Machine learning for shaft power prediction and analysis of fouling related performance deterioration

New Insights into the Propulsion Power Prediction of Cruise Ships

“…From an operational perspective, measures such as speed and trim optimization, best route prediction, or regular hull maintenance can lead to important energy savings. With respect to the latter point, hull fouling -i.e., accumulation of marine growth on the hull -could indeed lead to an increasing performance deterioration and hence an increase in energy consumption [4].…”

“…Finally, Laurie et al [4] tackle the subject of propulsion power prediction using ML models. Their study evaluates various ML models, trained with sensor data (10 second sampling rate) and weather data.…”

“…Step 2: split + cross-validation + test An indication of the existence of this cross-talk issue is the performance of the k-NN algorithm in evaluation papers [15], [4]. In particular, Laurie et al point out that for any data point in the validation/test sets, there exist very close analogs in the train set, when data is sampled at a high frequency [4].…”

“…Step 2: split + cross-validation + test An indication of the existence of this cross-talk issue is the performance of the k-NN algorithm in evaluation papers [15], [4]. In particular, Laurie et al point out that for any data point in the validation/test sets, there exist very close analogs in the train set, when data is sampled at a high frequency [4]. Figure 2 2) Overlapping windows: The problem of cross-talk is further exacerbated when overlapping windows are used for aggregation, such as in the case of simple moving averages [7].…”

“…3) Parsimony: the literature review exhibits already a wide range of ML models, from simple (e.g., multiple linear regression) to complex (e.g., recurrent neural networks) approaches. Laurie et al [4] suggest that parsimonious models such as k-NN can outperform most refined techniques for the prediction task. Therefore, one gap is related to a rigorous methodology to compare models' complexity, for instance by comparing small vs. large architectures within the same models family.…”