CN-121995294-A - Electronic voltage transformer error prediction method and system

CN121995294ACN 121995294 ACN121995294 ACN 121995294ACN-121995294-A

Abstract

The invention relates to the technical field of electric power, and provides an error prediction method and system of an electronic voltage transformer, wherein the error prediction method comprises the steps of collecting operation data of the electronic voltage transformer; combining physical characteristics and operation and maintenance experience of the electronic voltage transformer, constructing a physical information characteristic set of the electronic voltage transformer, combining physical constraint, constructing a physical loss function, replacing an original loss function of a XGBoost model to obtain a new learning model phy-XGB, inputting a historical error and the physical information characteristic set into the learning model phy-XGB, training to obtain a comparison prediction model of the electronic voltage transformer, calculating an average absolute SHAP value of the physical information characteristic in a time period, and determining the corresponding physical information characteristic needing to be focused. The method improves the accuracy of the measurement error prediction of the electronic voltage transformer, and adopts the SHAP value of each physical information characteristic in the time period to determine the focus of attention so as to guide the actual operation and maintenance.

Inventors

ZHANG ZHU
WANG JINGXUAN

Assignees

合肥工业大学

Dates

Publication Date: 20260508
Application Date: 20260408

Claims (10)

1. The electronic voltage transformer error prediction method is characterized by comprising the following steps of: s1, collecting operation data of an electronic voltage transformer in operation; S2, combining physical characteristics and operation and maintenance experience of the electronic voltage transformer to construct a physical information characteristic set of the electronic voltage transformer, wherein the physical information characteristic set comprises a load temperature sensitivity coefficient LTSC, an environment comprehensive coefficient ECI, a load temperature interaction coefficient LTQI, a load relative fluctuation coefficient RLF and a ratio difference trend coefficient RHT; s3, constructing a physical loss function by combining physical constraints, and replacing the original loss function of the XGBoost model to obtain a new learning model phy-XGB; S4, inputting the history error and the physical information feature set into a learning model phy-XGB, and training to obtain a comparison prediction model of the electronic voltage transformer; S5, calculating an average absolute SHAP value of the physical information features in the time period, and paying attention to the corresponding physical information features according to the descending order.
2. The method of claim 1, wherein the operation data in step 1 includes temperature, humidity, load, magnetic field strength, and historical ratio error values, and the following formula: Wherein, the Represents the temperature of the product and, Represents the humidity level of the water and, Representing the load, the load is represented by, Representing the strength of the magnetic field, Representing a historical ratio error value; for the temperature of the nth sample, For the humidity of the nth sample, For the load of the nth sample, For the magnetic field strength of the nth sample, For the ratio error of the nth sample, Representing the transpose.
3. The method for predicting error of electronic voltage transformer according to claim 2, wherein the load temperature sensitivity coefficient LTSC in step S2 is represented by the following formula: Wherein LTSC i is the load temperature sensitivity coefficient at the ith moment, In order to slide the window in the window, And For the load and temperature sequence within the window, Is a minimum value, avoids the denominator being 0, The average load in the window is used for normalization, eliminating the load and the magnitude difference of the temperature.
4. The method for predicting errors of an electronic voltage transformer according to claim 2, wherein the environmental integrated coefficient ECI in step S2 is as follows: wherein ECI i is the environmental comprehensive coefficient at the ith moment, 、、 Respectively the normalized values of the temperature, the humidity and the magnetic field intensity at the ith moment, 、、 Respectively, their corresponding weights.
5. The method for predicting error of electronic voltage transformer according to claim 2, wherein the load temperature interaction coefficient LTQI in step S2 is as follows: Wherein LTQI i is the load temperature interaction coefficient at the ith moment, 、 And (5) respectively setting standard values of the load and the temperature at the i time.
6. The method for predicting error of electronic voltage transformer according to claim 2, wherein the load relative fluctuation coefficient RLF in step S2 is as follows: Wherein RLF i is the load relative fluctuation coefficient at the ith moment, For a sequence of loads within a window of time, To take the standard deviation of the sequence.
7. The method for predicting error of electronic voltage transformer according to claim 2, wherein the ratio-difference trend coefficient RHT in step S2 is as follows: wherein RHT i is the comparison trend at the ith moment, Is the ratio difference at the i-th moment, Is a sliding window.
8. The method for predicting error of electronic voltage transformer according to claim 1, wherein the physical loss function in step S3 is as follows: Wherein, the The MSE, the original loss function of XGBoost model, cov (,) is covariance, as follows: Wherein, the For the true value of the i-th sample, The final predicted value of the ith sample is obtained, and n is the total number of samples; physical weight, P i is the load of the ith sample, As the average value of the load of the batch, The mean is predicted for the batch of samples.
9. The method for predicting the error of an electronic voltage transformer according to claim 1, wherein the step S5 comprises the steps of: s51, carrying out SHAP moment attribution analysis on a comparison difference prediction model of the electronic voltage transformer, and calculating to obtain SHAP value sets of comparison difference predicted values of the electronic voltage transformer by each physical information characteristic The following formula: Wherein s is any physical information characteristic, f shap-phyXGB is SHAP analysis calculation formula of learning model phy-XGB, and any element An ith time SHAP value which is a physical information feature j; s52, carrying out SHAP time period attribution analysis on a comparison difference prediction model of the electronic voltage transformer, and selecting a target time period That is, the m-th to n-th time points, the average absolute SHAP value of the target period is calculated as follows: Wherein, the An average absolute SHAP value of the physical information characteristic j of the time period; s53, arranging the average absolute SHAP values of the physical information features in a descending order, and determining the corresponding physical information features which need to be focused.
10. An electronic voltage transformer error prediction system is characterized by comprising the following modules: The acquisition module is used for acquiring operation data of the electronic voltage transformer in operation; The physical characteristic construction module is used for combining physical characteristics and operation and maintenance experience of the electronic voltage transformer to construct a physical information characteristic set of the electronic voltage transformer, wherein the physical information characteristic set comprises a load temperature sensitivity coefficient LTSC, an environment comprehensive coefficient ECI, a load temperature interaction coefficient LTQI, a load relative fluctuation coefficient RLF and a comparison trend coefficient RHT; the prediction model construction module is used for combining physical constraint to construct a physical loss function and replacing the original loss function of the XGBoost model to obtain a new learning model phy-XGB; The prediction model training module is used for inputting the history error and the physical information feature set into the learning model phy-XGB, and training to obtain a comparison prediction model of the electronic voltage transformer; and the SHAP value calculation module is used for calculating the average absolute SHAP value of the physical information characteristic in the time period, arranging the average absolute SHAP values in a descending order, and determining the physical information characteristic corresponding to the larger average absolute SHAP value as the important attention object.

Description

Electronic voltage transformer error prediction method and system Technical Field The invention relates to the technical field of electric power, in particular to an electronic voltage transformer error prediction method and system. Background The electronic voltage transformer is key equipment for voltage measurement and signal conversion in an electric power system, and is a core component for sensing a voltage state, wherein the capacitive voltage division type electronic voltage transformer (electronic voltage transformer) is widely applied to a high-voltage power grid of 110kV and above due to high insulation strength and low cost. However, the measurement error of the electronic voltage transformer is easily interfered by various factors, so that the long-term stability is difficult to maintain, and the safe operation and the fair settlement of the power grid are directly affected. Therefore, if the measurement error of the electronic voltage transformer can be perceived in advance and diagnosis is implemented, on one hand, metering disputes and relay protection misoperation caused by the measurement error of the electronic voltage transformer can be avoided in time, on the other hand, unnecessary operation and maintenance cost and power failure risk can be reduced, and finally, important support is provided for stable, economical and reliable operation of the power system. However, the error sensing research of the measuring equipment is more real-time, so that the dilemma of the traditional offline error sensing method is solved to a great extent. For example, chinese patent application CN118131109a, entitled "a capacitive voltage transformer error online self-detection method and system", proposes an error online monitoring method that is separated from a standard device, and determines a ratio difference change direction of a transformer through a bilateral threshold, but such a method can only reflect a current measurement error, only map a current operation state of a measurement device, and is difficult to perceive the measurement error in advance, and is more difficult to master a future operation state of the measurement device in advance. Meanwhile, error early sensing and system reliability analysis methods for similar devices are mainly divided into physical driving and data driving methods. Although the physical driving method can approximately sense the change direction of the measurement error in advance, the accurate error value is difficult to output, and direct and effective decision support cannot be provided for actual operation and maintenance. In addition, a huge and perfect knowledge base is needed for complete description of the physical change process of the equipment, which is difficult to realize in actual modeling, model parameters are needed to be determined through a series of assumptions, and thus, the modeling errors cannot be completely eliminated. A pure data driving method, for example, a prediction method based on a TCN network, which is disclosed in China patent application No. CN119862791A, an electronic voltage transformer error prediction method and system, has poor effect under actual working conditions, and the system precision is easily affected by system fluctuation, environment and equipment noise. The explanation for this error is that most students rely solely on conventional observations and do not incorporate any understanding of the internal physical mechanisms that produce the observation data system. In conclusion, the method for sensing the measurement error of the electronic voltage transformer in advance is less. The method for sensing the measurement errors of the mutual inductor is divided into physical driving and data driving, wherein the simple physical modeling cannot determine future error values, a complete knowledge base is difficult to construct and cannot support practical engineering application, and the simple data driving lacks physical rule constraint, is easy to be interfered by noise and has insufficient robustness and precision. And traditional machine learning and deep learning are often accompanied with a 'black box' dilemma, so that the influence of each characteristic is difficult to explain, and a targeted suggestion is difficult to provide for operation and maintenance. Disclosure of Invention The technical problem to be solved by the invention is how to improve the accuracy of the measurement error prediction of the electronic voltage transformer, and to adopt the SHAP value of each physical information characteristic in a time period to determine the focus of attention so as to guide the actual operation and maintenance. The invention solves the technical problems by the following technical means: the invention provides an electronic voltage transformer error prediction method, which comprises the following steps: s1, collecting operation data of an electronic voltage transformer in operation; S2, combining physical chara