A method for evaluating the load capacity of new energy transmission transformers

Abstract

The present invention discloses a method for evaluating the load capacity of a new energy transmission transformer, which specifically relates to the technical field of safe operation of power equipment. The method is used to solve the problem that existing evaluation methods ignore the cumulative degradation effect of frequent on-load tap changer operations on heat dissipation performance, resulting in inaccurate safe load boundaries. An interval time series is constructed by real-time recording of on-load tap changer operation timestamps, and a permutation entropy value is generated based on phase space reconstruction to trigger heat dissipation degradation evaluation. Three-dimensional data of the distributed optical fiber temperature and leakage magnetic field of the winding are synchronously acquired, and a temperature-magnetic field coupling network is constructed to extract structural characteristic parameters and generate a heat dissipation degradation evaluation value. The evaluation value is converted into a heat dissipation coefficient correction factor, and the transformer thermal circuit model parameters are dynamically corrected in combination with the real-time load current to generate a safe load boundary in an adaptively degraded state. Finally, a load capacity evaluation result including the corrected boundary is output, thereby realizing a dynamic correlation evaluation between the heat dissipation performance degradation and the safe load capacity of the transformer.

Description

Technical Field

[0001] The present invention relates to the technical field of safe operation of electric power equipment, and more specifically, to a method for evaluating the load capacity of a new energy transmission transformer. Background Art

[0002] Renewable energy generation is being integrated into the power system on a large scale, with wind farms and photovoltaic power stations centrally transmitting power through step-up transformers. Fluctuations in renewable energy output cause grid voltage to frequently deviate from the rated range. Transformer on-load tap changers (OLTCs) maintain voltage stability by adjusting the transformation ratio. The number of OLTC operations at renewable energy sites is significantly higher than at conventional substations, and their operating states are treated as independent events. Existing transformer load capacity assessment methods calculate safety limits based on the initial thermal characteristics of the equipment.

[0003] Existing assessment methods ignore the cumulative degradation effect of OLTC operation frequency on the transformer's heat dissipation performance. That is, the arc erosion pollutants generated by frequent operation continuously reduce the thermal conductivity of the cooling medium. However, the load capacity calculation is still based on the heat dissipation model under clean conditions, resulting in a disconnect between the assessment results and the actual operating status. Under the scenario of new energy fluctuations, the transformer is subjected to loads higher than the theoretical safety threshold for a long time, which accelerates insulation aging and increases the risk of failure. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a method for evaluating the load capacity of a new energy transmission transformer to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method for evaluating the load capacity of a new energy transmission transformer includes:

[0007] S1. Real-time record of the timestamp of each action of the transformer on-load tap-changer, calculate the interval time between adjacent actions based on the timestamp and construct an interval time series;

[0008] S2. Reconstruct the phase space of the interval time series to generate a permutation entropy value. If the permutation entropy value is lower than a preset permutation entropy threshold, a heat dissipation degradation assessment is triggered.

[0009] S3. When the heat dissipation degradation assessment is triggered, the distributed optical fiber temperature monitoring data and the three-dimensional distribution data of the leakage magnetic field of the transformer winding are simultaneously obtained;

[0010] S4. Constructing a coupling network based on the distributed optical fiber temperature monitoring data and the three-dimensional distribution data of the leakage magnetic field, and extracting the structural characteristic parameters of the coupling network to generate an actual heat dissipation degradation assessment value;

[0011] S5. Convert the actual heat dissipation degradation assessment value into a heat dissipation coefficient correction factor, and generate a corrected safe load boundary based on the transformer thermal circuit model, combining the heat dissipation coefficient correction factor and the real-time load current;

[0012] S6. Output the transformer load capacity assessment result including the revised safe load boundary.

[0013] In a preferred embodiment, the time stamp of each action of the transformer on-load tap changer is recorded in real time, the interval time between adjacent actions is calculated based on the time stamp, and the interval time sequence is constructed, including:

[0014] Record the exact time of each on-load tap-changer action as a timestamp;

[0015] Store the timestamps of consecutive actions in chronological order to form a timestamp sequence;

[0016] Calculate the time difference between adjacent timestamps as the action interval;

[0017] Arrange the action intervals in the order in which the actions occur to form an interval time sequence.

[0018] In a preferred embodiment, after performing phase space reconstruction on the interval time series, a permutation entropy value is generated. If the permutation entropy value is lower than a preset permutation entropy threshold, a heat dissipation degradation assessment is triggered, including:

[0019] Calculate time delay parameters and embedding dimension parameters based on interval time series;

[0020] Reconstruct the phase space of the interval time series according to the time delay parameter and the embedding dimension parameter to generate a set of state space vectors;

[0021] Symbolize each vector in the state space vector set to form a symbol sequence;

[0022] Count the frequencies of different permutation patterns in all symbol sequences;

[0023] Calculate the probability distribution based on the frequency of occurrence of the arrangement pattern;

[0024] Calculate the permutation entropy value based on the probability distribution using the information entropy formula;

[0025] Comparing the permutation entropy value with a preset permutation entropy threshold;

[0026] When the permutation entropy value is less than a preset permutation entropy threshold, a heat dissipation degradation assessment trigger signal is generated.

[0027] In a preferred embodiment, the time delay parameter is determined by the first zero-crossing point of the autocorrelation function, and the embedding dimension parameter is determined according to the rule that the number of permutation and combination patterns is equal to the factorial of the embedding dimension.

[0028] In a preferred embodiment, when the heat dissipation degradation assessment is triggered, the distributed optical fiber temperature monitoring data and the three-dimensional distribution data of the leakage magnetic field of the transformer winding are simultaneously acquired, including:

[0029] In response to a heat dissipation degradation assessment trigger signal, the temperature monitoring system and the magnetic field measurement system are simultaneously activated;

[0030] The temperature distribution data of the winding surface is collected by a distributed optical fiber temperature sensor array, and each sensor node obtains the spatial position coordinates and temperature value;

[0031] The three-axis magnetic field sensor array is used to collect magnetic induction intensity component data of discrete points in space, and the three-dimensional coordinates and magnetic field vector are obtained for each measurement point;

[0032] Mapping the temperature sensor spatial coordinates and the magnetic field measurement point spatial coordinates to a unified coordinate system;

[0033] Align the timestamps of temperature data and magnetic field data to the same time base;

[0034] Generate a temperature distribution dataset and a three-dimensional magnetic field distribution dataset containing spatial position information.

[0035] In a preferred embodiment, a coupling network is constructed based on distributed optical fiber temperature monitoring data and leakage magnetic field three-dimensional distribution data, and structural characteristic parameters of the coupling network are extracted to generate an actual heat dissipation degradation assessment value, including:

[0036] Establish a spatial correlation matrix between temperature nodes and magnetic field nodes, where the matrix elements represent the inverse of the physical distance between the temperature sensor and the magnetic field measurement point;

[0037] Construct a coupled network topology, where the temperature nodes and magnetic field nodes are connected through an association matrix to form bidirectional edges;

[0038] Calculate the betweenness centrality of all nodes in the coupled network;

[0039] Calculate the proportion of nodes whose betweenness centrality exceeds the preset betweenness threshold;

[0040] Calculate all edge weights in the coupled network;

[0041] Extract the statistical distribution characteristics of all edge weight values;

[0042] Identify the modular structure of the coupled network and divide the temperature-magnetic field coupled subnetworks by maximizing the modularity function;

[0043] Calculate the ratio of the average path length to the clustering coefficient of each coupled subnetwork;

[0044] Generate a multidimensional feature vector based on the statistical distribution characteristics of node ratio, edge weight value, number of coupled subnetworks and the ratio of average path length to clustering coefficient;

[0045] The multidimensional feature vector is input into the radial basis function interpolation model to calculate the actual heat dissipation degradation evaluation value.

[0046] In a preferred embodiment, the edge weight value is determined by the dot product of the temperature gradient vector and the magnetic field gradient vector.

[0047] In a preferred embodiment, the actual heat dissipation degradation assessment value is converted into a heat dissipation coefficient correction factor, and a corrected safe load boundary is generated based on the transformer thermal circuit model in combination with the heat dissipation coefficient correction factor and the real-time load current, including:

[0048] The basic heat dissipation coefficient in the transformer thermal circuit model is defined to characterize the heat dissipation capacity of the transformer under rated operating conditions;

[0049] Establish a piecewise linear mapping relationship between the actual heat dissipation degradation assessment value and the heat dissipation coefficient correction factor;

[0050] Obtain real-time load current data of transformer;

[0051] In the transformer thermal circuit model, the basic heat dissipation coefficient is multiplied by the heat dissipation coefficient correction factor to obtain the corrected heat dissipation coefficient;

[0052] Solve the thermal circuit differential equations based on the corrected heat dissipation coefficient and real-time load current;

[0053] Calculate the curve of winding hot spot temperature changing with time;

[0054] Determine the critical time point when the hot spot temperature of the winding reaches the limit temperature of the insulation material;

[0055] Generate safe load boundary curves based on critical time points and real-time load current.

[0056] In a preferred embodiment, the piecewise linear mapping relationship is obtained by calibration of historical fault data, and the thermal circuit differential equation group includes thermal capacitance parameters and thermal resistance parameters.

[0057] In a preferred embodiment, outputting the transformer load capacity assessment result including the revised safe load limit includes:

[0058] Convert the safety load boundary curve into a visual graphical format;

[0059] Generate load capacity assessment report text;

[0060] Compare the corresponding relationship between the current load current and the safe load boundary curve in real time;

[0061] When the real-time load current exceeds the allowable value of the safe load boundary curve, an early warning signal is triggered;

[0062] Integrate visualization graphics, load capacity assessment report text and warning signals into comprehensive assessment results;

[0063] The comprehensive evaluation results are output through the human-computer interaction interface.

[0064] Compared with the prior art, the present invention has the following beneficial effects:

[0065] 1. By innovatively establishing a dynamic correlation between on-load tap-changer (OLTC) operation frequency and heat dissipation degradation, the problem of inaccurate transformer safety assessments in scenarios with fluctuating renewable energy sources is resolved. Based on permutation entropy analysis of OLTC operation intervals, the system accurately captures the triggering timing of heat dissipation degradation. Phase space reconstruction technology analyzes the temporal chaotic characteristics of the operation intervals, initiating a deep assessment only when the sequence exhibits significant regularity (i.e., the permutation entropy falls below a threshold), avoiding inefficient calculations while ensuring the sensitivity of degradation detection. This triggering mechanism, based on operating behavior rather than fixed cycles, focuses system resources on critical degradation windows.

[0066] 2. By constructing a multi-physics field coupling network topology analysis method of temperature field and leakage magnetic field, different from conventional single parameter monitoring, the edge weight is formed by the spatial dot product of temperature gradient and magnetic field gradient, revealing the intrinsic correlation of thermal magnetic energy transfer at the physical level; through modular identification, the coupling sub-network is spontaneously divided to accurately locate the local heat dissipation abnormal area; the multi-dimensional features such as the betweenness centrality distribution and edge weight statistics are finally extracted to objectively quantify the overall degradation degree of the heat dissipation structure; the network characteristics are mapped into the heat dissipation coefficient correction factor through the radial basis function, and the thermal circuit model parameters are dynamically corrected so that the generated safe load boundary reflects the cumulative degradation effect of OLTC in real time; the adaptive evaluation of the transformer load capacity degradation with the switching action is realized, providing essential protection for the safe operation of transformers in new energy stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is a flow chart of a method for evaluating the load capacity of a new energy transmission transformer according to the present invention. DETAILED DESCRIPTION

[0068] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0069] Example: Figure 1The present invention provides a method for evaluating the load capacity of a new energy transmission transformer, comprising:

[0070] S1. Real-time record of the timestamp of each action of the transformer on-load tap-changer, calculate the interval time between adjacent actions based on the timestamp and construct an interval time series;

[0071] S2. Reconstruct the phase space of the interval time series to generate a permutation entropy value. If the permutation entropy value is lower than a preset permutation entropy threshold, a heat dissipation degradation assessment is triggered.

[0072] S3. When the heat dissipation degradation assessment is triggered, the distributed optical fiber temperature monitoring data and the three-dimensional distribution data of the leakage magnetic field of the transformer winding are simultaneously obtained;

[0073] S4. Constructing a coupling network based on the distributed optical fiber temperature monitoring data and the three-dimensional distribution data of the leakage magnetic field, and extracting the structural characteristic parameters of the coupling network to generate an actual heat dissipation degradation assessment value;

[0074] S5. Convert the actual heat dissipation degradation assessment value into a heat dissipation coefficient correction factor, and generate a corrected safe load boundary based on the transformer thermal circuit model, combining the heat dissipation coefficient correction factor and the real-time load current;

[0075] S6. Output the transformer load capacity assessment result including the revised safe load boundary.

[0076] S1. Real-time record the timestamp of each action of the transformer on-load tap-changer, calculate the interval time between adjacent actions based on the timestamp and construct the interval time sequence. The specific implementation is as follows:

[0077] A mechanical position sensor is installed on the on-load tap-changer's drive shaft. When it detects that the drive shaft's rotation angle reaches a threshold of 30 degrees, it sends a digital trigger signal to the control unit. Upon receiving the trigger signal, the control unit immediately reads the current time value from the high-precision clock module. This time value, based on Coordinated Universal Time, contains seven data fields: year, month, day, hour, minute, second, and millisecond. Time recording accuracy is ensured by a temperature-compensated crystal oscillator with a 16 MHz oscillation frequency. Within an operating temperature range of -40°C to +85°C, the time recording error does not exceed ±0.1 millisecond. Timestamp recording requires that three conditions are met simultaneously: the drive shaft's rotation angle exceeds 30 degrees, the drive motor current drops below 0.5 amps, and the mechanical vibration intensity is less than 5 meters per second squared.

[0078] Continuous action timestamp storage is managed using a circular buffer data structure. The buffer is designed to hold 100 timestamp data sets, each containing all seven time fields. Storage management follows a first-in, first-out principle: when a new timestamp is added, the oldest stored history is automatically overwritten. When the buffer is not full, empty slots are filled in index order. Buffer read and write operations are performed via a direct memory access controller, ensuring data storage within 5 microseconds. Buffer status is monitored in real time, triggering an early warning signal when storage utilization exceeds 90%.

[0079] The calculation of the interval between adjacent actions is performed using a time difference algorithm. Two adjacent timestamps are sequentially extracted from the stored timestamp sequence and converted to absolute time values ​​in milliseconds. This conversion is done by multiplying the hours by 3,600,000, the minutes by 60,000, and the seconds by 1,000. Finally, the millisecond value is added to obtain the total number of milliseconds. The difference between the two absolute time values ​​is the interval between actions, and the result is rounded to an integer millisecond. This calculation is performed in the hardware arithmetic logic unit, and a single calculation takes no more than 10 microseconds.

[0080] Interval time series are constructed using a dynamic array data structure. The calculated action interval values ​​are stored in an array in the order in which the actions occurred, with the earliest action interval value stored at array index 0 and the latest action interval value stored at the highest index. Data validity is verified before storage: if the interval is less than 10 milliseconds, it is considered a mechanical jitter interference signal and discarded. If the interval is greater than 86,400,000 milliseconds (24 hours), it is considered an equipment downtime event, and the current sequence is cleared and accumulation begins again. A dual-copy verification mechanism is used during data storage to ensure data integrity.

[0081] The determination of the minimum data volume for an interval time series is based on phase space reconstruction theory. According to the Takens embedding theorem in nonlinear dynamics, the minimum data dimension required for reconstructing a system state is twice the intrinsic dimension of the system plus one. Mechanical characteristic testing of transformer tap changers confirmed that the intrinsic dimension of their motion system is one-dimensional. Therefore, the minimum theoretical data volume is three action intervals. Considering a safety margin for engineering applications, a preset number is ten times the theoretical value, requiring the sequence to contain no fewer than 30 consecutive action intervals. This number was verified through analysis of 100 sets of historical data: when the data volume was less than 20, the permutation entropy fluctuation coefficient exceeded 30%, and when it reached 30, the fluctuation coefficient dropped to less than 10%.

[0082] The timestamp acquisition hardware utilizes a high-precision real-time clock chip with temperature compensation, connected to a 32-bit microcontroller via an integrated circuit bus. The chip incorporates a built-in temperature sensor and voltage compensation circuitry to ensure time drift is less than ±0.1 milliseconds per minute over the full temperature range. Interval calculations are performed within the microcontroller's interrupt service routine, with the interrupt priority set to the highest level. Data storage utilizes non-volatile ferroelectric memory, which supports 100,000,000 erase and write cycles and maintains data retention for over 10 years.

[0083] The exception handling mechanism in the action interval time calculation includes three scenario response plans: when it is detected that the interval time of five consecutive actions is less than 1000 milliseconds, the overload protection mode is activated, and valid data is recorded once every five actions; when a system time rollback exception is detected, the current sequence is automatically frozen and an alarm signal is triggered; when the memory unit verification error rate exceeds 0.1%, it automatically switches to the backup storage area to continue working.

[0084] The data completion mechanism used during interval time series construction addresses occasional data loss. When a single data point is missing due to a system failure, it is interpolated using the arithmetic mean of the two valid intervals before and after the missing data point. If more than three consecutive missing data points are present, the series is considered broken, and the current series is cleared and accumulation restarted. Interpolation is only enabled when the data missing rate is less than 5%. The interpolation formula is: the completion value is the sum of the previous and next valid intervals, divided by two.

[0085] The false trigger prevention mechanism during timestamp recording is implemented through multi-signal verification. In addition to the mechanical position sensor signal, the motor current curve and vibration sensor signal are also monitored simultaneously. All three signals must meet preset thresholds: position change greater than 30 degrees, current below 0.5 amps, and vibration intensity less than 5 meters per second squared. Signal verification utilizes parallel processing, ensuring verification time of less than 2 milliseconds.

[0086] Dynamic array management utilizes pre-allocated memory pool technology. Initially, storage space is allocated for 50 interval data units. When the data volume exceeds 80% of the current capacity, the capacity is automatically expanded by 50%. The expansion process uses a memory copy to ensure data continuity. The storage space release policy is to reclaim all memory resources when the sequence is reset, but retain the initial 50 data units of pre-allocated space.

[0087] S2. After reconstructing the phase space of the interval time series, a permutation entropy value is generated. If the permutation entropy value is lower than the preset permutation entropy threshold, a heat dissipation degradation assessment is triggered. The specific implementation is as follows:

[0088] The time delay parameter is calculated using the autocorrelation function analysis method. Taking an interval time series as input data, the series is first preprocessed to zero mean: the arithmetic mean of all elements in the series is calculated, and this mean is subtracted from each element. The autocorrelation function values ​​are then calculated for different lag steps: for each lag step k, the covariance between the original series and the series lagged k steps is calculated and then divided by the variance of the original series. The lag step k is incremented starting from 1, and the autocorrelation function value corresponding to each k is calculated until the first k value is found that makes the function value less than or equal to zero. This k value is the time delay parameter. For example, when the autocorrelation function value is negative for the first time when k = 3, the time delay parameter is determined to be 3. This parameter represents the decay rate of the internal correlation of the series and is used to determine the time interval for phase space reconstruction. The calculation process is performed in a 32-bit floating-point unit, and a single calculation takes no more than 5 milliseconds.

[0089] The embedding dimension parameter is determined based on the principle of permutation pattern diversity. Starting with a minimum dimension of m = 2, the iterative calculation begins: the interval time series is divided into continuous subsequences of length m, and the order of elements within each subsequence is analyzed. The order of elements is determined by comparing the values ​​of adjacent elements, forming a specific permutation pattern. The number of occurrences of different permutation patterns is counted and compared with the factorial value of m (i.e., 1 × 2 × ... × m). If the actual number of occurrences is less than the factorial of m, m is increased by 1 and the above process is repeated. When the actual number of occurrences reaches or exceeds the factorial of m for the first time, the iteration is stopped and the current m is determined as the embedding dimension parameter. This process ensures that the embedding dimension is large enough to distinguish different system states. The iteration limit is set to m = 6; if it exceeds this limit, m = 5 is automatically set.

[0090] Phase space reconstruction is performed according to the time delay parameter and the embedding dimension parameter. Taking an interval time series as input, for each starting position i in the sequence (i ranges from 1 to the sequence length minus (m-1) × the time delay parameter), a subsequence of m elements is extracted: starting at position i, a value is taken every time delay parameter, for a total of m values ​​forming an m-dimensional vector. The collection of all these vectors forms the reconstructed state space. The vector extraction process follows a temporal order, ensuring that each vector contains information about consecutive but non-adjacent time intervals. If the sequence length is insufficient, the reconstruction process automatically terminates with error code E201.

[0091] The symbolization of state-space vectors is achieved by sorting the elements by size. For each m-dimensional vector, the size of its internal elements is compared: the smallest element in the vector is marked as 0, the second smallest as 1, and so on, until the largest element is marked as m-1. When elements of the same value appear, they are marked in order of their original position in the vector, with those in the earlier positions receiving lower numbers. The symbolization result is represented as a sequence of numbers of length m, each corresponding to a specific element size arrangement. The symbolization results of all vectors form a set of symbol sequences. The processing is accelerated using a dedicated comparator array.

[0092] The frequency of occurrence of permutation patterns is counted using a hash table. An empty hash table is initialized to store the permutation patterns and their occurrence counts. Each symbol sequence in the symbol sequence set is traversed and used as a key in the hash table. If the key does not exist in the hash table, a new key is added and the count value is set to 1; if it already exists, the corresponding count value is increased by 1. After the traversal is completed, all permutation patterns and their occurrence frequencies are stored in the hash table. The statistical process excludes invalid patterns where all element values ​​are the same. The hash table capacity is preset to 1000 entries.

[0093] The probability distribution is calculated based on the frequency of occurrence of permutation patterns. First, the total number of symbol sequences (i.e., the number of elements in the state space vector set) is determined. Then, for each permutation pattern, its probability of occurrence is calculated by dividing the frequency of that pattern by the total number of symbol sequences. The probability values ​​for all permutations constitute the probability distribution. Probability values ​​are retained to four decimal places and normalized to ensure that the sum of all probabilities is strictly equal to 1. Normalization verification requires that the sum of probabilities deviate from 1 by less than 0.0001.

[0094] Permutation entropy is calculated using the information entropy formula. Taking a probability distribution as input, the following steps are performed for each permutation pattern: If the probability of the pattern is greater than zero, the probability value is multiplied by the base-2 logarithm of the probability value; if the probability is zero, the pattern is skipped. These product values ​​for all permutations are summed and then negated to obtain the permutation entropy value. The result is calculated to three decimal places. This calculation is performed in the floating-point unit, using a table lookup to accelerate the logarithmic calculation.

[0095] The preset permutation entropy threshold is determined based on historical data analysis. At least 100 time series samples are collected during normal transformer operation, and the permutation entropy value for each series is calculated. The arithmetic mean and standard deviation of all sample entropy values ​​are calculated. The preset permutation entropy threshold is set to the mean minus twice the standard deviation. This threshold is updated quarterly by collecting new operating data, recalculating the mean and standard deviation, and generating the updated threshold. The update process requires ≥50 new sample sets.

[0096] The permutation entropy value is compared to the threshold using a numerical comparator. The calculated permutation entropy value is converted to a 32-bit floating-point number and input into the digital comparator circuit along with the preset permutation entropy threshold. The comparator circuit outputs two states: a logic low (0) when the entropy value is greater than or equal to the threshold, and a logic high (1) when the entropy value is less than the threshold. The comparison result is stored in a status register for subsequent processing.

[0097] The trigger circuit generates the thermal degradation assessment trigger signal. When the comparator circuit outputs a logic high (1), the trigger circuit generates a 5V signal for 100 milliseconds. This signal is isolated via an optocoupler and transmitted to the main controller, simultaneously driving a red LED indicator. After the trigger signal is generated, a 500-ms blanking period automatically initiates, ignoring new trigger conditions during this period to prevent false triggering due to signal jitter. The signal rise time is less than 10 microseconds.

[0098] The phase space reconstruction boundary handling mechanism includes three response options: When the input sequence length is less than the minimum requirement (30 data points), error code E201 is returned and processing is suspended; when the time delay parameter is too large, resulting in insufficient valid vectors (fewer than 10 vectors), the embedding dimension parameter is automatically reduced to the minimum value of 2; and when an all-zero vector is detected, it is automatically skipped and not counted in the statistics. Error codes are reported through the system status register.

[0099] The exception handling process for permutation entropy calculations: When a probability distribution anomaly is detected (e.g., all probabilities are zero), the default safety entropy value of 1.0 is returned. When the calculated entropy value exceeds the theoretical range [0, log2(m!)], it is automatically truncated to the nearest boundary value. When the input parameters are invalid, the standard error code E202 is returned. The exception handling results are written to the diagnostic log and the system self-check procedure is triggered.

[0100] Linear interpolation is used to accurately locate the zero-crossing point in the time delay parameter calculation. When the autocorrelation function values ​​of two adjacent lag steps have opposite signs, linear interpolation is performed between the two points to calculate the precise lag step length at which the function value reaches zero. The interpolation formula is: k_zero = k1 + |r1| / (|r1|+|r2|), where r1 and r2 are adjacent autocorrelation values. This step length value is rounded to the nearest integer and used as the final time delay parameter.

[0101] k_zero: In the linear interpolation calculation of the autocorrelation function zero crossing, k_zero represents the exact lag step value corresponding to the interpolated autocorrelation function value equal to zero. This value is calculated using the interpolation formula using the autocorrelation function values ​​r1 and r2 of two adjacent lag steps k1 and k2 (k2 = k1 + 1), where k1 is the lag step reference point at which the autocorrelation function value changes from positive to negative.

[0102] k1: In linear interpolation calculations for autocorrelation function zero-crossings, k1 represents the last lag step value when the autocorrelation function changes from positive to negative. That is, when the lag step is k1, the autocorrelation function value is greater than zero, while when the lag step is k1+1, the autocorrelation function value is less than zero. k1 serves as the starting reference point for interpolation calculations.

[0103] Acceleration strategy during embedding dimension iteration: When m ≥ 4, a pattern type estimation method is used. By analyzing the pattern growth trends of the first three dimensions, the possible number of types in the current dimension is predicted, reducing the actual number of statistics. If the estimation error exceeds 20%, the algorithm falls back to full statistics mode.

[0104] Hardware optimization for symbolic processing: When the embedding dimension m is greater than 3, a dedicated comparator array is enabled to compare the size relationships of vector elements in parallel, reducing processing time to 1 / m of the original time. The comparator array contains 10 independent comparison units.

[0105] Normalization verification of probability distribution calculations: After the probability calculation is completed, all probability values ​​are summed and verified to be equal to 1. If the deviation exceeds 0.0001, the probability values ​​are adjusted proportionally to ensure that the sum is exactly 1. The adjustment formula is: p_corr = p_orig / sum(p).

[0106] To ensure accuracy in permutation entropy calculations, we use a piecewise linear approximation method to calculate the logarithmic function value, ensuring accuracy while avoiding complex floating-point operations. A 1024-entry lookup table is set up to store the common logarithm values, with an entry interval of 0.001.

[0107] p_corr: During the normalization process, p_corr represents the corrected probability value. When the original probability distribution sum(p) deviates from 1 by more than 0.0001, each original probability value p_orig is divided by the sum of probabilities sum(p) to obtain the normalized corrected probability value.

[0108] p_orig: In probability distribution calculations, p_orig represents the original probability value before adjustment. This value is directly calculated by dividing the frequency of the permutation pattern by the total number of symbol sequences. It is the initial probability value before normalization.

[0109] sum(p): In probability distribution verification, sum(p) represents the cumulative sum of all the original probability values ​​p_orig for each permutation pattern. This value is used to verify that the probability distribution meets the requirement of summing to 1. It is calculated by adding all the probability values ​​in the probability distribution dictionary.

[0110] Threshold update anomaly protection: When the standard deviation of the entropy value of a new sample is too large (exceeding 0.1), the threshold update is automatically frozen and a manual review is triggered. The update is performed during system idle time to avoid affecting real-time processing. Update data is stored in FRAM memory.

[0111] The signal generation circuit's anti-interference design uses a Schmitt trigger to shape the input signal, with a noise margin set to 200 millivolts. The output signal is transmitted via a twisted-pair cable, terminated with a 120-ohm terminating resistor.

[0112] S3. When the heat dissipation degradation assessment is triggered, the distributed optical fiber temperature monitoring data and the three-dimensional distribution data of the leakage magnetic field of the transformer winding are synchronously obtained. The specific implementation is as follows:

[0113] Distributed fiber-optic temperature monitoring data is acquired through the following technical solution: When the thermal degradation assessment trigger signal is activated, the central controller sends a start command to the fiber-optic interrogator. The fiber-optic interrogator controls a pulsed laser to emit 1550nm wavelength light pulses, which are transmitted via optical fiber to a distributed temperature sensor array. The sensor array is evenly spaced along the transformer winding axis. The node spacing is determined based on the winding's thermal conductivity characteristics: the axial temperature distribution curve is obtained by solving the heat conduction equation. The sensor array is denser in areas where the temperature change rate exceeds 2°C / cm. Specifically, a 50mm spacing is used in the main winding area and a 25mm spacing is used in the end area (10% of the winding length). Each sensor node calculates the temperature value by measuring the Stokes-to-Anti-Stokes intensity ratio of backscattered Raman light. The temperature value is equal to a calibration factor multiplied by the natural logarithm of the Stokes-to-Anti-Stokes intensity ratio. The calibration factor is obtained through factory oil bath constant temperature calibration at 40°C, 80°C, and 120°C.

[0114] The collection of three-dimensional distribution data of the leakage magnetic field is achieved through a three-axis Hall sensor array. The sensors are gridded in the outer space of the transformer, and the grid spacing is set based on the spatial attenuation characteristics of the leakage magnetic field: the magnetic field gradient distribution is calculated using Ampere's loop law, and the grid spacing is increased in areas where the gradient exceeds 10 microteslas / cm. The specific implementation is: a basic spacing of 200 mm is used around the outer periphery of the winding body, and the spacing is automatically increased to 100 mm in high-gradient areas such as core joints and clamp edges. Each three-axis Hall sensor simultaneously measures the magnetic induction intensity components in the X / Y / Z directions. The measured value is calculated through the linear relationship between the Hall voltage and the magnetic induction intensity, and the proportional coefficient is obtained by calibration of a standard magnetic field generator in the range of 100 microteslas to 10 milliteslas. A synchronous sampling and holding circuit is used during data acquisition to ensure that the three directional components are measured at the same microsecond.

[0115] The spatial coordinate mapping of the temperature sensor and the magnetic field measurement point is completed through spatial position alignment. A unified Cartesian coordinate system is established: the origin is located at the center point of the transformer base, which is determined by total station measurement and positioning. The coordinates of the temperature sensor are calculated through the installation parameters: the axial position Z is determined by the installation height, the circumferential position θ is determined by the installation angle, and the radial distance R is fixed to the winding radius. The coordinate conversion method is: the X coordinate is equal to the winding radius multiplied by the cosine of the angle θ, and the Y coordinate is equal to the winding radius multiplied by the sine of the angle θ. The coordinates of the magnetic field measurement point directly use the mechanical positioning coordinates of the mounting bracket. Each temperature sensor is associated with the three nearest magnetic field measurement points. The association rule is based on the principle of minimizing spatial distance: the Euclidean distance from the temperature sensor to all magnetic field measurement points is calculated, and the three points with the smallest distance are selected to establish a mapping relationship. The mapping relationship table is stored in the non-volatile memory.

[0116] The timestamp alignment of temperature and magnetic field data utilizes a layered time synchronization strategy. The first layer of synchronization is based on the IEEE1588 PTP protocol, achieving inter-system time synchronization accuracy of ±1 microsecond. The second layer of synchronization is based on acquisition trigger pulses. The central controller sends a global trigger pulse every second, with a pulse rising edge accuracy of ±100 nanoseconds. The third layer of synchronization is based on local high-precision clocks. Each data acquisition unit has a built-in temperature-compensated crystal oscillator with a clock drift rate of ±1 ppm. The alignment process uses the trigger pulse as the reference time point, calculates the time offset of each data packet, and aligns the asynchronous data to a unified time axis through linear interpolation. The interpolation method is: the target time value is equal to the previous time value plus the time difference divided by the time interval between the next and previous times, multiplied by the difference between the next and previous times.

[0117] The temperature distribution dataset is generated through spatial data reorganization. The collected discrete temperature data is reorganized by spatial position: first, the data is sorted in ascending order by the axial Z coordinate, and then the data at the same Z coordinate is sorted in ascending order by the circumferential angle θ. Linear interpolation is performed between adjacent sensor nodes to generate a continuous temperature distribution. The interpolation rule is: insert a data point every 10 mm in the axial spacing and every 5 degrees in the circumferential direction. The interpolation calculation method is: the temperature at the interpolation point is equal to the temperature of the previous node plus the position scale factor multiplied by the difference between the temperature of the next node and the temperature of the previous node. The dataset storage structure is a three-dimensional array: the first dimension index corresponds to the axial position, the second dimension index corresponds to the circumferential partition, and the third dimension stores the coordinates and temperature values. The temperature value accuracy is 0.1 degrees Celsius.

[0118] The generation of the three-dimensional magnetic field distribution dataset is completed through a spatial interpolation algorithm. Based on the magnetic field vector data of discrete measurement points, the inverse distance weighted interpolation method is used to generate a three-dimensional continuous distribution. The space surrounding the transformer is divided into a cubic grid, and the grid size is dynamically adjusted according to the magnetic field gradient: the basic grid size is 20 mm, and it is automatically encrypted to 10 mm in areas where the gradient exceeds 5 microtesla / mm. The magnetic field vector of each grid point is calculated by interpolating the data of the neighboring measurement points, and the interpolation weight is inversely proportional to the square of the distance. The interpolation calculation method is: the magnetic field value of the grid point is equal to the sum of the magnetic field values ​​of each measurement point multiplied by its weight coefficient, and then divided by the sum of the weight coefficients. The weight coefficient is equal to the inverse of the square of the distance from the measurement point to the grid point. The dataset is stored as a four-dimensional array, with the first three dimensions being spatial coordinate indexes and the fourth dimension storing the three magnetic field component values.

[0119] The installation and positioning of the distributed fiber optic temperature sensors are calibrated using a laser tracker. Reference points are marked on the winding surface before installation, and the three-dimensional coordinates of each sensor node are measured using a laser tracker with an accuracy of ±0.1 mm. After installation, any deviations from the designed coordinates exceeding 1 mm are re-adjusted. The sensor fiber is helically wound with a bend radius greater than 50 mm to prevent microbend losses from affecting temperature measurement accuracy. The fiber optic connectors use APC connectors with a return loss greater than 65 dB.

[0120] The triaxial magnetic field sensors are positioned and mounted using a non-magnetic aluminum alloy positioning frame. The frame has pre-set sensor mounting holes with a tolerance of ±0.5 mm. Before installation, a total station is used to measure the frame's reference point coordinates and establish coordinate system transformation parameters. After each sensor is installed, the actual coordinates are recorded, and compensation parameters are calibrated if the deviation from the designed coordinates exceeds 2 mm. The sensor cables are twisted-pair shielded, with the shield grounded at a single point.

[0121] Data acquisition synchronization is ensured by a hardware trigger circuit. A central controller generates a 1 Hz square wave trigger signal, which is transmitted to each acquisition unit via a 50 ohm coaxial cable. Trigger signal transmission delay is measured and compensated using time domain reflectometry with an accuracy of ±10 nanoseconds. Each acquisition unit initiates acquisition on the rising edge of the trigger signal, with an acquisition window time of 1 millisecond and a sampling rate of 10 kHz.

[0122] Coordinate mapping calibration mechanism: Key point coordinates are remeasured and coordinate conversion parameters updated during each transformer overhaul. During online monitoring, coordinate accuracy is verified by temperature distribution symmetry. The temperature difference between left and right symmetrical locations is calculated, and the coordinate calibration process is triggered when the average difference exceeds 10%. The calibration process is automated and takes less than 5 minutes.

[0123] Timestamp alignment is verified monthly by injecting a 5°C temperature step signal through a standard resistance heater and a 100 microtesla magnetic field step signal through a Helmholtz coil. The system measures the difference in temperature and magnetic field response time. If the difference exceeds 50 microseconds, compensation parameters are automatically adjusted. Test data is stored separately and does not affect normal operating data.

[0124] Dataset storage optimization: A discrete cosine transform compression algorithm is used, retaining 0.1°C accuracy for temperature data and 0.1 microtesla accuracy for magnetic field data. A compression ratio of 4:1 is set, and a 32-bit cyclic redundancy check (CRC) is added to the compressed data. Industrial-grade solid-state drives are used as storage media, with sufficient storage capacity for 30 days of continuous operation.

[0125] Sensor health monitoring: Real-time analysis of the statistical characteristics of each sensor data. The temperature sensor data range is limited to 0 to 150 degrees Celsius, with a rate of change limit of 10 degrees Celsius per second. The magnetic field sensor data range is limited to -1000 to +1000 microteslas, with a rate of change limit of 100 microteslas per second. If three consecutive samplings exceed the limit, an abnormality is flagged and the system automatically switches to the backup sensor channel. A self-calibration process is performed before the backup channel is activated.

[0126] Initialization of the temperature monitoring system: After power-up, a sensor scan is performed to establish a node location mapping table. The sampling frequency is set to 1 Hz, and the range is 0 to 150°C. A warm-up period of 10 minutes is allowed, during which data is marked as invalid. System status is reported in real time via a status word, including a readiness flag, error codes, and calibration status.

[0127] Initialization of the magnetic field measurement system: After power-on, perform sensor zero calibration to eliminate the influence of the geomagnetic field. Set the sampling frequency to 10 kHz and the measurement range to ±10 mTesla. Create a spatial grid index table and pre-allocate a memory buffer. After the system is initialized, send a ready signal to the central controller.

[0128] The coordinate conversion process is as follows: Read the temperature sensor installation parameter database and calculate the Cartesian coordinates of each node. Read the magnetic field sensor location database and create a spatial index. Execute a nearest neighbor search algorithm to associate each temperature node with the three closest magnetic field points. Generate a coordinate mapping table and verify data consistency.

[0129] Time base alignment is implemented as follows: The PTP master clock sends synchronization messages every second, and each slave device calibrates its local clock. A global trigger pulse is generated every second, with a rising edge accuracy of ±100 nanoseconds. The data acquisition unit records the local trigger reception time and stores the difference between the local trigger reception time and the theoretical time in a compensation table. The data timestamp is equal to the trigger time plus the local sampling time plus the compensation value.

[0130] Data integration quality control: After the temperature dataset is generated, a spatial continuity check is performed, triggering a review when the temperature difference between adjacent nodes exceeds 20 degrees Celsius. After the magnetic field dataset is generated, a divergence check is performed, and re-interpolation is performed when the divergence value exceeds the threshold. A data quality report is generated daily, including the percentage of valid data and outlier statistics.

[0131] S4. Construct a coupling network based on the distributed optical fiber temperature monitoring data and the three-dimensional distribution data of the leakage magnetic field, and extract the structural characteristic parameters of the coupling network to generate an actual heat dissipation degradation assessment value. The specific implementation is as follows:

[0132] The spatial correlation matrix is ​​constructed through three-dimensional coordinate distance calculation. The spatial position coordinates of the temperature sensor and the magnetic field measurement point generated in step S3 are obtained as input. For each temperature sensor node and each magnetic field measurement point node, the Euclidean distance between the two is calculated: the three-dimensional coordinate values ​​of the two nodes are obtained, the X-coordinate difference, the Y-coordinate difference, and the Z-coordinate difference are calculated, the three differences are squared and added together, and then the square root of the sum is calculated to obtain the distance value. The value of the matrix element is set to the reciprocal of the distance value. When the distance value is less than 1 mm, it is set to the reciprocal of 1 mm. The rows of the correlation matrix correspond to all temperature sensor nodes, and the columns correspond to all magnetic field measurement point nodes. The matrix dimension is the number of temperature nodes multiplied by the number of magnetic field nodes.

[0133] The coupled network topology is constructed based on a spatial correlation matrix. Each temperature sensor node and each magnetic field measurement point node is defined as a network node. For each element in the correlation matrix whose value is greater than the reciprocal of 0.1 mm, a bidirectional edge is established between the corresponding temperature node and magnetic field node. The network structure is stored using an adjacency list data structure, with each node recording information about all its connected neighbors. After construction, the network connectivity is checked, and isolated nodes without connecting edges are marked as invalid and excluded from subsequent calculations.

[0134] Node betweenness centrality is calculated using the Brandes algorithm. First, the number of shortest paths between all pairs of nodes in the network is calculated. Next, the number of shortest paths passing through each node is counted. Finally, the number of shortest paths passing through each node is divided by the total number of node pairs to obtain the betweenness centrality value for that node. The calculation uses the Floyd algorithm to solve for all-source shortest paths. When the number of nodes exceeds 500, stratified sampling is used to reduce computational complexity, with a sampling ratio of 10% of the total number of nodes.

[0135] The preset betweenness threshold is determined based on historical data analysis. Multiple sets of network data are collected from transformers under normal operating conditions. The betweenness centrality values ​​for all nodes in each set are calculated, followed by the mean and standard deviation of these values. The preset betweenness threshold is set to the mean plus twice the standard deviation. This threshold is updated quarterly, with the mean and standard deviation recalculated using the most recently collected operating data. The new data sample size must be at least 50 sets.

[0136] The edge weight value is calculated by the dot product of the temperature gradient vector and the magnetic field gradient vector. For each connecting edge, first obtain the temperature node position and the magnetic field node position connected by the edge. Temperature gradient vector calculation: At the temperature node position, obtain the adjacent temperature nodes in the X / Y / Z directions, and calculate the temperature change rate in the three directions by the central difference method. The rate of change is equal to the temperature value of the rear node minus the temperature value of the front node divided by the node spacing. The magnetic field gradient vector is calculated similarly. The edge weight value is equal to the X component of the temperature gradient vector multiplied by the X component of the magnetic field gradient vector, plus the product of the Y component, plus the product of the Z component. When the node is at the boundary, the forward or backward difference method is used to calculate the gradient.

[0137] The statistical distribution characteristics of edge weights are extracted using four statistics: the mean of all weights, the standard deviation of all weights, the skewness of all weights, and the kurtosis of all weights. The mean is calculated as the sum of all weights divided by the total number of edges; the standard deviation is calculated as the square root of the sum of the squares of the difference between each weight and the mean, divided by the total number of edges; the skewness is calculated as the third-order central moment of the weight divided by the cube of the standard deviation; and the kurtosis is calculated as the fourth-order central moment of the weight divided by the fourth power of the standard deviation, minus 3. Calculations are performed to four decimal places.

[0138] The modular structure of coupled networks is identified by maximizing the modularity function. Initially, each node is considered an independent community. All possible community merging schemes are then traversed, and the one that maximizes the modularity (Q) value is selected for merging. This process is repeated until the Q value increase is less than 0.001 or the number of iterations reaches 100. The modularity (Q) value is defined as the square of the sum of the node edge weights within a community minus the actual edge weight ratio within the community. The number of coupled subnetworks ultimately identified is recorded.

[0139] Calculation of average path length for coupled subnetworks: For each identified coupled subnetwork, the average shortest path length between all node pairs is calculated. Dijkstra's algorithm is used to calculate the shortest path length from each node to every other node, and the arithmetic average of these path lengths is then taken for all node pairs. When a subnetwork is disconnected, only node pairs within the largest connected component are evaluated, ignoring isolated nodes.

[0140] Calculate the clustering coefficient of a coupled subnetwork: For each node in the subnetwork, calculate the ratio of the actual number of edges between its neighboring nodes to the maximum possible number of edges. Then, take the arithmetic average of this ratio for all nodes. If the number of neighboring nodes is less than 2, the clustering coefficient of that node is defined as 0. Calculate the result to four decimal places.

[0141] Calculate the ratio of average path length to clustering coefficient: For each coupled subnetwork, divide the average path length by the clustering coefficient to obtain the ratio. If the clustering coefficient is 0, set the clustering coefficient to 0.001 and calculate the ratio again. After the calculation is completed, calculate the arithmetic mean of this ratio for all coupled subnetworks and use it as the fourth dimension eigenvalue.

[0142] The generated multidimensional feature vector contains four dimensions: the first dimension is the proportion of nodes whose betweenness centrality exceeds a preset betweenness threshold; the second dimension is the statistical distribution of edge weights, including mean, standard deviation, skewness, and kurtosis; the third dimension is the number of coupled subnetworks; and the fourth dimension is the arithmetic mean of the ratio of the average path length to the clustering coefficient across all coupled subnetworks. The feature vector is formatted as an array containing 8 elements in the following order: [node proportion, weight mean, weight standard deviation, weight skewness, weight kurtosis, number of subnetworks, path coefficient ratio mean].

[0143] The radial basis function interpolation model application process begins by loading the pre-trained model parameters, including the center point coordinates and weight coefficients. For each input feature vector, the Euclidean distance from the center point is calculated. This distance value is then substituted into a Gaussian kernel function to obtain the basis function value. The Gaussian kernel function is an exponential function consisting of the square of the distance divided by the negative of the kernel width. All basis function values ​​are multiplied by the corresponding weight coefficients and summed to obtain the actual thermal degradation assessment value. The model parameters are obtained through training with historical fault data using regularized least squares, with at least 200 training samples.

[0144] Abnormal handling of the spatial correlation matrix: when the distance between the temperature node and the magnetic field node is greater than 1000 mm, the matrix element is set to 0; when the coordinate data is missing, the average coordinate value of the adjacent nodes is used to supplement it; when the number of nodes does not match, the valid node subset is intercepted to construct the matrix and the abnormal log is recorded.

[0145] Optimizing betweenness centrality calculations: For large networks with more than 500 nodes, we randomly select 10% of the nodes as source nodes to calculate the shortest path and estimate betweenness centrality. The estimated result is multiplied by 10 and the final value is recorded within a ±15% error range.

[0146] Boundary processing of edge weight calculation: When a node is located at the boundary and the complete neighborhood cannot be obtained, the mirror method is used to supplement the virtual node value, and the virtual node value is equal to the boundary node value; when an abnormal value occurs in the gradient calculation, the median of the adjacent node gradient value is used instead, and the data quality flag is marked.

[0147] Modularity recognition acceleration strategy: Initial community divisions are quickly generated using the Louvain algorithm, followed by fine-tuning using the Newman algorithm. When the number of network nodes exceeds 1,000, the network is divided into multiple subgraphs for parallel processing, with the number of parallel threads equal to the number of processor cores.

[0148] Normalization of feature vectors: Subtract the historical mean of each dimension from the feature value, then divide by the historical standard deviation. Normalization parameters are updated synchronously with model parameters, with an update cycle of every 50 new valid data sets. The historical mean and standard deviation are calculated based on the most recent 1000 sets of normal data.

[0149] The radial basis function model update mechanism retrains model parameters every time 50 sets of valid operational data are added. The training process uses 5-fold cross-validation to select the regularization coefficient to prevent overfitting. Once the new model is validated with historical data, it is deployed when the prediction error decreases by more than 5%. Otherwise, the original model is retained.

[0150] The physical basis for gradient calculations: The temperature gradient vector reflects the direction of heat transfer, the magnetic field gradient vector reflects the trend of the leakage magnetic field, and the dot product of the two represents the strength of the thermal-magnetic coupling. This physical relationship serves as the theoretical foundation for edge weights, avoiding the subjectivity of manually setting weights.

[0151] The mathematical principle behind modularity identification is that maximizing the modularity function is equivalent to finding the natural community structure of a network. Through iterative optimization, the inherent correlation between temperature and magnetic field is discovered. The convergence of the algorithm is guaranteed by a modularity increment threshold of 0.001.

[0152] The design principles for constructing feature vectors are as follows: node ratio reflects the distribution of key nodes, weight statistics characterize the distribution of coupling strength, the number of subnetworks reflects system complexity, and the path coefficient ratio measures information transfer efficiency. Multi-dimensional features comprehensively cover the dimensions of heat dissipation degradation.

[0153] The rationale for applying radial basis functions is that they offer local approximation properties, making them suitable for handling nonlinear relationships. The Gaussian kernel ensures smooth transitions, the regularization term controls model complexity, and historical data training ensures reliable evaluation.

[0154] S5. Convert the actual heat dissipation degradation assessment value into a heat dissipation coefficient correction factor. Based on the transformer thermal circuit model, combine the heat dissipation coefficient correction factor and the real-time load current to generate a corrected safe load boundary. The specific implementation is as follows:

[0155] The determination of the basic heat dissipation coefficient is based on the thermodynamic principles of transformers. The basic heat dissipation coefficient characterizes the heat dissipation capacity of the transformer under rated operating conditions. Its physical meaning is the heat power that the transformer can dissipate per unit temperature difference. The method of obtaining it is: during the factory test phase of the transformer, operate it to a thermal equilibrium state under rated load and ambient temperature conditions, and measure the steady-state temperature rise of the winding and the total heat loss power. The basic heat dissipation coefficient is equal to the total heat loss power divided by the temperature difference between the winding and the ambient temperature. This parameter is stored in the system parameter database as an inherent property of the transformer and is loaded and used when the thermal circuit model is initialized. When the transformer is modified or overhauled, the basic heat dissipation coefficient needs to be retested and measured.

[0156] The piecewise linear mapping relationship between the actual heat dissipation degradation assessment value and the heat dissipation coefficient correction factor is established through calibration of historical operating data. Calibration process: The heat dissipation degradation assessment values ​​and corresponding actual heat dissipation capacity test data under different operating conditions of the transformer are collected. The actual heat dissipation capacity test is carried out during the transformer power outage and maintenance. The steady-state temperature rise is measured by injecting a constant heat source to infer the actual heat dissipation coefficient. The actual heat dissipation coefficient is divided by the basic heat dissipation coefficient to obtain the true correction factor. In the assessment value-correction factor coordinate system, three key turning points are selected: an assessment value of 0.3 corresponds to a correction factor of 1.0, an assessment value of 0.7 corresponds to a correction factor of 0.8, and an assessment value of 1.0 corresponds to a correction factor of 0.6. These three key points are connected by straight line segments to form a piecewise linear mapping relationship. The mapping relationship is stored in a parameterized form and supports runtime interpolation calculations.

[0157] Real-time transformer load current data is acquired through a measurement system. Precision current transformers (CTSs) with an accuracy of at least 0.2 are installed on the high-voltage and low-voltage busbars of the transformer. The collected raw current waveform data undergoes RMS calculations, with a calculation period of one power frequency cycle (20 milliseconds) to output 50 RMS current data points per second. Current values ​​are expressed in amperes, and the data is averaged per second to produce one load current data point per second. Data is transmitted in time-stamped data packets over redundant network channels.

[0158] The corrected heat dissipation coefficient is calculated using scalar multiplication. The base heat dissipation coefficient value is read from the parameter database to obtain the current heat dissipation coefficient correction factor. The corrected heat dissipation coefficient is equal to the base heat dissipation coefficient multiplied by the heat dissipation coefficient correction factor. The calculated result is range-checked: if the result is less than 0.6 times the base heat dissipation coefficient, it is forced to 0.6 times; if it is greater than 1.0 times the base heat dissipation coefficient, it is forced to 1.0 times. The calculation process log includes input and output values ​​and timestamps.

[0159] The thermal circuit differential equations are constructed based on the thermal circuit model topology. The model includes thermal capacitance parameters, which represent the winding's heat storage capacity, and thermal resistance parameters, which represent the resistance to heat conduction. The mathematical form of the equations is: the time derivative of the winding temperature is equal to the square of the load current multiplied by the heat generated by the winding resistance minus the heat dissipation coefficient multiplied by the temperature difference between the winding and the ambient temperature, divided by the thermal capacitance. Input parameters include the corrected heat dissipation coefficient, the real-time load current, the ambient temperature, and the initial winding temperature. The ambient temperature is the average value of thermometers located at multiple locations on the transformer body, while the initial winding temperature is a weighted average of distributed fiber-optic temperature monitoring data.

[0160] The thermal differential equations are solved using a variable-step fourth-order Runge-Kutta numerical method. The solver's initial step size is set to 1 second, and the temperature rate of change is evaluated at each computational step. When the absolute value of the temperature rate of change exceeds 5 Kelvin per minute, the step size is automatically reduced to 0.1 seconds; when the absolute value of the temperature rate of change is less than 1 Kelvin per minute, the step size is increased to 2 seconds. The solution continues until one of the following conditions is met: the absolute value of the temperature rate of change is less than 0.1 Kelvin per minute for 10 consecutive minutes (reaching steady state), or the cumulative solution time reaches 24 hours. Numerical stability is monitored in real time during the solution, and exception handling is triggered if the temperature value suddenly changes by more than 20 Kelvin.

[0161] The winding hotspot temperature curve is generated based on the results of solving the differential equation. The winding hotspot temperature value for each successful calculation step is recorded to form a time-temperature data series. The time axis uses Unix timestamp format with millisecond accuracy; the temperature unit is Celsius with one decimal place. The curve data is stored in a circular buffer, retaining the last 24 hours of data. A sliding average filter is applied to the raw data with a 60-second window width to eliminate high-frequency fluctuations.

[0162] The determination of the critical time point is achieved through a temperature threshold comparison algorithm. The limit temperature value of the insulation material is obtained, which is determined according to the insulation grade of the transformer: Class A insulation is 105 degrees Celsius, Class B is 130 degrees Celsius, and Class F is 155 degrees Celsius. In the hot spot temperature curve data sequence, a sequential scan is performed starting from the starting time to find the first data point whose temperature value is greater than or equal to the insulation limit temperature. If found, the time of this point is the critical time point; if it is not found at the end of the scan, the critical time point is set to 24 hours; if the starting temperature exceeds the limit, the critical time point is set to 0. The scanning process is optimized by binary search, first locating the temperature change section in the long-term data and then performing a fine search, with a search accuracy of 0.1 seconds.

[0163] The safe load boundary curve is generated based on the load current-critical time relationship. Fifteen load current levels are set within the range of 0.5 to 2.0 times the rated current, in steps of 0.1 times the rated current. For each current level, the load current value is fixed, and the thermal differential equations are resolved to calculate the critical time point under this constant load. The per-unit load current (actual current divided by rated current) is used as the abscissa, and the critical time is used as the ordinate to form a discrete data point set. A smooth curve is generated using a cubic spline interpolation algorithm, with the number of interpolation points set to 200. The curve data is stored as a piecewise polynomial, supporting fast runtime calculations.

[0164] Maintenance mechanism for piecewise linear mapping relationships: After each transformer overhaul, the mapping relationship is recalibrated based on the latest test data. The calibration process uses the least squares fitting method to optimize the slope of the broken line under three key constraints. The new calibration relationship must pass expert review before the system can be updated. During operation, the system records the correspondence between the evaluation value and the correction factor, triggering a calibration reminder when the deviation exceeds 10% for a sustained period.

[0165] Calibration of thermal circuit model parameters: Thermal capacity is calculated based on winding material properties and is equal to the copper density multiplied by the copper specific heat capacity multiplied by the winding mass. Thermal resistance is determined through transformer temperature rise testing and is equal to the steady-state temperature difference divided by the heat loss power. Parameter calibration is performed every five years or immediately upon any transformer structural changes. Calibration data is obtained from type test reports or field test records.

[0166] Real-time load current preprocessing: The raw current data is first subjected to outlier removal, using the 3σ criterion to identify and replace outliers. Phase correction is then performed to eliminate phase errors in the measurement system. Finally, band-limited filtering is performed to retain frequency components between 0.1 Hz and 100 Hz. The processed current data is then marked with a quality flag for subsequent use.

[0167] Initial conditions for solving the differential equations: The ambient temperature is averaged over the last hour to eliminate the effects of short-term fluctuations. The initial winding temperature is the 90th percentile of the distributed fiber optic temperature monitoring data, representing the hotspot temperature. If the initial temperature exceeds 90% of the insulation material's limit temperature, a conservative strategy is adopted, setting it to 90% of the limit temperature.

[0168] Hotspot temperature curve verification mechanism: Real-time verification is performed every two hours, comparing the calculated temperature with the actual fiber temperature measurement. Under steady-state conditions, a model calibration process is triggered if the deviation exceeds 5 degrees Celsius. Verification results are recorded in the operation log and used as a basis for model reliability assessment.

[0169] Conservative handling strategy for critical time points: When the load current exceeds 1.2 times the rated current, the critical time point is multiplied by a safety factor of 0.95; when it exceeds 1.5 times, it is multiplied by a safety factor of 0.9. For transient inrush current conditions, a dedicated algorithm is used to calculate the instantaneous heat accumulation effect.

[0170] Application of safe load boundary curves: Curve data is stored in a lookup table format, supporting real-time query of the allowable operating time at any load current. The system automatically generates a curve image file every hour, comparing the current curve with a historical baseline curve. An alert is triggered when the curve shape changes beyond a threshold, set to a rate of change of more than 10% in the area under the curve.

[0171] Online calibration of the thermal circuit model: Parameter optimization is performed monthly using normal operating data. Load current, ambient temperature, and measured hotspot temperature data are collected over 24 hours of continuous operation. A genetic algorithm is used to optimize thermal capacitance and thermal resistance parameters to minimize the root mean square error between the calculated and measured temperatures. The optimization process is performed on an offline server, and the new parameters are deployed after verification.

[0172] The physical meaning of the heat dissipation correction factor: The correction factor indicates the degree of reduction in the transformer's current heat dissipation capacity relative to rated operating conditions. A correction factor of 1.0 indicates perfect heat dissipation capacity, while a correction factor of 0.6 indicates a 40% reduction. This factor, a key input parameter of the thermal circuit model, directly affects the accuracy of hotspot temperature calculations.

[0173] Boundary conditions for the thermal differential equations: The ambient temperature is set at a monitoring point 5 meters from the transformer, updated hourly. The initial winding temperature is the highest temperature value provided by the fiber optic temperature measurement system, updated every 10 seconds. The thermal capacitance and thermal resistance parameters are assumed to be constant within the range of -40°C to +150°C.

[0174] Application scenarios for critical time points: When a transformer is facing overload, the system calculates the critical time point under the current load in real time, providing operators with a safe operating time window. This time point takes into account the thermal aging characteristics of the insulation material to ensure that the transformer operates within a safe temperature range.

[0175] The safe load boundary curve update strategy: Each time a new curve is generated, it is compared with the historical curve from the past 30 days. If the critical time point changes by more than 10% under the same load current, an early warning notification is automatically issued. Curve data is stored in encrypted form and includes a digital signature to verify integrity.

[0176] S6. Output the transformer load capacity assessment result including the corrected safe load boundary, which is specifically implemented as follows:

[0177] The safe load limit curve is converted into a visual graphical format using a graphics rendering engine. The data points of the safe load limit curve are read in. The curve is plotted against the per-unit load current (U / Pu) value (ranging from 0.5 to 2.0) and the permissible duration (in hours) on the vertical axis. Vector graphics technology is used to generate the curve: first, a coordinate system is established, with the horizontal axis labeled "Load Current / Rated Current" and the vertical axis labeled "Permissible Duration (Hours)." The data points are then connected to form a smooth curve, colored red. Finally, grid lines and tick marks are added. Real-time operating condition markers are generated through dynamic calculation: the current load current value is obtained and the corresponding point is marked on the curve. The marker is a 10-pixel-diameter circle filled with blue. The graphics format uses the SVG standard, supports lossless scaling, and is set to a resolution of 1920×1080 pixels.

[0178] The load capacity assessment report text is generated using a template engine. The report contains the following core sections: The first section describes the current safe load boundary curve version and generation time; the second section lists key parameters, including the maximum allowable load current, minimum allowable duration, and area under the curve; the third section analyzes historical trends and compares the current curve with last week's curve; and the fourth section provides operational recommendations. Text generation rules: Parameter values ​​are retrieved from the database in real time and inserted into a preset statement framework; the root mean square deviation of the current curve and the historical curve is calculated based on the trend; operational recommendations select predefined text based on the risk level of over-limit. The report language is Chinese, the font size is Songti size 4, and the paragraph spacing is 1.5 times.

[0179] Real-time comparison of the current load current against the safe load boundary curve is achieved through interpolation. This comparison is performed once per second: first, the current load current value is obtained and converted to per-unit value. Then, within the safe load boundary curve data point set, two adjacent data points are found (where the current value of the preceding point is less than the current value, and the current value of the following point is greater than the current value). Finally, the allowable duration is calculated using linear interpolation: the allowable duration is equal to the duration of the preceding point plus (the current value minus the current value of the preceding point) divided by the current value of the following point minus the current value of the preceding point, multiplied by the duration of the following point minus the duration of the preceding point. The result is rounded to one decimal place.

[0180] The warning signal triggering mechanism is based on a multi-level threshold design. When the actual operating time corresponding to the real-time load current exceeds 90% of the permitted duration, a level 1 warning (yellow) is triggered; when it exceeds 100%, a level 2 warning (orange) is triggered; and when it exceeds 110%, a level 3 warning (red) is triggered. The warning signal contains the following elements: warning level code, trigger time, percentage exceeded, and recommended action. Once the warning is generated, it is immediately sent to the message queue, and the audible alarm is activated. The warning persists until the limit-exceeding state ends, at which point a release notification is automatically sent.

[0181] The comprehensive assessment results are integrated using multi-source data fusion technology. A comprehensive data structure is created, consisting of three subcomponents: a visualization component stored as an SVG file object; a load capacity assessment report component stored as an HTML text object; and a warning signal component stored as a JSON message object. The integration process adds unified metadata: generation timestamp, transformer ID, and data version number. The data structure is encapsulated using binary serialization, limited to 1MB in size, and supports both network transmission and local storage.

[0182] The human-computer interaction interface (HCI) is implemented using web technology. The interface layout is divided into three areas: the left area embeds visual graphics, allowing users to hover their mouse to view coordinate values; the central area displays the load capacity assessment report text, with scrolling support; and the right area houses an early warning information panel, which updates the warning status in real time. The interface refreshes at a 1Hz frequency, using asynchronous loading technology to avoid lag. User interaction features include: curve zooming (supporting 1-5x magnification), report printing (generated in PDF format), and historical data comparison (overlaying multiple-day curves). The interface's theme color is industrial blue, conforming to power system visual standards.

[0183] Dynamic update of the load current-to-allowable duration curve: The visualization automatically updates when the safe load boundary curve is regenerated. The update process uses a smooth transition animation lasting 500 milliseconds. The old curve fades out while the new curve appears, avoiding visual jumps. The coordinate axis range is adaptively adjusted to ensure the complete display of the curve.

[0184] Real-time operating condition marker position calculation: The marker position is updated every 2 seconds. The current load current value is obtained and the coordinates of the corresponding point on the safe load boundary curve are calculated. The coordinate conversion formula is: the X coordinate equals the horizontal coordinate of the curve coordinate system origin plus (current current value minus minimum current value) divided by (maximum current value minus minimum current value) multiplied by the curve width; the Y coordinate is calculated similarly. A pulsating animation effect is added to the marker point, with a diameter that periodically varies between 8 and 12 pixels at a frequency of 1 Hz.

[0185] Key parameters for the load capacity assessment report are calculated as follows: the maximum allowable load current is calculated as the maximum value on the abscissa of the curve; the minimum allowable duration is calculated as the minimum value on the ordinate of the curve; and the area under the curve is calculated using the trapezoidal method of numerical integration: the curve is divided into 100 small intervals, and the area of ​​each trapezoid is accumulated. The result is rounded to two significant digits and the units are per unit, hours, and per unit·hour.

[0186] A library of recommended measures for early warning signals: Level 1 suggests "enhance temperature monitoring"; Level 2 suggests "reduce load or initiate cooling"; and Level 3 suggests "immediately reduce load and inspect." The measures library supports manual maintenance and is updated no more than once a month. The early warning information panel displays the 10 most recent warning records in reverse chronological order.

[0187] Exception handling in the human-computer interaction interface: When data loading fails, a placeholder graphic and error message are displayed; when the network is disconnected, local cached data is automatically switched; when a user operation times out, a session renewal dialog box pops up. The interface operation log records all user interaction events and is retained for 90 days.

[0188] Visualization elements: Add color bands below the curve to mark the safe zone: green (0%-90%), yellow (90%-100%), and red (>100%). Add unit labels to the coordinate axes: "Load current / rated current (per unit)" on the horizontal axis and "Permitted duration (hours)" on the vertical axis. Add a legend in the upper right corner of the graph to explain the curve and marked points.

[0189] Trend analysis in the load capacity assessment report: Calculate the average deviation between the current curve and the previous week's curve. Deviation calculation formula: Take 20 equally spaced current points and calculate the average absolute value of the difference in the allowable duration between them. Deviations less than 10% are marked "stable," 10% to 20% are marked "caution," and greater than 20% are marked "significant change." A text description of the analysis results should be inserted in the third paragraph of the report.

[0190] Real-time comparison boundary processing: When the load current is below the minimum current of the curve, the allowed duration is set to 24 hours; when it is above the maximum current, the allowed duration is set to 0. Comparison results are recorded once per second and stored in a time series database, retaining 30 days of historical data.

[0191] Warning signal transmission protocol: MQTT is used to publish warning messages, with the subject format being "transformer / ID / warning." The message body includes the warning level, trigger time, limit-exceeded percentage, and recommended measures. Subscribers include operation monitoring systems, mobile apps, and audible and visual alarms.

[0192] Storage solution for comprehensive assessment results: Each time a result is generated, a database record is created. This record includes the result ID, generation time, transformer ID, graphic file path, report text, and warning status. Optimized database indexing supports fast searches by time range. Archiving is performed daily at midnight, compressing old data to save space.

[0193] Access control on the human-computer interface: Administrators can modify parameter templates; operators can view all data; and guests can only view the current status. User logins utilize two-factor authentication, with sessions valid for 8 hours. Secondary password confirmation is required to operate sensitive functions.

[0194] The coordinate range of the safe load limit curve is: the minimum horizontal axis value is 0.5 times the rated current, and the maximum value is 2.0 times the rated current; the minimum vertical axis value is 0 hours, and the maximum value is 24 hours. The coordinate scale interval is: 0.1 times the rated current on the horizontal axis, and 1 hour on the vertical axis. The grid line density is: 1 unit per main grid, and 0.1 unit per auxiliary grid.

[0195] Visual feedback for real-time working condition markers: When a marker enters the yellow zone, its border turns yellow; when it enters the red zone, it flashes red. A tooltip displays next to the marker, indicating the remaining safety time under the current load. The tooltip automatically adjusts to avoid overlap.

[0196] The load capacity assessment report is generated automatically after each safe load boundary curve update; scheduled daily at midnight; or manually triggered. The report version number consists of a date and a sequence number in the format YYYYMMDD-NNN.

[0197] The warning signal is automatically lifted when the actual operating time falls below 85% of the permitted duration. The release notification includes the release time, duration, and maximum overrun percentage. The release notification uses the same transmission channel as the warning signal.

[0198] Data verification of comprehensive assessment results: After each generation, integrity checks are performed, including file size verification, digital signature verification, and data range checks. Automatic regeneration occurs if verification fails, and three consecutive failures trigger a system alarm.

[0199] The user interface is responsive: The layout adapts to different screen sizes, displaying a three-column layout on desktop and switching to a tabbed layout on mobile. Touch gestures support zooming, with a minimum touch target size of 40×40 pixels.

[0200] In the field of transformer condition assessment, this embodiment establishes a spatially coupled network of temperature and magnetic field gradients to achieve topological characterization of the degree of heat dissipation degradation. Edge weights are formed using the dot product of the temperature gradient vector and the leakage magnetic field gradient vector, physically revealing the inherent correlation between thermal and magnetic energy transfer. Modular identification spontaneously divides the temperature-magnetic field coupling subnetwork to discover hidden fault areas within the equipment. In the safety boundary generation phase, network features are dynamically integrated with the thermal circuit model. Multidimensional network features are mapped into heat dissipation coefficient correction factors based on radial basis function interpolation. By correcting the thermal circuit parameters in real time, the safety boundary dynamically adapts to the equipment's degradation state. The entire technical chain forms a tight closed loop: from physical field data acquisition to coupled network construction, from feature extraction to model correction, and ultimately generating an operational safety boundary. Local abnormal conduction paths are revealed through the spatial correlation matrix, and key failure nodes are identified through betweenness centrality, achieving a quantitative assessment of the degree of degradation.

[0201] The calculations involved in the embodiments are all dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to actual conditions.

[0202] It should be noted that the present invention can be deployed on the device itself to implement embedded applications, and can also be run on a PC or other terminal with a user interface, thereby meeting various hardware environments and usage requirements.

[0203] The above embodiments can be implemented in whole or in part via software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions or computer programs. When loaded or executed on a computer, the processes or functions described in the embodiments of this application are fully or partially performed. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired means (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server or data center that contains a collection of one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, tapes), optical media (e.g., DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0204] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and modules described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0205] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0206] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, and may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected to achieve the purpose of this embodiment according to actual needs.

[0207] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0208] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0209] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

[0210] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for evaluating the load capacity of a new energy transmission transformer, characterized in that: include: S1. Real-time record of the timestamp of each action of the transformer on-load tap-changer, calculate the interval time between adjacent actions based on the timestamp and construct an interval time series; S2. After reconstructing the phase space of the interval time series, a permutation entropy value is generated. If the permutation entropy value is lower than a preset permutation entropy threshold, a heat dissipation degradation assessment is triggered, including: Calculate time delay parameters and embedding dimension parameters based on interval time series; Reconstruct the phase space of the interval time series according to the time delay parameter and the embedding dimension parameter to generate a set of state space vectors; Symbolize each vector in the state space vector set to form a symbol sequence; Count the frequencies of different permutation patterns in all symbol sequences; Calculate the probability distribution based on the frequency of occurrence of the arrangement pattern; Calculate the permutation entropy value based on the probability distribution using the information entropy formula; Comparing the permutation entropy value with a preset permutation entropy threshold; When the permutation entropy value is less than a preset permutation entropy threshold, a heat dissipation degradation assessment trigger signal is generated; S3. When the heat dissipation degradation assessment is triggered, the distributed optical fiber temperature monitoring data and the three-dimensional distribution data of the leakage magnetic field of the transformer winding are simultaneously obtained; S4. Constructing a coupling network based on the distributed optical fiber temperature monitoring data and the three-dimensional distribution data of the leakage magnetic field, and extracting the structural characteristic parameters of the coupling network to generate an actual heat dissipation degradation assessment value; S5. Convert the actual heat dissipation degradation assessment value into a heat dissipation coefficient correction factor, and generate a corrected safe load boundary based on the transformer thermal circuit model, combining the heat dissipation coefficient correction factor and the real-time load current; S6. Output the transformer load capacity assessment result including the revised safe load boundary.

2. A method for evaluating the load capacity of a new energy transmission transformer according to claim 1, characterized in that: Record the timestamp of each action of the transformer on-load tap-changer in real time, calculate the interval time between adjacent actions based on the timestamps, and construct an interval time series, including: Record the exact time of each on-load tap-changer action as a timestamp; Store the timestamps of consecutive actions in chronological order to form a timestamp sequence; Calculate the time difference between adjacent timestamps as the action interval; Arrange the action intervals in the order in which the actions occur to form an interval time sequence.

3. A method for evaluating the load capacity of a new energy transmission transformer according to claim 1, characterized in that: The time delay parameter is determined by the first zero crossing point of the autocorrelation function, and the embedding dimension parameter is determined according to the rule that the number of permutation and combination patterns is equal to the factorial of the embedding dimension.

4. A method for evaluating the load capacity of a new energy transmission transformer according to claim 1, characterized in that: When the thermal degradation assessment is triggered, the distributed optical fiber temperature monitoring data and leakage magnetic field 3D distribution data of the transformer winding are acquired simultaneously, including: In response to a heat dissipation degradation assessment trigger signal, the temperature monitoring system and the magnetic field measurement system are simultaneously activated; The temperature distribution data of the winding surface is collected by a distributed optical fiber temperature sensor array, and each sensor node obtains the spatial position coordinates and temperature value; The three-axis magnetic field sensor array is used to collect magnetic induction intensity component data of discrete points in space, and the three-dimensional coordinates and magnetic field vector are obtained for each measurement point; Mapping the temperature sensor spatial coordinates and the magnetic field measurement point spatial coordinates to a unified coordinate system; Align the timestamps of temperature data and magnetic field data to the same time base; Generate a temperature distribution dataset and a three-dimensional magnetic field distribution dataset containing spatial position information.

5. The method for evaluating the load capacity of a new energy transmission transformer according to claim 1, characterized in that: A coupling network is constructed based on distributed optical fiber temperature monitoring data and three-dimensional leakage magnetic field distribution data. The structural characteristic parameters of the coupling network are extracted to generate the actual heat dissipation degradation assessment value, including: Establish a spatial correlation matrix between temperature nodes and magnetic field nodes, where the matrix elements represent the inverse of the physical distance between the temperature sensor and the magnetic field measurement point; Construct a coupled network topology, where the temperature nodes and magnetic field nodes are connected through an association matrix to form bidirectional edges; Calculate the betweenness centrality of all nodes in the coupled network; Calculate the proportion of nodes whose betweenness centrality exceeds the preset betweenness threshold; Calculate all edge weights in the coupled network; Extract the statistical distribution characteristics of all edge weight values; Identify the modular structure of the coupled network and divide the temperature-magnetic field coupled subnetworks by maximizing the modularity function; Calculate the ratio of the average path length to the clustering coefficient of each coupled subnetwork; Generate a multidimensional feature vector based on the statistical distribution characteristics of node ratio, edge weight value, number of coupled subnetworks and the ratio of average path length to clustering coefficient; The multidimensional feature vector is input into the radial basis function interpolation model to calculate the actual heat dissipation degradation evaluation value.

6. A method for evaluating the load capacity of a new energy transmission transformer according to claim 5, characterized in that: The edge weight value is determined by the dot product of the temperature gradient vector and the magnetic field gradient vector.

7. The method for evaluating the load capacity of a new energy transmission transformer according to claim 1, characterized in that: The actual heat dissipation degradation assessment value is converted into a heat dissipation coefficient correction factor. Based on the transformer thermal circuit model, the heat dissipation coefficient correction factor and the real-time load current are combined to generate a corrected safe load boundary, including: The basic heat dissipation coefficient in the transformer thermal circuit model is defined to characterize the heat dissipation capacity of the transformer under rated operating conditions; Establish a piecewise linear mapping relationship between the actual heat dissipation degradation assessment value and the heat dissipation coefficient correction factor; Obtain real-time load current data of transformer; In the transformer thermal circuit model, the basic heat dissipation coefficient is multiplied by the heat dissipation coefficient correction factor to obtain the corrected heat dissipation coefficient; Solve the thermal circuit differential equations based on the corrected heat dissipation coefficient and real-time load current; Calculate the curve of winding hot spot temperature changing with time; Determine the critical time point when the hot spot temperature of the winding reaches the limit temperature of the insulation material; Generate safe load boundary curves based on critical time points and real-time load current.

8. A method for evaluating the load capacity of a new energy transmission transformer according to claim 7, characterized in that: The piecewise linear mapping relationship is obtained by calibrating historical fault data, and the thermal circuit differential equation group includes heat capacity parameters and thermal resistance parameters.

9. The method for evaluating the load capacity of a new energy transmission transformer according to claim 1, characterized in that: Output transformer load capacity assessment results with revised safe load limits, including: Convert the safety load boundary curve into a visual graphical format; Generate load capacity assessment report text; Compare the corresponding relationship between the current load current and the safe load boundary curve in real time; When the real-time load current exceeds the allowable value of the safe load boundary curve, an early warning signal is triggered; Integrate visualization graphics, load capacity assessment report text and warning signals into comprehensive assessment results; The comprehensive evaluation results are output through the human-computer interaction interface.

Images