The Effect of Laziness on Agents for Large Scale Global Optimization.

摘要: Practical optimization problems from engineering often suffer non-convexity and rugged, multi-modal fitness or error landscapes are thus hard to solve. Especially in the high-dimensional case, a lack of derivatives entails additional challenges heuristics. High-dimensionality leads an exponential increase search space size tightens problem premature convergence. Parallelization for acceleration involves domain-specific knowledge data domain partition functional algorithmic decomposition. On other hand, fully decentralized agent-based procedures global based on coordinate descent gossiping have no specific decomposition needs can be applied arbitrary problems. Premature convergence mitigated by introducing laziness. We scrutinized effectiveness different levels laziness types first time approach real-world problem: predictive scheduling virtual power plant orchestration. The lazy agent turns out competitive superior non-lazy one standard heuristics many cases including real world problem.