Self-adaptive migration method for digital twinborn model of main power equipment of intelligent substation

Abstract

The invention discloses an adaptive migration method for a digital twinborn model of main power equipment of an intelligent substation. The adaptive migration method comprises the following steps: S1, acquiring a historical visible light image of the equipment; s2, preprocessing the historical visible light image, and constructing a training data set; s3, constructing an initial diagnosis model; s4, training the initial diagnosis model through the training data set to obtain an appearance abnormality diagnosis model; s5, acquiring a real-time visible light image; s6, inputting the real-time visible light image into the appearance abnormality diagnosis model to obtain a diagnosis result; and S7, when the severity of the abnormal state is increased, taking the original power main equipment digital twin model as a source template M0, carrying out appearance mapping adjustment based on the current real-time visible light image on the basis of M0, and obtaining a digital twin model Mn under a new working condition. According to the invention, continuous monitoring and prediction of equipment can be realized, so that the reliability and safety of a power system can be improved.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of smart substations, and specifically to a method for adaptively migrating a digital twin model of a smart substation power main equipment. Background Art

[0002] As the scale of power grids continues to expand, the number of primary and secondary equipment in smart substations continues to increase, resulting in a continuous increase in the workload of substation operation and maintenance, and insufficient manpower to support business needs. Substation equipment has accumulated a large amount of historical data during the operation and maintenance process, but most of the data is still in a "dormant state", failing to effectively tap its value and difficult to use it to guide production operations. As a rapidly developing digital information technology, digital twins use three-dimensional models and high-density dynamic data uploaded by sensors to achieve the mapping of substation equipment in virtual digital space, build digital twins that "run synchronously" with real equipment, and have functional advantages in digital scenarios, such as prediction and deduction, decision verification, etc., opening up new ways to promote equipment status perception, degradation trend analysis, and defect handling decision-making assistance.

[0003] However, the digital twin models used in most substations currently lack the ability to adapt to natural appearance changes such as equipment aging, corrosion, and damage, which results in the inability of operation and maintenance personnel to timely and intuitively grasp the status of substation equipment operation and management and assess the possibility of abnormal events. Therefore, developing an adaptive migration method for the digital twin model of the main power equipment in smart substations is of great significance for improving the operational reliability of digital twin substations and promoting lean operation and maintenance management and maintenance decision-making of substation equipment. Summary of the invention

[0004] In view of the technical problems existing in the prior art, the present invention provides a method for adaptively migrating a digital twin model of a smart substation power main equipment to achieve continuous monitoring and prediction of the equipment.

[0005] In order to solve the above technical problems, the technical solution proposed by the present invention is:

[0006] A method for adaptively migrating a digital twin model of a main power equipment in a smart substation comprises the following steps:

[0007] S1. Obtain historical visible light image data information of the main power equipment;

[0008] S2. Preprocess the historical visible light image data information obtained in step S1, classify them according to the abnormal state of the appearance of the main power equipment, and divide them into levels according to the severity, then make training labels corresponding to the abnormal state and construct a corresponding training data set;

[0009] S3, based on deep reinforcement learning convolutional neural network, build the initial diagnosis model of power main equipment;

[0010] S4, using the training data set obtained in step S2 as the input variable of the initial diagnosis model, performing model training, and obtaining a diagnosis model for abnormal appearance of the main power equipment;

[0011] S5. Obtaining real-time visible light image data information of the current power main equipment;

[0012] S6, inputting the real-time visible light image data information of the current power master device obtained in step S5 into the power master device appearance abnormality diagnosis model obtained in step S4, and obtaining the diagnosis result of the power master device appearance abnormality and severity;

[0013] S7. Based on the diagnosis result of step S6, when the severity of the abnormal state increases, the original digital twin model of the main power equipment is used as the source template M0. On the basis of the M0 template and based on the real-time visible light image data information of the current main power equipment, the appearance mapping is adjusted based on the three-dimensional point cloud model to obtain the digital twin model Mn under the new working condition. Finally, the digital twin model Mn is stored in the model library and imported into the replacement source template M0.

[0014] Preferably, the specific process of step S7 is:

[0015] S71, retrieving the three-dimensional point cloud model data from the database, performing image-point cloud registration on the three-dimensional point cloud model, and obtaining a point cloud-pixel index; after completing the point cloud-pixel index registration, performing image correction on the three-dimensional point cloud data, and generating an orthophoto image A as a reference image;

[0016] S72, by extracting the same-name control points in the three-dimensional point cloud and the two-dimensional image, solving the projection matrix M representing the index;

[0017] S73, orthographically projecting the three-dimensional point cloud of the substation equipment, projecting the three-dimensional laser point cloud onto a plane to generate a two-dimensional data form, and generating an orthographic projection image B of the point cloud elevation according to the projection matrix;

[0018] S74. According to the orthophoto image A, the point cloud elevation orthophoto projection image B and the projection matrix M, the pixel values ​​are assigned to the three-dimensional point clouds at their corresponding positions to obtain the infrared feature three-dimensional point clouds, and the digital twin model Mn under the new working conditions is obtained. Finally, the model is stored in the model library and imported to replace the original model.

[0019] Preferably, the specific process of obtaining the projection matrix in step S72 is:

[0020] Define S as the two-dimensional feature coordinates, u and v as the pixel plane coordinate system coordinates, and t(x, y, s) as the three-dimensional coordinates. The final conversion coefficient is solved by the conversion parameter s. x 、s yRepresents the conversion coefficients of the x and y axes in two dimensions, u0 and v0 represent the center points of two-dimensional pixels, and the projection matrix M is expressed as:

[0021]

[0022] in:

[0023]

[0024] Where: r1, r2, r3 and m1, m2, m3 are the intermediate conversion matrices of three dimensions; m 14 、m 24 、m 34 Indicates the conversion coefficient corresponding to m1, m2, and m3.

[0025] Preferably, the initial diagnostic model in step S3 includes an image preprocessing layer and an image diagnosis layer connected in series in sequence; the image preprocessing layer is constructed based on a scale-invariant feature conversion algorithm to extract feature points from the image data; the image diagnosis layer is constructed based on an enhanced deep convolutional neural network to further extract features from the preprocessed image, and the abnormal state of the appearance of the main power equipment is diagnosed according to the features of the image data to obtain an image diagnosis result.

[0026] Preferably, the image preprocessing layer is constructed by a scale-invariant feature conversion algorithm, comprising the following steps:

[0027] Gaussian convolution is used to construct a multi-scale space L(x,y,σ):

[0028] L(x,y,σ)=G(x,y,σ)*I(x,y)

[0029] Where L(x,y,σ) is defined as the scale space of the image, I(x,y) represents the pixel value of the image at the coordinate (x,y), G(x,y,σ) is a Gaussian function of varying scale, represents a Gaussian kernel centered at (x,y), σ is a scale parameter, and * is a convolution operation;

[0030] After obtaining the multi-scale space, a Gaussian difference pyramid is established, and the adjacent scale spaces are subtracted to obtain the Gaussian difference pyramid D(x, y, σ), that is:

[0031] D(x,y,σ)=L(x,y,kσ)-L(x,y,σ)

[0032] Where k is the multiple of the two scales.

[0033] Preferably, when extracting feature points from image data, the gradient directional features of the neighborhood pixels are used to obtain the directional parameters of the feature points, including the modulus m(x, y) and the phase angle θ(x, y), and the calculation formula is as follows:

[0034]

[0035] Preferably, the abnormal state of the appearance of the main power equipment in step S2 includes rust, stains and damage.

[0036] The present invention also discloses a computer program product, comprising a computer program, wherein the computer program executes the steps of the method described above when executed by a processor.

[0037] The present invention further discloses a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described above are executed.

[0038] The present invention also discloses a computer device, comprising a memory and a processor connected to each other, wherein a computer program is stored in the memory, and when the computer program is run by the processor, the steps of the method described above are executed.

[0039] Compared with the prior art, the advantages of the present invention are:

[0040] The present invention realizes the automatic diagnosis of abnormal appearance of power main equipment through the convolutional neural network model of deep reinforcement learning, reduces the need for manual analysis, and improves the efficiency and accuracy of diagnosis; by combining the new appearance map adjustment on the basis of the original digital twin model M0, a digital twin model Mn under the new working condition is generated, such a model can more accurately reflect the current state of the equipment, and provide a more accurate digital copy for the monitoring and maintenance of the equipment; the new digital twin model Mn is stored in the model library and replaces the original model, ensuring that the data in the model library is always the latest, which helps to realize the continuous monitoring and prediction of the equipment. Through continuous model updating, more accurate prediction and decision support can be provided for the operation of the power system, thereby improving the reliability and safety of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 The flowchart of the adaptive migration method of the present invention is shown in an embodiment.

[0042] Figure 2 It is a basic structural diagram of the enhanced deep convolutional neural network (EDCNN) of the present invention. DETAILED DESCRIPTION

[0043] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0044] like Figure 1 As shown, the method for adaptively migrating the digital twin model of the main power equipment of the smart substation according to the embodiment of the present invention comprises the following steps:

[0045] S1. Obtain historical visible light image data information of the main power equipment;

[0046] S2. Preprocess the visible light image data information obtained in step S1, classify the abnormal conditions such as rust, stains, and damage on the appearance of the main power equipment, and divide the levels according to the severity; then make training labels corresponding to the abnormal conditions and construct corresponding training data sets;

[0047] S3, based on deep reinforcement learning convolutional neural network, build the initial diagnosis model of power main equipment;

[0048] S4, using the training data obtained in step S2 as input variables of the initial diagnosis model, performing model training, and obtaining a diagnosis model for abnormal appearance of the main power equipment;

[0049] S5. Acquire real-time visible light image data information of the current power main device through cameras around the power main device;

[0050] S6, inputting the real-time visible light image data information of the current power main device obtained in step S5 into the power main device appearance abnormality diagnosis model obtained in step S4, and obtaining the diagnosis result of the power main device appearance abnormality (rust, stain, damage) and severity;

[0051] S7. Based on the diagnosis result of step S6, when the severity of the abnormal state increases, the original digital twin model of the main power equipment is used as the source template M0. On the basis of the M0 template and based on the real-time visible light image data information of the current main power equipment, the appearance mapping is adjusted based on the three-dimensional point cloud model to obtain the digital twin model Mn under the new working condition. Finally, the model is stored in the model library and imported to replace the original model.

[0052] In step S1, historical visible light image data information of the main power equipment of the built substation is mainly obtained, specifically including image information of abnormal working conditions such as rust, stains, and damage of the main power equipment.

[0053] In step S2, the visible light image data information in S1 is classified according to the abnormal operating conditions such as rust, stains, and damage of the main power equipment, and divided into levels according to the severity; the visible light image data information in S1 is preprocessed, and feature points of the image data are extracted and matched based on the scale-invariant feature conversion algorithm, and then training labels corresponding to the abnormal operating conditions are produced, and the corresponding training data set is constructed.

[0054] In step S3, the initial diagnosis model includes an image preprocessing layer and an image diagnosis layer connected in series. The image preprocessing layer is constructed based on the scale-invariant feature conversion algorithm to extract feature points from the image data; the image diagnosis layer is constructed based on the enhanced deep convolutional neural network to further extract features from the preprocessed image, and the abnormal state of the appearance of the main power equipment is diagnosed according to the image data features to obtain the image diagnosis result.

[0055] The image preprocessing layer is constructed by the scale-invariant feature conversion algorithm, including the following steps:

[0056] Gaussian convolution is used to construct a multi-scale space L(x,y,σ):

[0057] L(x,y,σ)=G(x,y,σ)*I(x,y)

[0058] Among them, L(x,y,σ) is defined as the scale space of the image, I(x,y) represents the pixel value of the image at the coordinate (x,y), G(x,y,σ) is a Gaussian function of varying scale, represents a Gaussian kernel centered at (x,y), σ is a scale parameter, and * is a convolution operation.

[0059] After obtaining the multi-scale space, a Gaussian difference pyramid is established, and the adjacent scale spaces are subtracted to obtain the Gaussian difference pyramid D(x, y, σ), that is:

[0060] D(x,y,σ)=L(x,y,kσ)-L(x,y,σ)

[0061] Where k is the multiple of the two scales.

[0062] When extracting feature points from image data, the extracted feature points only contain position and scale information, but lack direction information. Therefore, the gradient direction features of the neighborhood pixels are used to obtain the direction parameters of the feature points, including the modulus m(x, y) and the phase angle θ(x, y). The calculation is as follows:

[0063]

[0064] The enhanced deep convolutional neural network (EDCNN) has the ability to mine memory and nonlinear features when processing sequence data to build an initial diagnosis model for power equipment. The basic structure of EDCNN is as follows: Figure 2 shown.

[0065] Figure 2 Middle U t0-1 , U t0 and U t0+1 is the final prediction output result before t0, at t0 and after t0; L j is the loss function; s t0-1 、st0 and t0+1 is the predicted output result of the hidden layer before t0, at t0 and after t0; I t0-1 ,I t0 and I t0+1 The data are input before t0, at t0 and after t0 respectively.

[0066] In the process of propagating forward over time, EDCNN can be used and solved at a certain moment, the formula is as follows:

[0067] s t0 =λ1(μI t0 +ωs t0-1 +α)

[0068] Where: λ1 is the hyperbolic tangent activation function; μ, ω and α are network parameters.

[0069] The predicted output result H of the hidden layer at t0 t0 And the final prediction output result U t0 The calculation is as follows:

[0070] H t0 =β·s t0 +γ

[0071] U t0 =λ2(H t0 )

[0072] Where: λ2 is the Softmax activation function; β and γ are network parameters.

[0073] At time t0, the loss function L t0 The target output u can be used t0 and U t0 It can be expressed as the sample mean square error between:

[0074]

[0075] The specific steps of step S7 are as follows:

[0076] S71, retrieve the 3D point cloud model data from the database, use the 2D and 3D feature calibration method to perform image-point cloud registration on the 3D point cloud model to obtain a point cloud-pixel index; correct the original image of the 3D point cloud section to generate an orthophoto A as a reference image;

[0077] S72. By extracting the control points with the same name in the 3D point cloud and the 2D image, solve the projection matrix M representing the index. Define S as the 2D feature coordinates, u and v as the coordinates of the pixel plane coordinate system, t(x, y, s) as the 3D coordinates, and solve the final conversion coefficient through the conversion parameter s, s x 、sy Represents the conversion coefficients of the x and y axes in two dimensions, u0 and v0 represent the center points of two-dimensional pixels, and the matrix M expression is:

[0078]

[0079] in:

[0080]

[0081] Where: r1, r2, r3 and m1, m2, m3 are the intermediate conversion matrices of three dimensions; m 14 、m 24 、m 34 Indicates the conversion coefficient corresponding to m1, m2, and m3.

[0082] The three-dimensional point cloud-pixel index conversion matrix can be obtained from the above formula.

[0083] S73, ortho-project the three-dimensional point cloud of the substation equipment, project the three-dimensional laser point cloud onto a plane to generate a two-dimensional data format, each point cloud data is represented by (X, Y, Z) coordinates, projected onto the plane to generate the corresponding pixel point (u, v) to generate the point cloud elevation ortho-projection image B, and establish the point cloud pixel index, image A is used as a floating image, and the 3D-2D registration problem is converted into a 2D-2D registration problem (where A is the ortho-image of the image, and B is the ortho-projection image of the point cloud);

[0084] S74. According to the point cloud-pixel index, the pixel value is assigned to the three-dimensional point cloud at the corresponding position to obtain the infrared feature three-dimensional point cloud for visualization. The digital twin model Mn under the new working condition is obtained by using the three-dimensional point cloud visualization model, and finally the model is stored in the model library and imported to replace the original model.

[0085] The present invention analyzes the historical visible light image data information of the main power equipment, divides the levels according to the degree of abnormality of the appearance of the main power equipment, constructs a corresponding training data set, and constructs an initial diagnosis model based on a convolutional neural network model of deep reinforcement learning. The training data set is trained to obtain a diagnosis model for the abnormal appearance of the main power equipment, and the degree of abnormality of the appearance of the power equipment is judged. The original digital twin model of the main power equipment is used as the source template M0. On the basis of the M0 template, the appearance mapping is adjusted based on the three-dimensional point cloud model to obtain the digital twin model Mn under the new working condition. Finally, the model is stored in the model library and imported to replace the original model.

[0086] The present invention realizes the automatic diagnosis of abnormal appearance of power main equipment through the convolutional neural network model of deep reinforcement learning, reduces the need for manual analysis, and improves the efficiency and accuracy of diagnosis; by combining the new appearance map adjustment on the basis of the original digital twin model M0, a digital twin model Mn under the new working condition is generated, such a model can more accurately reflect the current state of the equipment, and provide a more accurate digital copy for the monitoring and maintenance of the equipment; the new digital twin model Mn is stored in the model library and replaces the original model, ensuring that the data in the model library is always the latest, which helps to realize the continuous monitoring and prediction of the equipment. Through continuous model updating, more accurate prediction and decision support can be provided for the operation of the power system, thereby improving the reliability and safety of the power system.

[0087] The present invention also discloses a computer program product, including a computer program, which executes the steps of the above method when executed by a processor. The present invention further discloses a computer-readable storage medium, which stores a computer program, which executes the steps of the above method when executed by a processor. The present invention also discloses a computer device, including a memory and a processor connected to each other, the memory stores a computer program, and the computer program executes the steps of the above method when executed by the processor. The product, medium and device of the present invention correspond to the above method and also have the advantages described in the above method.

[0088] The present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by hardware related to computer program instructions. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned method embodiment can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. Computer-readable storage media include: any entity or device that can carry computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. The memory is used to store computer programs and / or modules. The processor implements various functions by running or executing computer programs and / or modules stored in the memory, and calling data stored in the memory. The memory may include high-speed random access memory and may also include non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0089] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A method for adaptive migration of digital twin models of power main equipment in smart substations, characterized in that: Includes steps: S1. Obtain historical visible light image data information of the main power equipment; S2. Preprocess the historical visible light image data information obtained in step S1, classify them according to the abnormal state of the appearance of the main power equipment, and divide them into levels according to the severity, then make training labels corresponding to the abnormal state and construct a corresponding training data set; S3, based on deep reinforcement learning convolutional neural network, build the initial diagnosis model of power main equipment; S4, using the training data set obtained in step S2 as the input variable of the initial diagnosis model, performing model training, and obtaining a diagnosis model for abnormal appearance of the main power equipment; S5. Obtaining real-time visible light image data information of the current power main equipment; S6, inputting the real-time visible light image data information of the current power master device obtained in step S5 into the power master device appearance abnormality diagnosis model obtained in step S4, and obtaining the diagnosis result of the power master device appearance abnormality and severity; S7. Based on the diagnosis result of step S6, when the severity of the abnormal state increases, the original digital twin model of the main power equipment is used as the source template M0. On the basis of the M0 template and based on the real-time visible light image data information of the current main power equipment, the appearance mapping is adjusted based on the three-dimensional point cloud model to obtain the digital twin model Mn under the new working condition. Finally, the digital twin model Mn is stored in the model library and imported into the replacement source template M0.

2. The method for adaptive migration of the digital twin model of the main power equipment of the smart substation according to claim 1 is characterized in that: The specific process of step S7 is: S71, retrieving the three-dimensional point cloud model data from the database, performing image-point cloud registration on the three-dimensional point cloud model, and obtaining a point cloud-pixel index; after completing the point cloud-pixel index registration, performing image correction on the three-dimensional point cloud data, and generating an orthophoto image A as a reference image; S72, by extracting the same-name control points in the three-dimensional point cloud and the two-dimensional image, solving the projection matrix M representing the index; S73, orthographically projecting the three-dimensional point cloud of the substation equipment, projecting the three-dimensional laser point cloud onto a plane to generate a two-dimensional data form, and generating an orthographic projection image B of the point cloud elevation according to the projection matrix; S74. According to the orthophoto image A, the point cloud elevation orthophoto projection image B and the projection matrix M, the pixel values ​​are assigned to the three-dimensional point clouds at their corresponding positions to obtain the infrared feature three-dimensional point clouds, and the digital twin model Mn under the new working conditions is obtained. Finally, the model is stored in the model library and imported to replace the original model.

3. The method for adaptive migration of the digital twin model of the main power equipment of the smart substation according to claim 2 is characterized in that: The specific process of obtaining the projection matrix in step S72 is: Define S as the two-dimensional feature coordinates; u and v are the coordinates of the pixel plane coordinate system; t(x, y, s) is the three-dimensional coordinate, and the final conversion coefficient is solved by the conversion parameter s; s x 、s y Represents the conversion coefficients of the x and y axes in two dimensions; u0 and v0 represent the center points of two-dimensional pixels, then the projection matrix M is expressed as: in: Where: r1, r2, r3 and m1, m2, m3 are the intermediate conversion matrices of three dimensions; m 14 、m 24 、m 34 Indicates the conversion coefficient corresponding to m1, m2, and m3.

4. The method for adaptive migration of the digital twin model of the main power equipment of the smart substation according to claim 1, 2 or 3, characterized in that: The initial diagnosis model in step S3 includes an image preprocessing layer and an image diagnosis layer connected in series; An image preprocessing layer is constructed based on the scale-invariant feature conversion algorithm to extract feature points of image data. An image diagnosis layer is constructed based on the enhanced deep convolutional neural network to further extract features from the preprocessed image. According to the features of the image data, the abnormal state of the appearance of the main power equipment is diagnosed to obtain the image diagnosis results.

5. The method for adaptively migrating the digital twin model of the main power equipment of the smart substation according to claim 4 is characterized in that: The image preprocessing layer is constructed by the scale-invariant feature conversion algorithm, including the following steps: Gaussian convolution is used to construct a multi-scale space L(x,y,σ): L(x,y,σ)=G(x,y,σ)*I(x,y) Where L(x,y,σ) is defined as the scale space of the image, I(x,y) represents the pixel value of the image at the coordinate (x,y), G(x,y,σ) is a Gaussian function of varying scale, represents a Gaussian kernel centered at (x,y), σ is a scale parameter, and * is a convolution operation; After obtaining the multi-scale space, a Gaussian difference pyramid is established, and the adjacent scale spaces are subtracted to obtain the Gaussian difference pyramid D(x, y, σ), that is: D(x,y,σ)=L(x,y,kσ)-L(x,y,σ) In the formula, k is the multiple relationship between the two scales.

6. The method for adaptive migration of digital twin models of power main equipment of smart substation according to claim 4 is characterized in that: When extracting feature points from image data, the gradient directional features of the neighborhood pixels are used to obtain the directional parameters of the feature points, including the modulus m(x, y) and the phase angle θ(x, y). The calculation formula is as follows:

7. The method for adaptively migrating the digital twin model of the main power equipment of the smart substation according to claim 1, 2 or 3, characterized in that: The abnormal state of the appearance of the power master device in step S2 includes rust, stains and damage.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are performed.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the computer program performs the steps of the method according to any one of claims 1 to 7.

10. A computer device comprising a memory and a processor connected to each other, wherein a computer program is stored in the memory, characterized in that: When the computer program is executed by a processor, the computer program performs the steps of the method according to any one of claims 1 to 7.

Images