Research On Transformer Condition Evaluation Method Based On Multi-source Data

Power transformers play a key role in the efficient transmission of electricity between the power grid and its consumers as well as the interconnection of the main network,distribution network,and regional power grids.As a result,it is clear that keeping the power transformer in good working order is essential to the efficient and secure operation of the power system.A local or widespread blackout may result from a power transformer fault,which is directly tied to the reliability of the power system.Thus,it is crucial to take the appropriate steps to keep an eye on the power transformer’s state and achieve the goal of early problem detection through health evaluation.Dissolved gas analysis and infrared tracing are two reasonably mature theoretical investigations that have emerged in the current power transformer health monitoring field.Unfortunately,the majority of recent study findings concentrate on a particular level or structure of the operational condition of power transformers,which creates an information island in the findings and inadequate integration of the thorough assessment of transformer status.Power transformer operation status related information,including temperature monitoring,operation scheduling,meteorological environment,etc.,has gradually demonstrated the typical big data characteristics of large volume,variety,and fast growth in the context of smart grid promotion and energy Internet connectivity.Given this,comprehensive mining and analysis of power transformer status data based on artificial intelligence technology are proposed in this paper,and the transformer status evaluation system based on multi-source information fusion considering meteorological environment factors is further constructed.The research contents of this paper are as follows:1.In the face of massive data in the background of big data,corresponding pretreatment measures should be taken according to the different characteristics of the data.Firstly,the outlier detection method based on the quartile method is used to screen the original electrical data.Screen out the bad data generated by automatic and manual input errors and realize the preliminary preprocessing of electrical data.Then,the correlation analysis of transformer operation status is carried out on the original meteorological data through the grey correlation analysis theory.The data types with low correlation are discarded and the data with strong correlation are retained to realize the preliminary cleaning of the original data samples,to improve the quality of the sample data,and make full use of the sample data information for follow-up evaluation.2.Given data loss caused by temporary faults of sensors and information sampling and input equipment and abnormal data clearing in the aforementioned work,to ensure the integrity of collected historical data samples and the richness of potential information contained in them,and improve the accuracy of data analysis and mining,this paper compares and verifies various data filling methods.This paper proposes Hermite interpolation for continuous electrical data and a random forest filling method based on R language learning for non-continuous multi-type missing meteorological data.The feasibility of the filling process is verified by case studies.3.To improve the efficiency of data mining and solve the computational workload under massive electrical data,this paper proposes to convert the voltage,power,and other data indirectly related to the running state of the transformer into short circuit reactance data directly related to the winding state of the transformer.To realize the real-time measurement of transformer short-circuit reactance and solve the economic problem caused by installing an online monitoring device,this paper proposes the online monitoring and calculating circuit model of transformer winding deformation based on short-circuit reactance method.This model can calculate the real-time short-circuit reactance value of the transformer through the voltage,current,and power data acquired by the data acquisition and monitoring system of the power grid company,to realize the online acquisition and processing of short-circuit reactance value without an online monitoring device.The consistency between the calculation trend diagram of short circuit reactance and the evolution trend of the transformer winding health state is verified by case studies,and the validity of the model is confirmed.4.The Multiple-Kernel Learning Multiclass Relevance Vector Machine learning model,which can effectively fuse meteorological data and other data information from different channels and realize multi-source information fusion processing,is used in transformer operation status evaluation,and its advantages of good sparse performance,fast operation speed,and high classification accuracy are reasonably given play.Through the establishment,learning,and training of the Multiple-Kernel Learning Multiclass Relevance Vector Machine learning model,the transformer operation status evaluation based on multi-source data is realized,thus the transformer operation status evaluation system considering the information fusion under the meteorological environment is completed,and its feasibility is verified through the analysis of case studies.The findings demonstrate that the transformer status evaluation system suggested in this paper is capable of accurately assessing the transformer’s actual status and of successfully integrating information about the local meteorological environment.This allows for a more precise and applicable operation status evaluation for the various actual environments in which the equipment is used.