Transformer state monitoring method, device, equipment and storage medium

Abstract

This invention relates to the field of transformer monitoring, and more particularly to a method, apparatus, equipment, and storage medium for transformer condition monitoring. The method includes the following steps: collecting raw transformer data in its operating state to construct a multi-source monitoring data pool; physically decoupling and layering the multi-source monitoring data pool to construct a multi-dimensional monitoring feature set; tracking changes in the multi-dimensional monitoring feature set using a sliding time window to construct change curves for multiple data types; performing time-by-time limit detection on the change curves and marking exceeding-limit monitoring data points; and assessing the operational risk level and visualizing the exceeding-limit monitoring data points to construct a multi-parameter visualization view. This invention provides comprehensive and accurate perception of the transformer's operating status, improving the safety, reliability, and maintenance efficiency of transformer operation.

Description

Technical Field

[0001] This invention relates to the field of transformer monitoring, and in particular to a transformer condition monitoring method, device, equipment, and storage medium. Background Technology

[0002] During the long-term operation of transformers, the internal insulation materials, winding structures, and core components gradually age, deform, or degrade due to the combined effects of electrical, thermal, and mechanical stresses. Environmental temperature changes, load fluctuations, and external impacts can also induce potential faults such as partial discharge, overheating, and abnormal vibration. If these hidden dangers are not detected and effectively addressed in a timely manner, they can easily evolve into serious equipment failures, leading to power outages and even significant economic losses and safety risks. Continuous and accurate monitoring and assessment of transformer operating status has become a critical issue for the safe operation of power systems. Traditional transformer condition monitoring methods mainly rely on detection of single or limited parameters, such as oil temperature monitoring, load current monitoring, and periodic manual inspections. While these methods can reflect the basic operating status of the equipment to some extent, the sources of monitoring information are limited, making it difficult to comprehensively depict the complex operating characteristics of transformers. Furthermore, different fault modes often manifest themselves covertly in their early stages, making accurate identification difficult with a single detection method, easily leading to missed detections or misjudgments. Traditional monitoring methods are mostly offline or low-frequency detection, which cannot meet the real-time and refined management requirements of modern power systems. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention proposes a transformer condition monitoring method, device, equipment, and storage medium, thereby resolving at least one of the aforementioned technical problems.

[0004] To achieve the above objectives, the present invention provides a transformer condition monitoring method, comprising the following steps: Step S1: Collect raw data of the transformer in operation and construct a multi-source monitoring data pool; Step S2: Physically layer and decouple the multi-source monitoring data pool to construct a multi-dimensional monitoring feature set; Step S3: Perform sliding window change tracking on the multidimensional monitoring feature set to construct change curves for multiple data types; Step S4: Perform time-by-time limit detection on the change curve and mark the data points exceeding the limit; Step S5: Assess the operational risk level and perform visualization rendering on the over-limit monitoring data points to construct a multi-parameter visualization view.

[0005] This specification provides a transformer condition monitoring device for performing the transformer condition monitoring method described above, comprising: The data acquisition unit is used to collect raw data of the transformer in operation and build a multi-source monitoring data pool. Hierarchical units are used to physically decouple multi-source monitoring data pools and construct multi-dimensional monitoring feature sets. The tracking unit is used to track changes in a sliding time window of a multidimensional monitoring feature set and construct change curves for multiple data types. The deviation calculation unit is used to perform time-by-time limit detection on the change curve and mark the limit monitoring data points; The visualization unit is used to assess the operational risk level and render the visualization of the over-limit monitoring data points, and to construct a multi-parameter visualization view.

[0006] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the transformer condition monitoring method described in any of the preceding claims.

[0007] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the transformer condition monitoring method described in any of the preceding claims.

[0008] The beneficial effects of this invention are as follows: By collecting multi-source raw data such as voltage, current, temperature, oil level, vibration, and partial discharge, the actual operating status of the transformer can be comprehensively reflected from multiple dimensions, including electrical, thermal, and mechanical aspects, avoiding information gaps caused by monitoring a single parameter. Constructing a multi-source monitoring data pool facilitates centralized management and synchronous processing of data from different sources and sampling frequencies, laying a data foundation for subsequent feature extraction and risk analysis. Through physical layering and decoupling, data on electrical, thermal, and mechanical characteristics are processed in layers, avoiding mutual interference between different physical quantities and improving the accuracy of feature representation. Constructing a multi-dimensional monitoring feature set transforms complex raw data into feature parameters with clear physical meaning, facilitating subsequent change analysis and risk assessment. Using a sliding time window to continuously track monitoring features effectively captures the changing trends of parameters over time, avoiding reliance solely on instantaneous values ​​to determine the operating status. Change curves can reflect slowly evolving abnormal characteristics, such as abnormal temperature rise and insulation performance degradation, which is beneficial for early detection of potential risks. Change curves of different types of data provide intuitive and continuous analytical basis for subsequent limit-over-limit detection and trend judgment. By detecting limit exceedances at specific times, anomalies can be identified and marked promptly when monitored parameters exceed safety thresholds, improving the real-time performance of the monitoring system. Limit exceedance judgment based on change curves helps distinguish between instantaneous fluctuations and continuous anomalies, avoiding false alarms caused by short-term noise. Risk level classification of exceeding data points transforms complex operating states into intuitive risk levels, facilitating rapid assessment of equipment health by maintenance personnel. Multi-parameter visualization views visually present risk levels, abnormal parameters, and their changing trends, improving information acquisition efficiency. Attached Figure Description

[0009] Figure 1 This is a schematic flowchart of the steps of a transformer condition monitoring method according to the present invention; Figure 2 This is a detailed flowchart illustrating the implementation steps of step S1. Figure 3 This is a flowchart illustrating the detailed implementation steps of step S2. Detailed Implementation

[0010] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0011] This application provides a transformer condition monitoring method, apparatus, device, and storage medium. The executing entities of the transformer condition monitoring method, apparatus, device, and storage medium include, but are not limited to, mechanical equipment, data processing platforms, cloud server nodes, and network upload devices mounted on the system, which can be considered as general computing nodes in this application. The data processing platform includes, but is not limited to, at least one of an audio / image management system, an information management system, and a cloud data management system.

[0012] Please see Figures 1 to 3 This invention provides a transformer condition monitoring method, comprising the following steps: Step S1: Collect raw data of the transformer in operation and construct a multi-source monitoring data pool; Step S2: Physically layer and decouple the multi-source monitoring data pool to construct a multi-dimensional monitoring feature set; Step S3: Perform sliding window change tracking on the multidimensional monitoring feature set to construct change curves for multiple data types; Step S4: Perform time-by-time limit detection on the change curve and mark the data points exceeding the limit; Step S5: Assess the operational risk level and perform visualization rendering on the over-limit monitoring data points to construct a multi-parameter visualization view.

[0013] In the embodiments of the present invention, see Figure 1 The diagram below illustrates the steps of a transformer condition monitoring method according to the present invention. In this example, the steps of the transformer condition monitoring method include: Step S1: Collect raw data of the transformer in operation and construct a multi-source monitoring data pool; In this embodiment, multimodal sensors deployed at key locations of the transformer continuously collect raw operating data, including electrical signals (current, voltage, harmonics), thermal signals (winding temperature, oil temperature), mechanical vibration signals (axial vibration, lateral vibration, modal vibration frequencies), acoustic signals (acoustic energy and spectral distribution), and gas chemical signals (concentration and rate of change of gas components in oil). Each type of sensor is configured with an appropriate sampling frequency based on its physical characteristics. For example, the vibration sensor uses a sampling frequency of 10 kHz to capture rapid mechanical vibrations, the temperature sensor uses a sampling frequency of 1 Hz to reflect slow thermal changes, and the current and voltage sensors use a sampling frequency of 5 kHz to ensure complete harmonic details. During the acquisition process, to ensure data synchronization, the clocks of each sensor are calibrated and a global timestamp is generated to ensure time alignment of cross-modal data. After time alignment, all sensor data is aggregated into a multi-source monitoring data pool, with each data point appended with a timestamp, sensor type, and original measurement value.

[0014] Step S2: Physically layer and decouple the multi-source monitoring data pool to construct a multi-dimensional monitoring feature set; In this embodiment, current, voltage, and partial discharge data are classified into an electrical layer; temperature and oil temperature data into a thermal layer; vibration signals into a mechanical layer; acoustic signals into an acoustic layer; and oil gas composition data into a gas chemistry layer. Statistical analysis and signal processing are performed on the data from each physical layer, including the mean, variance, instantaneous peak value, and frequency domain power spectral density, to form preliminary feature vectors. Subsequently, wavelet transform or Fourier transform is performed on the data from each physical layer to extract time-frequency features, thereby obtaining transient anomalies and frequency distribution information in the time series.

[0015] Step S3: Perform sliding window change tracking on the multidimensional monitoring feature set to construct change curves for multiple data types; In this embodiment, a sliding time window method is used to track temporal changes and extract the dynamic variation patterns of features over time. The time window length (e.g., 10–30 seconds) and step size (e.g., 1–5 seconds) are set, and the mean, increment, rate of change, and cumulative change are calculated for the feature values ​​within each window, forming a temporal change increment sequence. Variation curves are generated for each physical layer feature, including temperature gradient curves, current harmonic amplitude curves, vibration mode energy curves, acoustic signature energy curves, and gas composition variation curves. Each variation curve reflects the evolution trend of a specific physical feature over time, facilitating the identification of sudden anomalies or long-term trend changes. For example, when the load suddenly increases, the temperature gradient curve rises rapidly within minutes, while the partial discharge energy curve fluctuates periodically within the same time window.

[0016] Step S4: Perform time-by-time limit detection on the change curve and mark the data points exceeding the limit; In this embodiment, an adaptive threshold method is used to perform time-by-time limit detection on the characteristics at each time point. The threshold is set based on the transformer's approved operating information (rated capacity, rated voltage, insulation class, and environmental boundary conditions) to establish a standard operating range, which is then dynamically adjusted according to the operating conditions to form an adaptive adjustment range. For example, the upper limit of winding temperature rise can be increased by 2–3 °C as the ambient temperature rises, and the upper limit of vibration mode is adjusted in real time according to load changes. If the characteristic value at each time point exceeds the adaptive upper and lower limits, it is marked as an over-limit monitoring data point, and the timestamp, characteristic type, and deviation magnitude are recorded. A continuous over-limit judgment rule can be introduced during the detection process, for example, only when 5 consecutive sampling points exceed the limit is it judged as a valid anomaly, in order to avoid instantaneous noise interference. Through this time-by-time limit detection method, a high-precision over-limit monitoring data set can be formed.

[0017] Step S5: Assess the operational risk level and perform visualization rendering on the over-limit monitoring data points to construct a multi-parameter visualization view.

[0018] In this embodiment, after obtaining the out-of-limit monitoring data points, their deviation and time-series evolution patterns are combined with historical data and safety rules to assess the operational risk level. The assessment method combines the deviation magnitude, duration, and fluctuation intensity to classify the out-of-limit state into three or five levels of risk: low, medium, and high. A temperature rise curve that continuously exceeds the threshold and gradually widens the deviation is determined to be high risk; a partial discharge curve with short-term fluctuations and rapid decline is determined to be medium risk; and occasional transient vibration deviation is determined to be low risk. Subsequently, the curve segments with different physical characteristics are differentiated and visualized: high-risk segments are displayed in red and highlighted, medium-risk segments are displayed in orange, and low-risk segments are displayed in green or semi-transparent. All curves are then superimposed on a unified time axis to generate a multi-parameter visualization view.

[0019] In this embodiment, see Figure 2 The diagram below illustrates the detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include: The transformer's raw data on its operating status is collected synchronously using multimodal sensors; Calculate the acquisition clock information of various sensors for the raw data of the transformer; A global timestamp reference is set based on the acquired clock information; the original transformer data is time-aligned based on the global timestamp reference to construct the first monitoring data; Self-noise identification is performed on the multimodal sensor to obtain the self-noise frequencies of multiple sensors; The first monitoring data is adaptively filtered based on the self-noise frequency to obtain the second monitoring data; Multi-source data fitting is performed on the second monitoring data to construct a multi-source monitoring data pool.

[0020] In this embodiment, during transformer operation status monitoring, raw operating data is simultaneously acquired using multi-modal sensors to comprehensively reflect the transformer's electrical and mechanical operating status. The acquired sensor types include data from multiple dimensions such as voltage, current, partial discharge, temperature, vibration, and oil-immersed medium gas concentration. Each sensor continuously acquires data at a fixed sampling frequency; for example, the vibration sensor samples at 10 kHz, the temperature sensor at 1 Hz, and the voltage and current sensors at 5 kHz. This synchronous acquisition mechanism ensures initial alignment of different modal data in the time dimension, while providing a high-precision raw data foundation for subsequent data fusion and intelligent visualization. During acquisition, sensor output signals are recorded in real time and appended with original acquisition timestamps to ensure traceability of each data point, forming a multi-modal transformer raw data set. After completing multi-modal data acquisition, the data acquisition clock of each sensor is analyzed to determine its specific time reference and sampling accuracy. This process obtains the acquisition clock frequency, start time, and drift characteristics by reading the sampling time information recorded in the sensor's internal clock register or the acquisition hardware. For example, the clock deviation of a vibration sensor might be ±5 ppm, and that of a temperature sensor might be ±10 ppm. By statistically analyzing the clocks of each sensor, we can identify time drift, delay, or inconsistent sampling frequencies between different modes. The acquired clock information provides a basis for establishing a global timestamp reference, ensuring accurate temporal matching of data from different modes, and also provides a synchronization reference for noise identification and adaptive filtering.

[0021] Based on the clock information collected by each sensor, a unified global timestamp reference is established, and the original data is time-aligned using linear interpolation or timestamp correction methods. For example, vibration sensor data with a higher sampling frequency and temperature data with a lower sampling frequency are mapped to a unified time grid, and the temperature data is linearly interpolated to each unified time point. For sensor data with sampling drift, the original timestamp is first corrected according to the clock deviation, and then aligned to ensure that data from different sensors at the same time point can accurately correspond. After processing, a first monitoring data set is formed. This data set is synchronized in the time dimension, with a unified data format, facilitating subsequent noise analysis, filtering, and multi-source fusion processing, while retaining the original sensor attributes and identification information. Based on the first monitoring data, self-noise analysis is performed on each sensor data to identify the inherent noise characteristics of the sensor itself. This step uses frequency domain analysis methods, such as Fourier transform or power spectral density analysis, to extract periodic or broadband noise components from the sensor output signal. For vibration sensors, their own mechanical vibration frequency can be identified, such as natural frequencies in the range of 50 Hz–200 Hz; for current and voltage sensors, high-frequency interference noise generated by the acquisition circuit can be extracted, such as spike signals above 100 kHz.

[0022] Based on the inherent noise frequencies identified by each sensor, the first monitoring data undergoes adaptive filtering to remove interference from the inherent noise of the sensors on signal analysis. Filtering methods include band-stop filtering, low-pass filtering, or adaptive Kalman filtering, with appropriate filtering parameters selected according to the noise characteristics of different sensors. For example, periodic noise with an inherent frequency of 120 Hz identified in the vibration sensor is suppressed by a band-stop filter while retaining the load vibration characteristics of 10–50 Hz; high-frequency spike noise in the current sensor is suppressed by a low-pass filter. After obtaining high-quality second monitoring data, the data from different modes are jointly fitted to construct a multi-source monitoring data pool. This step uses multivariate fitting methods, such as least squares, weighted regression, or Bayesian-based multi-source data fusion, to jointly model the data from different sensors and extract transformer state characteristics and potential correlations. For example, vibration, temperature, and partial discharge signals are jointly fitted to obtain the implicit patterns of local overheating or partial discharge trends in the transformer, and the reliability of the sensor data is evaluated through the fitting residuals.

[0023] In this embodiment, see Figure 3 The diagram below illustrates the detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include: Physically layer and decouple the multi-source monitoring data pool to obtain monitoring data of multiple physical types; The monitoring data is subjected to feature analysis to obtain multidimensional physical characteristics; the multidimensional physical characteristics include current harmonic components, voltage fluctuation amplitude, temperature rise gradient distribution, acoustic spectrum energy density, gas component concentration change rate, and vibration modes; Wavelet transform is performed on the multidimensional physical features to extract feature vectors; the feature vectors include time-domain statistical features and frequency-domain energy distribution features. Feature dimensionality reduction and key feature selection are performed on the feature vectors to construct a multi-dimensional monitoring feature set.

[0024] In this embodiment, the physical characteristics of sensor data from different modes are analyzed, and the raw data are classified into electrical, thermal, mechanical, and acoustic types. For example, data from current sensors, partial discharge sensors, and voltage sensors are classified into the electrical layer; data from temperature sensors and oil temperature sensors are classified into the thermal layer; vibration sensor data into the mechanical layer; acoustic sensor data into the acoustic layer; and data from oil gas sensors into the gas chemistry layer. The data from each physical layer is independently processed using time series analysis to ensure that the data from each layer remains consistent with the original monitoring signal in both time and space. Through physical layering and decoupling, abnormal signals caused by different physical mechanisms can be independently evaluated during subsequent feature extraction and analysis, reducing cross-modal interference.

[0025] Various monitoring data are analyzed for physical characteristics to extract multidimensional physical features. Electrical layer data is analyzed in the frequency domain to extract current harmonic components and voltage fluctuation amplitudes, quantifying electrical load status and power supply interference characteristics. Thermal layer data is analyzed using gradient analysis and time series fitting to extract temperature rise gradient distribution, reflecting the spatial heat conduction characteristics of windings and oil temperature. Acoustic layer data is analyzed using spectrum analysis and energy density calculation to extract acoustic signature spectrum energy density, reflecting acoustic anomalies caused by partial discharge and mechanical vibration. Gas chemical layer data is analyzed using differential analysis and rate of change calculation to extract the rate of change of gas component concentrations, reflecting partial discharge or oil decomposition characteristics. Mechanical layer data is analyzed using modal analysis to extract vibration mode characteristics, quantifying the mechanical structure response state. Discrete wavelet transform is used to extract energy distributions in different frequency bands from current harmonic signals and voltage fluctuation signals; wavelet decomposition is applied to the temperature rise gradient and vibration mode time series to extract fluctuation amplitudes at various scales; and instantaneous energy and rate of change characteristics are extracted from the acoustic signature spectrum and gas concentration change series. By using wavelet decomposition, each physical feature is mapped to a multidimensional feature vector, including time-domain statistical features (mean, variance, skewness, kurtosis) and frequency-domain energy distribution features (energy ratio and local energy density at different wavelet scales), forming a high-dimensional feature set that can simultaneously reflect the instantaneous state and long-term change patterns.

[0026] The original feature vectors are dimensionality reduced to eliminate redundant and highly correlated features, while retaining key indicators sensitive to transformer state changes. For example, correlation analysis is performed on current harmonics and voltage fluctuation energy characteristics, retaining principal components with a contribution exceeding 85%; for temperature rise gradient and vibration mode characteristics, features best reflecting local hotspots or mechanical anomalies are selected based on variance contribution rates; for acoustic waveform spectra and gas change rates, features with significant changes or abnormal response frequency ranges are selected. Through dimensionality reduction and key feature selection, a multi-dimensional monitoring feature set is finally constructed. Each sample contains core indicators of electrical, thermal, mechanical, acoustic, and gas chemistry, achieving data compression while preserving state information.

[0027] In this embodiment, step S3 includes the following steps: Set the sampling interval for parameter changes at different time points; Based on the sampling interval, a sliding window change tracking method is used to track the multidimensional monitoring feature set and extract the time-series change increment sequence. Calculate the first-order difference magnitude and second-order difference curvature of the time-series change increment sequence; Based on the first-order difference change amplitude and the second-order difference change curvature, polynomial fitting is performed to construct change curves for multiple data types.

[0028] In this embodiment, the sampling interval is set by analyzing the typical operating condition fluctuation cycle of the transformer and the sensor sampling characteristics. For example, temperature changes are usually slow processes with a typical timescale of 1–5 minutes, while transient changes in partial discharge, current harmonics, or vibration signals may occur in the millisecond to second range. To ensure that both slow trends and rapid disturbances can be captured simultaneously, a sliding sampling interval needs to be set on the multidimensional feature set, which can be a combination of multiple scales such as 1 second, 10 seconds, or 1 minute. The sampling interval not only affects the accuracy of subsequent sliding window feature extraction but also determines the temporal resolution and data processing efficiency of the time-series incremental sequence. After determining the sampling interval, the sliding window method is used to track time-series changes on the multidimensional monitoring feature set. The sliding window is usually set as a window of several consecutive sampling points, for example, a window length of 5–10 sampling intervals and a window step size of 1–2 sampling intervals. The feature change is calculated using the data within the window. For each window, the time-series change increment is extracted by calculating the difference between the mean of the current window feature and the mean of the previous window, forming an incremental sequence. Incremental sequences can quantify the rate and trend of change of different physical layer characteristics over time. For example, the temperature gradient increases by 0.2–0.5 ℃ / min over time, and the vibration mode energy increases by 0.05–0.1 m / s² with the change of load.

[0029] After obtaining the time-series increment sequence, the characteristic changes are quantitatively described through first-order and second-order difference analysis. First-order difference calculates the amplitude of continuous increment changes, reflecting the rate of change of the characteristic; for example, the first-order difference of current harmonic energy can show the instantaneous fluctuation intensity, and the first-order difference of temperature gradient shows the heating rate. Second-order difference calculates the curvature of change, describing the acceleration or concavity / convexity of the characteristic increment changes, and can capture local nonlinear or abrupt trends, such as the second-order difference peak generated by vibration energy during load abrupt changes. By performing first-order and second-order differences on characteristics of different physical layers respectively, a high-dimensional difference matrix is ​​formed, providing refined dynamic features for trend analysis and anomaly detection. The difference calculation process incorporates time interval standardization, so that the first-order difference represents the amplitude of change per unit time, and the second-order difference represents the rate of change per unit time, facilitating unified comparison and comprehensive analysis of features across physical layers.

[0030] Discrete first-order and second-order difference data are converted into continuously varying curves using polynomial fitting methods. Polynomial fitting employs polynomials of different orders (such as quadratic or cubic) to fit the difference sequence of each physical feature, smoothing instantaneous fluctuations and highlighting long-term trends. For example, using cubic polynomial fitting for the incremental curve of the temperature rise gradient can reveal a slow heating trend and sudden abnormal nodes; using quadratic polynomial fitting for the current harmonic difference sequence can display the peaks and troughs of periodic fluctuations. A continuously varying curve is generated for each physical type feature, including curves for the electrical, thermal, acoustic, mechanical, and gaseous chemical layers. The curves reflect the incremental changes and curvature trends over time, providing intuitive representation for multi-source visualization, anomaly detection, and condition assessment.

[0031] In this embodiment, step S4 includes the following steps: Identify the transformer's rated operating information; the rated operating information includes rated capacity, rated voltage level, insulation class, and operating environment boundary conditions; Based on the verified work information, the upper and lower limits of the standard work threshold are set to obtain the standard work range; The standard working range is dynamically corrected based on the operating conditions to obtain the adaptive adjustment range; The change curve is subjected to time-by-time limit detection based on the adaptive adjustment interval, and the data points exceeding the limit are marked.

[0032] In this embodiment, the rated operating information of the transformer is clearly defined to determine its design and safe operation constraints. The rated operating information includes the transformer's rated capacity, rated voltage level, insulation class, and operating environment boundary conditions. Rated capacity typically refers to the maximum power the transformer can continuously carry under rated load conditions, such as 500 kVA or 1000 kVA; rated voltage level refers to the main circuit design voltage, such as 10 kV or 35 kV; insulation class refers to the withstand voltage capacity and thermal stability class of the windings, core, and casing, such as Class A, Class F, or Class H, which directly determines the winding temperature rise and overload tolerance; operating environment boundary conditions include the ambient temperature range (-20~40 ℃), humidity range, and installation height, which affect the transformer's thermal stability and local insulation safety. After clarifying the rated operating information, standard operating threshold upper and lower limits are set according to the rated values ​​of each physical indicator and the design safety margin, forming a standard operating range. For example, the winding temperature rise threshold can be set according to the insulation class and rated capacity. The upper limit is usually the rated temperature rise value plus a safety margin (e.g., a rated temperature rise of 65 ℃ corresponds to an upper limit of 70 ℃), and the lower limit can be set to the lowest ambient temperature plus a safety margin. The upper limit of oil temperature is set to 85–90 ℃ according to the oil's heat resistance grade. The upper and lower limits of current and voltage thresholds are set according to the rated current and voltage levels. For example, a rated current of 500 A can allow ±10% fluctuation, forming a standard operating range of 450–550 A. Partial discharge and vibration amplitude are set according to empirical safety ranges. By setting upper and lower limits for each type of physical quantity, a standard operating range covering multiple dimensions such as electrical, thermal, mechanical, and acoustics is formed.

[0033] Based on the standard operating range, considering the impact of actual transformer operating conditions and environmental changes on thresholds, an adaptive adjustment range is formed through dynamic correction. This step combines real-time monitoring data and environmental factors. For example, when the ambient temperature rises to 35 ℃, the upper limit of the winding temperature rise can be appropriately increased by 2–3 ℃; when the load is consistently low, the lower limit of the temperature rise can be lowered to avoid misjudging abnormalities; vibration amplitude and partial discharge thresholds can be dynamically corrected based on current load fluctuations, switching operation frequency, and operating history. The dynamic correction method includes a linear correction model or an adaptive correction model based on empirical rules, comprehensively considering environmental factors, load status, and historical trends to adjust the upper and lower limits of the standard operating range so that the thresholds can reflect the actual safety limits under real-time operating conditions, forming an adaptive adjustment range and achieving intelligent adaptation to the transformer state boundaries. After obtaining the adaptive adjustment range, the aforementioned change curves (including current harmonics, temperature rise gradient, vibration modes, acoustic energy, and gas composition curves) are compared with the adaptive thresholds point by point over time to achieve time-by-time limit exceedance detection. For each type of physical characteristic at each time point, it is determined whether it exceeds the upper or lower limit of the adaptive range. For example, if the temperature curve reaches the upper limit of 85 ℃ at a certain moment, it is marked as an out-of-limit point; if the vibration modal energy exceeds the threshold of 0.15 m / s², it is also marked as an out-of-limit point. The out-of-limit detection not only considers the instantaneous peak value, but can also combine the duration threshold to determine the abnormal trend. For example, it is only marked as abnormal if 5 consecutive sampling points exceed the limit, avoiding false judgments due to noise.

[0034] In this embodiment, step S4 includes the following steps: The deviation of the over-limit monitoring data points is calculated to obtain the over-limit deviation. Perform time-series statistics on the out-of-limit deviation and extract deviation sequences at multiple time points; The mean, variance, and peak value of the deviation sequence are calculated, and trend discrimination is performed to obtain the deviation evolution pattern; the deviation evolution pattern includes continuous expansion of deviation, oscillating deviation, and gradual smooth convergence; Conduct an operational risk level assessment of deviations from the evolutionary model and generate risk assessment results; Based on the risk assessment results, a multi-parameter visualization view is constructed through visualization rendering; the multi-parameter visualization view is then uploaded to a cloud platform for storage.

[0035] In this embodiment, a standardized deviation index is formed by calculating the deviation degree, which is the ratio or absolute difference between the actual measured value and the upper or lower limit of the adaptive threshold. For example, if the transformer winding temperature rises to 88 ℃ at a certain moment, and the adaptive upper limit is 85 ℃, then the deviation degree can be defined as... This reflects the proportion of the out-of-limit amplitude to the threshold; the vibration mode energy deviation can be calculated similarly as... Deviation calculation not only quantifies the magnitude of anomalies but can also be used for subsequent trend analysis and risk assessment. Deviation is calculated independently for each physical feature, forming an out-of-limit deviation set. The data format includes timestamps, feature types, deviation magnitudes, and deviation directions (positive or negative). A sliding time window is used to statistically analyze the deviation changes of each physical feature over continuous time periods; for example, a deviation sequence is calculated every 5 minutes, forming a time series data matrix. The sequence records the maximum deviation, average deviation, and cumulative deviation count at each time point, reflecting the persistence, volatility, and cumulative effects of the anomaly. For example, in a temperature rise anomaly, if the deviation has been continuously increasing over the past 30 minutes, the deviation sequence shows a clear upward trend; for partial discharge fluctuations, the deviation sequence may exhibit periodic oscillations.

[0036] Statistical analysis methods are used to extract sequence features, including mean, variance, and peak value, to assess the overall level of deviation, fluctuation intensity, and anomalies in instantaneous peak values. The mean deviation reflects the persistence of the overall over-limit deviation, the variance reflects the volatility or stability of the over-limit, and the peak value represents the maximum instantaneous risk. For example, a mean temperature rise deviation of 0.03, variance of 0.002, and peak value of 0.05 indicates a moderate overall deviation but with short-term peak values; a variance of 0.01 and a peak value of 0.08 for vibration deviation indicates significant fluctuations and a potential risk of mechanical resonance. Based on these statistics, trend discrimination is performed, classifying deviation evolution patterns into typical types: continuously expanding deviation (mean and peak value gradually increase), oscillating fluctuation deviation (large variance, periodic changes in peak value), and gradually smoothing convergence (mean and peak value gradually decrease). The deviation patterns are matched with transformer operation safety rules and historical fault data to achieve operational risk level assessment. The assessment method can quantify the risk level based on the deviation magnitude, duration, fluctuation intensity, and peak value. For example, a continuously expanding deviation pattern with a peak deviation exceeding 0.05 can be classified as a high-risk state; an oscillating deviation pattern with a peak deviation between 0.02 and 0.05 can be classified as a medium-risk state; and a gradually smoothing convergence pattern with a mean deviation below 0.01 can be classified as a low-risk state. This quantitative risk assessment method converts multidimensional deviation information into single or multidimensional risk level indicators, including three-level (low, medium, high) or five-level (high) risk levels, forming a risk assessment result set. Data records include timestamps, physical characteristic types, deviation evolution patterns, and risk levels.

[0037] In this embodiment, the specific steps for performing visualization rendering based on the risk assessment results to construct a multi-parameter visualization view, and uploading the multi-parameter visualization view to the cloud platform for storage are as follows: Based on the risk assessment results, the change curves are time-matched, and the corresponding curve segments are marked. The corresponding curve segments are subjected to differentiated visualization processing, and visualization rendering is performed based on the changing curves of multiple data types to construct a multi-parameter visualization view; The multi-parameter visualization view is uploaded to the cloud platform for storage.

[0038] In this embodiment, by comparing the timestamps of the risk assessment results with the sampling time points of the change curves, the risk level at each time point is mapped to the corresponding change curve. For example, if the deviation of the winding temperature rise from its evolution pattern is determined to be high-risk during the period from 10:05 to 10:20, the temperature rise change curve for that period is marked as a high-risk segment; if the partial discharge peak is determined to be medium-risk during the period from 10:10 to 10:12, the corresponding curve segment is marked as medium-risk. During the time matching process, the difference in sampling intervals needs to be considered. For characteristic curves with different sampling frequencies, linear interpolation or time window alignment methods can be used to accurately map the risk results to high-resolution curve points. After marking the risk time periods, the curve segments corresponding to different risk levels are subjected to differentiated visualization processing, allowing users to intuitively distinguish the intensity and type of anomalies. Differentiation processing includes color coding, line shape changes, or transparency adjustments; for example, high-risk segments are displayed in red, medium-risk segments in orange, and low-risk segments in green. Simultaneously, the duration and deviation of the risk can be represented by line width or dashed lines. Subsequently, all physical characteristic curves (such as current harmonics, voltage fluctuations, temperature rise gradients, acoustic energy, vibration modes, and gas concentration changes) are overlaid on a unified visualization panel. A multi-parameter visualization view is constructed using axis synchronization, time alignment, and multi-layer rendering methods. For example, the left vertical axis displays the temperature gradient and oil temperature curves, the right vertical axis displays the vibration modes and partial discharge characteristics, and the horizontal axis is uniformly time-based. Before uploading, the visualization data undergoes standardization processing, including resolution optimization, layer compression, and metadata annotation, such as recording risk level zones, physical characteristic types, and timestamp ranges. The upload process ensures data integrity and access permissions through security protocols, employing encrypted transmission and version management methods. Each visualization rendering result is associated with a unique identifier to ensure historical data traceability. Cloud platform storage not only supports real-time visualization but also allows for cross-time and cross-site data comparison and analysis, providing a long-term data accumulation foundation for transformer operation monitoring and management.

[0039] In this embodiment, a transformer condition monitoring device is provided for executing the transformer condition monitoring method described above, including: The data acquisition unit is used to collect raw data of the transformer in operation and build a multi-source monitoring data pool. Hierarchical units are used to physically decouple multi-source monitoring data pools and construct multi-dimensional monitoring feature sets. The tracking unit is used to track changes in a sliding time window of a multidimensional monitoring feature set and construct change curves for multiple data types. The deviation calculation unit is used to perform time-by-time limit detection on the change curve and mark the limit monitoring data points; The visualization unit is used to assess the operational risk level and render the visualization of the over-limit monitoring data points, and to construct a multi-parameter visualization view.

[0040] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the transformer condition monitoring method described in any of the preceding claims.

[0041] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the transformer condition monitoring method described in any of the preceding claims.

[0042] In all respects, the embodiments should be regarded as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Therefore, it is intended that all variations falling within the meaning and scope of the equivalents of the application be included within the present invention.

[0043] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein are implemented in other embodiments without departing from the spirit or scope of the invention. The present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method of transformer condition monitoring, characterized by, Includes the following steps: Step S1: Collect raw data of the transformer in operation and construct a multi-source monitoring data pool; Step S2: Physically layer and decouple the multi-source monitoring data pool to construct a multi-dimensional monitoring feature set; Step S3: Perform sliding window change tracking on the multidimensional monitoring feature set to construct change curves for multiple data types; Step S4: Perform time-by-time limit detection on the change curve and mark the data points exceeding the limit; Step S5: Assess the operational risk level and perform visualization rendering on the over-limit monitoring data points to construct a multi-parameter visualization view.

2. The transformer condition monitoring method of claim 1, wherein, The specific steps of step S1 are as follows: The transformer's raw data on its operating status is collected synchronously using multimodal sensors; Calculate the acquisition clock information of various sensors for the raw data of the transformer; A global timestamp reference is set based on the acquired clock information; The original transformer data is time-aligned based on a global timestamp reference to construct the first monitoring data. Self-noise identification is performed on the multimodal sensor to obtain the self-noise frequencies of multiple sensors; The first monitoring data is adaptively filtered based on the self-noise frequency to obtain the second monitoring data; Multi-source data fitting is performed on the second monitoring data to construct a multi-source monitoring data pool.

3. The transformer condition monitoring method of claim 1, wherein, The specific steps of step S2 are as follows: Physically layer and decouple the multi-source monitoring data pool to obtain monitoring data of multiple physical types; Feature analysis is performed on the monitoring data to obtain multidimensional physical characteristics; The multidimensional physical characteristics include current harmonic components, voltage fluctuation amplitude, temperature rise gradient distribution, acoustic spectrum energy density, gas component concentration change rate, and vibration modes. Wavelet transform is performed on the multidimensional physical features to extract feature vectors; the feature vectors include time-domain statistical features and frequency-domain energy distribution features. Feature dimensionality reduction and key feature selection are performed on the feature vectors to construct a multi-dimensional monitoring feature set.

4. The transformer condition monitoring method of claim 1, wherein, Step S3 is as follows: Set the sampling interval for parameter changes at different time points; Based on the sampling interval, a sliding window change tracking method is used to track the multidimensional monitoring feature set and extract the time-series change increment sequence. Calculate the first-order difference magnitude and second-order difference curvature of the time-series change increment sequence; Based on the first-order difference change amplitude and the second-order difference change curvature, polynomial fitting is performed to construct change curves for multiple data types.

5. The transformer condition monitoring method of claim 1, wherein, The specific steps of step S4 are as follows: Identify the transformer's rated operating information; the rated operating information includes rated capacity, rated voltage level, insulation class, and operating environment boundary conditions; Based on the verified work information, the upper and lower limits of the standard work threshold are set to obtain the standard work range; The standard working range is dynamically corrected based on the operating conditions to obtain the adaptive adjustment range; The change curve is subjected to time-by-time limit detection based on the adaptive adjustment interval, and the data points exceeding the limit are marked.

6. The transformer condition monitoring method of claim 1, wherein, The specific steps of step S5 are as follows: The deviation of the over-limit monitoring data points is calculated to obtain the over-limit deviation. Perform time-series statistics on the out-of-limit deviation and extract deviation sequences at multiple time points; The mean, variance, and peak value of the deviation sequence are calculated, and trend discrimination is performed to obtain the deviation evolution pattern; the deviation evolution pattern includes continuous expansion of deviation, oscillating deviation, and gradual smooth convergence; Conduct an operational risk level assessment of deviations from the evolutionary model and generate risk assessment results; Based on the risk assessment results, perform visualization rendering to construct a multi-parameter visualization view; The multi-parameter visualization view is uploaded to the cloud platform for storage.

7. The transformer condition monitoring method of claim 6, wherein, The specific steps for performing visualization rendering based on the risk assessment results to construct a multi-parameter visualization view, and uploading the multi-parameter visualization view to the cloud platform for storage are as follows: Based on the risk assessment results, the change curves are time-matched, and the corresponding curve segments are marked. The corresponding curve segments are subjected to differentiated visualization processing, and visualization rendering is performed based on the changing curves of multiple data types to construct a multi-parameter visualization view; The multi-parameter visualization view is uploaded to the cloud platform for storage.

8. A transformer condition monitoring apparatus, characterised in that, For performing the transformer condition monitoring method as described in claim 1, including: The data acquisition unit is used to collect raw data of the transformer in operation and build a multi-source monitoring data pool. Hierarchical units are used to physically decouple multi-source monitoring data pools and construct multi-dimensional monitoring feature sets. The tracking unit is used to track changes in a sliding time window of a multidimensional monitoring feature set and construct change curves for multiple data types. The deviation calculation unit is used to perform time-by-time limit detection on the change curve and mark the limit monitoring data points; The visualization unit is used to assess the operational risk level and render the visualization of the over-limit monitoring data points, and to construct a multi-parameter visualization view. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8. When the processor executes the computer program, it implements the steps of the transformer condition monitoring method according to any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the transformer condition monitoring method according to any one of claims 1 to 7.

Images