A Study On NO_x Emission Prediction From Ship Main Engines Based On Multi-source Data

With the development of the global economy,the number of ships and the volume of transportation are increasing,and the resulting ship exhaust emissions are of great concern to the international community.Among them,nitrogen oxides(NO_x)not only threaten human health and the ecological environment,but also contribute to global climate change.In order to reduce NO_x emission from ships,the International Maritime Organization has developed relevant policies.Therefore,accurate prediction of ship NO_x emission under different sailing status and weather conditions is important for understanding emission patterns and establishing relevant emission reduction policies.However,current classical ship emission estimation methods fail to consider the influence of weather conditions on ship emissions;in addition,data quality and variety can also limit the accuracy of prediction models.Therefore,based on multiple sources of data such as NO_x monitoring data,Automatic Identification System(AIS)data and sailing weather data,this study uses classical methods and machine learning methods to construct NO_x emission prediction models for ship main engines,respectively.The main contents of this study are as follows:(1)Aiming at the problem that single data cannot provide comprehensive and complete information,this study collects NO_x emission data,AIS data and sailing weather data,and preprocesses the data,including outlier cleaning,interpolation of missing values from AIS data,and temporal and spatial expansion of weather data.Then,based on the ship sailing time and geographical location provided by AIS data,weather information is retrieved from weather data。And these data about ship sailing and weather conditions are fused with NO_x emission monitoring data to provide data support for building NO_x emission prediction models.(2)Aiming at the problem that the current classical estimation methods ignore the influence of sailing weather conditions,this study uses the real-time wind and wave current information in the constructed dataset to correct the speed-on-ground(SOG)by AIS data to obtain the ship’s static water speed(STW).Based on this,the ship main engine load optimization prediction model is established,and the ship fuel consumption and NO_x emission are assessed based on the fuel consumption rate and NO_x emission factor,and the accuracy of the model is verified by comparing with the actual measured data and other classical methods.(3)Aiming at the problems that assumptions of conditions in classical estimation methods and the influence of coupled multiple factors,this study introduces machine learning to predict NO_x emission,including multiple linear regression,minimum absolute shrinkage and selection operator regression,support vector machines,random forests(RF),decision trees,and BP neural networks.Then,based on the RF prediction model with good prediction performance,the Mean Decrease Impurity method and the Recursive Feature Elimination method are used to conduct feature importance analysis and screening.And the SHAP method is used to explain the influence of feature variables on the prediction model.The results show that the performance of the prediction model is effectively improved after feature screening,while reducing the model running time.In addition,the results of the SHAP analysis showed that weather factors both showed different patterns of influence on the output prediction of the ballast and full load NO_x emission prediction models,except for the positive influence of SOG on the output prediction of the two models.