An industrial internet-based digital operation management method, device and medium

Abstract

The application discloses a kind of based on industrial internet's digital operation management method, equipment and medium, it is related to operation management technical field, including, acquisition multi-source heterogeneous data, through cause-effect discovery algorithm extraction each enterprise internal process parameter and the cause-effect relationship between performance indicators, utilize deep learning model and encode cause-effect relationship as local knowledge vector;According to the expected performance gain value descending order sorting candidate knowledge vector, and collect the real performance data after target enterprise actual application candidate knowledge vector;Based on adjusted performance weight, the contribution degree of each enterprise in global knowledge value network is calculated by Shapley value algorithm, generates incentive allocation scheme and triggers smart contract execution.The application calculates the contribution degree of each enterprise in global knowledge value network by Shapley value algorithm, can be based on the dynamic performance weight of knowledge node accurately quantifies the marginal value contribution of each knowledge source in cross-enterprise collaboration.

Description

Technical Field

[0001] This invention relates to the field of operations management technology, and in particular to a digital operations management method, equipment and medium based on the industrial internet. Background Technology

[0002] Federated learning and secure multi-party computation technologies are now widely used in the industrial internet field to achieve cross-enterprise data collaboration and model training, while ensuring that local data remains within its domain. Digital twins and causal discovery algorithms are also being gradually introduced into industrial scenarios to build virtual simulation environments and uncover causal mechanisms between process parameters, providing decision support for production optimization. These technologies have laid a preliminary foundation for the secure sharing and collaborative optimization of industrial knowledge.

[0003] However, existing methods often focus on data or model aggregation, lacking dynamic measurement of the value of knowledge itself and credible incentive mechanisms. Knowledge contributors often fail to receive rewards commensurate with their actual value, leading to insufficient willingness among enterprises to share high-value knowledge. Furthermore, traditional incentive allocation methods are often based on data volume or transaction frequency, failing to reflect the true marginal contribution of knowledge in cross-enterprise applications, thus limiting the sustainable development of the industry's knowledge ecosystem. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a digital operation and management method based on the industrial internet to solve the problems of insufficient participation and weak ecological sustainability in cross-enterprise knowledge sharing due to the lack of value quantification and fair incentive mechanisms.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] Firstly, this invention provides a digital operation management method based on the industrial internet, comprising: collecting multi-source heterogeneous data; extracting causal relationships between internal process parameters and performance indicators of each enterprise through a causal discovery algorithm; encoding the causal relationships into local knowledge vectors using a deep learning model; performing federated secure aggregation of the enterprises' local knowledge vectors through a secure multi-party computation protocol to construct a global knowledge value network containing knowledge nodes, performance weights, and compatibility relationships; responding to a knowledge query request issued by a target enterprise; retrieving matching candidate knowledge vectors from the global knowledge value network; and inputting the candidate knowledge vectors into the target enterprise's local digital twin; and injecting... The process parameter adjustment strategy corresponding to the candidate knowledge vector is used to simulate the production status changes after the execution of the process parameter adjustment strategy by running a counterfactual reasoning engine, and the expected efficiency gain value is output. The candidate knowledge vectors are sorted in descending order according to the expected efficiency gain value, and the actual efficiency data after the target enterprise actually applies the candidate knowledge vectors is collected. The actual efficiency data is compared and analyzed with the expected efficiency gain value to generate a knowledge verification deviation signal. The efficiency weight of the corresponding knowledge node in the global knowledge value network is dynamically adjusted according to the knowledge verification deviation signal. Based on the adjusted efficiency weight, the contribution of each enterprise in the global knowledge value network is calculated by the Shapley value algorithm, an incentive allocation scheme is generated, and the smart contract is executed.

[0008] As a preferred embodiment of the digital operation management method based on the industrial internet described in this invention, the multi-source heterogeneous data includes real-time time-series data, production work order data, quality inspection data, and temperature and humidity data.

[0009] The specific steps for encoding causal relationships into local knowledge vectors using a deep learning model are as follows.

[0010] Multi-source heterogeneous data is input into the causal discovery algorithm. Based on the directed acyclic graph output by the causal discovery algorithm, the causal strength value between process parameter nodes and performance indicator nodes is extracted.

[0011] The features of each node and the corresponding causal strength value are input into the graph neural network encoder. The information of neighboring nodes is aggregated through the multi-layer message passing mechanism of the graph neural network encoder to generate a global graph representation vector as a local knowledge vector.

[0012] As a preferred embodiment of the digital operation and management method based on the industrial internet described in this invention, the secure multi-party computation protocol refers to a scheme that uses a homomorphic encryption algorithm to encrypt and transmit local knowledge vectors; the federated secure aggregation refers to a weighted average operation performed on the encrypted local knowledge vectors using a secure aggregation algorithm.

[0013] The specific steps for constructing a global knowledge value network that includes knowledge nodes, performance weights, and compatibility relationships are as follows.

[0014] Decrypt the local knowledge vector to obtain the global knowledge vector set, and calculate the cosine similarity between any two vectors in the global knowledge vector set.

[0015] Knowledge vectors with cosine similarity exceeding a preset compatibility association threshold are used to establish compatibility relationship edges. Performance weights are initialized for each knowledge node, and an adjacency matrix is ​​constructed to form a global knowledge value network.

[0016] As a preferred embodiment of the digital operation management method based on the industrial internet described in this invention, the knowledge query request refers to a formatted request initiated by the target enterprise for production optimization goals, which includes query conditions and constraint parameters; the retrieval matching refers to calculating the similarity between the features of the knowledge query request and the features of knowledge nodes in the global knowledge value network, and selecting candidate knowledge vectors.

[0017] As a preferred embodiment of the digital operation management method based on the industrial internet described in this invention, the process parameter adjustment strategy includes a temperature setpoint adjustment range, a pressure control target value change sequence, an equipment operating rate adjustment gradient, and a material proportion adjustment ratio.

[0018] The specific architecture of the counterfactual reasoning engine is as follows.

[0019] Receive the process parameter adjustment strategy corresponding to the candidate knowledge vector, and execute the intervention operation of the process parameter adjustment strategy in the local digital twin;

[0020] The simulation model of the local digital twin is run to simulate changes in production status, and the changes in performance indicators before and after the intervention are calculated as the expected performance gain value.

[0021] As a preferred embodiment of the digital operation management method based on the industrial internet described in this invention, the knowledge verification deviation signal includes two types: negative deviation signal and positive deviation signal generation.

[0022] The specific steps for generating the knowledge verification bias signal are as follows:

[0023] Obtain the actual performance data of the target company after applying the process parameter adjustment strategy, and calculate the absolute error between the actual performance data and the expected performance gain value;

[0024] The relative error rate is generated based on the ratio between the absolute error value and the expected performance gain value.

[0025] If the relative error rate exceeds the preset allowable deviation range, a negative deviation signal is generated; otherwise, a positive deviation signal is generated.

[0026] As a preferred embodiment of the digital operation management method based on the industrial internet described in this invention, the specific steps for dynamically adjusting the effectiveness weights of corresponding knowledge nodes in the global knowledge value network based on knowledge verification deviation signals are as follows:

[0027] Analyze the knowledge to verify the relative error rate in the deviation signal and identify the signal type;

[0028] The first coefficient of the performance weight is adjusted based on the negative deviation signal.

[0029] The second coefficient of the performance weight is adjusted based on the positive deviation signal.

[0030] The adjusted performance weights are then updated to the corresponding knowledge nodes in the global knowledge value network.

[0031] As a preferred embodiment of the digital operation management method based on the industrial internet described in this invention, the specific steps for generating the incentive allocation scheme and triggering the execution of the smart contract are as follows:

[0032] Traverse all knowledge nodes in the global knowledge value network after adjustment of their performance weights, and aggregate and calculate the sum of performance weights of the knowledge nodes belonging to each enterprise based on the affiliation of the knowledge nodes.

[0033] The sum of performance weights is used as the contribution value of each enterprise. The contribution ratio of each enterprise is allocated through the Shapley value algorithm. The contribution ratio is converted into token quantity and written into the blockchain smart contract, which triggers the smart contract to automatically execute the token transfer operation.

[0034] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the digital operation management method based on the industrial internet as described in the first aspect of the present invention.

[0035] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the digital operation management method based on the industrial internet as described in the first aspect of the present invention.

[0036] The beneficial effects of this invention are as follows: By calculating the contribution of each enterprise to the global knowledge value network using the Shapley value algorithm, the marginal value contribution of each knowledge source in cross-enterprise collaboration can be accurately quantified based on the dynamic effectiveness weight of knowledge nodes; the Shapley value algorithm calculates the change in the total value of the alliance before and after an enterprise joins by traversing all possible combinations of sub-alliances, thereby establishing a quantitative evaluation system based on multi-dimensional synergistic effects. This allocation mechanism based on cooperative game theory ensures that the incentive allocation strictly corresponds to the actual effectiveness gain generated by knowledge, and achieves automated token allocation through smart contracts, forming a verifiable and traceable value return system, effectively stimulating the intrinsic motivation of enterprises to continuously participate in high-quality knowledge sharing. Attached Figure Description

[0037] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a flowchart of a digital operation and management method based on the industrial internet.

[0039] Figure 2 A flowchart for constructing a global knowledge value network.

[0040] Figure 3 A flowchart for the operation of the counterfactual reasoning engine.

[0041] Figure 4 A flowchart for assigning incentives based on Shapley values. Detailed Implementation

[0042] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0043] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0044] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0045] Reference Figures 1-4 This is one embodiment of the present invention, which provides a digital operation management method based on the industrial internet, including the following steps:

[0046] S1: Collect multi-source heterogeneous data, extract the causal relationship between internal process parameters and performance indicators of each enterprise through causal discovery algorithm, and encode the causal relationship into local knowledge vector using deep learning model.

[0047] S1.1: Multi-source heterogeneous data includes real-time time series data, production work order data, quality inspection data, and temperature and humidity data.

[0048] Specifically, various production equipment and sensors in the factory are connected to data acquisition platforms (such as MES and QMS) to collect real-time time-series data, production work order data, quality inspection data, and temperature and humidity data through data interfaces.

[0049] Real-time time-series data includes time-series data collected by equipment sensors, such as equipment operating speed and current values; production work order data includes production order information and process execution records; quality inspection data includes product inspection results and defect statistics; and temperature and humidity data includes readings from environmental temperature and humidity sensors. After collection of multi-source heterogeneous data, it is stored in a unified data storage system for future reference.

[0050] S1.2: Input multi-source heterogeneous data into the causal discovery algorithm, and extract the causal strength value between process parameter nodes and performance indicator nodes based on the directed acyclic graph output by the causal discovery algorithm.

[0051] Specifically, the collected multi-source heterogeneous data will be used for variable identification to determine process parameter variables X (such as temperature setpoint and pressure control value) and performance indicator variables Y (such as production efficiency and product quality score).

[0052] The PC causal discovery algorithm was used to process process parameter variables and performance indicator variables.

[0053] Furthermore, the conditional independence between process parameter variables and performance indicator variables is calculated using a chi-square test. For any two variables X and Y, given the set of other variables Z, the chi-square statistic and p-value are calculated. The p-value is compared with a significance level of 0.05. If the p-value is greater than 0.05, then X and Y are considered independent given Z.

[0054] It should be noted that the p-value, also known as the probability value or significance probability, measures the likelihood of the observed data (or more extreme data) occurring under the assumption that "variables X and Y are conditionally independent" (called the null hypothesis). A significance level of 0.05 is a commonly used threshold in statistics to determine whether a hypothesis test result is statistically significant. It controls the probability of a Type I error (i.e., incorrectly rejecting the true null hypothesis) in hypothesis testing. A significance level of 0.05 represents a 5% probability of observing a more extreme test result if the null hypothesis is true, thus providing an acceptable margin of error for determining the conditional independence between variables.

[0055] Based on the results of the conditional independence test, construct an undirected skeleton graph. Initialize a completely undirected graph where each variable is a node and there is an edge between each pair of nodes. For each pair of variables (X, Y), perform a conditional independence test given a set of conditions for all possible subsets. If there exists a set of conditions Z such that X and Y are conditionally independent, remove the edge between X and Y.

[0056] After constructing the undirected skeleton graph, the edges are oriented to form a directed acyclic graph.

[0057] The edge orientation process consists of two main stages: step C1 and step C2.

[0058] Step C1: Identify and orient all potential V-structures (also known as collision points). Traverse all triple variables X, Y, Z in the graph that meet the conditions: If there exists an XYZ connection structure (i.e., X is connected to Y, and Y is connected to Z), and X and Z are not adjacent (i.e., there is no direct edge connection), and based on the previous conditional independence test, Y is not in the conditional independence set of X and Z, then orient the originally undirected edges XY and YZ to X→Y and Z→Y, thus forming a V-structure of X→Y←Z.

[0059] Step C2: After completing the V-structure orientation, apply the orientation propagation rules to avoid cyclic paths in the directed acyclic graph. The orientation propagation rules are as follows: If there is an already oriented edge X→Y, and there is an undirected edge YZ between Y and Z, but X and Z are not adjacent, then orient the undirected edge YZ to Y→Z; If there is a directed path X→Y→Z, and there is an undirected edge XZ between X and Z, then orient the undirected edge XZ to X→Z; If there is an undirected edge XY, and there is another directed path X→Z→Y, then orient the undirected edge XY to X→Y.

[0060] By applying V-structure recognition and directional propagation rules, the orientation of all edges in the directed acyclic graph (DAG) is completed. At this point, the PC causal discovery algorithm has completed the entire process of constructing the DAG, outputting a complete DAG. In the DAG, nodes represent process parameters or performance indicators, and edges represent causal relationships.

[0061] A superior approach is to construct a directed acyclic graph and quantify causal strength using a causal discovery algorithm (PC algorithm). This overcomes the inherent flaw of traditional methods that can only identify correlations rather than true causal relationships, avoids erroneous decisions caused by confused variables or spurious correlations, and thus accurately extracts the causal mechanism between process parameters and performance indicators.

[0062] Based on a directed acyclic graph (DAG), the causal strength value is estimated for each edge pointing from a process parameter node to a performance indicator node. Specifically, a linear regression model is used for estimation: with the process parameter variable as the independent variable and the performance indicator variable as the dependent variable, a linear regression model is fitted, and the regression coefficients are calculated as the causal strength values. The linear regression model uses the least squares method to estimate the parameters. The causal strength value represents the degree of influence of the process parameter on the performance indicator. The linear regression model adopts an on-demand training method, that is, for each causal relationship edge pointing from the process parameter node to the performance indicator node, the least squares method is used in real time to fit the current data to obtain the causal strength value.

[0063] For each node in the directed acyclic graph (DAG), node features are extracted from multi-source heterogeneous data. Node features include the historical mean, historical standard deviation, and latest observation of the variable. The calculation time window is a preset value covering recent production cycles (e.g., the last 30 days). Edge weights use the corresponding causal strength values. The node features and edge weights are organized into a directed acyclic graph data structure, where the node feature vector contains numerical features, and the edge weight matrix contains causal strength values.

[0064] S1.3: Input the features of each node and the corresponding causal strength value into the graph neural network encoder. Aggregate the information of neighboring nodes through the multi-layer message passing mechanism of the graph neural network encoder to generate a global graph representation vector as a local knowledge vector.

[0065] The node feature vectors and edge weight matrices are input into the graph neural network encoder. The graph neural network encoder adopts a two-layer graph convolutional network architecture: the first layer uses the ReLU activation function to aggregate the direct neighbor information of each node and update the node representation; the second layer uses the linear activation function to further aggregate the information of more distant neighbors.

[0066] In the message passing mechanism, the feature representation of each node in the graph neural network encoder The update formula is:

[0067] ;

[0068] In the formula, This represents the target node obtained after calculation. In the The new node feature representation of the layer, This indicates the index of the target node whose feature representation needs to be updated. Represents the first in a graph neural network layer, Represents the first in a graph neural network layer, Represents the ReLU activation function. Indicates the first The trainable weight matrix of the layer, Indicates a node Iterate through all neighboring nodes. For the index of neighboring nodes, This indicates a node. The set of all neighboring nodes, Indicates from neighboring nodes Point to target node The weight of the edge. Represents the target node The number of neighboring nodes, Representing neighboring nodes In the Layer node feature representation, Indicates the first Trainable bias vectors of the layer;

[0069] It should be noted that, These are the standard parameters for a graph neural network, learned through training data. The causal strength value is derived from the causal discovery algorithm. It is a standard component of neural network layers and is optimized through the backpropagation algorithm; It extracts node features from multi-source heterogeneous data, including specific features such as the historical mean, historical standard deviation, and latest observed value of variables;

[0070] A global average pooling layer is applied to average the feature representations of all nodes, generating a global graph representation vector. This global graph representation vector is the local knowledge vector.

[0071] Example: For a manufacturing company, process parameters include furnace temperature and press pressure, while performance indicators include product qualification rate and production cycle time. After collecting data on these variables, the PC causal discovery algorithm outputs a directed acyclic graph, showing a direct causal relationship between furnace temperature and product qualification rate; the linear regression estimate of the causal strength is 0.75. Node features include the mean (150°C) and standard deviation (5°C) of the furnace temperature, and the mean (100MPa) and standard deviation (2MPa) of the press pressure. After processing by the graph neural network encoder, a fixed-dimensional global graph representation vector is output as the local knowledge vector.

[0072] In a better way, the semantically rich local knowledge vectors generated by graph neural network encoding can be more deeply integrated with graph structure information and node features compared with traditional feature engineering methods (such as principal component analysis), thereby improving the discriminativeness and transferability of knowledge representation.

[0073] S2: Through a secure multi-party computation protocol, the enterprise's local knowledge vectors are aggregated in a federated manner to construct a global knowledge value network that includes knowledge nodes, performance weights, and compatibility relationships.

[0074] It should be noted that the secure multi-party computation protocol refers to a scheme that uses homomorphic encryption algorithms to encrypt and transmit local knowledge vectors; the federated secure aggregation refers to a weighted average operation on the encrypted local knowledge vectors using a secure aggregation algorithm.

[0075] S2.1: Decrypt the local knowledge vector to obtain the global knowledge vector set, and calculate the cosine similarity between any two vectors in the global knowledge vector set.

[0076] Specifically, a central aggregation server generates key pairs for the Paillier homomorphic encryption algorithm, including a public key and a private key. The public key is distributed to all enterprises participating in the federated security aggregation, while the private key is securely stored by the aggregation server. Each enterprise uses its public key to encrypt each dimension element of its local knowledge vector, generating an encrypted local knowledge vector. The encryption computation is performed locally, ensuring that the original local knowledge vector does not leave the enterprise's local environment.

[0077] Each enterprise transmits its encrypted local knowledge vectors to the central aggregation server via a secure network channel. Transport layer security protocols are used to ensure secure communication during the transmission process.

[0078] The central aggregation server receives encrypted local knowledge vectors from all participating enterprises. Federated secure aggregation employs a weighted averaging operation, where the aggregation server calculates the arithmetic mean of all encrypted local knowledge vectors. Because the Paillier homomorphic encryption algorithm supports the additive homomorphic property (meaning the sum of encrypted values ​​equals the sum of encrypted values), the averaging of the encrypted local knowledge vectors is performed in an encrypted state. The resulting aggregation is an encrypted global knowledge vector. The weights of each local knowledge vector in the weighted averaging operation are set to equal weights by default.

[0079] The aggregation server uses the stored Paillier homomorphic encryption algorithm private key to decrypt the encrypted global knowledge vector, obtaining a set of plaintext global knowledge vectors, which contains the aggregated knowledge representations of all participating enterprises.

[0080] For each global knowledge vector in the global knowledge vector set, calculate the pairwise cosine similarity between the global knowledge vector and all other global knowledge vectors in the global knowledge vector set.

[0081] The formula for calculating cosine similarity is:

[0082] Cosine similarity = ;

[0083] In the formula, and These represent two different global knowledge vectors. This represents the dot product operation. Let f(x) denote the Euclidean norm of the vectors, and calculate the similarity values ​​for all vector pairs.

[0084] S2.2: Establish compatibility relationship edges for knowledge vectors whose cosine similarity exceeds the preset compatibility association threshold, initialize performance weights for each knowledge node and construct an adjacency matrix to form a global knowledge value network.

[0085] Specifically, a compatibility association threshold is set. The compatibility association threshold is determined by calculating the weighted sum of the average and standard deviation of the cosine similarity between all pairs of vectors in the global knowledge vector set. The suggested value range is 0.7-0.85.

[0086] Iterate through all calculated cosine similarity values. When the cosine similarity value between a pair of global knowledge vectors exceeds the compatibility association threshold, establish an undirected compatibility relationship edge between the knowledge nodes represented by these two global knowledge vectors (the establishment process is based on the principle of undirected edge connection in graph theory). The compatibility relationship edge indicates that the process knowledge contained in the two knowledge nodes has high compatibility and synergy.

[0087] Initialize a performance weight for each knowledge node in the global knowledge vector set. The initial value of the performance weight is uniformly set to 1.0. The performance weight is used to quantify the expected effectiveness of the process knowledge represented by the knowledge node. The initial value of the performance weight is uniformly set to 1.0 based on the assumption that the potential effectiveness of all knowledge nodes is unknown and equal before practical verification; this assignment method ensures the neutrality of the initial network weights.

[0088] Based on the established compatibility edges, an adjacency matrix is ​​constructed to represent the structure of the global knowledge value network. The rows and columns of the matrix correspond to all knowledge nodes. If a compatibility edge exists between two knowledge nodes, it is marked as 1 at the corresponding position in the matrix; otherwise, it is marked as 0. The adjacency matrix, together with the set of knowledge nodes and the set of performance weights, forms the global knowledge value network.

[0089] Example: Suppose three companies participate in the aggregation, resulting in three global knowledge vectors after decryption. The cosine similarity between vectors A and B is calculated to be 0.85, exceeding the compatibility association threshold of 0.8. Therefore, a compatibility relationship edge is established between knowledge nodes A and B. After similarity calculation, a global knowledge value network is formed, containing three knowledge nodes, one compatibility edge, and all effective weights initially set to 1.0. A 3x3 adjacency matrix is ​​used to record the connection relationships.

[0090] A superior approach is to use secure multi-party computation and homomorphic encryption for federated aggregation. Under the premise of absolutely protecting enterprise data privacy (compared to the traditional centralized data pool that requires sharing of original data), a global knowledge value network is constructed to achieve cross-enterprise knowledge security integration and value measurement.

[0091] S3: In response to the knowledge query request issued by the target enterprise, retrieve matching candidate knowledge vectors from the global knowledge value network and input the candidate knowledge vectors into the target enterprise's local digital twin.

[0092] It should be noted that a knowledge query request refers to a formatted request initiated by the target enterprise for production optimization goals, which includes query conditions and constraint parameters; retrieval matching refers to calculating the similarity between the features of the knowledge query request and the features of knowledge nodes in the global knowledge value network, and selecting candidate knowledge vectors; the selection process involves calculating the cosine similarity between the features of the knowledge query request and the features of knowledge nodes in the global knowledge value network, sorting them from high to low similarity, and selecting the node before the position with the largest similarity difference as the candidate knowledge vector.

[0093] The local digital twin pre-training process is as follows: The target enterprise trains its local digital twin using its own historical production data. This historical production data includes heterogeneous data from multiple sources collected in step S1, namely real-time time-series data, production work order data, quality inspection data, and temperature and humidity data. The local digital twin is constructed using a Long Short-Term Memory (LSTM) network architecture. The LTM network takes a sequence of process parameter data as input and outputs a sequence of predicted performance indicator data. The training process uses mean squared error as the loss function and optimizes the LTM network parameters using a gradient descent algorithm to make the predicted output as close as possible to the actual historical performance indicator data. The trained local digital twin is capable of simulating the target enterprise's specific production processes.

[0094] S3.1: The target company generates a knowledge query request.

[0095] Specifically, the knowledge query request is a structured data object containing a production optimization target field and a constraint parameter field. The production optimization target field specifies the name of the performance indicator to be optimized (such as "product qualification rate") and the optimization direction (such as "maximize"). The constraint parameter field specifies the adjustable range of process parameters (such as "the upper limit of the heating furnace temperature adjustment is 180 degrees Celsius, and the lower limit is 120 degrees Celsius") and other business constraints (such as "the total energy consumption increase shall not exceed 5%)".

[0096] The knowledge query request is converted into a feature vector for similarity calculation (the conversion process uses one-hot encoding). The feature vector is composed of the following parts: a vector after one-hot encoding of the performance indicator names in the production optimization target field; a scalar representation of the optimization direction (maximization is encoded as one, minimization is encoded as zero); and a standardized numerical sequence of the upper and lower limits of the allowable adjustment range of each process parameter in the constraint parameter field.

[0097] S3.2: Calculate the feature similarity with knowledge nodes and select candidate knowledge vectors to be transmitted to the local digital twin.

[0098] Specifically, feature representations (i.e., global knowledge vectors) of all knowledge nodes are obtained from the global knowledge value network. The cosine similarity between the feature vector of the knowledge query request and the feature vector of each knowledge node is calculated.

[0099] All knowledge nodes are sorted in descending order based on the calculated cosine similarity scores. Global knowledge vectors corresponding to knowledge nodes whose cosine similarity ranking is higher than a preset similarity ranking threshold are selected as candidate knowledge vectors. The selection process ensures that the candidate knowledge vectors are highly relevant to the knowledge query request in terms of semantics and constraints.

[0100] It should be noted that the preset similarity ranking threshold is automatically determined by identifying the position with the largest difference between adjacent rankings in the cosine similarity ranking sequence between knowledge nodes and query requests. A suggested value range is 0.15-0.25. Specifically, after sorting nodes by similarity from high to low, the similarity difference between adjacent nodes is calculated (e.g., the difference between the 1st and 2nd ranked nodes, the difference between the 2nd and 3rd ranked nodes, etc.). The position with the largest difference indicates a sharp drop in similarity at that point. Nodes before the position with the largest difference can be considered as a candidate set highly relevant to the query, while nodes after the position with the largest difference have significantly reduced relevance. This method adaptively determines the size of the candidate set based on the results of each query.

[0101] The selected candidate knowledge vectors are transferred to the target company's local environment. Each candidate knowledge vector is sequentially input into the target company's local digital twin. The local digital twin provides a standardized data interface, receiving candidate knowledge vector inputs with dimensions consistent with the candidate knowledge vectors. The candidate knowledge vectors will be used to derive specific process parameter adjustment strategies in subsequent steps.

[0102] It should be noted that the target company's local environment refers to the company's own private computing environment, including infrastructure such as the company's intranet servers, production management system and data acquisition system; this infrastructure is physically or logically isolated from the company's external network to ensure that data and applications are always within the company's self-managed security boundary, while having the computing resources and data interfaces required to run the digital twin.

[0103] S4: Inject the process parameter adjustment strategy corresponding to the candidate knowledge vector into the local digital twin, run the counterfactual reasoning engine to simulate the production state change after the execution of the process parameter adjustment strategy, and output the expected efficiency gain value.

[0104] It should be noted that for each input candidate knowledge vector, a pre-trained fully connected neural network decoder is used to map the candidate knowledge vector to a specific process parameter adjustment strategy. The fully connected neural network decoder has a three-layer structure: the input layer has a dimension of 128 (consistent with the dimension of the candidate knowledge vector), the middle hidden layer uses 256 neurons to perform feature transformation and dimension expansion, and the output layer dimension is determined according to the content of the process parameter adjustment strategy.

[0105] The process parameter adjustment strategy is specifically decoded into four components: the temperature setpoint adjustment range (represented as minimum and maximum values), the pressure control target value change sequence (represented as a series of target values ​​arranged in chronological order), the equipment operating rate adjustment gradient (represented as the rate of change per minute), and the material ratio adjustment proportion (represented as the percentage change of each component). The fully connected neural network decoder uses a linear activation function, and the weight parameters of the fully connected neural network decoder are learned through training on historical data. The learning objective is to minimize the mean squared error between the decoded strategy and the truly effective strategy.

[0106] S4.1: Receive the process parameter adjustment strategy corresponding to the candidate knowledge vector, and execute the intervention operation of the process parameter adjustment strategy in the local digital twin.

[0107] Specifically, the counterfactual reasoning engine receives the decoded process parameter adjustment strategy. The engine operates the local digital twin, modifying its input parameters to implement the strategy. Specifically, it constrains the temperature setpoint in the local digital twin's simulation input parameters within the decoded adjustment range; replaces the local digital twin's pressure control target value sequence with the decoded change sequence; sets the rate of change of the local digital twin's equipment operating speed to the decoded adjustment gradient; and adjusts the local digital twin's material proportioning parameters according to the decoded proportions, simulating the application of the process knowledge represented by the candidate knowledge vector in the actual production environment. The simulation input parameters refer to the process parameter settings in the local digital twin used to simulate the actual production environment, including temperature setpoints, pressure control target values, equipment operating speed, and material proportions; these parameters are modified to simulate the execution effect of the process adjustment strategy.

[0108] S4.2: Run the local digital twin simulation model to simulate changes in production status and calculate the changes in performance indicators before and after intervention as the expected performance gain value.

[0109] Specifically, after the process parameter adjustment strategy is implemented, the simulation operation of the local digital twin is initiated. Based on the long short-term memory network model trained in step S3, the local digital twin uses the aforementioned modified process parameters as input to perform forward calculations, simulating changes in production status over a future period. The simulation outputs a predicted performance indicator data sequence. This performance indicator data sequence reflects the expected production results after applying the process parameter adjustment strategy.

[0110] The counterfactual reasoning engine calculates the changes in performance indicators before and after the intervention of process parameter adjustment strategies.

[0111] Specifically, the baseline performance index values ​​output by the local digital twin simulation are recorded when no process parameter adjustment strategy is applied (i.e., the current process parameters remain unchanged); the new performance index values ​​output by the simulation after applying the process parameter adjustment strategy are obtained. The expected performance gain value is obtained by calculating the absolute difference between the new performance index value and the baseline performance index value. The absolute difference quantifies the expected performance improvement after applying the process knowledge represented by the candidate knowledge vector.

[0112] A superior approach is to use strategy simulation verification based on digital twins and counterfactual reasoning, which replaces the high-cost and high-risk on-site trial and error debugging, significantly reducing the experimental cycle and resource consumption.

[0113] S5: Sort candidate knowledge vectors in descending order based on expected performance gain values, and collect actual performance data after the target enterprise actually applies the candidate knowledge vectors.

[0114] S5.1: Use the expected performance gain value as the sorting key to sort the candidate knowledge vectors in descending order to generate a recommendation sequence.

[0115] Specifically, the expected performance gain value corresponding to each candidate knowledge vector is obtained from the output of the counterfactual reasoning engine. The expected performance gain value is a numerical data point representing the expected change in performance indicators after applying the process parameter adjustment strategy corresponding to that candidate knowledge vector.

[0116] The quicksort algorithm is used to sort the candidate knowledge vector set. The sorting key is the expected performance gain value corresponding to each candidate knowledge vector. The quicksort algorithm recursively selects a pivot element and divides the set into subsets with expected performance gains greater than and less than the pivot element, thus completing the descending order of the entire set. After sorting, the candidate knowledge vector with the largest expected performance gain value is located at the beginning of the sequence.

[0117] The sorted candidate knowledge vectors, their corresponding process parameter adjustment strategies, and expected performance gains are organized into an ordered list to form a recommendation sequence. The recommendation sequence clarifies the priority order for applying different candidate knowledge vectors.

[0118] S5.2: Based on the priority of the recommended sequence, guide the target enterprise to apply the process parameter adjustment strategy corresponding to the candidate knowledge vector in sequence, and simultaneously record the change in performance indicators as real performance data.

[0119] Recommended sequences are presented to the target company's production management personnel through an interface. Following the priority order of the recommended sequences, the system prompts the application of the process parameter adjustment strategy corresponding to the top-ranked candidate knowledge vector. The process parameter adjustment strategy is presented in the form of clear parameter setting guidance, including the temperature setpoint adjustment range, the pressure control target value change sequence, the equipment operating rate adjustment gradient, and the specific values ​​for material ratio adjustment proportions. Based on this guidance, production management personnel can manually adjust the corresponding process parameter settings in the production management system or authorize the automatic control system to perform parameter adjustments.

[0120] After the process parameter adjustment strategy is applied, the production data acquisition process is initiated. Through interface integration with the enterprise's production management system and data acquisition and monitoring control system, multi-source heterogeneous data is continuously collected, including real-time time-series data, production work order data, and quality inspection data. Data acquisition covers a complete production cycle after the strategy is applied, such as 24 hours.

[0121] From the collected multi-source heterogeneous data, performance indicator data related to the application strategy objectives are extracted. The change in this performance indicator before and after the strategy application is calculated as the actual effectiveness data. The calculation method for the actual effectiveness data is consistent with the calculation method for the expected effectiveness gain value in step S4, that is, calculating the average change of the same performance indicator over the same period of time.

[0122] After completing the application and data recording of the first strategy, the system prompts the user to apply the process parameter adjustment strategy corresponding to the next candidate knowledge vector in the recommended sequence order, and repeats the aforementioned data collection and real performance data calculation process until all strategies in the recommended sequence have been applied and recorded, or the preset application limit has been reached.

[0123] It should be noted that the preset upper limit for the number of applications is determined by comprehensively considering the target company's actual production cycle length, the stability requirements of process adjustments, and the feasibility of data collection. For example, the specific value is usually set to 3-5 strategy application batches. This upper limit ensures that sufficient validation data is obtained while avoiding excessive process adjustments that may disrupt normal production.

[0124] S6: Compare and analyze the actual performance data with the expected performance gain value to generate a knowledge verification deviation signal; dynamically adjust the performance weight of the corresponding knowledge node in the global knowledge value network based on the knowledge verification deviation signal; the knowledge verification deviation signal includes two types: negative deviation signal and positive deviation signal.

[0125] S6.1: Obtain the actual performance data of the target enterprise after applying the process parameter adjustment strategy, and calculate the absolute error between the actual performance data and the expected performance gain value. The absolute error value is obtained by calculating the arithmetic difference between the actual performance data and the expected performance gain value.

[0126] The relative error rate is generated based on the ratio between the absolute error value and the expected performance gain value; the relative error rate is obtained by dividing the absolute error value by the expected performance gain value.

[0127] S6.2: If the relative error rate exceeds the preset allowable deviation range, a negative deviation signal is generated; otherwise, a positive deviation signal is generated.

[0128] Specifically, an allowable deviation range is set; this allowable deviation range is set to a fixed value. The fixed value is determined based on the relative error rate distribution between the expected performance gain value and the simulated real performance data on the local digital twin's validation set. A specific percentile (e.g., the 95th percentile) of the relative error rate distribution is taken as the fixed value. The simulated real performance data refers to the change in the predicted performance index output by the digital twin after simulating the process parameter adjustment strategy based on historical production data during the local digital twin validation phase. This change is used to compare and verify with the expected performance gain value.

[0129] It should be noted that during the digital twin development phase, historical data is used to divide the dataset into training and validation sets. The relative error of the digital twin's predictions is tested on the validation set, and the statistical distribution is analyzed.

[0130] The relative error rate is compared with the preset allowable deviation range. The judgment logic in the comparison process is as follows: if the relative error rate > the preset allowable deviation range, a knowledge verification deviation signal with the type labeled "negative deviation signal" is generated; the content of the negative deviation signal contains the calculated relative error rate value; otherwise (relative error rate ≤ preset allowable deviation range), a knowledge verification deviation signal with the type labeled "positive deviation signal" is generated; the content of the positive deviation signal also contains the calculated relative error rate value.

[0131] S6.3: Analyze the relative error rate in the knowledge verification deviation signal and identify the signal type; trigger the first coefficient adjustment of the performance weight based on the negative deviation signal; trigger the second coefficient adjustment of the performance weight based on the positive deviation signal. Update the adjusted performance weight to the corresponding knowledge node in the global knowledge value network.

[0132] Specifically, the knowledge verification deviation signal is analyzed, its relative error rate value is extracted, and the signal type ("negative deviation signal" or "positive deviation signal") is identified. Based on the different signal types, different multiplication coefficients are used to adjust the current performance weight of the corresponding knowledge node within the global knowledge value network.

[0133] If the signal type is a "negative deviation signal", the first coefficient is multiplied by the current performance weight to adjust the weight downward. If the signal type is a "positive deviation signal", the second coefficient is multiplied by the current performance weight to adjust the weight upward.

[0134] It should be noted that the determination of the first and second coefficients is based on the credit allocation concept in reinforcement learning, and is obtained through dynamic calibration by a feedback loop of historical weight adjustment effects.

[0135] Specifically, the process involves: initializing the first coefficient to a value less than 1 and the second coefficient to a value greater than 1 to achieve the basic functions of weight decay and enhancement, respectively; after each coefficient adjustment, monitoring the frequency of subsequent retrieval of the knowledge node and the actual effectiveness of its contribution; if the retrieval frequency of the knowledge node significantly decreases after the weight is lowered or the actual effectiveness of its subsequent contribution remains excellent, then the gap between the first coefficient and 1 is reduced by a fixed step size to mitigate excessive punishment; if the retrieval frequency of the knowledge node does not significantly decrease after the weight is lowered or the actual effectiveness of its subsequent contribution does not remain excellent, then the gap between the first coefficient and 1 is increased by a fixed step size to strengthen the punishment; if the retrieval frequency of the knowledge node does not increase after the weight is higher or the actual effectiveness of its subsequent contribution is mediocre, then the gap between the second coefficient and 1 is reduced by a fixed step size to avoid excessive reward; if the retrieval frequency of the knowledge node increases after the weight is higher or the actual effectiveness of its subsequent contribution remains excellent, then the gap between the second coefficient and 1 is increased by a fixed step size to enhance the reward; through this dynamic calibration mechanism, the first and second coefficients can adaptively converge to reasonable values ​​that accurately reflect the actual reliability of the knowledge node.

[0136] Furthermore, the adjusted performance weight values ​​are written into and overwrite the original performance weight values ​​of the knowledge nodes corresponding to the currently validated candidate knowledge vectors in the global knowledge value network. This completes the dynamic update of the performance weights of knowledge nodes in the global knowledge value network.

[0137] S7: Based on the adjusted performance weights, calculate the contribution of each enterprise in the global knowledge value network using the Shapley value algorithm, generate an incentive allocation scheme, and trigger the execution of smart contracts.

[0138] S7.1: Traverse all knowledge nodes in the global knowledge value network after adjustment of their performance weights, and aggregate and calculate the sum of performance weights of the knowledge nodes belonging to each enterprise based on the affiliation of the knowledge nodes.

[0139] Specifically, the system accesses the global knowledge value network, retrieves all knowledge nodes and their updated performance weights, and records the enterprise identifier of each knowledge node at the time of its generation. Based on the enterprise identifier, the knowledge nodes are assigned to their corresponding enterprises.

[0140] The effectiveness weight values ​​of all knowledge nodes belonging to the corresponding enterprise are summed and calculated.

[0141] The summation formula is expressed as:

[0142] ;

[0143] in, Indicates enterprise The sum of performance weights, Indicates ownership by the enterprise The set of knowledge nodes, Representing knowledge nodes Performance weights, total performance weights It is referred to as the company's own contribution value.

[0144] S7.2: The sum of performance weights is used as the enterprise's own contribution value. The contribution ratio of each enterprise is allocated through the Shapley value algorithm. The contribution ratio is converted into token quantity and written into the blockchain smart contract, triggering the smart contract to automatically execute the token transfer operation.

[0145] Specifically, all participating companies are considered as a general alliance. Each company's own contribution value Defined as the value created when a company forms its own sub-alliance. For any sub-alliance The total value of the sub-alliance By calculating sub-alliance The sum of the contributions of all member companies is obtained, expressed as:

[0146] ;

[0147] in, Represents the characteristic function. Indicates a pair alliance Iterate through each company in the list. It is an index for enterprises;

[0148] It should be noted that in the Shapley value algorithm, the feature function... A sub-consortium of firms, S (i.e., a subset of the set of firms), is defined. The total value or contribution that can be generated is specifically realized as the contribution value of each member within the alliance. The sum of these values ​​is used to quantitatively assess the synergistic value of different combinations of businesses.

[0149] Furthermore, to ensure a fair distribution of the total value of the alliance, for each enterprise... Calculate the corresponding Shapley value This represents the enterprise's contribution to the overall knowledge value network.

[0150] The formula for calculating the Shapley value is:

[0151] ;

[0152] In the formula, Indicates enterprise The Shapley value, Expressing support for the General League The company is not included. All possible subsets Perform a traversal and summation. Sub-alliance The base number, To represent factorial operation, Indicates the General League The base number, that is, the total number of participating enterprises. Indicates enterprise Join the sub-alliance The total value of the newly formed sub-alliances;

[0153] It should be noted that, Indicates enterprise For sub-alliance The marginal contribution of a company. The addition of the sub-alliance Additional value increment; It is a weighting factor used to calculate sub-coalitions. The probability of occurrence, the numerator is the sub-alliance The number of permutations, with the denominator being the total number of allies. The number of permutations ensures that the Shapley value satisfies the fairness axiom.

[0154] Contribution ratio of each enterprise It is calculated by taking a company's Shapley value as a percentage of the total Shapley values ​​of all companies. The formula is as follows:

[0155] ;

[0156] In the formula, Expressing support for the General League All enterprises Perform a traversal and summation. Represents all enterprises The Shapley value, This represents the sum of the Shapley values ​​of all enterprises, i.e., the total contribution.

[0157] The total number of tokens allocated in this incentive distribution is set to a fixed value. (For example, 1000), each enterprise Number of tokens to be allocated The calculation formula is: ;

[0158] Each enterprise Number of tokens to be allocated The corresponding enterprise blockchain address information is written as a parameter into a smart contract pre-deployed on the blockchain;

[0159] The corresponding allocation function in the smart contract is invoked, triggering the automatic execution of the smart contract. Based on the written allocation scheme, the smart contract automatically transfers a specified number of tokens from the liquidity pool account to the blockchain addresses of each enterprise, completing the incentive distribution.

[0160] In contrast to traditional incentive methods that simply allocate incentives based on data volume or transaction amount, this invention uses the Shapley value algorithm to accurately quantify the marginal contribution of each enterprise in knowledge integration, ensuring the fairness and verifiability of the allocation scheme. This effectively stimulates enterprises' willingness to continuously participate in sharing and promotes a virtuous cycle of the industry's knowledge ecosystem.

[0161] This embodiment also provides a computer device applicable to the digital operation management method based on the industrial internet, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the digital operation management method based on the industrial internet proposed in the above embodiment.

[0162] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0163] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the digital operation management method based on the industrial internet proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0164] In summary, this invention calculates the contribution of each enterprise to the global knowledge value network using the Shapley value algorithm. It accurately quantifies the marginal value contribution of each knowledge source in cross-enterprise collaboration based on the dynamic effectiveness weights of knowledge nodes. The Shapley value algorithm calculates the change in the total value of the alliance before and after an enterprise joins by traversing all possible combinations of sub-alliances, thereby establishing a quantitative evaluation system based on multi-dimensional synergistic effects. This allocation mechanism, based on cooperative game theory, ensures that the incentive allocation strictly corresponds to the actual effectiveness gain generated by the knowledge. Furthermore, it achieves automated token allocation through smart contracts, forming a verifiable and traceable value return system, effectively stimulating the intrinsic motivation of enterprises to continuously participate in high-quality knowledge sharing.

[0165] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A digital operation management method based on the industrial internet, characterized in that: include, Collect multi-source heterogeneous data, extract the causal relationship between internal process parameters and performance indicators of each enterprise through causal discovery algorithm, and encode the causal relationship into local knowledge vector using deep learning model; By using a secure multi-party computation protocol, the enterprise's local knowledge vectors are aggregated in a federated manner to construct a global knowledge value network that includes knowledge nodes, performance weights, and compatibility relationships. The specific steps are as follows. Decrypt the local knowledge vector to obtain the global knowledge vector set, and calculate the cosine similarity between any two vectors in the global knowledge vector set. Knowledge vectors with cosine similarity exceeding a preset compatibility association threshold are used to establish compatibility relationship edges. Performance weights are initialized for each knowledge node and an adjacency matrix is ​​constructed to form a global knowledge value network. In response to a knowledge query request from a target enterprise, the system retrieves matching candidate knowledge vectors from the global knowledge value network and inputs the candidate knowledge vectors into the target enterprise's local digital twin. Inject the process parameter adjustment strategy corresponding to the candidate knowledge vector into the local digital twin, run the counterfactual reasoning engine to simulate the production status change after the execution of the process parameter adjustment strategy, and output the expected efficiency gain value. Candidate knowledge vectors are sorted in descending order based on expected performance gain values, and real performance data after the target enterprise actually applies the candidate knowledge vectors is collected. The actual performance data is compared and analyzed with the expected performance gain value to generate a knowledge verification deviation signal; the performance weight of the corresponding knowledge node in the global knowledge value network is dynamically adjusted based on the knowledge verification deviation signal. Based on the adjusted performance weights, the contribution of each enterprise to the global knowledge value network is calculated using the Shapley value algorithm. An incentive allocation scheme is then generated, and the smart contract is executed. The specific steps are as follows: Traverse all knowledge nodes in the global knowledge value network after adjustment of their performance weights, and aggregate and calculate the sum of performance weights of the knowledge nodes belonging to each enterprise based on the affiliation of the knowledge nodes. The sum of performance weights is used as the contribution value of each enterprise. The contribution ratio of each enterprise is allocated through the Shapley value algorithm. The contribution ratio is converted into token quantity and written into the blockchain smart contract, which triggers the smart contract to automatically execute the token transfer operation.

2. The digital operation management method based on the industrial internet as described in claim 1, characterized in that: The multi-source heterogeneous data includes real-time time-series data, production work order data, quality inspection data, and temperature and humidity data; The specific steps for encoding causal relationships into local knowledge vectors using a deep learning model are as follows. Multi-source heterogeneous data is input into the causal discovery algorithm. Based on the directed acyclic graph output by the causal discovery algorithm, the causal strength value between process parameter nodes and performance indicator nodes is extracted. The features of each node and the corresponding causal strength value are input into the graph neural network encoder. The information of neighboring nodes is aggregated through the multi-layer message passing mechanism of the graph neural network encoder to generate a global graph representation vector as a local knowledge vector.

3. The digital operation management method based on the industrial internet as described in claim 2, characterized in that: The secure multi-party computation protocol refers to a scheme that uses homomorphic encryption algorithm to encrypt and transmit local knowledge vectors; the federated secure aggregation refers to a weighted average operation on the encrypted local knowledge vectors using a secure aggregation algorithm.

4. The digital operation management method based on the industrial internet as described in claim 3, characterized in that: The knowledge query request refers to a formatted request initiated by the target enterprise for production optimization goals, which includes query conditions and constraint parameters; the retrieval matching refers to calculating the similarity between the features of the knowledge query request and the features of knowledge nodes in the global knowledge value network, and selecting candidate knowledge vectors.

5. The digital operation management method based on the industrial internet as described in claim 4, characterized in that: The process parameter adjustment strategy includes the temperature setpoint adjustment range, the pressure control target value change sequence, the equipment operating rate adjustment gradient, and the material proportion adjustment ratio. The counterfactual reasoning engine has the following architecture. Receive the process parameter adjustment strategy corresponding to the candidate knowledge vector, and execute the intervention operation of the process parameter adjustment strategy in the local digital twin; The simulation model of the local digital twin is run to simulate changes in production status, and the changes in performance indicators before and after the intervention are calculated as the expected performance gain value.

6. The digital operation management method based on the industrial internet as described in claim 5, characterized in that: The knowledge verification deviation signal includes two types: negative deviation signal and positive deviation signal generation. The specific steps for generating the knowledge verification bias signal are as follows: Obtain the actual performance data of the target company after applying the process parameter adjustment strategy, and calculate the absolute error between the actual performance data and the expected performance gain value; The relative error rate is generated based on the ratio between the absolute error value and the expected performance gain value. If the relative error rate exceeds the preset allowable deviation range, a negative deviation signal is generated; otherwise, a positive deviation signal is generated.

7. The digital operation management method based on the industrial internet as described in claim 6, characterized in that: The specific steps for dynamically adjusting the effectiveness weights of corresponding knowledge nodes in the global knowledge value network based on the knowledge verification deviation signal are as follows. Analyze the knowledge to verify the relative error rate in the deviation signal and identify the signal type; The first coefficient of the performance weight is adjusted based on the negative deviation signal. The second coefficient of the performance weight is adjusted based on the positive deviation signal. The adjusted performance weights are then updated to the corresponding knowledge nodes in the global knowledge value network.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the digital operation management method based on the industrial internet as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the digital operation management method based on the industrial internet as described in any one of claims 1 to 7.

Images