Research On The Characteristics And Forecast Method Of Humidity In The Mogao Grottoes

Dunhuang is located in the desert arid region,the annual evaporation is thousands times the capacity of the rainfall.Because of the natural climate and the sparsely populated characteristic,the Dunhuang Mogao Grottoes has been preserved so far.In recent years,the development of the tourism industry has caused the originally stable micro-environment of the cave to be destroyed,thus affecting the preservation of cultural relics and paintings in the cave.According to the environmental monitoring data,it is time to analyze the mechanism of the disease and conclude a corresponding set of prevention and control measures,which is the only way to protect the historic sites and realize the sustainable development of the scenic spots.High humidity in the cave will produce water and salt chemical reactions in the murals,forming the salt cream,and then cause the mural surface from flaking,cracking.In this paper,aiming at the change of humidity which is one of the cave micro-environment factors,we hopes to grasp the change characteristics of the humidity in the cave and simulate its changing trend,so as to achieve the purpose of early prediction and prevention in advance.Thus we can provide the cave micro-environmental control with important technical support.In this paper,we analyze the humidity environment characteristics in the cave by comparing the annual variation of humidity inside and outside the cave,the change of humidity in the rainy season,the influence of the tourists' flow on the humidity in the cave.On this basis,we propose two humidity of Mogao Grottoes prediction models based on air exchange rate and recurrent neural network respectively.The former based on the physical conservation of mass of the air exchange inside and outside the house,predict the absolute humidity through calculating the exchange rate of air inside and outside the cave.Due to its physical characteristics,this model is deficient in practicability and long-term prediction.In order to solve this problem,based on historical monitoring data,a humidity prediction model based on recurrent neural network(LSTM)is proposed,which includes network structure design,network training and algorithm of forecasting process.The experimental results show that the proposed LSTM prediction model is more practicable and accurate in humidity time series analysis by comparing with the prediction model based on air exchange rate.