Research on domestic air conditioners long-term performance and evaluation index.

Performance of practical operation domestic air conditioner (AC) is an important evaluation index to estimate energy saving efficiency. In order to investigate characteristic of air conditioners long term performance (LTP) and to establish optimation design method of high LTP in multi-factors impact conditions, BP neural network prediction method has been applied. The training sample of LTP prediction BP neural network acquired form experimental result of practical operation domestic ACs and data of ACs dynamic LTP on-line monitor system. By a large size of training sample, the decision weights of multi-impact factors and LTP optimation strategies can be obtained. In order to establish a LTP prediction model, the performances of 26 practical operations domestic ACs have been tested. And the high temperature cooling condition performance, rated cooling condition performance, low temperature heating condition performance, rated heating condition performance of 26 samples has been obtained. 85% of testing sample results was served as training sample data and 15% of testing data was served as validation data to LTP prediction BP neural network. The result indicated that the prediction of LTP prediction BP neural network is convergence and error is less than 5% during the BP neural network training by 22 samples. The decision weights of time weighted high temperature cooling, rated cooling, low temperature heating, rated heating normalized performance value are 0.187, 0.203, 0.312, 0.298, respectively. For further increasing the prediction precision, practical operation domestic AC performance online monitor system and LTP online data acquisition website has been established for data acquisition to validate LTP prediction BP neural network.

实际运行的家用空调（AC）性能是评估其节能效率的重要指标。为探究空调长期性能（LTP）的特性，并建立多因素影响条件下高LTP的优化设计方法，本研究采用了BP神经网络预测方法。LTP预测BP神经网络的训练样本来源于实际运行家用空调的实验结果，以及空调动态LTP在线监测系统的数据。通过大规模训练样本，可得到多影响因素的决策权重及LTP优化策略。为建立LTP预测模型，本研究测试了26台实际运行家用空调的性能，获取了这些样本在高温制冷工况、额定制冷工况、低温制热工况及额定制热工况下的性能数据。将85%的测试样本结果作为LTP预测BP神经网络的训练样本数据，15%作为验证数据。结果表明，使用22个样本训练BP神经网络时，LTP预测BP神经网络的预测结果收敛，且误差小于5%。时间加权的高温制冷、额定制冷、低温制热及额定制热归一化性能值的决策权重分别为0.187、0.203、0.312和0.298。为进一步提高预测精度，本研究建立了实际运行家用空调性能在线监测系统及LTP在线数据采集网站，用于数据采集以验证LTP预测BP神经网络。