Analysis And Prediction Research Of High-Voltage Switchgear Contact Temperature Data

The safe operation of high-voltage switchgear is the premise of long-term, stable supply of power systems, and the temperature of the three-phase( A, B and C in the switchgear) contact is an important factor which directly maps the performance of the switchgear. In recent years, there had been several accidents caused by over-high temperature of the switchgear. If we can accurately predict the trend of future temperature change of the contact, we will be able to reduce the occurrence of thermal failure of the switchgear, which helps to improve the service life of the equipment.Firstly, by analysing the temperature data of high-voltage switchgear contact, the theise select two appropriate models. Then, simply describe the grey prediction model and multivariate linear regression model. Based on the background of the data, using the two methods to model and simulate. According to the grey prediction model, single contact point temperature data can be divided into three different time sequences to study the smoothness and smoothing treatment, trial lengths to get the optimal length of each time series. Then, using traditional method, Newton interpolation method and integral reconstruction method respectively to build background values and test the effect of GM(1,1) model predictions. For multiple linear regression models, analysis the switchboard contact from transverse and longitudinal regression respectively. By dividing the data into 5 different time intervals, established horizontal regression model. Take the average temperature and the character of the remaining two contacts in the same switchgear as independent variables to build longitudinal regression model. Then use the merge mode and activity chain model of Map-reduce to do multiple linear regression to achieve the parallel processing of large data sets.The actual simulation analysis of temperature of the 110 KV voltage switch cabinet contact showed that, in the prediction of grey model, 1 hour interval span of time series reconstruction integral under the background value of GM(1,1) predicting the best, fit for an hour early prediction of transient. While, in multiple linear regression, 1 hour interval longitudinal regression for time series forecasting best, suited for one hour in advance of long-term forecasts. To sum up, GM(1,1) takes advantage in the prediction of transient, and multiple linear regression suit for long-term prediction.