The Research And Application Of Online Distribution Network Reconfiguration Based On Improved Artificial Neural Network

By introducing the history of China’s power industry development at the begin-ning of this paper, we draw a conclusion that strengthening distribution network con-struction is vital. Hardware base and software prerequisites for distribution networkreconfiguration are gradually formed with regard to the fact that distribution automa-tion technologies are getting more sophisticated.The traditional distribution network implements reconfiguration computing onseason or month basis. However, considering the dynamic characteristic of load andfully tapping the potential of distribution network transmission, online distributionnetwork reconfiguration is imperative. Numerous reconfiguration algorithms based onheuristics and artificial intelligence has been researched, but they cannot be separatedfrom iterations process to recalculate the load trend, which makes it infeasible foronline application due to time requirement. But well trained neural networks are ableto present reconfiguration suggestions immediately. In order to make it happen, thispaper proposed a set of improved neural network for real-time distribution reconfi-guration solution and the main contents and works can be described as follow:First, improving standard Elman neural network structure, which means provingfeedback from output layer to hidden layer and it aims to avoid adverse impact on theoverall network when the feedback is missing. As a result, OIF-Elman network cansignificantly enhance the network generalization capabilities and output stability.After that, this paper presented a brand new heuristic algorithm as human groupoptimization to train neural network. Compared to classic training methods, it is su-perior in aspects like learning ability and solution precision, which can not only beapplied in reconfiguration solution but also other optimization questions in electricalfield and even other subjects.Then considering the fact that the combination of load nodes and load levels ofdistribution network will lead to oversized training sample sets causing ineffective-ness of study, this paper introduced validation fuzzy C-Mean clustering technologybased on Gerardo Beni in load sample classification, which significantly simplifiesthe neural network structure and reduces the size of the sample set without under-mining the quality of solution.At last, through the verification on IEEE30nodes testing network and compari- son to classical algorithm, it reflects excellent performance, which makes it an effec-tive and practical online distribution network reconfiguration solution.