Welding parameter prediction method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention discloses a welding parameter prediction method and device, electronic equipment and a storage medium, and relates to the technical field of welding. The method comprises the steps that a feature value of a first welding parameter at a first moment is obtained; inputting the characteristic value of the first welding parameter into a pre-trained prediction model to obtain a characteristic value of a second welding parameter at the next moment of the first moment output by the prediction model; wherein the second welding parameter comprises welding time and / or welding power, and the first welding parameter is an independent variable of the second welding parameter. According to the embodiment, the welding time and the welding power of the battery cell can be accurately predicted, the accuracy of the prediction result of the welding time and the welding power is effectively improved, meanwhile, the welding process can be intervened in time according to the prediction result, and the welding quality is improved.

Description

Technical Field

[0001] The present application relates to the field of welding technology. Specifically, the present application relates to a method, device, electronic device, and storage medium for predicting welding parameters. Background Art

[0002] Tab welding by ultrasonic is a key step in the production of battery cells, and its welding quality directly affects the performance, safety, and production cost of the battery cells. During the selection of welding parameters, welding time and welding power are two key factors. The welding time refers to the time length during which ultrasonic waves act on the welding point. If the welding time is too short, the welding may be insufficient, resulting in an insecure connection at the welding point, thereby affecting the overall performance of the battery cell. On the contrary, if the welding time is too long, overwelding or welding through may occur, damaging the battery cell material and affecting the product quality and safety. Therefore, correctly setting the welding time is crucial for ensuring welding quality. At the same time, the welding power is the intensity of the applied energy. If the power is too low, the welding may be insufficient, and if the power is too high, overwelding or welding through may occur, causing excessive heating and damage to the material.

[0003] In the traditional process of selecting ultrasonic welding parameters, key data such as welding time and welding power can only be obtained after actual welding experiments. Since the welding time and welding power can only be obtained afterwards, the identification and solution of welding problems are postponed. Therefore, it has become an urgent task to predict the welding time and welding power and intervene in a timely manner according to the prediction results to improve the welding quality of the battery. Summary of the Invention

[0004] Embodiments of the present application provide a method, device, electronic device, and storage medium for predicting welding parameters to solve the technical problem in the related art that the welding time and welding power can only be obtained afterwards, resulting in the postponement of the identification and solution of welding problems.

[0005] According to a first aspect of the embodiments of the present application, a method for predicting welding parameters is provided. The method includes: Obtaining the characteristic value of the first welding parameter at the first moment; wherein, the first welding parameter includes at least one of the material parameter of the welding operation object and the welding process parameter; Inputting the characteristic value of the first welding parameter into a pre-trained prediction model; wherein, the prediction model is trained by a graph convolutional network and a gated recurrent unit using the parameter values of the first welding parameter at multiple sample moments as training samples; each training sample further includes a training label, and the training label is the characteristic value of the second welding parameter at the next moment corresponding to the sample moment; wherein, the second welding parameter includes: welding time and / or welding power, and the first welding parameter is the independent variable of the second welding parameter; Obtain the eigenvalue of the second welding parameter at the next moment of the first moment output by the prediction model.

[0006] As an alternative implementation, the graph convolutional network includes: a graph convolutional module and an attention module; the gated recurrent network includes: a feature extraction module and a second welding parameter prediction module; Inputting the eigenvalue of the first welding parameter into a pre-trained prediction model to obtain the eigenvalue of the second welding parameter at the next moment of the first moment output by the prediction model includes: Input the eigenvalue of the first welding parameter into the graph convolutional module and the attention module respectively, obtain the target feature matrix output by the graph convolutional module, and the target feature vector output by the attention module; according to the target feature matrix and the target feature vector, obtain the updated eigenvalue of the first welding parameter; Input the updated eigenvalue of the first welding parameter into the feature extraction module to obtain the target hidden state output by the feature extraction module, where the target hidden state indicates the dependency relationship between the first welding parameters; input the target hidden state into the second welding parameter prediction module to output the eigenvalue of the second welding parameter at the next moment.

[0007] As an alternative implementation, the graph convolutional module includes at least one graph convolutional unit; there is a pre-determined mutual relationship between the first welding parameters; The step of inputting the eigenvalue of the first welding parameter into the graph convolutional module to obtain the target feature matrix output by the graph convolutional module includes: Determine the feature map at the first moment according to the eigenvalue of the first welding parameter; wherein, the nodes of the feature map are the first welding parameters, and the edges between the nodes are the pre-determined mutual relationships between the first welding parameters; Input the feature map into each graph convolutional unit in sequence to obtain the feature matrices output by each graph convolutional unit, and use the feature matrix output by the last graph convolutional unit as the target feature matrix.

[0008] As an alternative implementation, the attention module includes at least one attention unit and a global pooling unit; wherein, each attention unit includes an attention channel corresponding to the first welding parameter; The step of inputting the eigenvalue of the first welding parameter into the attention module to obtain the target feature vector output by the attention module includes: Determine the input data of each attention unit; wherein, for the first attention unit, the input data is the eigenvalue of the first welding parameter; for non-first attention units, the input data is the eigenvalue output after convolutional calculation by the previous attention unit; For each attention unit, input the input data corresponding to each first welding parameter into the corresponding attention channel, and perform convolution calculation in the corresponding attention channel to obtain the eigenvalue output by each attention channel; wherein, the weight values of each attention channel are different; Input the eigenvalues output by each attention channel in the last attention unit into the global pooling unit, extract the maximum value among all the eigenvalues in each attention channel as the eigenvalue after updating the corresponding first welding parameter, and form a one-dimensional feature vector with the eigenvalues after updating the first welding parameter as the target feature vector.

[0009] As an alternative implementation, the feature extraction module includes a plurality of gated recurrent units and a fully connected unit; The step of inputting the eigenvalue of the updated first welding parameter into the feature extraction module and obtaining the target hidden state output by the feature extraction module includes: For the first gated recurrent unit, the input data is the eigenvalue after updating the first welding parameter, and the output data is the hidden state of the first gated recurrent unit; For non-first gated recurrent units, the input data is the eigenvalue after updating the first welding parameter and the hidden state output by the previous gated recurrent unit, and the output is the hidden state of the corresponding gated recurrent unit; Input the hidden state output by the last gated recurrent unit into the fully connected unit for full connection processing to obtain the hidden state output by the fully connected unit as the target hidden state.

[0010] As an alternative implementation, the second welding parameter prediction module includes a plurality of gated recurrent units and a global pooling unit; The step of inputting the target hidden state into the second welding parameter prediction module and outputting the eigenvalue of the second welding parameter at the next moment includes: Determine the input data of each gated recurrent unit; wherein, for the first gated recurrent unit, the input data is the target hidden state; for non-first gated recurrent units, the input data is the hidden state output by the previous gated recurrent unit; For each gated recurrent unit, perform a dot product operation on its own output hidden state to output the predicted value of the second welding parameter; Input the predicted values of the second welding parameter output by each gated recurrent unit into the global pooling unit, and perform pooling processing on all the predicted values of the second welding parameter to obtain the eigenvalue of the second welding parameter at the next moment.

[0011] As an alternative implementation, the prediction model is generated in the following manner, including: Pre-acquire the parameter values of the first welding parameters and the parameter values of the second welding parameters at multiple moments, preprocess the parameter values of the first welding parameters and the parameter values of the second welding parameters at each moment to obtain the characteristic values of the first welding parameters and the characteristic values of the second welding parameters at each moment; wherein, the preprocessing includes at least one of data cleaning and normalization processing; According to the characteristic values of the first welding parameters and the characteristic values of the second welding parameters at each moment, determine the characteristic values of the first welding parameters at multiple sample moments and the characteristic values of the second welding parameters at the next moment corresponding to the sample moments; Using the characteristic values of the first welding parameters at the sample moments as training samples and the characteristic values of the second welding parameters at the next moment corresponding to the sample moments as training labels, train the initial prediction model until the training stop condition is satisfied to obtain the welding prediction model.

[0012] According to the second aspect of the embodiments of the present application, there is provided a welding parameter prediction device, the device includes: A first processing module, configured to obtain the characteristic values of the first welding parameters at the first moment; wherein, the first welding parameters include at least one of the material parameters of the welding operation object and the welding process parameters; A second processing module, configured to input the characteristic values of the first welding parameters into a pre-trained prediction model; wherein, the prediction model is trained by a graph convolutional network and a gated recurrent unit using the parameter values of the first welding parameters at multiple sample moments as training samples; each of the training samples further includes a training label, and the training label is the characteristic value of the second welding parameters at the next moment corresponding to the sample moment; wherein, the second welding parameters include: welding time and / or welding power, and the first welding parameters are the independent variables of the second welding parameters; A third processing module, configured to obtain the characteristic values of the second welding parameters at the next moment of the first moment output by the prediction model.

[0013] According to the third aspect of the embodiments of the present application, there is provided an electronic device, including: a memory, a processor, and a computer program stored on the memory, and the processor executes the computer program to implement the steps of the method according to any one of the first aspect.

[0014] According to the fourth aspect of the embodiments of the present application, there is provided a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the method according to any one of the first aspect.

[0015] The beneficial effects brought by the technical solutions provided by the embodiments of the present application are: In the embodiments of the present application, the parameter values of the first welding parameters at multiple sample times are used as training samples, and the eigenvalue of the second welding parameter at the next time of the corresponding sample time is used as a training label to train a graph convolutional network and a gated recurrent unit to obtain a prediction model; the eigenvalue of the first welding parameter at the first time is obtained, and the eigenvalue of the first welding parameter is input into the pre-trained prediction model to obtain the eigenvalue of the second welding parameter at the next time of the first time output by the prediction model. Compared with the traditional solution of obtaining the welding time and welding power through actual welding tests, the embodiments of the present application can accurately predict the welding time and welding power of the battery cell, effectively improve the accuracy of the prediction results of the welding time and welding power, instruct the staff to intervene in a timely manner according to the prediction results, and improve the welding quality. Description of the Drawings

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description in the embodiments of the present application.

[0017] Figure 1 It is a schematic flowchart of a welding parameter prediction method provided by an embodiment of the present application; Figure 2 It is a schematic structural diagram of a prediction model provided by an embodiment of the present application; Figure 3 It is a schematic structural diagram of a graph convolutional module provided by an embodiment of the present application; Figure 4 It is a schematic structural diagram of an attention module provided by an embodiment of the present application; Figure 5 It is a schematic structural diagram of a feature extraction module provided by an embodiment of the present application; Figure 6 It is a schematic structural diagram of a second welding parameter prediction module provided by an embodiment of the present application; Figure 7 It is a schematic structural diagram of a gated recurrent unit provided by an embodiment of the present application; Figure 8 It is a schematic flowchart of training a prediction model provided by an embodiment of the present application; Figure 9 It is a schematic structural diagram of a welding parameter prediction device provided by an embodiment of the present application; Figure 10 It is a schematic structural diagram of an electronic device provided by an embodiment of the present application. Detailed Embodiments

[0018] The embodiments of the present application will be described below with reference to the accompanying drawings in the present application. It should be understood that the embodiments described below with reference to the accompanying drawings are exemplary descriptions for explaining the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions of the embodiments of the present application.

[0019] Those skilled in the art of the present technology can understand that unless specifically stated otherwise, the singular forms "a", "an", "the" and "said" used herein may also include plural forms. It should be further understood that the terms "comprising" and "including" used in the embodiments of the present application mean that the corresponding features can be implemented as the presented features, information, data, steps, operations, elements and / or components, but do not exclude the implementation of other features, information, data, steps, operations, elements, components and / or their combinations supported by the art of the present technology. It should be understood that when we say an element is "connected" or "coupled" to another element, the one element can be directly connected or coupled to the other element, or it can mean that the one element and the other element establish a connection relationship through an intermediate element. In addition, the "connection" or "coupling" used here can include wireless connection or wireless coupling. The term "and / or" used here indicates at least one of the items defined by the term, for example, "A and / or B" can be implemented as "A", or implemented as "B", or implemented as "A and B".

[0020] In the embodiments of the present application, the term "module" or "unit" refers to a computer program with a predetermined function or a part of a computer program, which works together with other related parts to achieve a predetermined goal, and can be fully or partially implemented by using software, hardware (such as a processing circuit or a memory), or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be a part of the overall module or unit that includes the function of the module or unit.

[0021] To make the purpose, technical solutions and advantages of the present application clearer, the embodiments of the present application will be further described in detail below with reference to the accompanying drawings.

[0022] Tab ultrasonic welding is a key step in the production of battery cells, and its welding quality directly affects the performance, safety and production cost of the battery cells. During the selection of welding parameters, welding time and welding power are two key factors.

[0023] In the related art, key data such as welding time and welding power can only be obtained after actual welding experiments. Since the welding time and welding power can only be obtained afterwards, the identification and solution of welding problems are postponed. Therefore, it has become an urgent task to predict the welding time and welding power and intervene in a timely manner according to the prediction results to improve the welding quality of the battery.

[0024] The welding parameter prediction method, device, electronic device and storage medium provided by this application aim to solve the above technical problems in the prior art.

[0025] The technical solutions of the embodiments of this application and the technical effects produced by the technical solutions of this application will be described below through the description of several exemplary embodiments. It should be noted that the following embodiments can refer to, draw on or combine with each other. The same terms, similar features and similar implementation steps in different embodiments will not be described repeatedly.

[0026] It can be understood that in the welding parameter prediction method provided by the embodiments of this disclosure, any method step can be executed by an electronic device and / or a server. All steps in the method can be independently executed by the electronic device or the server, or jointly executed by the electronic device and the server.

[0027] Among them, the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The electronic device can be a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart voice interaction device (such as a smart speaker), a wearable electronic device (such as a smart watch), a vehicle-mounted terminal, a smart home appliance (such as a smart TV), an AR / VR device, etc., but is not limited thereto.

[0028] Subsequently, the embodiments of this disclosure will be introduced with the server as the execution subject. However, this does not constitute a limitation on the embodiments of this disclosure. The method provided by the embodiments of this disclosure can also be applied to the prediction of welding time and / or welding power during the welding process in other scenarios in addition to the welding scenario during the production of battery cells.

[0029] Figure 1 It is a schematic flowchart of a welding parameter prediction method provided by an embodiment of this application. As shown in the figure, the method includes: S101. Obtain the feature value of the first welding parameter at the first moment; wherein, the first welding parameter includes at least one of the material parameter and the welding process parameter of the welding operation object.

[0030] In the embodiments of this application, the welding operation object can be the battery cell. Taking the cell as an example, the first welding parameter and the second welding parameter provided by the embodiments of this disclosure will be described in detail below.

[0031] Specifically, in the embodiments of the present application, the material parameters include but are not limited to: the number of tab layers of the battery cell, the thickness of the foil material, the depth of the weld pattern, and the welding area; the welding process parameters include but are not limited to: energy, pressure, and amplitude. It should be noted that in the embodiments of the present disclosure, the specific material parameters and welding process parameters can be defined according to the actual situation, and the embodiments of the present application do not limit the specific parameter types of the material parameters and welding process parameters.

[0032] Specifically, in the embodiments of the present application, the welding process of the battery cell includes multiple consecutive moments. In the embodiments of the present application, the eigenvalue of the first welding parameter at any one of the multiple consecutive moments is collected, and the collection moment is used as the first moment.

[0033] Optionally, in the embodiments of the present application, the eigenvalue of the first welding parameter at the first moment can be obtained by means such as being automatically recorded by the welding device, measuring and outputting the eigenvalue of the first welding parameter by the set monitoring device and recording, etc. The specific obtaining method can be determined according to actual needs.

[0034] S102. Input the eigenvalue of the first welding parameter into a pre-trained prediction model; Among them, the prediction model is trained by a graph convolutional network and a gated recurrent unit using the parameter values of the first welding parameter at multiple sample moments as training samples; each training sample also includes a training label, and the training label is the eigenvalue of the second welding parameter at the next moment corresponding to the sample moment; among them, the second welding parameter includes: welding time and / or welding power, and the first welding parameter is the independent variable of the second welding parameter.

[0035] In the embodiments of the present application, the first welding parameter is the independent variable of the second welding parameter. In other words, there is a causal relationship between the first welding parameter and the second welding parameter; for example: the number of tab layers and the thickness of the foil material affect the heat conduction during welding and the required energy. The greater the number of tab layers and the thickness of the foil material, the greater the welding area or higher power is required to ensure a firm connection; an increase in the welding area may require a longer welding time or a higher amplitude to cover a larger area; in ultrasonic welding, the amplitude and pressure directly affect the welding energy, and the energy is related to the welding time and power. A higher amplitude may require a higher power, while a greater pressure may shorten the welding time. Therefore, on the premise that there is a causal relationship between the first welding parameter and the second welding parameter, the embodiments of the present application input the eigenvalue of the first welding parameter at the first moment into the pre-trained training model to obtain the eigenvalue of the second welding parameter.

[0036] Specifically, the prediction model is obtained through pre-training. In the embodiments of the present application, the prediction model includes a graph convolutional network and a gated recurrent network. Among them, the graph convolutional network extracts the spatial features of the input data through graph convolutional operations, and can effectively model the mutual relationship between the first welding parameters. For example, the interaction between the number of tab layers and the foil thickness. The gated recurrent network processes parameter changes through a gating mechanism and captures the dynamic changes of parameters during the welding process. By combining the spatial feature extraction ability of the graph convolutional network and the dynamic information processing ability of the gated recurrent network, the embodiments of the present application can capture the input data features more comprehensively, thereby significantly improving the prediction performance of the welding time and welding power.

[0037] Specifically, in the embodiments of the present application, during the pre-training of the prediction model, the eigenvalue of the first welding parameter at multiple moments during the welding process is collected in advance as sample data; at the same time, the eigenvalue of the second welding parameter at the next moment of multiple consecutive moments is collected as sample labels. The prediction model is trained according to the sample data and sample labels to obtain a trained prediction model, which can output the eigenvalue of the second welding parameter at the next moment according to the eigenvalue of the first welding parameter at the first moment input.

[0038] S103. Obtain the eigenvalue of the second welding parameter at the next moment of the first moment output by the prediction model.

[0039] In the embodiments of the present application, by inputting the eigenvalue of the first welding parameter at the first moment into the prediction model, the eigenvalue of the second welding parameter at the next moment of the first moment can be obtained. For example, if the first moment is 10:20:00, the time step is 1 second (i.e., welding once per second), then the next moment of the first moment is 10:20:01, that is, the prediction model outputs the eigenvalue of the second welding parameter at 10:20:01.

[0040] Optionally, since the welding of the battery cell is a continuous process, there may be errors in training the prediction model using the sample data and sample labels of a single moment, and the change trend of the welding time and welding power cannot be reflected. Therefore, in the embodiments of the present application, the eigenvalues of the first welding parameter at multiple consecutive moments are collected, and in chronological order, the eigenvalues of the first welding parameter at each moment are input into the prediction model in turn to obtain the eigenvalues of the second welding parameter at the next moment of each moment, so as to better reflect the change trend of the eigenvalues of the second welding parameter, enabling the staff to intervene in the welding process according to the change trend of the second welding parameter and improving the welding quality.

[0041] Optionally, in the embodiments of the present application, after obtaining the eigenvalue of the second welding parameter at the next moment of the first moment, the eigenvalue of the second welding parameter at the next moment can be compared with the preset range of the second welding parameter. If the eigenvalue of the second welding parameter at the next moment is within the preset range, no intervention by the staff is required. If the eigenvalue of the second welding parameter at the next moment is not within the preset range, at least one of the material parameters and the welding process parameters needs to be adjusted.

[0042] In the embodiments of the present application, the prediction model is trained with the parameter values of the first welding parameter at multiple sample moments through a graph convolutional network and a gated recurrent network. The graph convolutional network is good at capturing the spatial features of data, and the gated recurrent network is good at capturing the temporal features of data. Therefore, the prediction model can more accurately reflect the complex relationship between welding parameters, which makes the output eigenvalue of the second welding parameter more accurate, helps to achieve accurate prediction of the welding process, and solves the technical problem in the prior art that the welding time and welding power can only be obtained afterwards, resulting in the delay in the identification and solution of welding problems.

[0043] Based on the above embodiments, as an optional embodiment, the graph convolutional network includes: a graph convolutional module and an attention module; the gated recurrent network includes: a feature extraction module and a second welding parameter prediction module; Inputting the eigenvalue of the first welding parameter into the pre-trained prediction model to obtain the eigenvalue of the second welding parameter at the next moment of the first moment output by the prediction model includes: Inputting the eigenvalue of the first welding parameter into the graph convolutional module and the attention module respectively to obtain the target feature matrix output by the graph convolutional module and the target feature vector output by the attention module; obtaining the updated eigenvalue of the first welding parameter according to the target feature matrix and the target feature vector; Inputting the updated eigenvalue of the first welding parameter into the feature extraction module to obtain the target hidden state output by the feature extraction module, where the target hidden state indicates the dependence relationship between the first welding parameters; inputting the target hidden state into the second welding parameter prediction module to output the eigenvalue of the second welding parameter at the next moment.

[0044] Specifically, Figure 2A structural schematic diagram of a prediction model provided by an embodiment of the present application is shown in the figure. The prediction model 20 sequentially includes a graph convolutional network 201 and a gated recurrent network 202. Among them, the graph convolutional network 201 includes a graph convolutional module 30 and an attention module 40, and the gated recurrent network 202 includes a feature extraction module 50 and a second welding parameter prediction module 60. The eigenvalue of the first welding parameter at the first moment is input into the prediction model 20, and after being processed by the graph convolutional network 201 and the gated recurrent network 202, the eigenvalue of the second welding parameter at the next moment of the first moment is output.

[0045] In the embodiment of the present application, the eigenvalues of the first welding parameter are respectively input into the graph convolutional module 30 and the attention module 40. The graph convolutional module 30 extracts and fuses features by using the mutual relationship between the first welding parameters to form a target feature matrix, and the target feature matrix reflects the mutual relationship of the welding parameters in the graph structure. The attention module 40 outputs a target feature vector according to the weights of different feature channels, and the target feature vector highlights the feature channels that are most critical for predicting the welding parameters at the next moment. Then, the target feature matrix output by the graph convolutional module 30 and the target feature vector output by the attention module 40 are subjected to a dot product operation, and the result of the dot product operation is used to update the eigenvalue of the first welding parameter. The updated eigenvalue of the first welding parameter is input into the feature extraction module 50, and this module further processes these features and outputs a target hidden state. The target hidden state not only contains the feature information of the welding parameters themselves, but also implies the dependence relationship between these parameters, which is crucial for understanding the complex interactions in the welding process. Finally, the target hidden state is input into the second welding parameter prediction module 60, and this module uses the information in the hidden state to predict the eigenvalue of the second welding parameter at the next moment. Since the hidden state has captured the dependence relationship and key features between the welding parameters, the prediction result can more accurately reflect the change trend of the second welding parameter.

[0046] In the embodiment of the present application, by combining the graph convolutional module and the attention module, the feature information of the welding parameters is effectively extracted and utilized, and the prediction accuracy of the welding parameters at the next moment is improved.

[0047] On the basis of the above embodiments, as an optional embodiment, the graph convolutional module includes at least one graph convolutional unit; there is a predetermined mutual relationship between the first welding parameters; Inputting the eigenvalue of the first welding parameter into the graph convolutional module to obtain the target feature matrix output by the graph convolutional module includes: Determining the feature map at the first moment according to the eigenvalue of the first welding parameter; wherein, the nodes of the feature map are the first welding parameters, and the edges between the nodes are the predetermined mutual relationships between the first welding parameters; The feature maps are sequentially input into each graph convolutional unit to obtain the feature matrices output by each graph convolutional unit, and the feature matrix output by the last graph convolutional unit is used as the target feature matrix.

[0048] Specifically, Figure 3 As shown in the structural schematic diagram of a graph convolutional module provided by an embodiment of the present application, Figure 3 As shown, the graph convolutional module 30 includes at least one graph convolutional unit 301. Each graph convolutional unit 301 is connected in series. The eigenvalue of the first welding parameter at the first moment is input into the graph convolutional module 30, and through the convolutional operations of each graph convolutional unit 301 in sequence, the last graph convolutional unit 301 outputs the target feature matrix.

[0049] Specifically, in the embodiment of the present application, the first welding parameter is used as the node of the feature map, and according to the pre-determined mutual relationship between the welding parameters, the edges between the nodes are constructed; in this way, the feature map intuitively represents the welding parameters and their mutual relationship; the constructed feature map is sequentially input into each graph convolutional unit 301. The graph convolutional unit 301 can extract the feature information in the feature map by using convolutional operations. Through the stacking of multiple graph convolutional units 301, more complex features can be iteratively extracted from the low-level features; among them, each graph convolutional unit 301 will output a feature matrix, and this matrix contains the feature information extracted from the feature map. The feature matrix output by the last graph convolutional unit 301 is used as the target feature matrix, and the target feature matrix synthesizes the processing results of all graph convolutional units 301 and contains the high-level feature representation of the welding parameters.

[0050] It should be noted that in the embodiment of the present application, after obtaining the eigenvalue of the first welding parameter at the first moment, it is necessary to generate a corresponding feature map according to the eigenvalue of the first welding parameter. Among them, the nodes in the feature map are the first welding parameters, and the edges between the nodes are the mutual relationships between the first welding parameters. For example: the first welding parameters include: the number of tabs of the battery cell, the foil thickness, the weld pattern depth, the welding area, the energy, the pressure, and the amplitude. These welding parameters are used as the nodes of the feature map, and the mutual relationships between these welding parameters are used as the edges between the corresponding nodes; it should be noted that in the embodiment of the present application, the mutual relationship between the first welding parameters is pre-determined. For example, for the mutual relationship between the number of tabs and the foil thickness, there is an edge between the node of the number of tabs and the node of the foil thickness.

[0051] Furthermore, according to the feature map, the feature matrix X, the adjacency matrix A, and the degree matrix D of the nodes can be obtained; among them, the feature matrix A is used to describe the initial eigenvalues of each node, the adjacency matrix X is used to describe the connection relationships between each node, and the degree matrix is used to describe the number of edges connected to each node.

[0052] Furthermore, the normalized adjacent features of the feature map can be obtained. , where the formula (1) of the normalized adjacency matrix is shown as follows:

[0053] where, is 's degree matrix, is obtained by adding the self-loop matrix I to the adjacency matrix A, and the formula (2) is shown as follows:

[0054] Furthermore, input the feature map into the first graph convolutional unit 301, and output the feature matrix , where the calculation formula (3) of the feature matrix is shown as follows:

[0055] where, is the weight matrix of the first graph convolutional unit 301, is the activation function (e.g., ReLU).

[0056] When the number of graph convolutional units 301 is greater than 1, input the feature matrix output by the first graph convolutional unit 301 into each subsequent graph convolutional unit 301, and sequentially obtain the feature matrices output by each graph convolutional unit 301. In the embodiment of the present application, the feature matrix output by the last graph convolutional unit 301 is used as the target feature matrix; where the calculation formula (4) of the target feature matrix is shown as follows:

[0057] where, is the feature matrix output by the th graph convolutional unit 301; is the feature matrix output by the lth graph convolutional unit 301, is the weight matrix of the lth graph convolutional unit 301. When l is equal to 1, it is the feature matrix output by the second graph convolutional unit 301; when l is equal to 2, it is the feature matrix output by the third graph convolutional unit 301, and so on, until the feature matrix output by the last graph convolutional unit 301 is obtained as the target feature matrix.

[0058] In the embodiment of the present application, by processing the feature map through the graph convolutional unit, feature information can be effectively extracted from the first welding parameters; meanwhile, the target feature matrix contains the high-level feature representation of the welding parameters, which is of great significance for subsequent tasks such as predicting the second welding parameters and optimizing the welding process.

[0059] Based on the above embodiments, as an alternative embodiment, the attention module includes at least one attention unit and a global pooling unit; wherein, each attention unit includes an attention channel corresponding to the first welding parameter; Inputting the eigenvalue of the first welding parameter into the attention module to obtain the target feature vector output by the attention module, including: Determine the input data of each attention unit; wherein, for the first attention unit, the input data is the eigenvalue of the first welding parameter; for non-first attention units, the input data is the eigenvalue output after convolutional calculation by the previous attention unit; For each attention unit, input the input data corresponding to each first welding parameter into the corresponding attention channel, and perform convolutional calculation in the corresponding attention channel to obtain the eigenvalues output by each attention channel; wherein, the weight values of each attention channel are different; Input the eigenvalues output by each attention channel in the last attention unit into the global pooling unit, extract the maximum value among all the eigenvalues in each attention channel as the updated eigenvalue of the corresponding first welding parameter, and form a one-dimensional feature vector with the updated eigenvalues of the first welding parameter as the target feature vector.

[0060] Specifically, Figure 4 FIG. is a schematic structural diagram of an attention module provided by an embodiment of the present application, as Figure 4 shown: The attention module 40 includes at least one attention unit 401 and a global pooling unit 402; wherein, the attention units 401 are connected in sequence and then connected in series with the global pooling unit 402. Input the eigenvalue of the first welding parameter at the first moment into the attention module 40 to output the target feature vector.

[0061] Specifically, in the embodiments of the present application, for the first attention unit 401, the input data is the eigenvalue of the first welding parameter; for non-first attention units 401, the input data is the eigenvalue output after the convolutional calculation of the previous attention unit 401. This design enables each attention unit 401 to perform further feature extraction and update based on the processing result of the previous unit. Further, each attention unit 401 includes multiple attention channels, and each channel corresponds to a different weight value. The input data corresponding to each first welding parameter is input into the corresponding attention channel, and convolutional calculation is performed in each attention channel to obtain the eigenvalue output by the channel. Since the weight values of each channel are different, the features extracted by each channel will also be different. Then, the eigenvalue outputs of each attention channel in the last attention unit 401 are input into the global pooling unit 402, and the global pooling unit 402 extracts the maximum value among all the eigenvalue in each attention channel as the updated eigenvalue of the corresponding first welding parameter. This design helps to retain the most significant feature information in each channel. Finally, the updated eigenvalues of the first welding parameter are combined into a one-dimensional feature vector, which contains the updated eigenvalues obtained after all welding parameters are processed by multiple attention units 401 and the global pooling unit 402, and can be regarded as the target feature vector.

[0062] In the embodiments of the present application, through the processing of multiple attention units and attention channels, the eigenvalue of the first welding parameter can be gradually extracted and updated, making the finally obtained feature vector more accurate and comprehensive. At the same time, the different weight values of each attention channel help the model to focus on different feature information, improving the generalization ability and robustness of the model. In addition, the global pooling unit extracts the maximum value in each channel, which helps to retain the most significant feature information and reduce the dimension of the feature vector, thereby reducing the computational complexity. The finally obtained target feature vector contains the updated eigenvalues of all the first welding parameters, effectively improving the prediction accuracy of the subsequent second welding parameter.

[0063] Based on the above embodiments, as an optional embodiment, the feature extraction module includes multiple gated recurrent units and a fully connected unit. Input the updated eigenvalue of the first welding parameter into the feature extraction module to obtain the target hidden state output by the feature extraction module, including: For the first gated recurrent unit, the input data is the updated eigenvalue of the first welding parameter, and the output data is the hidden state of the first gated recurrent unit. For non-first gated recurrent units, the input data is the updated eigenvalue of the first welding parameter and the hidden state output by the previous gated recurrent unit, and the output is the hidden state of the corresponding gated recurrent unit. The hidden state output by the last gated recurrent unit is input into a fully connected unit for fully connected processing to obtain the hidden state output by the fully connected unit as the target hidden state.

[0064] Specifically, Figure 5 FIG. is a schematic structural diagram of a feature extraction module provided by an embodiment of the present application. As Figure 5 shown: The feature extraction module 50 includes a plurality of gated recurrent units 501 and a fully connected unit 502; wherein, the input data of each gated recurrent unit 501 is the parameter value of the first welding parameter updated at the first moment, and the hidden state output by the previous gated recurrent unit 501; the last gated recurrent unit 501 and the fully connected unit 502 are connected in series, and the last gated recurrent unit inputs the output hidden state into the fully connected unit 502, and the fully connected unit 502 outputs the target hidden state.

[0065] Specifically, for the first gated recurrent unit 501, its input is the feature value after the first welding parameter is updated. The first welding parameter includes at least one of a material parameter and a welding process parameter, and they are obtained after preprocessing and feature extraction of the feature value of the first welding parameter; for the first gated recurrent unit 501, the output is the hidden state of the first gated recurrent unit 501, and this hidden state contains the information in the feature value of the first welding parameter.

[0066] Specifically, for non-first gated recurrent units 501, its input includes two parts: one is the feature value after the first welding parameter is updated (this part is the same in all gated recurrent units 501), and the other is the hidden state output by the previous gated recurrent unit 501. Such a design enables each gated recurrent unit 501 to utilize the information of the previous unit, thereby transmitting and integrating information in the entire sequence; in addition, the output of the non-first gated recurrent unit 501 is the hidden state of the current gated recurrent unit 501, and this state contains all relevant information from the start of the sequence to the current position.

[0067] Specifically, the hidden state output by the last gated recurrent unit 501 is input into the fully connected unit 502, and the fully connected unit 502 further processes the hidden state, usually by linear transformation and activation functions (such as ReLU, sigmoid, etc.) to extract higher-level features. The output of the fully connected unit 502 is used as the target hidden state, and this state can be used for subsequent tasks such as classification, regression, or sequence generation.

[0068] In the embodiments of the present application, each gated recurrent unit outputs a hidden state, and these states are deep representations of the input data. As the data is passed between the gated recurrent units, features are gradually extracted and deepened. The target hidden state output by the fully connected unit can be used as the input for subsequent tasks; since the fully connected layer has high flexibility, it can adapt to different task requirements.

[0069] Based on the above embodiments, as an optional embodiment, the second welding parameter prediction module includes a plurality of gated recurrent units and a global pooling unit; Input the target hidden state into the second welding parameter prediction module, and output the eigenvalue of the second welding parameter at the next moment, including: Determine the input data of each gated recurrent unit; among them, for the first gated recurrent unit, the input data is the target hidden state; for non-first gated recurrent units, the input data is the hidden state output by the previous gated recurrent unit; For each gated recurrent unit, perform a dot product operation on the hidden state output by itself, and output the predicted value of the second welding parameter; Input the predicted values of the second welding parameter output by each gated recurrent unit into the global pooling unit, and perform pooling processing on all the predicted values of the second welding parameter to obtain the eigenvalue of the second welding parameter at the next moment.

[0070] Specifically, Figure 6 FIG. is a schematic structural diagram of a second welding parameter prediction module provided by an embodiment of the present application, as Figure 6 shown: The second welding parameter prediction module 60 includes a plurality of gated recurrent units 601 and a global pooling unit 602; among them, each gated recurrent unit 601 inputs the result of performing a dot product operation on the hidden state output by itself into the global pooling unit 602; at the same time, the hidden state output by the previous gated recurrent unit 601 serves as the input hidden state of the subsequent gated recurrent unit 601.

[0071] Specifically, in the embodiments of the present application, for the first gated recurrent unit 601, the input data is the target hidden state, and this state may contain prior information or context information related to the welding task; for non-first gated recurrent units 601, the input data is the hidden state output by the previous gated recurrent unit 601, so that each gated recurrent unit 601 can use the information of the previous unit to update its hidden state.

[0072] Specifically, in the embodiments of the present application, each gated recurrent unit 601 outputs an updated hidden state, and performs a dot product operation on the output hidden state to generate a predicted value of the second welding parameter from the hidden state. The predicted values of the second welding parameter output by each gated recurrent unit 601 are input into the global pooling unit 602, and all the predicted values are summarized or averaged to generate a single, representative predicted value as the eigenvalue of the second welding parameter at the next moment.

[0073] In the embodiments of the present application, through the gated recurrent unit, the information in the sequence can be integrated to capture the changing trend of the welding parameter over time. The global pooling unit performs a summarization process on all the predicted values, which helps to generate the eigenvalue of the second welding parameter at the next moment.

[0074] Figure 7 FIG. shows a schematic structural diagram of a gated recurrent unit provided by an embodiment of the present application. As shown in the figure, the gated recurrent unit includes a reset gate and an update gate. For the current gated recurrent unit, according to the input data and the hidden state output by the previous gated recurrent unit , the reset gate and the update gate are calculated through the current gated recurrent unit. The formula (5) of the reset gate and the formula (6) of the update gate are as follows:

[0075]

[0076] In the formula, is a fully connected layer and the activation function Sigmoid function, and are the weights of the update gate and the reset gate respectively, and are the bias amounts of the reset gate and the update gate respectively.

[0077] According to the reset gate , the target input and the hidden state output by the previous gated recurrent unit , the candidate hidden state is obtained. The reset gate controls how the hidden state output by the previous gated recurrent unit flows into the candidate hidden state of the current gated recurrent unit. The hidden state output by the previous gated recurrent unit contains all the historical information of the eigenvalue of the first welding parameter. Therefore, the reset gate can be used to discard the historical information irrelevant to the prediction.

[0078] According to the update gate , the hidden state output by the previous gated recurrent unit and the candidate hidden state , obtain the hidden state output by the current gated recurrent unit , the update gate can control how the hidden state should be updated by the candidate hidden state .

[0079] The formula (7) of the candidate hidden state and the formula (8) of the hidden state output by the current gated recurrent unit are as follows:

[0080]

[0081] In the formula, tanh refers to the hyperbolic tangent activation function, refers to the corresponding weight when calculating the candidate hidden state.

[0082] Specifically, the input data of each gated recurrent unit in the feature extraction module are the eigenvalue of the first welding parameter at the first moment, that is: each in the above formulas (5)-(7) for each gated recurrent unit is the eigenvalue of the first welding parameter at the first moment; further, in the embodiments of the present application, starting from the second gated recurrent unit, represents the hidden state output by the previous gated recurrent unit, and here "t" can be understood as the serial number of the gated recurrent unit.

[0083] Specifically, in the second welding parameter prediction module, the input of each gated recurrent unit is only the hidden state. In other words, in the embodiments of the present application, in the above formulas (5)-(7) can be regarded as 0, which can be understood as that the embodiments of the present application have improved the conventional gated recurrent unit.

[0084] It should be noted that for the first gated recurrent unit, since there is no previous gated recurrent unit, therefore, the hidden state output by the previous gated recurrent unit can be initialized, that is: determine the initial hidden state. Usually, the initial hidden state can be set to a zero vector or a random small value vector.

[0085] Based on the above embodiments, as an optional embodiment, the prediction model is generated in the following manner, including: Pre-acquire the parameter values of the first welding parameters and the parameter values of the second welding parameters at multiple moments, and preprocess the parameter values of the first welding parameters and the parameter values of the second welding parameters at each moment to obtain the eigenvalue of the first welding parameter and the eigenvalue of the second welding parameter at each moment; wherein, the preprocessing includes at least one of data cleaning and normalization processing; According to the eigenvalue of the first welding parameter and the eigenvalue of the second welding parameter at each moment, determine the eigenvalue of the first welding parameter at multiple sample moments and the eigenvalue of the second welding parameter at the next moment corresponding to the sample moment; Use the eigenvalue of the first welding parameter at the sample moment as the training sample, and use the eigenvalue of the second welding parameter at the next moment corresponding to the sample moment as the training label, and train the initial prediction model until the training stop condition is met to obtain the welding prediction model.

[0086] Specifically, Figure 8 As shown in the figure, which is a schematic flow chart for training a prediction model provided by an embodiment of the present application. In the embodiment of the present application, the parameter values of the first welding parameters (such as material parameters and welding process parameters) at multiple moments and the parameter values of the second welding parameters (such as welding time and welding power) at the next moment of multiple moments are pre-collected as sample data. These parameter values should cover data under different welding conditions to ensure the generalization ability of the model.

[0087] Specifically, preprocess the collected sample data, including: cleaning, removing outliers and missing values to ensure the accuracy and integrity of the data; optionally, the preprocessing of the sample data also includes normalization processing, converting parameter values with different dimensions to the same scale to improve the training efficiency and prediction accuracy of the model; the normalization method can be selected according to the data characteristics, such as min-max normalization, Z-score standardization, etc.

[0088] Specifically, according to the preprocessed data, determine the eigenvalue of the first welding parameter at each moment and the eigenvalue of the second welding parameter at the next moment; use the eigenvalue of the first welding parameter at the sample moment as the training sample, and use the eigenvalue of the second welding parameter at the next moment corresponding to the sample moment as the training label. The sample set constructed in this way can reflect the change relationship of the welding parameters over time and provide a basis for predicting the welding result at the next moment.

[0089] Specifically, in the embodiment of the present application, a graph convolutional network and a gated recurrent network are used as the initial prediction model. Input the constructed sample set into the prediction model to be trained for training, and continuously adjust the model parameters to reduce the prediction error. When the prediction model meets the training stop condition (such as the prediction error reaches the preset threshold, the number of training iterations reaches the upper limit, etc.), stop training and obtain the final prediction model.

[0090] Specifically, after obtaining the final prediction model, the eigenvalue of the first parameter at the first moment is input into the prediction model to obtain the eigenvalue of the second welding parameter at the next moment of the first moment.

[0091] Optionally, the Adam algorithm can be selected to train the prediction model during the training phase. Among them, the Adam algorithm is one of the commonly used optimization algorithms in deep neural networks, which combines the advantages of momentum optimization and adaptive learning rate, effectively updates network parameters and accelerates model convergence. The core of Adam is to combine the first moment (mean) and second moment (variance) estimates of the gradient, calculate these two using exponential weighted moving average, and perform bias correction on the estimates.

[0092] Optionally, the mean squared error (MSE) is used as the loss function in this embodiment of the application to measure the gap between the model prediction value and the actual value. The mean squared error is commonly used in regression problems and is suitable for situations where continuous numerical values need to be predicted. It calculates the average of the squares of the differences between the prediction value and the actual value, thereby quantifying the size of the prediction error, helping to optimize the model to minimize the prediction error, and improving the accuracy and precision of the prediction. The calculation formula (9) of the mean squared error MSE is as follows:

[0093] In the formula, MSE represents the mean squared error loss function, represents the actual value, represents the prediction value, and n is the number of samples.

[0094] Optionally, R² is also used as the convergence criterion for the model in this embodiment of the application. The mean squared error (MSE) is used to measure the gap between the model prediction value and the actual value, while R² evaluates the ability of the model to explain the data variance. In a regression task, R² is used to measure the goodness of fit of the model, and its value ranges from 0 to 1. The closer the value is to 1, the better the model fits. By introducing R² as the convergence criterion, the fitting situation during the model training process can be monitored, and the prediction ability of the model can be evaluated, thereby further optimizing the model and improving its prediction accuracy and stability in practical applications. The calculation formula (10) of R² is as follows:

[0095] Among them, is denoted as the sum of squared residuals, which is used to describe the sum of the squares of the differences between the model prediction value and the actual value; ‌is denoted as the total sum of squared deviations, which is used to describe the sum of the squares of the differences between the actual value and the mean.‌

[0096] Specifically, in the embodiments of the present application, based on the trained prediction model, the eigenvalue of the first welding parameter at the current moment obtained can be used as the input of the model, and the predicted value of the second welding parameter at the next moment output by the model can be obtained. Then, according to the predicted value of the output second welding parameter, it can be determined whether it is necessary to intervene in the welding process to improve the welding quality.

[0097] In the embodiments of the present application, through data preprocessing and feature extraction, the influence of noise and redundant information on the model is reduced, and the prediction accuracy is improved. At the same time, the prediction model can help to discover and solve potential problems in advance, improve the welding efficiency and quality, contribute to reducing the scrap rate and rework rate, and thus reduce the production cost.

[0098] Figure 9 FIG. is a schematic structural diagram of a welding parameter prediction device provided by an embodiment of the present application. As shown in the figure, the device may include: a first processing module 901, a second processing module 902, and a third processing module 903.

[0099] The first processing module 901 is configured to obtain the eigenvalue of the first welding parameter at the first moment; wherein, the first welding parameter includes at least one of the material parameter of the welding operation object and the welding process parameter; The second processing module 902 is configured to input the eigenvalue of the first welding parameter into a pre-trained prediction model; wherein, the prediction model is trained by a graph convolutional network and a gated recurrent unit using the parameter values of the first welding parameter at multiple sample moments as training samples; each training sample further includes a training label, and the training label is the eigenvalue of the second welding parameter at the next moment corresponding to the sample moment; wherein, the second welding parameter includes: welding time and / or welding power, and the first welding parameter is the independent variable of the second welding parameter; The third processing module 903 is configured to obtain the eigenvalue of the second welding parameter at the next moment of the first moment output by the prediction model.

[0100] The welding parameter prediction device in the embodiments of the present application can execute the welding parameter prediction method provided by the embodiments of the present application, and its implementation principle is similar. The actions performed by each module in the welding parameter prediction device in the embodiments of the present application correspond to the steps in the welding parameter prediction method in the embodiments of the present application. For the detailed function descriptions of each module of the welding parameter prediction device, reference may specifically be made to the descriptions in the corresponding methods shown above, and details are not described herein again.

[0101] In the embodiment of the present application, the parameter values of the first welding parameters at multiple sample times are used as training samples, and the characteristic values of the second welding parameters at the next time of the corresponding sample time are used as training labels to train a graph convolutional network and a gated recurrent unit to obtain a prediction model; the characteristic values of the first welding parameters at the first time are obtained, and the characteristic values of the first welding parameters are input into the pre-trained prediction model to obtain the characteristic values of the second welding parameters at the next time of the first time output by the prediction model. Compared with the traditional solution of obtaining the welding time and welding power through actual welding tests, the embodiment of the present application can accurately predict the welding time and welding power of the battery cell, effectively improve the accuracy of the prediction results of the welding time and welding power, instruct the staff to intervene in a timely manner according to the prediction results, and improve the welding quality.

[0102] Figure 10 FIG. is a schematic structural diagram of an electronic device provided by an embodiment of the present application, as Figure 10 shown, the electronic device 4000 includes: a processor 4001 and a memory 4003. Among them, the processor 4001 and the memory 4003 are connected, such as connected through a bus 4002. Optionally, the electronic device 4000 may further include a transceiver 4004, and the transceiver 4004 may be used for data interaction between the electronic device and other electronic devices, such as data sending and / or data receiving, etc. It should be noted that in actual applications, the transceiver 4004 is not limited to one, and the structure of the electronic device 4000 does not constitute a limitation to the embodiment of the present application.

[0103] The processor 4001 may be a CPU (Central Processing Unit, central processor), a general-purpose processor, a DSP (Digital Signal Processor, data signal processor), an ASIC (Application Specific Integrated Circuit, application-specific integrated circuit), an FPGA (Field Programmable Gate Array, field programmable gate array) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It can implement or execute various exemplary logical blocks, modules and circuits described in connection with the disclosure of the present application. The processor 4001 may also be a combination that implements a computing function, such as a combination including one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0104] The bus 4002 may include a path for transmitting information among the above components. The bus 4002 can be a PCI (Peripheral Component Interconnect) bus, an EISA (Extended Industry Standard Architecture) bus, or the like. The bus 4002 can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience of representation, Figure 10 it is only represented by a thick line in Figure 10 , but it does not mean that there is only one bus or one type of bus.

[0105] The memory 4003 can be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or it can also be an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory), or other optical disc storage, optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium that can be used to carry or store computer programs and can be read by a computer, which is not limited herein.

[0106] The memory 4003 is used to store the computer program for implementing the embodiments of the present application and is controlled by the processor 4001 to execute. The processor 4001 is used to execute the computer program stored in the memory 4003 to implement the steps shown in the foregoing method embodiments.

[0107] Among them, the electronic device package may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Tablet Computers), PMPs (Portable Multimedia Players), vehicle terminals (such as vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 10 The shown electronic device is only an example and should not bring any limitation to the functions and usage scopes of the embodiments of the present disclosure.

[0108] The embodiments of the present application provide a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps and corresponding contents shown in the foregoing method embodiments can be implemented. Compared with the prior art, it can achieve: It should be noted that the computer-readable medium described above in the present disclosure can be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of a computer-readable storage medium can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium can be any tangible medium that contains or stores a program, which can be used by or in conjunction with an instruction execution system, apparatus, or device. In the present disclosure, a computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium can be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0109] The embodiments of the present application also provide a computer program product, including a computer program, which can implement the steps and corresponding contents of the foregoing method embodiments when executed by a processor. Compared with the prior art, it can achieve: The terms "first", "second", "third", "fourth", "1", "2", etc. (if any) in the description and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present application described here can be implemented in an order other than that shown or described in words.

[0110] It should be understood that although the flowchart of the embodiments of the present application indicates various operation steps by arrows, the execution order of these steps is not limited to the order indicated by the arrows. Unless otherwise clearly stated herein, in some implementation scenarios of the embodiments of the present application, the implementation steps in each flowchart can be executed in other orders according to requirements. In addition, some or all of the steps in each flowchart may include multiple sub-steps or multiple stages based on the actual implementation scenario. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage among these sub-steps or stages can also be executed at different times respectively. In the scenario where the execution times are different, the execution order of these sub-steps or stages can be flexibly configured according to requirements, and the embodiments of the present application do not limit this.

[0111] The above are only optional implementation manners of some implementation scenarios of the present application. It should be noted that for those of ordinary skill in the art, without departing from the technical concept of the solution of the present application, adopting other similar implementation means based on the technical idea of the present application also belongs to the protection scope of the embodiments of the present application.

Claims

1. A welding parameter prediction method, characterized in that: The method comprises: Acquire a characteristic value of a first welding parameter at a first moment; wherein the first welding parameter includes at least one of a material parameter of a welding operation object and a welding process parameter; Inputting the characteristic value of the first welding parameter into a pre-trained prediction model; wherein the prediction model is obtained by training a graph convolutional network and a gated recurrent network using the parameter values ​​of the first welding parameter at multiple sample moments as training samples; each of the training samples also includes a training label, and the training label is the characteristic value of the second welding parameter at the next moment of the corresponding sample moment; wherein the second welding parameter includes: welding time and / or welding power, and the first welding parameter is an independent variable of the second welding parameter; A characteristic value of a second welding parameter at a moment next to the first moment output by the prediction model is obtained.

2. The welding parameter prediction method according to claim 1, characterized in that: The graph convolution network includes: a graph convolution module and an attention module; the gated recurrent network includes: a feature extraction module and a second welding parameter prediction module; Inputting the characteristic value of the first welding parameter into a pre-trained prediction model to obtain the characteristic value of the second welding parameter at the next moment after the first moment output by the prediction model, comprises: Inputting the eigenvalues ​​of the first welding parameter into a graph convolution module and an attention module respectively, obtaining a target feature matrix output by the graph convolution module and a target feature vector output by the attention module; obtaining an updated eigenvalue of the first welding parameter according to the target feature matrix and the target feature vector; The updated characteristic value of the first welding parameter is input into the feature extraction module to obtain a target hidden state output by the feature extraction module, wherein the target hidden state indicates a dependency relationship between the first welding parameters; the target hidden state is input into the second welding parameter prediction module to output a characteristic value of the second welding parameter at the next moment.

3. The welding parameter prediction method according to claim 2, characterized in that: The graph convolution module includes at least one graph convolution unit; there is a predetermined relationship between the first welding parameters; The step of inputting the characteristic value of the first welding parameter into a graph convolution module to obtain a target characteristic matrix output by the graph convolution module includes: Determine a characteristic graph at the first moment according to the characteristic value of the first welding parameter; wherein the nodes of the characteristic graph are the first welding parameters, and the edges between the nodes are the predetermined relationships between the first welding parameters; The feature map is sequentially input into each graph convolution unit to obtain a feature matrix output by each graph convolution unit, and the feature matrix output by the last graph convolution unit is used as the target feature matrix.

4. The welding parameter prediction method according to claim 3, characterized in that: The attention module includes at least one attention unit and a global pooling unit; wherein each attention unit includes an attention channel corresponding to the first welding parameter; The step of inputting the characteristic value of the first welding parameter into an attention module to obtain a target characteristic vector output by the attention module comprises: Determine the input data of each attention unit; wherein, for the first attention unit, the input data is the characteristic value of the first welding parameter; for non-first attention units, the input data is the characteristic value output after convolution calculation of the previous attention unit; For each attention unit, input data corresponding to each first welding parameter is input into the corresponding attention channel, and convolution calculation is performed in the corresponding attention channel to obtain the characteristic value output by each attention channel; wherein the weight value of each attention channel is different; The eigenvalues ​​output by each attention channel in the last attention unit are input into the global pooling unit, and the maximum value of all eigenvalues ​​in each attention channel is extracted as the corresponding eigenvalue after the first welding parameter is updated. The eigenvalues ​​after the first welding parameter is updated are combined into a one-dimensional feature vector as the target feature vector.

5. The welding parameter prediction method according to claim 4, characterized in that: The feature extraction module includes a plurality of gated recurrent units and a fully connected unit; The step of inputting the updated characteristic value of the first welding parameter into the characteristic extraction module to obtain the target hidden state output by the characteristic extraction module comprises: For the first gated recurrent unit, the input data is the updated characteristic value of the first welding parameter, and the output data is the hidden state of the first gated recurrent unit; For non-first gated recurrent unit, the input data is the updated eigenvalue of the first welding parameter and the hidden state output by the previous gated recurrent unit, and the output is the hidden state of the corresponding gated recurrent unit; The hidden state output by the last gated recurrent unit is input into the fully connected unit for full connection processing, and the hidden state output by the fully connected unit is obtained as the target hidden state.

6. The welding parameter prediction method according to claim 5, characterized in that: The second welding parameter prediction module includes a plurality of gated recurrent units and a global pooling unit; The step of inputting the target hidden state into the second welding parameter prediction module and outputting the characteristic value of the second welding parameter at the next moment comprises: Determine the input data of each gated recurrent unit; wherein, for the first gated recurrent unit, the input data is the target hidden state; for non-first gated recurrent units, the input data is the hidden state output by the previous gated recurrent unit; For each gated recurrent unit, a dot product operation is performed on the hidden state of its own output to output a predicted value of the second welding parameter; The predicted values ​​of the second welding parameters output by each gated cycle unit are input into the global pooling unit, and the predicted values ​​of all the second welding parameters are pooled to obtain the characteristic value of the second welding parameter at the next moment.

7. The welding parameter prediction method according to any one of claims 1 to 6, characterized in that: The prediction model is generated by: Acquire parameter values ​​of the first welding parameter and parameter values ​​of the second welding parameter at multiple moments in advance, preprocess the parameter values ​​of the first welding parameter and the parameter values ​​of the second welding parameter at each moment, and obtain characteristic values ​​of the first welding parameter and the second welding parameter at each moment; wherein the preprocessing includes at least one of data cleaning and normalization processing; Determine the characteristic value of the first welding parameter at a plurality of sample moments and the characteristic value of the second welding parameter at a moment next to the corresponding sample moment according to the characteristic value of the first welding parameter and the characteristic value of the second welding parameter at each moment; The first welding parameter characteristic value at the sample moment is used as a training sample, and the characteristic value of the second welding parameter at the next moment of the corresponding sample moment is used as a training label. The initial prediction model is trained until the training stop condition is met to obtain the welding prediction model.

8. A welding parameter prediction device, characterized in that: The device comprises: A first processing module, configured to obtain a characteristic value of a first welding parameter at a first moment; wherein the first welding parameter includes at least one of a material parameter of a welding operation object and a welding process parameter; A second processing module is used to input the characteristic value of the first welding parameter into a pre-trained prediction model; wherein the prediction model is obtained by training a graph convolution network and a gated recurrent network using the parameter values ​​of the first welding parameter at multiple sample moments as training samples; each of the training samples also includes a training label, and the training label is the characteristic value of the second welding parameter at the next moment of the corresponding sample moment; wherein the second welding parameter includes: welding time and / or welding power, and the first welding parameter is an independent variable of the second welding parameter; The third processing module is used to obtain the characteristic value of the second welding parameter at the next moment after the first moment output by the prediction model.

9. An electronic device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

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

Images