Multi-modal sensing industrial process technological index prediction method

Through the multimodal perception method, the feature extraction and fusion of structured and unstructured data is solved, and the problem of low prediction accuracy of process indicators in industrial processes is achieved, achieving higher prediction accuracy and stability of industrial processes.

CN118982275APending Publication Date: 2024-11-19CENT SOUTH UNIV
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411005146.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-25
Publication Date
2024-11-19

AI Technical Summary

Technical Problem

The accuracy of process index prediction in the prior art in industrial processes is low, mainly because the correlation and complementarity between multimodal data is ignored, resulting in insufficient accuracy of process index prediction.

Method used

Using multimodal perception of industrial process technology index prediction method, by obtaining structured and unstructured data, using feature extraction models and encoders to feature extraction and fusion of data, constructing a comprehensive loss function for optimization, and obtaining the final fusion feature And make predictions to improve the prediction accuracy of process indicators.

Benefits of technology

The accuracy of process index prediction is improved, and by combining information within and between modes, the accuracy of fusion characteristics is enhanced, effectively improving the stability and efficiency of industrial processes.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118982275A_ABST
    Figure CN118982275A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of industrial process data, and provides a multi-modal sensing industrial process technical index prediction method. The method comprises the following steps: performing feature extraction on unstructured data to obtain deep features, and encoding structured data to obtain shallow features; fusing the deep features and the shallow features by using a second encoder to obtain fused features; constructing a comprehensive loss function according to all the fusion features, and optimizing the second encoder to obtain an optimized second encoder; using the optimized second encoder to obtain a final fusion feature, and using the prediction model to predict the final fusion feature to obtain a process index prediction value; constructing a prediction loss function based on the process index prediction value, and performing fine adjustment on the prediction model to obtain an optimized prediction model; and performing index prediction by using the optimized second encoder and the optimized prediction model to obtain a process index prediction result. According to the method, the accuracy of process index prediction can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of industrial process data technology, and in particular to a method for predicting industrial process technology indicators based on multimodal perception. Background Art

[0002] With the increasing scale and complexity of modern industrial processes, online prediction of key indicators is essential for real-time identification of production operation conditions, guiding on-site workers to make adjustments, and ensuring the safe, efficient and stable operation of complex industrial processes. However, most industrial sites have harsh environments, complex production mechanisms, and many key process indicators cannot be directly measured. Due to the lack of online detection methods, operators mainly rely on multi-dimensional comprehensive analysis and observation of vision, hearing, measurement data and experience to estimate key process indicators and adjust the production process. However, this "artificial perception, experience decision-making" method has problems such as strong subjectivity and low reliability, which can easily lead to large fluctuations in process indicators and short stable operation cycles. It makes it impossible to efficiently control the operating parameters under dynamic conditions and it is difficult to ensure the stable and efficient operation of industrial processes.

[0003] The widespread application of different types of sensors and monitoring equipment in industrial processes has enriched the dimensions of industrial process data collection and prompted industrial process data to present multimodal characteristics. Multimodal data refers to a collection of data from different sources, with different forms of expression, but can jointly describe the same thing or process. In industrial processes, multimodal data usually includes image data obtained by image monitoring equipment, audio data obtained by audio monitoring equipment, and sensor data such as temperature, pressure, and flow. These multimodal data can reflect the operating status and process characteristics of industrial processes from different angles. Among them, sensor data is mostly structured data, while image and audio data are unstructured data.

[0004] However, data of various modes are often interrelated and mutually influential, jointly determining the changing trend of industrial process indicators. With the development of image recognition, audio processing and artificial intelligence technology, image recognition technology is used to replace human visual perception to obtain unstructured image data and preprocess it, audio processing technology is used to replace human auditory perception to obtain unstructured audio data and preprocess it, machine learning algorithms are used to preprocess structured sensor data, and artificial intelligence technology is integrated to simulate the multimodal perception ability of operators. A prediction model that can comprehensively consider the relationship between multiple modal data is constructed to achieve online and accurate estimation of key process indicators of industrial processes, provide operators with more accurate guidance and suggestions, and avoid problems such as large fluctuations in product quality, short stable operation cycle and low production efficiency caused by subjective arbitrariness. However, existing technologies usually only rely on single-mode data to predict process indicators, ignoring the correlation and complementarity between multimodal data, resulting in low accuracy of process indicator prediction. Summary of the invention

[0005] The present application provides a method for predicting industrial process indicators based on multimodal perception, which can solve the problem of low accuracy in predicting process indicators.

[0006] In a first aspect, an embodiment of the present application provides a method for predicting industrial process indicators based on multimodal perception, and the method for predicting industrial process indicators includes:

[0007] Acquire structured data and unstructured data of multiple target industrial processes; structured data is data represented by a unified structure, and unstructured data is data represented by different structures;

[0008] Using the feature extraction model to extract features from each unstructured data to obtain deep features of each unstructured data, and using the first encoder to encode each structured data to obtain shallow features corresponding to each structured data;

[0009] The deep features and shallow features corresponding to each target industrial process are fused using the second encoder to obtain fused features corresponding to each target industrial process;

[0010] A comprehensive loss function is constructed according to all fused features, and the second encoder is optimized by using the comprehensive loss function to obtain an optimized second encoder; the comprehensive loss function is used to represent the accuracy of all fused features, the difference between each fused feature and multiple other similar fused features, and the difference between each fused feature and multiple other dissimilar fused features;

[0011] The optimized second encoder is used to obtain the final fusion feature corresponding to each target industrial process, and the prediction model is used to predict each final fusion feature to obtain the predicted value of the process index corresponding to each target industrial process;

[0012] A prediction loss function is constructed based on the predicted values ​​of all process indicators, and the prediction model is fine-tuned according to the prediction loss function to obtain an optimized prediction model;

[0013] The feature extraction model, the first encoder, the optimized second encoder and the optimized prediction model are used to predict the indicators of the industrial process to be predicted, and the prediction results of the process indicators of the industrial process to be predicted are obtained.

[0014] Optionally, a feature extraction model is used to extract features from each unstructured data to obtain deep features of each unstructured data, including:

[0015] By formula:

[0016]

[0017] O mean,r (x, y) = mean (a,b)∈W(x,y) O n (a, b)

[0018]

[0019] Get the deep features of n-th unstructured data

[0020] Among them, Flatten() represents flattening processing, mean,n represents the pooling feature of the nth unstructured data, O n (x, y) represents the feature map of the nth unstructured data, W(x, y) represents the pooling window centered at the coordinates (x, y) in the feature map, (a, b) represents the coordinates within the pooling window, and f CNN () represents the activation function, I() represents the feature map, i and j represent the position offset of the convolution kernel, K(i, j) represents the convolution kernel, a = [V / 2], b = [M / 2], V×M represents the size of the convolution kernel, n = 1, 2, ..., N, N represents the number of target industrial processes.

[0021] Optionally, each structured data is encoded using a first encoder to obtain shallow features corresponding to each structured data, including:

[0022] By formula:

[0023]

[0024] Calculate the shallow features of the nth structured data

[0025] Among them, f e () represents the activation function of the first encoder, x n Represents the nth structured data, w e represents the weight matrix, b e represents the bias vector, n=1, 2, ..., N, where N represents the number of target industrial processes.

[0026] Optionally, the deep features and shallow features corresponding to each target industrial process are fused using a second encoder to obtain fused features corresponding to each target industrial process, including:

[0027] By formula:

[0028]

[0029] Get the nth fusion feature

[0030] in, represents the activation function of the second encoder, and Both represent weight matrices, represents the bias vector, represents the nth deep feature, represents the nth shallow feature, n = 1, 2, ..., N, where N represents the number of target industrial processes.

[0031] Optionally, construct a comprehensive loss function based on all fused features, including:

[0032] Construct a reconstruction error loss function based on all fused features;

[0033] For each target industrial process, calculate the similarity between the fusion features of the target industrial process and the fusion features of each other target industrial process, and sort all the similarities from large to small, and take the other target industrial processes corresponding to the first multiple similarities as similar industrial processes of the target industrial process, and take the other target industrial processes corresponding to the last multiple similarities as dissimilar industrial processes of the target industrial process;

[0034] According to all similar industrial processes and all dissimilar industrial processes corresponding to each target industrial process, a geometric feature loss function within the working condition and a geometric feature loss function between working conditions are constructed;

[0035] The reconstruction error loss function, the geometric feature loss function within the working condition and the geometric feature loss function between working conditions are fused to obtain the comprehensive loss function.

[0036] Optionally, the reconstruction error loss function is:

[0037]

[0038] in, represents the value of the reconstruction error loss function, A unified representation of the deep and shallow features of the nth target industrial process, represents the deep features corresponding to the nth target industrial process, represents the shallow features corresponding to the nth target industrial process, express The reconstructed data, represents the decoder activation function, represents the weight matrix, represents the bias term, represents the nth fused feature, n=1, 2, ..., N, where N represents the number of fused features;

[0039] The geometric feature loss function within the working condition is:

[0040]

[0041] in, represents the value of the geometric feature loss function in the working condition, K intra,n represents the number of similar industrial processes to the nth target industrial process, Indicates The fusion characteristics of similar industrial processes, A matrix representing all similar industrial processes of the nth target industrial process;

[0042] The geometric feature loss function between working conditions is:

[0043]

[0044] in, represents the value of the geometric feature loss function between working conditions, K inter,n represents the number of dissimilar industrial processes to the nth target industrial process, Indicates The fusion characteristics of dissimilar industrial processes, A matrix representing all dissimilar industrial processes to the nth target industrial process.

[0045] Optionally, the comprehensive loss function is:

[0046]

[0047] in, represents the value of the comprehensive loss function, and Both represent weights.

[0048] Optionally, optimizing the second encoder by using a comprehensive loss function to obtain an optimized second encoder includes:

[0049] Determine whether the value of the comprehensive loss function is less than the preset loss function value;

[0050] If yes, the second encoder is used as the optimized second encoder;

[0051] Otherwise, adjust the parameters in the feature extraction model and the second encoder, and return to the step of using the second encoder to fuse the deep features and shallow features corresponding to each target industrial process to obtain the fused features corresponding to each target industrial process.

[0052] Optionally, a prediction model is used to predict each final fusion feature to obtain a predicted value of a process indicator corresponding to each target industrial process, including:

[0053] By formula:

[0054]

[0055] Calculate the predicted value of the nth process indicator

[0056] Among them, f p () represents the activation function of the prediction model, represents the weight matrix, represents the bias vector, represents the nth final fusion feature, n=1, 2, ..., N, where N represents the number of target industrial processes.

[0057] Optionally, the prediction loss function is:

[0058]

[0059] Among them, J Fine-tuning Represents the value of the prediction loss function, y n Indicates the true value of the nth process indicator.

[0060] In a second aspect, an embodiment of the present application provides a multi-modal sensing industrial process technology indicator prediction device, comprising:

[0061] An acquisition module, which acquires structured data and unstructured data of multiple target industrial processes; structured data is data represented by a unified structure, and unstructured data is data represented by different structures;

[0062] The feature extraction module uses the feature extraction model to extract features from each unstructured data to obtain deep features of each unstructured data, and uses the first encoder to encode each structured data to obtain shallow features corresponding to each structured data;

[0063] A fusion module, which uses the second encoder to fuse the deep features and shallow features corresponding to each target industrial process to obtain a fusion feature corresponding to each target industrial process;

[0064] An optimization module, which constructs a comprehensive loss function according to all fused features, and optimizes the second encoder using the comprehensive loss function to obtain an optimized second encoder; the comprehensive loss function is used to represent the accuracy of all fused features, the difference between each fused feature and multiple other similar fused features, and the difference between each fused feature and multiple other dissimilar fused features;

[0065] The prediction module uses the optimized second encoder to obtain the final fusion feature corresponding to each target industrial process, and uses the prediction model to predict each final fusion feature to obtain the predicted value of the process indicator corresponding to each target industrial process;

[0066] The fine-tuning module builds a prediction loss function based on the predicted values ​​of all process indicators, and fine-tunes the prediction model according to the prediction loss function to obtain an optimized prediction model;

[0067] The indicator prediction module uses the feature extraction model, the first encoder, the optimized second encoder, and the optimized prediction model to predict the indicators of the industrial process to be predicted, and obtains the process indicator prediction results of the industrial process to be predicted.

[0068] In a third aspect, an embodiment of the present application provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned multimodal perception industrial process technology indicator prediction method when executing the above-mentioned computer program.

[0069] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned multi-modal perception industrial process technology indicator prediction method.

[0070] The above solution of the present application has the following beneficial effects:

[0071] In an embodiment of the present application, structured data and unstructured data of multiple target industrial processes are obtained, and then a feature extraction model is used to extract features from each unstructured data to obtain deep features of each unstructured data, and each structured data is encoded using a first encoder to obtain shallow features corresponding to each structured data, and then the deep features and shallow features corresponding to each target industrial process are fused using a second encoder to obtain fused features corresponding to each target industrial process, and then a comprehensive loss function is constructed based on all fused features, and the second encoder is optimized using the comprehensive loss function to obtain an optimized second encoder, and then the optimized second encoder is used to obtain the final fused features corresponding to each target industrial process, and each final fused feature is predicted using a prediction model to obtain a predicted value of a process indicator corresponding to each target industrial process, and then a prediction loss function is constructed based on all predicted values ​​of process indicators, and the prediction model is fine-tuned according to the prediction loss function to obtain an optimized prediction model, and finally the feature extraction model, the first encoder, the optimized second encoder, and the optimized prediction model are used to perform indicator prediction on the industrial process to be predicted to obtain a prediction result of the process indicator of the industrial process to be predicted. Among them, using different feature extraction methods for structured data and unstructured data respectively can improve the accuracy of feature extraction, improve the accuracy of deep features and shallow features, and then improve the accuracy of fused features. Constructing a comprehensive loss function can combine the information within and between modalities to describe the accuracy and difference of fused features. Using the comprehensive loss function to obtain the final fused features can further improve the accuracy of fused features. Using precise fused features for indicator prediction can effectively improve the accuracy of process indicator prediction results.

[0072] Other beneficial effects of the present application will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0074] Figure 1 A flowchart of a method for predicting industrial process indicators using multimodal sensing provided in one embodiment of the present application;

[0075] Figure 2 A curve chart of the predicted value of concentrate grade provided in one embodiment of the present application;

[0076] Figure 3A schematic diagram of the structure of a multi-modal sensing industrial process technology indicator prediction device provided in one embodiment of the present application;

[0077] Figure 4 A schematic diagram of the structure of a terminal device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0078] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0079] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0080] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0081] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0082] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0083] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0084] In order to solve the problem of low accuracy of existing process indicator prediction, the embodiment of the present application provides a multi-modal perception industrial process process indicator prediction method, which obtains structured data and unstructured data of multiple target industrial processes, and then uses a feature extraction model to extract features from each unstructured data to obtain deep features of each unstructured data, and uses a first encoder to encode each structured data to obtain shallow features corresponding to each structured data, and then uses a second encoder to fuse the deep features and shallow features corresponding to each target industrial process to obtain fused features corresponding to each target industrial process, and then constructs a deep feature extraction model based on all fused features. The comprehensive loss function is used to optimize the second encoder by using the comprehensive loss function to obtain the optimized second encoder, and then the optimized second encoder is used to obtain the final fusion feature corresponding to each target industrial process, and the prediction model is used to predict each final fusion feature to obtain the process index prediction value corresponding to each target industrial process, and then the prediction loss function is constructed based on all the process index prediction values, and the prediction model is fine-tuned according to the prediction loss function to obtain the optimized prediction model, and finally the feature extraction model, the first encoder, the optimized second encoder, and the optimized prediction model are used to predict the index of the industrial process to be predicted, and the process index prediction result of the industrial process to be predicted is obtained. Among them, different feature extraction methods are used for structured data and unstructured data respectively, which can improve the accuracy of feature extraction, improve the accuracy of deep features and shallow features, and then improve the accuracy of fusion features. The comprehensive loss function is constructed, which can combine the information within and between modalities to express the accuracy and difference of fusion features, and the comprehensive loss function is used to obtain the final fusion feature, which can further improve the accuracy of fusion features, and the use of accurate fusion features for index prediction can effectively improve the accuracy of process index prediction results.

[0085] Next, an exemplary description is given of the multimodal sensing industrial process technology indicator prediction method provided in this application.

[0086] like Figure 1 As shown, the multimodal sensing industrial process technology index prediction method provided by the present application includes the following steps:

[0087] Step 11, obtaining structured data and unstructured data of multiple target industrial processes.

[0088] The above-mentioned structured data is data represented by a unified structure, usually in the form of two-dimensional data, strictly complying with data format and length specifications (such as numbers, etc.), and unstructured data is data represented by different structures (such as pictures, audio, etc.). The above-mentioned target industrial process is an industrial process with true values ​​of process indicators. The process indicators can be indicators such as product purity, water content, impurity content, density, etc. All target industrial processes are the same industrial process. The difference is that different target industrial processes correspond to different times. For example, if the first target industrial process is an industrial process at 8 o'clock, the second target industrial process can be an industrial process at 9 o'clock. And the structured data and unstructured data of all target industrial processes are data at historical moments, which are mainly used to train the relevant models for process indicator prediction mentioned below.

[0089] Exemplarily, the target industrial process is a potassium salt flotation industrial production process, and the acquired data are froth image data, concentrate audio data, and sensor measurement data, wherein the froth image data and concentrate audio data are unstructured data, and the sensor measurement data is structured data. In subsequent calculations, only one unstructured data and one structured data are required.

[0090] In some embodiments of the present application, structured data and unstructured data of a target industrial process may be acquired through devices such as sensors, cameras, and recorders.

[0091] Step 12: Use the feature extraction model to extract features from each unstructured data to obtain deep features of each unstructured data, and use the first encoder to encode each structured data to obtain shallow features corresponding to each structured data.

[0092] In some embodiments of the present application, the steps of extracting features from each unstructured data using a feature extraction model to obtain deep features of each unstructured data, and encoding each structured data using a first encoder to obtain shallow features corresponding to each structured data are specifically as follows:

[0093] In the first step, the feature extraction model is used to extract features from each unstructured data to obtain the deep features of each unstructured data.

[0094] Specifically, the feature extraction model is expressed as:

[0095]

[0096] O mean,r (x, y) = mean (a,b)∈W(x,y) O n (a, b)

[0097]

[0098] Get deep features of n-th unstructured data

[0099] Among them, flatten() represents flattening processing, mean,n represents the pooling feature of the nth unstructured data, O n (x, y) represents the feature map of the nth unstructured data, W(x, y) represents the pooling window centered at the coordinates (x, y) in the feature map, (a, b) represents the coordinates within the pooling window, and f CNN () represents the activation function, I() represents the feature map, i and j represent the position offset of the convolution kernel, K(i, j) represents the convolution kernel, a = [V / 2], b = [M / 2], V×M represents the size of the convolution kernel, n = 1, 2, ..., N, N represents the number of target industrial processes.

[0100] It can be understood that the above formula for obtaining deep features is the expression of the feature extraction model.

[0101] In the second step, each structured data is encoded using the first encoder to obtain the shallow features corresponding to each structured data.

[0102] Specifically, the first encoder uses the formula:

[0103]

[0104] Calculate the shallow features of the nth structured data

[0105] Among them, f e () represents the activation function of the first encoder, X n Represents the nth structured data, w e represents the weight matrix, b e represents the bias vector, n=1, 2, ..., N, where N represents the number of target industrial processes.

[0106] It can be understood that the above formula for calculating shallow features is the expression of the first encoder.

[0107] It should be noted that the accuracy of the shallow features can be described by constructing a loss function of the first encoder, and the first encoder is trained using the loss function, that is, when the value of the loss function is greater than a preset loss function value (such as 1.2), the parameters of the first encoder are adjusted, the shallow features are reacquired and the value of the loss function is calculated until the value of the loss function is less than or equal to the preset loss function value. The expression of the loss function is:

[0108]

[0109] Among them, J AE represents the value of the loss function, Represents the reconstructed data of the nth structured data, f d () represents the decoder activation function, w d represents the weight matrix, b d Represents the bias term.

[0110] Exemplarily, the above steps can be run using computer software such as matlab and python to obtain deep features of unstructured data and shallow features of structured data.

[0111] It is worth mentioning that according to the different data forms of unstructured data and structured data, different feature extraction methods can improve the accuracy of feature extraction and improve the accuracy of deep and shallow features in describing information.

[0112] Step 13: Use the second encoder to fuse the deep features and shallow features corresponding to each target industrial process to obtain the fused features corresponding to each target industrial process.

[0113] Specifically, the second encoder uses the formula:

[0114]

[0115] Get the nth fusion feature

[0116] in, represents the activation function of the second encoder, and Both represent weight matrices, represents the bias vector, represents the nth deep feature, represents the nth shallow feature, n=1, 2, ..., N, N represents the number of fused features.

[0117] Exemplarily, the second encoder can be run using computer software such as matlab, python, etc. to obtain fusion features.

[0118] It is worth mentioning that by fusing deep features and shallow features, the fused features can include information from both structured data and unstructured data.

[0119] Step 14, constructing a comprehensive loss function based on all fused features, and optimizing the second encoder using the comprehensive loss function to obtain an optimized second encoder.

[0120] The above comprehensive loss function is used to represent the accuracy of all fused features, the difference between each fused feature and multiple other similar fused features, and the difference between each fused feature and multiple other dissimilar fused features.

[0121] In some embodiments of the present application, the steps of constructing a comprehensive loss function according to all fusion features, and optimizing the second encoder using the comprehensive loss function to obtain the optimized second encoder are specifically as follows:

[0122] In the first step, a reconstruction error loss function is constructed based on all fused features.

[0123] Specifically, the reconstruction error loss function is:

[0124]

[0125] in, represents the value of the reconstruction error loss function, A unified representation of the deep and shallow features of the nth target industrial process, represents the deep features corresponding to the nth target industrial process, represents the shallow features corresponding to the nth target industrial process, express The reconstruction features of represents the decoder activation function, represents the weight matrix, represents the bias term, represents the nth fused feature, n=1, 2, ..., N, where N represents the number of fused features.

[0126] In the second step, for each target industrial process, the similarity between the fusion features of the target industrial process and the fusion features of each other target industrial process is calculated, and all similarities are sorted from large to small. The other target industrial processes corresponding to the first multiple similarities are regarded as similar industrial processes of the target industrial process, and the other target industrial processes corresponding to the last multiple similarities are regarded as dissimilar industrial processes of the target industrial process.

[0127] Exemplarily, the similarity between the fusion features of the target industrial process and the fusion features of other target industrial processes may be calculated using a cosine similarity calculation formula.

[0128] In the third step, the intra-condition geometric feature loss function and the inter-condition geometric feature loss function are constructed according to all similar industrial processes and all dissimilar industrial processes corresponding to each target industrial process.

[0129] Specifically, the geometric feature loss function within the working condition is:

[0130]

[0131] in, represents the value of the geometric feature loss function in the working condition, K intra,n represents the number of similar industrial processes to the nth target industrial process, Indicates The fusion characteristics of similar industrial processes, A matrix representing all similar industrial processes to the nth target industrial process.

[0132] Specifically, the geometric feature loss function between working conditions is:

[0133]

[0134] in, represents the value of the geometric feature loss function between working conditions, K inter,n represents the number of dissimilar industrial processes to the nth target industrial process, Indicates The fusion characteristics of dissimilar industrial processes, A matrix representing all dissimilar industrial processes to the nth target industrial process.

[0135] The fourth step is to fuse the reconstruction error loss function, the intra-condition geometric feature loss function, and the inter-condition geometric feature loss function to obtain a comprehensive loss function.

[0136] Specifically, the comprehensive loss function is:

[0137]

[0138] in, represents the value of the comprehensive loss function, and Both represent weights.

[0139] The fifth step is to determine whether the value of the comprehensive loss function is less than the preset loss function value.

[0140] If so, the second encoder is used as the optimized second encoder.

[0141] Otherwise, adjust the parameters in the feature extraction model and the second encoder, and return to the step of using the second encoder to fuse the deep features and shallow features corresponding to each target industrial process to obtain the fused features corresponding to each target industrial process.

[0142] Exemplarily, if the value of the comprehensive loss function is 5, which is greater than the preset loss function value of 1.2, the parameters in the second encoder are adjusted, and the step of using the second encoder to fuse the deep features and shallow features corresponding to each target industrial process to obtain the fused features corresponding to each target industrial process is returned, and the value of the comprehensive loss function is recalculated. At this time, the value of the comprehensive loss function is 0.9, which is less than the preset loss function value. The second encoder at this time is used as the optimized second encoder.

[0143] It is worth mentioning that by constructing a comprehensive loss function, it is possible to express three aspects: the accuracy of the fused features, the difference between the fused features and other features with the same modality, and the difference between the fused features and other features with different modalities, so that the performance of the second encoder optimized by the comprehensive loss function meets expectations.

[0144] Step 15, using the optimized second encoder to obtain the final fusion feature corresponding to each target industrial process, and using the prediction model to predict each final fusion feature to obtain the process indicator prediction value corresponding to each target industrial process.

[0145] In some embodiments of the present application, the deep features and shallow features of each target industrial process are fused using the optimized second encoder to obtain the final fused features corresponding to each target industrial process. The above prediction model is used to predict each final fused feature to obtain the process indicator prediction value corresponding to each target industrial process:

[0146] The prediction model is based on the formula:

[0147]

[0148] Calculate the predicted value of the nth process indicator

[0149] Among them, f p () represents the activation function of the prediction model, represents the weight matrix, represents the bias vector, represents the nth final fusion feature, n=1, 2, ..., N, where N represents the number of target industrial processes.

[0150] It can be understood that the above formula for calculating the predicted value of the process indicator is the formula of the prediction model.

[0151] Exemplarily, the process indicator may be water content, impurity content, product purity, etc., and the process indicator prediction value is the predicted value of the set process indicator to be predicted, such as 20% water content, 0.1% impurity content or 97% product purity, etc.

[0152] It is worth mentioning that the final fusion feature obtained by utilizing the feature extraction model and the second encoder to achieve performance prediction has high accuracy, thereby improving the accuracy of the predicted value of the process indicator.

[0153] Step 16, constructing a prediction loss function based on the predicted values ​​of all process indicators, and fine-tuning the prediction model according to the prediction loss function to obtain an optimized prediction model.

[0154] Specifically, the above prediction loss function is:

[0155]

[0156] Among them, J Fine-tuning represents the value of the prediction loss function, Represents the predicted value of the nth process indicator, y n Indicates the true value of the nth process indicator.

[0157] The above-mentioned step of fine-tuning the prediction model according to the prediction loss function to obtain the optimized prediction model is specifically as follows: determine whether the value of the prediction loss function is less than the preset value of the prediction loss function, if so, use the prediction model as the optimized prediction model, otherwise, adjust the parameters of the prediction model, and return to the step of using the prediction model to predict each final fusion feature to obtain the predicted value of the process indicator corresponding to each target industrial process.

[0158] Exemplarily, if the value of the prediction loss function is 3.1, which is greater than the preset value of the prediction loss function of 2.1, the parameters of the prediction model are adjusted, and the step of using the prediction model to predict each final fusion feature to obtain the predicted value of the process indicator corresponding to each target industrial process is returned, and the value of the prediction loss function is recalculated. At this time, the value of the prediction loss function is 1.8, which is less than 2.1. The prediction model at this time is used as the optimized prediction model.

[0159] Step 17, using the feature extraction model, the first encoder, the optimized second encoder, and the optimized prediction model to perform index prediction on the industrial process to be predicted, and obtain a process index prediction result of the industrial process to be predicted.

[0160] The above-mentioned industrial process to be predicted is an industrial process that needs to be predicted for process indicators. The industrial process to be predicted and the target industrial process are the same industrial process, but the industrial process to be predicted does not have a real process indicator prediction value and needs to be predicted.

[0161] Specifically, the structured data and unstructured data of the industrial process to be predicted at the current moment are obtained, and the optimized feature extraction model is used to perform feature extraction on the unstructured data to obtain deep features. The first encoder is used to extract features on the structured data to obtain shallow features, and then the optimized second encoder is used to fuse them to obtain the final fused features of the industrial process to be predicted. Finally, the optimized prediction model is used to perform indicator prediction on the final fused features to obtain the process indicator prediction results of the industrial process to be predicted at the current moment.

[0162] Exemplarily, the process indicator may be water content, impurity content, product purity, etc., and the above process indicator prediction result may be a process indicator prediction value obtained according to the final fusion characteristics of the industrial process to be predicted.

[0163] It is worth mentioning that using different feature extraction methods for structured data and unstructured data respectively can improve the accuracy of feature extraction, improve the accuracy of deep features and shallow features, and then improve the accuracy of fused features. Constructing a comprehensive loss function can combine the information within and between modalities to describe the accuracy and difference of fused features. Using the comprehensive loss function to obtain the final fused features can further improve the accuracy of fused features. Using accurate fused features for indicator prediction can effectively improve the accuracy of process indicator prediction results.

[0164] The following is an illustrative example of the multimodal sensing industrial process technology indicator prediction method of the present application.

[0165] The potassium salt flotation process is an important part of potassium salt mineral processing and plays a vital role in the effectiveness of the production process. The quality of potassium salt flotation performance is directly related to the quality and output of the final potassium salt product and even energy consumption and resource utilization. Realizing online inspection of key process indicators in the potassium salt flotation process is of great significance for guiding on-site workers to make operational adjustments and is the basis for achieving safe and stable operation of the potassium salt flotation process. However, due to the harsh environment of the potassium salt flotation site, key process indicators are difficult to detect online, so the prediction of key process indicators has become an alternative way to guide on-site workers.

[0166] In this case, data for nearly two months were collected from the potash flotation site, and industrial verification experiments were conducted. Under the premise of ensuring data quality, a total of 224 sets of data samples were collected after excluding abnormal production status data caused by equipment maintenance and shutdown operations. Each set of data samples includes unstructured data, structured data and the corresponding true values ​​of process indicators. The data samples are divided into two parts, training set and test set, according to the chronological order. The first 160 sets of samples are training sets, and the remaining 64 sets of samples are test sets, which are used to evaluate the effectiveness of the method of this application in the prediction task of potash flotation process indicators.

[0167] In order to verify the effectiveness of the method of the present application in predicting process indicators in the process of potassium salt flotation, a comparative analysis was performed using the adaptive attention-driven manifold regularized autoencoder (AAMR-AE) method without considering feature fusion, the prediction method of the integrated convolutional neural network and autoencoder (CNN+AE), autoencoder (AE, AutoEncoder), graph dynamic autoencoder (GDAE, Graph Dynamic Autoencoder), low-rank reconstruction autoencoder (LRAE, Low-rankReconstruction Autoencoder) and multi-manifold autoencoder (MMAE, Multimanifold Autoencoder) to verify the advantages of the method of the present application. The process indicators obtained after the process indicator prediction experiment using the above data sets and all methods are as follows: The prediction results of concentrate grade are as follows Figure 2 As shown in the figure, the horizontal axis represents the number of samples, the horizontal axis represents the concentrate grade, the unit is percentage %, the solid line with dots represents the true value of the concentrate grade, the solid line with the right triangle represents the concentrate grade prediction value curve of AAMR-AE, the solid line with the left triangle represents the concentrate grade prediction value curve of CNN+AE, the solid line with diamonds represents the concentrate grade prediction value curve of LRAE, the solid line with the upper triangle represents the concentrate grade prediction value curve of MMAE, the solid line with a cross represents the concentrate grade prediction value curve of FDAE, and the solid line with a star represents the concentrate grade prediction value curve of the method (BiM-AE) of the present application.

[0168] The results of the root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE) and relative coefficient (R2, R-Square) of the prediction results are shown in Table 1.

[0169] method RMSE MAE MAPE(%) R2 AAMR-AE 0.69±0.01 0.54±0.01 1.82±0.04 0.71±0.01 CNN+AE 0.83±0.04 0.67±0.03 2.26±0.10 0.57±0.04 LRAE 0.75±0.03 0.60±0.02 2.04±0.08 0.65±0.03 MMAE 0.81±0.02 0.69±0.02 2.34±0.06 0.59±0.02 GDAE 0.77±0.03 0.55±0.03 1.82±0.11 0.63±0.03 This application 0.52±0.01 0.38±0.01 1.29±0.04 0.83±0.01

[0170] Table 1

[0171] It can be seen from this that the prediction accuracy and stability of the prediction results obtained by the method of the present application are higher than those of other methods.

[0172] The following is an exemplary description of the multimodal sensing industrial process technology indicator prediction device provided in this application.

[0173] like Figure 3 As shown, the embodiment of the present application provides a multi-modal sensing industrial process technology index prediction device, and the industrial process technology index prediction device 300 includes:

[0174] An acquisition module 301 acquires structured data and unstructured data of a plurality of target industrial processes; structured data is data represented by a unified structure, and unstructured data is data represented by different structures;

[0175] The feature extraction module 302 uses the feature extraction model to extract features from each unstructured data to obtain deep features of each unstructured data, and uses the first encoder to encode each structured data to obtain shallow features corresponding to each structured data;

[0176] A fusion module 303 uses a second encoder to fuse the deep features and shallow features corresponding to each target industrial process to obtain a fusion feature corresponding to each target industrial process;

[0177] The optimization module 304 constructs a comprehensive loss function according to all fused features, and optimizes the second encoder using the comprehensive loss function to obtain an optimized second encoder; the comprehensive loss function is used to represent the accuracy of all fused features, the difference between each fused feature and multiple other similar fused features, and the difference between each fused feature and multiple other dissimilar fused features;

[0178] The prediction module 305 uses the optimized second encoder to obtain the final fusion feature corresponding to each target industrial process, and uses the prediction model to predict each final fusion feature to obtain the process indicator prediction value corresponding to each target industrial process;

[0179] A fine-tuning module 306 constructs a prediction loss function based on the predicted values ​​of all process indicators, and fine-tunes the prediction model according to the prediction loss function to obtain an optimized prediction model;

[0180] The index prediction module 307 uses the feature extraction model, the first encoder, the optimized second encoder, and the optimized prediction model to perform index prediction on the industrial process to be predicted, and obtains the process index prediction result of the industrial process to be predicted.

[0181] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0182] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0183] like Figure 4 As shown, an embodiment of the present application provides a terminal device. The terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 4 Only one processor is shown in the figure), a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100, wherein the processor D100 implements the steps of any of the above-mentioned method embodiments when executing the computer program D102.

[0184] Specifically, when the processor D100 executes the computer program D102, it obtains structured data and unstructured data of multiple target industrial processes, and then uses a feature extraction model to extract features from each unstructured data to obtain deep features of each unstructured data, and uses a first encoder to encode each structured data to obtain shallow features corresponding to each structured data, and then uses a second encoder to fuse the deep features and shallow features corresponding to each of the target industrial processes to obtain fused features corresponding to each of the target industrial processes, and then constructs a comprehensive loss function based on all fused features, and uses the comprehensive loss function to optimize the second encoder to obtain an optimized second encoder, and then uses the optimized second encoder to obtain the final fused features corresponding to each target industrial process, and uses the prediction model to predict each final fused feature to obtain a predicted value of a process indicator corresponding to each of the target industrial processes, and then constructs a prediction loss function based on all the predicted values ​​of the process indicators, and fine-tunes the prediction model according to the prediction loss function to obtain an optimized prediction model, and finally uses the feature extraction model, the first encoder, the optimized second encoder, and the optimized prediction model to perform indicator prediction on the industrial process to be predicted to obtain a prediction result of the process indicator of the industrial process to be predicted. Among them, using different feature extraction methods for structured data and unstructured data respectively can improve the accuracy of feature extraction, improve the accuracy of deep features and shallow features, and then improve the accuracy of fused features. Constructing a comprehensive loss function can combine the information within and between modalities to describe the accuracy and difference of fused features. Using the comprehensive loss function to obtain the final fused features can further improve the accuracy of fused features. Using precise fused features for indicator prediction can effectively improve the accuracy of process indicator prediction results.

[0185] The processor D100 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0186] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may also be an external storage device of the terminal device D10, such as a plug-in hard disk, a smart memory card (SMC, SmartMedia Card), a secure digital (SD, Secure Digital) card, a flash card (Flash Card), etc. equipped on the terminal device D10. Further, the memory D101 may also include both an internal storage unit of the terminal device D10 and an external storage device. The memory D101 is used to store an operating system, an application program, a boot loader (BootLoader), data and other programs, such as the program code of the computer program, etc. The memory D101 may also be used to temporarily store data that has been output or is to be output.

[0187] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the above-mentioned method embodiments can be implemented.

[0188] An embodiment of the present application provides a computer program product. When the computer program product runs on a terminal device, the terminal device can implement the steps in the above-mentioned method embodiments when executing the computer program product.

[0189] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium that can carry the computer program code to the multimodal perception industrial process process indicator prediction method device / terminal device. For example, a USB flash drive, a mobile hard disk, a magnetic disk or an optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0190] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0191] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0192] The above is a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles described in the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A multi-modal perception method for predicting industrial process indicators. It is characterized in that include: Acquire structured data and unstructured data of multiple target industrial processes; the structured data is data represented by a unified structure, and the unstructured data is data represented by different structures; Using a feature extraction model to extract features from each of the unstructured data to obtain deep features of each of the unstructured data, and using a first encoder to encode each of the structured data to obtain shallow features corresponding to each of the structured data; Using a second encoder to fuse the deep features and shallow features corresponding to each of the target industrial processes to obtain fused features corresponding to each of the target industrial processes; A comprehensive loss function is constructed according to all fused features, and the second encoder is optimized by using the comprehensive loss function to obtain an optimized second encoder; the comprehensive loss function is used to represent the accuracy of all fused features, the difference between each fused feature and multiple other similar fused features, and the difference between each fused feature and multiple other dissimilar fused features; Using the optimized second encoder to obtain the final fusion feature corresponding to each of the target industrial processes, and using the prediction model to predict each of the final fusion features to obtain a predicted value of the process indicator corresponding to each of the target industrial processes; Constructing a prediction loss function based on the predicted values ​​of all process indicators, and fine-tuning the prediction model according to the prediction loss function to obtain an optimized prediction model; The feature extraction model, the first encoder, the optimized second encoder, and the optimized prediction model are used to perform index prediction on the industrial process to be predicted, and a process index prediction result of the industrial process to be predicted is obtained.

2. The method for predicting industrial process indicators according to claim 1, It is characterized in that The feature extraction model is used to extract features from each of the unstructured data to obtain deep features of each of the unstructured data, including: By formula: O mean,r (x,y)=mean (a,b)∈W(x,y) O n (a,b) Get the deep features of n-th unstructured data Among them, flatten() represents flattening processing, mean,n represents the pooling feature of the nth unstructured data, O n (x, y) represents the feature map of the nth unstructured data, W(x, y) represents the pooling window centered on the coordinates (x, y) in the feature map, (a, b) represents the coordinates within the pooling window, and f CNN ( ) represents an activation function, I ( ) represents the feature map, i and j represent the position offset of the convolution kernel, K(i,j) represents the convolution kernel, a=[V / 2], b=[M / 2], V×M represents the size of the convolution kernel, n=1,2,...,N, and N represents the number of target industrial processes.

3. The industrial process technology indicator prediction method according to claim 1, It is characterized in that The using the first encoder to encode each of the structured data to obtain the shallow features corresponding to each of the structured data includes: By formula: Calculate the shallow features of the nth structured data Among them, f e ( ) represents the activation function of the first encoder, x n represents the nth structured data, w e represents the weight matrix, b e represents the bias vector, n=1,2,...,N, and N represents the number of target industrial processes.

4. The industrial process technology indicator prediction method according to claim 1, It is characterized in that The using the second encoder to fuse the deep features and shallow features corresponding to each of the target industrial processes to obtain the fused features corresponding to each of the target industrial processes includes: By formula: Get the nth fusion feature in, represents the activation function of the second encoder, and Both represent weight matrices, represents the bias vector, represents the nth deep feature, represents the nth shallow feature, n = 1, 2, ..., N, and N represents the number of target industrial processes.

5. The industrial process technology indicator prediction method according to claim 1, It is characterized in that The comprehensive loss function is constructed based on all fusion features, including: Construct a reconstruction error loss function based on all fused features; For each of the target industrial processes, respectively, calculate the similarity between the fusion feature of the target industrial process and the fusion feature of each other target industrial process, and sort all the similarities from large to small, and take the other target industrial processes corresponding to the first multiple similarities as similar industrial processes of the target industrial process, and take the other target industrial processes corresponding to the last multiple similarities as dissimilar industrial processes of the target industrial process; Constructing an intra-operating condition geometric feature loss function and an inter-operating condition geometric feature loss function according to all similar industrial processes and all dissimilar industrial processes corresponding to each target industrial process; The reconstruction error loss function, the intra-working condition geometric feature loss function and the inter-working condition geometric feature loss function are fused to obtain a comprehensive loss function.

6. The method for predicting industrial process indicators according to claim 5, It is characterized in that The reconstruction error loss function is: in, represents the value of the reconstruction error loss function, A unified representation of the deep and shallow features of the nth target industrial process, represents the deep features corresponding to the nth target industrial process, represents the shallow features corresponding to the nth target industrial process, express The reconstructed data, represents the decoder activation function, represents the weight matrix, represents the bias term, represents the nth fused feature, n=1,2,...,N, N represents the number of fused features; The geometric feature loss function in the working condition is: in, represents the value of the geometric feature loss function in the working condition, K intraq,n represents the number of similar industrial processes to the nth target industrial process, Indicates The fusion characteristics of similar industrial processes, A matrix representing all similar industrial processes of the nth target industrial process; The geometric feature loss function between the working conditions is: in, represents the value of the geometric feature loss function between working conditions, K inter,n represents the number of dissimilar industrial processes to the nth target industrial process, Indicates The fusion characteristics of dissimilar industrial processes, A matrix representing all dissimilar industrial processes of the nth target industrial process.

7. The method for predicting industrial process indicators according to claim 6, It is characterized in that The comprehensive loss function is: in, represents the value of the comprehensive loss function, and Both represent weights.

8. The method for predicting industrial process indicators according to claim 1, It is characterized in that The step of optimizing the second encoder by using the comprehensive loss function to obtain an optimized second encoder includes: Determine whether the value of the comprehensive loss function is less than a preset loss function value; If yes, use the second encoder as the optimized second encoder; Otherwise, adjust the parameters in the feature extraction model and the second encoder, and return to the step of using the second encoder to fuse the deep features and shallow features corresponding to each of the target industrial processes to obtain the fused features corresponding to each of the target industrial processes.

9. The method for predicting industrial process indicators according to claim 1, It is characterized in that The method of using the prediction model to predict each of the final fusion features to obtain a predicted value of a process indicator corresponding to each of the target industrial processes includes: By formula: Calculate the predicted value of the nth process indicator Among them, f p ( ) represents the activation function of the prediction model, represents the weight matrix, represents the bias vector, represents the nth final fusion feature, n = 1, 2, ..., N, and N represents the number of target industrial processes.

10. The method for predicting industrial process indicators according to claim 9, It is characterized in that The prediction loss function is: Among them, J Fine-tuning Represents the value of the prediction loss function, y n Indicates the true value of the nth process indicator.

Citation Information

Cited By

  • Microgrid scheduling optimization method based on multiple modes and related equipment

    CN119742870A

  • Data acquisition method for standard structured data set architecture of intelligent SOP

    CN119886575A