Production operation benchmarking system and method based on multi-source data analysis

Abstract

The invention discloses a production operation benchmarking system and method based on multi-source data analysis, and belongs to the technical field of production operation benchmarking, and the system comprises a multi-source confrontation fusion module, a production situation awareness map module, a self-adaptive diagnosis analysis module, a cross-domain collaborative deduction strategy module, a hidden variable benchmarking analysis module and a cross-scene migration module. According to the method, enterprise resources are quantized into a matrix, cross-domain resource allocation is optimized in combination with a collaborative gain function, collaborative gains are measured through a nuclear norm and a Frobenius norm, multi-target weighting is combined, priorities of different fields are balanced, the global optimality of a collaborative strategy is improved, the enterprise resource state is synchronized to a digital twinborn model, and the collaborative resource allocation efficiency is improved. A strategy execution process is simulated, burst interference is injected, strategy stability is verified, a cross-domain cooperation efficiency comprehensive value is calculated through a four-flow state, a single-domain independent optimization mean value is compared, and cooperation gain is quantified.

Description

Technical Field

[0001] This invention belongs to the field of production and operation benchmarking technology, specifically referring to a production and operation benchmarking system and method based on multi-source data analysis. Background Technology

[0002] With the development of industry and intelligent manufacturing, enterprise production and operation data are characterized by multi-source heterogeneity, high-dimensional dynamics, and noise interference. Existing production and operation analysis systems generally have many problems.

[0003] However, existing benchmarking methods for production and operations still have certain shortcomings. Existing methods do not fully quantify enterprise resources and model them as constraints in resource allocation, making it difficult for cross-domain collaboration strategies to balance the priorities of multiple domains with overall efficiency. The lack of a collaborative gain function to measure cross-domain linkages leads to a lack of scientific basis for resource allocation strategies. Furthermore, the benchmarking system does not isolate the systematic impact of external variables on business indicators, resulting in distorted capability assessment results that fail to reflect the true level of the enterprise. Therefore, a production and operations benchmarking system and method based on multi-source data analysis is proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a production operation benchmarking system and method based on multi-source data analysis to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a production operation benchmarking system and method based on multi-source data analysis, including a multi-source adversarial fusion module, a production situation awareness map module, an adaptive diagnostic analysis module, a cross-domain collaborative inference strategy module, a latent variable benchmarking analysis module, and a cross-scenario migration module;

[0006] The multi-source adversarial fusion module dynamically allocates trusted weights through an adversarial feature distillation network based on the input from external data sources, performs contaminated data filtering and heterogeneous data fusion, and outputs an anti-interference fusion data stream.

[0007] The production situation awareness map module constructs a dynamic production situation awareness map by simulating entity behavior coupling through an embodied cognition engine based on the anti-interference fusion data stream.

[0008] The adaptive diagnostic analysis module generates an adaptive diagnostic strategy by performing counterfactual reasoning and multi-objective trade-offs based on the causal chain of the dynamic production situation awareness map and real-time business needs.

[0009] The cross-domain collaborative inference strategy module generates cross-domain collaborative optimization strategies based on diagnostic strategies and enterprise resource constraints, and verifies the execution effect through a digital twin model;

[0010] The latent variable benchmarking analysis module, based on the collaborative optimization strategy, strips out external variables through counterfactual causal comparison, performs pure capability space mapping, and outputs a biased benchmarking score matrix.

[0011] The cross-scene migration module deconstructs knowledge units through a meta-transfer capsule network based on the benchmarking scoring matrix, performs capsule recombination and scene constraint injection, outputs a set of transferable capsule rules, and feeds them back to the multi-source adversarial fusion module.

[0012] Preferably, the multi-source adversarial fusion module is wirelessly connected to the production situation awareness map module, the production situation awareness map module is wirelessly connected to the adaptive diagnostic analysis module, the adaptive diagnostic analysis module is wirelessly connected to the cross-domain collaborative inference strategy module, the cross-domain collaborative inference strategy module is wirelessly connected to the latent variable benchmarking analysis module, the latent variable benchmarking analysis module is wirelessly connected to the cross-scene migration module, and the cross-scene migration module is wirelessly connected to the multi-source adversarial fusion module.

[0013] Preferably, the multi-source adversarial fusion module acquires raw data from heterogeneous systems, converts unstructured data into a unified vector representation, aligns asynchronous data sources with timestamps, eliminates data time-series deviations, and quickly filters data that is obviously abnormal or has a high missing rate.

[0014] Specifically, an adversarial feature distillation network is constructed, including a teacher model and a student model. The robustness of the student model is enhanced through adversarial training. An initial credibility score is calculated based on the historical accuracy, update frequency, and source authority of the data source. During adversarial training, the weights are dynamically updated through gradient feedback. Adversarial samples are identified by the prediction differences between the teacher and student models. For suspected contaminated data, interpolation or deletion is performed based on the statistical characteristics of neighboring data points.

[0015] Different modal data are mapped to a unified semantic space, asynchronous data are weighted and averaged according to time windows, the fusion strategy is dynamically adjusted according to the performance of downstream tasks, and an anti-interference fused data stream is output.

[0016] Preferably, the production situation awareness map module acquires anti-interference fusion data streams, distinguishes between static and dynamic data, and associates sensor data with text descriptions through the multimodal perception capabilities of the embodied cognition engine to form semantic tags;

[0017] Specifically, key entities are extracted from the data stream, combined with the hierarchical relationships defined by ontology, and the physical interactions between entities are simulated through an embodied cognition engine. Implicit relationships are then deduced by combining historical behavior patterns.

[0018] Based on real-time data and preset rules, behavioral sequences of entities are generated. The feasibility of these behaviors is verified using a simulation engine, the impact of entity behaviors on the overall situation is quantified, and a production situation awareness map is constructed.

[0019] Nodes: Entity, Status, Timestamp;

[0020] Edges: Relationships between entities, triggering conditions for behaviors;

[0021] New nodes or edges are added based on the real-time data stream, and when key events occur, the potential impact of related entities is re-inferred through the embodied cognition engine.

[0022] Preferably, the adaptive diagnostic analysis module includes the following steps for causal chain extraction and modeling: obtaining entities and their temporal causal relationships from the dynamic production situation awareness map module, and receiving real-time business requirements from external input.

[0023] Suppose there exist i causal chains C1, C2, ..., C i The weight w of each causal chain i Based on real-time business requirements R i The dynamic adjustment and comprehensive diagnostic strategy D are implemented using the following formula:

[0024]

[0025] It should be understood that w i Based on Bayesian update rules, and by adjusting weights using historical data and real-time feedback, the weighted causal chain S is used as the initial input to the diagnostic strategy, representing the priority ranking of key issues.

[0026] Preferably, the adaptive diagnostic analysis module includes a counterfactual reasoning step comprising: assuming, based on a comprehensive diagnostic strategy, that for variable X... j Intervention is carried out to calculate the effect on target Y. j The effect is achieved by the following formula:

[0027]

[0028] In the formula, Y cr Z represents the target value in a counterfactual hypothetical scenario, x′ represents the variable value after intervention, and Z represents the target value in a counterfactual hypothetical scenario. j 'm' represents the external variable (such as supply chain stability), and 'm' represents the dimension of the virtual scenario.

[0029] Preferably, the adaptive diagnostic analysis module includes a multi-objective trade-off step comprising: adjusting Y... cr Normalization is performed to eliminate dimensional differences, and then the normalized Y is combined. cr And a comprehensive diagnostic strategy, calculating an adaptive diagnostic strategy, the formula is as follows:

[0030]

[0031] In the formula, U represents the adaptive diagnostic strategy, and α k Y represents the weight of the k-th objective. cr,k Y represents the target value of the k-th target in a counterfactual hypothetical scenario. min,k Y represents the minimum possible value of the k-th objective. max,k Let S represent the maximum possible value of the k-th objective, p represent the number of objectives, and log(S) represent the nonlinear effect of the balanced comprehensive diagnostic strategy.

[0032] Specifically, based on the counterfactual reasoning results and the adaptive diagnostic strategy, multiple candidate strategies are generated. The generated strategies are then input into the digital twin model to simulate the execution effect. The strategy parameters are adjusted based on the simulation results, and the effectiveness of the strategies is verified through the latent variable benchmarking analysis module.

[0033] Preferably, the cross-domain collaborative inference strategy module includes the following steps for generating cross-domain collaborative optimization strategies: receiving an adaptive diagnostic strategy U, quantifying enterprise resources and modeling them as resource constraints, and decomposing the adaptive diagnostic strategy into executable actions for each domain based on the objectives and resource constraints of the adaptive diagnostic strategy. The formula for generating the cross-domain collaborative optimization strategy is as follows:

[0034]

[0035] In the formula, S represents the cross-domain collaboration strategy, which is a decision combination of allocating resources and actions across multiple domains. (t+1) argm represents the optimized coordination strategy. S ax represents the maximization operation, which seeks the strategy S that maximizes the objective function;

[0036] Specifically, This represents a multi-objective weighted sum, where the weights are w. k Measuring different objectives f k priority, w k f represents the target weight. k (S) represents the k-th objective function, λ represents the cooperative gain adjustment coefficient, and ΔΠ(S,E,U) represents the cooperative gain function;

[0037] Subkect represents the constraint condition, E represents the resource allocation matrix, which shows the resource usage of each domain, A represents the resource demand matrix, which shows the resource demand of each adaptive diagnostic strategy S, and C represents the resource constraint condition.

[0038] E·A≤C represents a resource constraint, U adap Let ΔΠ(S,E,U) represent the feasible space of the adaptive diagnostic strategy.

[0039] It should be understood that the combined gain of the two domains is calculated by weighting U, and T ::ij (S,E) represents a flow coupling tensor slice, describing the cross-domain linkage relationship under the cross-domain cooperation strategy and resource allocation E. * Denotes the nuclear norm, |||| F Let P, L, F, and I represent the Frobenius norm, and let P, L, F, and I represent the four flows (personnel flow, logistics flow, capital flow, and information flow). This indicates the overall efficiency of cross-domain collaboration. This represents the mean of the efficiency of independent optimization in a single domain.

[0040] Specifically, the validated digital twin model is invoked, the real-time enterprise resource status is synchronized to the digital twin model, the generated collaborative strategy is injected into the digital twin model, its execution process is simulated, key indicators are fed back in real time through the model, the simulation results are compared with the strategy objectives, sudden interference is simulated, and the stability of the strategy is verified.

[0041] Preferably, the latent variable benchmarking analysis module records the business performance under the current collaborative strategy, and calculates the virtual result by assuming that the external variables have not been disturbed through counterfactual reasoning. Comparison with actual results Counterfactual results Quantifying the impact of external variables:

[0042] Align the distribution of business indicators under different external conditions to a unified standard, eliminate the systematic influence of external variables on the indicators by mean alignment, and output pure capability indicators;

[0043] Latent variable modeling: Define latent variables that influence the benchmarking results, establish the relationship between observed variables and latent variables, map the debiased indicators to the pure capability space through kernel principal component analysis, update the latent variable model through online learning based on real-time business status, normalize the debiased indicators, dynamically adjust the scoring weights according to the collaborative optimization strategy, and construct the scoring matrix:

[0044] Line: Benchmark object;

[0045] Column: Scoring Dimensions;

[0046] Value: Weighted score

[0047] Output format:

[0048] Bias-free benchmarking scoring matrix:

[0049]

[0050] Where m represents the number of benchmark objects and n represents the number of scoring dimensions.

[0051] Preferably, the cross-scene migration module maps the score values ​​to the [0, 1] interval according to the bias-free benchmarking score matrix to ensure that the score dimensions are consistent across all scenes;

[0052] Transform the rating matrix into transferable knowledge units:

[0053] K-Means is used to cluster the rows of the rating matrix to identify similar strategy patterns;

[0054] Extract the feature vectors corresponding to the cluster centers;

[0055] Meta-transfer capsule network construction:

[0056] The standardized bias-free benchmarking score matrix is ​​encoded into a low-dimensional vector, local features are extracted by a convolutional neural network to generate primary capsules, and the output of the primary capsules is aggregated into higher-order capsules by a dynamic routing algorithm.

[0057] Capsule deconstruction and recombination:

[0058] Capsule deconstruction: Decode the output vector of the digital capsule to recover its corresponding policy features, and separate the magnitude and direction of the capsule vector through backpropagation;

[0059] Capsule recombination: The deconstructed knowledge units are linearly combined according to the scenario weights, and the coupling coefficient is updated iteratively to ensure that the recombinated capsules meet the constraints of the new scenario.

[0060] The constraints of the new scenario are injected into the recombined capsule network. The generator is a meta-transfer capsule network that generates candidate strategies. The discriminator is an adversarial network based on scenario constraints that evaluates the feasibility of the strategies. The discriminator provides multi-dimensional gradients to guide the generator in adjusting the capsule parameters.

[0061] Training is accelerated by using a negative sampling module, taking into account both strategies that users are interested in and those they are not, and the recombined capsule output is transformed into interpretable rules;

[0062] Based on the capsule attributes, a predefined rule template is matched, and the capsule vector is converted into a natural language rule through a sequence-to-sequence model. The rule is verified to meet the scenario constraints. The collaborative gain after the rule application is predicted through counterfactual reasoning and fed back to the multi-source adversarial fusion module.

[0063] Preferably, a production and operation benchmarking method based on multi-source data analysis includes the following steps:

[0064] S1. Based on the input from external data sources, the system dynamically allocates trusted weights through an adversarial feature distillation network to filter contaminated data and fuse heterogeneous data, outputting an anti-interference fused data stream.

[0065] S2. Based on the anti-interference fusion data stream, simulate entity behavior coupling through the embodied cognition engine to construct a dynamic production situation awareness map;

[0066] S3. Based on the causal chain of the dynamic production situation awareness map and real-time business needs, perform counterfactual reasoning and multi-objective trade-offs to generate adaptive diagnostic strategies.

[0067] S4. The cross-domain collaborative deduction strategy module generates a cross-domain collaborative optimization strategy based on the diagnostic strategy and enterprise resource constraints, and verifies the execution effect through a digital twin model.

[0068] S5. Based on the collaborative optimization strategy, external variables are removed by counterfactual causal comparison, pure capability space mapping is performed, and the biased benchmarking score matrix is ​​output.

[0069] S6. Based on the benchmarking scoring matrix, knowledge units are deconstructed through meta-transfer capsule networks, capsule recombination and scenario constraint injection are performed, and a set of transferable capsule rules is output and fed back to step S1.

[0070] Compared with the prior art, the beneficial effects of the present invention are:

[0071] 1. This invention quantifies enterprise resources into a matrix, optimizes cross-domain resource allocation by combining a synergy gain function, measures synergy gain by using the nuclear norm and Frobenius norm, and improves the global optimality of the synergy strategy by combining multi-objective weighting and balancing the priorities of different domains. It synchronizes the enterprise resource status to a digital twin model, simulates the strategy execution process and injects sudden disturbances to verify the stability of the strategy. It calculates the comprehensive value of cross-domain synergy efficiency through the four-flow state and compares it with the mean of single-domain independent optimization to quantify the synergy gain.

[0072] 2. This invention calculates virtual results by assuming no disturbance to external variables and compares them with actual results, quantifying the systematic impact of external variables on indicators and outputting a pure capability indicator. Through kernel principal component analysis, the unbiased indicators are mapped to the pure capability space, and the latent variable model is dynamically updated by online learning to adapt to changes in business status. The weights of the scoring dimensions are dynamically adjusted according to the collaborative optimization strategy to generate a row and column unbiased benchmarking scoring matrix, improving the objectivity and comparability of the benchmarking results. Through mean alignment, the distribution of business indicators under different external conditions is unified to a standard space, eliminating the interference of environmental differences on the benchmarking results.

[0073] 3. This invention identifies similar strategy patterns through K-Means clustering, extracts cluster center feature vectors, and constructs high-order capsules using a dynamic routing algorithm to achieve low-dimensional representation and flexible recombination of knowledge units. New scenario constraints are injected into the capsule network, and the feasibility of the strategy is optimized through generator-discriminator adversarial training to ensure that the transferred rules meet the scenario requirements. The capsule vectors are converted into natural language rules through a sequence-to-sequence model, and the collaborative gain is predicted by counterfactual reasoning to improve the interpretability of the transferred strategy. A negative sampling module is introduced to take into account the training of strategies that users are interested in and not interested in, thereby improving the model's generalization ability.

[0074] 4. This invention achieves adaptive prioritization of diagnostic strategies by adjusting the weights of causal chains based on Bayesian update rules, combined with historical data and real-time feedback. It simulates hypothetical scenarios, quantifies the impact of intervention measures on the target, breaks through the limitations of traditional experience-driven approaches, eliminates dimensional differences through normalization processing, and finds the optimal solution among conflicting targets by combining a nonlinear balance mechanism. Candidate strategies are injected into the digital twin model, and the execution effect is fed back in real time and parameters are corrected, significantly improving the reliability of strategy implementation. Attached Figure Description

[0075] Figure 1 This is a schematic diagram of the structure of a production operation benchmarking system based on multi-source data analysis according to the present invention;

[0076] Figure 2 This invention provides an operational flow diagram for a production and operation benchmarking system based on multi-source data analysis. Figure 1 ;

[0077] Figure 3 This invention provides an operational flow diagram for a production and operation benchmarking system based on multi-source data analysis. Figure 2 ;

[0078] Figure 4 This invention provides an operational flow diagram for a production and operation benchmarking system based on multi-source data analysis. Figure 3 ;

[0079] Figure 5 This is a schematic diagram of the structure of a production operation benchmarking method based on multi-source data analysis according to the present invention. Detailed Implementation

[0080] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0081] Example

[0082] Please see Figures 1-5 As shown, the present invention provides a technical solution including a multi-source adversarial fusion module, a production situation awareness map module, an adaptive diagnostic analysis module, a cross-domain collaborative inference strategy module, a latent variable benchmarking analysis module, and a cross-scenario migration module;

[0083] The multi-source adversarial fusion module dynamically allocates trusted weights through an adversarial feature distillation network based on the input from external data sources, performs contaminated data filtering and heterogeneous data fusion, and outputs an anti-interference fusion data stream.

[0084] The production situation awareness map module constructs a dynamic production situation awareness map by simulating entity behavior coupling through an embodied cognition engine based on the anti-interference fusion data stream.

[0085] The adaptive diagnostic analysis module performs counterfactual reasoning and multi-objective trade-offs based on the causal chain of the dynamic production situation awareness map and real-time business needs, and generates an adaptive diagnostic strategy.

[0086] The cross-domain collaborative inference strategy module generates cross-domain collaborative optimization strategies based on diagnostic strategies and enterprise resource constraints, and verifies the execution effect through a digital twin model;

[0087] The latent variable benchmarking analysis module, based on the collaborative optimization strategy, strips out external variables through counterfactual causal comparison, performs pure capability space mapping, and outputs a biased benchmarking score matrix.

[0088] The cross-scene migration module deconstructs knowledge units through a meta-transfer capsule network based on the benchmarking scoring matrix, performs capsule recombination and scene constraint injection, outputs a set of transferable capsule rules, and feeds them back to the multi-source adversarial fusion module.

[0089] Preferably, the multi-source adversarial fusion module is wirelessly connected to the production situation awareness map module, the production situation awareness map module is wirelessly connected to the adaptive diagnostic analysis module, the adaptive diagnostic analysis module is wirelessly connected to the cross-domain collaborative inference strategy module, the cross-domain collaborative inference strategy module is wirelessly connected to the latent variable benchmarking analysis module, the latent variable benchmarking analysis module is wirelessly connected to the cross-scene migration module, and the cross-scene migration module is wirelessly connected to the multi-source adversarial fusion module.

[0090] Preferably, the multi-source adversarial fusion module acquires raw data from heterogeneous systems, converts unstructured data into a unified vector representation, aligns asynchronous data sources with timestamps, eliminates data time-series deviations, and quickly filters data that is obviously abnormal or has a high missing rate.

[0091] Specifically, an adversarial feature distillation network is constructed, including a teacher model and a student model. The robustness of the student model is enhanced through adversarial training. An initial credibility score is calculated based on the historical accuracy, update frequency, and source authority of the data source. During adversarial training, the weights are dynamically updated through gradient feedback. Adversarial samples are identified by the prediction differences between the teacher and student models. For suspected contaminated data, interpolation or deletion is performed based on the statistical characteristics of neighboring data points.

[0092] Different modal data are mapped to a unified semantic space, asynchronous data are weighted and averaged according to time windows, the fusion strategy is dynamically adjusted according to the performance of downstream tasks, and an anti-interference fused data stream is output.

[0093] Preferably, the production situation awareness map module acquires anti-interference fusion data streams, distinguishes between static and dynamic data, and associates sensor data with text descriptions through the multimodal perception capabilities of the embodied cognition engine to form semantic tags;

[0094] Specifically, key entities are extracted from the data stream, combined with the hierarchical relationships defined by ontology, and the physical interactions between entities are simulated through an embodied cognition engine. Implicit relationships are then deduced by combining historical behavior patterns.

[0095] Based on real-time data and preset rules, behavioral sequences of entities are generated. The feasibility of these behaviors is verified using a simulation engine, the impact of entity behaviors on the overall situation is quantified, and a production situation awareness map is constructed.

[0096] Nodes: Entity, Status, Timestamp;

[0097] Edges: Relationships between entities, triggering conditions for behaviors;

[0098] In this embodiment, new nodes or edges are added based on the real-time data stream, and when a critical event occurs, the potential impact of related entities is re-derived through the embodied cognition engine.

[0099] Preferably, the adaptive diagnostic analysis module includes the following steps for causal chain extraction and modeling: obtaining entities and their temporal causal relationships from the dynamic production situation awareness map module, and receiving real-time business requirements from external input.

[0100] Suppose there exist i causal chains C1, C2, ..., C i The weight w of each causal chain i Based on real-time business requirements R i The dynamic adjustment and comprehensive diagnostic strategy D are implemented using the following formula:

[0101]

[0102] In this embodiment, w iBased on Bayesian update rules, and by adjusting weights using historical data and real-time feedback, the weighted causal chain S is used as the initial input to the diagnostic strategy, representing the priority ranking of key issues.

[0103] Preferably, the adaptive diagnostic analysis module includes a counterfactual reasoning step comprising: assuming, based on a comprehensive diagnostic strategy, that for variable X... j Intervention is carried out to calculate the effect on target Y. j The effect is achieved by the following formula:

[0104]

[0105] In the formula, Y cr Z represents the target value in a counterfactual hypothetical scenario, x′ represents the variable value after intervention, and Z represents the target value in a counterfactual hypothetical scenario. j 'm' represents the external variable (such as supply chain stability), and 'm' represents the dimension of the virtual scenario.

[0106] Preferably, the adaptive diagnostic analysis module includes a multi-objective trade-off step comprising: adjusting Y... cr Normalization is performed to eliminate dimensional differences, and then the normalized Y is combined. cr And a comprehensive diagnostic strategy, calculating an adaptive diagnostic strategy, the formula is as follows:

[0107]

[0108] In the formula, U represents the adaptive diagnostic strategy, and α k Y represents the weight of the k-th objective. cr,k Y represents the target value of the k-th target in a counterfactual hypothetical scenario. min,k Y represents the minimum possible value of the k-th objective. max,k Let S represent the maximum possible value of the k-th objective, p represent the number of objectives, and log(S) represent the nonlinear effect of the balanced comprehensive diagnostic strategy.

[0109] In this embodiment, multiple candidate strategies are generated based on counterfactual reasoning results and adaptive diagnostic strategies. The generated strategies are then input into a digital twin model to simulate the execution effect. The strategy parameters are adjusted based on the simulation results, and the effectiveness of the strategies is verified through a latent variable benchmarking analysis module.

[0110] Preferably, the cross-domain collaborative inference strategy module includes the following steps for generating cross-domain collaborative optimization strategies: receiving an adaptive diagnostic strategy U, quantifying enterprise resources and modeling them as resource constraints, and decomposing the adaptive diagnostic strategy into executable actions for each domain based on the objectives and resource constraints of the adaptive diagnostic strategy. The formula for generating the cross-domain collaborative optimization strategy is as follows:

[0111]

[0112] In the formula, S represents the cross-domain collaboration strategy, which is a decision combination of allocating resources and actions across multiple domains. (t+1) argm represents the optimized coordination strategy. S ax represents the maximization operation, which seeks the strategy S that maximizes the objective function;

[0113] Specifically, This represents a multi-objective weighted sum, where the weights are w. k Measuring different objectives f k priority, w k f represents the target weight. k (S) represents the k-th objective function, λ represents the cooperative gain adjustment coefficient, and ΔΠ(S,E,U) represents the cooperative gain function;

[0114] In this embodiment, subkect represents the constraint condition, E represents the resource allocation matrix, which shows the resource occupancy of each domain, A represents the resource demand matrix, which shows the resource demand of each adaptive diagnostic strategy S, and C represents the resource constraint condition.

[0115] E·A≤C represents a resource constraint, U adap Let ΔΠ(S,E,U) represent the feasible space of the adaptive diagnostic strategy.

[0116] It should be understood that the combined gain of the two domains is calculated by weighting U, and T ::ij (S,E) represents a flow coupling tensor slice, describing the cross-domain linkage relationship under the cross-domain cooperation strategy and resource allocation E. * Denotes the nuclear norm, || || F Let P, L, F, and I represent the Frobenius norm, and let P, L, F, and I represent the four flows (personnel flow, logistics flow, capital flow, and information flow). This indicates the overall efficiency of cross-domain collaboration. This represents the mean of the efficiency of independent optimization in a single domain.

[0117] In this embodiment, a verified digital twin model is invoked, the real-time enterprise resource status is synchronized to the digital twin model, the generated collaborative strategy is injected into the digital twin model, its execution process is simulated, key indicators are fed back in real time through the model, the simulation results are compared with the strategy objectives, sudden interference is simulated, and the stability of the strategy is verified.

[0118] Preferably, the latent variable benchmarking analysis module records the business performance under the current collaborative strategy, and calculates the virtual result by assuming that the external variables have not been disturbed through counterfactual reasoning. Comparison with actual results Counterfactual results Quantifying the impact of external variables:

[0119] Align the distribution of business indicators under different external conditions to a unified standard, eliminate the systematic influence of external variables on the indicators by mean alignment, and output pure capability indicators;

[0120] In this embodiment, latent variable modeling involves defining latent variables that influence the benchmarking results, establishing the relationship between observed variables and latent variables, mapping the debiased indicators to the pure capability space through kernel principal component analysis, updating the latent variable model through online learning based on real-time business status, normalizing the debiased indicators, dynamically adjusting the scoring weights according to the collaborative optimization strategy, and constructing the scoring matrix.

[0121] Line: Benchmark object;

[0122] Column: Scoring Dimensions;

[0123] Value: Weighted score

[0124] Output format:

[0125] Bias-free benchmarking scoring matrix:

[0126]

[0127] Where m represents the number of benchmark objects and n represents the number of scoring dimensions.

[0128] Preferably, the cross-scene migration module maps the score values ​​to the [0, 1] interval according to the bias-free benchmarking score matrix to ensure that the score dimensions are consistent across all scenes;

[0129] Transform the rating matrix into transferable knowledge units:

[0130] K-Means is used to cluster the rows of the rating matrix to identify similar strategy patterns;

[0131] Extract the feature vectors corresponding to the cluster centers;

[0132] Meta-transfer capsule network construction:

[0133] The standardized bias-free benchmarking score matrix is ​​encoded into a low-dimensional vector, local features are extracted by a convolutional neural network to generate primary capsules, and the output of the primary capsules is aggregated into higher-order capsules by a dynamic routing algorithm.

[0134] Capsule deconstruction and recombination:

[0135] Capsule deconstruction: Decode the output vector of the digital capsule to recover its corresponding policy features, and separate the magnitude and direction of the capsule vector through backpropagation;

[0136] Capsule recombination: The deconstructed knowledge units are linearly combined according to the scenario weights, and the coupling coefficient is updated iteratively to ensure that the recombinated capsules meet the constraints of the new scenario.

[0137] In this embodiment, the constraints of the new scene are injected into the recombined capsule network. The generator is a meta-transfer capsule network that generates candidate strategies. The discriminator is an adversarial network based on scene constraints that evaluates the feasibility of the strategies. The discriminator provides multi-dimensional gradients to guide the generator in adjusting the capsule parameters.

[0138] Training is accelerated by using a negative sampling module, taking into account both strategies that users are interested in and those they are not, and the recombined capsule output is transformed into interpretable rules;

[0139] Based on the capsule attributes, a predefined rule template is matched, and the capsule vector is converted into a natural language rule through a sequence-to-sequence model. The rule is verified to meet the scenario constraints. The collaborative gain after the rule application is predicted through counterfactual reasoning and fed back to the multi-source adversarial fusion module.

[0140] Preferably, a production and operation benchmarking method based on multi-source data analysis includes the following steps:

[0141] S1. Based on the input from external data sources, the system dynamically allocates trusted weights through an adversarial feature distillation network to filter contaminated data and fuse heterogeneous data, outputting an anti-interference fused data stream.

[0142] S2. Based on the anti-interference fusion data stream, simulate entity behavior coupling through the embodied cognition engine to construct a dynamic production situation awareness map;

[0143] S3. Based on the causal chain of the dynamic production situation awareness map and real-time business needs, perform counterfactual reasoning and multi-objective trade-offs to generate adaptive diagnostic strategies.

[0144] S4. The cross-domain collaborative deduction strategy module generates a cross-domain collaborative optimization strategy based on the diagnostic strategy and enterprise resource constraints, and verifies the execution effect through a digital twin model.

[0145] S5. Based on the collaborative optimization strategy, external variables are removed by counterfactual causal comparison, pure capability space mapping is performed, and the biased benchmarking score matrix is ​​output.

[0146] S6. Based on the benchmarking scoring matrix, knowledge units are deconstructed through meta-transfer capsule networks, capsule recombination and scenario constraint injection are performed, and a set of transferable capsule rules is output and fed back to step S1.

[0147] Working principle: The adversarial feature distillation network dynamically allocates the credible weights of the data source, initializes them with indicators such as historical accuracy and update frequency, and optimizes the weights in real time through gradient feedback. Unstructured data is converted into a unified vector representation. Asynchronous data sources eliminate time series bias through timestamp alignment. The prediction difference between the teacher model and the student model is used to identify adversarial examples. The statistical characteristics of neighboring data points are combined to correct contaminated data. Finally, the heterogeneous data is mapped to a unified semantic space and outputs an anti-interference fused data stream.

[0148] Based on anti-interference data streams, an embodied cognition engine simulates the physical interactions and behavioral coupling between entities, constructing a dynamic graph. Sensor data and text descriptions are correlated through multimodal perception to generate semantic tags. Key entities are extracted and nodes and edges are defined based on ontological hierarchical relationships. Real-time data drives dynamic graph updates. When key events are triggered, potential impacts are re-derived, quantifying the impact of entity behavior on the global situation, achieving real-time perception and evolutionary modeling of production scenarios. Temporal causal chains are extracted from the situation graph, and the weights of these chains are dynamically adjusted based on real-time business needs to generate preliminary diagnostic strategies. Weights are iteratively optimized based on Bayesian update rules. Counterfactual reasoning simulates the impact of variable intervention on the target, quantifying multi-objective conflicts. After normalization, an adaptive strategy is generated using a nonlinear equilibrium mechanism, and the feasibility of the strategy is verified through a digital twin model. Parameters are corrected using latent variable benchmarking analysis, and candidate strategies are output. The cross-domain collaborative inference strategy module receives the diagnostic strategy, quantifies enterprise resources and models them as constraints, decomposes the strategy into cross-domain executable actions, and then... Multi-objective weighted and collaborative gain function optimization of resource allocation, combined with digital twin model to synchronize real-time resource status, simulate strategy execution effect and verify stability under sudden interference, and finally output cross-domain collaborative strategy to balance the priority of each domain and global efficiency. Counterfactual reasoning is used to remove the impact of external variables on business performance, calculate the difference between virtual and actual results, and generate pure capability index after mean alignment. Combined with kernel principal component analysis, the debiased index is mapped to a unified capability space, the scoring weight is dynamically adjusted, and row and column debiased benchmarking scoring matrix is ​​constructed to eliminate environmental interference. After standardizing the debiased scoring matrix, K-Means clustering is used to identify similar strategy patterns, extract feature vectors and input them into meta-transfer capsule network. The capsule network extracts local features through convolution, dynamically routes to generate higher-order capsules, and after deconstruction, knowledge units are reorganized according to the constraints of the new scenario. Adversarial network evaluates the feasibility of strategy and optimizes parameters. Sequence to sequence model transforms capsule vectors into interpretable rules and feeds them back to multi-source adversarial fusion module to realize cross-scenario strategy transfer and closed-loop optimization.

[0149] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their likenesses.

[0150] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A production operation benchmarking system based on multi-source data analysis, characterized in that: It includes a multi-source adversarial fusion module, a production situation awareness map module, an adaptive diagnostic analysis module, a cross-domain collaborative inference strategy module, a latent variable benchmarking analysis module, and a cross-scenario migration module; The adaptive diagnostic analysis module performs counterfactual reasoning and multi-objective trade-offs based on the causal chain of the dynamic production situation awareness map and real-time business needs, and generates an adaptive diagnostic strategy. The cross-domain collaborative inference strategy module generates cross-domain collaborative optimization strategies based on diagnostic strategies and enterprise resource constraints, and verifies the execution effect through a digital twin model; The latent variable benchmarking analysis module, based on a collaborative optimization strategy, removes external variables through counterfactual causal comparison. Perform pure capability space mapping and output the bias-free benchmarking score matrix.

2. The production operation benchmarking system based on multi-source data analysis according to claim 1, characterized in that: The multi-source adversarial fusion module acquires raw data from heterogeneous systems, converts unstructured data into a unified vector representation, aligns asynchronous data sources with timestamps, and eliminates data timing deviations. An adversarial feature distillation network is constructed, including a teacher model and a student model. The robustness of the student model is enhanced through adversarial training. During adversarial training, the weights are dynamically updated through gradient feedback. The prediction differences between the teacher and student models are used to identify adversarial samples. For suspected contaminated data, interpolation is performed by combining the statistical characteristics of neighboring data points. The fusion strategy is dynamically adjusted according to the performance of downstream tasks, and an anti-interference fused data stream is output.

3. The production operation benchmarking system based on multi-source data analysis according to claim 1, characterized in that: The production situation awareness map module acquires anti-interference fusion data streams, distinguishes between static and dynamic data, and associates sensor data with text descriptions through the multimodal perception capabilities of the embodied cognition engine to form semantic tags. Based on real-time data and preset rules, behavioral sequences of entities are generated. The feasibility of these behaviors is verified using a simulation engine, the impact of entity behaviors on the overall situation is quantified, and a production situation awareness map is constructed. Nodes: Entity, Status, Timestamp; Edges: Relationships between entities, triggering conditions for behaviors; New nodes or edges are added based on the real-time data stream, and when key events occur, the potential impact of related entities is re-inferred through the embodied cognition engine.

4. A production operation benchmarking system based on multi-source data analysis according to claim 1, characterized in that: The adaptive diagnostic analysis module includes a causal chain extraction and modeling step that involves: obtaining entities and their temporal causal relationships from the dynamic production situation awareness map module, and receiving real-time business requirements from external input. Assume there are i causal chains C1, C2, ..., C... i The weight w of each causal chain i Based on real-time business requirements R i The dynamic adjustment and comprehensive diagnostic strategy D are implemented using the following formula:

5. A production operation benchmarking system based on multi-source data analysis according to claim 4, characterized in that: The adaptive diagnostic analysis module includes a counterfactual reasoning step: based on the comprehensive diagnostic strategy, let the variable X... j Intervention is carried out to calculate the effect on target Y. j The effect is achieved by the following formula: In the formula, Y cr Z represents the target value in a counterfactual hypothetical scenario, x′ represents the variable value after intervention, and Z represents the target value in a counterfactual hypothetical scenario. j represents the external variable, and m represents the dimension of the virtual scenario.

6. A production operation benchmarking system based on multi-source data analysis according to claim 5, characterized in that: The adaptive diagnostic analysis module includes a multi-objective trade-off step: normalizing the objective value under the counterfactual virtual scenario, and combining it with the normalized Y. cr And a comprehensive diagnostic strategy, calculating an adaptive diagnostic strategy, the formula is as follows: In the formula, U represents the adaptive diagnostic strategy, and α k Y represents the weight of the k-th objective. cr,k Y represents the target value of the k-th target in a counterfactual hypothetical scenario. min,k Y represents the minimum possible value of the k-th objective. max,k Let S represent the maximum possible value of the k-th objective, p represent the number of objectives, and log(S) represent the nonlinear effect of the balanced comprehensive diagnostic strategy.

7. A production operation benchmarking system based on multi-source data analysis according to claim 1, characterized in that: The cross-domain collaborative inference strategy module includes the following steps for generating cross-domain collaborative optimization strategies: receiving the adaptive diagnostic strategy U, quantifying enterprise resources and modeling them as resource constraints, and decomposing the adaptive diagnostic strategy into executable actions for each domain based on the objectives and resource constraints of the adaptive diagnostic strategy. The formula for generating the cross-domain collaborative optimization strategy is as follows: U∈U adap , In the formula, S represents the cross-domain collaboration strategy, S (t+1) This represents the optimized collaboration strategy. This indicates a maximize operation. w represents a multi-objective weighted sum. k f represents the target weight. k (S) represents the k-th objective function, λ represents the cooperative gain adjustment coefficient, ΔΠ(S,E,U) represents the cooperative gain function, subkect represents the constraint condition, E represents the resource allocation matrix, A represents the resource demand matrix, C represents the resource constraint condition, E·A≤C represents the resource constraint condition, and U adap This represents the feasible space for adaptive diagnostic strategies.

8. A production operation benchmarking system based on multi-source data analysis according to claim 1, characterized in that: The latent variable benchmarking analysis module records the business performance under the current collaborative strategy. Through counterfactual reasoning, assuming that the external variables have not been disturbed, it calculates the virtual results, compares the actual results with the counterfactual results, quantifies the impact of the external variables, aligns the distribution of business indicators under different external conditions to a unified standard, eliminates the systematic impact of external variables on the indicators through mean alignment, and outputs pure capability indicators. Latent variable modeling: Define latent variables that affect the benchmarking results, establish the relationship between observed variables and latent variables, map the debiased indicators to the pure capability space through kernel principal component analysis, update the latent variable model through online learning based on real-time business status, normalize the debiased indicators, dynamically adjust the scoring weights according to the collaborative optimization strategy, and generate a debiased benchmarking scoring matrix.

9. A production operation benchmarking system based on multi-source data analysis according to claim 1, characterized in that: The cross-scene migration module maps the score values ​​to the [0, 1] interval according to the bias-free benchmarking score matrix, ensuring that the score dimensions are consistent across all scenes. Meta-transfer capsule network construction: The standardized bias-free benchmarking score matrix is ​​encoded into a low-dimensional vector, local features are extracted through a convolutional neural network to generate primary capsules, and the output of the primary capsules is aggregated into higher-order capsules through a dynamic routing algorithm for capsule deconstruction and recombination. The constraints of the new scenario are injected into the recombined capsule network. The generator is a meta-transfer capsule network that generates candidate strategies. The discriminator is an adversarial network based on scenario constraints that evaluates the feasibility of the strategies. The discriminator provides multi-dimensional gradients to guide the generator in adjusting the capsule parameters. Training is accelerated by using a negative sampling module, which transforms the recombined capsule output into interpretable rules. Based on the capsule attributes, a predefined rule template is matched, and the capsule vector is converted into a natural language rule through a sequence-to-sequence model. The rule is verified to meet the scenario constraints. The collaborative gain after the rule application is predicted through counterfactual reasoning and fed back to the multi-source adversarial fusion module.

10. A production operation benchmarking method based on multi-source data analysis implemented according to claim 1, characterized in that, Includes the following steps: S1. Based on the input from external data sources, the system dynamically allocates trusted weights through an adversarial feature distillation network to filter contaminated data and fuse heterogeneous data, outputting an anti-interference fused data stream. S2. Based on the anti-interference fusion data stream, simulate entity behavior coupling through the embodied cognition engine to construct a dynamic production situation awareness map; S3. Based on the causal chain of the dynamic production situation awareness map and real-time business needs, perform counterfactual reasoning and multi-objective trade-offs to generate adaptive diagnostic strategies. S4. The cross-domain collaborative deduction strategy module generates a cross-domain collaborative optimization strategy based on the diagnostic strategy and enterprise resource constraints, and verifies the execution effect through a digital twin model. S5. Based on the collaborative optimization strategy, external variables are removed by counterfactual causal comparison, pure capability space mapping is performed, and the biased benchmarking score matrix is ​​output. S6. Based on the benchmarking scoring matrix, knowledge units are deconstructed through meta-transfer capsule networks, capsule recombination and scenario constraint injection are performed, and a set of transferable capsule rules is output and fed back to step S1.

Images