<p>The objective function and constraints of Dynamic Multi-objective Optimization Problems change over time, which affects the diversity of optimization solutions and prediction accuracy. Most existing Dynamic Multi-objective Optimization Evolutionary Algorithms (DMOEAs) not only ignore the impact of initial population diversity on evolution, but also rarely consider nonlinear changes in prediction strategies. Therefore, a DMOEA based on innovative hybrid initialization and ensemble prediction strategy, coupled with dynamic mutation adjustment is proposed. The hybrid initialization strategy innovatively adds multi-level chaotic mapping perturbations and random enhancements to the uniformly distributed population generated by Latin Hypercube Sampling, effectively expands the diversity and spatial coverage of the initial population, improves the global search capability of solutions. The key highlight of ensemble model prediction strategy is the accurate prediction of nonlinear changes. The most suitable model can be selected based on different types of changes, reducing reliance on individual model prediction errors. The dynamic mutation adjustment strategy uses pareto entropy to monitor the distribution of pareto front in real time, adaptively regulates the balance of exploration and development in changes, and perform well in real-time response to environmental changes and maintaining solution set diversity. Compared with other advanced algorithms, the experimental results show that the proposed algorithm not only responds quickly to dynamic environmental changes, but also achieves outstanding results in convergence and diversity.</p>