Transformer abnormity early warning method and device
By constructing the transformer's thermal power curve and frequency domain feature vector, the modeling process of transformer abnormal identification is simplified, the problems of complex modeling and large data volume in the existing technology are solved, and simpler and more effective abnormal identification is achieved.
Patent Information
- Application Number
- CN202510208867.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-25
- Publication Date
- 2025-06-13
AI Technical Summary
The existing data-driven transformer abnormality recognition method has complex modeling process and large data volume, making it difficult to promote.
By obtaining the ambient temperature, equipment temperature curve, vibration curve and current curve, a thermal power curve is constructed, and frequency domain characteristics are extracted from these curves in segments, frequency domain feature vectors are constructed, and inputted into the state classification model to determine the transformer state.
The data source and sample size requirements are simplified, the model construction process is simpler, suitable for promotion, and effectively identify the abnormal state of the transformer.
Smart Images
Figure CN120145104A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of transformer operation monitoring, and particularly to a transformer anomaly warning method and device. Background Art
[0002] Power transformers are one of the main equipment in power plants and substations. The functions of transformers are various. They can not only step up the voltage to send electrical energy to the power consumption area, but also step down the voltage to various levels of operating voltages to meet the power consumption needs.
[0003] Currently, the main transformer anomaly identification methods are based on physical detection methods, that is, by collecting and obtaining the dielectric loss factor of the transformer during off-line operation to comprehensively, accurately and effectively detect the equipment; another method is a data-driven transformer anomaly identification method, which usually uses big data as characteristic parameters reflecting the performance of the transformer, and identifies the abnormal state of the transformer through data-driven methods.
[0004] Among them, the data-driven anomaly identification method usually uses artificial intelligence algorithms to implement. It requires a large amount of data, and the process of modeling and applying the model is complex, which is not convenient for popularization.
[0005] Based on this, it is necessary to develop a transformer anomaly warning method and device. Summary of the Invention
[0006] Embodiments of the present invention provide a transformer anomaly warning method and device to solve the problem that the modeling process of the data-driven anomaly identification method is relatively complex.
[0007] In a first aspect, embodiments of the present invention provide a transformer anomaly warning method, including:
[0008] Obtain the ambient temperature, the device temperature curve, the vibration curve, and the current curve;
[0009] According to the ambient temperature and the device temperature curve, construct a heat generation power curve, where the heat generation power curve represents the curve of the transformer's heat generation power changing with time;
[0010] Respectively extract a plurality of frequency domain features in segments from the heat generation power curve, the vibration curve, and the current curve, and construct them into a frequency domain feature vector;
[0011] Input the frequency domain feature vector into a state classification model, and determine the state of the transformer according to the output of the state classification model.
[0012] In a possible implementation manner, the constructing a heat generation power curve according to the ambient temperature and the device temperature curve includes:
[0013] Obtain the mass of the transformer, the specific heat capacity of the transformer, the equivalent heat dissipation surface of the transformer, and the thermal resistance of the transformer;
[0014] According to the ambient temperature, the device temperature curve, the heat capacity of the transformer, the mass of the transformer, the specific heat capacity of the transformer, the equivalent heat dissipation surface of the transformer, and the thermal resistance of the transformer, construct a first equation expressing the relationship between the device temperature, the ambient temperature, and the heat generation power;
[0015] Construct a heat generation model according to the first equation;
[0016] Construct a heat generation power curve according to the heat generation model, the ambient temperature, and the device temperature curve.
[0017] In a possible implementation manner, the first equation is:
[0018]
[0019] In the formula, P is the heat generation power, R is the thermal resistance, m is the mass of the transformer, c is the specific heat capacity of the transformer, T is the temperature of the transformer, t is the time variable, S is the effective heat dissipation surface of the transformer, and T E is the ambient temperature.
[0020] In a possible implementation manner, the heat generation model is:
[0021]
[0022] In the formula, P is the heat generation power, R is the thermal resistance, m is the mass of the transformer, c is the specific heat capacity of the transformer, T is the temperature of the transformer, t is the time variable, S is the effective heat dissipation surface of the transformer, and T E is the ambient temperature, τ is the time constant, and e is the natural constant.
[0023] In a possible implementation manner, respectively extract a plurality of frequency domain features from the heat generation power curve, the vibration curve, and the current curve in segments, and construct them into a frequency domain feature vector, including:
[0024] For each of the heat generation power curve, the vibration curve, and the current curve, respectively perform the following steps:
[0025] Obtain the fundamental wave period;
[0026] Starting from a predetermined position of the curve, cut out a curve segment that is several times the fundamental wave period as the curve segment to be processed;
[0027] Extract a plurality of frequency domain features from the curve segment to be processed according to the fundamental wave period, and add the extracted plurality of frequency domain features to the frequency domain queue;
[0028] If the traversal of the curve is not completed, perform an offset on the predetermined position and jump to start from the predetermined position of the slave curve, and cut a curve segment several times the fundamental period as the curve segment to be processed;
[0029] Otherwise, use multiple data in the frequency domain queue as multiple frequency domain feature elements and add them to the frequency domain feature vector.
[0030] In a possible implementation manner, the extracting multiple frequency domain features from the curve segment to be processed according to the fundamental period includes:
[0031] Extract multiple frequency domain features from the curve segment to be processed according to the second formula and the fundamental period, where the second formula is:
[0032]
[0033] In the formula, CF(k) is the k-th frequency domain eigenvalue, cN is the total number of data obtained after discretization of the curve segment to be processed, CDW(cn) is the cn-th data obtained after discretization of the curve segment to be processed, e is the natural constant, j is the imaginary unit, ω 0 is the frequency corresponding to the fundamental period, PN is the number of data obtained after discretization of the curve segment to be processed within the fundamental period duration, and π is the pi.
[0034] In a possible implementation manner, the construction process of the state classification model includes:
[0035] Obtain multiple frequency domain feature vector samples and multiple coefficient arrays, where each frequency domain feature vector sample corresponds to a transformer state;
[0036] Substitute the multiple coefficient arrays into the classification basic model respectively to obtain multiple intermediate models;
[0037] Substitute the multiple frequency domain feature vector samples into each intermediate model in turn, and construct multiple model outputs into an output array, so as to obtain multiple output arrays;
[0038] Determine the fitness of the intermediate model according to each output array and the transformer state corresponding to the multiple frequency domain feature vector samples, so as to obtain multiple fitnesses;
[0039] If the multiple fitnesses are all less than the fitness threshold, select three coefficient arrays with the best fitness from the multiple coefficient arrays as three target arrays, adjust the other coefficient arrays according to the three target arrays, and jump to the step of substituting the multiple coefficient arrays into the classification basic model respectively to obtain multiple intermediate models;
[0040] Otherwise, substitute the coefficient array with the best fitness into the classification basic model to obtain the state classification model.
[0041] In a possible implementation manner, the classification basic model is:
[0042]
[0043] In the formula, class is the output of the classification model, w an is the pre - coefficient, CFS(an) is the an - th element of the frequency - domain feature vector, and aN is the total number of elements of the frequency - domain feature vector.
[0044] In a possible implementation manner, determining the fitness of the intermediate model according to each output array and the transformer states corresponding to the multiple frequency - domain feature vector samples includes:
[0045] For each output array, perform the following steps respectively:
[0046] Arrange the data in the output array in ascending order of numerical value to obtain a data queue;
[0047] Set the transformer state labels for each data in the data queue according to the multiple frequency - domain feature vector samples;
[0048] Determine the fitness of the intermediate model according to the third formula and the transformer state labels set for each data in the data queue, where the third formula is:
[0049]
[0050] In the formula, Fitness is the fitness of the intermediate model, QTT sn is the number of frequency - domain feature vector samples of the sn - th transformer state, Rank sn_max is the highest rank of the data in the data queue with the transformer state label being the sn - th transformer state, Rank sn_min is the lowest rank of the data in the data queue with the transformer state label being the sn - th transformer state, and sN is the number of transformer state labels.
[0051] In a second aspect, an embodiment of the present invention provides a transformer abnormal warning device, including:
[0052] A monitoring curve acquisition module, configured to acquire the environmental temperature, equipment temperature curve, vibration curve, and current curve;
[0053] A heat generation curve construction module, configured to construct a heat generation power curve according to the environmental temperature and the equipment temperature curve, where the heat generation power curve represents the curve of the transformer heat generation power changing with time;
[0054] A frequency-domain feature vector construction module, configured to respectively and segmentally extract a plurality of frequency-domain features from the heat generation power curve, the vibration curve, and the current curve, and construct them into a frequency-domain feature vector;
[0055] And,
[0056] A transformer state determination module, configured to input the frequency-domain feature vector into a state classification model, and determine the state of the transformer according to the output of the state classification model.
[0057] Advantageous effects of the embodiments of the present invention compared with the prior art:
[0058] The present invention provides a transformer abnormal warning method and device. First, it obtains the environmental temperature, the equipment temperature curve, the vibration curve, and the current curve; then constructs a heat generation power curve according to the environmental temperature and the equipment temperature curve, where the heat generation power curve represents the curve of the heat generation power of the transformer changing with time; then respectively and segmentally extracts a plurality of frequency-domain features from the heat generation power curve, the vibration curve, and the current curve, and constructs them into a frequency-domain feature vector; finally, inputs the frequency-domain feature vector into a state classification model, and determines the state of the transformer according to the output of the state classification model. The abnormal recognition method adopted in the embodiments of the present invention requires less data sources. The classification model is constructed based on the frequency features extracted from the curves, requires fewer sample sizes than intelligent algorithms, the model and the model construction process are simpler, and it is suitable for popularization. BRIEF DESCRIPTION OF THE DRAWINGS
[0059] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.
[0060] Figure 1 It is a flowchart of the implementation of the transformer abnormal warning method provided by the embodiments of the present invention;
[0061] Figure 2 It is a schematic diagram of the principle of the fitness calculation process of the intermediate model provided by the embodiments of the present invention;
[0062] Figure 3 It is a schematic structural diagram of the transformer abnormal warning device provided by the embodiments of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0063] In the following description, specific details such as specific system structures and technologies are presented for the purpose of illustration rather than limitation, so as to thoroughly understand the embodiments of the present invention. However, those skilled in the art should clearly understand that the present invention can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid unnecessary details from interfering with the description of the present invention.
[0064] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will be described through specific embodiments in conjunction with the accompanying drawings.
[0065] Figure 1 The implementation flowchart of the transformer abnormal warning method provided for the embodiments of the present invention is described in detail as follows:
[0066] Step 101, obtain the ambient temperature, the device temperature curve, the vibration curve, and the current curve.
[0067] Step 102, construct a heat generation power curve according to the ambient temperature and the device temperature curve, where the heat generation power curve represents the curve of the heat generation power of the transformer changing with time.
[0068] In some embodiments, the constructing a heat generation power curve according to the ambient temperature and the device temperature curve includes:
[0069] Obtain the mass of the transformer, the specific heat capacity of the transformer, the equivalent heat dissipation surface of the transformer, and the thermal resistance of the transformer;
[0070] According to the ambient temperature, the device temperature curve, the heat capacity of the transformer, the mass of the transformer, the specific heat capacity of the transformer, the equivalent heat dissipation surface of the transformer, and the thermal resistance of the transformer, construct a first equation expressing the relationship between the device temperature, the ambient temperature, and the heat generation power;
[0071] According to the first equation, construct a heat generation model;
[0072] According to the heat generation model, the ambient temperature, and the device temperature curve, construct a heat generation power curve.
[0073] In some embodiments, the first equation is:
[0074]
[0075] In the formula, P is the heat generation power, R is the thermal resistance, m is the mass of the transformer, c is the specific heat capacity of the transformer, T is the temperature of the transformer, t is the time variable, S is the effective heat dissipation surface of the transformer, and T E is the ambient temperature.
[0076] In some embodiments, the heat generation model is as follows:
[0077]
[0078] In the formula, P is the heat generation power, R is the thermal resistance, m is the mass of the transformer, c is the specific heat capacity of the transformer, T is the temperature of the transformer, t is the time variable, S is the effective heat dissipation surface of the transformer, T E is the ambient temperature, τ is the time constant, and e is the natural constant.
[0079] Exemplarily, as described above, currently for the analysis of transformer anomalies, one is through physical detection methods, which usually do not easily detect potential hidden dangers and anomalies. The other is through data-driven methods to analyze the state of the transformer, usually using artificial intelligence algorithms, such as artificial neural network algorithms for analysis. This data-driven anomaly analysis method usually requires a complex modeling process and a large amount of samples, thus limiting its promotion. The embodiments of the present invention extract features related to the state of the transformer from aspects such as ambient temperature, equipment temperature curve, vibration curve, and current curve, and evaluate the state of the transformer through a state analysis model constructed by feature analysis.
[0080] To achieve the above object, in terms of heat generation, which is an important index for evaluating transformers, the present invention attempts to restore the transformer heat generation curve (heat generation power curve) from aspects of ambient temperature and equipment temperature curve. For this purpose, the embodiments of the present invention obtain the mass of the transformer, the specific heat capacity of the transformer, the equivalent heat dissipation surface of the transformer, and the thermal resistance of the transformer, and construct a first equation describing the heat generation power and equipment temperature, ambient temperature based on the above parameters:
[0081]
[0082] In the formula, P is the heat generation power, R is the thermal resistance, m is the mass of the transformer, c is the specific heat capacity of the transformer, T is the temperature of the transformer, t is the time variable, S is the effective heat dissipation surface of the transformer, T E is the ambient temperature.
[0083] The above equation is a first-order differential equation. After transformation, the heat generation model is obtained:
[0084]
[0085] In the formula, P is the heat generation power, R is the thermal resistance, m is the mass of the transformer, c is the specific heat capacity of the transformer, T is the temperature of the transformer, t is the time variable, S is the effective heat dissipation surface of the transformer, T E is the ambient temperature, τ is the time constant, and e is the natural constant.
[0086] Input the data queues obtained by discretizing the ambient temperature and the device temperature curve into the heat generation model, and the heat generation power data queue is obtained. By fitting these heat generation power data queues, the heat generation power curve can be obtained.
[0087] Step 103: Extract multiple frequency domain features from the heat generation power curve, the vibration curve, and the current curve respectively in segments, and construct them into a frequency domain feature vector.
[0088] In some embodiments, the extracting multiple frequency domain features from the heat generation power curve, the vibration curve, and the current curve respectively in segments and constructing them into a frequency domain feature vector includes:
[0089] For each of the heat generation power curve, the vibration curve, and the current curve, perform the following steps respectively:
[0090] Obtain the fundamental period;
[0091] Starting from a predetermined position of the curve, cut out a curve segment that is several times the fundamental period as the curve segment to be processed;
[0092] Extract multiple frequency domain features from the curve segment to be processed according to the fundamental period, and add the extracted multiple frequency domain features to the frequency domain queue;
[0093] If the traversal of the curve is not completed, offset the predetermined position and jump to the step of starting from the predetermined position of the curve, cutting out a curve segment that is several times the fundamental period as the curve segment to be processed;
[0094] Otherwise, take the multiple data in the frequency domain queue as multiple frequency domain feature elements and add them to the frequency domain feature vector.
[0095] In some embodiments, the extracting multiple frequency domain features from the curve segment to be processed according to the fundamental period includes:
[0096] Extract multiple frequency domain features from the curve segment to be processed according to the second formula and the fundamental period, where the second formula is:
[0097]
[0098] In the formula, CF(k) is the kth frequency domain eigenvalue, cN is the total number of data obtained after discretizing the curve segment to be processed, CDW(cn) is the cnth data obtained after discretizing the curve segment to be processed, e is the natural constant, j is the imaginary unit, ω 0 is the frequency corresponding to the fundamental period, PN is the number of data obtained after discretizing the curve segment to be processed within the fundamental period duration, and π is the pi.
[0099] Exemplarily, in order to facilitate the extraction of information related to the transformer state contained in the curves and simplify the data calculation amount and the complexity of the model, the embodiments of the present invention extract the frequency domain features segment by segment for each of the heat generation power curve, the vibration curve, and the current curve, and construct the extracted frequency domain features into a frequency domain feature vector. Based on this frequency domain vector, the state of the transformer can be analyzed more conveniently.
[0100] For each curve, a curve segment is taken from a predetermined position, and multiple frequency domain features are extracted using the second formula:
[0101]
[0102] In the formula, CF(k) is the k-th frequency domain feature value, cN is the total number of data obtained after discretization of the curve segment to be processed, CDW(cn) is the cn-th data obtained after discretization of the curve segment to be processed, e is the natural constant, j is the imaginary unit, ω 0 is the frequency corresponding to the fundamental wave period, PN is the number of data obtained after discretization of the curve segment to be processed within the fundamental wave period duration, and π is the pi.
[0103] After the frequency domain features are extracted, the extraction position is offset (the offset length may be less than the curve segment length). After the offset, a curve segment is taken from the curve again, and the above steps of frequency domain feature extraction are performed until the curve is traversed. Then, the extracted frequency domain features are constructed into a frequency domain queue; after the frequency domain queue is extracted for each curve, the data in the multiple frequency domain queues can be extracted and arranged in a preset order to obtain a frequency domain feature vector.
[0104] Step 104: Input the frequency domain feature vector into the state classification model, and determine the state of the transformer according to the output of the state classification model.
[0105] In some embodiments, the construction process of the state classification model includes:
[0106] Obtain multiple frequency domain feature vector samples and multiple coefficient arrays, where each frequency domain feature vector sample corresponds to a transformer state;
[0107] Substitute the multiple coefficient arrays into the classification basic model respectively to obtain multiple intermediate models;
[0108] Substitute the multiple frequency domain feature vector samples into each intermediate model in turn, and construct the obtained multiple model outputs into an output array, so as to obtain multiple output arrays;
[0109] Determine the fitness of the intermediate model according to each output array and the transformer state corresponding to the multiple frequency domain feature vector samples, so as to obtain multiple fitnesses;
[0110] If all of the multiple fitness values are less than the fitness threshold, select the coefficient arrays with the three best fitness values from the multiple coefficient arrays as the three target arrays, adjust the other coefficient arrays according to the three target arrays, and jump to the step of substituting the multiple coefficient arrays into the classification basic model respectively to obtain multiple intermediate models;
[0111] Otherwise, substitute the coefficient array with the best fitness value into the classification basic model to obtain the state classification model.
[0112] In some embodiments, the classification basic model is:
[0113]
[0114] In the formula, class is the output of the classification model, w an is the pre - coefficient, CFS(an) is the an - th element of the frequency - domain feature vector, and aN is the total number of elements of the frequency - domain feature vector.
[0115] In some embodiments, determining the fitness of the intermediate model according to each output array and the transformer states corresponding to the multiple frequency - domain feature vector samples includes:
[0116] For each output array, perform the following steps respectively:
[0117] Arrange the data in the output array in ascending order of numerical value to obtain a data queue;
[0118] Set transformer state labels for each data in the data queue according to the multiple frequency - domain feature vector samples;
[0119] Determine the fitness of the intermediate model according to the third formula and the transformer state labels set for each data in the data queue, where the third formula is:
[0120]
[0121] In the formula, Fitness is the fitness of the intermediate model, QTT sn is the number of frequency - domain feature vector samples of the sn - th transformer state, Rank sn_max is the highest rank in the data queue where the transformer state label is the sn - th transformer state, Rank sn_min is the lowest rank in the data queue where the transformer state label is the sn - th transformer state, and sN is the number of transformer state labels.
[0122] Exemplarily, in the embodiments of the present invention, the frequency-domain feature vector is input into the state classification model, and the state of the transformer is determined according to the numerical range where the value output by the model is located. The numerical range and the state classification model are constructed together.
[0123] As Figure 2 shown, specifically, in terms of determining the model and the numerical range, the embodiments of the present invention obtain a plurality of frequency-domain feature vector samples 201. The obtaining process of these plurality of frequency-domain feature vector samples 201 is the same as the obtaining steps of the frequency-domain feature vector in the foregoing steps, and they are all obtained by constructing from the ambient temperature, the equipment temperature curve, the vibration curve, and the current curve. At the same time, each frequency-domain feature vector sample also corresponds to a transformer state 205. Then, a plurality of coefficient arrays are obtained and initialized, and these coefficient arrays are respectively substituted into the classification basic model to obtain a plurality of intermediate models 202. The formula of the basic model is:
[0124]
[0125] In the formula, class is the output of the classification model, w an is the preposed coefficient, CFS(an) is the an-th element of the frequency-domain feature vector, and aN is the total number of elements of the frequency-domain feature vector.
[0126] For each intermediate model 202, a plurality of frequency-domain feature vector samples 201 are respectively input into the intermediate model 202 to obtain a plurality of results. These results are constructed into an output array 203, and the data in the output array 203 are arranged according to the numerical size to construct a data queue 204. For a specific transformer state, there are a plurality of frequency-domain feature vector samples 201 corresponding to it. The ratio of these samples to the ranking interval that the transformer state may span can be used as the fitness, which can reflect the rationality of the intermediate model 202 (the rationality of the coefficient array substituted into the intermediate model 202). Therefore, the embodiments of the present invention use the third formula to evaluate the fitness of the model:
[0127]
[0128] In the formula, Fitness is the fitness of the intermediate model, QTT sn is the number of frequency-domain feature vector samples of the sn-th transformer state, Rank sn_max is the highest ranking of the transformer state label of the sn-th transformer state in the data queue, Rank sn_min is the lowest ranking of the transformer state label of the sn-th transformer state in the data queue, and sN is the number of transformer state labels. For Figure 2There are a total of two states 205 of the middle square. After the frequency-domain feature vector samples 201 corresponding to the state 205 of the square are input into the middle model 202 and the obtained results are sorted, they are respectively in the third position and the first position. Then, the fitness calculated according to the third formula is 2 / 3. Similarly, the fitness is calculated for other states (such as the state 205 of the circle) and the average value is taken, then the fitness of the middle model 202 is obtained.
[0129] Similarly, for other models (the middle models 202 constructed based on other coefficient arrays), the fitness is also obtained. If these fitness values are all less than the threshold, select the three coefficient arrays with the best fitness as the targets, adjust the data of other coefficient arrays, and repeat the above steps of inputting multiple frequency-domain feature vector samples 201; if there is a fitness value greater than the threshold, select the middle model corresponding to the maximum fitness as the state classification model. At the same time, set the numerical interval according to the state corresponding to the output of the state classification model.
[0130] In the embodiment of the present invention, first, the environmental temperature, the device temperature curve, the vibration curve, and the current curve are obtained; then, according to the environmental temperature and the device temperature curve, a heat generation power curve is constructed, where the heat generation power curve represents the curve of the heat generation power of the transformer changing with time; then, multiple frequency-domain features are respectively extracted in segments from the heat generation power curve, the vibration curve, and the current curve and constructed into a frequency-domain feature vector; finally, the frequency-domain feature vector is input into the state classification model, and according to the output of the state classification model, the state of the transformer is determined. The abnormal recognition method adopted in the embodiment of the present invention requires less data sources. The classification model is constructed based on the frequency features extracted from the curves, requires fewer sample sizes than intelligent algorithms, the model and the model construction process are simpler, and it is suitable for popularization.
[0131] It should be understood that the magnitudes of the sequence numbers of the steps in the above embodiments do not mean the order of execution. The order of execution of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.
[0132] The following is the device embodiment of the present invention. For the details not described in detail herein, reference can be made to the corresponding method embodiments above.
[0133] Figure 3 The structural schematic diagram of the transformer abnormal warning device provided by the embodiment of the present invention is shown. For the convenience of description, only the parts related to the embodiment of the present invention are shown and are described in detail as follows:
[0134] As Figure 3As shown in the figure, the transformer abnormal warning device 3 includes: a monitoring curve acquisition module 301, a heat generation curve construction module 302, a frequency domain feature vector construction module 303, and a transformer state determination module 304, where:
[0135] The monitoring curve acquisition module 301 is used to acquire the ambient temperature, the equipment temperature curve, the vibration curve, and the current curve;
[0136] The heat generation curve construction module 302 is used to construct a heat generation power curve according to the ambient temperature and the equipment temperature curve, where the heat generation power curve represents the curve of the transformer heat generation power changing with time;
[0137] The frequency domain feature vector construction module 303 is used to respectively extract a plurality of frequency domain features in segments from the heat generation power curve, the vibration curve, and the current curve, and construct them into a frequency domain feature vector;
[0138] The transformer state determination module 304 is used to input the frequency domain feature vector into the state classification model, and determine the state of the transformer according to the output of the state classification model.
[0139] In the above embodiments, the descriptions of each embodiment have their own emphases. For the parts not detailed or recorded in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.
[0140] Those of ordinary skill in the art can realize that the templates, units, and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of the present invention.
[0141] If the above-mentioned module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such understanding, to implement all or part of the processes in the above-mentioned implementation manners of the method of the present invention, it can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-mentioned implementation manners of each transformer abnormal warning method can be realized. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable medium can include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory, random access memory, electrical carrier signal, telecommunication signal, and software distribution medium, etc.
[0142] The above-mentioned implementation manners are only used to illustrate the technical solutions of the present invention, rather than limiting them; although the present invention has been described in detail with reference to the foregoing implementation manners, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing implementation manners, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various implementation manners of the present invention, and should all be included in the protection scope of the present invention.
Claims
1. A transformer abnormality early warning method, characterized in that: include: Obtain ambient temperature, equipment temperature curve, vibration curve and current curve; Constructing a heat generation power curve according to the ambient temperature and the device temperature curve, wherein the heat generation power curve represents a curve of the heat generation power of the transformer changing with time; Extracting a plurality of frequency domain features from the heat generation power curve, the vibration curve and the current curve in sections respectively, and constructing them into frequency domain feature vectors; The frequency domain feature vector is input into a state classification model, and the state of the transformer is determined according to the output of the state classification model.
2. The transformer abnormality early warning method according to claim 1 is characterized in that: The step of constructing a heat generation power curve according to the ambient temperature and the device temperature curve includes: Obtain the mass of the transformer, the specific heat capacity of the transformer, the equivalent heat dissipation surface of the transformer, and the thermal resistance of the transformer; Constructing a first equation expressing the relationship between the device temperature and the ambient temperature and the heat generation power according to the ambient temperature, the device temperature curve, the heat capacity of the transformer, the mass of the transformer, the specific heat capacity of the transformer, the equivalent heat dissipation surface of the transformer and the thermal resistance of the transformer; According to the first equation, a heat generation model is constructed; A heat generation power curve is constructed according to the heat generation model, the ambient temperature and the device temperature curve.
3. The transformer abnormality early warning method according to claim 2 is characterized in that: The first equation is: In the formula, P is the heat generation power, R is the thermal resistance, m is the mass of the transformer, c is the specific heat capacity of the transformer, T is the temperature of the transformer, t is the time variable, S is the effective heat dissipation surface of the transformer, T E is the ambient temperature.
4. The transformer abnormality early warning method according to claim 2 is characterized in that: The heat generation model is: In the formula, P is the heat generation power, R is the thermal resistance, m is the mass of the transformer, c is the specific heat capacity of the transformer, T is the temperature of the transformer, t is the time variable, S is the effective heat dissipation surface of the transformer, T E is the ambient temperature, τ is the time constant, and e is the natural constant.
5. The transformer abnormality early warning method according to claim 1 is characterized in that: The extracting of multiple frequency domain features from the heat generation power curve, the vibration curve and the current curve in sections to construct frequency domain feature vectors includes: For each of the heat generation power curve, the vibration curve, and the current curve, the following steps are performed respectively: Get the fundamental wave period; Starting from a predetermined position of the curve, cutting a curve segment that is several times the fundamental wave period as the curve segment to be processed; Extracting a plurality of frequency domain features from the to-be-processed curve segment according to the fundamental wave period, and adding the extracted plurality of frequency domain features into a frequency domain queue; If the traversal of the curve is not completed, the predetermined position is offset and the step of jumping to the predetermined position of the curve and cutting a curve segment that is several times the fundamental wave period as the curve segment to be processed; Otherwise, the multiple data in the frequency domain queue are added to the frequency domain feature vector as multiple frequency domain feature elements.
6. The transformer abnormality early warning method according to claim 5 is characterized in that: The extracting a plurality of frequency domain features from the curve segment to be processed according to the fundamental wave period includes: A plurality of frequency domain features are extracted from the curve segment to be processed according to the second formula and the fundamental wave period, wherein the second formula is: Where CF(k) is the kth frequency domain eigenvalue, cN is the total number of data obtained after the curve segment to be processed is discretized, CDW(cn) is the cnth data obtained after the curve segment to be processed is discretized, e is a natural constant, j is an imaginary unit, ω0 is the frequency corresponding to the fundamental period, PN is the number of data obtained after the curve segment to be processed is discretized within the duration of the fundamental period, and π is the circumference of a circle.
7. The transformer abnormality early warning method according to any one of claims 1 to 6, characterized in that: The construction process of the state classification model includes: Obtain multiple frequency domain feature vector samples and multiple coefficient arrays, wherein each frequency domain feature vector sample corresponds to a transformer state; Substituting the multiple coefficient arrays into the classification basic model respectively to obtain multiple intermediate models; Substituting the plurality of frequency domain feature vector samples into each intermediate model in turn, and constructing the obtained plurality of model outputs into an output array, thereby obtaining a plurality of output arrays; Determining the fitness of the intermediate model according to the transformer state corresponding to each output array and the plurality of frequency domain feature vector samples, thereby obtaining a plurality of fitnesses; If the multiple fitnesses are all less than the fitness threshold, three coefficient arrays with the best fitnesses are selected from the multiple coefficient arrays as three target arrays, other coefficient arrays are adjusted according to the three target arrays, and the process jumps to the step of substituting the multiple coefficient arrays into the classification basic models respectively to obtain multiple intermediate models; Otherwise, substitute the coefficient array of the best fitness into the classification basic model to obtain the state classification model.
8. The transformer abnormality early warning method according to claim 7 is characterized in that: The classification basic model is: In the formula, class is the classification model output, w an is the prefix coefficient, CFS(an) is the an-th element of the frequency domain eigenvector, and aN is the total number of elements in the frequency domain eigenvector.
9. The transformer abnormality early warning method according to claim 7, characterized in that: The step of determining the fitness of the intermediate model according to the transformer state corresponding to each output array and the plurality of frequency domain feature vector samples comprises: For each output array, perform the following steps: Arrange the data in the output array in order of numerical value to obtain a data queue; Setting a transformer state label for each data in the data queue according to the plurality of frequency domain feature vector samples; According to the third formula and each data of the data queue, a transformer state label is set to determine the fitness of the intermediate model, wherein the third formula is: In the formula, Fitness is the fitness of the intermediate model, QTT sn is the number of frequency domain feature vector samples of the sn-th transformer state, Rank sn_max The transformer state label is the highest rank of the sn-th transformer state in the data queue, Rank sn_min The transformer state label is the lowest position of the sn-th transformer state in the data queue, and sN is the number of transformer state labels.
10. A transformer abnormality early warning device, characterized in that: Used to implement the transformer abnormality early warning method according to any one of claims 1 to 9, the transformer abnormality early warning device comprises: Monitoring curve acquisition module, used to obtain ambient temperature, equipment temperature curve, vibration curve and current curve; A heat generation curve construction module, used to construct a heat generation power curve according to the ambient temperature and the device temperature curve, wherein the heat generation power curve represents a curve of the heat generation power of the transformer changing with time; A frequency domain feature vector construction module, used to extract a plurality of frequency domain features from the heat generation power curve, the vibration curve and the current curve in sections, respectively, and construct them into frequency domain feature vectors; as well as, The transformer state determination module is used to input the frequency domain feature vector into a state classification model, and determine the state of the transformer according to the output of the state classification model.