Transformer electromagnetic-mechanical information fusion method based on fuzzy support vector machine

Abstract

A transformer electromagnetic-mechanical information fusion method based on a fuzzy support vector machine belongs to the technical field of transformers, and comprises the following steps: dividing an electromagnetic subsystem and a mechanical subsystem of a transformer, extracting characteristic parameters of each subsystem, forming a characteristic parameter vector set, analyzing a characteristic parameter corresponding relation and a change condition, and forming a global multi-source information base; extracting current and vibration information of a transformer unbalanced operation virtual / physical entity by using a fuzzy support vector machine; and obtaining fusion feature indexes, dividing state grades, constructing a transformer winding state quantitative evaluation and fuzzy clustering system, and realizing information physical fusion of a virtual / physical entity and an electromagnetic-mechanical subsystem.

Description

Technical Field

[0001] The present invention belongs to the technical field of transformers, and in particular relates to a method based on the application of electromagnetic-mechanical physical information fusion of power transformers. Background Art

[0002] Power transformers are core components of power grids, and their operating status is crucial to the safety and stability of the power system. Existing statistics and transformer accident analysis indicate that mechanical winding failure is a major cause of transformer failure. Therefore, identifying and evaluating the operating status of transformer windings is of great research value and engineering significance. Using observable electrical parameters to describe otherwise observable physical characteristics is key to transformer winding operating status monitoring and identification.

[0003] Therefore, the prior art urgently needs a new solution to solve the above problems. Summary of the Invention

[0004] The technical problem to be solved by the present invention is: to provide a transformer electromagnetic-mechanical information fusion method based on fuzzy support vector machine, establish a transformer virtual / physical entity, construct electromagnetic and mechanical subsystems, and based on the electromagnetic-mechanical coupling principle, link the subsystems through key physical information parameters to achieve information-physical fusion of virtual / physical entities and electromagnetic-mechanical subsystems.

[0005] The transformer electromagnetic-mechanical information fusion method based on fuzzy support vector machine includes the following steps, which are performed in sequence:

[0006] Step 1: Transformer system division

[0007] The transformer is divided into electromagnetic subsystem and mechanical subsystem, and a virtual simulation model is established to obtain the multi-source characteristic information of the transformer. The electromagnetic subsystem information includes port voltage u, winding current i, harmonic THD, winding leakage magnetic B σ The mechanical subsystem information includes the winding electromagnetic force F, winding vibration g, Young's modulus and Poisson's ratio. A global multi-source information library of transformer windings is established with time points as indexes.

[0008] Step 2: Information Fusion

[0009] Fuzzy support vector machine and FSVM clustering method are used to fuse the virtual / physical data of electromagnetic subsystem and mechanical subsystem, and the characteristic indicators of transformer winding status are integrated and quantitatively evaluated to assess the operating status of transformer winding.

[0010] In the step 2, the change characteristics of the electromagnetic subsystem are represented by the winding current, and the change characteristics of the mechanical subsystem are represented by the vibration acceleration.

[0011] The method of fusing the fuzzy support vector machine and the FSVM clustering method in step 2 is:

[0012] The current information is used as the feature vector X and the vibration information is used as the feature vector Y to form the sample set x. The fuzzy membership of the sample set is calculated to obtain the fuzzy training set. is the cluster center, r is the cluster radius, and the degree of membership is determined based on the distance;

[0013]

[0014] Where n is the number of samples, μ is the sample membership; δ is a minimum positive real number to ensure that the membership is not zero; S is the fuzzy training set, and i is the random sample label;

[0015] Introducing the radial basis kernel function K(x i , x j ) Map the training points to the high-order space, solve the fuzzy optimal classification function by quadratic programming, solve the k-classification problem based on one-to-many combination, and obtain k binary classifiers.

[0016]

[0017] The kernel function is used to calculate the class center distance and quantitatively evaluate the classification results of different sample spaces. i (x).

[0018]

[0019]

[0020] In the formula, C is the penalty coefficient, ω and b are the normal vector and offset of the optimal classification hyperplane respectively, α i is the Lagrange multiplier (α i ≥0);

[0021] Normalize the sample space feature information:

[0022]

[0023] Calculate the fusion feature index:

[0024]

[0025] Where r is the fusion weight;

[0026] By integrating characteristic indicators and comprehensively considering the changes in the electromagnetic-mechanical characteristics of the transformer winding, the state subspace mapped by each sample space can be clustered:

[0027] class(x)=argmax{f1(x),f2(x),…,f k(x)}

[0028] Where class(x) is the state space clustering function and arg max is the feature parameter multi-classification functional.

[0029] Through the above-mentioned design scheme, the present invention can bring the following beneficial effects: a transformer electromagnetic-mechanical information fusion method based on a fuzzy support vector machine is used to divide the transformer electromagnetic subsystem and mechanical subsystem, extract the characteristic parameters of each subsystem, form a characteristic parameter vector set, analyze the corresponding relationship and change of the characteristic parameters, and form a global multi-source information database; use a fuzzy support vector machine to extract the current and vibration information of the virtual / physical entity of the transformer's unbalanced operation; obtain fusion characteristic indicators, divide the state levels, construct a transformer winding state quantitative evaluation and fuzzy clustering system, and realize the information-physical fusion of virtual / physical entities and electromagnetic-mechanical subsystems. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The present invention will be further described below with reference to the accompanying drawings and specific embodiments:

[0031] Figure 1 This is a schematic diagram of the transformer electromagnetic-mechanical information fusion method based on fuzzy support vector machine of the present invention.

[0032] Figure 2 This is a schematic diagram of the electromagnetic-mechanical coupling principle of the transformer electromagnetic-mechanical information fusion method based on the fuzzy support vector machine of the present invention.

[0033] Figure 3 This is a flow chart of the transformer electromagnetic-mechanical information fusion method based on fuzzy support vector machine of the present invention. DETAILED DESCRIPTION

[0034] Transformer electromagnetic-mechanical information fusion method based on fuzzy support vector machine, such as Figure 1 As shown in Figure 1, electromagnetic-mechanical cyber-physical fusion includes electromagnetic / mechanical information fusion and virtual (virtual simulation) and real (physical monitoring) data fusion. The transformer is divided into electromagnetic subsystems and mechanical subsystems to obtain multi-source and multi-feature information of the transformer, where the transformer winding electromagnetic subsystem information includes port voltage u, winding current i, harmonic THD, winding leakage magnetic B σ The mechanical subsystem includes time-varying parameters such as the winding electromagnetic force F and winding vibration g, as well as constant material parameters such as Young's modulus and Poisson's ratio. The electromagnetic force (torque) of the transformer winding is generated by electromagnetic coupling and influences its vibration characteristics. Therefore, the electromagnetic force is used as a link to correlate winding current and vibration. A fuzzy support vector machine (SVM) is used to integrate electromagnetic and mechanical virtual / physical data to quantitatively evaluate the winding operating status and perform fuzzy clustering.

[0035] Among them, the electromagnetic subsystem information acquisition method is:

[0036] Taking the three-phase three-column dry-type experimental transformer (SSG-15kVA2kV / 1kV) as an example, a dynamic model experimental platform was built to collect the electromagnetic subsystem parameters under different working conditions of the transformer.

[0037] The electromagnetic subsystem information collection process is as follows:

[0038] 1) Set different operating conditions of the transformer and connect PT and CT sensors to monitor the voltage and current parameters of the electromagnetic subsystem;

[0039] 2) Debug the sensor and current monitoring module to collect transformer port voltage and current information in real time;

[0040] 3) Filter the voltage and current to remove noise and interference information, and dynamically store them using the operating status and time point as labels.

[0041] Maxwell / Ansys was used to construct a proportional finite element virtual entity. Some physical parameters of the transformer are shown in Table 1. The virtual entity was divided into a magnetic field connectivity domain, a current excitation domain, and a mechanical stress domain, corresponding to the core, winding, and other components, maintaining spatial consistency with the physical entity.

[0042] Table 1 Three-phase transformer parameters

[0043]

[0044]

[0045] Note: Positive Young's modulus and Poisson's ratio are obtained through material tension and compression tests.

[0046] In the current-carrying excitation domain, a circular current excitation is applied to the winding, and the entire magnetic field connection domain is subject to the parallel external boundary condition of the magnetic field lines. The rest are natural boundary conditions, that is, it is assumed that the magnetic field lines inside the transformer box are connected and closed, and do not diffuse outward through the box.

[0047] The mathematical model and solution of the transformer electromagnetic subsystem are realized through electromagnetic coupling. By digital modeling and simulation of the electromagnetic subsystem, the evolution process of the multi-source physical information of the transformer winding is described at different time and space scales, and a multi-source information database is constructed at the time scale. The specific principle is shown in Figure 2 The main steps are as follows:

[0048] 1) Electromagnetic coupling modeling and simulation. The global multi-source information database uses time as a link tag to retrieve t in the electromagnetic sub-information database. k Electromagnetic information at time t k The winding current is used as the excitation to set the current-carrying excitation domain and the magnetic field connection domain, solve the electromagnetic spatiotemporal distribution of the winding and obtain the electromagnetic field characteristic parameters, such as the current i k , magnetic flux leakage Bσk wait.

[0049] 2) Information acquisition iterative process criterion. If the absolute convergence norm is less than the convergence criterion value, or the coupling cycle reaches the set number of times, the iteration ends. Links are established based on the time point index, and the calculation results are stored in the global multi-source information library. Otherwise, the coupling parameter i k+1 Input the magnetic field model and solve the magnetic field at the next moment.

[0050] Assume that the information collection process of the mechanical subsystem is as follows:

[0051] 1) Set up different operating conditions for the transformer and arrange vibration measurement points. Connect the piezoelectric accelerometer and debug the vibration monitoring module to monitor the mechanical subsystem parameters.

[0052] 2) Collect transformer winding vibration information in real time, filter the vibration signal to remove noise and interference signals, and store it with operating status and time point as labels.

[0053] Virtual Simulation

[0054] The material parameters corresponding to the virtual model were then set using the data in Table 1. The mechanical stress domain, based on the specific structure of the experimental transformer, applied a preload using a constant uniform load and constrained the winding base to rigidly connect the three degrees of freedom. The transformer's mechanical subsystem was solved using sequential electromagnetic-mechanical coupling [ The specific steps are as follows:

[0055] 1) Using time as index, search t in the global multi-source information database k The electromagnetic information at each moment is used to obtain the temporal and spatial distribution of the electromagnetic force of the winding and to calculate the electromagnetic force at t k Dynamic storage for tags.

[0056] 2) With t k At the moment, the electromagnetic force is used as the excitation to set the mechanical stress domain, solve the winding vibration distribution and obtain the mechanical field characteristic parameter vibration acceleration g. Then enter the next moment g k+1 Calculation.

[0057] 3) Combine electromagnetic and mechanical information using time points as indexes to form multi-source information.

[0058] Specifically, the fusion algorithm adopted in the present invention is:

[0059] In the data mining process of electromagnetic-mechanical multi-source information of transformer windings, the implicit attributes of the sample space and its nonlinear mapping relationship with the state space are relatively complex. Fuzzy Support Vector Machine (FSVM) is used to solve this problem. This algorithm has great advantages in solving multi-attribute samples, nonlinear problems and fuzzy clustering. It is suitable for feature fusion and state assessment of transformer winding current and vibration signals. The current information is used as the feature vector X and the vibration information is used as the feature vector Y to form the sample set x. The fuzzy membership of the sample set is calculated to obtain the fuzzy training set. Assume is the cluster center, r is the cluster radius, and the degree of membership is determined based on the distance.

[0060]

[0061] Where n is the number of samples, μ is the sample membership; δ is a minimum positive real number to ensure that the membership is not zero; S is the fuzzy training set.

[0062] The multi-source heterogeneous characteristics of electromagnetic-mechanical virtual / physical entity data make it difficult to divide the sample set in the original space, so the radial basis kernel function K(x i , x j ) Map the training points to the high-order space

[17] , the fuzzy optimal classification function is obtained by solving the quadratic programming, based on the one-to-many combination idea

[18] Solve the k-classification problem and obtain k binary classifiers.

[0063]

[0064] The kernel function is used to calculate the class center distance and quantitatively evaluate the classification results of different sample spaces. i (x).

[0065]

[0066] Where C is the penalty coefficient, ω and b are the normal vector and offset of the optimal classification hyperplane respectively, and α i is the Lagrange multiplier (α i ≥0).

[0067] Normalize the sample space feature information:

[0068]

[0069] Calculate the fusion feature index:

[0070]

[0071] Where r is the fusion weight.

[0072] By integrating characteristic indicators and comprehensively considering the changes in the electromagnetic-mechanical characteristics of the transformer winding, the state subspace mapped by each sample space can be clustered:

[0073] class(x)=argmax{f1(x),f2(x),…,f k (x)} (6)

[0074] Where class(x) is the state space clustering function and argmax is the feature parameter multi-classification functional.

[0075] Fusion Solution

[0076] By extracting electromagnetic-mechanical information characteristic parameters, a global multi-source characteristic information database is formed, and the information-physical fusion is performed by combining the FSVM clustering method. The fusion process diagram is shown in the following figure. Figure 3 shown.

[0077] The basic process of cyber-physical fusion is as follows:

[0078] 1) Experimentally monitor transformer electrical information (winding current) and vibration information (winding vibration). Based on virtual entity electromagnetic coupling, the magnetic field connectivity domain and current-carrying excitation domain are calculated to obtain winding current, magnetic flux leakage, and electromagnetic force results. Virtual vibration information is obtained by calculating the mechanical stress domain through sequential electromagnetic-mechanical coupling. Based on this, a global multi-source information database of transformer windings is constructed, indexed by time points.

[0079] 2) The changing characteristics of the electromagnetic subsystem are characterized by winding current, and the changing characteristics of the mechanical subsystem are characterized by vibration acceleration. Combining virtual and real data information, fusion characteristic indicators are calculated and quantitatively evaluated.

[0080] 3) Obtain the time-domain current signal characteristic parameters and the time-frequency domain characteristic parameters of the winding vibration signal, expand the sample feature set, and form a multi-source feature information library.

[0081] 4) Set the training set and sample set, use the FSVM clustering method to fuse the transformer electromagnetic-mechanical virtual / physical entity data, and evaluate the operating status of the transformer winding.

[0082] The calculation conditions, legends, etc. in the embodiments of the present invention are only used to further illustrate the present invention and are not exhaustive and do not constitute a limitation on the scope of protection of the claims. Those skilled in the art can conceive of other substantially equivalent alternatives based on the inspiration gained from the examples of the present invention without creative work, and all of them are within the scope of protection of the present invention.

Claims

1. A transformer electromagnetic-mechanical information fusion method based on fuzzy support vector machine is characterized by: The method comprises the following steps, and the following steps are performed in sequence: Step 1: Transformer system division The transformer is divided into electromagnetic subsystem and mechanical subsystem, and a virtual simulation model is established to obtain the multi-source characteristic information of the transformer. The electromagnetic subsystem information includes port voltage u, winding current i, harmonic THD, winding leakage magnetic B σ The mechanical subsystem information includes the winding electromagnetic force F, winding vibration g, Young's modulus and Poisson's ratio. A global multi-source information library of transformer windings is established with time points as indexes. Step 2: Information Fusion Fuzzy support vector machine and FSVM clustering method are used to fuse the virtual / physical data of electromagnetic subsystem and mechanical subsystem, and the characteristic indicators of transformer winding status are integrated and quantitatively evaluated to assess the operating status of transformer winding.

2. The transformer electromagnetic-mechanical information fusion method based on fuzzy support vector machine according to claim 1 is characterized by: In the step 2, the change characteristics of the electromagnetic subsystem are represented by the winding current, and the change characteristics of the mechanical subsystem are represented by the vibration acceleration.

3. The transformer electromagnetic-mechanical information fusion method based on fuzzy support vector machine according to claim 1 is characterized by: The method of fusing the fuzzy support vector machine and the FSVM clustering method in step 2 is: The current information is used as the feature vector X and the vibration information is used as the feature vector Y to form the sample set x. The fuzzy membership of the sample set is calculated to obtain the fuzzy training set. is the cluster center, r is the cluster radius, and the degree of membership is determined based on the distance; Where n is the number of samples, μ is the sample membership; δ is a minimum positive real number to ensure that the membership is not zero; S is the fuzzy training set, and i is the random sample label; Introducing the radial basis kernel function K(x i , x j ) Map the training points to the high-order space, solve the fuzzy optimal classification function by quadratic programming, solve the k-classification problem based on one-to-many combination, and obtain k binary classifiers. The kernel function is used to calculate the class center distance and quantitatively evaluate the classification results of different sample spaces f i (x). In the formula, C is the penalty coefficient, ω and b are the normal vector and offset of the optimal classification hyperplane respectively, α i is the Lagrange multiplier (α i ≥0); Normalize the sample space feature information: Calculate the fusion feature index: Where r is the fusion weight; By integrating characteristic indicators and comprehensively considering the changes in the electromagnetic-mechanical characteristics of the transformer winding, the state subspace mapped by each sample space is clustered: class(x)=argmax{f1(x),f2(x),…,f k (x)} Where class(x) is the state space clustering function and argmax is the feature parameter multi-classification functional.

Images