Real-time reconstruction method, system and storage medium for cable fire resistance test temperature with multi-element heat output

Abstract

The present invention discloses a real-time reconstruction method for cable fire resistance test temperature for multi-element heat output, comprising the following steps: Step 1: Collecting temperature data: 11) Using a combination of grey correlation and time series response to determine the temperature collection points in a fire resistance detection device; 12) For the temperature collection points selected in the fire resistance detection device, collecting the temperature data of the blowtorch at each temperature collection point, and obtaining the temperature series data of each temperature collection point; Step 2: Temperature prediction: 21) Using the temperature series data of each temperature collection point as the input of a sliding time window, obtaining the temperature series data of m steps ahead; 22) Inputting the temperature series data output by the sliding time window into a graph neural network GCN, and using Fourier convolution operation to extract features from the temperature series data; 23) Inputting the features extracted by the graph neural network GCN into a recursive neural network SRU to obtain the cable combustion temperature data.

Description

Technical Field

[0001] The present invention belongs to the technical field of cable fire resistance performance detection, and specifically relates to a method, system and storage medium for real-time reconstruction of cable fire resistance performance test temperature for multiple heat output. Background Art

[0002] Determining the fire resistance of cables is of great significance for ensuring building fire safety and reducing secondary injuries at the fire scene. The key to testing the fire resistance of cables is to use the national standard heating curve to heat the cable fire resistance, and then test the power-on state of the cable under heating conditions. In order to ensure the objectivity and correctness of the test results during the cable fire resistance test, the heating temperature of the cable fire resistance during the test process must be consistent with the standard heating curve. Among them, in order to ensure that the temperature of the cable fire resistance heating process is consistent with the standard heating curve, it is necessary to accurately obtain the temperature inside the fire resistance testing equipment.

[0003] For the temperature detection in the fire resistance performance detection device, the traditional method mostly adopts the hardware direct test method. There are two main types of temperature test methods based on hardware, namely the thermocouple method and the infrared image method. In terms of thermocouples, the Chinese patent application with publication number CN117487987A discloses a blast furnace iron mouth temperature detection and erosion alarm device, which can intuitively obtain the heating rate at each iron tapping and the average temperature of the iron tapping over a period of time. The Chinese patent application with publication number CN116256073A discloses a thermocouple structure, a longitudinal temperature detection device and a longitudinal diffusion furnace, which accurately measures the temperature of the reaction chambers at different longitudinal positions set in the longitudinal diffusion furnace through the longitudinal arrangement structure. The Chinese patent with publication number CN214372951U discloses a temperature measurement device for a thermocouple calibration furnace, including a temperature detection device body and a shell. The above direct measurement method based on thermocouple sensors can reflect the internal temperature of the equipment in real time to a certain extent, and can be used for the temperature detection task in the cable fire resistance performance detection process. However, thermocouples can usually only obtain the temperature of a specific point and cannot be used to reflect the change state of the overall temperature field. However, in the task of cable fire resistance performance testing, it is necessary to obtain all the temperature states on the cable laying line in order to obtain a control plan.

[0004] On this basis, it is proposed to use visual devices such as infrared thermal imagers to obtain changes in the temperature field. For example, the Chinese patent application with publication number CN105004447A discloses a visual-based temperature monitoring system; the Chinese design patent with publication number CN307071131S discloses an infrared temperature monitoring alarm; the Chinese patent with publication number CN112362164B discloses a temperature monitoring method, device, electronic device and storage medium for a device. Although the above-mentioned method based on infrared thermal imagers can obtain changes in the temperature field, it is often difficult to apply to temperature detection in the detection furnace. Due to safety conditions and other constraints, it is necessary to use a certain device to cover the detection furnace to prevent high temperature hazards. Therefore, the method based on visual images is difficult to apply to combustion detection of fire resistance detection furnaces.

[0005] In order to realize the temperature monitoring inside the furnace, the researchers proposed an indirect modeling method of the temperature field based on the available data. In this regard, the Chinese patent application with publication number CN113420429A discloses a temperature data modeling method based on the data modeling method, which obtains the internal temperature of the equipment in the machine room by collecting data of the powered environment, the external fixed environment and the temperature control equipment; the Chinese patent application with publication number CN117824875A discloses a temperature monitoring method for heating equipment, which uses CAE simulation to achieve modeling of the temperature field of the internal equipment. Although there are a large number of temperature field modeling methods, the existing modeling methods are still limited in applicability to the temperature detection in the furnace for fire resistance performance testing. The main reasons are as follows:

[0006] (1) The temperature source in the fire resistance performance testing device is mainly obtained by the combustion of multiple fixed torches. Compared with the above models, its temperature field has more data. In addition, the temperature in the fire resistance performance testing device is mostly affected by the input time series flow, and its temperature change is more complicated. The existing technology usually lacks research on multi-input response and is difficult to be directly used for the task of modeling the temperature field in the fire resistance performance testing device.

[0007] (2) The temperature calculation in the fire resistance performance detection device is large. When the existing technology is modeling the temperature, the impact of temperature data is mostly based on a simple neural network model or simulation. However, the temperature detection of the fire resistance performance detection device involves a lot of calculations, and the traditional temperature field modeling method with a large number of parameters is often difficult to apply to the rapid response of the temperature field in the furnace. For this reason, it is difficult to apply it to the combustion task of the temperature field in the furnace. Summary of the invention

[0008] In view of this, the purpose of the present invention is to provide a real-time reconstruction method, system and storage medium for cable fire resistance test temperature with multi-element heat output, which can quickly reconstruct the temperature field in the cable furnace based on the blowtorch flow rate, obtain the real-time change of the cable fire resistance test temperature at key positions, and realize accurate and rapid prediction of the temperature field in the furnace.

[0009] In order to achieve the above object, the present invention provides the following technical solutions:

[0010] The present invention first proposes a real-time reconstruction method for cable fire resistance test temperature for multi-element heat output, comprising the following steps:

[0011] Step 1: Collect temperature data

[0012] 11) The temperature collection points in the fire resistance performance testing device are determined by combining grey correlation with time series response;

[0013] 12) For the temperature collection points selected in the fire resistance performance testing device, the temperature data of the torch at each temperature collection point is collected to obtain the temperature series data of each temperature collection point;

[0014] Step 2: Temperature prediction

[0015] 21) Taking the temperature series data of each temperature collection point as the input of the sliding time window, obtaining the temperature series data of m steps ahead;

[0016] 22) Input the temperature series data output by the sliding time window into the graph neural network GCN, and use Fourier convolution operation to extract features of the temperature series data;

[0017] 23) The features extracted by the graph neural network GCN are input into the recurrent neural network SRU to obtain the cable combustion temperature data.

[0018] Furthermore, in step 11), the method for determining the temperature collection point in the fire resistance performance detection device is:

[0019] Construct the temperature time series change expression:

[0020] f(t)=f(t0)+f[t0,t1](t―t0)+…+f[t0,…,t n ](t―t0)…(t―t n―1 )+f[t,t0,…,t n ]+(t―t0)…(t―t n―1 )(t―t n )

[0021] Where: t0, t1, …, t n―1 and t nis the temperature data collection time; f(.) is the difference remainder calculation function constructed based on grey correlation;

[0022] Constructing a correlation model for fire resistance performance testing devices based on grey correlation:

[0023]

[0024] Where: γ(y,x i ) is the correlation value between the temperature measurement point and the cable combustion temperature; y is the difference remainder data; x i represents the combustion temperature value obtained by the i-th temperature measurement sensor, where i = 1, 2, 3, ..., N, N is the upper limit number of temperature sensors arranged; y(t) is the temperature value obtained at the t-th moment; x i (k) is the kth temperature value obtained at the i-th temperature measurement point; y(k) represents the temperature output value; is the absolute difference between the ith temperature data and the cable combustion temperature data at the temperature point in the fire resistance test device; based on the absolute difference Construct the difference matrix:

[0025]

[0026] Where: Δ min and Δ max That is, the minimum and maximum values ​​in the difference matrix; ρ is the resolution coefficient;

[0027] Based on the obtained association value γ(y,x i ), select the temperature collection points corresponding to the key parameters that can characterize the changes in the temperature field.

[0028] Further, in step 21), the temperature sequence data consisting of the first n steps of temperature data collected by the ith temperature collection point is:

[0029]

[0030] Use the sliding time window to get the temperature series data of m steps ahead:

[0031]

[0032] in: The temperature data of the first j steps collected at the i-th temperature collection point; for The temperature data is obtained after the sliding time window slides forward m steps.

[0033] Furthermore, the mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) are used to evaluate the performance of the sliding time window:

[0034]

[0035]

[0036] Where: x(t) is the measured temperature value, is the temperature value predicted by the model; m is the number of time window steps passed.

[0037] Furthermore, in step 22), the method of using graph neural network GCN to extract features from temperature series data is:

[0038] The image expression of temperature series data is defined as G = (V, E, W); where G is the graph structure, V is the node set, E represents the edge set, and W represents the adjacency transformation matrix; and:

[0039]

[0040] Where: w ij Represents the connection weight relationship between node i and node j;

[0041] Get the degree matrix D based on the constructed adjacency transformation matrix:

[0042]

[0043] Where: wx k is the element value in the adjacency matrix; D ii is the ith temperature data of the ith temperature collection point;

[0044] Construct the Laplace transform matrix of the measured temperature:

[0045]

[0046] Where: L represents the Laplacian matrix of the graph neural network; is the normalized Laplace matrix; U is the eigenvector of the normalized Laplace matrix; D is the degree matrix; Λ is the diagonal matrix of eigenvalues; and:

[0047] Λ=diag([λ0,λ1,...,λ N ])

[0048] Where: λ0,λ1,...,λ N are the values ​​on the diagonal of the diagonal matrix;

[0049] Use Laplace transformation to perform feature transposition and input:

[0050] X·H=F ―1 (F(X)⊙F(H))=U((UT X)⊙(U T H))

[0051] Where: X represents the information signal matrix in the node matrix V; H is the convolution kernel matrix in the graph neural network GCN; ⊙ represents the Hadamard product; F ―1 and F denotes Fourier transform and inverse Fourier transform.

[0052] Furthermore, in the process of training the graph neural network GCN, an improved loss function matrix based on the lifting ratio is adopted:

[0053]

[0054] Where: P MAE , P RMSE and P MAPE Respectively represent the mean absolute error coefficient, the root mean square error coefficient, and the prediction accuracy error; MAE i 、RMSE i and MAPE i are the mean absolute error, root mean square error and error coefficient involved in the i-th Epoch respectively; MAE1, RMSE1 and MAPE1 are the mean absolute error, root mean square error and average precision respectively.

[0055] Further, in step 23), the principle of the recursive neural network SRU is:

[0056]

[0057] Where: H is the cable combustion temperature data output from the fire resistance performance detection device; C is the state matrix inside the model; t represents the time; G is the activation function; F t For the Gate of Oblivion; is the input gate.

[0058] Furthermore, it is characterized in that: it also includes step 3: lightweight deployment: detecting invalid neurons in the model by traversing the loop;

[0059] Perform a presence or absence test on each neuron to obtain the sensitivity of the neuron to the final result:

[0060]

[0061] Where: N i,0 With N i,1 They represent the output corresponding to the original state matrix model when the i-th neuron weight parameter is set to 0; μ j is the information evaluation matrix constructed based on MAE, RMSE, and MAPE; ε j is the sensitivity coefficient; Nj is the total number of neurons in a layer; N i Indicates the number of neural network layers;

[0062] The importance of neurons in a certain layer is determined by the sensitivity coefficient, and the neural network is pruned by deleting neurons with low importance.

[0063] The present invention also proposes a real-time reconstruction system for cable fire resistance test temperature with multiple heat outputs, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the method described above when executing the program.

[0064] The present invention also proposes a storage medium, wherein a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0065] The beneficial effects of the present invention are:

[0066] The present invention provides a real-time reconstruction method for cable fire resistance test temperature with multi-element heat output. First, the grey correlation is combined with the time series response. Based on the grey correlation, the time series response is integrated. Then, the structure of the fire resistance detection device is considered to realize the accurate selection of the temperature collection point of each blowtorch of the fire resistance detection device, which can be beneficial to realize the accurate prediction of the cable combustion temperature; then, the sliding time window is used to obtain the temperature data of m steps ahead to ensure the coherent input and prediction of the effective temperature data; finally, the graph neural network GCN is used, and the Fourier convolution operation is used to extract the features of the temperature time series data, and the cable combustion temperature data is output through the recursive neural network SRU; in summary, the present invention provides a real-time reconstruction method for cable fire resistance test temperature with multi-element heat output. The temperature field in the cable furnace can be quickly reconstructed based on the blowtorch flow rate, the real-time change of the cable fire resistance detection temperature at the key position can be obtained, and the accurate and rapid prediction of the temperature field in the furnace can be realized.

[0067] In the real-time reconstruction method of cable fire resistance test temperature with multi-element heat output of the present invention, lightweight deployment is used to characterize the performance value in the model pruning process, and reverse iterative parameter optimization is performed. By traversing and searching, invalid or less influential neurons in the model are reduced, thereby realizing iterative optimization of the model and facilitating the embedded implementation of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to make the purpose, technical solution and beneficial effects of the present invention clearer, the present invention provides the following drawings for illustration:

[0069] Figure 1It is a flow chart of the real-time reconstruction method of cable fire resistance test temperature for multiple heat output of the present invention;

[0070] Figure 2 This is a diagram showing the location of the temperature sensors in the fire resistance testing device;

[0071] Figure 3 Hardware components involved in temperature collection;

[0072] Figure 4 It is the data acquisition interface of the temperature acquisition system;

[0073] Figure 5 This is the installation condition diagram for on-site testing of the fire resistance performance testing device;

[0074] Figure 6 The overall framework for temperature prediction;

[0075] Figure 7 This is the schematic diagram of the sliding time window;

[0076] Figure 8 The structural diagram of the graph neural network GCN and the recurrent neural network SRU;

[0077] Fig. 9 It is the structural diagram of the recursive neural network SRU;

[0078] Fig.10 A schematic diagram of the principle of pruning a neural network. DETAILED DESCRIPTION

[0079] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.

[0080] like Figure 1 As shown, the real-time reconstruction method of cable fire resistance test temperature for multi-element heat output in this embodiment includes the following steps:

[0081] Step 1: Collect temperature data

[0082] 11) The temperature collection points in the fire resistance performance testing device are determined by combining grey correlation and time series response.

[0083] Specifically, the method for determining the temperature collection point in the fire resistance performance testing device is:

[0084] Construct the temperature time series change expression:

[0085] f(t)=f(t0)+f[t0,t1](t―t0)+…+f[t0,…,t n](t―t0)…(t―t n―1 )+f[t,t0,…,t n ]+(t―t0)…(t―t n―1 )(t―t n )

[0086] Where: t0, t1, …, t n―1 and t n is the time of temperature data collection; f(.) is the difference remainder calculation function constructed based on grey correlation.

[0087] Constructing a correlation model for fire resistance performance testing devices based on grey correlation:

[0088]

[0089] Where: γ(y,x i ) is the correlation value between the temperature measurement point and the cable combustion temperature; y is the difference remainder data; x i represents the combustion temperature value obtained by the i-th temperature measurement sensor, where i = 1, 2, 3, ..., N, N is the upper limit number of temperature sensors arranged; y(t) is the temperature value obtained at the t-th moment; x i (k) is the kth temperature value obtained at the i-th temperature measurement point; y(k) represents the temperature output; It is the absolute difference between the i-th temperature data and the cable combustion temperature data at the temperature point in the fire resistance performance testing device.

[0090] Based on absolute difference Construct the difference matrix:

[0091]

[0092] Where: Δ min and Δ max That is, the minimum and maximum values ​​in the difference matrix; ρ is the resolution coefficient;

[0093] Based on the obtained association value γ(y,x i ), considering the number of input parameters, we can select key parameters that can characterize the change of temperature field, and then select the temperature collection points corresponding to the key parameters that can characterize the change of temperature field, so as to reduce the parameter input in the model and increase the calculation rate and accuracy of the subsequent model by reducing the parameters.

[0094] 12) For the temperature collection points selected in the fire resistance performance testing device, the temperature data of the blowtorch at each temperature collection point is collected to obtain the temperature series data of each temperature collection point.

[0095] like Figure 2As shown in the figure, it is a diagram of the arrangement of temperature sensors in the fire resistance performance detection device. Specifically, in this embodiment, 48 thermocouples are used to collect the temperature of the burner in the combustion furnace at the beginning, that is, 24 temperature collection points are determined in the initial state, and each temperature collection point is installed to collect the temperature of the corresponding burner. 24 temperature collection points are selected by combining gray association and time series response, that is, 24 thermocouples are finally used to build a temperature collection system.

[0096] Temperature acquisition mainly involves hardware and software. For the hardware design of temperature acquisition, this embodiment adopts a method based on LabVIEW combined with thermocouples. The acquisition hardware involved is as follows: Figure 3 As shown. The hardware part uses K-type thermocouples. In this embodiment, a total of 24 thermocouples are used to build a temperature acquisition system; on this basis, this embodiment further uses NI's temperature data acquisition card to collect temperature data. Further, the temperature data acquisition system built with LabVIEW is used to convert temperature data, such as Figure 4 The on-site test installation conditions of the fire resistance performance testing device are as follows: Figure 5 As shown, the field temperature experiment data is collected by wrapping a red thermocouple with white asbestos.

[0097] Step 2: Temperature prediction

[0098] 21) The temperature series data of each temperature collection point is used as the input of the sliding time window to obtain the temperature series data m steps ahead. Specifically, the overall architecture of temperature prediction is as follows Figure 6 Specifically, in the sliding window process involved in this embodiment, as shown in FIG. Figure 7 As shown, firstly, the combustion temperatures in the first n early fire resistance performance detection devices measured by the ith sensor are used to construct the temperature sequence data Furthermore, a single-step sliding method is used to slide a single cell forward. Based on the sliding single cell, the temperature series data of the next moment is constructed. According to this mode, the temperature data can be input and predicted effectively. In this embodiment, by using the temperature prediction of m steps ahead, the temperature inside the fire resistance performance detection device can be obtained very accurately.

[0099] Specifically, the principle of the sliding time window is: let the temperature sequence data composed of the first n steps of temperature data collected by the i-th temperature collection point be:

[0100]

[0101] Use the sliding time window to get the temperature series data of m steps ahead:

[0102]

[0103] in: The temperature data of the first j steps collected at the i-th temperature collection point; for The temperature data is obtained after the sliding time window slides forward m steps.

[0104] In this embodiment, the mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) are used to evaluate the performance of the sliding time window:

[0105]

[0106] Where: x(t) is the measured temperature value; is the temperature value predicted by the model; m is the number of time window steps passed.

[0107] 22) Input the temperature series data output by the sliding time window into the graph neural network GCN, and use Fourier convolution operation to extract features of the temperature series data, such as Figure 8 shown.

[0108] Specifically, in this embodiment, the method of using graph neural network GCN to extract features from temperature series data is:

[0109] The image expression of temperature series data is defined as G = (V, E, W); where G is the graph structure, V is the node set, E represents the edge set, and W represents the adjacency transformation matrix; and:

[0110]

[0111] Where: w ij Represents the connection weight relationship between node i and node j;

[0112] Get the degree matrix D based on the constructed adjacency transformation matrix:

[0113]

[0114] Where: w ik is the element value in the adjacency matrix; D ii is the ith temperature data of the ith temperature collection point;

[0115] Construct the Laplace transform matrix of the measured temperature:

[0116]

[0117] Where: L represents the Laplacian matrix of the graph neural network; is the normalized Laplace matrix; U is the eigenvector of the normalized Laplace matrix; D is the degree matrix; Λ is the diagonal matrix of eigenvalues; and:

[0118] Λ=diag([λ0,λ1,...,λ N ])

[0119] Where: λ0,λ1,...,λ N are the values ​​on the diagonal of the diagonal matrix;

[0120] Use Laplace transformation to perform feature transposition and input:

[0121] X·H=F ―1 (F(X)⊙F(H))=U((U T X)⊙(U T H))

[0122] Where: X represents the information signal matrix in the node matrix V; H is the convolution kernel matrix in the graph neural network GCN; ⊙ represents the Hadamard product; F ―1 and F denotes Fourier transform and inverse Fourier transform.

[0123] In the process of training the graph neural network GCN, an improved loss function matrix based on the boost ratio is adopted:

[0124]

[0125] Where: P MAE , P RMSE and P MAPE Represents the mean absolute error, root mean square error and error coefficient value respectively; MAE i 、RMSE i and MAPE i are the mean absolute error, root mean square error and error coefficient involved in the i-th Epoch respectively; MAE1, RMSE1 and MAPE1 are the mean absolute error, root mean square error and error value respectively.

[0126] 23) Input the features extracted by the graph neural network GCN into the recurrent neural network SRU to obtain the cable combustion temperature data, such as Fig. 9 shown.

[0127] Specifically, the principle of the recursive neural network SRU is:

[0128]

[0129] Wherein: H is the cable combustion temperature data outputted by the fire resistance performance detection device; C is the state matrix inside the model; t represents the time; G is the activation function, and the activation function of this embodiment adopts the Tanh activation function; F t For the Gate of Oblivion; is the input gate.

[0130] Step 3: Lightweight deployment: Detect invalid neurons in the model by traversing the loop, such as Fig.10 shown.

[0131] Perform a presence or absence test on each neuron to obtain the sensitivity of the neuron to the final result:

[0132]

[0133] Where: N i,0 With N i,1 They represent the output corresponding to the original state matrix model when the i-th neuron weight parameter is set to 0; μ j is the information evaluation matrix constructed based on MAE, RMSE, and MAPE; ε j is the sensitivity coefficient; N j is the total number of neurons in a layer; N i Indicates the number of neural network layers;

[0134] The importance of neurons in a certain layer is determined by the sensitivity coefficient, and the neural network is pruned by deleting neurons with low importance.

[0135] This embodiment also proposes a real-time reconstruction system for cable fire resistance test temperature with multiple heat outputs, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the method described above in this embodiment is implemented.

[0136] This embodiment further proposes a 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 in this embodiment are implemented.

[0137] The above-described embodiments are only preferred embodiments for fully illustrating the present invention, and the protection scope of the present invention is not limited thereto. Equivalent substitutions or changes made by those skilled in the art based on the present invention are within the protection scope of the present invention. The protection scope of the present invention shall be subject to the claims.

Claims

1. A real-time reconstruction method for cable fire resistance test temperature with multi-element heat output, characterized in that: The steps include: Step 1: Collect temperature data 11) The temperature collection points in the fire resistance performance testing device are determined by combining grey correlation with time series response; 12) For the temperature collection points selected in the fire resistance performance testing device, the temperature data of the torch at each temperature collection point is collected to obtain the temperature series data of each temperature collection point; Step 2: Temperature prediction 21) Taking the temperature series data of each temperature collection point as the input of the sliding time window, obtaining the temperature series data of m steps ahead; 22) Input the temperature series data output by the sliding time window into the graph neural network GCN, and use Fourier convolution operation to extract features of the temperature series data; 23) Input the features extracted by the graph neural network GCN into the recursive neural network SRU to obtain the cable combustion temperature data; In the step 11), the method for determining the temperature collection point in the fire resistance performance detection device is: Construct the temperature time series change expression: f(t)=f(t0)+f[t0,t1](t―t0)+…+f[t0,…,t n ](t―t0)…(t―t n―1 )+f[t,t0,…,t n ]+(t―t0)…(t―t n―1 )(t―t n ) Where: t0, t1, …, t n―1 and t n is the temperature data collection time; f(.) is the difference remainder calculation function constructed based on grey correlation; Constructing a correlation model for fire resistance performance testing devices based on grey correlation: Where: γ(y,x i ) is the correlation value between the temperature measurement point and the cable combustion temperature; y is the difference remainder data; x i represents the combustion temperature value obtained by the i-th temperature measurement sensor, where i = 1, 2, 3, ..., N, N is the upper limit number of temperature sensors arranged; y(t) is the temperature value obtained at the t-th moment; x i (k) is the kth temperature value obtained at the i-th temperature measurement point; y(k) represents the output temperature value; is the absolute difference between the ith temperature data and the cable combustion temperature data at the temperature point in the fire resistance test device; based on the absolute difference Construct the difference matrix: Where: Δ min and Δ max That is, the minimum and maximum values ​​in the difference matrix; ρ is the resolution coefficient; Based on the obtained association value γ(y,x i ), select the temperature collection points corresponding to the key parameters that can characterize the changes in the temperature field.

2. The method for real-time reconstruction of cable fire resistance test temperature for multiple heat output according to claim 1, characterized in that: In step 21), the temperature sequence data consisting of the first n steps of temperature data collected by the ith temperature collection point is: Use the sliding time window to get the temperature series data of m steps ahead: in: The temperature data of the first j steps collected at the i-th temperature collection point; for The temperature data is obtained after the sliding time window slides forward m steps.

3. The method for real-time reconstruction of cable fire resistance test temperature for multiple heat output according to claim 2, characterized in that: The mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) are used to evaluate the performance of the sliding time window: Where: x(t) is the measured temperature value, is the temperature value predicted by the model; m is the number of time window steps passed.

4. The method for real-time reconstruction of cable fire resistance test temperature for multiple heat output according to claim 1 is characterized in that: In the step 22), the method of using the graph neural network GCN to extract features from the temperature series data is: The image expression of temperature series data is defined as G = (V, e, W); where G is the graph structure, V is the node set, E represents the edge set, and W represents the adjacency transformation matrix; and: Where: w ij Represents the connection weight relationship between node i and node j; Get the degree matrix D based on the constructed adjacency transformation matrix: Where: w ik is the element value in the adjacency matrix; D ii is the ith temperature data of the ith temperature collection point; Construct the Laplace transform matrix of the measured temperature: Where: L represents the Laplacian matrix of the graph neural network; is the normalized Laplace matrix; U is the eigenvector of the normalized Laplace matrix; D is the degree matrix; Λ is the diagonal matrix of eigenvalues; and: Λ=diag([λ0,λ1,...,λ N ]) Where: λ0,λ1,...,λ N are the values ​​on the diagonal of a diagonal matrix; Use Laplace transformation to perform feature transposition and input: X·H=F ―1 (F(X)⊙F(H))=U((U T X)⊙(U T H)) Where: X represents the information signal matrix in the node matrix V; H is the convolution kernel matrix in the graph neural network GCN; ⊙ represents the Hadamard product; F ―1 and F denotes Fourier transform and inverse Fourier transform.

5. The method for real-time reconstruction of cable fire resistance test temperature for multiple heat output according to claim 4 is characterized in that: In the process of training the graph neural network GCN, an improved loss function matrix based on the boost ratio is adopted: Where: P MAE , P RMSE and P MAPE Respectively represent the average error coefficient percentage, root mean square error coefficient percentage, and average prediction accuracy error percentage; MAE i 、RMSE i and MAPE i are the mean absolute error, root mean square error and error coefficient involved in the i-th Epoch; MAE i 、RMSE i and MAPE i are the mean square error, root mean square error, and average precision error of the i-th point respectively.

6. The method for real-time reconstruction of cable fire resistance test temperature for multiple heat output according to claim 1 is characterized in that: In step 23), the principle of the recursive neural network SRU is: Where: H is the cable combustion temperature data output from the fire resistance performance detection device; C is the state matrix inside the model; t represents the time; G is the activation function; F t For the Gate of Oblivion; is the input gate.

7. The method for real-time reconstruction of cable fire resistance test temperature for multiple heat output according to any one of claims 1 to 6, characterized in that: It also includes step 3: lightweight deployment: detecting invalid neurons in the model by traversing the loop; Perform a presence or absence test on each neuron to obtain the sensitivity of the neuron to the final result: Where: N i,0 With N i,1 They represent the output corresponding to the original state matrix model when the i-th neuron weight parameter is set to 0; μ j is the information evaluation matrix constructed based on MAE, RMSE, and MAPE; ε j is the sensitivity coefficient; N j is the total number of neurons in a layer; N i Indicates the number of neural network layers; The importance of neurons in a certain layer is determined by the sensitivity coefficient, and the neural network is pruned by deleting neurons with low importance.

8. A real-time reconstruction system for cable fire resistance test temperature with multi-element heat output, characterized by: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method according to any one of claims 1 to 7 is implemented.

9. A storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images