模拟演化计算的效率加速理论及其实现策略

Simulated evolutionary computation (SEC in short) is a family of AI technologies for searching and optimization. They work through simulating nature evolution and genetics rules. After development more than 20 years, SEC technologies have been very extensively applied in science and engineering. Their efficiency is, however, not high, particularly, as that as expected. Through identifying the main causes of low efficiency of the currently known SEC algorithms, a series of efficiency speed-up strategies together with related theories is developed in this project. With the developed strategies, the classical genetic algorithm, a typical SEC technology, can be sped up in efficiency by several (2-20) orders of magnitude. The proposed efficiency speed-up strategies have been applied successfully to solving a set of difficult optimization problems including the inverse problem of fractal image compression, the maximal independent set problems and multimodal, mutiobjective optimization problems. We also systematically analysed convergence issue of various SEC algorithms, and developed a generic convergence theory of SEC. The obtained results in the project, consisting of 14 papers published in international and domestic journals, have great impact on further development of SEC technologies.

模拟演化计算是借鉴生物进化机制求解复杂问题的人工智能技术。该类技术生物基础坚实，应用极为广泛，但计算效率低，数学基础薄弱。本项研究围绕提高模拟演化算法计算效率，建立其加速理论。主要包括：有效避免和利用resampling的原理与方法；非elitist型执行呗缘氖樟怖砺郏恢秩河牖肪辰换プ饔媒Ｐ图盎诳善唇颖嗦牒团懦阕拥目焖僮允视κ迪值取