面向大规模电网优化调度的纵横交叉群智能优化方法研究

Optimal dispatching of power grid is a kind of high dimensional non-convex optimization problem which can produce remarkable economic benefit. In the field of electric power industry, it is a hot spot to solve such NP problem by using swarm intelligent optimization methods. In the previous study, the applicant first discovered the dimensional local optimum phenomenon and proposed a high performance crisscross optimization algorithm, which has been successfully applied to solve a large-scale economic dispatch problem with 24000 variables and has made great progress in addressing premature convergence and dimensionality disaster. Although crisscross optimization algorithm exhibits good performance in the practical application, its intrinsic mechanism is not clear. Inspired by this, this project is to carry out the following study: First, the global convergence of the crisscross optimization algorithm will be analyzed in depth, and then the chain propagation mechanism of the dimensional local optimality in the population will be revealed. Furthermore, the principle of the catalytic acceleration for the hybrid method using crisscross optimization will be expounded. Finally, this project will explore the distributed parallel computing mechanism and method of the crisscross optimization algorithm and build a multi-agent optimization system based on middleware in local network so as to carry out the confirmatory research. The above research has important scientific and practical significance in developing the theory system of crisscross optimization algorithm, extending the theoretical intension of swarm intelligent optimization method and its application to the optimization decision problems of power grid.

电网优化调度是一类可产生显著经济效益的高维非凸优化问题，利用群智能优化方法求解此类NP难题是电力领域的一个研究热点。申请者在前期研究中，首次发现维局部最优现象并提出一种高性能纵横交叉算法，该方法已成功用于求解一个含24000变量的超大规模电网经济调度问题，在解决早熟收敛和维数灾方面取得重要进展。纵横交叉算法虽然实际应用效果良好，但内在机理尚不清楚。受此启发，本项目拟开展如下研究：首先，深入分析纵横交叉算法全局收敛特性；然后，揭示维局部最优性在种群内的链式传播机理；其次，阐明纵横交叉融合方法的催化加速原理；最后，探索面向电网大规模优化调度问题的纵横交叉算法分布式并行计算机制与方法，并搭建局域网下基于中间件的多智能体优化系统开展验证性研究。上述研究对于发展纵横交叉算法理论体系，扩展群智能优化方法的理论内涵及其在电网优化决策问题中的适用范围具有重要的科学与工程实践意义。

群智能优化方法面临大规模电网调度此类高维非凸优化问题时，计算过程往往会出现早熟问题而导致调度结果质量不理想，同时也会出现计算耗时过长而无法满足超短期实时调度要求。本课题以高性能纵横交叉算法相关理论机制及其并行计算研究为主线，旨在助力群智能优化算法解决高维非凸优化问题面临的早熟收敛和计算效率共性问题。项目完成了纵横交叉CSO算法马尔科夫链模型及收敛性分析，探索了面向深度学习二次训练的CSO全局收敛能力；项目揭示了维局部最优性在种群内的链式动态传播机理，以多目标环境经济调度问题为突破口，提出了一种自适应CSO扩展因子的最优筛选方法，同时发现了一种有效寻找多目标复杂优化问题最优Pareto前沿的多目标CSO优化方法；项目揭示了纵横交叉CSO算法融合方法的种群更新机制，以灰狼算法突破口，探究了CSO横交叉算法融合方法的催化加速原理与加速过程，提出了为其它智能优化方法催化加速的一般性方法；项目构建了完全并行化的分解计算模型，重点研究了纵横交叉算法并行计算方法，以电力系统多区域经济优化调度问题为突破口，深入研究了具备保护数据隐私，降低解问题维度，减轻通信负担的CSO分布式并行计算一般性方法。上述获得的研究结果不仅有助于拓展群智能优化算法的理论内涵，同时为CSO在复杂约束条件下高维非凸实际工程问题中的应用提供了重要的理论与应用基础。