A transformer fault monitoring method and system based on artificial intelligence

Abstract

The present invention relates to the field of data processing technology, and in particular to a transformer fault monitoring method and system based on artificial intelligence. The method includes the steps of: collecting an input current data set and an output current data set, obtaining a characteristic deviation value of each data in the input current data set, and obtaining a first oscillation amplitude of the input current data set based on the characteristic deviation value; obtaining a second oscillation amplitude of the input current data set, and obtaining a differential weight according to the first oscillation amplitude and the second oscillation amplitude; obtaining a load change of the power transformer according to the consistency of the change of the input current and the output current of the power transformer, and correcting the differential weight according to the load change to obtain a corrected differential weight; inputting the corrected differential weight into the DE algorithm to optimize the convolution kernel parameters of the CNN network, and then monitoring the fault of the power transformer. The present invention improves the accuracy of transformer fault monitoring.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a transformer fault monitoring method and system based on artificial intelligence. Background Art

[0002] The power transformer is the most critical static electrical equipment in the operation of the power grid. Whether it can operate smoothly is related to the safety and power supply reliability of the entire power system. Once the power transformer fails and stops operating, it will not only cause its own destruction, but also interrupt the power supply, causing great economic losses to users, and even cause major accidents. Therefore, efficient and accurate monitoring of power transformers is of great significance to the safe operation of the power system. The CNN model can effectively learn and extract features from complex raw data, such as extracting fault features from current, voltage and other data collected by transformer sensors. The optimization algorithm can be combined with the CNN model to realize autonomous learning and training of convolution kernel parameters to achieve transformer fault monitoring. Since the DE algorithm (differential evolution algorithm) is a kind of heuristic parallel search technology, the advantage of this algorithm is its strong optimization ability, which is suitable for optimizing hyperparameters in the CNN model. Therefore, the DE algorithm can be combined to optimize the parameters of the convolution kernel in the CNN model, and then the CNN model can be used to realize transformer fault monitoring.

[0003] The patent application document with the publication number CN109002892A is an implementation method for an improved DE-GWO algorithm. The DE-GWO algorithm is formed by introducing the differential evolution strategy (DE algorithm) and the GWO algorithm, and the three operations of mutation, crossover and selection of differential evolution are used to iteratively update the position and objective function of the gray wolf individuals; a nonlinear function convergence factor strategy is introduced to improve the optimization performance of the DE-GWO algorithm and obtain the global optimal solution.

[0004] The differential weight of the DE algorithm plays a key role in the above method. Its role is mainly reflected in controlling the degree and influence of the differential mutation operation. The differential weight determines how to generate new candidate solutions through mutation operations. Larger values ​​lead to larger mutations, resulting in new solutions further out in the search space; smaller The value produces smaller variation, which helps to maintain the stability and convergence of the population. It is known that transformer fault monitoring is a complex real-time task, and the fault type and working conditions of the transformer may vary with time and environment, resulting in different characteristics of the current data. The selection of differential weights will affect the optimization effect of the DE algorithm, and then affect the effect of parameter optimization of the convolution kernel in the CNN model, which ultimately affects the accuracy of the transformer monitoring results. Summary of the invention

[0005] In order to solve the technical problem that the selection of differential weights will affect the optimization effect of the DE algorithm, thereby affecting the effect of parameter optimization of the convolution kernel in the CNN model, and ultimately leading to inaccurate transformer monitoring results, the present invention provides a transformer fault monitoring method and system based on artificial intelligence.

[0006] In a first aspect, the present invention provides a transformer fault monitoring method based on artificial intelligence, which adopts the following technical solution:

[0007] A transformer fault monitoring method based on artificial intelligence comprises the following steps:

[0008] Collecting an input current data set and an output current data set; obtaining a characteristic value deviation of each data in the input current data set, wherein the characteristic value deviation is the deviation between the value of each data in the input current data set and the mean of all data; taking the difference between the maximum value and the minimum value of the characteristic value deviation of all data in the input current data set as the first oscillation amplitude of the input current data set; obtaining a Fourier curve of the input current data set; obtaining a second oscillation amplitude of the input current data set, wherein the second oscillation amplitude is positively correlated with the kurtosis and skewness of the Fourier curve of the input current data set;

[0009] Get differential weights ; B a second oscillation amplitude representing an input current data set; A represents the first oscillation amplitude of the input current data set; Represents the preset first weight; represents a preset second weight; exp() represents an exponential function with a natural constant as the base; the load change of the power transformer is obtained, and the load change is negatively correlated with the difference between the corresponding data in the input current data set and the output current data set; the difference between the differential weight and the load change of the power transformer is used as a corrected differential weight; according to the corrected differential weight, the fault of the power transformer is monitored.

[0010] The innovation of the present invention lies in adaptively acquiring differential weights according to the oscillation amplitude of the data in the input current data set and the load change of the power transformer, applying the obtained differential weights to the DE algorithm, optimizing the convolution kernel parameters in the CNN model, and carrying the optimized CNN model to the transformer fault monitoring system for fault monitoring, thereby improving the optimization effect of the DE algorithm, and further improving the effect of parameter optimization of the convolution kernel in the CNN model, making the transformer monitoring results more accurate.

[0011] Preferably, the step of obtaining the characteristic value deviation of each data in the input current data set includes:

[0012] ;

[0013] In the formula, Represents the characteristic deviation value of the i-th data in the input current data set; Represents the value of the i-th data in the input current data set; represents the mean of all data in the input current data set; || represents the absolute value symbol.

[0014] The characteristic value deviation of each data in the input current data set can reflect the oscillation of the numerical value of each data, so as to facilitate the subsequent acquisition of the first oscillation amplitude of the input current data set according to the characteristic value deviation.

[0015] Preferably, the acquiring of the Fourier curve of the input current data set includes:

[0016] Perform Fourier transform on the input current data set and obtain the Fourier curve of the input current data set.

[0017] Preferably, the step of obtaining the second oscillation amplitude of the input current data set includes:

[0018] ;

[0019] In the formula, B a second oscillation amplitude representing an input current data set; K The kurtosis of the Fourier curve representing the input current data set; S The skewness of the Fourier curve representing the input current data set.

[0020] This facilitates obtaining the differential weight according to the second oscillation amplitude of the input current data set.

[0021] Preferably, obtaining the load change of the power transformer includes:

[0022] ;

[0023] In the formula, Z Represents the load change of the power transformer; Represents the value of the i-th data in the input current data set; Represents the value of the data in the output current data set corresponding to the sampling time of the i-th data in the input current data set; N Represents the number of data in the input current data set; || represents the absolute value symbol; D represents the Pearson correlation coefficient between the input current dataset and the input current dataset; norm () represents the normalization function.

[0024] The load variation of the power transformer is taken into account, so that the differential weight can be corrected according to the load variation later.

[0025] Preferably, monitoring the fault of the power transformer according to the modified differential weight comprises:

[0026] The modified differential weights are input into the DE algorithm to optimize the convolution kernel parameters in the CNN model. The optimized CNN model is deployed into the power transformer fault monitoring system. The collected input current data set is input into the optimized CNN model to perform fault judgment on the power transformer.

[0027] The optimization effect of the DE algorithm is improved, which in turn improves the effect of parameter optimization of the convolution kernel in the CNN model, making the transformer monitoring results more accurate.

[0028] Preferably, the collecting of the input current data set and the output current data set includes:

[0029] Obtain any power transformer that has been used for one year, obtain input current data and output current data of the power transformer for nearly one year, and obtain an input current data set and an output current data set.

[0030] In a second aspect, the present invention provides a transformer fault monitoring system based on artificial intelligence, which adopts the following technical solution:

[0031] A transformer fault monitoring system based on artificial intelligence includes: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned transformer fault monitoring method based on artificial intelligence is implemented.

[0032] By adopting the above technical solution, the above-mentioned artificial intelligence-based transformer fault monitoring method is generated into a computer program and stored in a memory to be loaded and executed by a processor, so that a terminal device is made according to the memory and the processor for easy use.

[0033] The present invention has the following technical effects: the present invention obtains the oscillation amplitude of the input current data set according to the fluctuation of the data in the input current data set; then obtains the load change of the power transformer according to the change consistency of the corresponding data of the input current data set and the output current data set; finally, according to the oscillation amplitude of the input current data set and the load change of the power transformer, the differential weight is adaptively obtained, the obtained differential weight is applied to the DE algorithm, the convolution kernel parameters in the CNN model are optimized, and the optimized CNN model is carried on the transformer fault monitoring system for fault monitoring, which improves the optimization effect of the DE algorithm, and then improves the effect of optimizing the parameters of the convolution kernel in the CNN model, so that the transformer monitoring result is more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] By reading the detailed description below with reference to the accompanying drawings, the above and other purposes, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-restrictive manner, and the same or corresponding reference numerals represent the same or corresponding parts.

[0035] Figure 1 This is a flow chart of a method for transformer fault monitoring based on artificial intelligence in an embodiment of the present invention. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0037] It should be understood that when the terms "first", "second", etc. are used in the claims, descriptions, and drawings of the present invention, they are only used to distinguish different objects, rather than to describe a specific order. The terms "include" and "comprise" used in the description and claims of the present invention indicate the presence of the described features, wholes, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their collections.

[0038] The embodiment of the present invention discloses a transformer fault monitoring method based on artificial intelligence, referring to Figure 1 , comprising steps S1 to S3:

[0039] S1: Collect input current data set and output current data set.

[0040] In an embodiment of the present invention, any power transformer that has been used for one month is obtained and recorded as a target power transformer. Input current data and output current data of the target power transformer in the past month are obtained to obtain an input current data set and an output current data set.

[0041] S2: Obtain a characteristic deviation value of each data in the input current data set, and based on the characteristic deviation value, obtain a first oscillation amplitude of the input current data set; obtain a second oscillation amplitude of the input current data set, and obtain a differential weight according to the first oscillation amplitude and the second oscillation amplitude; obtain a load change of the power transformer according to the consistency of changes in the input current and output current of the power transformer, correct the differential weight according to the load change, and obtain a corrected differential weight.

[0042] Step S2 includes steps S20-S21, which are specifically as follows:

[0043] S20: Obtain a characteristic deviation value of each data in the input current data set, and based on the characteristic deviation value, obtain a first oscillation amplitude of the input current data set; obtain a second oscillation amplitude of the input current data set, and obtain a differential weight according to the first oscillation amplitude and the second oscillation amplitude.

[0044] It should be noted that it is known that changes in the working state and environmental conditions of the power transformer will cause changes in the characteristic distribution of the current data. Therefore, adaptively adjusting the differential weight can make the DE algorithm more flexible to adapt to these changes, so that the final monitoring model is more suitable for the fault monitoring of the power transformer. It is known that the internal fault of the power transformer often causes the current waveform to oscillate rapidly or increase the frequency component. By monitoring the current oscillation amplitude, these abnormal characteristics can be effectively captured. These abnormal characteristics are often indicators of the fault state of the transformer. Therefore, it is necessary to adjust the differential weight to respond to these oscillation characteristics to help the system detect potential internal faults of the transformer more sensitively. In addition, because the ratio and waveform of the input and output currents of the power transformer will change under different load conditions, by analyzing the load changes of the power transformer, it is possible to understand the difference in the current waveform characteristics under normal operation and abnormal state of the transformer. Therefore, it is necessary to adjust the differential weight to cope with the load changes of the power transformer, which helps the system maintain stable monitoring capabilities under dynamic working conditions. In summary, the differential weight can be obtained based on the oscillation amplitude of the collected input current data set and the load changes of the power transformer;

[0045] The oscillation amplitude of the data in the input current data set here and the load change of the power transformer are not contradictory. This is because the current oscillation change is usually caused by specific faults or abnormal conditions inside the transformer, such as short circuits between windings, ground faults or insulation loss. These problems cause the current waveform to oscillate or fluctuate rapidly and irregularly in time. The load change of the power transformer refers to the change in the load (i.e., the input and output current) borne by the transformer. This change may be the result of normal operation, such as fluctuations in grid load or changes in user demand.

[0046] It should be further explained that, since there are certain changes in the data in the input current data set collected by the power transformer, in order to comprehensively evaluate the oscillation amplitude of the input current data set, it is necessary to combine the frequency domain and time domain for evaluation. Among them, the difference between the value of each data in the input current data set and the mean of all data in the time domain can simply and intuitively measure the fluctuation of the data; in the frequency domain, the kurtosis and skewness of the Fourier curve of the input current data set can be obtained through Fourier transform, which can describe the sharpness and asymmetry of the data in the input current data set, and then reflect the oscillation characteristics of the input current data set.

[0047] In an embodiment of the present invention, the characteristic deviation value of each data in the input current data set is obtained:

[0048] ;

[0049] In the formula, Represents the characteristic deviation value of the i-th data in the input current data set; Represents the value of the i-th data in the input current data set; represents the mean of all data in the input current data set; || represents the absolute value symbol.

[0050] The difference between the maximum value and the minimum value of the characteristic deviation values ​​of all data in the input current data set is used as the first oscillation amplitude of the input current data set;

[0051] The input current data set is subjected to Fourier transform, and the Fourier curve of the input current data set and the skewness and kurtosis of the Fourier curve of the input current data set are obtained; it should be noted that Fourier transform is a prior art, and in the embodiments of the present invention, it will not be described in detail.

[0052] Get the second oscillation amplitude of the input current data set:

[0053] ;

[0054] In the formula, B a second oscillation amplitude representing an input current data set; K The kurtosis of the Fourier curve representing the input current data set; S The skewness of the Fourier curve representing the input current data set. It should be noted that the Fourier transform algorithm, skewness and kurtosis are well-known technologies and will not be described in detail in the embodiments of the present invention.

[0055] It should be noted that when the oscillation amplitude of the input current data set is large, it usually means that there may be obvious abnormalities or faults inside the transformer. Therefore, in order to enhance the sensitivity of the DE algorithm to such abnormalities, it is possible to consider increasing the differential weight. A larger differential weight can make the variation more drastic, making it easier to discover and respond to the fault characteristics represented by the current data oscillation.

[0056] Get the differential weights:

[0057] ;

[0058] In the formula, F represents the differential weight; B a second oscillation amplitude representing an input current data set; A represents the first oscillation amplitude of the input current data set; Represents the preset first weight; represents a preset second weight; exp() represents an exponential function with a natural constant as the base; in the embodiment of the present invention, the preset first weight , preset the second weight In other embodiments, the implementer may preset the values ​​of the first weight and the second weight according to the specific implementation situation; Represents the oscillation amplitude of the input current data set. The larger the value, The larger the value of The value range is [0,2].

[0059] S21: Obtain the load change of the power transformer according to the consistency of the change of the input current and the output current of the power transformer, and correct the differential weight according to the load change to obtain the corrected differential weight.

[0060] It should be noted that load changes are inevitable during the operation of power transformers. Therefore, the differential weights obtained based on the oscillation amplitude of the data in the input current data set may not be able to fully adapt to these changes, and the obtained differential weights need to be corrected according to the load changes of the power transformer. It is known that larger load changes will increase the range of variation of current data, which may lead to false alarms or omissions in fault monitoring. At this time, the differential weights need to be reduced to reduce their interference with fault detection. In addition, because the ratio of input and output currents of power transformers will change under different load conditions, when the load increases, the input and output currents of the power transformer will increase proportionally to meet the power requirements. Therefore, by analyzing the difference between the corresponding data in the input current data set and the output current data set at the same time, the consistency of the input current and output current changes of the power transformer is reflected. If the changes in the input current and output current of the power transformer are more consistent, the load change of the power transformer is greater.

[0061] In an embodiment of the present invention, the load change of the power transformer is obtained:

[0062] ;

[0063] In the formula, Z Represents the load change of the power transformer; Represents the value of the i-th data in the input current data set; Represents the value of the data in the output current data set corresponding to the sampling time of the i-th data in the input current data set; N represents the number of data in the input current data set; it should be noted that the number of data in the input current data set and the output current data set is the same; || represents the absolute value symbol; D represents the Pearson correlation coefficient between the input current dataset and the input current dataset; norm () represents the normalization function; Represents the cumulative sum of the absolute values ​​of the differences between all corresponding data in the input current data set and the output current data set. The smaller the value, the greater the consistency of the input current and output current changes of the power transformer, which means the greater the load change of the power transformer; It reflects the correlation between the input current data set and the output current data set. The larger the value, the The more credible the value.

[0064] Get the modified differential weights:

[0065] ;

[0066] In the formula, R represents modified differential weight; F represents the differential weight; Z represents the load change of the power transformer; e is a natural constant; The value range of Z is constrained to be between [0,1].

[0067] S3: Monitor the fault of the power transformer according to the modified differential weight.

[0068] It should be noted that after obtaining the modified differential weight, it needs to be applied to the DE algorithm and combined with the CNN model to monitor the faults of the power transformer.

[0069] In the embodiment of the present invention, a number of power transformers that have been in use for a full month are obtained, and input current data of each power transformer in the past month is obtained to form a historical current data sequence.

[0070] Construct a fault detection network, which is a CNN network, and input the modified differential weight into the DE algorithm to optimize the convolution kernel parameters in the CNN network to obtain an optimized CNN network, where the loss function of the optimized CNN network is the mean square error. Manually label the historical current data sequence, where the historical current data sequence collected when there is a fault is labeled as 1, and the historical current data sequence collected when there is no fault is labeled as 0, to obtain a labeled historical current data sequence. All labeled historical current data sequences constitute a data set.

[0071] The fault detection network is trained using the labeled historical current data sequence to obtain a trained fault detection network. The currently collected input current data set is input into the trained fault detection network to obtain a fault detection result. When the fault detection result is that a fault exists, an early warning is issued.

[0072] An embodiment of the present invention further discloses an artificial intelligence-based transformer fault monitoring system, comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, an artificial intelligence-based transformer fault monitoring method according to the present invention is implemented.

[0073] The above system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface, and their configuration and functions are known in the art, so they will not be described in detail here.

[0074] In the present invention, the aforementioned memory may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium may be any suitable magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory, a dynamic random access memory, a static random access memory, an enhanced dynamic random access memory, a high bandwidth memory, a hybrid storage cube, etc., or any other medium that can be used to store the required information and can be accessed by an application, a module, or both. Any such computer storage medium may be part of the device or accessible or connectable to the device.

[0075] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

[0076] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A transformer fault monitoring method based on artificial intelligence, characterized in that: Includes steps: collecting an input current data set and an output current data set; Obtaining a characteristic value deviation of each data in the input current data set, wherein the characteristic value deviation is a deviation between a value of each data in the input current data set and a mean value of all data; The difference between the maximum value and the minimum value of the characteristic value deviation of all data in the input current data set is used as the first oscillation amplitude of the input current data set; Obtain the Fourier curve of the input current data set; Acquire a second oscillation amplitude of the input current data set, wherein the second oscillation amplitude is positively correlated with the kurtosis and skewness of the Fourier curve of the input current data set; Get differential weights ; B a second oscillation amplitude representing an input current data set; A represents the first oscillation amplitude of the input current data set; Represents the preset first weight; represents the preset second weight; exp() represents an exponential function with a natural constant as the base; Obtaining a load change of the power transformer, wherein the load change is negatively correlated with a difference between corresponding data in the input current data set and the output current data set; using a difference between the differential weight and the load change of the power transformer as a modified differential weight; The modified differential weights are input into the DE algorithm to optimize the convolution kernel parameters in the CNN model. The optimized CNN model is deployed into the power transformer fault monitoring system. The collected input current data set is input into the optimized CNN model to perform fault judgment on the power transformer.

2. The transformer fault monitoring method based on artificial intelligence according to claim 1 is characterized in that: The step of obtaining the characteristic value deviation of each data in the input current data set includes: ； In the formula, Represents the characteristic deviation value of the i-th data in the input current data set; Represents the value of the i-th data in the input current data set; represents the mean of all data in the input current data set; || represents the absolute value symbol.

3. The transformer fault monitoring method based on artificial intelligence according to claim 1 is characterized in that: The obtaining of the Fourier curve of the input current data set comprises: Perform Fourier transform on the input current data set and obtain the Fourier curve of the input current data set.

4. The transformer fault monitoring method based on artificial intelligence according to claim 1 is characterized in that: The obtaining a second oscillation amplitude of the input current data set comprises: ； In the formula, B a second oscillation amplitude representing an input current data set; K The kurtosis of the Fourier curve representing the input current data set; S The skewness of the Fourier curve representing the input current data set.

5. The transformer fault monitoring method based on artificial intelligence according to claim 1 is characterized in that: The obtaining of the load change of the power transformer comprises: ； In the formula, Z Represents the load change of the power transformer; Represents the value of the i-th data in the input current data set; Represents the value of the data in the output current data set corresponding to the sampling time of the i-th data in the input current data set; N Represents the number of data in the input current data set; || represents the absolute value symbol; D represents the Pearson correlation coefficient between the input current dataset and the input current dataset; norm () represents the normalization function.

6. The transformer fault monitoring method based on artificial intelligence according to claim 1 is characterized in that: The collecting of the input current data set and the output current data set includes: Obtain any power transformer that has been used for one year, obtain input current data and output current data of the power transformer for nearly one year, and obtain an input current data set and an output current data set.

7. A transformer fault monitoring system based on artificial intelligence, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, an artificial intelligence-based transformer fault monitoring method according to any one of claims 1 to 6 is implemented.

Images