Study On Multi-factor Temperature Prediction System For Substation Equipment

Substation as a key part of the smart grid,the stable operation of its equipment is related to the safety of the entire grid.Most substation equipment breakouts result from overheat.Thermal failure can be detected by the equipment surface temperature monitoring.And forecasting temperature trend in future ahead can achieve preventive measures to reduce the economic losses caused by power plant equipment failure.At present,more equipment temperature monitoring methods are used for real-time on-line monitoring and short-term temperature prediction.As for short-term temperature prediction,it is often realized by historical temperature data.It does not take into account the relevant equipment and environmental factors that affect the temperature of the equipment.Learning from the methods and process of data mining,a temperature prediction model for substation equipment is put forward in this thesis,while considering the influence of equipment operating parameters and environmental factors on the temperature.Firstly,the data preprocessing of the equipment historical data samples is carried out by outlier rejection scheme based on the neighborhood density,thus the adverse effects of the abnormal samples on the prediction model are eliminated.Then the hybrid feature selection scheme which is combined of genetic algorithm and rough-set is conducted,choosing the key features of sample data,and eliminating redundant features.Finally,based on the outliers rejection and feature selection,the random sequence temperature prediction and timing sequence temperature prediction of the samples using the recurrent extreme learning machine prediction model are proposed.Finally,the forecasting model proposed in this thesis is applied to the actual system,and a set of substation equipment temperature prediction software system is designed for an 110 kV substation of State Grid Corporation.Compared with other similar models,the recurrent extreme learning machine forecasting model adopted in this thesis is verified to be superior to the comparison algorithm in both random sequence sample set and the time series sample set.And it can solve the functional requirements of scattered forecast and real-time prediction of the software system,and provide a technical guarantee for the smooth operation of substation equipment.