| dc.description.abstract | Measurements from on-board monitoring systems of vessels are now increasingly available for data analysis, and their relevance is growing as the ship industry enters the digitilization phase. The analysis of data from vessels (in operation) can be used to verify the power system design in general and improve the electrical power load analysis in particular. The electrical power load analysis is a standard procedure used for the dimensioning of vessel power systems and is mainly based on the electrical load lists. The power demand is determined by assigning a load factor to each electrical consumer according to the experience from similar ships. This method has limitations, because the output is only a single number and the range of possible combinations of loads remains unknown. Therefore, making correct assumptions regarding a range of expected loads enables more appropriate sizing of the power plant. The assumptions about operational profile and the real power load range used in various operational modes are thus crucial.
Data mining approach can help to overcome challenges in vessel power system design and might help to understand the factors that influence power requirements; it can also be applied for more precise dimensioning of the power plant characterized by less margins. Designers of a vessel's power systems can benefit from data mining to enhance their knowledge of the power systems and to validate ship power systems and their personal experience by high-level data analysis and advanced statistical modeling with an aim to improve power system design. The analysis of power generated signals can enable to understand the properties of a power system, and it can allow to dimension more efficient, hybrid power systems, where the power source is chosen using a complementary approach based on data mining. On the basis of the data, statistical models of power requirement and influential power requirement variables can be developed. The biggest advantage of statistical models is that they allow to extract and visualize the relationships between input and output variables; these relationships can then be applied to understand factors that influence power consumption, improve some of the low-fidelity simulations, and perform more precise dimensioning of the power plant with less margins according to real power requirement.
Highlights of the thesis include analysis of real measurements from a platform supply vessel (PSV) with diesel {electric configuration and are as follows :
• The crucial and fundamental step of data mining cycle is data preprocessing, which is challenging because many distortions appear in real measurements. Before the signals are analyzed, they should be synchronized and preprocessed without introducing the delay. In this research, the discrete Fourier transformation (DFT) digital filtering is proposed as a tool for efficient sensor data preprocessing, which does not introduce a time delay and has low numerical cost. This makes the method appropriate for big data preprocessing.
• Because the power demand varies significantly in vessel operational tasks, splitting the data between vessel operations for the analysis of vessel power system is a critical step. Investigation of the probability distribution function of power generated/consumed allows to estimate the most frequent values, the maximal and minimal values of power, as well as let to study design margins, and to validate power system design in each operation.
• The rarely occurring high power demands are critical for power system design and optimization and can be studied by the real power range analysis. Therefore, an advanced methodology to quantify variability in the generated power is proposed, which explains the tails of probability distributions of a power signal based on signal decom- position. The proposed methodology can facilitate the selection of the optimal size, number, and configuration of generators or batteries when designing new power systems.
• Key factors that influence power demand and quantification of their contribution are studied using an appropriately chosen statistical model. The generalized additive model (GAM), which allows to model nonlinearities, was developed to determine the relationship between power consumed and the key influential factors for a power system. The importance of feature extraction for statistical modeling is described on the basis of Hilbert Transform to improve the model. The results of the data-driven model are verified with a simulation model to show that the former is within the boundaries of power requirements from simulations. | nb_NO |
