Method and system for predicting health degree degradation trend of test incubator cluster
By constructing an adaptive spatiotemporal neural network and utilizing encoding, decoding, and multi-head attention modules, the gradient vanishing problem in traditional recurrent neural networks when learning long-term features is solved. This enables accurate prediction of the health degradation trend of test chamber clusters, lowers the professional threshold, and improves prediction efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- AEROSPACE SCI & IND INERTIA TECH CO LTD
- Filing Date
- 2024-10-30
- Publication Date
- 2026-05-01
AI Technical Summary
Traditional recurrent neural networks may suffer from gradient vanishing due to error propagation when learning long-term features, leading to inaccurate predictions of the health degradation trend of test chamber clusters.
We employ a mask-based adaptive spatiotemporal neural network approach. By constructing an adaptive spatiotemporal neural network and utilizing an encoding module, a decoding module, and a multi-head attention module, combined with the spatiotemporal unit STGRU and mask map convolution operations, we extract the temporal and spatial features of the test chamber cluster. This approach captures the periodicity of the temporal dimension and the heterogeneity of the spatial dimension, thus avoiding the gradient vanishing problem.
It enables accurate prediction of the health degradation trend of test chamber clusters, lowers the professional threshold, avoids the gradient vanishing problem, and improves the accuracy and efficiency of prediction.
Smart Images
Figure CN121960087A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of predicting the health degradation trend of test chamber clusters, and in particular to a method and system for predicting the health degradation trend of test chamber clusters. Background Technology
[0002] Test chambers are essential testing equipment in aerospace, shipbuilding, weaponry, electronic communications, petrochemical, automotive, and scientific research fields. They are commonly used to test and determine whether the parameters and performance of electrical and electronic products and other materials still meet predetermined requirements after temperature changes in high and low temperature testing environments, thus providing test basis for product design, qualification, and factory inspection.
[0003] When the health of a test chamber deteriorates, without effective maintenance and repair, it will eventually lead to equipment failure. Since industrial production is a complex, interconnected process, the degradation of critical testing equipment like test chambers will severely impact product quality, increasing defect rates and jeopardizing the healthy operation of the entire industrial production system. Therefore, employing efficient health monitoring technologies and degradation trend prediction methods during the degradation phase of the test chamber to determine its remaining service life, and then arranging reasonable maintenance work accordingly, is crucial to saving maintenance costs, improving product quality, preventing production accidents, and increasing company profits.
[0004] In practical industrial production systems, several test chambers are often combined into a chamber cluster for use. The spatial relationship between these test chambers is a non-Euclidean spatial structure, so the data collected can be regarded as spatiotemporal graph data. However, traditional recurrent neural networks may encounter the gradient vanishing problem caused by error propagation when learning long-term features. Summary of the Invention
[0005] This invention provides a method and system for predicting the health degradation trend of a cluster of test chambers, which can solve the technical problem of gradient vanishing caused by error propagation when recurrent neural networks learn long-term features in the prior art.
[0006] According to one aspect of the present invention, a method for predicting the health degradation trend of a test chamber cluster based on a mask adaptive spatiotemporal neural network is provided. The method includes: Step 1, constructing an adaptive spatiotemporal neural network, which consists of an encoding module, a decoding module, and a multi-head attention module. A multi-head attention module is embedded between the encoding and decoding modules to connect them while simultaneously emphasizing important features in the latent variables; Step 2, inputting real-time current, voltage, signal strength, and thermal stress index of the test chamber cluster into the encoding module, which then outputs the latent variables at all times; Step 3, utilizing the multi-head attention module to emphasize important features in the latent variables output by the encoding module to output the latent variables after adaptive fusion. Step four, the decoding module processes the latent variables after adaptive fusion. The health degradation value of each test chamber in the test chamber cluster is predicted by decoding output.
[0007] Furthermore, in step two, the latent variable H corresponding to each time step is obtained by the spatiotemporal feature calculation of the spatiotemporal unit STGRU. t H t =u t ⊙H t-1 +(1-u t )⊙C t , in, The encoded representation matrix is assigned to the test chamber cluster to identify the heterogeneity in different spatial dimensions of the chambers; The encoding representation matrix is assigned to different time points and is used as an identification matrix to identify which hour of the day the current time falls within. The encoded representation matrix is assigned to different time points to identify which day of the week the given time falls on; This is the first weight matrix. This is the second weight matrix. b is the third weight matrix. C b is the first bias coefficient. r b is the second bias coefficient. u The third bias coefficient; MGConv(X) is the mask image convolution operation, X t Let C be the input data at time t. t u is used to encode intermediate variables. t For encoding update gate, r t Let σ be the encoding reset gate, d be the activation function, d be the hyperparameter, and h be the dimension of the latent variables of each chamber at a certain time.
[0008] Furthermore, the formula for calculating the mask image convolution operation MGConv(X) is MGConv(X) = P f XW f +P b XW b +A adp XW a ,in, This is the fourth weight matrix. This is the fifth weight matrix. The sixth weight matrix, which needs to be learned during model training; The mask matrix during forward information diffusion is calculated using the following formula: The mask matrix during backward information diffusion is calculated using the following formula: It is the first representation matrix learned during model training. It is the second representation matrix learned during model training. This is the initial adjacency matrix for testing the temperature chamber cluster; A adp The global mask matrix is calculated using the following formula: It is the third representation matrix learned during model training. It is the fourth representation matrix learned during model training, where X is the input data.
[0009] Furthermore, in step three, the latent variables are added after the attention mechanism is implemented. The calculation formula is in, In the attention mechanism, K corresponds to the query. n V corresponds to the key in the attention mechanism. n In the attention mechanism, W corresponds to the value. h This is the weight matrix.
[0010] Furthermore, in step three, the latent variables after adaptive fusion... The calculation formula is in, It is a parameter that can be learned along with the model training.
[0011] Furthermore, in step four, the decoded latent variable H' t According to H' t =u' t ⊙H' t-1 +(1-u' t )⊙C' t , where u' t To decode the update gate, C' t This is an intermediate variable for decoding.
[0012] Furthermore, in step four, C' t and u' t According to in The decoding representation matrix assigned to the test chamber cluster is used to identify the heterogeneity in different spatial dimensions of the chambers; The decoding representation matrix is assigned to different time points, and the identification matrix is used to identify which hour of the day the current time falls within; The decoded representation matrix is assigned to different time points to identify which day of the week the current time falls on; d is a hyperparameter. This is the seventh weight matrix. This is the eighth weight matrix. b' is the ninth weight matrix. C b' is the fourth bias coefficient. r b' is the fifth bias coefficient. u The sixth bias coefficient, r t 'Reset the gate for decoding.'
[0013] According to another aspect of the present invention, a test chamber cluster health degradation trend prediction system based on a mask adaptive spatiotemporal neural network is provided. The test chamber cluster health degradation trend prediction system uses the test chamber cluster health degradation trend prediction method described above to predict the health degradation trend. The test chamber cluster health degradation trend prediction system consists of three parts: an encoding module, a decoding module, and a multi-head attention module. The encoding module inputs the real-time current, voltage, signal strength, and thermal stress index of the test chamber cluster. The decoding module outputs the predicted health degradation value of each test chamber in the test chamber cluster. The multi-head attention module is embedded between the encoding module and the decoding module to connect the two and simultaneously emphasize important features in the latent variables.
[0014] According to another aspect of the present invention, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method for predicting the health degradation trend of a test chamber cluster based on a mask adaptive spatiotemporal neural network as described above.
[0015] According to another aspect of the present invention, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the method for predicting the health degradation trend of a test chamber cluster based on a mask adaptive spatiotemporal neural network as described above.
[0016] The present invention provides a method for predicting the health degradation trend of test chamber clusters based on a mask-adaptive spatiotemporal neural network. The spatiotemporal neural network proposed in this invention can not only extract the temporal and spatial features of the test chamber cluster data, but also simultaneously capture the periodicity of the temporal dimension and the heterogeneity of the spatial dimension. Secondly, when using a graph neural network to extract the spatial features of the test chamber cluster data, the spatiotemporal neural network in this invention does not require predefining the graph structure of the test chamber cluster, which greatly reduces the professional threshold for using this method. Finally, when using a recurrent neural network, the spatiotemporal neural network in this invention avoids the gradient vanishing problem caused by error propagation that may occur when traditional recurrent neural networks learn long-term temporal features. Attached Figure Description
[0017] The accompanying drawings, which form part of this specification, are provided to further illustrate embodiments of the invention and, together with the textual description, explain the principles of the invention. It is obvious that the drawings described below are merely some embodiments of the invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.
[0018] Figure 1 A schematic diagram of the structure of a test chamber cluster health degradation trend prediction system provided according to a specific embodiment of the present invention is shown. Detailed Implementation
[0019] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the present invention or its application or use. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0020] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.
[0021] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values of the components and steps set forth in these embodiments do not limit the scope of the invention. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following figures denote similar items; therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.
[0022] like Figure 1 As shown in the figure, a method for predicting the health degradation trend of a test chamber cluster based on a mask adaptive spatiotemporal neural network is provided according to a specific embodiment of the present invention. The method includes: Step 1, constructing an adaptive spatiotemporal neural network, which consists of an encoding module, a decoding module, and a multi-head attention module. A multi-head attention module is embedded between the encoding module and the decoding module to connect the encoding module and the decoding module, and to emphasize important features in the latent variables; Step 2, inputting the real-time current, voltage, signal strength, and thermal stress index of the test chamber cluster into the encoding module, and outputting the latent variables at all times; Step 3, using the multi-head attention module to emphasize important features in the latent variables output by the encoding module to output the latent variables after adaptive fusion. Step four, the decoding module processes the latent variables after adaptive fusion. The health degradation value of each test chamber in the test chamber cluster is predicted by decoding output.
[0023] This configuration provides a method for predicting the health degradation trend of test chamber clusters based on a mask-adaptive spatiotemporal neural network. The spatiotemporal neural network proposed in this invention can not only extract the temporal and spatial features of the test chamber cluster data, but also simultaneously capture the periodicity of the temporal dimension and the heterogeneity of the spatial dimension. Secondly, when using a graph neural network to extract the spatial features of the test chamber cluster data, the spatiotemporal neural network in this invention does not require predefining the graph structure of the test chamber cluster, which greatly reduces the professional threshold for using this method. Finally, when using a recurrent neural network, the spatiotemporal neural network in this invention avoids the gradient vanishing problem caused by error propagation that may occur when traditional recurrent neural networks learn long-term temporal features.
[0024] Specifically, in this invention, in order to predict the health degradation trend of the test chamber cluster, it is first necessary to construct an adaptive spatiotemporal neural network. The adaptive spatiotemporal neural network consists of three parts: an encoding module, a decoding module, and a multi-head attention module. The multi-head attention module is embedded between the encoding module and the decoding module to connect the encoding module and the decoding module, while giving extra attention to important features in the latent variables.
[0025] Furthermore, after constructing the adaptive spatiotemporal neural network, the real-time current, voltage, signal strength, and thermal stress index of the test chamber cluster can be input into the encoding module, and the encoding module will output the latent variables at all times.
[0026] Specifically, in step two, the latent variable H corresponding to each time step is obtained by the spatiotemporal feature calculation of the spatiotemporal unit STGRU. t H t =u t ⊙H t-1 +(1-u t )⊙C t , in, The encoded representation matrix is assigned to the test chamber cluster to identify the heterogeneity in different spatial dimensions of the chambers; The encoding representation matrix is assigned to different time points and is used as an identification matrix to identify which hour of the day the current time falls within. The encoded representation matrix is assigned to different time points to identify which day of the week the given time falls on; This is the first weight matrix. This is the second weight matrix. b is the third weight matrix. C b is the first bias coefficient. r b is the second bias coefficient. u The third bias coefficient; MGConv(X) is the mask image convolution operation, X t Let C be the input data at time t. t u is used to encode intermediate variables. t For encoding update gate, r t Let σ be the encoding reset gate, d be the activation function, d be the hyperparameter, and h be the dimension of the latent variables of each chamber at a certain time.
[0027] The formula for calculating the mask image convolution operation MGConv(X) is MGConv(X) = P f XW f +P b XW b +A adp XW a ,in, This is the fourth weight matrix. This is the fifth weight matrix. The sixth weight matrix, which needs to be learned during model training; The mask matrix during forward information diffusion is calculated using the following formula: The mask matrix during backward information diffusion is calculated using the following formula: It is the first representation matrix learned during model training. It is the second representation matrix learned during model training. This is the initial adjacency matrix for testing the temperature chamber cluster; A adp The global mask matrix is calculated using the following formula: It is the third representation matrix learned during model training. It is the fourth representation matrix learned during model training, where X is the input data.
[0028] Furthermore, after the encoding module outputs the latent variables at all time points, a multi-head attention module can be used to emphasize important features in the latent variables output by the encoding module to output the latent variables after adaptive fusion.
[0029] Specifically, in step three, the latent variables are added after the attention mechanism is implemented. The calculation formula is in, In the attention mechanism, K corresponds to the query. n V corresponds to the key in the attention mechanism. n In the attention mechanism, W corresponds to the value. h This is the weight matrix.
[0030] Latent variables after adaptive fusion The calculation formula is in, It is a parameter that can be learned along with the model training.
[0031] Furthermore, after obtaining the latent variables after adaptive fusion... Then, the decoding module processes the latent variables after adaptive fusion. The health degradation value of each test chamber in the test chamber cluster is predicted by decoding output.
[0032] Specifically, in step four, the decoded latent variable H' t According to H' t =u' t ⊙H' t-1 +(1-u't )⊙C' t , where u' t To decode the update gate, C' t For decoding intermediate variables. C' t and u' t According to in, The decoding representation matrix assigned to the test chamber cluster is used to identify the heterogeneity in different spatial dimensions of the chambers; The decoding representation matrix is assigned to different time points, and the identification matrix is used to identify which hour of the day the current time falls within; The decoded representation matrix is assigned to different time points to identify which day of the week the current time falls on; d is a hyperparameter. This is the seventh weight matrix. This is the eighth weight matrix. b' is the ninth weight matrix. C b' is the fourth bias coefficient. r b' is the fifth bias coefficient. u The sixth bias coefficient, r t 'Reset the gate for decoding.'
[0033] According to another aspect of the present invention, a test chamber cluster health degradation trend prediction system based on a mask adaptive spatiotemporal neural network is provided. This test chamber cluster health degradation trend prediction system uses the test chamber cluster health degradation trend prediction method described above to predict the health degradation trend. The test chamber cluster health degradation trend prediction system consists of three parts: an encoding module, a decoding module, and a multi-head attention module. The encoding module inputs the real-time current, voltage, signal strength, and thermal stress index of the test chamber cluster. The decoding module outputs the predicted health degradation value of each test chamber in the test chamber cluster. The multi-head attention module is embedded between the encoding module and the decoding module to connect the two and simultaneously emphasize important features in the latent variables.
[0034] This configuration provides a system for predicting the health degradation trend of test chamber clusters based on a mask-adaptive spatiotemporal neural network. Using the spatiotemporal neural network proposed in this invention, not only can the temporal and spatial features of the test chamber cluster data be extracted, but the periodicity of the temporal dimension and the heterogeneity of the spatial dimension can also be captured simultaneously. Secondly, when using a graph neural network to extract the spatial features of the test chamber cluster data, the spatiotemporal neural network in this invention does not require predefining the graph structure of the test chamber cluster, which greatly reduces the professional threshold for using this method. Finally, when using a recurrent neural network, the spatiotemporal neural network in this invention avoids the gradient vanishing problem caused by error propagation that may occur when traditional recurrent neural networks learn long-term temporal features.
[0035] According to another aspect of the present invention, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method for predicting the health degradation trend of a test chamber cluster based on a mask adaptive spatiotemporal neural network as described above.
[0036] According to another aspect of the present invention, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the method for predicting the health degradation trend of a test chamber cluster based on a mask adaptive spatiotemporal neural network as described above.
[0037] To gain a further understanding of the present invention, the following description is provided in conjunction with... Figure 1 The present invention provides a detailed description of the method and system for predicting the health degradation trend of test chamber clusters based on mask adaptive spatiotemporal neural networks.
[0038] like Figure 1 As shown, this invention proposes a mask-adaptive spatiotemporal neural network based on basic graph neural networks and recurrent neural networks, combined with the characteristics of test chamber cluster data, and uses it to solve the problem of predicting the health degradation trend of test chamber clusters.
[0039] The technical solution adopted in this invention patent is as follows:
[0040] The mask-adaptive spatiotemporal neural network is based on a "sequence-to-sequence" architecture, consisting of an encoding module, a decoding module, and a multi-head attention module. The encoding module takes into account real-time data such as current, voltage, signal strength, and heat stress index of the test chamber cluster, and decodes to output the predicted health degradation value of each test chamber in the cluster. The multi-head attention module is embedded between the encoding and decoding modules to connect the two while simultaneously emphasizing important features in the latent variables.
[0041] Both the encoding and decoding modules are composed of spatiotemporal units (STGRUs), which mainly include spatiotemporal feature operations and mask image convolution operations.
[0042] The multi-head attention module adds multi-head attention mechanism operations to the latent variables output by the encoder module, then adaptively fuses them with the initial output of the encoder, and finally inputs them to the decoding module.
[0043] As attached Figure 1 As shown, the adaptive spatiotemporal neural network consists of three parts: an encoding module, a decoding module, and a multi-head attention module.
[0044] 1) Encoding module
[0045] The input to the encoding module is real-time data such as current, voltage, signal strength, and heat stress index of the test chamber cluster. Where T is the duration of the input data in one iteration, N is the number of chambers in the test chamber cluster, and C is the dimension of the data. The output of the encoding module is the latent variables at all time points. h is the dimension of the latent variables of each temperature chamber at a certain time, and is a hyperparameter. The latent variable H at each time step is... t The calculation formula is obtained from the spatiotemporal characteristics of the spatiotemporal unit STGRU:
[0046] H t =u t ⊙H t-1 +(1-u t )⊙C t
[0047] Among them, the encoding update gate u t Encoding reset gate r t With encoding intermediate variable C t The calculation formula is:
[0048] C t =tanh((MGConv([X t ,(r t ⊙H t-1 )])+[E dev E day E week ])W C +b C )
[0049] r t =σ(MGConv([X t H t-1 ])+[E dev E day E week ])W r +b r )
[0050] u t =σ(MGConv([X t H t-1 ])+[E dev E day E week ])W u +b u )
[0051] in, The encoded representation matrix is assigned to the test chamber cluster to identify the heterogeneity in different spatial dimensions of the chambers; The encoding representation matrix is assigned to different time points and is used as an identification matrix to identify which hour of the day the current time falls within. The encoded representation matrix is assigned to different time points to identify which day of the week the given time falls on; This is the first weight matrix. This is the second weight matrix. b is the third weight matrix. C b is the first bias coefficient. r b is the second bias coefficient. u The third bias coefficient is the one that needs to be learned during model training; MGConv(X) is the mask image convolution operation, where X is the third bias coefficient. t Let C be the input data at time t. t u is an intermediate variable. t To update the gate, r t Let σ be the encoding reset gate, d be the activation function, d be the hyperparameter, and h be the dimension of the latent variables of each chamber at a given time. The formula for calculating MGConv(X) is:
[0052] MGConv(X)=P f XW f +P b XW b +A adp XW a
[0053] in, This is the fourth weight matrix. This is the fifth weight matrix. The sixth weight matrix, which needs to be learned during model training; The mask matrix during forward information diffusion is calculated using the following formula: The mask matrix during backward information diffusion is calculated using the following formula: It is the first representation matrix learned during model training. It is the second representation matrix learned during model training. This is the initial adjacency matrix for the test chamber cluster, which can be predefined based on the connection relationships between the chambers; A adp The global mask matrix is calculated using the following formula: It is the third representation matrix learned during model training. It is the fourth representation matrix learned during model training, where X is the input data.
[0054] 2) Multi-head attention module
[0055] This invention embeds a multi-head attention module between the encoding and decoding modules. This module connects the encoding and decoding modules while simultaneously emphasizing important features in the latent variables. After adding the attention mechanism, the latent variables... The calculation formula is: W h As a weight matrix, each test chamber has its own separate head to incorporate the attention mechanism, and its calculation formula is as follows:
[0056]
[0057] in, In the attention mechanism, K corresponds to the query. n V corresponds to the key in the attention mechanism. n The formulas for calculating the value in the corresponding attention mechanism are as follows: To obtain the weight matrix H during model training t 'H' represents the latent variable output by the decoder at time t. n This is the set of latent variables at all times in the encoder output.
[0058] Finally, the latent variables after incorporating the attention mechanism will be... By adaptively fusing the original latent variable H input to the encoding module, the input to the decoding module at time t can be obtained. The calculation formula is as follows:
[0059]
[0060] in, It is a parameter that can be learned along with the model training.
[0061] 3) Decoding module
[0062] The input to the decoding module is the latent variables adaptively fused from the output of the multi-head attention module. The calculation formula was explained in the previous section. The intermediate output of the decoding module is the latent variable at all time points. The latent variable H' at each time step t The calculation formula is obtained from the spatiotemporal characteristics of the spatiotemporal unit STGRU:
[0063] H' t =u' t ⊙H' t-1 +(1-u' t )⊙C' t
[0064] Among them, the decoding update gate u't Decode Reset Gate t 'With decoding intermediate variable C' t The calculation formula is:
[0065]
[0066] in, The decoding representation matrix assigned to the test chamber cluster is used to identify the heterogeneity in different spatial dimensions of the chambers; The decoding representation matrix is assigned to different time points, and the identification matrix is used to identify which hour of the day the current time falls within; The decoded representation matrix is assigned to different time points to identify which day of the week the current time falls on; d is a hyperparameter. This is the seventh weight matrix. This is the eighth weight matrix. b' is the ninth weight matrix. r b' is the fifth bias coefficient. u The sixth bias coefficient, r, needs to be learned during model training. t ' is the reset gate. MGConv(X) is the mask image convolution operation, and its calculation formula is the same as that of the mask image convolution operation in the encoding module.
[0067] Finally, the output H' of the decoding module at the last moment is... t+T The input is fed into a three-layer fully connected layer, and the output is a vector of the predicted health degradation trend of the test device cluster.
[0068] In summary, this invention provides a method for predicting the health degradation trend of test chamber clusters based on a mask-adaptive spatiotemporal neural network. Using the spatiotemporal neural network proposed in this invention, not only can the temporal and spatial features of the test chamber cluster data be extracted, but the periodicity of the temporal dimension and the heterogeneity of the spatial dimension can also be captured simultaneously. Secondly, when using a graph neural network to extract the spatial features of the test chamber cluster data, the spatiotemporal neural network in this invention does not require predefining the graph structure of the test chamber cluster, which greatly reduces the professional threshold for using this method. Finally, when using a recurrent neural network, the spatiotemporal neural network in this invention avoids the gradient vanishing problem caused by error propagation that may occur when traditional recurrent neural networks learn long-term temporal features.
[0069] For ease of description, spatial relative terms such as "above," "on top of," "on the upper surface of," "above," etc., are used herein to describe the spatial positional relationship of a device or feature as shown in the figures to other devices or features. It should be understood that spatial relative terms are intended to encompass different orientations in use or operation beyond the orientation of the device as described in the figures. For example, if the device in the figures were inverted, a device described as "above" or "on top of" other devices or structures would subsequently be positioned as "below" or "under" other devices or structures. Thus, the exemplary term "above" can include both "above" and "below." The device may also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatial relative descriptions used herein will be interpreted accordingly.
[0070] Furthermore, it should be noted that the use of terms such as "first" and "second" to define components is merely for the purpose of distinguishing the corresponding components. Unless otherwise stated, the above terms have no special meaning and therefore should not be construed as limiting the scope of protection of this invention.
[0071] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for predicting the health degradation trend of a test chamber cluster based on a mask-adaptive spatiotemporal neural network, characterized in that, The method for predicting the health degradation trend of the test chamber cluster includes: Step 1: Construct an adaptive spatiotemporal neural network. The adaptive spatiotemporal neural network consists of three parts: an encoding module, a decoding module, and a multi-head attention module. The multi-head attention module is embedded between the encoding module and the decoding module to connect the encoding module and the decoding module, while giving more attention to important features in the latent variables. Step 2: Input the real-time current, voltage, signal strength, and heat stress index of the test chamber cluster into the encoding module, and the encoding module outputs the latent variables at all times; Step 3: The multi-head attention module is used to emphasize important features in the latent variables output by the encoding module to output the latent variables after adaptive fusion. Step four, the decoding module processes the latent variables after adaptive fusion. The health degradation value of each test chamber in the test chamber cluster is predicted by decoding output.
2. The method for predicting the health degradation trend of a test chamber cluster based on a mask adaptive spatiotemporal neural network according to claim 1, characterized in that, In step two, the latent variable H corresponding to each time step is obtained by spatiotemporal feature calculation of the spatiotemporal unit STGRU. t H t =u t ⊙H t-1 +(1-u t )⊙C t , in, The encoded representation matrix is assigned to the test chamber cluster to identify the heterogeneity in different spatial dimensions of the chambers; The encoding representation matrix is assigned to different time points and is used as an identification matrix to identify which hour of the day the current time falls within. The encoded representation matrix is assigned to different time points to identify which day of the week the given time falls on; This is the first weight matrix. This is the second weight matrix. b is the third weight matrix. C b is the first bias coefficient. r b is the second bias coefficient. u The third bias coefficient; MGConv(X) is the mask image convolution operation, X t Let C be the input data at time t. t u is used to encode intermediate variables. t For encoding update gate, r t Let σ be the encoding reset gate, d be the activation function, d be the hyperparameter, and h be the dimension of the latent variables of each chamber at a certain time.
3. The method for predicting the health degradation trend of a test chamber cluster based on a mask adaptive spatiotemporal neural network according to claim 2, characterized in that, The formula for calculating the mask image convolution operation MGConv(X) is MGConv(X) = P f XW f +P b XW b +A adp XW a ,in, This is the fourth weight matrix. This is the fifth weight matrix. The sixth weight matrix, which needs to be learned during model training; The mask matrix during forward information diffusion is calculated using the following formula: The mask matrix during backward information diffusion is calculated using the following formula: It is the first representation matrix learned during model training. It is the second representation matrix learned during model training. This is the initial adjacency matrix for testing the temperature chamber cluster; A adp The global mask matrix is calculated using the following formula: It is the third representation matrix learned during model training. It is the fourth representation matrix learned during model training, where X is the input data.
4. The method for predicting the health degradation trend of a test chamber cluster based on a mask adaptive spatiotemporal neural network according to any one of claims 1 to 3, characterized in that, In step three, the latent variables are added after the attention mechanism is introduced. The calculation formula is in, In the attention mechanism, K corresponds to the query. n V corresponds to the key in the attention mechanism. n In the attention mechanism, W corresponds to the value. h This is the weight matrix.
5. The method for predicting the health degradation trend of a test chamber cluster based on a mask adaptive spatiotemporal neural network according to claim 4, characterized in that, In step three, the latent variables after adaptive fusion The calculation formula is in, It is a parameter that can be learned along with the model training.
6. The method for predicting the health degradation trend of a test chamber cluster based on a mask adaptive spatiotemporal neural network according to claim 5, characterized in that, In step four, the decoded latent variable H' t According to H' t =u' t ⊙H' t-1 +(1-u' t )⊙C' t , where u' t To decode the update gate, C' t This is an intermediate variable for decoding.
7. The method for predicting the health degradation trend of a test chamber cluster based on a mask adaptive spatiotemporal neural network according to claim 6, characterized in that, In step four, C' t and u' t According to in, The decoding representation matrix assigned to the test chamber cluster is used to identify the heterogeneity in different spatial dimensions of the chambers; The decoding representation matrix is assigned to different time points, and the identification matrix is used to identify which hour of the day the current time falls within; The decoded representation matrix is assigned to different time points to identify which day of the week the current time falls on; d is a hyperparameter. This is the seventh weight matrix. This is the eighth weight matrix. b' is the ninth weight matrix. C b' is the fourth bias coefficient. r b' is the fifth bias coefficient. u The sixth bias coefficient, r t 'Reset the gate for decoding.' 8. A system for predicting the health degradation trend of a test chamber cluster based on a mask-adaptive spatiotemporal neural network, characterized in that, The test chamber cluster health degradation trend prediction system uses the test chamber cluster health degradation trend prediction method as described in any one of claims 1 to 7 to predict the health degradation trend. The test chamber cluster health degradation trend prediction system consists of three parts: an encoding module, a decoding module, and a multi-head attention module. The encoding module inputs the real-time current, voltage, signal strength, and heat stress index of the test chamber cluster. The decoding module outputs the predicted health degradation value of each test chamber in the test chamber cluster. The multi-head attention module is embedded between the encoding module and the decoding module to connect the two and simultaneously emphasize important features in the latent variables.
9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the method for predicting the health degradation trend of a test chamber cluster based on a mask adaptive spatiotemporal neural network as described in any one of claims 1 to 7.
10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for predicting the health degradation trend of a test chamber cluster based on a mask adaptive spatiotemporal neural network as described in any one of claims 1 to 7.