Abstract: As a massive number of renewable energies are connected to the grid, the traditional power system model has been incapable of meeting the complex demands of modern power systems. To adapt to the trend of synergistic power generation from multiple energy types, this study constructs a novel power system model with thermal power generation as the main source and renewable energy as a supplementary source, integrating multiple renewable energy sources such as wind, solar, nuclear, hydro, biomass, and tidal energy. Because of the challenge of multi-objective trade-offs between generator cost and carbon emission targets for novel power systems, an intelligent optimization method is needed to dynamically adjust the output of each generator to utilize the full benefits of each energy type. Therefore, this study proposes the non-dominated sorting moth-flame optimizer-bidirectional encoder representations from Transformers algorithm. Bidirectional encoder representations from Transformers (BERT), as a large language model, has excelled in handling complex data relationships. The BERT can quickly generate massive scheduling strategies by learning the relationship between the active power of output generators and load forecasts produced by the non-dominated sorting moth-flame optimization algorithm. The simulation results show that compared with the non-dominated sorting moth-flame optimizer, the multi-objective grey wolf algorithm, and the multi-objective ant lion algorithm: (1) the proposed algorithm can find smaller Pareto curves for the objective values of the generator cost and the carbon emissions; (2) the proposed algorithm achieves 69.3%, 61.4% and 90.9% faster computation speed than the comparison algorithms, respectively, and has the highest computational efficiency.

Key words: bidirectional encoder representations from Transformers, non-dominated sorting moth-flame optimizer, large language model, novel power system, generator costs, carbon emissions