摘要
建筑能耗短期预测对实时性要求较高,传统神经网络存在收敛速度慢的缺点。为此,采用LM算法改进标准BP神经网络,建立了基于LM算法的建筑能耗预测模型。首先通过理论说明该算法的先进性,然后设计一套建筑能耗数据采集系统和建立基于LMBP神经网络的建筑能耗预测模型,最后采集某建筑一个月的整点电量作为预测模型的实验数据。实验结果表明,该模型明显提高了训练速度,且预测精度满足实际需求,说明了LMBP神经网络适用于建筑能耗短期预测。
The traditional neural network is too slow in term of convergence speed to meet the high real-time requirements of short-term prediction of building energy consumption. Therefore, LM algorithm is adopted instead of conventional BP algorithm to establish the building energy consumption model. Firstly through theoretical description of the advanced algorithm,then design a set of data acquisition system to monitor building energy consumption and set up the prediction model based on LMBP neural network. Finally a building’ s 24-hour power consumption data for one month is collected by the data acquisition system as the experimental samples to verify the model. Empirical results show that the LMBP neural network prediction model significantly improves the training speed,precisely enough to meet the actual demand. Thus,the model is adequate for short-term prediction of building energy consumption.
出处
《计算机技术与发展》
2015年第6期216-218,223,共4页
Computer Technology and Development
基金
上海市教委学科专业建设资助项目(XKCZ1212)
关键词
建筑能耗
数据采集
短期预测
神经网络
BP算法
LM算法
building energy consumption
data acquisition
short-term prediction
neural network
Back Propagation algorithm
Levenberg-Marquardts algorithm