基于LM-BP神经网络的建筑能耗在线预测

Online Forecast of Building Energy Consumption Based on LM BP Neural Network

由于神经网络不需要建立复杂的数学模型,因此基于BP神经网络的建筑能耗预测引起广泛关注.但标准BP神经网络收敛速度慢,不适于建筑能耗在线预测,因此采用了LevenbergMarquardts算法对标准BP神经网络加以改进,并将其应用于某建筑未来24小时的整点电量负荷预测.实验结果表明,改进的神经网络明显提高了训练速度,为建筑短期负荷的在线预测提供了一种方法.

Since neural networks do not require establish complex mathematical models, BP neural network has been paid widespread attention in building energy consumption prediction. However, the convergence rate of standard BP algorithm is so slow that is not suitable to predict building energy consumption online. To solve this problem, the Levenberg-Marquardts algorithm was introduced to improve the standard BP neural network,and then applied it to a building power load forecast in next 24 hours. Experimental results show that the improved neural network obviously improves the training speed and put a method to predict short-term building energy consumption online.