基于修正组合模型的包头市用水量预测分析

Prediction of water consumption in Baotou based on amended combination model

【目的】建立基于马尔科夫链修正的组合灰色神经网络模型,为城市用水量的准确预测提供支撑。【方法】在运用灰色关联性分析法确定用水量变化的主要影响因素的基础上,建立了基于马尔科夫链修正的组合灰色神经网络模型,将该模型应用于包头市2009和2010年的用水量预测,并将其预测结果与灰色模型、BP神经网络模型和组合灰色神经网络模型的预测结果进行比较。【结果】包头市用水量受人口、国内生产总值、工业总产值、建成区绿化覆盖率、耕地面积及工业用水重复利用率的影响较大,利用建立的基于马尔科夫链修正的组合灰色神经网络模型对包头市2009和2010年用水量进行预测,并与实际用水量进行比较后表明,其相对误差分别为0.16%和2.16%,均方根相对误差为1.53%,而灰色模型、BP神经网络模型和组合灰色神经网络模型的均方根相对误差分别为4.34%,3.08%和1.99%。可见,基于马尔科夫链修正的组合灰色神经网络模型的预测效果最好。【结论】基于马尔科夫链修正的组合灰色神经网络模型结合了各模型的优势,预测精度较高。

Abstract. [Objective] This study established a combination model of Grey model and BP neural net- work revised by Markov chain to better forecast water consumption. [Method] Main factors influencing change of water consumption in Baotou were analyzed by gray correlation analysis, and a combination model of Grey model and BP neural network revised by Markov chain was established. The model was used to forecast water consumption in 2009 and 2010 in Baotou,and the model prediction was compared with re- sults of Grey model,BP neural network and combined Grey neural network. [Result] Water consumption in Baotou was influenced by population, GDP, total industrial output value, built-up area greenbelt cover rate, cultivated area,and industrial water recycling rate. Comparison of water consumption in 2009 and 2010 of Baotou predicted by the established combination model of Grey model and BP neural network revised by Markov chain and actual water consumption indicated that the relative errors were 0.16% and 2.16% re- spectively,and the relative error of root mean square was 1.53%. The relative errors of root mean square for Grey model,BP neural network and combined Grey neural network were 4.34%,3.08~ and 1.99%, respectively. The combination model of Grey and BP neural network revised by Markov chain was the best.[Conclusion] Combined Grey-neural network model revised by Markov chain had high prediction accuracy.