通过总结材料优化设计过程特征，归纳出一类大量存在但兼具小数据、多目标、偏好性和动态性等特征的复杂工程优化问题。拟采用集成学习和增量学习技术对小样本数据和持续新增样本数据协同构建高精度的动态多目标代理模型，并对代理模型建立复杂环境（多目标、偏好性和动态性）下的进化过程状态评估及信号反馈机制，结合粒子群优化算法在快速收敛、易于响应动态变化环境和易于选择偏好多目标最优解等方面的优势，设计具有自适应平衡开发与开采过程能力的多目标粒子群优化算法，进而求解这类问题在用户偏好导引下的区域Pareto最优解集和前沿，从而为这类复杂工程优化问题提供一套高效实用的数据建模与优化求解方法。并联合研究NdFeB稀土永磁材料成分和工艺的智能优化设计新方法，将数据驱动的多目标进化优化技术引入材料工程，为计算材料学提供新的技术途径，也为复杂工程进化优化算法的理论研究成果提供工程应用和验证案例。

The complex engineering optimization problem (CEOP) with some characteristics, such as small data, multiobjective, preference and dynamics, can be concluded from the process of material optimization design. The machine learning technique including ensemble learning and incremental learning will be employed to collaboratively build the dynamic multiobjective surrogate model for CEOP according to the small historical data and the continuous new data. A mechanism for evaluating the evolutionary statuses and feeding back the evolutionary signals will be established to monitor the complex evolutionary environment of CEOP. An adaptive multiobjective Particle Swarm Optimization (AMOPSO) with the ability of balancing exploration and exploitation process will be designed to optimize CEOPs so as to obtain the regional Pareto solutions and front under the user’s preference. Thereafter, a high-performance and practical approach will be proposed for modeling and optimizing CEOP. Meanwhile, a new material design method will be intensively investigated to optimize the ingredient and processing technic for the NdFeB rare-earth permanent magnet materials. Therefore, the data-driven evolutionary multiobjective optimization will be a new and intelligent approach for the computational materials science. As well, the material optimization design will be a good application and verification case of the data-driven evolutionary algorithm.