GIS temperature field rapid calculation method based on nonlinear dimension reduction method of local linear interpolation and related device

Abstract

The invention belongs to the technical field of power equipment temperature field calculation, and discloses a GIS temperature field rapid calculation method based on a nonlinear dimensionality reduction method of local linear interpolation and a related device, and the method comprises the steps: obtaining preset working condition data of a GIS, inputting the preset working condition data into an agent model, and obtaining a low-dimensional projection vector Yr under the preset working condition data; calculating a homogenized Euclidean distance matrix dr of the low-dimensional projection vector Yr and a low-dimensional projection matrix Y in a training set when the low-dimensional projection vector Yr and the low-dimensional projection matrix Y are used for training the proxy model; arranging each row of dr from small to large, and selecting samples in the low-dimensional projection Y corresponding to the minimum k distance values for numbering; selecting corresponding elements in Y corresponding to sample numbers to form a neighborhood matrix of Yr; selecting high-dimensional temperature samples Xq corresponding to the first k low-dimensional projections in the dr to form a reconstruction matrix; calculating a local reconstruction weight matrix Wr by using the neighborhood matrix; and reconstructing the low-dimensional projection vector Yr to an established high-dimensional space by using the weight matrix Wr and the reconstruction matrix to obtain GIS thermal field distribution under preset working condition data.

Description

Technical Field

[0001] The present invention belongs to the technical field of temperature field calculation of electric power equipment, and in particular relates to a GIS temperature field fast calculation method and related devices based on a nonlinear dimensionality reduction method of local linear interpolation. Background Art

[0002] The core of the digitalization of power equipment is the visualization and evaluability of status data throughout its life cycle. With the increasing demand for digitalization and intelligentization of equipment, the utilization of sensor monitoring data and the guidance of operation and maintenance still require the support of multi-physics field simulation technology. However, the multi-physics field simulation solution is highly nonlinear, and the equipment has a large number of grid nodes in the multi-scale complex structure, resulting in a slow solution speed, which restricts the application of online real-time visual evaluation. Therefore, in order to adapt to the demand for visualization of the characteristic status of equipment in the integration of digital technology and traditional power grids, the original common key technology for real-time simulation of the multi-physics field characteristics of equipment is the key to achieving second-level rapid calculation of single and multi-fields in key parts of the equipment. This will enable the digitalization of power equipment and effectively support the inherent safety of power equipment.

[0003] For example, gas insulated switchgear (GIS) has numerous components and geometric models spanning different scales. When using the finite element method to calculate the distribution of multi-physics fields, the meshing requirements are extremely strict, the number of nodes is huge, and the iterative convergence speed during the multi-physics field coupling simulation process is slow. Even on high-performance cluster computers, the calculation time may be as long as several days. This has become a bottleneck for characteristic visualization and real-time evaluation and analysis in the field of panoramic perception and information fusion of power equipment, and cannot meet the urgent needs of real-time monitoring and feedback of digital twins.

[0004] In summary, in-depth research and development of multi-physical field quantity dimensionality reduction and rapid calculation methods for GIS equipment is of vital importance for improving the accuracy of GIS equipment status monitoring and fault warning, realizing the efficient and highly reliable application of GIS equipment digital twin technology, and supporting the visualization analysis and in-depth mining of perception data throughout the life cycle of GIS equipment. Summary of the Invention

[0005] In order to solve the problems existing in the prior art, the purpose of the present invention is to provide a GIS temperature field fast calculation method and related devices based on a nonlinear dimensionality reduction method of local linear interpolation. The present invention can realize the fast calculation of the GIS temperature field and ensure the accuracy of the calculation results while improving the calculation efficiency.

[0006] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0007] A fast calculation method for GIS temperature field based on nonlinear dimensionality reduction method using local linear interpolation includes:

[0008] Obtaining preset operating condition data of the GIS, wherein the operating condition data includes GIS current load and ambient temperature;

[0009] The preset working condition data is input into the trained proxy model, and the low-dimensional projection vector Y under the preset working condition data is calculated. r ;Wherein, the proxy model is trained using different working condition data of GIS;

[0010] Using the low-dimensional projection vector Y r And the low-dimensional projection matrix Y in the training set used to train the proxy model, calculate the low-dimensional projection vector Y r The normalized Euclidean distance matrix d between the low-dimensional projection matrix Y r ;

[0011] For the normalized Euclidean distance matrix d r Each row of is arranged from small to large, and the normalized Euclidean distance matrix d is selected r The sample number in the low-dimensional projection Y corresponding to the smallest k distance values;

[0012] Select the corresponding elements in the low-dimensional projection Y corresponding to the sample number to form a low-dimensional projection vector Y r Neighborhood matrix of

[0013] Select the normalized Euclidean distance matrix d r The high-dimensional temperature samples X in the acquired temperature matrix D corresponding to the first k low-dimensional projections q , forming a reconstruction matrix; where q=1,2,…,k;

[0014] Using the low-dimensional projection vector Y r Neighborhood matrix, calculate the local reconstruction weight matrix W r ;

[0015] Using the weight matrix W r And the reconstruction matrix, the low-dimensional projection vector Y r Reconstruct into the established high-dimensional space to obtain the GIS thermal field distribution under the preset working condition data.

[0016] Preferably, the training process of the proxy model includes:

[0017] Determine the value ranges of the GIS input variables current load and external ambient temperature;

[0018] The value range of the input variable current load and the external ambient temperature is sampled to obtain the input parameter set P;

[0019] By using simulation software, the GIS thermal field distribution under the input parameter set P is calculated to obtain the temperature matrix D in the high-dimensional space. m×n ={X i |i=1,2,…,n}, where X i The sample point results of the simulation calculation under each working condition. The rows of the matrix represent the number of finite element nodes, the columns represent the number of working conditions, and n is the total number of working conditions.

[0020] Calculate the temperature matrix D m×n Any sample point result X i The neighborhood set of ;

[0021] According to the result X of any sample point i Neighborhood set, calculate the global weight matrix W;

[0022] Calculate the symmetric sparse matrix M using the global weight matrix W;

[0023] Utilizing the symmetric sparse matrix M, calculating the projection matrix Y of the temperature matrix in the high-dimensional space in the low-dimensional space;

[0024] The proxy model is trained with the input parameter set P as input and the projection matrix Y as output to obtain a trained proxy model.

[0025] Preferably, the temperature matrix D is calculated m×n Any sample point result X i The neighborhood set of , including:

[0026] Calculate the temperature matrix D m×n Sample point result X i The normalized Euclidean distance between the sample point and the other sample point results is used to obtain the distance matrix d of the sample point;

[0027] Sort each row of the distance matrix d from small to large;

[0028] Select the number of the sample point in the temperature matrix D corresponding to the smallest k distance values in each row of the distance matrix d;

[0029] The sample composition of each sample point result X in the temperature matrix D corresponding to the number of the sample point i The neighborhood set of .

[0030] Preferably, according to the result X of any sample point i The neighborhood set of , calculates the global weight matrix W, including:

[0031] According to the sample point result X i The Gram matrix is calculated based on the samples in the neighborhood set

[0032] Gbc =(X i -X b )(X i -X c ) T

[0033] Among them, X b , X c is the sample point result X i The sample point results within the neighborhood set of ;

[0034] According to the Gram matrix G i Calculate the sample point result X i The local linear weight matrix w i , the calculation formula is as follows:

[0035]

[0036] Where Γ is a k×1 column vector of all ones;

[0037] Each sample point result X i The corresponding weight matrix w i Fill in the global weight matrix W according to the following rules to obtain the global weight matrix W; the rule is: if the qth working condition is a neighbor of the ith working condition, then W iq =w iq , otherwise W iq =0.

[0038] Preferably, the symmetric sparse matrix M is calculated using the global weight matrix W by the following formula:

[0039] M=(IW) T (IW)

[0040] in, is the identity matrix.

[0041] Preferably, the symmetric sparse matrix M is used to calculate the projection matrix Y of the temperature matrix in the high-dimensional space in the low-dimensional space, including:

[0042] Perform eigenvalue decomposition on the sparse matrix M to obtain the eigenvalue λ and the eigenvector matrix corresponding to the eigenvalue

[0043] Mv=λv

[0044] Where λ is the eigenvalue and v is the eigenvector matrix;

[0045] Arrange the obtained eigenvalues λ from small to large;

[0046] Take the eigenvectors corresponding to the first [2, d+1] eigenvalues and combine them to obtain the temperature matrix D in the high-dimensional space m×n Projection matrix in low-dimensional space

[0047]

[0048] Preferably, the weight matrix W is used as follows: r And the reconstruction matrix, the low-dimensional projection vector Y r Reconstruct into the established high-dimensional space to obtain the GIS thermal field distribution under the preset working condition data:

[0049]

[0050] in, is the GIS thermal field distribution under the reconstructed preset working condition data; W rq The local reconstruction weight matrix is calculated using the neighborhood matrix of the low-dimensional projection vector under the preset working condition data.

[0051] The present invention also provides a GIS temperature field calculation system, comprising:

[0052] Data acquisition module: used to obtain preset operating condition data of GIS, including GIS current load and ambient temperature;

[0053] Calculation module: used to input the preset working condition data into the trained proxy model and calculate the low-dimensional projection vector Y under the preset working condition data r ;Wherein, the proxy model is trained using different working condition data of GIS;

[0054] Distance calculation module: used to use the low-dimensional projection vector Y r And the low-dimensional projection matrix Y in the training set used to train the proxy model, calculate the low-dimensional projection vector Y r The normalized Euclidean distance matrix d between the low-dimensional projection matrix Y r ;

[0055] Sorting module: used to normalize the Euclidean distance matrix d r Each row of is arranged from small to large, and the normalized Euclidean distance matrix d is selected r The sample number in the low-dimensional projection Y corresponding to the smallest k distance values;

[0056] Neighborhood matrix building module: used to select the corresponding elements in the low-dimensional projection Y corresponding to the sample number to form the low-dimensional projection vector Y r Neighborhood matrix of

[0057] Reconstruction matrix building module: used to select the normalized Euclidean distance matrix d r The high-dimensional temperature samples X in the acquired temperature matrix D corresponding to the first k low-dimensional projections q , forming a reconstruction matrix; where q=1,2,…,k;

[0058] Local reconstruction weight matrix calculation module: used to use the low-dimensional projection vector Y r Neighborhood matrix, calculate the local reconstruction weight matrix W r ;

[0059] Reconstruction module: used to use the weight matrix W r And the reconstruction matrix, the low-dimensional projection vector Y r Reconstruct into the established high-dimensional space to obtain the GIS thermal field distribution under the preset working condition data.

[0060] The present invention also provides an electronic device, comprising:

[0061] one or more processors;

[0062] a storage device having one or more programs stored thereon;

[0063] When the one or more programs are executed by the one or more processors, the one or more processors implement the GIS temperature field fast calculation method based on the nonlinear dimensionality reduction method of local linear interpolation as described above in the present invention.

[0064] The present invention also provides a storage medium storing a computer program, wherein when the computer program is executed by a processor, the method for quickly calculating the GIS temperature field based on the nonlinear dimensionality reduction method of local linear interpolation as described above is implemented.

[0065] Compared with the prior art, the present invention has the following beneficial effects:

[0066] The present invention is a method for rapidly calculating the GIS temperature field based on a nonlinear dimensionality reduction method using local linear interpolation. Considering that the dimensions of the neural network input and output parameters should not be too many, which will affect the training accuracy and efficiency, and in order to improve the calculation speed, the sample temperature data is first dimensionalized, and then the temperature field is rapidly calculated using a neural network. This significantly reduces the calculation time of the thermal field while ensuring that the impact on the calculation accuracy is within a certain control range. Furthermore, the present invention's nonlinear dimensionality reduction method based on local linear interpolation proposes a normalized distance method to find the nearest neighbor sample points, which can avoid the problem of insufficient neighbor measurement in the traditional Euclidean distance method. The Euclidean distance is improved and designed to overcome the situation where the uneven distribution of samples causes unreasonable neighborhood construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is a flow chart of the method for quickly calculating the GIS temperature field based on the nonlinear dimensionality reduction method of local linear interpolation of the present invention;

[0068] FIG2( a ) is a temperature distribution cloud diagram of an original sample of GIS equipment according to an embodiment of the present invention;

[0069] FIG2( b ) is a temperature distribution cloud diagram of the GIS equipment after d=10th order dimensionality reduction in an embodiment of the present invention. DETAILED DESCRIPTION

[0070] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0071] See also Figure 1 The present invention provides a fast calculation method for GIS temperature field based on a nonlinear dimensionality reduction method of local linear interpolation, comprising the following steps:

[0072] Step 1: Determine the range of input variables current load and external ambient temperature.

[0073] Step 2: Sample the variable current load and the external ambient temperature to obtain the input parameter set P = {P (i) |i=1,2,…,n}. The input parameter set P has n samples, P (i) is a matrix with q rows and 1 column, the matrix P (i) Each element in represents the parameter value of different current carrying capacity and ambient temperature under different input state parameters, P (i) =[P i1 P i2 …P iq ] T , the subscript q represents the number of different input state parameters.

[0074] Step 3: Use commercial simulation software to calculate the GIS thermal field distribution under the input parameter set P and obtain the temperature matrix D in the high-dimensional space m×n ={X i |i=1,2,…,n}, where X i The sample point results of the simulation calculation under each working condition, the temperature matrix D in the high-dimensional space m×n The rows represent the number of finite element nodes, and the columns represent the number of load cases.

[0075] Step 4.1: Calculate the sample point result X i The normalized Euclidean distance between the sample point results and other sample point results is used to obtain the distance matrix d of the sample point results. The calculation formula of the elements in the distance matrix d is as follows:

[0076]

[0077] Among them, M iis the sample point result X i And the temperature matrix D in high-dimensional space m×n The mean distance between the other sample points in M j is the sample point result X j And the temperature matrix D in high-dimensional space m×n The mean distance between the other sample points in ||X i -X j || is the sample point result X i and the sample point result X j The Euclidean distance between As the denominator, for sparse samples, divide by the larger value; for dense samples, divide by the smaller The uniform distance obtained in this way can achieve the effect of smoothing the sample distribution, thereby overcoming the disadvantage that the dimensionality reduction effect of traditional local nonlinear dimensionality reduction is easily affected by the uneven distribution of samples.

[0078] Step 4.2: Sort each row of the distance matrix d from small to large.

[0079] Step 4.3: Select the temperature matrix D in the high-dimensional space corresponding to the k smallest distance values in each row of the distance matrix d m×n The number of the sample point result in .

[0080] Step 4.3: The temperature matrix D in the high-dimensional space corresponding to the sample point result number m×n The samples in the result X of each sample point i The neighborhood set of .

[0081] Step 5.1: Based on the sample point result X i Samples in the neighborhood set of , calculate the Gram matrix

[0082] G bc =(X i -X b )(X i -X c ) T (2)

[0083] Among them, X b , X c is the sample point result X i The sample point results within the neighborhood set of .

[0084] Step 5.2: According to the Gram matrix G i Calculate the sample point result X i The local linear weight matrix (corresponding to k nearest neighbor points). Local linear weight matrix w i The calculation formula is shown in (7):

[0085]

[0086] where Γ is a k×1 column vector of all ones.

[0087] Step 5.3: Set each sample point result X i The corresponding weight matrix w i Fill in the global weight matrix according to the following rules Get the global weight matrix W: If the qth working condition is a neighbor of the i-th working condition, then W iq =w iq , otherwise W iq =0.

[0088] Step 6.1: Calculate the symmetric sparse matrix according to formula (4)

[0089] M=(IW) T (IW) (4)

[0090] in, is the identity matrix.

[0091] Step 6.2: Perform eigenvalue decomposition on the symmetric sparse matrix M to obtain the eigenvalue λ and the eigenvector matrix corresponding to the eigenvalue

[0092] Mv=λv (5)

[0093] Where λ is the eigenvalue and v is the eigenvector matrix.

[0094] Step 6.3: Arrange the obtained eigenvalues λ from small to large.

[0095] Step 6.4: Take the eigenvectors corresponding to the first [2, d+1] eigenvalues λ and combine them to obtain the sample set (i.e. the temperature matrix D in the high-dimensional space) m×n ) is the projection matrix in low-dimensional space

[0096]

[0097] Step 7.1: Construct a surrogate model using machine learning algorithms such as LSTM, BP, and RBF neural networks, or regression models such as Kriging and response surface surrogate models. The surrogate model's input training set is the input parameter set P, and its output training set is the low-dimensional projection matrix Y in the low-dimensional space under different input parameters. Train these neural networks or regression models to obtain a trained surrogate model.

[0098] Step 7.2: Input the n-parameter P containing the specific current load and ambient temperature parameters into the trained proxy model. (r) (i.e. preset working condition data), the low-dimensional projection vector Y under this parameter combination (i.e. preset working condition data) can be quickly obtained r .

[0099] Step 8.1: Calculate the low-dimensional projection vector Y according to formula (1) r The normalized Euclidean distance matrix d between the low-dimensional projection vector Y in the training set r .

[0100] Step 8.2: Normalize the Euclidean distance matrix d r Each row is arranged from small to large.

[0101] Step 8.3: Get d r The sample number in the low-dimensional projection vector Y corresponding to the smallest k distance values in .

[0102] Step 8.4: Select the corresponding element in the low-dimensional projection Y corresponding to the sample number in the low-dimensional projection vector Y to form the low-dimensional projection vector Y r The neighborhood matrix of .

[0103] Step 8.5: Select the normalized Euclidean distance matrix d r The temperature matrix D of the high-dimensional space corresponding to the first k low-dimensional projections m×n High-dimensional temperature samples X in q (q=1,2,…,k), forming the reconstruction matrix.

[0104] Step 8.6: Calculate Y according to formula (2) r Gram matrix G of the neighborhood matrix r

[0105] Step 8.7: Calculate the local reconstruction weight matrix W according to formula (3) r ={W rq |q=1,2,…,k}.

[0106] Step 8.8: According to formula (7), restore and reconstruct the low-dimensional temperature column vector to the original high-dimensional data to obtain the reconstructed data (i.e. GIS thermal field distribution under preset working condition data).

[0107]

[0108] Example

[0109] See also Figure 1 The rapid calculation method of the thermal field of GIS equipment in this embodiment includes the following steps:

[0110] Step 1: Establish a thermal field simulation model for GIS equipment, divide the mesh of the thermal field simulation model into sections, and set the thermal conductivity parameters of the materials required for the thermal field simulation model analysis and calculation of the GIS equipment; the thermal conductivity parameters include thermal conductivity coefficient and convection heat transfer coefficient;

[0111] Step 2.1: Determine that the input variable current load is 3000A and the ambient temperature range is 15℃-40℃.

[0112] Step 2.2: Sample 40 sets of data for the variable current load and external ambient temperature to obtain the input parameter set P = {P (i) |i=1,2,…,n}.

[0113] Step 2.3: Use commercial software to simulate and calculate the three-dimensional thermal field distribution of the GIS busbar air chamber under 40 working conditions, obtain the temperature values at 23993 grid nodes (dimension m = 23923), and construct the sample set D m×40 ={X i |i=1,2,…,40},X i =[x1 x2…x m ], x i Represents the temperature of each grid node.

[0114] Step 3.1: Calculate the sample point X i The normalized Euclidean distance between the sample points and other sample points is used to obtain the distance matrix d of the sample points. The calculation process of the elements in d adopts formula (1);

[0115] Step 3.2: Sort each row of the distance matrix d from small to large.

[0116] Step 3.3: Get the temperature matrix D corresponding to the 8 smallest distance values in each row of the distance matrix d m×n Sample number

[0117] Step 3.3: Temperature matrix D corresponding to the sample number m×n The sample points in the sample point X i The neighborhood set of .

[0118] Step 4.1: According to X i The samples in the neighborhood set of , calculate the Gram matrix by formula (2)

[0119] Step 4.2: According to the Gram matrix G i Calculate X i The local linear weight matrix w i ={w iq |q=1,2,…,8}(corresponding to 8 neighboring points). Local linear weight matrix wi The calculation formula is shown in formula (3). In this embodiment, Γ is an 8×1 column vector of all 1s.

[0120] Step 4.3: Place each X i The corresponding w i Fill in the global weight matrix according to the following rules If the qth working condition is a neighbor of the ith working condition, then W iq =w iq , otherwise W iq =0.

[0121] Step 5.1: Calculate the symmetric sparse matrix according to the above formula (4) Where I is a 40×40 identity matrix.

[0122] Step 5.2: Perform eigenvalue decomposition on the matrix M to obtain the eigenvalue λ and the eigenvector matrix v corresponding to the eigenvalue = {v i |i=1,2,…,40}.

[0123] Step 5.3: Arrange the obtained eigenvalues λ from small to large.

[0124] Step 5.4: Take the eigenvectors corresponding to the first [2, 10+1] eigenvalues and combine them to obtain the projection matrix of the sample set in the low-dimensional space

[0125]

[0126] Step 6.1: Use an RBF neural network to construct a proxy model. The model input training set is the input parameter set P, and the output training set is the projection matrix Y in the low-dimensional space under different input parameters. Train the neural network and regression model to obtain a trained proxy model.

[0127] Step 6.2: Based on the trained proxy model, the input current is 3000A and the ambient temperature is 15°C. The low-dimensional projection vector Y under this parameter combination can be quickly obtained. r .

[0128] Step 7.1: Calculate the low-dimensional projection vector Y according to formula (1) r The normalized Euclidean distance matrix d between the low-dimensional projection Y in the training set r .

[0129] Step 7.2: Distance matrix d r Arrange from smallest to largest.

[0130] Step 7.3: Get d r The sample number in the low-dimensional projection Y corresponding to the 8 smallest distance values.

[0131] Step 7.3: Construct a low-dimensional projection vector Y based on the elements in the low-dimensional projection Y corresponding to the sample number r The neighborhood matrix of .

[0132] Step 7.4: Choose the distance matrix d r The high-dimensional temperature samples X corresponding to the first 8 low-dimensional projections q (q=1,2,…,8) constitute the reconstruction matrix.

[0133] Step 7.5: Calculate Y according to the above formula (2) r The Gram matrix G r

[0134] Step 7.6: Calculate the local reconstruction weight matrix W according to the above formula (3) r ={W rq |q=1,2,…,8}.

[0135] Step 7.7: According to the above formula (7), the low-dimensional temperature column vector is restored and reconstructed to the original high dimension to obtain the reconstructed data

[0136] The original sample temperature cloud of the GIS equipment temperature field (shown in Figure 2(a)) and the thermal field distribution after nonlinear dimensionality reduction using local linear interpolation (shown in Figure 2(b)) show that the temperature field distribution characteristics of the sample data and the results calculated after d=10-order dimensionality reduction are essentially the same as those of the full-order model, with the maximum values occurring at the same location on the lower busbar of the GIS. However, the maximum values of the sample data and the results calculated after d=10-order truncation and dimensionality reduction differ slightly, at 52.0079°C and 50.5858°C, respectively, a difference of approximately 1.60°C. The small error between the full-order model and the reduced-order model (fast calculation) demonstrates the rationality and accuracy of the proposed dimensionality reduction method.

[0137] In addition, an embodiment of the present invention further provides a system for implementing the GIS temperature field fast calculation method based on the nonlinear dimensionality reduction method of local linear interpolation of the present invention, the system comprising:

[0138] Data acquisition module: used to obtain preset operating condition data of GIS, including GIS current load and ambient temperature;

[0139] Calculation module: used to input the preset working condition data into the trained proxy model and calculate the low-dimensional projection vector Y under the preset working condition data r ;Wherein, the proxy model is trained using different working condition data of GIS;

[0140] Distance calculation module: used to use the low-dimensional projection vector Y rAnd the low-dimensional projection matrix Y in the training set used to train the proxy model, calculate the low-dimensional projection vector Y r The normalized Euclidean distance matrix d between the low-dimensional projection matrix Y r ;

[0141] Sorting module: used to normalize the Euclidean distance matrix d r Each row of is arranged from small to large, and the normalized Euclidean distance matrix d is selected r The sample number in the low-dimensional projection Y corresponding to the smallest k distance values;

[0142] Neighborhood matrix building module: used to select the corresponding elements in the low-dimensional projection Y corresponding to the sample number to form the low-dimensional projection vector Y r Neighborhood matrix of

[0143] Reconstruction matrix building module: used to select the normalized Euclidean distance matrix d r The high-dimensional temperature samples X in the acquired temperature matrix D corresponding to the first k low-dimensional projections q , forming a reconstruction matrix; where q=1,2,…,k;

[0144] Local reconstruction weight matrix calculation module: used to use the low-dimensional projection vector Y r Neighborhood matrix, calculate the local reconstruction weight matrix W r ;

[0145] Reconstruction module: used to use the weight matrix W r And the reconstruction matrix, the low-dimensional projection vector Y r Reconstruct into the established high-dimensional space to obtain the GIS thermal field distribution under the preset working condition data.

[0146] The embodiments of the present invention also provide corresponding electronic devices and computer-readable storage media for implementing the solutions provided by the embodiments of the present invention.

[0147] In which, the device includes a memory and a processor, the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device executes the GIS temperature field fast calculation method based on the nonlinear dimensionality reduction method of local linear interpolation described in any embodiment of the present application.

[0148] The storage medium stores a computer program, wherein when the computer program is executed by the processor, the method for quickly calculating the GIS temperature field based on the nonlinear dimensionality reduction method of local linear interpolation described in any embodiment of the present application is implemented.

[0149] Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0150] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A fast calculation method for GIS temperature field based on a nonlinear dimensionality reduction method using local linear interpolation is characterized by: include: Obtaining preset operating condition data of the GIS, wherein the operating condition data includes GIS current load and ambient temperature; The preset working condition data is input into the trained proxy model, and the low-dimensional projection vector Y under the preset working condition data is calculated. r ;Wherein, the proxy model is trained using different working condition data of GIS; Using the low-dimensional projection vector Y r And the low-dimensional projection matrix Y in the training set used to train the proxy model, calculate the low-dimensional projection vector Y r The normalized Euclidean distance matrix d between the low-dimensional projection matrix Y r ; For the normalized Euclidean distance matrix d r Each row of is arranged from small to large, and the normalized Euclidean distance matrix d is selected r The sample number in the low-dimensional projection Y corresponding to the smallest k distance values; Select the corresponding elements in the low-dimensional projection Y corresponding to the sample number to form a low-dimensional projection vector Y r Neighborhood matrix of Select the normalized Euclidean distance matrix d r The high-dimensional temperature samples X in the acquired temperature matrix D corresponding to the first k low-dimensional projections q , forming a reconstruction matrix; where q=1,2,…,k; Using the low-dimensional projection vector Y r Neighborhood matrix, calculate the local reconstruction weight matrix W r ; Using the weight matrix W r And the reconstruction matrix, the low-dimensional projection vector Y r Reconstruct into the established high-dimensional space to obtain the GIS thermal field distribution under the preset working condition data.

2. The GIS temperature field calculation method according to claim 1, characterized in that: The training process of the proxy model includes: Determine the value ranges of the GIS input variables current load and external ambient temperature; The value range of the input variable current load and the external ambient temperature is sampled to obtain the input parameter set P; By using simulation software, the GIS thermal field distribution under the input parameter set P is calculated to obtain the temperature matrix D in the high-dimensional space. m×n ={X i |i=1,2,…,n}, where X i The sample point results of the simulation calculation under each working condition. The rows of the matrix represent the number of finite element nodes, the columns represent the number of working conditions, and n is the total number of working conditions. Calculate the temperature matrix D m×n Any sample point result X i The neighborhood set of ; According to the result X of any sample point i Neighborhood set, calculate the global weight matrix W; Calculate the symmetric sparse matrix M using the global weight matrix W; Utilizing the symmetric sparse matrix M, calculating the projection matrix Y of the temperature matrix in the high-dimensional space in the low-dimensional space; The proxy model is trained with the input parameter set P as input and the projection matrix Y as output to obtain a trained proxy model.

3. The GIS temperature field calculation method according to claim 2, characterized in that: Calculate the temperature matrix D m×n Any sample point result X i The neighborhood set of , including: Calculate the temperature matrix D m×n Sample point result X i The normalized Euclidean distance between the sample point and the other sample point results is used to obtain the distance matrix d of the sample point; Sort each row of the distance matrix d from small to large; Select the number of the sample point in the temperature matrix D corresponding to the smallest k distance values in each row of the distance matrix d; The sample composition of each sample point result X in the temperature matrix D corresponding to the number of the sample point i The neighborhood set of .

4. The GIS temperature field calculation method according to claim 2, characterized in that: According to the result X of any sample point i The neighborhood set of , calculates the global weight matrix W, including: According to the sample point result X i The Gram matrix G is calculated by the samples in the neighborhood set of i : G bc =(X i -X b )(X i -X c ) T Among them, X b , X c is the sample point result X i The sample point results within the neighborhood set of ; According to the Gram matrix G i Calculate the sample point result X i The local linear weight matrix w i , the calculation formula is as follows: Where Γ is a k×1 column vector of all ones; Each sample point result X i The corresponding weight matrix w i Fill in the global weight matrix W according to the following rules to obtain the global weight matrix W; the rule is: if the qth working condition is a neighbor of the ith working condition, then W iq =w iq , otherwise W iq =0.

5. The GIS temperature field fast calculation method based on the nonlinear dimensionality reduction method of local linear interpolation according to claim 2 is characterized in that: The symmetric sparse matrix M is calculated using the global weight matrix W by the following formula: M=(I-W) T (I-W) in, is the identity matrix.

6. The method for rapid calculation of GIS temperature field based on the nonlinear dimensionality reduction method of local linear interpolation according to claim 2 is characterized in that: Using the symmetric sparse matrix M, calculating the projection matrix Y of the temperature matrix in the high-dimensional space in the low-dimensional space includes: Perform eigenvalue decomposition on the sparse matrix M to obtain the eigenvalue λ and the eigenvector matrix corresponding to the eigenvalue Mv=λv Where λ is the eigenvalue and v is the eigenvector matrix; Arrange the obtained eigenvalues λ from small to large; Take the eigenvectors corresponding to the first [2, d+1] eigenvalues and combine them to obtain the temperature matrix D in the high-dimensional space m×n Projection matrix in low-dimensional space 7. The method for rapidly calculating GIS temperature field based on the nonlinear dimensionality reduction method using local linear interpolation according to claim 2 is characterized in that: By using the following formula, the weight matrix W r And the reconstruction matrix, the low-dimensional projection vector Y r Reconstruct into the established high-dimensional space to obtain the GIS thermal field distribution under the preset working condition data: in, is the GIS thermal field distribution under the reconstructed preset working condition data; W rq The local reconstruction weight matrix is calculated using the neighborhood matrix of the low-dimensional projection vector under the preset working condition data.

8. GIS temperature field calculation system, characterized by: include: Data acquisition module: used to obtain preset operating condition data of GIS, including GIS current load and ambient temperature; Calculation module: used to input the preset working condition data into the trained proxy model and calculate the low-dimensional projection vector Y under the preset working condition data r ;Wherein, the proxy model is trained using different working condition data of GIS; Distance calculation module: used to use the low-dimensional projection vector Y r And the low-dimensional projection matrix Y in the training set used to train the proxy model, calculate the low-dimensional projection vector Y r The normalized Euclidean distance matrix d between the low-dimensional projection matrix Y r ; Sorting module: used to normalize the Euclidean distance matrix d r Each row of is arranged from small to large, and the normalized Euclidean distance matrix d is selected r The sample number in the low-dimensional projection Y corresponding to the smallest k distance values; Neighborhood matrix building module: used to select the corresponding elements in the low-dimensional projection Y corresponding to the sample number to form the low-dimensional projection vector Y r Neighborhood matrix of Reconstruction matrix building module: used to select the normalized Euclidean distance matrix d r The high-dimensional temperature samples X in the acquired temperature matrix D corresponding to the first k low-dimensional projections q , forming a reconstruction matrix; where q=1,2,…,k; Local reconstruction weight matrix calculation module: used to use the low-dimensional projection vector Y r Neighborhood matrix, calculate the local reconstruction weight matrix W r ; Reconstruction module: used to use the weight matrix W r And the reconstruction matrix, the low-dimensional projection vector Y r Reconstruct into the established high-dimensional space to obtain the GIS thermal field distribution under the preset working condition data.

9. An electronic device, characterized in that: include: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors implement the GIS temperature field fast calculation method based on the nonlinear dimensionality reduction method of local linear interpolation as described in any one of claims 1 to 7.

10. A storage medium, characterized in that: A computer program is stored thereon, wherein when the computer program is executed by a processor, the method for quickly calculating the GIS temperature field using the nonlinear dimensionality reduction method based on local linear interpolation as claimed in any one of claims 1 to 7 is implemented.

Images