Remote monitoring and detection system for transformer load

Abstract

The present application belongs to the technical field of power equipment monitoring, and particularly relates to a transformer load remote monitoring and detecting system. The system comprises a data acquisition module, an edge computing module, a load adjustment module, a load prediction module and a load analysis module. The data acquisition module acquires transformer secondary side load power, current, vibration signal and winding temperature data; the edge computing module processes load early warning, including data denoising; the load adjustment module calculates and adjusts the primary side load power; the load prediction module constructs a model to predict future primary side load power; and the load analysis module analyzes future load conditions according to the prediction results. The system realizes remote real-time monitoring, improves data processing accuracy, load adjustment accuracy and prediction capability, and guarantees safe and efficient operation of the transformer.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power equipment monitoring, and particularly relates to a transformer load remote monitoring and detecting system. BACKGROUND

[0002] As a key device in the power system, the load state of the transformer directly affects the safety and efficiency of power transmission. At present, the traditional transformer load monitoring method mainly relies on manual inspection or local monitoring, which has problems such as untimely data collection, delayed abnormal load response, and insufficient prediction accuracy. In actual operation, transformer load fluctuation can cause winding temperature rise and abnormal vibration, and long-term overload or unstable load can cause equipment failure and even power outage. At the same time, the existing monitoring system has poor noise processing effect on collected data, affecting the accuracy of load analysis; load adjustment is mainly based on a single parameter, without fully combining temperature, vibration and other multi-dimensional characteristics, so it is difficult to achieve accurate control; there is no effective load prediction mechanism, which cannot predict future load trends in advance, and cannot meet the needs of intelligent and remote monitoring of the power system. Therefore, there is an urgent need for a transformer load monitoring and detecting system that can realize remote real-time monitoring, accurate data processing, dynamic load adjustment and reliable load prediction. SUMMARY

[0003] The technical problems in the above background of the application are solved by proposing a transformer load remote monitoring and detecting system.

[0004] In order to achieve the above purpose, the technical solution adopted by the application is as follows: a data acquisition module, an edge computing module, a load adjustment module, a load prediction module, and a load analysis module are included.

[0005] The data acquisition module is used to collect the transformer secondary side load power, current, transformer vibration signal data and transformer winding temperature data in the monitoring period.

[0006] The edge computing module includes a data processing unit and a fast response unit, and is used to process obvious load warning conditions.

[0007] The load adjustment module is used to calculate the primary side load power according to the secondary side load power, and adjust the primary side load power according to the transformer winding temperature data and the transformer vibration signal data, so that the adjusted primary side load power is obtained.

[0008] The load prediction module collects the adjusted primary side load power in a period of time and other data in this period of time, integrates them into training data, constructs a load prediction model, inputs the training data to train the trained load prediction model, inputs the newly collected data to predict, and predicts the primary side load power in the future period.

[0009] Load Analysis Module: This module is used to analyze the transformer load based on the predicted primary-side load power for future periods, and to obtain the transformer load for those future periods.

[0010] Preferably, the data processing unit in the edge computing module is used to perform noise reduction processing on the collected data to obtain noise-removed data, specifically implemented as follows:

[0011] First, initialize the search space, defining the number of modes K and the penalty factor. The range of values ​​for each group Perform VMD decomposition to obtain K intrinsic mode components (IMFs);

[0012] Calculate the biobjective function, including energy concentration. and information entropy gain The energy concentration The calculation method is as follows ,in The information entropy gain is expressed as the variance. The calculation method is as follows: first, calculate the entropy of the original signal. Then calculate the weighted entropy sum after decomposition. , ,in For the energy of the k-th IMF, The total energy of all IMFs, The information entropy of the k-th IMF is calculated, and then the information entropy gain is calculated. ;

[0013] Genetic algorithms are used to find the best candidate. and For a two-fitness function, iteratively find the optimal one. ;

[0014] Next, the VMD decomposition yielded... For each IMF, a screening index is calculated, including permutation entropy (PE), trend consistency (TC), and frequency domain correlation (FC). The three are standardized and weighted, and the screening index is calculated. Based on a preset threshold, only IMF components with screening indices greater than or equal to the threshold are retained.

[0015] For the retained IMF components, wavelet threshold denoising is used for secondary purification to obtain... For the noisy IMF, calculate its residual distribution compared to the retained IMF components. Furthermore, it uses adversarial networks to learn residual patterns and generate repair residuals. ;

[0016] Finally, the data is reconstructed to obtain the denoised data. .

[0017] As preferred, after the denoising processing is implemented, there is also a verification operation, which is specifically implemented as:

[0018] Firstly, historical same period data is extracted, and the same process is used for denoising to obtain historical denoised data;

[0019] The dynamic similarity of the current denoised data and the historical denoised data is calculated, and the calculation method is , wherein is the data sequence after the current denoising, is the data sequence after the historical denoising, and L is the data length. If the dynamic similarity is less than the set threshold value, the screening threshold value is adjusted reversely, and the denoising is re-executed until the similarity requirement is met.

[0020] As preferred, the implementation of the fast response unit in the edge computing module for obvious load warning situation is that if the real-time secondary side load power exceeds 1.3 times of the device set value, the fast response is directly performed, and the warning is triggered.

[0021] As preferred, the calculation method of the load adjustment module for calculating the primary side load power according to the secondary side load power is:

[0022] Firstly, the harmonic additional loss power is calculated , wherein, is the total power of the secondary side load, is the load current harmonic distortion rate, is a load type correction coefficient;

[0023] The primary side input power is , wherein, is the primary side input power, is the transformer efficiency.

[0024] As preferred, the load current harmonic distortion rate reflects the distortion degree of the transformer secondary side load current waveform deviating from the sine wave, and the calculation method is the ratio of the effective value of the current harmonic component to the effective value of the fundamental component.

[0025] As preferred, the load adjustment module also has a physical correction unit, which is used to combine the transformer winding temperature and vibration signal characteristics to perform secondary correction on the basic primary side power, and the specific implementation is:

[0026] Firstly, a correlation formula of dynamic characteristics and transformer additional loss is constructed, and the relationship between vibration and loss under different loads is fitted through experiments, and the formula is: , wherein, The vibration acceleration effective value is collected and the vibration signal characteristics are collected synchronously with the power calculation period, and the vibration acceleration effective value is obtained by root mean square calculation, The rated value of the vibration acceleration, The vibration main frequency offset is collected and the vibration signal characteristics are collected synchronously with the power calculation period, and the vibration main frequency offset is obtained by fast Fourier transform, Represent the main frequency under rated load;

[0027] Then, combined with the temperature factor, the temperature correction coefficient is obtained by correction, and the calculation method is: , wherein, The temperature threshold value is:

[0028] The original side input power basic value is corrected by the correction coefficient, and the formula is: .

[0029] As preferred, the load adjustment module also has a load characteristic self-learning unit, which is used to identify the load state according to historical period data, and iteratively optimize To adapt to the actual load characteristics, the implementation steps are:

[0030] First, the load power fluctuation variance is calculated by calculating the load power fluctuation of the secondary side in the historical period , then the THD data collected in the historical period is averaged to obtain the period average harmonic distortion rate ;

[0031] The load power fluctuation variance and the period average harmonic distortion rate are standardized to obtain the standardized and , and the load characteristic deviation degree is constructed , and the calculation method is , and compared with the set load characteristic deviation threshold ;

[0032] If it is less than or equal to , no change is made, and if it is greater than , the optimization is triggered, .

[0033] As preferred, the load prediction module constructs a load prediction model using a bidirectional long short-term memory network, and a residual interaction is added between its attention layer and full connection layer, and the specific implementation is:

[0034] First, the short-term key features output by the attention layer are received ;

[0035] The long-term sequence of the training data is called, and the long-term trend component is calculated by sliding window ;

[0036] Residual error of short-term feature and long-term trend ;

[0037] Through the gating unit Wherein is a weight matrix, G is a gating value controlling the compensation ratio of the residual error, the output , into the full connection layer.

[0038] As preferred, the load analysis module is used to analyze the transformer load condition according to the predicted primary side load power of the future period, and the implementation of the transformer load condition of the future period is:

[0039] First, set the safe load interval of the primary side input power , for the primary side load power of each predicted time t in the future ;

[0040] If , calculate the abnormality index , wherein, is the rated primary side input power;

[0041] If , the load change rate is checked, and the calculation method is , wherein is the detection of the predicted time, is the primary side load power of the last predicted time, is the load change rate; wherein, if , it is determined that the primary side power is stable and normal, if , it is determined that the primary side input power exists sudden fluctuation, and the abnormality index is calculated.

[0042] If , the abnormality index is calculated.

[0043] The obtained abnormality index is standardized, and if the abnormality index is greater than the set abnormality index threshold value, the transformer load warning is carried out.

[0044] Compared with the prior art, the advantages and positive effects of the present application are:

[0045] 1. VMD decomposition combined with genetic algorithm optimization is adopted, the modal parameters are optimized through the double objective functions of energy concentration degree and information entropy gain, and the effective IMF component is reserved; the noise component is repaired by using the residual error law of the adversarial network learning, combined with the dynamic similarity test of historical data, the data quality is greatly improved, and the problem of poor traditional noise processing effect is solved.

[0046] 2、 Based on the secondary side power calculation primary side power, combined with winding temperature, vibration signal to build correction model, through the physical correction unit optimization twice; load characteristics self-learning unit iteration optimization type coefficient, adaptive actual load characteristics, overcome the limitations of single parameter adjustment, improve the accuracy of power calculation.

[0047] 3、 Adopt bidirectional LSTM network, increase residual interaction between attention layer and full connection layer, fuse short-term key features and long-term trend components, dynamically compensate residual through gate unit, improve future period primary side power prediction accuracy, solve the problem of traditional prediction deficiency.

[0048] 4、 Set the safe load interval, calculate the abnormal index for different power conditions, trigger the warning combined with the standardization processing, compared with the simple threshold judgment, more comprehensive reflect the load state, improve the reliability of early warning. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to more clearly illustrate the technical scheme of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description, obviously, the drawings in the following description are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0050] Figure 1 The structure flow diagram of the transformer load remote monitoring detection system. DETAILED DESCRIPTION

[0051] In order to more clearly illustrate the above-mentioned purposes, features and advantages of the present application, the following will further illustrate the present application combined with the drawings and embodiments. It should be noted that, in the case of no conflict, the embodiments of the present application and the features in the embodiments can be combined with each other.

[0052] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, however, the present application can also be implemented in other ways different from the description, therefore, the present application is not limited to the specific embodiments disclosed in the following description.

[0053] Embodiment, the transformer as the core equipment of power system, its load state is directly related to the safety and economy of power transmission and distribution. However, the existing technology has the problems of monitoring lag and insufficient prediction accuracy, therefore, the present application proposes a transformer load remote monitoring detection system, the specific implementation process is as shown in Figure 1 The transformer load remote monitoring detection system includes a data acquisition module, an edge computing module, a load adjustment module, a load prediction module and a load analysis module.

[0054] The data acquisition module is configured to acquire the secondary side load power and current of the transformer, vibration signal data of the transformer and winding temperature data of the transformer in a monitoring period.

[0055] The edge computing module includes a data processing unit and a fast response unit, and is configured to process the load warning condition with obvious cases. The data processing unit is configured to perform denoising processing on the acquired data to obtain denoised data, and the denoising processing is specifically implemented as follows: first, an initial search space is searched, and a value range of a modal number K and a penalty factor is defined. For each group , VMD decomposition is performed to obtain K intrinsic modal components (IMFs). The double-objective function includes an energy concentration and an information entropy gain , the energy concentration is calculated as follows: , wherein represents a variance, and the information entropy gain is calculated as follows: , wherein is the energy of the kth IMF, is the total energy of all IMFs, is the information entropy of the kth IMF, and then the information entropy gain is calculated. and are used as the double-objective function, and the optimal is iteratively found. Then, the K IMFs obtained by VMD decomposition are calculated for a screening index, including permutation entropy PE, trend consistency TC and frequency domain correlation FC, the three are standardized and weighted, the screening index is calculated, and the threshold is preset to judge, only the IMF component with a screening index greater than or equal to the threshold is reserved. For the reserved IMF component, wavelet threshold denoising is used for secondary purification to obtain For the noise IMF, the residual distribution of the noise IMF and the reserved IMF component is calculated, and an adversarial network is used to learn the residual law to generate a repaired residual .

[0056] After the data denoising operation, there is also a verification operation, which is specifically implemented as follows: first, historical same-period data is extracted, and the same process is used for denoising to obtain historical denoised data; the dynamic similarity between the current denoised data and the historical denoised data is calculated as follows: , wherein is a current denoised data sequence, is a historical denoised data sequence, L is a data length, if the dynamic similarity is less than a set threshold value, the screening threshold value is adjusted reversely, the denoising is re-executed until the similarity requirement is met.

[0057] The implementation of the fast response unit in the edge computing module to handle the obvious load warning condition is that if the real-time secondary side load power exceeds 1.3 times of the device set value, the fast response is directly performed to trigger the warning. Specifically, in order to achieve the effect of capturing sudden overload in real time, the system continuously monitors the secondary side load power signal after multi-stage denoising. Once it is detected that the power value at a certain time exceeds 1.3 times of the rated value of the device, a hardware interrupt request is sent to the edge controller, and an overload warning message is sent to the local human-machine interface and the control center.

[0058] The load adjustment module is used to calculate the primary side load power according to the secondary side load power, and adjust the primary side load power according to the transformer winding temperature data and the transformer vibration signal data, so as to obtain the adjusted primary side load power. The calculation method of the load adjustment module for calculating the primary side load power according to the secondary side load power is as follows: first, the harmonic additional loss power is calculated , wherein, is the total power of the secondary side load, is the load current harmonic distortion rate, is the load type correction coefficient; the primary side input power is , wherein, is the primary side input power, is the transformer efficiency.

[0059] The load current harmonic distortion rate reflects the distortion degree of the transformer secondary side load current waveform deviating from the sine wave, and the calculation method is the ratio of the current harmonic component effective value to the fundamental component effective value. Specifically, first, the collected secondary side current signal is subjected to fast Fourier transform (FFT), and the amplitude corresponding to the fundamental component and the amplitude corresponding to each high-order harmonic component are extracted; then the total harmonic effective value obtained by squaring and taking the square root of the fundamental component effective value and the effective value of all harmonic components is calculated; finally, the harmonic distortion rate in percentage form is obtained by the ratio of the current harmonic component effective value to the fundamental component effective value. This index directly reflects the proportion of non-fundamental energy in the current waveform, and can be used to evaluate the influence of nonlinear load on transformer additional loss and temperature rise, and provide an important basis for subsequent loss correction and operation optimization.

[0060] The load adjustment module also has a physical correction unit, which is used for secondary correction of the basic primary power in combination with the transformer winding temperature and vibration signal characteristics, and is specifically implemented as follows: first, a correlation formula of dynamic characteristics and transformer additional loss is constructed, and the relationship between vibration and loss under different loads is fitted through experiments, and the formula is: wherein, is the effective value of vibration acceleration, the vibration signal characteristics are collected synchronously with the power calculation period, and the effective value is obtained through root mean square calculation, represents the rated value of vibration acceleration, is the vibration main frequency offset, the vibration signal characteristics are collected synchronously with the power calculation period, and the vibration main frequency offset is obtained through fast Fourier transform, represents the main frequency under rated load; then, the temperature factor is combined to perform correction again, and a temperature correction coefficient is obtained, and the calculation method is as follows: wherein, is the temperature threshold; the original primary input power basic value is corrected through the correction coefficient, and the formula is: .

[0061] The load adjustment module also has a load characteristic self-learning unit, which is used for identifying the load state according to historical period data and iteratively optimizing to adapt to the actual load characteristics, and the implementation steps are as follows:

[0062] First, the load power fluctuation variance is calculated by calculating the fluctuation of the secondary load power in the historical period, then the THD data collected in the historical period is averaged to obtain the period average harmonic distortion rate ; the load power fluctuation variance and the period average harmonic distortion rate are standardized to obtain the standardized and , and the load characteristic deviation degree is constructed, and the calculation method is as follows: wherein is a weight coefficient, and the load characteristic deviation degree threshold is compared; if it is less than or equal to , no change is made, and if it is greater than , the optimization is triggered, .

[0063] Load prediction module: collect the primary side load power after adjustment for a period of time, and other data in this period of time, including secondary side current, harmonic distortion rate, winding temperature, vibration characteristics, and environmental temperature data, integrate into training data, construct a load prediction model, input training data for training to obtain a trained load prediction model, input newly collected data for prediction, and predict the primary side load power in the future period. Specifically, the load prediction module first aligns the primary side load power collected in a period of time and the data in the same period in time sequence. Then, the preprocessed time sequence signal is divided into multiple training samples by using the sliding window method. In order to improve the generalization ability of the model, data enhancement operations such as window span expansion and noise disturbance are introduced in the training set. Next, a deep time sequence model based on bidirectional long short-term memory network (Bi-LSTM) is selected for prediction operation, and a residual interaction is added between its attention layer and fully connected layer, which is specifically implemented as follows: first, receive the short-term key features output by the attention layer ; call the long-term sequence of the training data, calculate the long-term trend component through the sliding window ; calculate the residual of the short-term feature and the long-term trend ; pass through the gate unit , where is a weight matrix, G is a gate value controlling the compensation ratio of the residual, and the output is transmitted to the fully connected layer.

[0064] Load analysis module: used to analyze the transformer load condition according to the predicted primary side load power in the future period, and obtain the transformer load condition in the future period. Specifically, first set the safe load interval of the primary side input power , and calculate the primary side load power for each prediction time t in the future; if , calculate the abnormality index , where is the rated primary side input power; if , check the load change rate, and the calculation method is , where is the prediction time detection, is the primary side load power at the last prediction time, is the load change rate; if , it is determined that the primary side power is stable and normal, if , it is determined that there is a sudden fluctuation in the primary side input power, and the abnormality index is calculated; if , the abnormality index is calculated; finally, the obtained abnormality index is standardized, and when the determined abnormality index is greater than the set abnormality index threshold value, the transformer load warning is performed.

[0065] The above merely provides the preferred embodiment of the present application, and is not intended to limit the present application to other forms. Any person skilled in the art can make modifications and variations to the present application without departing from the spirit and scope of the present application. Therefore, the scope of the present application shall be defined and protected by the appended claims.

Claims

1. A transformer load remote monitoring and detection system, characterized by, It comprises a data acquisition module, an edge computing module, a load adjustment module, a load prediction module, and a load analysis module. The data acquisition module is used to collect the secondary-side load power, current, vibration signal data, and winding temperature data of the transformer within a monitoring period. The edge computing module comprises a data processing unit and a rapid response unit, and is used to process obvious load warning conditions. The load adjustment module is used to calculate the primary-side load power according to the secondary-side load power, and adjust the primary-side load power according to the winding temperature data and vibration signal data of the transformer, so that the adjusted primary-side load power is obtained. The load prediction module collects the adjusted primary-side load power in a period of time and other data in the period of time, integrates the data into training data, constructs a load prediction model, inputs the training data to train the load prediction model, inputs newly collected data to make a prediction, and obtains the primary-side load power in a future period of time. The load analysis module is used to analyze the transformer load condition according to the predicted primary-side load power in the future period of time, and obtain the transformer load condition in the future period of time. The data processing unit in the edge computing module is used to perform denoising processing on the collected data to obtain noise-removed data, and the specific implementation is as follows: First, initialize the search space, define the value range of the modal number K and the penalty factor For each group , perform VMD decomposition to obtain K intrinsic modal components IMFs; computing the dual objective function including energy concentration and information entropy gain , the energy concentration is calculated as , wherein the information entropy gain is calculated as, first calculating the original signal entropy , then calculating the weighted entropy sum after decomposition , , wherein is the energy of the kth IMF, is the total energy of all IMFs, is the information entropy of the kth IMF, and then the information entropy gain is calculated; Genetic algorithm is used to search the optimal parameters of the model, and the optimal parameters are found by using the double fitness function of genetic algorithm. and as the double fitness function, the optimal parameters are found by iteration. ; Then, the VMD decomposed IMFs are calculated to obtain a screening index, including permutation entropy PE, trend consistency TC and frequency domain correlation FC, and the three are standardized and weighted to calculate the screening index, and a preset threshold is used for judgment, and only the IMF component with a screening index greater than or equal to the threshold is retained. Then, the VMD decomposed IMFs are calculated to obtain a screening index, including permutation entropy PE, trend consistency TC and frequency domain correlation FC, and the three are standardized and weighted to calculate the screening index, and a preset threshold is used for judgment, and only the IMF component with a screening index greater than or equal to the threshold is retained. For the remaining IMF components, secondary purification is carried out using wavelet threshold denoising to obtain For the noise IMF, the residual distribution of the noise IMF and the remaining IMF components is calculated And the residual error is generated using the residual error rule learned by the adversarial network ; Finally, the reconstruction is performed to obtain the de-noised data as .

2. The transformer load remote monitoring and detection system of claim 1, wherein, After the denoising processing is implemented, there is a verification operation, and the specific implementation is as follows: First, extract historical contemporaneous data, use the same process to perform denoising, and obtain historical denoised data. The dynamic similarity of the current de-noising data and the historical de-noising data is calculated in a manner that wherein is a current de-noised data sequence, is a historical de-noised data sequence, L is a data length, and if the dynamic similarity is less than a set threshold value, the screening threshold value is reversely adjusted, and the de-noising is re-executed until the similarity requirement is met.

3. The transformer load remote monitoring and detection system of claim 1, wherein, The rapid response unit in the edge computing module processes obvious load warning conditions, and if the real-time secondary-side load power exceeds 1.3 times of the device set value, the rapid response is directly performed, and the warning is triggered.

4. The transformer load remote monitoring and detection system of claim 1, wherein, The load adjustment module is used to calculate the primary-side load power according to the secondary-side load power. First, the harmonic additional loss power is calculated wherein, is the total power of the secondary side load, is the harmonic distortion rate of the load current, is the load type correction coefficient; The primary input power is wherein, is the primary input power, is the transformer efficiency.

5. The transformer load remote monitoring and detection system of claim 4, wherein, The load current harmonic distortion rate The distortion degree of the secondary side load current waveform of the transformer deviating from the sine wave is reflected, and the calculation method is the ratio of the effective value of the current harmonic component to the effective value of the fundamental component.

6. The transformer load remote monitoring and detection system of claim 4, wherein, The load adjustment module further comprises a physical correction unit, which is used to combine the winding temperature and vibration signal characteristics of the transformer to perform secondary correction on the basic primary-side power, and the specific implementation is as follows: Firstly, the correlation formula between dynamic characteristics and additional loss of transformer is constructed, and the relationship between vibration and loss under different loads is fitted through experiment, and the formula is: wherein, is the effective value of vibration acceleration, the vibration signal characteristics are collected synchronously with the power calculation period, and is obtained through root mean square calculation, represents the rated value of vibration acceleration, is the vibration main frequency offset, the vibration signal characteristics are collected synchronously with the power calculation period, and is obtained through fast Fourier transform, represents the main frequency under rated load; Then, the temperature factor is combined again to make a correction, and a temperature correction coefficient is obtained, and the calculation method is: wherein, is a temperature threshold value; The primary input power base value is corrected by a correction coefficient, and the formula is: .

7. The transformer load remote monitoring and detection system of claim 4, wherein, The load adjustment module further comprises a load characteristic self-learning unit, which is configured to identify a load state according to historical period data and iteratively optimize The actual load characteristics are adapted by implementing the following steps: First, the load power fluctuation variance is calculated by counting the fluctuation of the secondary side load power in the historical period , and then the average harmonic distortion rate is obtained by averaging the THD data collected in the historical period . Standardization is performed on the load power fluctuation variance and the periodic average harmonic distortion rate to obtain a standardized and , a load characteristic deviation degree is constructed The calculation method is is compared with a set load characteristic deviation degree threshold if less than or equal to then no change is made, if greater than then a optimization, is triggered.

8. The transformer load remote monitoring and detection system of claim 1, wherein, The load prediction module constructs a load prediction model by using a bidirectional long short-term memory network, and increases residual interaction between the attention layer and the fully connected layer, and the specific implementation is as follows: short-term key features first received from the attention layer output ; calling long-term sequences of training data, computing long-term trend components by sliding windows ; computing a residual of short-term features from long-term trends ; By the gating unit wherein is a weight matrix, G is a gating value controlling the compensation ratio of the residual, the output is passed to a fully connected layer.

9. The transformer load remote monitoring and detection system of claim 1, wherein, The load analysis module is used to analyze the transformer load condition according to the predicted primary-side load power in the future period of time, and obtain the transformer load condition in the future period of time. First, set the safe load interval of the primary input power , for each future predicted time t primary load power If , the abnormality index is calculated wherein is the rated primary input power; If , the load change rate is checked, and the calculation method is , wherein is the detection at the prediction time, is the original side load power at the last prediction time, is the load change rate; wherein, if , it is determined that the original side power is stable and normal, if , it is determined that there is a sudden fluctuation in the original side input power, and the abnormal index is calculated; If , calculate an anomaly index ; The abnormality index obtained The transformer load is warned when the abnormality index is greater than a set abnormality index threshold after normalization.

Images