Comparison of calculation methods for switching impulse discharge voltage of 50% air gap between rod and plate

Author： GOZ Electric Time：2024-10-25 09:34:43 Read：13

The rod-plate long air gap operation impulse discharge test was carried out at the outdoor test site (55m above sea level) of the State Grid UHV DC test base in Beijing. The test equipment mainly consists of a 200kV/480kJ impulse voltage generator and a capacitive voltage divider. The upper end of the rod electrode is suspended by a long composite insulator string, and the lower end of the insulator string is connected to the power line and steel pipe. 126 sets of rod-plate positive polarity operation impulse test data were used as samples to train and test the intelligent model.

Comparison with other algorithm model results

To verify the advantages of the algorithm proposed in this paper over the traditional machine learning algorithm, this section uses three algorithms, namely the classic BP neural network (BPNN), support vector machine (SVM), and random forest (RF), to construct the discharge voltage prediction model. Among them, the input variables, training sets, and test sets selected for the training model are the same as above. Since the empirical risk minimization algorithm converges faster than the constant risk minimization algorithm, the total number of iterations of BPNN is set to 10000, and the other parameter settings are the same as above. SVM uses the cuckoo algorithm to optimize hyperparameters, and its parameter settings are the same as those in the literature. The parameter selection of RF is simpler than the first two algorithms, and its settings are the same as those in the literature.

The 50% discharge voltage of the rod-plate gap calculated by the regularized IRM-NN model is closest to the test value, and the strong generalization of the model in the test temperature range of -0.9℃~29.0℃ is verified. In summary, the model in this paper can be extended to the air gap discharge voltage calculation and meteorological correction scenarios outside the test conditions of the training samples. Its calculation results are accurate and reliable, and it is suitable for a wide range of electrode size changes.

In order to propose a calculation method for the 50% operating impulse discharge voltage of the rod-plate length air gap outside the test conditions that can be extended, this paper selects the regularized IRM-NN algorithm to construct an intelligent prediction model. 76 sets of rod-plate gap discharge test data with a temperature range of 5℃~25℃ are used as training samples, and the rod-plate gap distance, air pressure, dry temperature, and absolute humidity are selected as input variables to train the model. Then, the generalization performance of the model and the calculation accuracy of extrapolation to other environments were tested on test samples with a wider range of temperature and humidity. The prediction effects of the proposed model were compared with those of three traditional machine learning models, and the following conclusions were obtained:

1) The factors affecting the discharge of the rod-plate air gap were systematically analyzed, including gap structure, voltage waveform, voltage polarity and meteorological factors. Combined with the sample data characteristics of this paper, the rod-plate gap length, air pressure, temperature and absolute humidity were extracted as the key feature quantities of the rod-plate gap operation discharge.

2) A method for calculating the discharge voltage of the rod-plate gap 50% operation impulse based on regularized IRM-NN was proposed. Compared with traditional intelligent algorithms, this method is less prone to overfitting problems and can be extended to other application scenarios outside the test conditions.

3) In the training model based on the rod-plate gap operation impulse test data in the temperature range of 5℃~25℃, in addition to the samples in the temperature range of 5℃~25℃, the test set also contains samples with temperatures less than or equal to 5℃ and greater than 25℃. The maximum absolute percentage error of the model on the test set is 5.1%, and the average absolute percentage error is only 1.6%, which verifies that the model not only has high prediction accuracy within the test conditions of the training samples, but can also be effectively extended to prediction targets outside the test conditions.

4) The prediction effects of the proposed model are compared with those of three other common machine learning models. The results show that the accuracy and stability of the proposed model are higher than those of other models, and the calculation accuracy on samples outside the test range of the training samples is significantly higher than that of other models.

It should be noted that the proposed model is suitable for the prediction of 50% discharge voltage of positive polarity operation impulse of rod-plate length gap of more than 2m with a rod electrode radius less than the critical radius, and can be extended to temperature and humidity environments outside the test range. Since the samples are all collected from test data at an altitude of 55m, the effect of the model for other altitudes remains to be verified. In subsequent research, we will supplement test data at other altitudes to extend the calculation method to a larger range of application.