Bi-level multi-objective stochastic optimization of multi-energy systems in remote mining areas with joint chance constraints

With the substantial depletion of existing mining resources, the safe and stable operation of multi-energy systems in remote mining areas has become a crucial research objective for most energy and mining companies. However, challenges remain, including large prediction errors in renewable energy sources and insufficient power supply from external power grids. To address this issue, a method based on an improved least-squares generative adversarial network is proposed to learn the distribution of time-of-use renewable energy output prediction errors and generate scenarios to model joint chance constraints. Simultaneously, a bi-layer multi-objective stochastic optimization model is established, proposing a method for rigorously relaxing joint chance constraints and a comprehensive linearization approach. The lower-layer model is modeled as a canonical mixed-integer linear programming model directly guiding scheduling, while the upper-layer model is a three-objective optimization with only continuous variables, aiming to reduce overall scheduling costs, total carbon emissions, and total electricity purchases from external power grids. This aims to provide appropriate key parameters for the scheduling problem. To efficiently and effectively solve this problem, a novel adaptive non-dominated sorting evolutionary algorithm with Halton sequence is proposed, along with a decision-making algorithm to determine the final solution. This paper provides extensive case studies and discussions, including comparisons of the solution performance of different base solvers under a unified algorithm framework and with the same population size and number of evolutions, comparisons of solution performance with different population initializations, and comparisons of solution performance with different important parameter settings. It also includes comparisons of solution performance under more than 10 different algorithm frameworks, supplemented by linearization methods for different norms and demonstrations of solution results, and finally obtains a decision solution based on a re-decision algorithm. This research comprehensively provides a scheduling scheme for multi-energy systems in remote mining areas that considers multiple indicators throughout the entire process.