Dynamic Multi-Objective Optimization Based on Integrated Incremental Long Short-Term Memory and Inverse Model Prediction Strategy

Dynamic multi-objective optimization problems (DMOPs) are widely applied in scheduling, resource allocation, and complex task assignments, requiring algorithms to respond quickly to environmental changes. Existing prediction methods do not id=R2fully id=R3utilizemake use of historical information and rely on simplistic linear or nonlinear assumptions, which fail to effectively handle complex environmental dynamics in practice, especially leading to increased prediction errors over long periods of change. This paper id=R3proposesintroduces a novel prediction framework based on incremental LSTM and inverse models (IncLSTM-IM). For nonlinear changes, the IncLSTM is employed to learn the evolutionary patterns of the pareto optimal set (POS), fully exploiting the capacity for handling sequential data and learning long-term dependencies to guide offspring solutions id=R3toward thebeing closer to next POS. For linear changes, an inverse model is introduced to predict using local historical information. Furthermore, a population selection mechanism based on clustering is implemented to combine diverse offspring populations. IncLSTM-IM overcomes the limitations of existing prediction methods by effectively capturing complex environmental changes and enhancing population diversity through cluster-based id=R3population selection. Experimental results, compared with existing algorithms, demonstrate the robust performance and adaptability of the proposed id=R3approachalgorithm.