The yachting industry is embracing sustainable practices in response to global environmental concerns, particularly those outlined in the Paris Agreement, which aim to reduce greenhouse gas emissions and limit global temperature increases. Although diesel engines are most commonly used in yachts and are a major contributor to greenhouse gas emissions, efforts are being made to minimise energy consumption through the use of alternative fuels. Auxiliary systems, such as heating, ventilation and air conditioning (HVAC), are also being studied for efficiency improvements. As yachts spend a considerable amount of time at anchor or in port, these systems account for a significant percentage of the total consumption.

This thesis focuses on estimating the energy demand of HVAC systems onboard yachts. In collaboration with De Voogt Naval Architects, part of Feadship, this research uses a grey box model approach, a combination of a white box and a black box, to estimate and evaluate HVAC systems. White box models are based on first principles or physics and are transparent models. This contrasts with black box models, which model a system based solely on observed data, without any prior knowledge of the system. These models are primarily machine learning algorithms such as artificial neural networks. Black box models perform better than white box models within the range of the trained data, but white box models excel at prediction outside this range. The integration of both models can combine the advantages of both into a grey box model.

The white box model described in this report is a type of predictive model that combines theoretical knowledge with empirical data to estimate the heat load for HVAC systems on yachts. The model uses theoretical principles derived from various methods, such as an ISO heat load balance, and incorporates empirical data collected from sensors installed on yachts, including temperature and humidity measurements. This data is used to simulate the performance of the HVAC system under different weather conditions.

The heat load estimation of the white box model is used as input for the greybox model combined with various sensor data. It makes use of a perceptron artificial neural network (ANN), that can learn from the data and adjust its predictions. The hyperparameters of the ANN are chosen and validated using a kfold cross validation. The final calculations are performed with the optimal configuration.

The results of the optimal model are validated against the power consumption recorded in the voyage data. The grey box model achieves a mean absolute percentage error (MAPE) accuracy of 91.3%, which is an improvement compared to the performance of the solo black box. Specifically, the MAPE accuracy of the grey box model is 0.9% higher along with similar improvements in other metrics.

Finally, the study provides an insight on the possible application of the grey box model. The solo white and black box models can also be used apart from the grey box model. A simple extrapolation analysis is conducted to provide insights on the capabilities of the model. With the implementation of these models, Feadship can predict the heat load and energy demand for varying weather conditions and ranges of data.