Knowledge graph-based intelligent matching method and system for industrial production parameters
By using a knowledge graph-based approach and employing convolutional neural networks and multi-spatial projection techniques, suitable parameter combinations are generated, solving the problem of low parameter configuration accuracy in industrial production and achieving efficient and accurate parameter matching and dynamic adjustment.
Patent Information
- Application Number
- PCT/CN2025/114627
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-14
- Filing Date
- 2025-08-14
- Publication Date
- 2026-11-19
AI Technical Summary
In existing technologies, the configuration accuracy of industrial production parameters is low, making it difficult to improve production efficiency and product quality.
A knowledge graph-based approach is adopted to acquire factory production process data, preprocess it to generate standardized process tensors and constraint vectors, use convolutional neural networks to extract equipment feature vectors and project them into hyperbolic and spherical spaces, and combine fully connected layers and target optimization modules to generate suitable parameter combinations.
It improves the accuracy and efficiency of parameter matching, can dynamically adapt to changes in the production environment, ensures that parameter combinations comply with real-time process rules and safety constraints, and enhances the efficiency and safety of the production process.
Smart Images

Figure CN2025114627_19112026_PF_FP_ABST
Abstract
Description
A Knowledge Graph-Based Intelligent Matching Method and System for Industrial Production Parameters Technical Field
[0001] This application relates to the field of industrial control technology, and more specifically, to a method and system for intelligent matching of industrial production parameters based on knowledge graphs. Background Technology
[0002] With the advancement of the Industrial Internet and intelligent manufacturing, the demand for intelligence, automation, and efficiency in industrial production processes is becoming increasingly urgent. Industrial production involves a large number of complex parameters, such as equipment parameters, process parameters, and environmental parameters. The matching and optimization of these parameters are crucial for improving production efficiency, reducing costs, and ensuring product quality. Knowledge graphs, as a structured knowledge representation method, can effectively integrate and manage this parameter information, providing a foundation for intelligent matching.
[0003] In practical applications, extracting and integrating valuable knowledge from industrial production data is a complex process. It requires handling data of different formats and structures, and resolving conflicts and redundancies between pieces of knowledge. Especially when dealing with multi-dimensional factors such as parameter ranges, correlations, and constraints, the accuracy of parameter configuration can easily be low. Summary of the Invention
[0004] This application provides a knowledge graph-based intelligent matching method and system for industrial production parameters, which can at least partially solve the problem of low parameter configuration accuracy in industrial production.
[0005] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.
[0006] According to one aspect of this application, a knowledge graph-based intelligent matching method for industrial production parameters is provided, comprising: acquiring process data during factory production; preprocessing the process data to generate standardized process tensors and constraint vectors; extracting equipment feature vectors from the process tensors using a convolutional neural network, projecting the equipment feature vectors into hyperbolic space and spherical space to generate a first projection and a second projection respectively, and then concatenating them to generate a concatenated vector; decomposing a factor matrix from the concatenated vector, generating a linear combination vector of the constraint vectors based on preset constraint weights, and generating a first tensor kernel based on the linear combination vector and the factor matrix; mapping the constraint vectors to the tensor space through a preset fully connected layer and fusing them with the first tensor kernel to generate a second tensor kernel; inputting the second tensor kernel into a preset target optimization module to solve for a suitable parameter combination, and sending the parameter combination to a control terminal.
[0007] In this application, based on the aforementioned scheme, the process data includes production data, equipment data, environmental data, and configuration files that change with the process stages during production.
[0008] In this application, based on the aforementioned scheme, the preprocessing of the process data to generate standardized process tensors and constraint vectors includes: aligning the process data based on the timestamps of the process data to obtain time-uniform process data; generating process tensors from the process data according to three dimensions: time, parameter type, and device node; extracting constraint rules from the configuration file, and constructing time-related constraint vectors based on the constraint rules.
[0009] In this application, based on the aforementioned scheme, the step of extracting device feature vectors from the process tensor using a convolutional neural network, projecting the device feature vectors into hyperbolic space and spherical space to generate a first projection and a second projection respectively, and then concatenating them to generate a concatenated vector, includes: extracting local features from the process tensor using a convolutional neural network as the device feature vectors; mapping the device feature vectors onto hyperbolic space to generate a first projection; mapping the device feature vectors onto spherical space to generate a second projection; and concatenating the first projection and the second projection to generate a concatenated vector.
[0010] In this application, based on the aforementioned scheme, the step of decomposing a factor matrix from the concatenated vector, generating a linear combination vector of the constraint vectors based on preset constraint weights, and generating a first tensor kernel based on the linear combination vector and the factor matrix includes: decomposing a factor matrix from the concatenated vector, wherein the factor matrix represents the potential correlation between data of different dimensions; performing linear processing on the constraint vectors based on preset constraint weights to generate a linear combination vector; and performing operations on the linear combination vector and the factor matrix to generate a first tensor kernel.
[0011] In this application, based on the aforementioned scheme, the step of mapping the constraint vector to a tensor space through a preset fully connected layer and fusing it with the first tensor kernel to generate a second tensor kernel includes: mapping the constraint vector to a tensor space through a preset fully connected layer to generate an adjustment tensor, wherein the adjustment tensor has the same dimension as the first tensor kernel; and fusing the adjustment tensor with the first tensor kernel to generate a second tensor kernel.
[0012] In this application, based on the aforementioned scheme, the step of inputting the second tensor kernel into a preset target optimization module to obtain a suitable parameter combination includes: inputting the second tensor kernel into a preset target optimization module, obtaining a suitable parameter combination through gradient descent; and sending the parameter combination to a control terminal.
[0013] According to one aspect of this application, a knowledge graph-based intelligent matching system for industrial production parameters is provided, comprising:
[0014] The acquisition unit is used to acquire process data during the factory production process.
[0015] The preprocessing unit is used to preprocess the process data to generate standardized process tensors and constraint vectors;
[0016] The projection unit is used to extract the device feature vector from the process tensor through a convolutional neural network, project the device feature vector into hyperbolic space and spherical space to generate a first projection and a second projection respectively, and then stitch them together to generate a stitched vector.
[0017] The decomposition unit is used to decompose the concatenated vector into a factor matrix, generate a linear combination vector of the constraint vector based on preset constraint weights, and generate a first tensor kernel based on the linear combination vector and the factor matrix.
[0018] The fusion unit is used to map the constraint vector to the tensor space through a preset fully connected layer, and fuse it with the first tensor kernel to generate a second tensor kernel;
[0019] The parameter unit is used to input the second tensor kernel into a preset target optimization module, solve for the appropriate parameter combination, and send the parameter combination to the control terminal.
[0020] According to one aspect of this application, a computer-readable medium is provided having a computer program stored thereon, which, when executed by a processor, implements the knowledge graph-based intelligent matching method for industrial production parameters as described in the above embodiments.
[0021] According to one aspect of this application, an electronic device is provided, comprising: one or more processors; and a storage device for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement the knowledge graph-based intelligent matching method for industrial production parameters as described in the above embodiments.
[0022] According to one aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the knowledge graph-based intelligent matching method for industrial production parameters provided in the various optional implementations described above.
[0023] In summary, the above process integrates and standardizes multidimensional production data, utilizes convolutional neural networks and multi-spatial projection to extract equipment features, and enhances feature representation capabilities. Constraints are explicitly integrated into factor matrices and tensor kernel generation to ensure accurate matching. Through the generation of structured process tensors and constraint vectors, raw industrial data is transformed into a "entity-relationship-attribute" triple form. Production equipment serves as the core entity, process parameters and environmental factors as attributes, and hierarchical relationships and parameter dependencies between equipment are quantified using hyperbolic / spherical projection, forming a dynamically updatable industrial knowledge graph. This graph is stored in tensor form, supporting efficient multidimensional relationship reasoning. Negative curvature geometry represents the equipment hierarchy, causing the distance between parent and child nodes in the knowledge graph to grow exponentially; for example, the distance from the central control system to production line equipment is greater than the distance between equipment within the production line, accurately reflecting the actual factory topology. Directional cosines are used to maintain symmetry, ensuring the equivalent expression of bidirectional dependencies in the knowledge graph.
[0024] The process rule configuration file is transformed into constraint edges of a knowledge graph, and an adaptive mapping from rules to the tensor space is achieved through a fully connected layer. This mechanism enables the knowledge graph to automatically switch constraints at different process stages. When multiple constraints conflict, they are automatically balanced using learned weight coefficients, such as the trade-off between energy consumption and quality. Finally, the gradient descent method is used to optimize and solve for the appropriate parameter combination, achieving intelligent matching and dynamic adjustment of production parameters, thereby improving the accuracy and efficiency of parameter matching during the production process.
[0025] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description
[0026] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.
[0027] Figure 1 schematically illustrates a flowchart of a knowledge graph-based intelligent matching method for industrial production parameters in one embodiment of this application.
[0028] Figure 2 schematically illustrates a flowchart of the generation process tensor and constraint vector in one embodiment of this application.
[0029] Figure 3 schematically illustrates a knowledge graph-based intelligent matching system for industrial production parameters in one embodiment of this application.
[0030] Figure 4 shows a schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application. Detailed Implementation
[0031] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art.
[0032] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this application.
[0033] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.
[0034] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.
[0035] The implementation details of the technical solution of this application are described below:
[0036] Figure 1 shows a flowchart of a knowledge graph-based intelligent matching method for industrial production parameters according to an embodiment of this application. Referring to Figure 1, the knowledge graph-based intelligent matching method for industrial production parameters includes at least steps S110 to S160, which are described in detail below:
[0037] S110, acquire process data during the factory production process.
[0038] In one embodiment of this application, process data during production is collected in real time through a sensor network deployed on factory equipment. This sensor network includes sensors for parameters such as temperature, pressure, and flow rate. The raw data is then uploaded to a central database via an Industrial Internet of Things (IIoT) protocol. The data is stored and categorized by timestamp, device node, and parameter type, forming a raw heterogeneous dataset.
[0039] In one embodiment of this application, process data includes production data, equipment data, environmental data, and configuration files that change with the process stages during production.
[0040] Optionally, the raw data can be preliminarily cleaned through preset edge nodes, such as filtering noise and filling missing values, and multi-source data streams can be aligned by timestamps and stored in structured databases, time-series databases, and file systems. Environmental data and equipment operation logs are associated and tagged, and configuration files are dynamically updated through version management, forming a traceable data link and providing a complete and consistent input source for subsequent standardized processing.
[0041] S120, preprocess the process data to generate standardized process tensors and constraint vectors.
[0042] In one embodiment of this application, the rule base in the process configuration file, such as safety thresholds and energy consumption limits, is parsed to extract static constraints (such as maximum equipment power) and dynamic constraints (such as time-period temperature limits). Dynamic constraints are associated with the process tensor through timestamp matching to construct a multi-dimensional constraint vector: numerical constraints are directly quantized, and logical constraints (such as equipment interlock rules) are converted into binary identifiers. Finally, a constraint vector matrix synchronized with the time dimension of the process tensor is generated, providing real-time rule input for downstream optimization modules.
[0043] As shown in Figure 2, in one embodiment of this application, the process data is preprocessed to generate standardized process tensors and constraint vectors, including:
[0044] S210, based on the timestamp of the process data, align the process data to obtain time-uniform process data;
[0045] S220, generate a process tensor from the process data according to three dimensions: time, parameter type, and device node;
[0046] S230, extract constraint rules from the configuration file, and construct a time-related constraint vector based on the constraint rules.
[0047] In one embodiment of this application, process data is stored as a structured file, including production data (such as temperature and pressure), equipment data (equipment connection relationships), and environmental data (such as humidity and vibration) from different sensors and devices. Constraints are defined in a configuration file, including static limits and dynamic rules, where static limits can be the maximum power of the equipment, and dynamic rules can be safety thresholds that change over time.
[0048] Multi-source data may experience timestamp misalignment due to differences in sensor sampling frequencies or transmission delays. By using a sliding window and interpolation, we ensure that data from all devices is fully aligned within the same time window, preventing subsequent analysis biases caused by missing or asynchronous data. This embodiment uses the smallest time unit, such as one second, to uniformly align the timestamps of multi-source data. For example, if data from a device is missing in a certain second, it is filled in using linear interpolation or nearest-neighbor values.
[0049] For example, a sliding window is used to process asynchronous data. For instance, a 1-minute window is defined, covering 60 time points, and sliding in 10-second increments. Missing data within the window is filled using linear or interpolation polynomials; high-frequency or low-frequency data is adjusted to a uniform frequency through aggregation or upsampling.
[0050] In this embodiment, process data is generated into process tensors based on time, parameter type, and device node. For example, each position of the three-dimensional process tensor represents the value of a certain type of parameter on a specific device at a certain time point. The standardized three-dimensional tensor can be directly input into the embedding module to generate a geometric representation of the device and parameters, improving the accuracy of hierarchical relationship modeling.
[0051] Constraint rules are extracted from the configuration file to construct a time-dependent dynamic constraint vector. For example, a dynamic rule that reduces the temperature threshold by 10% during the night shift will be updated in the constraint vector in real time. Constraints (such as safety thresholds) may change over time or at different stages of the process. Through the dynamic constraint vector, subsequent optimization modules can load the latest rules in real time, ensuring that parameter matching conforms to both current physical limitations and adapts to changes in the production environment. By combining the constraint vector with standardized data, the optimization algorithm considers real-time rules when searching for optimal parameters, avoiding invalid or dangerous solutions. Dynamic constraints support flexible adjustments to process rules, such as temporarily relaxing energy consumption limits during equipment maintenance, improving the system's ability to respond to sudden demands.
[0052] In the aforementioned process, data loading and preprocessing are the cornerstones of the intelligent matching system for industrial production parameters. Through time alignment, standardization, and dynamic constraint integration, the original heterogeneous data is transformed into a structured and standardized input format, eliminating interference from dimensions and asynchronicity while embedding real-time process rules. This process significantly improves the accuracy and reliability of subsequent knowledge graph construction and parameter optimization, providing a high-quality data foundation for intelligent manufacturing.
[0053] S130, using a convolutional neural network, extract the device feature vector from the process tensor, project the device feature vector into hyperbolic space and spherical space to generate a first projection and a second projection respectively, and then stitch them together to generate a stitched vector.
[0054] In one embodiment of this application, a standardized process tensor is processed using a three-dimensional convolutional neural network to extract device features layer by layer. The convolutional kernel slides along the time, parameter, and device dimensions to capture local patterns, such as temperature fluctuations in the device during specific process stages. Pooling layers compress redundant information, ultimately outputting a low-dimensional feature vector for each device. These vectors encode the device's operating characteristics and interaction patterns, providing input for subsequent spatial mapping.
[0055] In one embodiment of this application, a device feature vector is extracted from the process tensor using a convolutional neural network. The device feature vector is then projected into hyperbolic space and spherical space to generate a first projection and a second projection, respectively. These projections are then concatenated to generate a concatenated vector. The process includes:
[0056] Local features are extracted from the process tensor using a convolutional neural network and used as the device feature vector.
[0057] The device feature vectors are mapped onto hyperbolic space to generate a first projection;
[0058] The device feature vectors are mapped onto a spherical space to generate a second projection;
[0059] The first projection and the second projection are stitched together to generate a stitched vector.
[0060] In one embodiment of this application, the process tensor obtained through preprocessing includes various process parameters of different devices at different time points, such as temperature and pressure. For each device node, its corresponding multidimensional time-series data is extracted, that is, the data of the device at all times and for all parameters. The time-series data of each device is analyzed by a convolutional neural network. By identifying local patterns, such as short-term fluctuations and dependencies between parameters, the complex data of each device is compressed into a compact feature vector. For example, the temperature and pressure change trends of a certain device will be transformed into a low-dimensional vector, representing the key features of its operating state.
[0061] The device feature vectors are mapped to a hyperbolic space, where distances increase exponentially with hierarchy depth, similar to a tree structure. For example, the distance between the central control system and lower-level devices is magnified in the hyperbolic space, while the distances between devices at the same level are relatively small, generating a first projection h. j for:
[0062] Where k represents the curvature of the hyperbolic space, v j Let ||v| represent the device feature vector, j represent the device identifier, and ||v| represent the device feature vector. j || represents the norm of the device's eigenvector, and sinh(·) represents the hyperbolic sine function, which is used to nonlinearly amplify the vector norm and enhance the differences in hierarchical structure.
[0063] By adjusting the length of the eigenvectors using a nonlinear function, the hierarchical differences can be more clearly reflected in hyperbolic space. The exponentially growing distance characteristic of hyperbolic space is naturally suitable for representing tree-like hierarchical relationships, such as factory equipment topology. For example, the embedding distance between the central control system and the lower-level devices is much greater than the distance between devices at the same level.
[0064] The device feature vector is mapped onto a spherical space to generate a second projection s. j for:
[0065] Where τ represents the radius parameter of the spherical space.
[0066] Spherical space measures similarity by direction rather than distance, making it suitable for representing symmetrical relationships. For example, spherical space characterizes symmetric dependencies through directional similarity, avoiding the directional biases of traditional Euclidean space; the vector directions of two cooperating devices in spherical space will be highly similar; the temperatures of devices A and B influence each other, and their spherical projection vector directions will be similar. By adjusting the direction and magnitude of the feature vectors, symmetry can be maintained in spherical space, avoiding relationship biases caused by distance calculations.
[0067] Next, the first projection and the second projection are stitched together, and the embedding vectors of all devices are arranged according to time and device node to generate the stitching vector M. j =[h j s j By hybridizing manifolds to integrate hyperbolic and spherical geometry, and simultaneously modeling hierarchy and symmetry, this approach combines two spatial representations to enhance the model's ability to model complex technological relationships and address the problem of insufficient single-space representation capabilities.
[0068] Hyperbolic space excels at hierarchical relationships, while spherical space excels at symmetry relationships. Combining the two can simultaneously capture equipment topology and parameter reciprocity. The projection results from hyperbolic and spherical spaces are concatenated into a single composite vector. For example, if hyperbolic space generates a 64-dimensional vector and spherical space generates a 32-dimensional vector, the concatenation results in a 96-dimensional hybrid representation. This fusion of hierarchical structure and symmetry provides a more comprehensive geometric representation for subsequent analysis. Through multi-space fusion, the model's tolerance to complex process relationships is enhanced. For instance, even if some data noise interferes with hierarchical information, the symmetry modeling in spherical space can still provide supplementary information. If the embedded vector of a device deviates from the expected hierarchy or symmetry relationship, an alarm can be triggered, such as when the device is out of production line management or parameter dependencies break.
[0069] After extracting local features of equipment using a convolutional neural network, a dual representation of industrial equipment relationships is achieved through the collaborative projection of hyperbolic and spherical spaces. Hyperbolic space characterizes the equipment hierarchy topology through exponential distance variations, such as the tree-like structure of workshop-production line-equipment; spherical space maintains parameter symmetry dependencies through directional similarity, such as the bidirectional influence of collaborating equipment. This hybrid geometric embedding overcomes the expressive limitations of traditional Euclidean space, improving the accuracy of equipment hierarchy relationship recognition in scenarios such as semiconductor manufacturing.
[0070] S140, decompose the factor matrix from the spliced vector, generate a linear combination vector of the constraint vector based on the preset constraint weights, and generate the first tensor kernel based on the linear combination vector and the factor matrix.
[0071] In one embodiment of this application, a factor matrix is decomposed from the concatenated vector, a linear combination vector of the constraint vectors is generated based on preset constraint weights, and a first tensor kernel is generated based on the linear combination vector and the factor matrix, including:
[0072] A factor matrix is decomposed from the concatenated vector, and the factor matrix represents the potential correlation between data in different dimensions;
[0073] The constraint vector is linearly processed based on the preset constraint weights to generate a linear combination vector;
[0074] The linear combination vector and the factor matrix are operated on to generate the first tensor kernel.
[0075] In one embodiment of this application, a factor matrix is decomposed from the concatenated vector, and the factor matrix represents the potential correlation between data of different dimensions. Specifically, the original high-dimensional data, such as three-dimensional data of time, parameters, and devices, is decomposed into a low-dimensional core tensor and multiple factor matrices. By decomposing the high-dimensional data, core interaction patterns are preserved and redundant information is reduced. Core interaction patterns or long-term trends of parameters between devices are extracted from massive amounts of device data.
[0076] For example, in semiconductor manufacturing, decomposed core tensors may capture key patterns, such as high-frequency interactions with specific devices during high-temperature periods. Tensor decomposition significantly reduces data dimensionality and computational resource consumption. For instance, what would have required processing millions of data points can be reduced to analyzing only a thousand core patterns after decomposition, resulting in a speed increase of tens of times.
[0077] Then, based on the preset constraint weights, the constraint vector is linearly processed to generate a linear combination vector H(ɑ):
[0078] Where i represents the identifier of the constraint vector, p represents the total number of constraint vectors, α represents the index of the preset tensor network rank, and controls the relational complexity; ω i c represents the constraint weights learned through training. i (t) represents the value of the i-th constraint vector at time t, such as the temperature threshold.
[0079] By directly incorporating process constraints into the model, parameter combinations that violate these constraints are assigned high energy values, thus being naturally eliminated during the optimization process. This is similar to the red light mechanism in a traffic system, automatically preventing dangerous operations. Through energy constraints, the model automatically avoids invalid or dangerous solutions. For example, in lithium battery production, process parameters that easily lead to uneven coating under high humidity conditions are suppressed to ensure that the output always meets safety requirements.
[0080] Then, operations are performed on the linear combination vector and the factor matrix to generate the first tensor kernel T. ijk for:
[0081] in, represents the factor matrices decomposed from the concatenated vector, corresponding to device dimension i, relation dimension j, and attribute dimension k, respectively; Tr(·) represents the matrix trace operation, used to extract the core features of factor interactions; β represents the inverse temperature parameter; ɑ represents the index of the preset tensor network rank; and A represents the maximum value of the preset tensor network rank.
[0082] Through the above calculation process, the energy function is adjusted in real time as constraints change over time. For example, if the temperature threshold decreases during the night shift, the model will immediately lower the corresponding energy value to ensure that the parameters match the new rules. Real-time injection of dynamic constraints enables the system to respond quickly to changes in the production environment. For example, when equipment enters maintenance mode, energy consumption limits are temporarily relaxed, and the model automatically adjusts its parameter matching strategy.
[0083] By transforming complex process parameter matching into a controlled optimization process, high-risk solutions are dynamically filtered based on energy mechanisms, while dimensionality reduction techniques are used to improve computational efficiency. This method not only achieves high-precision matching but also endows the system with the ability to dynamically adapt and make transparent decisions, providing a reliable technological foundation for intelligent manufacturing.
[0084] S150, the constraint vector is mapped to the tensor space through a preset fully connected layer and fused with the first tensor kernel to generate the second tensor kernel.
[0085] In one embodiment of this application, the constraint vector is mapped to a tensor space through a preset fully connected layer and fused with the first tensor kernel to generate a second tensor kernel, including:
[0086] The constraint vector is mapped to the tensor space through a preset fully connected layer to generate an adjustment tensor, the adjustment tensor having the same dimension as the first tensor kernel;
[0087] The adjusted tensor is fused with the first tensor kernel to generate a second tensor kernel.
[0088] In one embodiment of this application, dynamic constraint fusion first receives the tensor kernel and real-time constraints from the previous step. A fully connected neural network is used to convert dynamic constraints (such as temperature thresholds or energy consumption limits) into specific adjustments to the tensor kernel. A pre-defined fully connected layer maps the constraint vector to the tensor space, generating an adjustment tensor with the same dimensions as the first tensor kernel. The association between constraints and tensor adjustments is automatically learned through training data, and the fully connected layer converts the constraints into adjustments to each dimension of the tensor kernel. For example, when the temperature threshold increases, the interaction weights of related devices need to be reduced; if the current environment requires reduced energy consumption, the fully connected layer generates corresponding weight adjustment values for the tensor kernel to suppress the intensity of high-energy-consumption parameter combinations.
[0089] The adjusted tensor is fused with the first tensor kernel to generate a second tensor kernel. Specifically, a linear function is used to element-wise add the activated adjusted tensor to the original tensor kernel to generate the final corrected result T′. ijk for:
[0090] T′ ijk =T ijk +σ(ΔT ijk )
[0091] Where σ(·) represents a linear function.
[0092] While preserving the original interaction pattern, constraint-driven adjustments are injected. For example, when a device has a high risk of overheating, its relevant weights are reduced. This preserves existing knowledge of the original tensor kernel, such as historical process patterns, while simultaneously incorporating the effects of real-time constraints.
[0093] Dynamic constraint fusion enables the system to respond in real time to changes in process rules while maintaining the stability of historical experience. For example, during sudden equipment maintenance, the system automatically reduces the weight of relevant parameters to ensure production safety without completely discarding long-accumulated effective patterns. This mechanism balances flexibility and reliability, providing dynamic adaptability for intelligent optimization.
[0094] S160, the second tensor kernel is input into the preset target optimization module to solve for the appropriate parameter combination, and the parameter combination is sent to the control terminal.
[0095] In one embodiment of this application, the second tensor kernel is input into a preset target optimization module to solve for a suitable parameter combination, including:
[0096] The second tensor kernel is input into the preset target optimization module, and the appropriate parameter combination is obtained by solving the gradient descent method.
[0097] The parameter combination is sent to the control terminal.
[0098] In one embodiment of this application, the dynamically adjusted weight matrix (tensor kernel) is first matched with real-time sensor data, and an optimization algorithm is used to find the optimal parameter combination. During the optimization process, the fit between the parameters and current production requirements is balanced, while avoiding excessive parameter fluctuations. For example, in the injection molding process, temperature and pressure parameters are dynamically adjusted to ensure stable product quality.
[0099] The core of the optimization is to balance real-time data matching and parameter stability. By comparing the adjusted weight matrix with real-time data, the error is calculated and the parameters are gradually optimized, while stability constraints are introduced to prevent extreme changes. Anomaly detection is achieved by decomposing the weight matrix to identify potential weak interaction patterns. When key indicators fall below a safety threshold, an alarm is triggered, such as device sensor failure or abnormal parameter dependencies.
[0100] This method can precisely adapt to dynamic production environments, improving process efficiency and reducing risks. For example, in steel smelting, optimized parameter combinations can save more than 10% of energy consumption, while anomaly detection can provide early warnings of equipment failures, reducing downtime. Through closed-loop feedback and continuous improvement, the efficiency and safety of the manufacturing process are ensured.
[0101] After generating the parameter combination, the combination is sent to the control terminal so that the operation of the relevant working equipment can be controlled through the control terminal.
[0102] This application's technical solution involves acquiring process data from a factory production process; preprocessing the process data to generate standardized process tensors and constraint vectors; extracting equipment feature vectors from the process tensors using a convolutional neural network; projecting the equipment feature vectors into hyperbolic space and spherical space to generate a first projection and a second projection, respectively, and then concatenating them to generate a concatenated vector; decomposing a factor matrix from the concatenated vector; generating a linear combination vector of the constraint vectors based on preset constraint weights; generating a first tensor kernel based on the linear combination vector and the factor matrix; mapping the constraint vectors to the tensor space through a preset fully connected layer and fusing them with the first tensor kernel to generate a second tensor kernel; inputting the second tensor kernel into a preset target optimization module to solve for a suitable parameter combination; and sending the parameter combination to a control terminal.
[0103] In summary, the above process integrates and standardizes multidimensional production data, utilizes convolutional neural networks and multi-spatial projection to extract equipment features, and enhances feature representation capabilities. Constraints are explicitly integrated into factor matrices and tensor kernel generation to ensure accurate matching. Through the generation of structured process tensors and constraint vectors, raw industrial data is transformed into a "entity-relationship-attribute" triple form. Production equipment serves as the core entity, process parameters and environmental factors as attributes, and hierarchical relationships and parameter dependencies between equipment are quantified using hyperbolic / spherical projection, forming a dynamically updatable industrial knowledge graph. This graph is stored in tensor form, supporting efficient multidimensional relationship reasoning. Negative curvature geometry characterizes equipment hierarchical relationships, causing the distance between parent and child nodes in the knowledge graph to grow exponentially (e.g., the distance from the central control system to production line equipment > the distance between equipment within the production line), accurately reflecting the actual factory topology. Directional cosines are used to maintain symmetry, ensuring the equivalent expression of bidirectional dependencies in the knowledge graph.
[0104] The process rule configuration file is transformed into constraint edges of a knowledge graph, and an adaptive mapping from rules to the tensor space is achieved through a fully connected layer. This mechanism enables the knowledge graph to automatically switch constraints at different process stages. When multiple constraints conflict, they are automatically balanced using learned weight coefficients, such as the trade-off between energy consumption and quality. Finally, the gradient descent method is used to optimize and solve for the appropriate parameter combination, achieving intelligent matching and dynamic adjustment of production parameters, thereby improving the accuracy and efficiency of parameter matching during the production process.
[0105] The following describes an embodiment of the apparatus described in this application, which can be used to execute the knowledge graph-based intelligent matching method for industrial production parameters described in the above embodiments of this application. It is understood that the apparatus can be a computer program (including program code) running on a computer device, for example, the apparatus is an application software; the apparatus can be used to execute the corresponding steps in the method provided in the embodiments of this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the knowledge graph-based intelligent matching method for industrial production parameters described above.
[0106] Figure 3 shows a block diagram of a knowledge graph-based intelligent matching system for industrial production parameters according to an embodiment of this application.
[0107] Referring to Figure 3, an embodiment of the knowledge graph-based intelligent matching system for industrial production parameters according to this application includes:
[0108] Acquisition unit 310 is used to acquire process data during the factory production process;
[0109] Preprocessing unit 320 is used to preprocess the process data to generate standardized process tensors and constraint vectors;
[0110] Projection unit 330 is used to extract device feature vectors from the process tensor through a convolutional neural network, project the device feature vectors in hyperbolic space and spherical space to generate a first projection and a second projection respectively, and then stitch them together to generate a stitched vector.
[0111] Decomposition unit 340 is used to decompose the concatenated vector into a factor matrix, generate a linear combination vector of the constraint vector based on preset constraint weights, and generate a first tensor kernel based on the linear combination vector and the factor matrix.
[0112] The fusion unit 350 is used to map the constraint vector to the tensor space through a preset fully connected layer, and fuse it with the first tensor kernel to generate a second tensor kernel;
[0113] The parameter unit 360 is used to input the second tensor kernel into a preset target optimization module, solve for the appropriate parameter combination, and send the parameter combination to the control terminal.
[0114] In this application, based on the aforementioned scheme, the process data includes production data, equipment data, environmental data, and configuration files that change with the process stages during production.
[0115] In this application, based on the aforementioned scheme, the preprocessing of the process data to generate standardized process tensors and constraint vectors includes: aligning the process data based on the timestamps of the process data to obtain time-uniform process data; generating process tensors from the process data according to three dimensions: time, parameter type, and device node; extracting constraint rules from the configuration file, and constructing time-related constraint vectors based on the constraint rules.
[0116] In this application, based on the aforementioned scheme, the step of extracting device feature vectors from the process tensor using a convolutional neural network, projecting the device feature vectors into hyperbolic space and spherical space to generate a first projection and a second projection respectively, and then concatenating them to generate a concatenated vector, includes: extracting local features from the process tensor using a convolutional neural network as the device feature vectors; mapping the device feature vectors onto hyperbolic space to generate a first projection; mapping the device feature vectors onto spherical space to generate a second projection; and concatenating the first projection and the second projection to generate a concatenated vector.
[0117] In this application, based on the aforementioned scheme, the step of decomposing a factor matrix from the concatenated vector, generating a linear combination vector of the constraint vectors based on preset constraint weights, and generating a first tensor kernel based on the linear combination vector and the factor matrix includes: decomposing a factor matrix from the concatenated vector, wherein the factor matrix represents the potential correlation between data of different dimensions; performing linear processing on the constraint vectors based on preset constraint weights to generate a linear combination vector; and performing operations on the linear combination vector and the factor matrix to generate a first tensor kernel.
[0118] In this application, based on the aforementioned scheme, the step of mapping the constraint vector to a tensor space through a preset fully connected layer and fusing it with the first tensor kernel to generate a second tensor kernel includes: mapping the constraint vector to a tensor space through a preset fully connected layer to generate an adjustment tensor, wherein the adjustment tensor has the same dimension as the first tensor kernel; and fusing the adjustment tensor with the first tensor kernel to generate a second tensor kernel.
[0119] In this application, based on the aforementioned scheme, the step of inputting the second tensor kernel into a preset target optimization module to obtain a suitable parameter combination includes: inputting the second tensor kernel into a preset target optimization module, obtaining a suitable parameter combination through gradient descent; and sending the parameter combination to a control terminal.
[0120] This application's technical solution involves acquiring process data from a factory production process; preprocessing the process data to generate standardized process tensors and constraint vectors; extracting equipment feature vectors from the process tensors using a convolutional neural network; projecting the equipment feature vectors into hyperbolic space and spherical space to generate a first projection and a second projection, respectively, and then concatenating them to generate a concatenated vector; decomposing a factor matrix from the concatenated vector; generating a linear combination vector of the constraint vectors based on preset constraint weights; generating a first tensor kernel based on the linear combination vector and the factor matrix; mapping the constraint vectors to the tensor space through a preset fully connected layer and fusing them with the first tensor kernel to generate a second tensor kernel; inputting the second tensor kernel into a preset target optimization module to solve for a suitable parameter combination; and sending the parameter combination to a control terminal. By integrating and standardizing multidimensional production data, convolutional neural networks and multi-spatial projection are used to extract equipment features, thereby enhancing feature representation capabilities. Constraints are explicitly incorporated into factor matrices and tensor kernel generation to ensure that the matching conforms to reality. Finally, the gradient descent method is used to optimize and solve the combination of adaptation parameters, thereby achieving intelligent matching and dynamic adjustment of production parameters and improving the accuracy and efficiency of parameter matching in the production process.
[0121] Figure 4 shows a schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application.
[0122] It should be noted that the computer system of the electronic device in this embodiment is only an example and should not impose any limitations on the function and scope of use of the embodiments of this application.
[0123] In this embodiment, the computer system includes a central processing unit 401, which can perform various appropriate actions and processes based on a program stored in a read-only memory 402 or a program loaded from a storage section 408 into a random access memory 403, such as executing the knowledge graph-based intelligent matching method for industrial production parameters described in the above embodiment. The random access memory 403 also stores various programs and data required for system operation. The central processing unit 401, the read-only memory 402, and the random access memory 403 are interconnected via a bus 404. An input / output interface 405 is also connected to the bus 404.
[0124] The following components are connected to the input / output interface 405: an input section 406 including a keyboard, mouse, etc.; an output section 407 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 408 including a hard disk, etc.; and a communication section 409 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 409 performs communication processing via a network such as the Internet. A drive 410 is also connected to the input / output interface 405 as needed. A removable medium 411, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 410 as needed so that computer programs read from it can be installed into the storage section 408 as needed.
[0125] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 409, and / or installed from removable medium 411. When the computer program is executed by central processing unit 401, it performs various functions defined in the system of this application.
[0126] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium or 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 thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying a computer-readable computer program. The transmitted data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The 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 connection with an instruction execution system, apparatus, or device. The computer program contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.
[0127] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.
[0128] The units described in the embodiments of this application can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.
[0129] According to one aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various alternative implementations described above.
[0130] In another aspect, this application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to implement the knowledge graph-based intelligent matching method for industrial production parameters described in the above embodiments.
[0131] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.
[0132] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the method according to the embodiments of this application.
[0133] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.
[0134] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.
Claims
1. An intelligent matching method for industrial production parameters based on a knowledge graph, characterized in that, The method comprises the following steps: acquiring process data in a factory production process; preprocessing the process data to generate a standardized process tensor and a constraint vector; extracting a device feature vector from the process tensor through a convolutional neural network, projecting the device feature vector in hyperbolic space and spherical space to generate a first projection and a second projection respectively, and then splicing the first projection and the second projection to generate a spliced vector; decomposing a factor matrix from the spliced vector, generating a linear combination vector of the constraint vector based on a preset constraint weight, and generating a first tensor core based on the linear combination vector and the factor matrix; mapping the constraint vector to a tensor space through a preset fully connected layer, fusing the constraint vector with the first tensor core, and generating a second tensor core; inputting the second tensor core into a preset target optimization module to obtain an adaptive parameter combination, and sending the parameter combination to a control terminal. 2.The knowledge graph-based intelligent matching method for industrial production parameters according to claim 1, characterized in that, The process data includes production data, device data, environment data, and configuration files that change with process stages in the production process. 3.The knowledge graph-based intelligent matching method for industrial production parameters according to claim 2, characterized in that, The preprocessing of the process data to generate a standardized process tensor and a constraint vector comprises the following steps: aligning the process data based on the timestamps of the process data to obtain time-unified process data; generating a process tensor from the process data according to time, parameter type, and device node; extracting constraint rules from the configuration files and constructing time-associated constraint vectors based on the constraint rules. 4.The knowledge graph-based intelligent matching method for industrial production parameters according to claim 1, characterized in that, The method for extracting a device feature vector from the process tensor through a convolutional neural network, projecting the device feature vector in hyperbolic space and spherical space to generate a first projection and a second projection respectively, and then splicing the first projection and the second projection to generate a spliced vector comprises the following steps: extracting local features from the process tensor through a convolutional neural network as the device feature vector; mapping the device feature vector to hyperbolic space to generate a first projection; mapping the device feature vector to spherical space to generate a second projection; splicing the first projection and the second projection to generate a spliced vector. 5.The knowledge graph-based intelligent matching method for industrial production parameters according to claim 1, characterized in that, The method for decomposing a factor matrix from the spliced vector, generating a linear combination vector of the constraint vector based on a preset constraint weight, and generating a first tensor core based on the linear combination vector and the factor matrix comprises the following steps: decomposing a factor matrix from the spliced vector, wherein the factor matrix represents the potential association relationship between data of different dimensions; linearly processing the constraint vector based on a preset constraint weight to generate a linear combination vector; operating the linear combination vector and the factor matrix to generate a first tensor core. 6.The knowledge graph-based intelligent matching method for industrial production parameters according to claim 1, characterized in that, The method for mapping the constraint vector to a tensor space through a preset fully connected layer, fusing the constraint vector with the first tensor core, and generating a second tensor core comprises the following steps: mapping the constraint vector to a tensor space through a preset fully connected layer to generate an adjustment tensor, wherein the adjustment tensor has the same dimension as the first tensor core; fusing the adjustment tensor with the first tensor core to generate a second tensor core. 7.The knowledge graph-based intelligent matching method for industrial production parameters according to claim 1, characterized in that, The second tensor core is input into a preset target optimization module, and an adaptive parameter combination is obtained by solving, and the parameter combination is sent to a control terminal, including: The second tensor core is input into a preset target optimization module, and an adaptive parameter combination is obtained by solving, and the parameter combination is sent to a control terminal, including: The parameter combination is sent to a control terminal.
8. An intelligent matching system for industrial production parameters based on a knowledge graph, characterized in that, Including: An acquisition unit is configured to acquire process data in a factory production process; A preprocessing unit is configured to preprocess the process data to generate a standardized process tensor and a constraint vector; A projection unit is configured to extract a device feature vector from the process tensor through a convolutional neural network, project the device feature vector in hyperbolic space and spherical space to generate a first projection and a second projection respectively, and then splice the first projection and the second projection to generate a spliced vector; A decomposition unit is configured to decompose a factor matrix from the spliced vector, generate a linear combination vector of the constraint vector based on a preset constraint weight, and generate a first tensor core based on the linear combination vector and the factor matrix; A fusion unit is configured to map the constraint vector to a tensor space through a preset full connection layer, and fuse the constraint vector with the first tensor core to generate a second tensor core; A parameter unit is configured to input the second tensor core into a preset target optimization module, obtain an adaptive parameter combination by solving, and send the parameter combination to a control terminal. 9.The knowledge graph-based intelligent matching system for industrial production parameters according to claim 8, characterized in that, The process data includes production data, device data, environment data, and configuration files that change with process stages in the production process. 10.The knowledge graph-based intelligent matching system for industrial production parameters according to claim 9, characterized in that, The process data is preprocessed to generate a standardized process tensor and a constraint vector, including: Aligning the process data based on timestamps of the process data to obtain time-unified process data; Generating a process tensor according to three dimensions of time, parameter type, and device node from the process data; Extracting constraint rules from the configuration file and constructing a constraint vector associated with time based on the constraint rules.
