A novel interval prediction method with adaptive calibration for ship energy consumption prediction

Area of Science:

• Maritime Engineering
• Data Science
• Energy Management

Background:

• Accurate Ship Energy Consumption (SEC) prediction is crucial for operational efficiency and abnormality detection.
• Existing methods often provide deterministic SEC values, failing to capture uncertainties in complex maritime environments.

Purpose of the Study:

• To develop a novel interval prediction method for SEC that quantifies uncertainty.
• To improve energy efficiency management and abnormality detection in maritime operations.

Main Methods:

• Proposed Gradient-resolved Adaptive Calibration for Energy consumption (GrACE) method.
• Formulated SEC interval prediction as a multi-objective learning problem (interval quality and width).
• Implemented adaptive calibration for aggregated prediction intervals and validated on five neural networks.

Main Results:

• GrACE consistently achieved superior interval quality across validated neural networks.
• Achieved a Prediction Interval Coverage Probability of 0.8976 at the 90% confidence level using a Convolutional Neural Network.
• Reduced Mean Prediction Interval Width (MPIW) by 56.55% compared to the classical LUBE method.

Conclusions:

• GrACE effectively provides reliable uncertainty quantification for SEC.
• The method offers valuable insights for energy efficiency and abnormality detection in complex operating conditions.
• GrACE enhances the reliability of SEC predictions by incorporating uncertainty.