Study On Running Section Similarity Matching Based On Spatio-temporal Information Of Power Grid

In order to simplify the complex scheme checking work in the process of making work tickets for power grid,reduce the dependence on the work experience of power grid dispatchers,and at the same time excavate more value from the spatio-temporal information of power grid,realize the preservation and appreciation of data assets,and further promote the development of power grid management towards automation and intellectualization,this paper presents a method of similarity matching of historical operation sections of power grid based on space-time information of power grid.Expect to take the specific data of the running section as the fulcrum and the similarity matching of the running section as the way to obtain more valuable historical information,so as to realize the intelligent decision-making of the operation mode of the power grid and provide help for the intelligent management of the operation and maintenance of the power system.Firstly,through the existing research and management experience,this paper constructs a feature database and a statistical database which can comprehensively characterize the information of power grid operation cross-section,so as to complete the dimension reduction of massive sample data.On this basis,the similarity matching index system of historical operation section is established,which makes the similarity degree between each operation section and the current operation section of the system visually displayed by numerical method,thus providing a theoretical basis for the similarity ranking of running section and further determining the more valuable running section.Considering that the existing machine learning algorithms still focus on clustering and classification of sample data,they can not accomplish the task of similarity matching well.Firstly,this paper improves the traditional semi-supervised K-means algorithm,including a series of optimization measures to determine the initial clustering center and the number of clustering,which make it more suitable for the similarity matching of running sections mentioned in this paper.Then,the improved semi-supervised K-means algorithm is used to complete the preliminary screening of the historical running section of the power grid,and the number of samples is greatly reduced by clustering.Finally,using the proposed similarity matching index system,the samples in the clustered target class are evaluated and sorted,and the most valuable historical running section is finally determined.At present,in-depth learning is still mainly used in the fields of picture and speech recognition,but it is seldom used in power system.So this paper refers to and learns the technology and method of image recognition.According to the need of the algorithm,the running section data is arranged into a specific format,and the training samples for supervised learning are constructed.Then,the information of power flow,load,node voltage and grid structure of the running section is included as much as possible in each training sample,which minimizes the missing of key features and ensures the highest matching accuracy.Finally,the labeled sample data are used for supervised learning to train and optimize the parameters of deep convolution neural network.Through data-driven approach,we extract features from raw data from low-level to high-level,from concrete to abstract.The trained model can make full use of the deep framework to mine the potential mapping relationship between sample data and similarity results,and ultimately get a higher matching accuracy.