Bayesian Optimization-Based Performance Enhancement of Pneumatic-Hybrid Diesel Generators for Remote Power Systems

Diesel generator sets remain a primary source of electricity in many remote and isolated areas due to their robustness, dispatchability, and ease of deployment. But their high fuel consumption and emissions of pollutants pose economic challenges and environmental concerns, especially in regions with high fuel transportation costs. Improving the efficiency of diesel-based power generation is essential for enhancing the sustainability and reliability of off-grid energy systems.

Pneumatic hybridization of diesel engines is a promising method of enhancing brake thermal efficiency and specific power output. In this approach, compressed air is used to assist the intake process and modify in-cylinder thermodynamic conditions, potentially improving combustion efficiency while reducing fuel consumption and emissions.

This study proposes a data-efficient optimization framework combining Diesel-RK engine-cycle simulation with Bayesian optimization. The Diesel-RK model is treated as a computational black-box function that maps controllable engine operating variables to performance and emissions-related outputs. The optimization problem is formulated with multiple objectives, including maximizing brake thermal efficiency and minimizing brake-specific fuel consumption, while satisfying operational constraints, such as the required electrical power output and the allowable peak in-cylinder pressure.

Bayesian optimization is particularly suited to optimization problems involving computationally expensive black-box models such as Diesel-RK engine simulations. In contrast, widely used population-based metaheuristic algorithms, including genetic algorithms, particle swarm optimization, and differential evolution, typically rely on large populations of candidate solutions and repeated generations of objective function evaluations to converge toward optimal solutions. When each evaluation requires a detailed engine-cycle simulation, these methods can result in high computational costs. Bayesian optimization constructs a probabilistic surrogate model of the objective function using a Gaussian process. In addition, it guides the search using an acquisition function that balances exploration of the design space and exploitation of promising operating regions. This surrogate-based strategy allows for efficient identification of candidate optimal solutions while significantly reducing the number of expensive simulation evaluations.

The proposed framework aims to determine optimal operating conditions for pneumatic-hybrid diesel generators for remote power applications. The analysis will quantify the influence of pneumatic assistance on brake thermal efficiency, fuel consumption, and emissions indicators, including nitrogen oxides and soot formation. In addition, the study will evaluate the computational efficiency of Bayesian optimization relative to conventional metaheuristic methods, focusing on convergence behavior and the number of required simulation evaluations.

The proposed methodology provides a systematic framework for optimizing pneumatic-hybrid diesel generators used in remote and off-grid energy systems, aiming to improve fuel efficiency, reduce operating costs, and mitigate environmental impacts associated with diesel-based electricity generation.