Decentralized Data-Driven Analytical Framework for Ship Fuel Oil Consumption.

Abstract

In the context of maritime operations, the advent of advanced technologies has paved the way for predictive analytics to optimize energy consumption. In this research, we introduce an AI and blockchain-assisted intelligent and secure framework for predicting energy consumption in ships to enhance efficiency and sustainability. In this context, we used a standard energy consumption dataset, which comprises Co2 emissions and energy consumption features; therefore, we first employed a regression model that predicted Co2 emissions in ships. Based on the prediction, we create the target labels in the dataset, i.e., ship with poor engine (1) and ship with good engine (0). Subsequently, we ap- plied decentralized training on the dataset using federated learning (FL) for the binary classification problem. For that, we utilized an artificial neural network (ANN) in FL that efficiently categorized the ships based on their energy consumption features. Fur- thermore, we considered a tamper-proof technology, i.e., blockchain technology, that confronts data tampering attacks on FL-trained weights. In that context, we developed a smart contract that ensures valid FL-trained weights get shared with FL clients and the global model. To guarantee the performance of the proposed framework, we assess it by considering different evaluation metrics, such as FL client’s training accuracy, training loss, validation curve, regression error rate, and blockchain’s transaction and execution cost. The synergy of AI and blockchain highlights their combined impact on revolutionizing energy consumption prediction in the maritime industry. This not only refines predictive accuracy but also ensures the confidentiality and integrity of the predicted data.