Growing concerns on the energy crisis impose great challenges in development and deployment of the smart grid technologies into the existing electrical power system. A key enabling technology in smart grid is distributed generation, which refers to the technology that power generating sources are located in a highly distributed fashion and each customer is both a consumer and a producer for energy. An important optimization problem in distributed generation design is the insertion of distributed generators (DGs), which are often renewable resources exploiting e.g., photovoltaic, hydro, wind, ocean energy. In this paper, a new power loss filtering based sensitivity guided cross entropy (CE) algorithm is proposed for the distributed generator insertion problem. This algorithm is based on the advanced CE optimization technique which exploits the idea of importance sampling in performing optimization. Our experimental results demonstrate that on large distribution networks, our algorithm can largely reduce (up to 179.3%) power loss comparing to a state-of-the-art sensitivity guided greedy algorithm with small runtime overhead. In addition, our algorithm runs about 5× faster than the classical CE algorithm due to the integration of power loss filtering and sensitivity optimization. Moreover, all existing techniques only test on very small distribution systems (usually with < 50 nodes) while our experiments are performed on the distribution networks with up to 5000 nodes, which matches the realistic setup. These demonstrate the practicality of the proposed algorithm.
Adaptive Day-Ahead Prediction of Resilient Power Distribution Network Partitions
