| The acceleration of urban process has led to the emergence of light pollution at night,and light monitoring has been incorporated into the environmental governance system of smart city.The existing research on light environment mostly focuses on expressing the characteristics of light distribution in the spatial dimension,and lacks the research on exploring the characteristics of light evolution from the dual dimensions of time and space.The maturity of digital and machine learning tools provides theoretical and technical support for the research on space-time distribution and accurate dynamic monitoring of light environment.Therefore,taking the light environment of Dalian as the research object,this paper introduces the prediction concept into the urban light monitoring system.The temporal and spatial evolution characteristics of lighting in Dalian are studied by using long-time series remote sensing images and multi parameter ground measured data;The optimal machine learning prediction model is determined through the error evaluation system,the light prediction program based on the optimal model is constructed,and the application value of the prediction program in the urban light monitoring system is explored.The specific conclusions are as follows:(1)Based on remote sensing and measured long-time series light data,the temporal and spatial distribution characteristics of light in Dalian are analyzed.Firstly,the univariate linear regression method is used to evaluate the change of lighting time series in China for the night light remote sensing images from 1992 to 2020.The average annual change rate of light in Dalian in 29 years is 35.5%;Then,seven representative lighting areas in Dalian are selected to conduct quantitative and visual analysis on the measured and new measured data of existing lighting from 2016 to 2021,so as to control the multi-scale temporal and spatial distribution of lighting in Dalian.(2)Focus on analyzing the influencing factors of lighting from the perspective of time and space.Using the long-term continuous monitoring light data,the correlation analysis method is used to determine the action mechanism of lunar phase,cloud layer and air quality on light.It is found that these factors are negatively correlated with light;At the same time,there is also a variation law of light every night and different seasons,so as to provide variable factors for the following prediction research.(3)Build a multivariable light prediction model based on machine learning.The multivariate time series data are substituted into ARIMA and LSTM prediction models respectively.After the evaluation of error evaluation index,the generalization and fitting ability of the latter are better than the former;Therefore,the light prediction program based on LSTM model is constructed in MATLAB to provide a technical tool for the data analysis of longperiod light dynamic monitoring.(4)Explore the application value of urban light prediction program.The urban lighting monitoring system is divided into urban surface layer and urban skyline layer.The former uses intelligent communication system and the latter uses prediction program.The program regularly feeds back to the communication according to the monitoring results to realize realtime intelligent regulation,so as to build an accurate and dynamic monitoring system of urban lighting.To sum up,this paper uses numerical models,neural network algorithms,machine learning and other technical means to provide a new research perspective for light environment research,as well as theoretical and technical methods for urban light monitoring. |
