Study on Switching Impulse Discharge Voltage of Rod/Ball-Plate Long Gap

Author： GOZ Electric Time：2024-10-03 09:28:00 Read：20

Air gap is an important external insulation form in power transmission and transformation projects. The determination of its insulation strength mainly relies on high-voltage discharge tests. Establishing a discharge model and predicting the gap discharge voltage through simulation calculations is one of the research goals of "computational high voltage engineering" in recent years. Since the French Les Renardières research group summarized the physical process of long air gap discharge through a series of experimental studies in the 1970s, domestic and foreign scholars have established a large number of physical models to simulate long gap discharges and calculated the impulse discharge voltage of typical structural gaps such as rods and plates. Limited by the complexity of long gap discharges, the calculation process of the physical model still involves many simplified assumptions, and it is currently difficult to apply it to the discharge calculation of complex engineering gaps.

In recent years, the use of artificial intelligence-driven data analysis technology to predict the discharge voltage of air gaps has become another way to achieve insulation calculation. This type of method uses the electric field distribution parameters, voltage waveform characteristics and atmospheric environment parameters that affect air discharge as the input of the machine learning model, and establishes a multi-dimensional nonlinear relationship between the input and the discharge voltage through the training model, thereby realizing the prediction of the gap insulation strength. The relevant literature divides the electrostatic field simulation calculation area of a typical electrode short air gap into the electrode surface, discharge channel, inter-electrode path and the entire area, and extracts the electric field distribution characteristics from it to train the support vector machine (SVM) model, realizing the prediction of the breakdown voltage of short gaps such as ball gaps, rod-plates, and ball-plates. The relevant literature uses geometric structural parameters such as gap distance, tower width, and grading ring size and atmospheric parameters such as air temperature, air pressure, and humidity as inputs to establish the Adaboost-SVR model to predict and analyze the impulse discharge voltage of DC transmission towers, and the error is within the engineering allowable range. Related research has verified the feasibility of applying data-driven models to air gap insulation prediction.

The data-driven model mainly focuses on the mathematical statistics and correlation between the gap insulation strength and various influencing factors, ignoring the discharge evolution process full of randomness and uncertainty. The accuracy of the results depends on the input features, sample size and intelligent algorithm. For the small sample problem of air gap discharge voltage prediction, SVM has achieved good prediction results due to its structural risk minimization principle and is currently a more suitable algorithm model. For input features, the key lies in how to characterize the gap structure. Only simple geometric parameters such as electrode size and gap distance are used, which cannot reflect the rich three-dimensional spatial structure of the gap, and it is difficult for the prediction model to achieve good training results in the case of small samples. Since the gap structure corresponds to the electrostatic field distribution one by one, the relevant literature extracts dozens of feature quantities related to electric field strength, energy, gradient, non-uniformity, etc. from the finite element simulation results to describe the gap structure. For the air gap of the transmission line tower, the relevant literature sets a rectangular area between the split conductor and the tower body or crossarm to extract the electric field distribution characteristics, but there is no basis for setting the size of the area. Related literature further simplifies the feature extraction area to the shortest geometric path from the high-voltage end fitting to the tower body, and combines the SVM model to preliminarily realize the discharge voltage prediction of complex tower gaps.

For long air gaps, the electric field distribution on the surface of the high-voltage electrode exceeds the limit value, which is the cause of the discharge initiation, and the development of the discharge through the two poles has a strong correlation with the gap distance. Therefore, when extracting the electric field distribution characteristics, the area near the high-voltage electrode and the inter-electrode path should be fully considered. Based on the previous work, this paper takes the typical gaps of rod-plate and ball-plate as examples, proposes a more reasonable electric field distribution feature set, establishes a prediction model based on the least squares support vector machine (LS-SVM), and optimizes the parameters through the improved gray wolf algorithm, realizing the prediction of the operating impulse discharge voltage of rod-plate and ball-plate long gaps of different geometric sizes.

In this paper, for the prediction of the insulation strength of the rod/ball-plate gap, a feature set to characterize the gap electric field distribution is proposed, and a discharge voltage prediction model based on IGWO optimized LS-SVM is established. The example verification is carried out, and the main conclusions are as follows:

1) A conical field with a vertex angle of θ and a bottom surface of x·U is constructed in the gap with the end of the rod/ball electrode as the vertex, and the shortest path between the electrodes is formed by connecting the high and low voltage electrodes. From it, 66-dimensional feature quantities such as electric field intensity, energy, gradient, non-uniformity, potential, and path length are defined and extracted respectively, which can effectively characterize the influence of the gap structure on the insulation strength of the rod/ball-plate gap.

2) The electric field distribution feature set after dimensionality reduction by the maximum information coefficient method is used as the input parameter of the LS-SVM model, and the model is trained in combination with the IGWO algorithm. The operating impulse discharge voltage of the rod/ball-plate gap can be predicted. The MAPE of the prediction result is 3.2%, the maximum relative error is 8.3%, and the U50-d curve has a similar change trend with the experimental results.