMA,GM and ANN,3 single forecasting models and 8 compound forecasting models are built to forecast the coal consumption of China. To forecast and analyze the development trend of China’ s coal consumption in next 10 years,the optimal model would be selected through parameter evaluation of R,MAE,MAPE and RMSE. The results show that the parameters MAE,MAPE and RMSE of optimal combination weighting model are smaller,and the prediction effect is obviously better than single prediction and simple combination model. The weight of optimum weighted composition model ARIMA-GM-ANN for China’ s coal consumption is（ 0. 73,0. 09,0. 18）. The growth trend of coal consumption is divided into three stages： ＂slow rise＂, ＂rapid growth period＂, ＂declining period＂and ＂stable period ＂. Coal consumption peaked at about 4. 314 billion tons in 2013. After 2020,coal consumption stabilized at about 3. 55 billion tons.