AdOpt: An Adaptive Optimization Framework for Large-scale Power Distribution Systems

Optimizing self-evolving and dynamically changing systems is a grand challenge. In order to apply optimizations almost all conventional optimization techniques require a runtime system model. However, system models and their solution techniques vary in their strengths and limitations. For a rigid system, a single system model is acceptable. But if the system is constantly changing its structure then a rigid model is not able to represent the system properly, resulting in an inefficient use of technique in some cases. Therefore, in this paper we propose a framework for an optimization engine that adapts the optimization technique based on the system state. The adaptation involves selection of techniques based on historical statistics and current data, and dynamic generation of a model at runtime. This runtime model is then used to apply a relevant optimization technique to find a desired optimization plan for the system. We have evaluated the proposed framework on an electricity distribution system. Our results show that the proposed framework is adaptable, fast and able to manage numerous situations.