一种基于多元特征循环神经网络的建筑冷负荷软测量方法

MULTI-FEATURE RECURRENT NEURAL NETWORK FOR BUILDING COOLING LOAD SOFT SENSING

准确的冷负荷计算是采暖通风空调系统实时调控的基础,对建筑节能具有重大意义。软测量是实现冷负荷在线实时估计的方法,为提高软测量模型的预测性能,基于集成算法的思想提出多元特征循环神经网络,集成五种机器学习算法。为验证模型预测性能,在三个来自不同类型建筑的建筑能耗仿真数据集进行对比实验。实验结果表明,提出的模型对比其基模型预测性能大幅提升,与其他集成模型和时间序列预测模型相比评价指标上表现更好,并具有良好的预测稳定性。

Real-time control on HVAC system,of great significance to building conservation,is based on accurate cooling load real-time calculation,which can be realized by soft sensing.In order to improve the prediction performance of soft sensing model,a multi-feature recurrent neural network based on Stacking algorithm is proposed,which integrates five machine learning algorithms.In order to verify the prediction performance of the model,comparative experiments were carried out on three building energy simulation data sets from different types of buildings.The experimental results show that the prediction performance of the proposed model is greatly improved compared with the base models,and the MAE and RMSE are the best compared with other integrated models and time series forecasting models,and have good prediction stability.