Abstract:

With the increasing integration of renewable energy， traditional power system models can no longer meet the complex demands of modern power systems. To adapt to the trend of multi-energy collaborative power generation， a new-type power system model was developed， primarily based on thermal power generation with renewable energy as supplementary sources. Due to the multi-objective trade-offs between power generation costs and carbon emission targets in the new-type power system， an intelligent optimization method was required to dynamically adjust the output of each generating unit and fully leverage the advantages of various energy sources. Therefore， a non-dominated sorting moth-flame optimization algorithm based on bidirectional encoder representations from transformers （NSMFO-BERT） was proposed. As a large model， BERT excelled in handling complex data relationships. By learning from NSMFO， it established the relationship between the active power of generating units and load forecasting， rapidly developing scheduling strategies for a large number of generating units. Simulation results showed that compared to NSMFO， the multi-objective grey wolf algorithm， and the multi-objective ant lion algorithm， NSMFO-BERT could find a Pareto curve with lower target values for power generation costs and carbon emissions. Furthermore， the computation speed of the proposed algorithm was 69.3%， 61.4%， and 90.9% faster than the aforementioned algorithms， respectively. It demonstrated strong generalization ability， suitable for addressing large-scale power system scheduling problems.

Key words: bidirectional encoder representations from transformers, non-dominated sorting moth-flame optimization algorithm, large model, new-type power system, power generation cost, carbon emissions