A transformer fault prediction method and system

Abstract

The application discloses a transformer fault prediction method and system, and relates to the technical field of power equipment.The method comprises the following steps: collecting the start-stop action times of each time window, and determining the temperature control response lag coefficient of each time window by analyzing the temperature control lag level of the cooling control system in the power transformer when the cooling control system faces power mutation; analyzing the non-steady-state operation degree of the cooling control system in the corresponding time window to determine whether the temperature control state of the cooling control system affects the normal operation of the power transformer, and issuing a transformer fault prediction instruction; combining the power consumption drift and voltage frequency spectrum disturbance of the power transformer in each time window, analyzing the non-normal operation degree of the power transformer in the corresponding time window, performing digital model analysis on the non-normal operation index of the power transformer in each time window, generating a corresponding grade fault prediction signal, and performing corresponding voice broadcast reminding.

Description

A method and system for predicting transformer faults Technical Field

[0001] This invention relates to the field of power equipment technology, specifically to a method and system for predicting transformer faults. Background Technology

[0002] As a core component of the power system, the operating status of transformers directly affects the stability and continuity of power supply. In the modern smart grid architecture, more and more transformers are being integrated into multi-dimensional monitoring networks to achieve real-time perception of key parameters and dynamic assessment of operating status. Therefore, the in-depth extraction and analysis of temperature control response, power fluctuation and spectral disturbance characteristics of power transformer cooling system behavior has become an important issue that needs to be addressed in fault prediction technology.

[0003] While existing technologies have gradually incorporated sensing of temperature, voltage, and current parameters into transformer fault prediction methods, most rely on static threshold judgments and rule-based decision-making, lacking dynamic modeling of unsteady behavior and sensitive identification of time-series evolution. This limitation is particularly evident in predicting complex faults caused by abnormal operation of the cooling control system in transformers. Firstly, existing technologies struggle to model the interaction between abnormal start-up and shutdown behavior of the cooling control system and dynamic imbalances in power consumption and temperature, making it difficult to effectively identify operational anomalies caused by temperature control response lag. Secondly, traditional methods struggle to establish causal chains and predictive indices for nonlinear symptoms such as power consumption drift and voltage spectrum disturbances in transformers caused by the cooling control system, severely restricting the ability to predict deep-level transformer faults.

[0004] The root cause of the above situation lies in the fact that when the cooling control system starts and stops frequently or the temperature control mechanism is unbalanced, the internal heat load of the transformer cannot be released in time, causing overheating of the transformer surface and abnormal operation, which in turn induces unsteady oscillations, manifested as frequent abnormal disturbances in the voltage spectrum and power consumption drift. If these abnormal behaviors in the time-series evolution process are not identified in time, it will not only lead to physical damage such as internal thermal breakdown, dielectric aging, and coil insulation degradation, but also cause regional power outages and unstable operation of the power grid system, with a wide range of impacts and high costs. Therefore, it is urgent to build an intelligent transformer fault prediction method and system to capture the impact of temperature control imbalance on transformer power consumption and spectrum disturbances. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a transformer fault prediction method and system, which solves the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a transformer fault prediction method, comprising the following steps:

[0007] S1: Based on the multiple sets of sensors installed inside the power transformer and the set window duration, collect the number of start-stop actions in each time window, and determine the temperature control response lag coefficient of each time window by analyzing the temperature regulation lag level of the cooling control system in the power transformer when facing power surges.

[0008] S2: Correlate the number of start-stop actions in each time window with the temperature control response lag coefficient of the corresponding time window, analyze the degree of unsteady operation of the cooling control system in the corresponding time window, in order to determine whether the temperature control status of the cooling control system affects the normal operation of the power transformer, and issue a transformer fault prediction command.

[0009] S3: After receiving the transformer fault prediction command, based on the impact of the cooling control system on the power transformer, and combined with the power consumption drift and voltage spectrum disturbance of the power transformer in each time window, analyze the degree of abnormal operation of the power transformer in the corresponding time window, and obtain the abnormal operation index of the power transformer in each time window.

[0010] S4: Perform numerical simulation analysis on the abnormal operation index of power transformers in each time window; generate corresponding fault prediction signals and execute corresponding voice broadcast reminders.

[0011] Preferably, step S1 specifically includes:

[0012] S11. Multiple sets of sensors are installed inside the power transformer, specifically including: a high-response digital trigger sensor is deployed at the output relay port of the cooling control system circuit in the power transformer; a dual-channel power monitoring unit is connected between the main output circuit and the control cabinet; and an infrared temperature sensor is installed on the surface of the transformer shell.

[0013] S12. Based on the deployment of high-response digital trigger sensors, record in real time the number of abnormal starts and stops of the cooling control system, and combine with the set window duration to count the number of start and stop actions in each time window.

[0014] S13. Based on the dual-channel power monitoring unit connected between the main output circuit and the control cabinet, and combined with the set window duration, the power status of the power transformer in the corresponding time window is monitored in real time to obtain the effective power value of each monitoring time point in each time window.

[0015] Preferably, in step S14, based on the effective power values ​​at each monitoring time point within each time window, and in conjunction with the deployment of infrared temperature sensors, the number of high-temperature monitoring moments and the number of high-power monitoring moments within each time window are counted, specifically including:

[0016] Infrared temperature sensors are installed on the surface of the transformer shell. The temperature of the transformer shell surface is monitored in real time within the corresponding time window, and the surface temperature value at each monitoring time point in each time window is obtained. The average surface temperature of each time window is determined by combining the statistical averaging algorithm.

[0017] The surface temperature values ​​at each monitoring time point in each time window are compared with the average surface temperature of the corresponding time window. The monitoring time points where the surface temperature value at each monitoring time point in each time window exceeds the average surface temperature of the corresponding time window are recorded as high temperature monitoring moments. The number of high temperature monitoring moments in each time window is counted.

[0018] Based on the effective power values ​​at each monitoring time point in each time window in S13, and combined with the statistical averaging algorithm, the mean effective power value for each time window is determined.

[0019] The effective power value at each monitoring time point in each time window is compared with the average effective power value of the corresponding time window. The monitoring time points where the effective power value at each monitoring time point in each time window exceeds the average effective power value of the corresponding time window are recorded as high power monitoring moments. The number of high power monitoring moments in each time window is counted.

[0020] Preferably, in step S15, the number of high-temperature monitoring moments in each time window is correlated with the number of high-power monitoring moments in the corresponding time window. After dimensionless processing, the temperature control lag level exhibited by the cooling control system in the power transformer when facing sudden power changes is analyzed to determine the temperature control response lag coefficient for each time window. Specifically: In the formula, This represents the temperature control response lag coefficient for the corresponding time window. This indicates the number of high-temperature monitoring moments within the corresponding time window. This represents the number of high-power monitoring moments within the corresponding time window, where N represents the number of monitoring time points within the corresponding time window.

[0021] Preferably, step S2 specifically includes:

[0022] S21. Correlate the number of start-stop actions in each time window with the temperature control response lag coefficient of the corresponding time window. After normalization, analyze the degree of unsteady operation of the cooling control system in the corresponding time window and determine the temperature control imbalance coefficient of each time window, specifically: In the formula, This represents the temperature control imbalance coefficient for the corresponding time window. This indicates the number of start / stop actions within the corresponding time window. This represents the temperature control response lag coefficient for the corresponding time window. and All represent weight values.

[0023] Preferably, in step S22, the temperature control imbalance coefficient of each time window is compared and analyzed with a preset imbalance threshold to determine whether the temperature control status of the cooling control system affects the normal operation of the power transformer, and a transformer fault prediction command is issued accordingly, specifically including:

[0024] The temperature control imbalance coefficient of each time window is compared and analyzed with the preset imbalance threshold. When the temperature control imbalance coefficient of the corresponding time window exceeds the imbalance threshold, it indicates that the cooling control system is in an unbalanced state in the current time window, and the current time window is marked as an unbalanced time window. When the temperature control imbalance coefficient of the corresponding time window does not exceed the imbalance threshold, it indicates that the cooling control system is not in an unbalanced state in the current time window, and the current time window is marked as a normal time window.

[0025] The number of imbalance time windows is statistically analyzed. When the number of imbalance time windows exceeds 10% of the total number of time windows, it indicates that the temperature control imbalance of the cooling control system is affecting the normal operation of the power transformer. At this time, a transformer fault prediction command is issued. Conversely, it indicates that the temperature control of the cooling control system is normal and does not affect the normal operation of the power transformer. At this time, a level 3 fault prediction signal is issued.

[0026] Preferably, step S3 specifically includes:

[0027] S31. Upon receiving the transformer fault prediction command, based on the effective power values ​​at each monitoring time point within each time window, and combined with the fluctuation intensity averaging algorithm, determine the power consumption drift rate for each time window, specifically as follows: In the formula, This represents the power consumption drift rate for the corresponding time window. This represents the effective power difference between two adjacent monitoring time points within the corresponding time window, where i = 1, 2, 3, ..., N, and N represents the number of monitoring time points within the corresponding time window.

[0028] S32. Based on the dual-channel power monitoring unit connected between the main output circuit and the control cabinet, and combined with the set window duration, the voltage waveform signal at the output terminal of the power transformer is monitored in real time within the corresponding time window. The collected voltage waveform signal is processed by a window function and then subjected to spectral decomposition. After time-domain sampling, the voltage spectrum amplitude at each monitoring time point in each time window is obtained. Combined with a statistical averaging algorithm, and after dimensionless processing, the degree of voltage spectrum disturbance of the power transformer in the corresponding time window is analyzed to determine the voltage spectrum disturbance coefficient of each time window. Specifically: In the formula, This represents the voltage spectrum perturbation coefficient for the corresponding time window. This represents the voltage spectrum amplitude at the corresponding monitoring time point within the corresponding time window. This represents the voltage spectrum amplitude average value for the corresponding time window.

[0029] Preferably, in step S33, the power consumption drift rate of each time window is correlated with the voltage spectrum disturbance coefficient of the corresponding time window, and combined with the temperature control imbalance coefficient of the corresponding time window, after dimensionless processing, the degree of abnormal operation of the power transformer in the corresponding time window is analyzed to obtain the abnormal operation index of the power transformer in each time window, specifically: In the formula, This indicates the abnormal operation index for the corresponding time window. This indicates the degree to which temperature control imbalance affects power consumption drift within the corresponding time window. This indicates the degree of impact of temperature control imbalance on voltage spectrum disturbances within the corresponding time window.

[0030] Preferably, step S4 specifically includes:

[0031] S41. Perform numerical simulation analysis on the abnormal operation index of the power transformer in each time window, and establish a two-dimensional coordinate system with the performance value of the abnormal operation index as the vertical axis and the time window as the horizontal axis. Plot the abnormal operation index of the power transformer in each time window on the two-dimensional coordinate system in the form of broken lines to obtain several abnormal operation broken lines.

[0032] S42. Merge the angles formed by each abnormal operation line and the horizontal line, and combine them with the statistical mean calculation algorithm to obtain the average angle of the abnormal operation line. When the average angle exceeds 5°, it indicates that the temperature control imbalance in the cooling control system is affecting the normal operation of the power transformer and causing a power transformer fault. At this time, a first-level fault prediction signal is generated. When the average angle does not exceed 5°, it indicates that the temperature control imbalance in the cooling control system is affecting the normal operation of the power transformer, but does not cause a power transformer fault. At this time, a second-level fault prediction signal is generated.

[0033] S43. Upon receiving a fault prediction signal of the corresponding level, execute the corresponding voice broadcast reminder, specifically including:

[0034] When a Level 1 fault prediction signal is received, the voice will repeatedly announce: "The current cooling control system is out of temperature control, which has caused a power transformer failure."

[0035] When a level 2 fault prediction signal is received, the voice will repeatedly announce "The current cooling control system is experiencing temperature control imbalance, but this has not caused a power transformer fault";

[0036] When a Level 3 fault prediction signal is received, the voice will repeatedly announce, "The current temperature control of the cooling system is normal and has not caused a power transformer fault."

[0037] A transformer fault prediction system includes a data acquisition module, a steady-state analysis module, an anomaly analysis module, and a fault prediction module.

[0038] The data acquisition module is used to collect the number of start-stop actions in each time window based on multiple sets of sensors deployed inside the power transformer and in combination with the set window duration. By analyzing the temperature control lag level exhibited by the cooling control system in the power transformer when facing sudden power changes, the temperature control response lag coefficient of each time window is determined.

[0039] The steady-state analysis module is used to correlate the number of start-stop actions in each time window with the temperature control response lag coefficient of the corresponding time window, analyze the degree of non-steady-state operation of the cooling control system in the corresponding time window, determine whether the temperature control status of the cooling control system affects the normal operation of the power transformer, and issue transformer fault prediction commands.

[0040] The anomaly analysis module is used to analyze the degree of abnormal operation of the power transformer in the corresponding time window after receiving the transformer fault prediction command, based on the impact of the cooling control system on the power transformer, and combined with the power consumption drift and voltage spectrum disturbance of the power transformer in each time window, and obtain the abnormal operation index of the power transformer in each time window.

[0041] The fault prediction module is used to perform numerical simulation analysis on the abnormal operation index of power transformers in various time windows; generate corresponding fault prediction signals and execute corresponding voice broadcast reminders.

[0042] This invention provides a method and system for predicting transformer faults, which has the following advantages:

[0043] (1) This invention provides a transformer fault prediction method and system, which breaks through the limitations of relying on static threshold judgment and rule base decision-making in the prior art. For the first time, it realizes the progressive modeling and comprehensive evaluation of the nonlinear interaction between the temperature control response lag, power consumption drift and voltage spectrum disturbance of the cooling control system. By constructing a temperature control imbalance coefficient that is related to the number of start-stop actions and the temperature control response lag coefficient, and integrating the power consumption change rate and the spectrum disturbance index, an abnormal operation index is generated. Then, the dynamic evolution trend is extracted through numerical model analysis, realizing transformer fault prediction and graded early warning based on the influence of temperature control. It has good time sensitivity and system adaptability, and improves the accuracy, systematicness and foresight of transformer fault prediction.

[0044] (2) By deploying high-response digital trigger sensors and infrared temperature sensing units in the cooling control system loop of the power transformer, and combining them with the set window length, the start-stop frequency, surface temperature rise and high power consumption fluctuation behavior of the cooling control system are dynamically monitored. Two core parameters, temperature control response lag coefficient and temperature control imbalance coefficient, are established, realizing time series modeling and dimensionless standardized expression of the temperature control state of the cooling control system. By cross-analyzing the high temperature monitoring time and the high power monitoring time, the response lag degree of temperature control behavior in power fluctuation is identified, thereby effectively solving the problem that traditional methods are difficult to capture the interaction between start-stop non-steady-state behavior and temperature control imbalance, and significantly enhancing the cognitive ability and control accuracy of the cooling control system in the power transformer.

[0045] (3) In the process of judging the impact of temperature control imbalance of the cooling control system on the operating status of power transformer, the abnormal operation index model is constructed to quantify the influence of temperature control imbalance coefficient on power consumption drift and spectrum disturbance. Combined with the digital-analog angle analysis method, the abnormal trend evolution of transformer in different time windows is identified. By setting the imbalance threshold and the proportion of time window, a preliminary prediction is made. The power consumption drift rate and voltage spectrum disturbance coefficient are superimposed to form a multidimensional causal chain discrimination, which effectively solves the limitations of traditional methods in predicting complex faults caused by abnormal operation of the cooling control system in transformers. This mechanism not only improves the sensitivity and classification ability of transformer fault judgment, but also effectively avoids the risk of false alarm and missed alarm, and has good practical value and system integration ability. Attached Figure Description

[0046] Figure 1 is a schematic flowchart of a transformer fault prediction method according to the present invention;

[0047] Figure 2 is a block diagram of a transformer fault prediction system according to the present invention;

[0048] Figure 3 is a logical diagram of the overall logic thinking of the transformer fault prediction method of the present invention;

[0049] Figure 4 is a diagram showing the device connection relationship in S1 of the present invention. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] Example 1

[0052] Please refer to Figures 1 and 3. This invention provides a transformer fault prediction method, including the following steps:

[0053] S1: Based on the multiple sets of sensors installed inside the power transformer and the set window duration, collect the number of start-stop actions in each time window, and determine the temperature control response lag coefficient of each time window by analyzing the temperature regulation lag level of the cooling control system in the power transformer when facing power surges.

[0054] S2: Correlate the number of start-stop actions in each time window with the temperature control response lag coefficient of the corresponding time window, analyze the degree of unsteady operation of the cooling control system in the corresponding time window, in order to determine whether the temperature control status of the cooling control system affects the normal operation of the power transformer, and issue a transformer fault prediction command.

[0055] S3: After receiving the transformer fault prediction command, based on the impact of the cooling control system on the power transformer, and combined with the power consumption drift and voltage spectrum disturbance of the power transformer in each time window, analyze the degree of abnormal operation of the power transformer in the corresponding time window, and obtain the abnormal operation index of the power transformer in each time window.

[0056] S4: Perform numerical simulation analysis on the abnormal operation index of power transformers in each time window; generate corresponding fault prediction signals and execute corresponding voice broadcast reminders.

[0057] In this embodiment, a transformer fault prediction method constructs a complete technical system for identifying abnormal behavior in cooling control systems and analyzing the dynamic evolution of transformer operating states. This overcomes the response lag and identification bias problems caused by existing methods relying on static rules. Through the joint extraction of the number of start-stop actions and the temperature control response lag coefficient in S1, high-precision quantitative identification of the temperature control hysteresis behavior of the cooling control system in the face of power surges is achieved, effectively revealing the nonlinear response characteristics between the dynamic temperature control and power changes in the cooling system. In S2, by constructing a time-window-level temperature control imbalance judgment mechanism, not only is normalized modeling of the unsteady-state behavior of the cooling control system achieved, but prediction commands can also be accurately generated based on the imbalance window ratio, demonstrating… The system significantly improves the time sensitivity of fault early warning. In S3 and S4, quantitative analysis of power consumption drift rate and voltage spectrum disturbance coefficient is further introduced to form an abnormal operation index that integrates cooling behavior, power fluctuation and spectrum anomaly. The system also uses the digital-analog angle trend analysis method to judge the level of operation and provide voice prompts, which enhances the ability to identify the deep fault evolution path of transformers and provide risk warnings. Overall, the constructed deep learning prediction process has significant time sensitivity, non-steady-state adaptability and multi-parameter collaborative analysis capabilities. It can realize a closed-loop judgment mechanism for the entire process from temperature control lag behavior recognition to abnormal operation index evaluation to graded fault signal generation, which significantly improves the safety of transformer operation and the level of intelligent operation and maintenance.

[0058] Example 2

[0059] Please refer to Figures 1 and 4. Specifically, the steps in S1 include:

[0060] S11. Multiple sets of sensors are installed inside the power transformer, specifically including: a high-response digital trigger sensor is deployed at the output relay port of the cooling control system circuit in the power transformer; a dual-channel power monitoring unit is connected between the main output circuit and the control cabinet; and an infrared temperature sensor is installed on the surface of the transformer shell.

[0061] S12. Based on the deployment of high-response digital trigger sensors, record in real time the number of abnormal starts and stops of the cooling control system, and combine with the set window duration to count the number of start and stop actions in each time window.

[0062] The number of start-stop actions in each time window refers to the total number of start-up and shutdown actions of the cooling control system within a set statistical window duration, reflecting the system's operational activity and response stability during that time period. This parameter is obtained in real time through high-response digital trigger sensors deployed at the output relay ports of the cooling control system. The sensors can accurately capture the relay's action signals and generate digital trigger records each time a start-up or shutdown event occurs. The system performs segmented statistical analysis on these records according to the set time windows to obtain the number of start-stop actions within the corresponding time window. This parameter plays a crucial role in fault prediction, revealing whether the cooling system exhibits non-steady-state behaviors such as frequent start-stops, control fluctuations, or abnormal responses. It is an important basic data for judging whether the temperature control is stable and whether there are potential faults.

[0063] S13. Based on the dual-channel power monitoring unit connected between the main output circuit and the control cabinet, and combined with the set window duration, the power status of the power transformer in the corresponding time window is monitored in real time to obtain the effective power value of each monitoring time point in each time window.

[0064] The effective power value for each time window refers to the sequence of actual power output values ​​of the main output circuit of the power transformer obtained through continuous sampling within a set statistical time window. It reflects the actual load operation status and power fluctuation of the transformer during that time period. This data is obtained through a dual-channel power monitoring unit deployed between the main output circuit of the transformer and the control cabinet. The monitoring unit can simultaneously sense voltage and current signals and continuously output power measurement results. These power values ​​are sampled according to a set time interval and assigned to the corresponding time window, thus forming a series of monitoring time point-power value data sequences. This parameter is the basis for analyzing whether the transformer load is stable, whether there is abnormal power drift, sudden change or irregular fluctuation, and provides key support for judging whether temperature control lag affects power output and for calculating the subsequent abnormal operation index.

[0065] S14. Based on the effective power values ​​at each monitoring time point within each time window, and in conjunction with the deployment of infrared temperature sensors, count the number of high-temperature monitoring moments and the number of high-power monitoring moments within each time window, specifically including:

[0066] Infrared temperature sensors are installed on the surface of the transformer shell. The temperature of the transformer shell surface is monitored in real time within the corresponding time window, and the surface temperature value at each monitoring time point in each time window is obtained. The average surface temperature of each time window is determined by combining the statistical averaging algorithm.

[0067] The surface temperature values ​​at each monitoring time point in each time window are compared with the average surface temperature of the corresponding time window. The monitoring time points where the surface temperature value at each monitoring time point in each time window exceeds the average surface temperature of the corresponding time window are recorded as high temperature monitoring moments. The number of high temperature monitoring moments in each time window is counted.

[0068] Based on the effective power values ​​at each monitoring time point in each time window in S13, and combined with the statistical averaging algorithm, the mean effective power value for each time window is determined.

[0069] The effective power value at each monitoring time point in each time window is compared with the average effective power value of the corresponding time window. The monitoring time points where the effective power value at each monitoring time point in each time window exceeds the average effective power value of the corresponding time window are recorded as high power monitoring moments. The number of high power monitoring moments in each time window is counted.

[0070] The number of high-temperature monitoring moments and the number of high-power monitoring moments in each time window refer to the number of monitoring points within a set time window where the surface temperature or effective power value of the power transformer exceeds the corresponding average value within that time window. These are important statistical parameters for measuring the frequency of abnormal temperature and power fluctuations. Surface temperature values ​​are acquired by deploying infrared temperature sensors and compared with the average temperature of the window; any exceeding this value is recorded as a high-temperature monitoring moment. Similarly, effective power values ​​are collected by a dual-channel power monitoring unit and compared with the average power value of the window; any exceeding this value is recorded as a high-power monitoring moment. These two statistical values ​​together reveal whether the power transformer is operating under high heat load and high power consumption during a specific period. They are key evidence for identifying the dynamic coupling relationship between temperature control response hysteresis and power surges, and are also core input parameters for subsequently calculating the temperature control response hysteresis coefficient. They play an important role in identifying cooling system performance fluctuations and judging early signs of faults.

[0071] S15. Correlate the number of high-temperature monitoring moments in each time window with the number of high-power monitoring moments in the corresponding time window. After dimensionless processing, analyze the temperature control lag level exhibited by the cooling control system in the power transformer when facing sudden power changes, in order to determine the temperature control response lag coefficient for each time window. Specifically: In the formula, This represents the temperature control response lag coefficient for the corresponding time window. This indicates the number of high-temperature monitoring moments within the corresponding time window. This represents the number of high-power monitoring moments within the corresponding time window, where N represents the number of monitoring time points within the corresponding time window.

[0072] The formula in S15 calculates the difference between the number of high-temperature monitoring moments (Cgtᵏ) and the number of high-power monitoring moments (Cggᵏ) within the same time window, and then normalizes it to the total number of monitoring points (N) using a dimensionless method. Essentially, it maps the causal chain mentioned in the background technology—that temperature control lag leads to difficulty in timely heat load release—with a quantifiable index: when the cooling control system experiences a delay in response to sudden power fluctuations, the power surge is detected before the heat dissipation from the casing. The larger the difference between the two and the higher its proportion across all monitoring moments, the more delayed the temperature control response, and the more likely the equipment is to enter an unsteady, high-heat-risk state. The formula, by differentiating between power and temperature at abnormal moments rather than using a simple ratio, eliminates the influence of sampling frequency and window length, and strengthens the directional correlation between power surges and temperature rise lags. This directly serves the subsequent progressive calculation of the temperature control imbalance coefficient and the abnormal operation index. The temperature control response lag coefficient Xzhᵏ plays a crucial role in measuring the timeliness of the cooling system's thermal regulation in the entire method. It is the first key bridging parameter in the fault prediction logic chain, rising from the underlying sensor data. Its accuracy directly determines the reliability of subsequent fault warning classifications.

[0073] In this embodiment, a three-dimensional collaborative sensing network consisting of a high-response digital trigger sensor, a dual-channel power monitoring unit, and an infrared temperature sensing array is constructed inside the power transformer. This network simultaneously captures three types of high-speed dynamic quantities within a single time window: the start-up and shutdown actions of the cooling control system, power transients, and surface temperature rise of the casing. This provides a high-spatiotemporal consistency foundation of raw data for subsequent algorithms. The high-response digital trigger sensor records abnormal start-up or shutdown behaviors with millisecond-level action feedback accuracy. The dual-channel power monitoring unit quantifies power fluctuations in the main output circuit and the control cabinet from a dual-view perspective, avoiding the blind spots of traditional single-channel sampling in monitoring phase shifts and load imbalances. The infrared temperature sensing array captures the details of the surface temperature field evolution in real time through area scanning, providing support for the time-series localization of abnormal temperature rises. Based on this, the number of start-ups and shutdowns, the effective power sequence, and the surface temperature sequence are first extracted using a window sliding statistical method. Then, a mean-adaptive threshold algorithm is used to mark the high-power and high-temperature monitoring times, thereby establishing the high-temperature monitoring time. This method uses the coupling ratio of the number of high-power monitoring moments to eliminate differences in window duration, sampling frequency, and ambient temperature background using dimensionless normalization, thereby accurately quantifying the temperature control response lag of the cooling control system when facing power surges. The resulting temperature control response lag coefficient not only reveals the dynamic imbalance level of the temperature control link in real time but also provides input features of a unified scale for subsequent deep learning models, enabling early identification of potential fault modes. This method transforms multi-source high-speed sensor information into a single, interpretable, and engineering-comparable core indicator, overcoming the limitations of traditional technologies in comprehensively evaluating the three-dimensional coupling relationship of start-stop frequency, power surge, and temperature rise exceeding the threshold. Simultaneously, based on windowed statistics and dimensionless processing mechanisms, it can be deployed across transformers of different capacities and operating environments, significantly improving the universality and reliability of the fault prediction model, reducing false alarm and missed alarm rates, and providing power operation and maintenance personnel with visualized and traceable diagnostic evidence, ultimately achieving early detection of transformer cooling control anomalies.

[0074] Example 3

[0075] Please refer to Figure 1. Specifically, the steps in S2 include:

[0076] S21. Correlate the number of start-stop actions in each time window with the temperature control response lag coefficient of the corresponding time window. After normalization, analyze the degree of unsteady operation of the cooling control system in the corresponding time window and determine the temperature control imbalance coefficient of each time window, specifically: In the formula, This represents the temperature control imbalance coefficient for the corresponding time window. This indicates the number of start / stop actions within the corresponding time window. This represents the temperature control response lag coefficient for the corresponding time window. and All represent weight values.

[0077] The formula in S21 constructs a temperature control imbalance coefficient by weighting and fusing the number of start-stop actions within a time window with the temperature control response lag coefficient. This coefficient is used to quantitatively analyze whether the cooling control system is in a non-steady-state operating state within that time window. This logic directly corresponds to the existing technology pointed out in the background section, which struggles to identify the causal chain between cooling control anomalies and operational anomalies. The number of start-stop actions reflects the frequency of behavior at the cooling system control level, while the temperature control response lag coefficient reflects the delay in its thermal response. The integrated formula forms the temperature control imbalance coefficient, which uses start-stop frequency as the input response excitation term and temperature control lag as the feedback performance term, forming a two-way coupled evaluation mechanism of control and feedback. This breaks through the traditional judgment mode that relies solely on static temperature thresholds, with two weighted values... and By fitting historical operating data, the system is set to balance the relative impact of start-stop frequency and temperature control lag on the degree of temperature control imbalance, making the calculation results more consistent with the actual operating characteristics of the transformer. The role of the temperature control imbalance coefficient is that it can not only serve as an early criterion in fault precursor identification, but also be used to further construct a fault prediction command triggering mechanism. It is the core mediating variable connecting sensor data and operating status judgment, which strengthens the prediction system's ability to characterize unsteady, nonlinear, and multi-factor intertwined behaviors, and truly realizes the calculable, comparable, and early warning processing of the dynamic characteristics of cooling control.

[0078] S22. Compare and analyze the temperature control imbalance coefficient of each time window with the preset imbalance threshold to determine whether the temperature control status of the cooling control system affects the normal operation of the power transformer, and issue a transformer fault prediction command accordingly, specifically including:

[0079] The temperature control imbalance coefficient of each time window is compared and analyzed with the preset imbalance threshold. When the temperature control imbalance coefficient of the corresponding time window exceeds the imbalance threshold, it indicates that the cooling control system is in an unbalanced state in the current time window, and the current time window is marked as an unbalanced time window. When the temperature control imbalance coefficient of the corresponding time window does not exceed the imbalance threshold, it indicates that the cooling control system is not in an unbalanced state in the current time window, and the current time window is marked as a normal time window.

[0080] The number of imbalance time windows is statistically analyzed. When the number of imbalance time windows exceeds 10% of the total number of time windows, it indicates that the temperature control imbalance of the cooling control system is affecting the normal operation of the power transformer. At this time, a transformer fault prediction command is issued. Conversely, it indicates that the temperature control of the cooling control system is normal and does not affect the normal operation of the power transformer. At this time, a level 3 fault prediction signal is issued.

[0081] In this embodiment, the temperature control imbalance coefficient and its comparison mechanism constructed through S2 not only successfully integrate the start-stop behavior of the cooling control system with the temperature control response lag characteristics into a single characterization index, but also establish a mathematical judgment model for unsteady-state dynamic analysis, breaking through the crude mode of traditional cooling anomaly judgment that relies on empirical rules or static parameter comparison. Specifically, in S21, after normalizing the start-stop frequency and temperature control response lag degree within each time window, the resulting temperature control imbalance coefficient can dynamically quantify the response deviation level of the cooling system within that window, truly reflecting the unsteady-state operation state of the system caused by frequent start-stop or temperature control lag, and possessing high time sensitivity and computational portability. In S22, an imbalance threshold is further introduced to classify the temperature control imbalance coefficient. By comparing and setting a global judgment threshold based on the proportion of imbalance time windows, the accuracy and robustness of identifying cooling system anomalies are improved, and the risk of false alarms is significantly reduced. This enables effective perception of interference from local behavioral anomalies to the global operating status. Furthermore, the adoption of a quantitative standard that the number of imbalance time windows exceeds 10% of the total number of time windows avoids interference from individual anomaly windows on the overall judgment results. This allows for a comprehensive consideration of cooling link stability based on statistical strategies, improving the reliability and engineering feasibility of the judgment. More importantly, it enhances the fault prediction system's dynamic discrimination capability against transformer cooling performance anomalies, serving as a key triggering basis for subsequent multi-parameter fusion decisions and the triggering of level fault signals. This fundamentally strengthens the accuracy, logic, and adaptability of the prediction mechanism.

[0082] Example 4

[0083] Please refer to Figure 1. Specifically, the steps in S3 include:

[0084] S31. Upon receiving the transformer fault prediction command, based on the effective power values ​​at each monitoring time point within each time window, and combined with the fluctuation intensity averaging algorithm, determine the power consumption drift rate for each time window, specifically as follows: In the formula, This represents the power consumption drift rate for the corresponding time window. This represents the effective power difference between two adjacent monitoring time points within the corresponding time window, where i = 1, 2, 3, ..., N, and N represents the number of monitoring time points within the corresponding time window.

[0085] The formula in S31 uses a fluctuation intensity averaging algorithm to sum the absolute values ​​of the effective power differences between adjacent monitoring time points within a time window and then normalize them, thereby obtaining the power consumption drift rate for the corresponding time window. Its core purpose is to reflect the stability or the severity of fluctuations in the power output of the power transformer during that period. This formula directly addresses the problem pointed out in the background technology: existing technologies lack the ability to dynamically model the non-steady-state fluctuation behavior of power consumption, especially during the abnormal operation stage induced by the imbalance of the cooling control system, where the transformer power output is prone to drastic fluctuations. By performing time series differencing on the effective power sequence, the power consumption drift rate can accurately capture the energy consumption disturbance trend caused by cooling runaway, serving as an important basic indicator for subsequent judgment of abnormal operation index. Its role is to not only quantify the average intensity of power changes in a local time period, but also to establish a logical channel between the abnormal control of the cooling system and the actual load performance. It is a key link in identifying the causal chain from temperature control imbalance to power fluctuations and then to potential faults, effectively improving the time sensitivity and trend judgment of the abnormal load evolution process.

[0086] S32. Based on the dual-channel power monitoring unit connected between the main output circuit and the control cabinet, and combined with the set window duration, the voltage waveform signal at the output terminal of the power transformer is monitored in real time within the corresponding time window. The collected voltage waveform signal is processed by a window function and then subjected to spectral decomposition. After time-domain sampling, the voltage spectrum amplitude at each monitoring time point in each time window is obtained. Combined with a statistical averaging algorithm, and after dimensionless processing, the degree of voltage spectrum disturbance of the power transformer in the corresponding time window is analyzed to determine the voltage spectrum disturbance coefficient of each time window. Specifically: In the formula, This represents the voltage spectrum perturbation coefficient for the corresponding time window. This represents the voltage spectrum amplitude at the corresponding monitoring time point within the corresponding time window. This represents the voltage spectrum amplitude average value for the corresponding time window.

[0087] The formula in S32 and the background technology point out that abnormal cooling control can easily induce transformer energy consumption fluctuations and unstable operation. This is highly correlated with the problem. By accumulating and averaging the differences between continuously sampled power points, the fluctuation intensity of power output within a time window is effectively characterized, reflecting whether the transformer is in a non-steady-state operation state with high-frequency oscillations and load disturbances. In the corresponding spectral disturbance analysis, the voltage spectral disturbance coefficient of each time window is used to quantify the unstable performance of the output voltage in the frequency domain. Its calculation is based on window function processing and spectral decomposition after time-domain sampling to obtain the spectral amplitude of each monitoring point, and the spectrum... After comparing the amplitude average, the degree of dispersion beyond the amplitude can be identified; this coefficient essentially makes the thermoelectric interference path caused by the imbalance of the cooling system explicit, captures its impact on voltage stability from the frequency domain perspective, and forms a dual-channel synergistic index with the power consumption drift rate; the role of the voltage spectrum disturbance coefficient is that it not only reveals the frequency anomaly and power quality degradation caused by temperature control imbalance, but also provides a structured input for the construction of the abnormal operation index. It is an important basis for identifying potential insulation aging, voltage distortion and system oscillation of transformers, and improves the system's global perception of the transmission effect of cooling link anomalies at the electrical characteristic level.

[0088] S33. Correlate the power consumption drift rate of each time window with the voltage spectrum disturbance coefficient of the corresponding time window, and combine it with the temperature control imbalance coefficient of the corresponding time window. After dimensionless processing, analyze the degree of abnormal operation of the power transformer in the corresponding time window, and obtain the abnormal operation index of the power transformer in each time window, specifically: In the formula, This indicates the abnormal operation index for the corresponding time window. This indicates the degree to which temperature control imbalance affects power consumption drift within the corresponding time window. This indicates the degree of impact of temperature control imbalance on voltage spectrum disturbances within the corresponding time window.

[0089] The formula in S33 is established from the perspective of modeling the mutual coupling effects of multiple variables. By weighting and coupling the power consumption drift rate, voltage spectrum disturbance coefficient, and temperature control imbalance coefficient, it forms an abnormal operation index, which reflects the overall degree of abnormal operation of the power transformer within a specific time window. This formula is highly consistent with the core issue in the background technology that abnormal cooling control leads to power consumption drift and voltage spectrum disturbance, and existing methods struggle to construct a causal chain. By multiplying the temperature control imbalance coefficient by the power consumption drift rate and the voltage spectrum disturbance coefficient respectively, it characterizes the intensity of the disturbance effect of temperature control imbalance on power consumption and the spectrum, and integrates them into a unified whole through a square root function. The abnormal state intensity is expressed to construct a complete logical chain: abnormal response of cooling control system → thermal control lag → energy consumption fluctuation and voltage disturbance → comprehensive abnormal operation trend; the advantage of this formula is that it does not simply list multiple abnormal indicators in parallel, but constructs interactive quantities, which fully reflects the transmission and amplification effect between multiple factors; the role of the abnormal operation index is that, as the final characterization indicator of multi-source monitoring data fusion, it has a clear physical meaning and trend visualization capability. It is a key parameter used to judge the operational health, issue fault level warnings, and guide intervention strategies, which significantly enhances the overall grasp and prediction capability of the deep-level abnormal evolution path of transformers.

[0090] The following is an example of S3 calculation:

[0091] 1. Time window setting (window number k)

[0092] Time window duration: 6 minutes

[0093] Number of monitoring points N: 6 (sampling once per minute)

[0094] 2. Sampling data

[0095] Effective power values ​​Pv: 810, 820, 805, 825, 815, 830 (unit: kW)

[0096] Voltage spectrum amplitude Fpp: 1.02, 1.05, 1.00, 1.08, 1.04, 1.06 (unit: power)

[0097] 3. Calculation process

[0098] Power drift rate Lpvᵏ:

[0099] Lpvᵏ=(|820-810|+|805-820|+...+|830-815|) / 6=16.67, where / represents the division sign;

[0100] Voltage spectrum perturbation coefficient Rrdᵏ:

[0101] Calculate the deviation of each point from the mean and take the average, Rrdᵏ=0.025;

[0102] Temperature control imbalance coefficient Xwtᵏ:

[0103] The number of start-stop actions, Tqtᵏ, is 4; the temperature control response hysteresis coefficient, Xzhᵏ, is 0.06. =0.1, =1.0;

[0104] The temperature control imbalance coefficient Xwtᵏ = 0.1 × 4 + 1 × 0.06 = 0.46;

[0105] Abnormal operation index Zfzᵏ:

[0106] ;

[0107] In this embodiment, the abnormal operation index analysis mechanism constructed in S3 essentially realizes the systematic quantification and causal chain fusion of the bidirectional influence path of cooling system temperature control imbalance on power consumption behavior and voltage disturbance, breaking through the fragmented judgment mode of traditional technology that can only independently analyze power fluctuations or spectrum anomalies. First, the power consumption drift rate calculated by the fluctuation intensity averaging algorithm in S31 can accurately reflect the power instability of the power transformer within a unit time window. Especially when the cooling system regulation is unbalanced, the nonlinear jump, fall, or oscillation behavior of power can be dynamically captured, providing a quantitative basis for abnormal energy consumption trends. Second, in S32, the frequency domain characteristics of the voltage signal are obtained by combining window function and spectrum decomposition technology. Through the disturbance coefficient formed by the deviation of the voltage spectrum amplitude from the mean, the frequency domain disturbance characteristics of the output voltage caused by temperature control anomalies or load imbalance are effectively revealed, no longer... The previous approach limited the coarse discrimination of time-domain waveforms. More importantly, it introduced a temperature control imbalance coefficient as an intermediary variable in S33, calculating its impact on power consumption drift and spectral disturbances. This resulted in the fusion of three indices to generate a unified abnormal operation index. This index not only possesses good physical meaning and interpretability but also serves as a comprehensive measure of operational stability at the time window level, enabling early detection and prediction of potential fault states. This mechanism effectively solves the previous problem of integrating and modeling the three-element chain of abnormal cooling control, power consumption drift, and frequency disturbances. It is particularly suitable for identifying system-level operational disturbances caused by local temperature control imbalances and maintains stable judgment capabilities even in scenarios with multiple interfering factors. Therefore, this step enhances the ability to identify fault evolution trends under transformer operating conditions and establishes a solid data and model foundation for subsequent numerical model analysis and fault classification.

[0108] Example 5

[0109] Please refer to Figure 1. Specifically, the steps in S4 include:

[0110] S41. Perform numerical simulation analysis on the abnormal operation index of the power transformer in each time window, and establish a two-dimensional coordinate system with the performance value of the abnormal operation index as the vertical axis and the time window as the horizontal axis. Plot the abnormal operation index of the power transformer in each time window on the two-dimensional coordinate system in the form of broken lines to obtain several abnormal operation broken lines.

[0111] Several abnormal operation lines are trend lines formed by connecting the abnormal operation indices of the power transformer calculated within a continuous time window in chronological order. Each line corresponds to the trajectory of the operating status change within an operating stage or analysis period. These lines are plotted in a two-dimensional coordinate system with the time window as the horizontal axis and the abnormal operation index as the vertical axis, which can intuitively reflect the dynamic fluctuation characteristics of the transformer's operating status over time. Its core function is to transform the originally scattered and isolated values ​​within the time window into a continuous operating trajectory with visualized trends and morphological characteristics, which is convenient for subsequent geometric morphological analysis to determine whether there is a systematic abnormal trend or fault evolution process.

[0112] S42. Merge the angles formed by each abnormal operation line and the horizontal line, and combine them with the statistical mean calculation algorithm to obtain the average angle of the abnormal operation line. When the average angle exceeds 5°, it indicates that the temperature control imbalance in the cooling control system is affecting the normal operation of the power transformer and causing a power transformer fault. At this time, a first-level fault prediction signal is generated. When the average angle does not exceed 5°, it indicates that the temperature control imbalance in the cooling control system is affecting the normal operation of the power transformer, but does not cause a power transformer fault. At this time, a second-level fault prediction signal is generated.

[0113] S43. Upon receiving a fault prediction signal of the corresponding level, execute the corresponding voice broadcast reminder, specifically including:

[0114] When a Level 1 fault prediction signal is received, the voice will repeatedly announce: "The current cooling control system is out of temperature control, which has caused a power transformer failure."

[0115] When a level 2 fault prediction signal is received, the voice will repeatedly announce "The current cooling control system is experiencing temperature control imbalance, but this has not caused a power transformer fault";

[0116] When a Level 3 fault prediction signal is received, the voice will repeatedly announce, "The current temperature control of the cooling system is normal and has not caused a power transformer fault."

[0117] In this embodiment, the abnormal operation index trend judgment mechanism introduced in S4 realizes the key transformation of power transformer operation status from numerical anomaly identification to trend evolution judgment, significantly improving the logical level and discrimination accuracy of fault warning. Specifically, in S41, a two-dimensional coordinate system is constructed by using the abnormal operation index within each time window as the vertical axis and the corresponding window number as the horizontal axis, and an abnormal operation polyline is generated, forming a continuous and structured expression of the operation trajectory. This eliminates the reliance on traditional discrete point-based single-point judgment, thereby capturing the dynamic evolution characteristics of the system state over time. Based on this, S42 further proposes the average angle between the polyline and the horizontal line as a global stability evaluation index. By statistically analyzing the angle formed by the local slopes of each segment of the polyline and taking the average, a numerical-analog fusion analysis is achieved. When the average angle exceeds 5°, it clearly reflects a continuous upward trend in the abnormal index, i.e., cooling control. The system temperature control imbalance is no longer a temporary fluctuation, but a substantial failure affecting the overall operation of the transformer, thus triggering a first-level fault prediction signal. When the included angle does not exceed 5°, it is judged as an unsteady fluctuation under controlled conditions, and only a second-level prediction signal is issued. This analysis strategy avoids the common error of misjudging instantaneous high values ​​as serious faults, and strengthens the accurate control of trend stability and fault evolution persistence. In S43, the corresponding level of fault prediction signals are distinguished by content and given clear semantic prompts through a voice broadcast mechanism, providing operation and maintenance personnel with intuitive and timely intervention basis, realizing a complete closed-loop processing from data to semantics and from state to response. In summary, this step, by mapping the complex state index to geometric trends and extracting angle features, not only breaks through the limitations of point value anomaly identification, but also realizes the level classification and active interaction of prediction signals, improving the ability to judge the early fault trend of the transformer and the efficiency of operation and maintenance response.

[0118] Example 6

[0119] Please refer to Figures 1 and 2. Specifically, a transformer fault prediction system includes a data acquisition module, a steady-state analysis module, an anomaly analysis module, and a fault prediction module.

[0120] The data acquisition module is used to collect the number of start-stop actions in each time window based on multiple sets of sensors deployed inside the power transformer and in combination with the set window duration. By analyzing the temperature control lag level exhibited by the cooling control system in the power transformer when facing sudden power changes, the temperature control response lag coefficient of each time window is determined.

[0121] The steady-state analysis module is used to correlate the number of start-stop actions in each time window with the temperature control response lag coefficient of the corresponding time window, analyze the degree of non-steady-state operation of the cooling control system in the corresponding time window, determine whether the temperature control status of the cooling control system affects the normal operation of the power transformer, and issue transformer fault prediction commands.

[0122] The anomaly analysis module is used to analyze the degree of abnormal operation of the power transformer in the corresponding time window after receiving the transformer fault prediction command, based on the impact of the cooling control system on the power transformer, and combined with the power consumption drift and voltage spectrum disturbance of the power transformer in each time window, and obtain the abnormal operation index of the power transformer in each time window.

[0123] The fault prediction module is used to perform numerical simulation analysis on the abnormal operation index of power transformers in various time windows; generate corresponding fault prediction signals and execute corresponding voice broadcast reminders.

[0124] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for predicting transformer faults, characterized in that: Includes the following steps: S1: Based on multiple sensors deployed inside the power transformer and combined with the set window duration, the number of start-stop actions in each time window is collected. By analyzing the temperature control lag level exhibited by the cooling control system in the power transformer when facing power surges, the temperature control response lag coefficient of each time window is determined. S2: The number of start-stop actions in each time window is correlated with the corresponding temperature control response lag coefficient. The degree of non-steady-state operation of the cooling control system in the corresponding time window is analyzed to determine whether the temperature control state of the cooling control system affects the normal operation of the power transformer, and a transformer fault prediction command is issued. S3: After receiving the transformer fault prediction command, based on the impact of the cooling control system on the power transformer, and combined with the power consumption drift and voltage spectrum disturbance of the power transformer in each time window, the degree of abnormal operation of the power transformer in the corresponding time window is analyzed, and the abnormal operation index of the power transformer in each time window is obtained. S4: Perform numerical simulation analysis on the abnormal operation index of power transformers in each time window; generate corresponding fault prediction signals and execute corresponding voice broadcast reminders.

2. The transformer fault prediction method according to claim 1, characterized in that: The specific steps of S1 include: S11, installing multiple sets of sensors inside the power transformer, specifically including: deploying a high-response digital trigger sensor at the output relay port of the cooling control system circuit in the power transformer; connecting a dual-channel power monitoring unit between the main output circuit and the control cabinet; and installing an infrared temperature sensor on the surface of the transformer shell; S12, based on the deployment of the high-response digital trigger sensor, recording the number of abnormal starts and stops of the cooling control system in real time, and combining the set window duration to count the number of start and stop actions in each time window; S13, based on the dual-channel power monitoring unit connected between the main output circuit and the control cabinet, and combined with the set window duration, monitoring the power status of the power transformer in real time within the corresponding time window, and obtaining the effective power value at each monitoring time point in each time window.

3. The transformer fault prediction method according to claim 2, characterized in that: S14. Based on the effective power values ​​at each monitoring time point within each time window, and considering the presence of infrared temperature sensors, count the number of high-temperature monitoring moments and high-power monitoring moments within each time window. Specifically, this includes: real-time monitoring of the surface temperature of the power transformer casing within the corresponding time window, based on the infrared temperature sensors deployed on the transformer casing surface and the set window duration; obtaining the surface temperature values ​​at each monitoring time point within each time window; determining the average surface temperature for each time window using a statistical averaging algorithm; comparing the surface temperature values ​​at each monitoring time point within each time window with the average surface temperature for the corresponding time window, and then calculating the average surface temperature for each time window. The monitoring time points in each monitoring time window whose surface temperature values ​​exceed the average surface temperature of the corresponding time window are recorded as high-temperature monitoring moments, and the number of high-temperature monitoring moments in each time window is counted. Based on the effective power values ​​of each monitoring time point in each time window in S13, and combined with the statistical averaging algorithm, the average effective power of each time window is determined. The effective power values ​​of each monitoring time point in each time window are compared with the average effective power of the corresponding time window, and the monitoring time points in each time window whose effective power values ​​exceed the average effective power of the corresponding time window are recorded as high-power monitoring moments, and the number of high-power monitoring moments in each time window is counted.

4. The transformer fault prediction method according to claim 3, characterized in that: S15. Correlate the number of high-temperature monitoring moments in each time window with the number of high-power monitoring moments in the corresponding time window. After dimensionless processing, analyze the temperature control lag level exhibited by the cooling control system in the power transformer when facing sudden power changes, in order to determine the temperature control response lag coefficient for each time window. Specifically: In the formula, This represents the temperature control response lag coefficient for the corresponding time window. This indicates the number of high-temperature monitoring moments within the corresponding time window. This represents the number of high-power monitoring moments within the corresponding time window, where N represents the number of monitoring time points within the corresponding time window.

5. The transformer fault prediction method according to claim 4, characterized in that: The specific steps of S2 include: S21, correlating the number of start-stop actions in each time window with the temperature control response lag coefficient of the corresponding time window, and after normalization, analyzing the degree of unsteady operation of the cooling control system in the corresponding time window, and determining the temperature control imbalance coefficient of each time window, specifically: In the formula, This represents the temperature control imbalance coefficient for the corresponding time window. This indicates the number of start / stop actions within the corresponding time window. This represents the temperature control response lag coefficient for the corresponding time window. and All represent weight values.

6. The transformer fault prediction method according to claim 5, characterized in that: S22. Compare and analyze the temperature control imbalance coefficient of each time window with the preset imbalance threshold to determine whether the temperature control status of the cooling control system affects the normal operation of the power transformer, and issue a transformer fault prediction command accordingly. Specifically, this includes: comparing and analyzing the temperature control imbalance coefficient of each time window with the preset imbalance threshold; when the temperature control imbalance coefficient of the corresponding time window exceeds the imbalance threshold, it indicates that the cooling control system is in an imbalanced state in the current time window, and the current time window is marked as an imbalanced time window; when the temperature control imbalance coefficient of the corresponding time window does not exceed the imbalance threshold, it indicates that the cooling control system is not in an imbalanced state in the current time window, and the current time window is marked as a normal time window; count the number of imbalanced time windows; when the number of imbalanced time windows exceeds 10% of the total number of time windows, it indicates that the temperature control imbalance of the cooling control system affects the normal operation of the power transformer, and a transformer fault prediction command is issued at this time; otherwise, it indicates that the temperature control of the cooling control system is normal and does not affect the normal operation of the power transformer, and a level three fault prediction signal is issued at this time.

7. The transformer fault prediction method according to claim 6, characterized in that: The specific steps of S3 include: S31, after receiving the transformer fault prediction command, based on the effective power value at each monitoring time point in each time window, and combined with the fluctuation intensity averaging algorithm, the power consumption drift rate of each time window is determined, specifically as follows: In the formula, This represents the power consumption drift rate for the corresponding time window. This represents the effective power difference between two adjacent monitoring time points within the corresponding time window, i = 1, 2, 3, ..., N, where N represents the number of monitoring time points within the corresponding time window; S32. Based on the dual-channel power monitoring unit connected between the main output circuit and the control cabinet, and combined with the set window duration, the voltage waveform signal at the output terminal of the power transformer within the corresponding time window is monitored in real time. The collected voltage waveform signal is processed by a window function and then subjected to spectral decomposition. After time-domain sampling, the voltage spectrum amplitude at each monitoring time point in each time window is obtained. Combined with a statistical averaging algorithm, after dimensionless processing, the degree of voltage spectrum disturbance of the power transformer in the corresponding time window is analyzed to determine the voltage spectrum disturbance coefficient of each time window, specifically: In the formula, This represents the voltage spectrum perturbation coefficient for the corresponding time window. This represents the voltage spectrum amplitude at the corresponding monitoring time point within the corresponding time window. This represents the voltage spectrum amplitude average value for the corresponding time window.

8. The transformer fault prediction method according to claim 7, characterized in that: S33. Correlate the power consumption drift rate of each time window with the voltage spectrum disturbance coefficient of the corresponding time window, and combine it with the temperature control imbalance coefficient of the corresponding time window. After dimensionless processing, analyze the degree of abnormal operation of the power transformer in the corresponding time window, and obtain the abnormal operation index of the power transformer in each time window, specifically: In the formula, This indicates the abnormal operation index for the corresponding time window. This indicates the degree to which temperature control imbalance affects power consumption drift within the corresponding time window. This indicates the degree of impact of temperature control imbalance on voltage spectrum disturbances within the corresponding time window.

9. A transformer fault prediction method according to claim 8, characterized in that: The specific steps in S4 include: S41, performing numerical simulation analysis on the abnormal operation index of the power transformer in each time window, and establishing a two-dimensional coordinate system with the performance value of the abnormal operation index as the vertical axis and the time window as the horizontal axis. Plotting the abnormal operation index of the power transformer in each time window on the two-dimensional coordinate system in the form of broken lines, thereby obtaining several abnormal operation broken lines; S42, merging the angles formed by each abnormal operation broken line and the horizontal line, and combining the statistical mean calculation algorithm to obtain the average angle of the abnormal operation broken lines. When the average angle exceeds 5°, it indicates that the temperature control imbalance in the cooling control system affects the normal operation of the power transformer and causes a power transformer fault. At this time, a first-level fault prediction signal is generated. When the average angle does not exceed 5°, it indicates that the temperature control imbalance in the cooling control system affects the normal operation of the power transformer but does not cause a power transformer fault. At this time, a second-level fault prediction signal is generated. S43. Upon receiving a fault prediction signal of the corresponding level, execute the corresponding voice broadcast reminder, specifically including: when a level 1 fault prediction signal is received, broadcast the voice message "The current cooling control system is out of temperature control balance, which has caused a power transformer fault" repeatedly; when a level 2 fault prediction signal is received, broadcast the voice message "The current cooling control system is out of temperature control balance, which has not caused a power transformer fault" repeatedly; when a level 3 fault prediction signal is received, broadcast the voice message "The current cooling control system is in normal temperature control, which has not caused a power transformer fault".

10. A transformer fault prediction system, used to implement the transformer fault prediction method according to any one of claims 1 to 9, characterized in that: The system includes a data acquisition module, a steady-state analysis module, an anomaly analysis module, and a fault prediction module. The data acquisition module uses multiple sensors deployed within the power transformer and a set window duration to collect the number of start-stop actions in each time window. It then analyzes the temperature control lag level exhibited by the cooling control system in the power transformer when facing sudden power fluctuations to determine the temperature control response lag coefficient for each time window. The steady-state analysis module correlates the number of start-stop actions in each time window with the corresponding temperature control response lag coefficient to analyze the degree of non-steady-state operation of the cooling control system in the corresponding time window. This determines whether the temperature control state of the cooling control system affects the normal operation of the power transformer and issues a transformer fault prediction command. The anomaly analysis module, upon receiving the transformer fault prediction command, analyzes the degree of abnormal operation of the power transformer in the corresponding time window based on the impact of the cooling control system on the power transformer and the power consumption drift and voltage spectrum disturbance of the power transformer in each time window, obtaining the abnormal operation index of the power transformer in each time window. The fault prediction module performs numerical simulation analysis on the abnormal operation index of the power transformer in each time window, generates a corresponding level of fault prediction signal, and executes a corresponding voice broadcast reminder.

Images