Application Research Of Machine Learning In Indoor Intelligent Lighting Engineering

As an important part of the building,the lighting system is rapidly developing with the rise of intelligent buildings.Illumination calculation,which can be used for lighting system design,indoor illumination calibration and luminaire control,is a necessary part of lighting system engineering.In traditional illuminance calculation method,the utilization factor method is complicated because of the process of searching tables,which brings inconvenience to the design and control of the lighting system.As a research hotspot in artificial intelligence,machine learning is good at solving classification and regression problems,and has been widely used in lighting engineering.In this paper,the machine learning methods were applied to the lighting engineering.To simplify the calculation process of the luminaire utilization factor while the calculation accuracy is guaranteed,machine learning methods were used to improve the utilization factor method.And an intelligent indoor lighting control system was designed based on the clustering algorithm and the improved illuminance calculation method.Intelligent lighting control system.The main research points include the following three parts:(1)In view of the complexity and inaccuracy of searching table method of calculating luminaire utilization factor in traditional luminaire calculation method,the utilization coefficient table of the luminaire was fitted by machine learning models to improve luminaire calculation method.Utilization factor calculation model based on neural network,the utilization factor calculation model based on support vector machine and the utilization factor calculation model based on random forest were established respectively while using utilization factor data of the YG1-1 fluorescent tube as dataset.The cross-validation and control variable method were used to adjust the hyperparameter of those models.Comparative experiment about the accuracy were conducted between the three models to find the optimal model.By using the trained model to calculate the luminaire utilization factor,the calculation process can be simplified and calculation accuracy can be improved.(2)To reduce the relative error rate of calculation results and improve the convergence speed,the utilization factor calculation model based on neural network was optimized.Optimization strategies include weight initialization method optimization,update algorithm optimization and network structure optimization.In weight initialization method optimization,it is proposed to use the particle swarm optimization algorithm to find the initial weight and experiment results were compared with other initialization methods.In update algorithm optimization,the relative error rates of various gradient descent training algorithms are compared and analyzed.In network structure optimization,it is proposed to use the two neural networks to calculate the utilization factor when the effective floor reflectance is 0.2 and the utilization factor correction coefficient when the effective floor reflectance is not 0.2,so as to make full use of all the data in the luminaire utilization factor table.Finally,Comparative experiment were conducted to verify effectiveness of the three optimization strategies on the utilization factor calculation model based on neural network.(3)According to the fact that the traditional lighting system can not control the lighting area in real time and intelligently,resulting in waste of electric energy,an intelligent lighting control system based on improved DBSCAN algorithm is designed.The position information of the indoor personnel is obtained by sensor;the position information is converted into two-dimensional coordinate data on the illumination plane and the improved DBSCAN clustering algorithm is used to cluster the data to find a relatively densely distributed area;The illuminance calculation based on utilization factor method is performed on each area to calculate and the number of lamps to be turned or the luminous flux each lamp needs to emit.In the illuminance calculation process,the neural network model proposed in(2)is used to calculate the utilization lamp,which can improve the calculation accuracy.Energy saving can be achieved through real-time,intelligent control of the lighting area.