Design And Implementation Of Auxiliary Decision System For Operation And Inspection Of Substation Equipment In Power Grid

With the rapid development of the economy,the demand for electricity is becoming stronger,the scale of power transmission and transformation network is getting larger and larger,and a large amount of unstructured power text data has been accumulated.These data contain rich professional value and can be used for defect reasoning Classification,analysis and early warning,and maintenance decision-making and other key areas.However,due to the complexity and variability of actual failure factors of power transformation equipment and the subjective initiative of maintenance engineers in written records,the format of most text data is far from the specification for defects in power transmission and transformation equipment stipulated in the State Grid Standards,and it is impossible to use computer systems Automated and efficient processing.(1)Constructing the Defect Category Inference Dataset of Substation Equipment.Currently,the publicly available data sets in the field of substation equipment only have time series and image data sets,and lack data sets for text analysis.Therefore,this paper uses inspection records and defect classification data for manual labeling to construct a data set for inferring substation equipment defect categories.Among them,there are 13694 pieces of patrol inspection record data and 5374 pieces of defect classification data.(2)Design and implement a natural language inference model based on tiny weights.In the actual production environment of the power company,there is a lack of graphics card processors for fast reasoning of algorithm models,so it is necessary to design a lightweight deep learning model to assist maintenance engineers in text analysis.This paper first uses the principal component analysis method to compress the word vector dimension,which simplifies the network structure compared with methods such as the fully connected network structure,and the calculation is simple and fast;Then use depth-separable convolution as the encoder of the encoding layer.Compared with structures such as cyclic neural network and long-term short-term memory network,it not only improves the operation speed,but also greatly reduces the weight parameters of the model;And use the enhanced residual connection mechanism to extract the deep semantic features of the text to make up for the lack of generalization ability of the network structure;Then use the enhanced residual connection mechanism to extract the deep semantic features of the text to make up for the lack of generalization ability of the network structure;Finally,instance standardization is introduced to fuse the semantic features of text pairs to improve the inference ability of the network structure.(3)Design and implement substation maintenance plan management process.This paper first provides the network construction function of the power substation,and intends to divide the power substation into four levels of "substation-interval equipment-substation equipment-equipment components" according to the structure,function and characteristics.Then determine the information that needs to be recorded for the substation equipment,which are the commissioning time,voltage level,manufacturer,and equipment model.Next determine the information that needs to be recorded for the equipment components,which are the regular maintenance items and historical maintenance information.Finally,it provides the maintenance plan editing function,which divides a year into 12 time periods based on the month cycle,and initializes the maintenance plan according to the rules.The maintenance engineer can refer to the data recorded in the power substation network for further modification and optimization.