Estimation of operational gaseous emissions in ship life cycle with machine learning method

The maritime industry has been one of the most important landmarks in the development of human civilization since the beginning of human history. The vessels have been using the natural forces such as the referral system for a long time; at first, they have begun to use coal with the industrial revolution and later on the petroleum and its derivatives as the technology develops. These types of fuels, called fossil fuels, contain very high levels of carbon. The burning of these fuels by the internal combustion process of modern engine systems which provide energy production and therefore the propulsion of the ship; result in carbon dioxide, carbon monoxide, methane, sulfur oxides, nitrogen oxides and nitrous oxide, particulate matter and black carbonand non-methane volatile organic compounds. It is known that there are also a wide variety of heavy metal fumes in the combustion process in addition to aforementioned gases. All these gaseous emissions are wastes with very deleterious impacts to human health and the environment. The damage to the environment can be summarized as global warming, acid rain and ground-level ozone whereas the harm to human health can be summarized as circulatory and respiratory system disorders. For these reasons, it is extremely important to estimate, reduce and control ship emissions. In this study, firstly total emission amounts were calculated from the daily noon report of nine bulk carriers, and then these data were put into regression analysis. The relationship between the main characteristics of the ship, which are the deadweight tonnage and the block coefficient, was assessed and formulas based on the deadweight tonnage and blok coefficient values were developed in order to estimate future potential emissions during the ship's pre-design phase. For these formulas are based upon only two constants, the results are not accepted as reliable. Thus, in the second phase, by using voyage time, engine revolutions per minute, speed, displacement, weather condition, sea condition and average draft data as input and total emission data as output, it was aimed to find out which neural network model predicts the emission amount closer to the real data. Neural networks is accepted as a reliable method, which can be used to calculate the impacts of dynamic operation conditions on emissions. The difference between the real and calculated data is approximately 1.57 %. In the third phase, the best result was applied to two different routes, north and south, during january and june in Atlantic and Pacific Oceans, in order to estimate the emission amounts. The routes have different voyage time, sea and weather conditions. Then, the total fuel consumption data was calculated by using the emission amount and a fuel cost analysis was realized. In this context, it was clearly seen that there is a correlation between weather and sea conditions and emission amounts. It is concluded that less emission is occurred in the south route, which is calmer weather and sea conditions, in both january and june. For the distance between North and South Atlantic routes is small, South Atlantic route provided more eligible results in terms of both environmental and economic performances. The distance between North and South Pacific routes is very much, thus, although the emission per distance is lesser in south route, north route provided better results in a holistic perspective. Decision support system will help to estimate the total emission amounts on estimated routes while the ships are still in the preliminary design phase. Thus, once the routes and main characteristics of the ship are determined, the emission amounts, fuel consumption and fuel costs can be predicted at an early stage of a ship's life cycle.