基于BP神经网络的海口住宅室内气温预测

Forecast of residential indoor temperature based on BP neural network in Haikou

利用2014年1月—2015年12月海南海口居民住宅室内小气候观测数据和室外气象要素,基于BP神经网络构建海口住宅室内气温预测模型,通过与统计回归模型相比来综合评估模型的预测效果。结果表明:BP模型预测精度较高,不同季节预测精度存在差异,春秋季精度最高、夏季精度最差,其中地温加入对模型预测精度有较大改进。不同季节采用地温的BP模型对室内气温的预测值与观测值的均方根误差分别为0.25℃、0.62℃、0.26℃和0.52℃,平均绝对误差均小于0.5℃,即误差均在合理范围内。且预测精度(RMSE)较统计回归模型分别提高了26.5%、34.7%、56.7%和25.7%。可见该模型可以满足海口居民住宅室内气温的预测需求,可为室内居住环境、建筑能耗研究提供有效的基础数据。

By using indoor microclimate observation data in Haikou residential building and outdoor meteorological elements during the period from January in 2014 to December in 2015,the BP neural network forecast models were developed to predict the seasonal residential indoor temperature in Haikou,then,they were compared with the statistical regression models to evaluate the prediction effects comprehensively. The results show that,the BP models have higher prediction accuracy,there are differences in different seasons,the best result is in spring and autumn,while the worst result is in summer,and addition of ground temperature can improve a lot on the prediction precision. The root mean squared errors between the predicted and the observed residential indoor temperature in the different seasons are 0. 25 ℃,0. 62 ℃,0. 26 ℃ and 0. 52 ℃,the mean absolute error are both less than0. 5 ℃,that is,errors of the BP neural network forecast models are in reasonable range. What＇s more,prediction accuracy（ RMSE） increases by 26. 5%,34. 7%,56. 7% and 25. 7% than that of statistical regression models. It is obvious that the BP models can meet the forecast requirements for residential building indoor temperature in Haikou,which can provide effective basic data for the research of indoor living environment and building energy consumption.