APM₁₀ Concentration Level Prediction Model Based On Continuous HMM For Lan Zhou City

With the rapid growth of economy in our country, the constant improvements of industrialization and urbanization, environmental pollution is becoming more and more serious. The "fog" has become the focus of the national people’s topic to talk, it crisis with the normal life of people, is harmful to people’s health. In recent years, across the country are taking steps to control the fog haze, curb the further deterioration of air pollution. Investigate its fundamental, fog in the real harm to people’s health is suspended particles in the air PM10 and PM2.5.Therefore, in order to reduce air pollution on human health harm, it is real meaningful to detect suspended particulate matter(PM10 and PM2.5) concentration effectively and control it reasonably, and analyze existing data, design the appropriate model to predict the concentration of PM10 to play a preventive role. As we all know, air pollution in Lanzhou city in th e country is very serious, and the high concentration of PM10 is one of the important factors that lead to air pollution. In this paper, we analyzed the relationship between the concentration of PM10 in Lanzhou City and its correlative meteorological condition, continuous meteorological conditions: temperature, relative humidity, wind speed and PM10 concentrations of the previous day as observation variables, the PM10 concentration level division as the hidden states, to establish continuous observation variables of hidden Markov model, on this basis, a 24-hour advance forecast model based on continuous observation condition hidden Markov model. For the goal, to achieve the 24-hour advance forecast Lanzhou summer PM10 concentrations purposes.In the specific experiments, We select 2007--2010 summer meteorological data and PM10 pollution data in Lan Zhou city as the training set, and 2011 summer meteorological data and PM10 pollution data in Lan Zhou city as the test set, establish and validate the 24 hours in advance to predictive model. The predictive value of the experiment compared with the predicted value of three classic prediction algorithms. Experimental results show that, the prediction effect of our model is little than other models, it can play an effective role to predict the vast majority concentration level of PM10.