Method for identifying abnormal operation of DC magnetic bias of power transformer

By building a physical physical dynamic mode platform and virtual simulation body, combined with the Transformer-KOA model, the problem of low fault identification accuracy in monitoring of DC bias magnetic abnormal operation of transformer is solved, and faster and more accurate fault warning and identification are achieved.

CN119066566APending Publication Date: 2024-12-03NORTHEAST DIANLI UNIVERSITY
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411060262.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-05
Publication Date
2024-12-03

Smart Images

  • Figure CN119066566A_ABST
    Figure CN119066566A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of transformers, and provides a method for identifying direct-current magnetic bias abnormal operation of a power transformer, which is scientific, reasonable, efficient and practical and realizes fault early warning and identification under the condition of direct-current magnetic bias abnormal operation of the power transformer. Acquiring electromagnetic and mechanical parameters to form a sample set information base; and electromagnetic-mechanical multi-information association is constructed, and a fitting formula of current representation vibration is searched. Based on a Transform-KOA sample learning method, a current time domain signal is decomposed, clustering division is carried out in combination with a fitting formula, a transformer direct current magnetic bias abnormal operation early warning scheme is formed, the continuous optimization problems that a traditional algorithm is prone to falling into local optimum, the convergence speed is low, population diversity is lacked and the like are effectively solved, and the early warning method is suitable for transformer direct current magnetic bias abnormal operation early warning. The method can achieve the recognition of the abnormal operation state of the DC magnetic bias of the transformer more quickly and accurately.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of transformers, and in particular relates to a method for identifying abnormal operation of a power transformer with direct current magnetic bias. Background Art

[0002] The power transformer is the core component of the power system, responsible for increasing and decreasing the voltage to achieve efficient transmission of electric energy. It ensures the safety of the power grid through electrical isolation, while regulating the voltage to maintain system stability. The transformer can also limit the short-circuit current, protect the power grid equipment, and play a key role in improving the quality of power and the economy of system operation. Therefore, the reliability and performance of the transformer directly affect the normal operation of the entire power system. Therefore, it is of great research value and engineering significance to identify and warn the abnormal operation of the transformer. Using measurable electrical parameters to describe the physical characteristics that are difficult to measure is the key to monitoring and identifying the abnormal operation of the transformer. At present, the monitoring of abnormal operation of transformer DC bias magnetization generally adopts real-time current monitoring, Hall effect sensor monitoring, vibration analysis and other methods, but there are often problems such as the accuracy and stability of the sensor being greatly affected by the ambient temperature, the low accuracy of fault identification, and the difficulty of extracting complex signals.

[0003] Therefore, a new technical solution is urgently needed in the prior art to solve this problem. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method for identifying abnormal operation of a power transformer using direct current bias magnetic field to solve the technical problems of low fault identification accuracy and difficulty in extracting complex signals in monitoring abnormal operation of direct current bias magnetic field of the current transformer.

[0005] A method for identifying abnormal operation of a power transformer with DC bias magnetic field comprises the following steps, which are performed in sequence:

[0006] Step 1, build a physical entity dynamic model platform, the physical entity dynamic model platform includes a transformer that is proportionally reduced to an actual large transformer, a primary and secondary windings of the transformer, a current monitoring unit, a voltage monitoring unit, a harmonic current adjustment unit, a vibration monitoring unit, a noise monitoring unit, a vibration measurement sensor and a load adjustment unit, determine the material property parameters of each component of the transformer, measure and obtain the electrical information and vibration information of the transformer under different operating modes, wherein the electrical information includes current and voltage, and the vibration information includes the vibration acceleration of the core and the winding measuring points; construct a virtual simulation entity based on the physical entity, solve the magnetic field model in the virtual simulation entity to obtain winding leakage, winding electromagnetic force, winding vibration acceleration, core main magnetic flux, core electromagnetic force, and core vibration acceleration; the different operating modes include normal operation of the transformer and abnormal operation of DC bias magnetization;

[0007] Step 2: define the DC disturbance level based on the no-load current, use the DC disturbance level as a label, use time t as an index, fuse the information obtained in step 1 by the physical entity of the transformer and the virtual simulation entity, and store them in the corresponding information library, wherein the information library is divided into an electrical information library and a mechanical information library, which together constitute a global information library. Based on the global information library, use the transformer simulation current, voltage, load rate, and component vibration information to compare the transformer physical current, voltage, load rate, and transformer component vibration information, and verify the accuracy of the simulation model through the set evaluation criteria; the transformer components include a core and a winding;

[0008] Step 3: Use the DC disturbance level and vibration acceleration to build a mapping relationship to form a transformer DC bias magnetic abnormality operation state domain, wherein the transformer DC bias magnetic abnormality operation state domain is divided into a normal operation domain, a bias magnetic reminder domain, a bias magnetic abnormality domain, and a bias magnetic alarm domain according to a set threshold range;

[0009] Step 4: Based on the global information database, collect the simulated current, real current and vibration acceleration of transformer components, form a training set and a test set according to the set ratio, use the training set to train the Transformer model, and use the Kepler optimization algorithm KOA to optimize the autocorrelation operation learning rate and the number of neurons in the Transformer model, set the false alarm rate of the transformer operation status as the loss function, and set the number of iterations to obtain the trained Transformer-KOA model;

[0010] Step 5: Input the test set samples into the trained Transformer-KOA model and output the transformer operating status: normal operating status, bias magnetization reminder status, bias magnetization abnormal status, and bias magnetization alarm status, so as to realize the recognition of abnormal DC bias magnetization operating status of the transformer.

[0011] The process of measuring the transformer electrical information and vibration information under different operating modes in step 1 is as follows:

[0012] 1) Set different operation modes of the transformer, debug the voltage regulation unit and current monitoring module, collect the voltage and current of the measuring point in real time, and filter the voltage and current to remove noise and interference information;

[0013] 2) In view of the operation of the transformer subjected to DC disturbance, vibration measurement points are arranged, piezoelectric accelerometers are connected as vibration measurement sensors, and the vibration monitoring module is debugged. The vibration information of the transformer components is collected in real time under different modes, and the vibration signal is subjected to anti-noise processing.

[0014] The DC disturbance level in step 2 is defined as:

[0015] I h =hI0;

[0016] In the formula, I h is the DC level, h is the DC multiple, and I0 is the no-load current.

[0017] The Transformer model in step 4 is composed of an input layer, a certain number of hidden layers and an output layer, which can be expressed as:

[0018] f t =σ g (w f ·[h t-1 ,x t ]+b f );

[0019] i t =σ g (w i ·[h t-1 ,x t ]+b i );

[0020] o t =σ g (w o ·[h t-1 ,x t ]+b o );

[0021] C t =f t *C t-1 +i t *tanh(W C ·[h t-1 ,x t ]+b C );

[0022] h t =o t *tanh(C t );

[0023] Among them, f t ,i t and t are the forget, input and output gate activation vectors respectively, t is the layer index, σ g is the sigmoid activation function, C t is the unit activation vector, h t is the output vector, h t-1 is the hidden state of the previous time step, x t is the input of the current time step, b f is the bias vector of the forget gate, w f is the weight matrix of the forget gate, w i is the weight matrix of the input gate, bi is the bias vector of the input gate, w o is the weight matrix of the output gate, b o is the bias vector of the output gate, C t-1 is the cell state at the previous time step, W C is the weight matrix of the candidate cell state, b C is the bias vector of the candidate cell state.

[0024] The optimization process of the Kepler optimization algorithm KOA in step 4 is as follows:

[0025] 1) Population initialization

[0026]

[0027] e i =rand [0,1] ,i=1,...,N;

[0028] T i =|r|,i=1,...,N;

[0029] Where: P i j represents the initial position of the planet for the jth decision variable, T i , e i They represent the corresponding orbital period and eccentricity respectively, r is a randomly generated value between 0 and 1, rand [0,1] is a random vector in the range [0,1], and They represent the upper and lower bounds of the j-th decision variable, N represents the number of candidate solutions in the search space, and d represents the dimension of the problem to be solved;

[0030] 2) Define the Sun's gravity

[0031] The main reason why planets orbit the sun is that the sun's gravity keeps them in their orbits. Each planet has its own gravity, which is proportional to the size of the planet. The closer a planet is to the sun, the greater its orbital speed, and vice versa.

[0032] The sun's gravity is defined as follows:

[0033]

[0034] Where: M sn and m pn represents the normalized value of the mass of the sun and planets, ε is a very small value; μ(t) is a function that decays exponentially with time t and is defined to control the search accuracy; r1 represents a randomly generated value between 0 and 1, R in Represents the distance from the sun S to the target planet Pi The Euclidean distance between

[0035] 3) Calculate the speed of the target planet

[0036] A planet's speed increases as it approaches the sun and decreases as it moves away from the sun to avoid being pulled toward the sun. The speed of the planet is expressed as:

[0037]

[0038] Where: V i (t) represents the speed of the planet at time t; P a and P b represents two random positions in the planet population; P i Indicates the position of the planet at the corresponding time; R in (t) represents the sun S and P at time t i Euclidean distance; r3 and r4 are randomly generated values ​​between 0 and 1, r5 is a vector of random values ​​between 0 and 1, α, β, L represent the selection variables after folding; U1 and U2 represent the selection criteria; P i,up , P i,low They represent the upper and lower limits of the planetary positions respectively;

[0039] 4) Escape from local optimum

[0040] In the solar system, most objects rotate counterclockwise, they all rotate around their own axis, but some celestial bodies revolve around the sun in a clockwise direction. KOA uses this behavior to change the search direction using the flag γ to escape the local optimal area;

[0041] 5) Update planetary positions

[0042]

[0043] Where: P i (t+1) is the new position of the planet after the update; P i (t) is the position of the planet before the update; S(t) is the optimal position of the sun S so far; r is a number randomly generated based on the normal distribution; γ represents the search direction; F(t) represents the gravity of the sun at time t; U is the selection criterion, which is: embedding the self-correction and generalization representation modules based on the fusion feature index ξ, comparing the self-correction result with the decoder output to improve the accuracy of morphological recognition, and merging the generalization representation with the morphological recognition result to improve the generalization ability of class domain division;

[0044] 6) Update the distance to the sun

[0045]

[0046] Where: Pa (t), P b (t) represent two random positions in the planetary population at time t, and h represents the adaptive factor;

[0047] 7) Elite Strategy

[0048]

[0049] After optimization, if the fitness is the best, update the planet position P i '(t+1), otherwise the planet position remains unchanged, f(P i (t+1))、f(P i (t)) represent the planet position fitness at time t+1 and time t, respectively.

[0050] Through the above design scheme, the present invention can bring the following beneficial effects:

[0051] The present invention provides a scientific, reasonable, efficient and practical method for realizing fault warning and identification under abnormal operation of power transformers with DC bias. The method uses a physical entity dynamic model platform and a virtual simulation entity of the transformer to obtain electromagnetic and mechanical parameters and form a sample set information library. Electromagnetic-mechanical multi-information association is constructed to find a fitting formula for current characterization of vibration. Based on the Transformer-KOA sample learning method, the current time domain signal is decomposed, and clustering is performed in combination with the fitting formula to form an abnormal operation warning scheme for the transformer with DC bias.

[0052] The present invention is based on a global information library, collects winding current and component vibration signals, forms a training set and a test set, uses the training set to train a Transformer model, and uses the Kepler optimization algorithm to optimize the autocorrelation operation learning rate and the number of neurons in the Transformer model, effectively solving the continuous optimization problems existing in traditional algorithms, such as easy to fall into local optimum, slow convergence speed, lack of population diversity, etc., and can more quickly and accurately realize the recognition of abnormal operating status of transformer DC bias magnetic field. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The present invention is further described below with reference to the accompanying drawings and specific embodiments:

[0054] Figure 1 A flow chart of a method for identifying abnormal operation of a power transformer with DC bias magnetic flux in the present invention;

[0055] Figure 2 It is a test wiring schematic diagram of a physical entity dynamic model platform in a method for identifying abnormal operation of a power transformer with direct current bias magnetic field according to the present invention;

[0056] Figure 3A transformer virtual simulation entity and a distribution diagram of measuring points in a method for identifying abnormal operation of a power transformer with DC bias magnetic field according to the present invention;

[0057] Figure 4a It is a multi-physical information association diagram of simulation information of measuring point ① in an embodiment of a method for identifying abnormal operation of a power transformer with DC bias magnetic field according to the present invention;

[0058] Figure 4b It is a multi-physical information association diagram of the experimental information of the measuring point No. ① in an embodiment of a method for identifying abnormal operation of a power transformer with DC bias magnetic field according to the present invention;

[0059] Figure 4c It is a multi-physical information association diagram of simulation information of measuring point No. ③ in an embodiment of a method for identifying abnormal operation of a power transformer with DC bias magnetic field according to the present invention;

[0060] Figure 4d It is a multi-physical information association diagram of the experimental information of the measuring point No. ① in an embodiment of a method for identifying abnormal operation of a power transformer with DC bias magnetic field according to the present invention;

[0061] Figure 5 The present invention discloses a flow chart of optimizing a Transformer model based on KOA in a method for identifying abnormal operation of a power transformer with DC bias magnetic field. DETAILED DESCRIPTION

[0062] The objective of the present invention is achieved by the following technical scheme: A method for identifying abnormal operation of a power transformer with DC bias magnetic field is a method for warning abnormal operation of a transformer with DC bias magnetic field based on a Transformer-KOA sample learning method, which comprises the following steps:

[0063] 1. Simulation and experimental data acquisition

[0064] Transformer electromagnetic-mechanical information collection includes electromagnetic and mechanical entity information collection and electromagnetic and mechanical simulation data collection. The transformer core and winding are divided into electromagnetic and mechanical subsystems, and key physical feature information is extracted respectively. The electromagnetic information includes port voltage u, current i, load rate η, DC level I h And leakage magnetic B σ etc.; mechanical information includes time-varying parameters such as winding electromagnetic force F and vibration acceleration g. In the data extraction process, physical entities can directly extract u, i, I h , η, g and other characteristic information; u, i, I h ,η,g,B σ and F, etc. The temporal and spatial consistency of the simulation / physical entity and multi-source feature information is checked to verify the effectiveness of the simulation / physical entity and feature information acquisition, and the current is further used to characterize the vibration of the transformer under DC disturbance.

[0065] 2. Time Series Prediction Using Transformer

[0066] The Transformer network is a recurrent neural network (RNN) that is suitable for sequence prediction problems. Unlike standard feedforward neural networks, the Transformer network has feedback connections. It forwards the output of one layer to the next layer, and forwards the output of one layer to itself. The Transformer unit includes a storage unit that keeps information in memory for a long time. A set of gates is used to control the time to store, forget, and output information. This architecture allows the Transformer model to learn longer-term dependencies.

[0067] 3. Kepler Optimization Algorithm (KOA)

[0068] Kepler optimism algorithm (KOA), the algorithm is inspired by Kepler's laws of planetary motion. There are four basic operators that affect the trajectory of a planet around the sun: the planet's position, mass, gravity, and orbital velocity. These operators represent the basis for the mathematical model built on which the proposed algorithm is built. In KOA, each planet and its position is considered as a candidate solution, and the best solution so far (the sun) is randomly updated through an optimization process.

[0069] 3. DC bias magnetic anomaly identification solution

[0070] By extracting the electromagnetic-mechanical information features of the transformer and correlating the multivariate feature information, the Transformer-KOA model is formed. Figure 1 The identification scheme of the abnormal operation warning method of the DC bias magnetic of the power transformer shown is as follows:

[0071] 1) Build a physical entity dynamic model platform to measure the transformer electrical information (current, voltage) and vibration information (vibration acceleration of the core and winding measurement points); build a virtual entity based on the physical entity, and solve the magnetic field model to obtain winding leakage flux, winding electromagnetic force, winding vibration acceleration, core main magnetic flux, core electromagnetic force, and core vibration acceleration.

[0072] 2) Taking the DC disturbance level as the label and time t as the index, the physical entity and simulation entity information of the transformer are integrated and stored in the information base, where the information base is divided into an electrical information base and a mechanical information base, which together constitute a global information base. Based on the global information base, the transformer simulation current, voltage, load rate, and component vibration acceleration information are compared with the transformer entity current, voltage, load rate, and component vibration acceleration information to verify the accuracy of the simulation model, effectively reflecting the reliability of the simulation model evaluation entity.

[0073] 3) The DC disturbance level and vibration acceleration are used to construct the mapping relationship between simulation and experiment, forming the transformer DC bias abnormal operation state domain: normal operation, bias reminder, bias abnormality, and bias alarm.

[0074] 4) Based on the global information library, the winding current and component vibration signals are collected to form a training set and a test set. The training set is used to train the Transformer model, and the Kepler optimization algorithm is used to optimize the autocorrelation operation learning rate and the number of neurons in the Transformer model. The initial search space is the autocorrelation operation learning rate ∈ [0.001, 0.01], the number of neurons ∈ [3, 12], the false alarm rate of the transformer operation status is set as the loss function, and the number of iterations is set to 100. For specific optimization methods, see Figure 5 ; Effectively solve the continuous optimization problems existing in traditional algorithms, such as easy to fall into local optimality, slow convergence, and lack of population diversity.

[0075] 5) Input the test set samples into the trained Transformer-KOA model and output the transformer operating status: normal operation, bias magnetization reminder, bias magnetization abnormality, and bias magnetization alarm. This enables the transformer DC bias magnetization abnormality operating status recognition.

[0076] Example:

[0077] The present invention will be further described below in conjunction with specific embodiments:

[0078] A method for identifying abnormal operation of a power transformer with direct current magnetic bias, comprising the following steps:

[0079] 1. Extraction and verification of electromagnetic-mechanical information

[0080] 1.1 Physical Entity Information Acquisition

[0081] Combined with the actual operation of large transformers, a proportionally reduced three-phase experimental transformer BSS-1000VA was selected as the research object. The primary and secondary windings of the transformer are connected in star-delta connection, and the specific parameters are shown in Table 1.

[0082] Table 1 Three-phase transformer parameters

[0083]

[0084]

[0085] Note: Positive Young's modulus and Poisson's ratio are obtained through material tension and compression tests.

[0086] The DC disturbance level is defined as:

[0087] I h =hI0;

[0088] In the formula, I h is the DC disturbance level, h is the DC multiple, and I0 is the no-load current.

[0089] The process of collecting electromagnetic-mechanical information of transformer is as follows:

[0090] 1) Set different operation modes of the transformer, debug the sensor and current monitoring module, collect the port voltage u and current i information in real time, filter the voltage and current, and remove noise and interference information;

[0091] 2) According to the operation of the transformer under DC disturbance, arrange the vibration measuring points, namely, measuring point ①, measuring point ②, measuring point ③, and measuring point ④, see the attached Figure 3 , connect the piezoelectric acceleration sensor and debug the vibration monitoring module, collect transformer winding vibration information in real time under different modes, and perform anti-noise processing on the vibration signal;

[0092] 3) The material properties of transformer components were determined by using multifunctional material tension and compression tests. The specific information is shown in Table 1;

[0093] 4) Using the DC disturbance level as a label and time t as an index, the transformer entity current, voltage, load rate, and component vibration acceleration information are stored.

[0094] 1.2 Simulation Information Acquisition

[0095] 1) Digital modeling is carried out by actually measuring the electrical, mechanical and material data of the transformer. Through a large number of simulation and experimental comparisons, the model parameters and field settings are repeatedly corrected to achieve the consistency of the simulation in terms of entity and operating characteristics, so that the simulation model can effectively reflect the electromagnetic mechanical process of the physical entity.

[0096] 2) Use finite element simulation software to construct a proportional finite element virtual simulation entity, see the attached figure Figure 2 ,Some physical parameters of the transformer (silicon steel sheet model is 50WW800) are shown in Table 1. The virtual entity is divided into magnetic field connection domain, current excitation domain and mechanical stress domain, corresponding to the core, winding and other components, and the space is consistent with the physical entity.

[0097] 3) The virtual entity material parameter settings are consistent with the information in Table 1. During the simulation process, the mesh division uses differentiated tetrahedral meshes. The core winding components that need to be accurately calculated select "fine" units, and the pads and pull plates select "rough" units. The perfect matching layer uses hexahedral sweeping. The solver selects the "full coupling solution method". Apply excitation and boundary: ① The winding excitation in the electromagnetic system is set to a circular current, and the magnetic field boundary is set to the parallel boundary condition of the magnetic field lines, and the rest are set to natural boundary conditions; ② The transformer base in the mechanical system is set to a fixed constraint condition, and the rest is subject to the roller support constraint condition, that is, the transformer base is fixed, and only the axial vibration of the component is considered.

[0098] 4) Using the DC disturbance level as a label and time t as an index, store the transformer simulation results: voltage, current, load rate, component vibration acceleration, main / leakage flux, component force, etc.

[0099] 1.3 Construction of global information database and verification of real and fake information

[0100] 1) Taking the DC disturbance level as the label and time t as the index, the physical entity and simulation entity information of the transformer are integrated and stored in the information database, where the information database is divided into an electrical information database and a mechanical information database, which together constitute a global information database.

[0101] 2) Based on the global information database, the transformer simulation current, voltage, load rate, and component vibration acceleration information are compared with the transformer entity current, voltage, load rate, and component vibration acceleration information to verify the accuracy of the simulation model and effectively reflect the reliability of the simulation model evaluation entity.

[0102] 2. Current-vibration correlation

[0103] Through simulation and experiments, it is shown that when DC current invades, the vibration of transformer components intensifies and has obvious fluctuation characteristics. Taking the winding as an example, the component of its vibration signal in the 100Hz frequency band (twice the frequency of power frequency excitation) accounts for a large proportion. The vibration acceleration size g_100Hz in the 100Hz frequency band is selected as the fingerprint quantity to conduct an in-depth analysis of the vibration acceleration of the measuring point.

[0104] With the load rate as the x-coordinate, the DC disturbance level as the y-coordinate, and the component vibration acceleration as the z-coordinate, relevant data is extracted from the global information library and the MATLAB fitting toolbox is used to perform data fitting. The electromagnetic-vibration information mapping relationship between measuring points ① and ③ is set as shown in the attached figure. Figures 4a to 4d shown.

[0105] When the transformer is subjected to DC intrusion, the vibration of the components intensifies, and the mapping relationship between the vibration acceleration and the DC disturbance level can be expressed in the form of a mathematical function. Taking the load rate of 50% as an example, the mapping fitting function between the DC disturbance level and the vibration acceleration is as follows:

[0106] f(β,I h %) = p1(I h % ) 3 +p2(I h % ) 2 +p3(I h %)+p4 (1);

[0107] Where p1, p2, p3, and p4 represent the fitting formula coefficients respectively.

[0108] Based on the current-vibration correlation, taking the winding vibration measuring point ① and the load rate of 50% as an example, the transformer vibration state domain is divided into four states: normal operation, bias magnetization reminder, bias magnetization abnormality, and bias magnetization alarm.

[0109] Table 2 Classification of transformer vibration states

[0110]

[0111] 3. Transformer Algorithm

[0112] The Transformer network is a recurrent neural network (RNN) that is suitable for sequence prediction problems. Unlike standard feedforward neural networks, the Transformer network has feedback connections. It forwards the output of one layer to the next layer, and forwards the output of one layer to itself. The Transformer unit includes a storage unit that keeps information in memory for a long time. A set of gates is used to control the time to store, forget, and output information. This architecture allows the Transformer model to learn longer-term dependencies.

[0113] The Transformer network consists of an input layer, a certain number of hidden layers, and an output layer:

[0114] f t =σ g (w f ·[h t-1 ,x t ]+b f ) (2);

[0115] i t =σ g (w i ·[h t-1 ,x t ]+b i ) (3);

[0116] o t =σ g (w o ·[ht-1 ,x t ]+b o ) (4);

[0117] C t =f t *C t-1 +i t *tanh(W C ·[h t-1 ,x t ]+b C ) (5);

[0118] h t =o t *tanh(C t ) (6);

[0119] Among them, f t ,i t and t are the forget, input, and output gate activation vectors, respectively; t is the layer index; σ g is the sigmoid activation function; C t is the unit activation vector; h t is the output vector.

[0120] 4. Kepler Optimization Algorithm (KOA)

[0121] Kepler optimism algorithm (KOA), which is inspired by Kepler's laws of planetary motion. There are four basic operators that affect the trajectory of a planet around the sun: the planet's position, mass, gravity, and orbital velocity. These operators represent the basis for the mathematical model on which the proposed algorithm is built. In KOA, each planet and its position is considered as a candidate solution, and the best solution so far (the sun) is randomly updated through an optimization process. The optimization process is as follows:

[0122] 1) Population initialization

[0123]

[0124] e i =rand [0,1] ,i=1,...,N (8);

[0125] T i =|r|,i=1,...,N (9);

[0126] Where: P i j represents the initial position of the planet for the jth decision variable, T i , e iThey represent the corresponding orbital period and eccentricity respectively, r is a randomly generated value between 0 and 1, rand [0,1] is a random vector in the range [0,1], and They represent the upper and lower bounds of the j-th decision variable, N represents the number of candidate solutions in the search space, and d represents the dimension of the problem to be solved.

[0127] 2) Define the Sun's gravity

[0128] The main reason why planets orbit the Sun is that the Sun's gravity keeps them in their orbits. Each planet has its own gravitational pull, which is proportional to the size of the planet. The closer a planet is to the Sun, the greater its orbital speed, and vice versa.

[0129] The sun's gravity is defined as follows:

[0130]

[0131] Where: M sn and m pn represents the normalized value of the mass of the sun and the planets, ε is a very small value; μ(t) is a function that decays exponentially with time t and is defined to control the search accuracy; e i represents the eccentricity of the planetary orbit with a value between 0 and 1; r1 represents a randomly generated value between 0 and 1, R in Represents the distance from the sun S to the target planet P i The Euclidean distance between .

[0132] 3) Calculate the speed of the target planet

[0133] A planet's speed increases as it moves closer to the Sun and decreases as it moves away from the Sun. If an object is very close to the Sun, the Sun's gravity is quite strong, and the planet will try to increase its speed to avoid being pulled towards the Sun.

[0134]

[0135] Where: V i (t) represents the planet P i The speed at time t, P a and P b Represents two random planets a, b, R in the planet population in (t) represents the distance between the sun S and the planet P at time t i , r3 and r4 are randomly generated values ​​between 0 and 1, and r5 is a vector of random values ​​between 0 and 1.

[0136] 4) Escape from local optimum

[0137] In the solar system, most objects rotate counterclockwise, they rotate around their own axis, but some celestial bodies orbit the sun clockwise. KOA uses this behavior to escape the local optimal area by changing the search direction using the flag γ.

[0138] 5) Update planetary positions

[0139]

[0140] Where: P i (t+1) is the new position of the planet after the update, P i (t) is the position of the planet before the update, S(t) is the optimal position of the sun S so far, r is a number randomly generated based on a normal distribution, and U is the selection criterion: the self-correction and generalization representation modules are embedded based on the fusion feature index ξ, the self-correction result is compared with the decoder output to improve the accuracy of morphological recognition, and the generalization representation is fused with the morphological recognition result to improve the generalization ability of class domain division.

[0141] 6) Update the distance to the sun

[0142]

[0143] 7) Elite Strategy

[0144]

[0145] If the fitness is the best after optimization, update the planet position P i '(t+1) Otherwise, the position of the planet remains unchanged.

[0146] 5. DC bias magnetic anomaly identification solution

[0147] By extracting the electromagnetic-mechanical information characteristics of the transformer, correlating the multivariate feature information, and combining the Transformer-KOA model, an identification scheme for the abnormal operation early warning method of DC bias magnetic field of power transformer is formed.

[0148] 1) Build a physical entity dynamic model platform to measure the transformer electrical information (current, voltage) and vibration information (vibration acceleration of the core and winding measurement points); build a virtual entity based on the physical entity, and solve the magnetic field model to obtain winding leakage flux, winding electromagnetic force, winding vibration acceleration, core main magnetic flux, core electromagnetic force, and core vibration acceleration.

[0149] 2) Using the DC disturbance degree as a label and time t as an index, the physical entity and simulation entity information of the transformer are integrated and stored in the information base, where the information base is divided into an electrical information base and a mechanical information base, which together constitute a global information base. Based on the global information base, the transformer simulation current, voltage, load rate, and component vibration acceleration information are compared with the transformer entity current, voltage, load rate, and component vibration acceleration information to verify the accuracy of the simulation model, effectively reflecting the reliability of the simulation model evaluation entity.

[0150] 3) The DC disturbance level and vibration acceleration are used to construct the mapping relationship between simulation and experiment, forming the transformer DC bias abnormal operation state domain: normal operation, bias reminder, bias abnormality, and bias alarm.

[0151] 4) Based on the global information base, the winding current and component vibration signals are collected to form a training set and a test set. The training set is used to train the Transformer model, and the Kepler optimization algorithm (Formula 7-14) is used to optimize the autocorrelation operation learning rate and the number of neurons in the Transformer model. The initial search space is the autocorrelation operation learning rate ∈ [0.001, 0.01], the number of neurons ∈ [3, 12], the false alarm rate of the transformer operation status is set as the loss function, and the number of iterations is set to 100. See the attached for specific optimization methods. Figure 5 The Transformer-KOA model effectively solves the continuous optimization problems existing in traditional algorithms, such as easy to fall into local optimality, slow convergence, and lack of population diversity.

[0152] 5) Input the test set samples into the trained Transformer-KOA model and output the transformer operating status: normal operation, bias magnetization reminder, bias magnetization abnormality, and bias magnetization alarm. This enables the transformer DC bias magnetization abnormality operating status recognition.

Claims

1. A method for identifying abnormal operation of a power transformer with DC bias magnetic field, characterized in that: The process includes the following steps, which are performed in sequence: Step 1, build a physical entity dynamic model platform, the physical entity dynamic model platform includes a transformer that is proportionally reduced to an actual large transformer, a primary and secondary windings of the transformer, a current monitoring unit, a voltage monitoring unit, a harmonic current adjustment unit, a vibration monitoring unit, a noise monitoring unit, a vibration measurement sensor and a load adjustment unit, determine the material property parameters of each component of the transformer, measure and obtain the electrical information and vibration information of the transformer under different operating modes, wherein the electrical information includes current and voltage, and the vibration information includes the vibration acceleration of the core and the winding measuring points; construct a virtual simulation entity based on the physical entity, solve the magnetic field model in the virtual simulation entity to obtain winding leakage, winding electromagnetic force, winding vibration acceleration, core main magnetic flux, core electromagnetic force, and core vibration acceleration; the different operating modes include normal operation of the transformer and abnormal operation of DC bias magnetization; Step 2: define the DC disturbance level based on the no-load current, use the DC disturbance level as a label, use time t as an index, fuse the information obtained in step 1 by the physical entity of the transformer and the virtual simulation entity, and store them in the corresponding information library, wherein the information library is divided into an electrical information library and a mechanical information library, which together constitute a global information library. Based on the global information library, use the transformer simulation current, voltage, load rate, and component vibration information to compare the transformer physical current, voltage, load rate, and transformer component vibration information, and verify the accuracy of the simulation model through the set evaluation criteria; the transformer components include a core and a winding; Step 3: Use the DC disturbance level and vibration acceleration to build a mapping relationship to form a transformer DC bias magnetic abnormality operation state domain, wherein the transformer DC bias magnetic abnormality operation state domain is divided into a normal operation domain, a bias magnetic reminder domain, a bias magnetic abnormality domain, and a bias magnetic alarm domain according to a set threshold range; Step 4: Based on the global information database, collect the simulated current, real current and vibration acceleration of transformer components, form a training set and a test set according to the set ratio, use the training set to train the Transformer model, and use the Kepler optimization algorithm KOA to optimize the autocorrelation operation learning rate and the number of neurons in the Transformer model, set the false alarm rate of the transformer operation status as the loss function, and set the number of iterations to obtain the trained Transformer-KOA model; Step 5: Input the test set samples into the trained Transformer-KOA model and output the transformer operating status: normal operating status, bias magnetization reminder status, bias magnetization abnormal status, and bias magnetization alarm status, so as to realize the recognition of abnormal DC bias magnetization operating status of the transformer.

2. The method for identifying abnormal operation of a power transformer with DC bias magnetic field according to claim 1 is characterized in that: The process of measuring the transformer electrical information and vibration information under different operating modes in step 1 is as follows: 1) Set different operation modes of the transformer, debug the voltage regulation unit and current monitoring module, collect the voltage and current of the measuring point in real time, and filter the voltage and current to remove noise and interference information; 2) In view of the operation of the transformer subjected to DC disturbance, vibration measurement points are arranged, piezoelectric accelerometers are connected as vibration measurement sensors, and the vibration monitoring module is debugged. The vibration information of the transformer components is collected in real time under different modes, and the vibration signal is subjected to anti-noise processing.

3. The method for identifying abnormal operation of a power transformer with DC bias magnetic field according to claim 1 is characterized in that: The DC disturbance level in step 2 is defined as: I h =hI0; In the formula, I h is the DC level, h is the DC multiple, and I0 is the no-load current.

4. The method for identifying abnormal operation of a power transformer with DC bias magnetic field according to claim 1 is characterized in that: The Transformer model in step 4 is composed of an input layer, a certain number of hidden layers and an output layer, which can be expressed as: f t =σ g (w f ·[h t-1 ,x t ]+b f ); i t =σ g (w i ·[h t-1 ,x t ]+b i ); the t =s g (w o ·[h t-1 ,x t ]+b o ); C t =f t *C t-1 +i t *tanh(W C ·[h t-1 ,x t ]+b C ); h t =o t *tanh(C t ); Among them, f t ,i t and t are the forget, input and output gate activation vectors respectively, t is the layer index, σ g is the sigmoid activation function, C t is the unit activation vector, h t is the output vector, h t-1 is the hidden state of the previous time step, x t is the input of the current time step, b f is the bias vector of the forget gate, w f is the weight matrix of the forget gate, w i is the weight matrix of the input gate, b i is the bias vector of the input gate, w o is the weight matrix of the output gate, b o is the bias vector of the output gate, C t-1 is the cell state at the previous time step, W C is the weight matrix of the candidate cell state, b C is the bias vector of the candidate cell state.

5. The method for identifying abnormal operation of a power transformer with DC bias magnetic field according to claim 1 is characterized in that: The optimization process of the Kepler optimization algorithm KOA in step 4 is as follows: 1) Population initialization it is i =row [0,1] ,i=1,…,N; T i =|r|,i=1,...,N; Where: P i j represents the initial position of the planet for the jth decision variable, T i , e i They represent the corresponding orbital period and eccentricity respectively, r is a randomly generated value between 0 and 1, rand [0,1] is a random vector in the range [0,1], and They represent the upper and lower bounds of the j-th decision variable, N represents the number of candidate solutions in the search space, and d represents the dimension of the problem to be solved; 2) Define the Sun's gravity The main reason why planets orbit the sun is that the sun's gravity keeps them in their orbits. Each planet has its own gravity, which is proportional to the size of the planet. The closer a planet is to the sun, the greater its orbital speed, and vice versa. The sun's gravity is defined as follows: Where: M sn and m pn represents the normalized value of the mass of the sun and planets, ε is a very small value; μ(t) is a function that decays exponentially with time t and is defined to control the search accuracy; r1 represents a randomly generated value between 0 and 1, R in Represents the distance from the sun S to the target planet P i The Euclidean distance between 3) Calculate the speed of the target planet A planet's speed increases as it approaches the sun and decreases as it moves away from the sun to avoid being pulled toward the sun. The speed of the planet is expressed as: Where: V i (t) represents the speed of the planet at time t; P a and P b represents two random positions in the planet population; P i Indicates the position of the planet at the corresponding time; R in (t) represents the sun S and P at time t i The Euclidean distance; r3 and r4 are randomly generated values ​​between 0 and 1, r5 is a vector of random values ​​between 0 and 1, α, β, represents the selection variable after folding; U1 and U2 represent the selection criteria; P i,up , P i,low They represent the upper and lower limits of the planetary positions respectively; 4) Escape from local optimum In the solar system, most objects rotate counterclockwise, they all rotate around their own axis, but some celestial bodies revolve around the sun in a clockwise direction. KOA uses this behavior to change the search direction using the flag γ to escape the local optimal area; 5) Update planetary positions P i (t+1)=P i (t)+γ×V i (t);+(F(t)+|r|)×U×(S(t)-P i (t)) Where: P i (t+1) is the new position of the planet after the update; P i (t) is the position of the planet before the update; S(t) is the optimal position of the sun S so far; r is a number randomly generated based on the normal distribution; γ represents the search direction; F(t) represents the gravity of the sun at time t; U is the selection criterion, which is: embedding the self-correction and generalization representation modules based on the fusion feature index ξ, comparing the self-correction result with the decoder output to improve the accuracy of morphological recognition, and merging the generalization representation with the morphological recognition result to improve the generalization ability of class domain division; 6) Update the distance to the sun Where: P a (t), P b (t) represent two random positions in the planetary population at time t, and h represents the adaptive factor; 7) Elite Strategy After optimization, if the fitness is the best, update the planet position P i '(t+1), otherwise the planet position remains unchanged, f(P i (t+1))、f(P i (t)) represent the planet position fitness at time t+1 and time t, respectively.

Citation Information

Cited By

  • Fault prediction method and system for high-voltage switch operating mechanism

    CN121388955A