室内工作面天然采光照度分布实时预测方法

照射进室内的天然光随时间变化往往是非线性的,在空间上的分布也往往是不均匀的,随着智慧照明技术的快速发展,准确舒适的调光决策依赖于可靠的天然采光照度分布基础数据收集方法。本文采用两种机器学习算法:随机森林(Random Forest)和BP反向传播神经网络(Back Propagation Neural Network),以四种主要输入特征(传感器照度信息、天空亮度信息、时间信息、其他信息(房间尺寸、窗地比、室内平均采光系数、建筑朝向))在实测方法下对室内工作面照度分布进行了实时预测。结果表明,随机森林模型在测试集中的回归决定系数R2为0.826;BP反向传播神经网络模型在测试集中的回归决定系数R2为0.739,随机森林的表现相对较好。两种机器学习算法在室内照度分布预测方面具备发展潜力,在未来建筑的照明智慧化调光及间接节能方面具有正向促进作用。

The Real-time Prediction Method of Daylighting Illuminance Distribution of Working Plane

With the rapid development of intelligent lighting technology,accurate and comfortable dimming decision-making depends on reliable basic data collection methods of natural daylighting illuminance distribution.Two machine learning algorithms,Random Forest and BP Back Propagation Neural Network,are used to predict daylighting illuminance distribution of working plane.Input features include four main parameters(sensor illumination information,sky brightness information,time information,and other information(room size,window to ground ratio,indoor average daylighting coefficient,window direction)).Under the conditions of actual measurement,the illuminance distribution of working plane is predicted in real time.The results show that under the experimental conditions,the regression coefficient R² of random forest model in the test set is 0.826;The regression coefficient R² of BP back propagation neural network model in test set is 0.739.Under the condition of real measurement,the performance of random forest is relatively good.The two machine learning algorithms have development potential in the prediction of indoor illumination distribution,and play a positive role in intelligent dimming and indirect energy saving of future building lighting.