부산대학교 도서관

This dissertation is a study on the air pollution prediction systems for the air quality management around port. There is growing interest in the health problems of air pollution, and according to the OECD report(2016) particulate matters were highlighted as major cause of premature deaths in South Korea. Regarding the degree of contribution to air pollution, it has been studied that the contribution concentration of domestic vessels to PM-2.5 based on 2016 standards is 0.57㎍/㎥, which is about 0.2% of the annual average 26㎍/㎥. By region, it has been reported that Busan has the highest contribution of 1.39㎍/㎥. The contribution of coal-fired power generation facilities to air pollution is 0.51㎍/㎥. Thus, the degree of contribution to air pollution by domestic vessels is similar to that by power generation facilities. It is necessary to develop an air quality prediction system that can predict air quality and prevent damage as well as emission reduction policies to prevent damage to health and property due to fine dust around ports. Therefore, this research tried to develop a hybrid model system that predicts air quality by applying machine learning and numerical models. First, as a result of predicting the air quality around Busan Port using a numerical model, the IOA of daily forecast result was 0.72 at Busan North Port and 0.63 at Busan New Port. The IOA of the hourly forecast results were 0.68 and 0.61, respectively. In addition, the RMSE results of the daily forecast results were analyzed to be 9.7㎍/㎥ at Busan North Port and 12.2㎍/㎥ at Busan New Port, and the RMSE results at the hourly forecast were analyzed to be 13.6 ㎍/㎥ and 16.1㎍/㎥, respectively. It was shown that the statistical value of the hourly forecast result is low. Second, RNN-LSTM was applied to a machine learning model in order to predict the air quality around Busan Port. The optimum conditions for predicting the air quality of Busan North Port are predicted to be IOA 0.97451 and an RMSE of 4.88㎍/㎥ as a result of applying observation data of air quality, weather and the number of vessels of 2,000 tons or more. The optimum conditions for Busan New Port were predicted to be IOA 0.96998 and RMSE 5.87㎍/㎥ based on the results predicted using air quality observation data, weather and the number of berths of all vessels at Busan New Port. The results predicted using machine learning were shown to be superior in performance to the results predicted using a numerical model. Third, a hybrid model was developed to utilize the advantages of spatial prediction of numerical models and the strength of superior prediction performance of machine learning. To verify the hybrid model results, three observation points were selected each around Busan North Port and Busan New Port. The IOA of the numerical model prediction results of the points around Busan North Port was analyzed to be 0.68, and the RMSE was 12.27 ~ 13.56㎍/㎥. However, the IOA of the hybrid model prediction results was 0.94 ~ 0.97, and the RMSE was 5.29 ~ 6.81㎍/㎥. The IOA of the numerical model prediction results around Busan New Port was between 0.59 and 0.68, and the RMSE was between 12.56 and 19.28㎍/㎥, whereas the IOA of the prediction result of the hybrid model was between 0.89 and 0.92, and RMSE between 7.74 and 10.87㎍/㎥. When the hybrid model was applied, the prediction performance was excellent. Finally, a conceptual design was performed to establish an air quality model system. An input data process system was designed and constructed for the air quality prediction. The numerical model system components were designed and operated using the meteorological model WRF, the emission model SMOKE, and the transport and diffusion model CMAQ. The machine learning model is designed to use the RNN-LSTM model. Hybrid model constituted an interface that can use numerical model results and machine learning model results, and designed a data assimilation method. If a prediction service for the air quality around Busan Port is established, it is expected that the limit point of monitoring only the observation data of the air quality can be overcome.