Enterprise gene generation method and system based on enterprise multidimensional data

By building an enterprise knowledge graph and using machine learning technology to generate enterprise genes, the problem of lack of global optimization methods in enterprise digital transformation is solved, intelligent decision-making support and continuous optimization of enterprise production and operation is achieved, and data-driven decision-making capabilities are improved.

CN118503491BActive Publication Date: 2024-11-05BEIJING LONGDUN DATA CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410962468.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-18
Publication Date
2024-11-05
Estimated Expiration
2044-07-18

AI Technical Summary

Technical Problem

In the process of digital transformation of enterprises, existing technologies lack the method of portraying and optimizing the overall characteristics and operation mechanism of the enterprise from a global perspective, and the integration and integration of multiple technologies have not been achieved, and data-driven enterprise insights and decision-making optimization capabilities are insufficient.

Method used

By collecting multi-source heterogeneous data of enterprises, building enterprise knowledge graphs, extracting structured and unstructured data features for feature fusion, generating multi-dimensional features of enterprises, and using machine learning technology to train enterprise gene generation models, combining multi-objective optimization algorithms to build and optimize the enterprise genes to achieve global optimization and intelligent decision-making support for enterprise production and operations.

Benefits of technology

It realizes global optimization of enterprise production and operation, improves production plan formulation efficiency, reduces the burden of manual decision-making, can monitor and adapt to production abnormalities in real time, improves the robustness of production plan execution, and continuously iterates and optimizes enterprise knowledge assets through version management .

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118503491B_ABST
    Figure CN118503491B_ABST
Patent Text Reader

Abstract

The present invention provides a method and system for generating enterprise genes based on enterprise multidimensional data, which relates to the field of data processing technology and is used to solve the problem that the existing technology has single data and cannot fully understand the enterprise genes. The present invention includes: collecting and preprocessing enterprise multi-source heterogeneous data to construct an enterprise knowledge graph; extracting structured and unstructured data features of the enterprise knowledge graph, performing feature fusion, and obtaining enterprise multidimensional features; inputting enterprise multidimensional features into a pre-trained enterprise gene generation model to generate initial enterprise genes, constructing an enterprise gene multi-objective optimization model, solving the enterprise gene multi-objective optimization model through a multi-objective optimization algorithm, and obtaining optimized enterprise genes; detecting production anomalies, and if production anomalies are detected, re-optimizing the optimized enterprise genes according to a preset production anomaly handling strategy, while introducing version management and update mechanisms to achieve iterative optimization and continuous evolution of enterprise genes.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to data processing technology, and in particular to a method and system for generating enterprise genes based on enterprise multidimensional data. Background Art

[0002] With the development of new-generation information technologies such as industrial Internet, big data, and artificial intelligence, digital transformation and upgrading of enterprises has become a key measure to enhance the core competitiveness of enterprises. However, enterprises are still facing many challenges in the process of digital transformation, such as prominent data island problems, difficulty in optimizing business processes, and lack of decision-making basis. Therefore, there is an urgent need for a new technical solution that can fully tap the value of multi-source heterogeneous data of enterprises, characterize the key characteristics of enterprises, and assist enterprises in optimizing decision-making.

[0003] In the existing technology, there have been some explorations and attempts for enterprise digital transformation and optimization decision-making. For example, by building a manufacturing process knowledge map and intelligent optimization model, intelligent optimization and decision-making of the production process can be achieved; or by using big data analysis and machine learning technology to mine and model equipment operation data, equipment health status monitoring and fault prediction can be achieved.

[0004] However, existing technologies mainly focus on optimizing a specific business area or link, and lack methods to characterize and optimize the overall characteristics and operating mechanisms of the enterprise from a global perspective. At the same time, most existing technologies independently apply a single technology and have not yet achieved the integration and fusion of multiple technologies. The data-driven enterprise insight and decision-making optimization capabilities need to be further enhanced.

[0005] Therefore, the purpose of the present invention is to provide a method and system for generating enterprise genes based on enterprise multidimensional data to solve the above-mentioned technical problems existing in the prior art. The present invention integrates multi-source heterogeneous enterprise data and uses data mining, machine learning, knowledge graph and other technologies to construct a digital "enterprise gene" that reflects the key characteristics and operation mechanism of the enterprise, and on this basis, realizes the global optimization and intelligent decision-making support of the enterprise's production and operation, providing new technical means for the digital transformation and upgrading of enterprises. Summary of the invention

[0006] The embodiments of the present invention provide a method and system for generating enterprise genes based on enterprise multi-dimensional data, which can solve the problems in the prior art.

[0007] According to a first aspect of the embodiments of the present invention,

[0008] A method for generating enterprise genes based on enterprise multidimensional data is provided, comprising:

[0009] Collecting enterprise multi-source heterogeneous data, wherein the enterprise multi-source heterogeneous data includes structured data and unstructured data, preprocessing the enterprise multi-source heterogeneous data to obtain standard enterprise data, constructing an enterprise knowledge graph based on the obtained standard enterprise data, extracting structured data features and unstructured data features in the enterprise knowledge graph, and performing feature fusion to obtain enterprise multi-dimensional features;

[0010] Inputting the enterprise's multi-dimensional features into a pre-trained enterprise gene generation model to generate an initial enterprise gene, constructing an enterprise gene multi-objective optimization model based on the initial enterprise gene, solving the enterprise gene multi-objective optimization model through a multi-objective optimization algorithm, and obtaining an optimized enterprise gene;

[0011] The optimized enterprise gene is decoded and converted into an enterprise production plan, and combined with real-time production operation data, production anomalies are detected. If a production anomaly is detected, the optimized enterprise gene is re-optimized according to a preset production anomaly handling strategy. At the same time, a version management and update mechanism is introduced to achieve iterative optimization and continuous evolution of the enterprise gene.

[0012] In an optional embodiment,

[0013] Based on the obtained standard enterprise data, the enterprise knowledge graph is constructed, the structured data features and unstructured data features in the enterprise knowledge graph are extracted, and the features are fused to obtain the enterprise multi-dimensional features including:

[0014] According to the obtained enterprise knowledge graph, by constructing positive sample pairs and negative sample pairs, the initial embedding representation of the nodes in the enterprise knowledge graph is trained, and the contrastive learning loss function is introduced. The initial embedding representation of the node is updated by minimizing the contrastive learning loss function to obtain the node embedding vector;

[0015] Construct a multi-layer graph attention network. In the first layer of the multi-layer graph attention network, the node embedding vector is used as the initial feature representation of the current node, the attention coefficient between the current node and its neighboring nodes is calculated, and the features of all neighboring nodes are weighted summed based on the attention coefficient to obtain the aggregated feature representation of the current node.

[0016] In each layer of the multi-layer graph attention network, the aggregated feature representations of the nodes output by the previous layer are concatenated, and nonlinear transformation is performed through the feedforward neural network to obtain the target node features, and the target node feature representation is passed to the next layer. When reaching the last layer of the multi-layer graph attention network, the target node features of all nodes are aggregated using a global pooling operation to obtain the feature representation vector of the enterprise knowledge graph;

[0017] According to the obtained feature representation vector of the enterprise knowledge graph, the multi-layer graph attention network is trained end-to-end using a back-propagation algorithm to obtain a trained multi-layer graph attention network model;

[0018] The trained multi-layer graph attention network model is used to extract the structured data features and unstructured data features in the enterprise knowledge graph, and then perform feature fusion to obtain the multi-dimensional features of the enterprise.

[0019] In an optional embodiment,

[0020] Input the enterprise's multi-dimensional features into the pre-trained enterprise gene generation model to generate the initial enterprise genes including:

[0021] Build a conditional generative adversarial network model that meets production constraints, including a generator and a discriminator;

[0022] The generator receives random noise and conditional variables as input, generates enterprise genes that meet production constraints through transposed convolution layers and batch normalization layers, and sets a resource constraint layer at the last layer of the generator to map the generated enterprise genes to the resource consumption space, and calculates the difference between the amount of resources consumed by the enterprise genes and the actual amount of resources available to the enterprise as the resource constraint loss;

[0023] The discriminator receives the enterprise gene generated by the generator as input, generates the discrimination result of the enterprise gene through the convolution layer and the batch normalization layer, and at the same time, sets the task constraint discrimination layer in the last layer of the discriminator to predict the probability that the enterprise gene belongs to the preset production task category, and calculates the task constraint loss based on the prediction result and the real label of the preset production task category;

[0024] The comprehensive loss function of the conditional generative adversarial network model is constructed based on the generator loss, discriminator loss and constraint loss. The constraint loss includes resource constraint loss and task constraint loss. By minimizing the comprehensive loss function, the parameters of the conditional generative adversarial network model are updated. The iteration is repeated until the preset termination condition is met, and finally a trained enterprise gene generation model is obtained.

[0025] The extracted multi-dimensional features of the enterprise are mapped into conditional variables using a pre-trained feature encoder, and random noise is sampled from a Gaussian distribution. The obtained conditional variables and random noise are input into the trained enterprise gene generation model to generate the initial enterprise gene.

[0026] In an optional embodiment,

[0027] The calculation formula for the comprehensive loss function of the conditional generative adversarial network model constructed based on the generator loss, discriminator loss, and constraint loss is as follows:

[0028] ;

[0029] Among them, L total represents the resource constraint loss, L G represents the generator loss, L D represents the discriminator loss, m represents the number of available resources for an enterprise, n represents the dimension of resource consumption in the generated enterprise gene, and r ij represents the consumption of the i-th resource in the j-th dimension of the generated enterprise gene, g ij represents the value of the i-th resource in the j-th dimension of the generated enterprise gene, R i represents the actual available quantity of the i-th resource, k represents the number of preset production task categories, and y l Indicates l The true category labels of the production tasks, Indicates l is the predicted class label probability of the production task, λ1 represents the hyperparameter for balancing the generator loss and the constraint loss, and λ2 represents the hyperparameter for balancing the discriminator loss and the constraint loss.

[0030] In an optional embodiment,

[0031] Based on the initial enterprise gene, constructing an enterprise gene multi-objective optimization model includes:

[0032] Construct objective functions, which include minimizing the completion time of production tasks, minimizing production costs, and maximizing resource utilization; construct constraints, which include the delivery time constraints, cost constraints, and resource constraints of production tasks;

[0033] Taking the initial enterprise gene as an optimization variable, constructing an enterprise gene multi-objective optimization model according to the objective function and constraint conditions;

[0034] The calculation formula of the objective function is as follows:

[0035] ;

[0036] Where J represents the objective function, α represents the weight coefficient of completion time, A represents the total number of production tasks, C a represents the completion time of production task a, T a represents the delivery time of production task a, T avg represents the average delivery time of all production tasks, β represents the weight coefficient of production cost, C budget represents the total production cost budget, γ represents the weight coefficient of resource utilization, B represents the number of production resource types, R b represents the actual available quantity of production resource b, and R represents the total available quantity of production resource b.

[0037] In an optional embodiment,

[0038] The enterprise gene multi-objective optimization model is solved by a multi-objective optimization algorithm, and the optimized enterprise genes include:

[0039] The enterprise gene is divided into multiple submodules, each of which corresponds to a key element of the production plan. Each submodule is encoded, and a set of initial position vectors is generated based on the enterprise's historical data and expert experience to form an initial particle swarm, where each particle represents a potential optimization solution;

[0040] According to the constructed enterprise gene multi-objective optimization model, the objective function value and the degree of satisfaction of the constraint conditions of each particle are calculated. At the same time, the fuzzy measurement mechanism is introduced to combine the objective function value and the degree of satisfaction of the constraint conditions to obtain the comprehensive fitness value of the particle;

[0041] Compare the current comprehensive fitness value of the particle with the historical optimal comprehensive fitness value. If the current comprehensive fitness value is greater than the historical optimal comprehensive fitness value, update the individual optimal solution of the corresponding particle and update the global optimal solution of the particle swarm at the same time.

[0042] Based on the individual optimal solution of each particle and the global optimal solution of the particle swarm, an adaptive mutation and crossover algorithm is used to dynamically adjust the crossover position and crossover probability according to the similarity of the particles to generate new particles;

[0043] A multi-stage constraint processing mechanism is used for the newly generated particles. Particles that violate hard constraints are repaired and particles that violate soft constraints are punished. The global optimal solution of the particle swarm is further optimized through a local search strategy to obtain a new particle swarm.

[0044] Repeat the iteration until the adaptive termination condition is met, terminate the optimization process, and output the optimized enterprise gene non-dominated solution set. Based on the obtained enterprise gene non-dominated solution set, it is visualized and interactive screening and comparison tools are provided for decision makers to select satisfactory enterprise gene solutions as the final optimized enterprise genes based on their preferences and trade-offs.

[0045] In an optional embodiment,

[0046] The optimized enterprise gene is decoded and converted into an enterprise production plan, and combined with real-time production operation data, production anomalies are detected. If a production anomaly is detected, the optimized enterprise gene is re-optimized according to the preset production anomaly handling strategy, and a version management and update mechanism is introduced to achieve iterative optimization and continuous evolution of the enterprise gene, including:

[0047] Decode the optimized enterprise genes, extract key information of the enterprise's production tasks, and use the rule-based production plan generation algorithm to convert the extracted key information into an executable enterprise production plan;

[0048] Collect production operation data in real time and perform preprocessing, build a production anomaly detection model based on a machine learning algorithm, and use the production anomaly detection model to identify production anomalies and corresponding production anomaly types;

[0049] Automatically trigger the production plan re-optimization process according to the production exception type and the preset exception handling strategy, use the heuristic optimization algorithm to generate the optimized production plan, and update the enterprise gene according to the optimized production plan, use the evolutionary computing algorithm to search for the optimal enterprise gene, and record the updated version of the current enterprise gene;

[0050] Utilizing a distributed version control system based on blockchain, an enterprise gene version dependency graph is established for each updated version of the current enterprise gene recorded, and smart contracts are designed to control the access rights and update rules of the enterprise gene version. At the same time, a visual enterprise gene management platform is developed to manage and update the enterprise gene version.

[0051] According to a second aspect of the embodiments of the present invention,

[0052] Provided is an enterprise gene generation system based on enterprise multidimensional data, comprising:

[0053] The first unit is used to collect enterprise multi-source heterogeneous data, wherein the enterprise multi-source heterogeneous data includes structured data and unstructured data, pre-process the enterprise multi-source heterogeneous data to obtain standard enterprise data, construct an enterprise knowledge graph based on the obtained standard enterprise data, extract structured data features and unstructured data features in the enterprise knowledge graph, and perform feature fusion to obtain enterprise multi-dimensional features;

[0054] The second unit is used to input the enterprise multi-dimensional features into the pre-trained enterprise gene generation model to generate initial enterprise genes, build an enterprise gene multi-objective optimization model based on the initial enterprise genes, solve the enterprise gene multi-objective optimization model through a multi-objective optimization algorithm, and obtain an optimized enterprise gene;

[0055] The third unit is used to decode the optimized enterprise gene and convert it into an enterprise production plan, and detect production anomalies in combination with real-time production operation data. If a production anomaly is detected, the optimized enterprise gene is re-optimized according to a preset production anomaly handling strategy. At the same time, a version management and update mechanism is introduced to realize iterative optimization and continuous evolution of the enterprise gene.

[0056] According to a third aspect of the embodiments of the present invention,

[0057] An electronic device is provided, comprising:

[0058] processor;

[0059] a memory for storing processor-executable instructions;

[0060] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0061] According to a fourth aspect of the embodiments of the present invention,

[0062] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0063] Beneficial effects of the present invention:

[0064] (1) The present invention can efficiently integrate multi-source heterogeneous data inside and outside the enterprise by preprocessing and extracting features from structured data and unstructured data, and adopt an effective feature fusion method, so as to make full use of the enterprise's data assets and provide comprehensive data support for production decisions.

[0065] (2) The present invention constructs an enterprise knowledge graph to formalize and systematically store the enterprise's domain knowledge, forming the enterprise's own knowledge base. The knowledge graph has the characteristics of semantics and structure, and can support intelligent knowledge reasoning and retrieval, providing knowledge support for enterprise decision-making.

[0066] (3) The present invention uses machine learning technology to train the enterprise gene generation model, and combined with the multi-dimensional feature data of the enterprise, it can intelligently generate the initial production plan gene, greatly improve the efficiency of production plan formulation, and reduce the burden of manual decision-making. Based on the initial gene, the enterprise gene multi-objective optimization model is constructed, and the advanced optimization algorithm is used to solve it. It can optimize multiple goals at the same time and generate high-quality production plan solutions. Combined with real-time production operation data, production abnormalities can be monitored in real time. Once an abnormality is found, the production plan gene is dynamically optimized according to the preset strategy, so that the production plan can adapt to the changing production environment in real time and improve the robustness of plan execution.

[0067] (4) By introducing version management and update mechanisms, the present invention enables the production planning gene to be continuously iterated and optimized, and the optimization results are fed back to the knowledge base, continuously enriching the enterprise's knowledge assets and continuously evolving the enterprise's production decision-making capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a flowchart of a method for generating enterprise genes based on enterprise multi-dimensional data according to an embodiment of the present invention;

[0069] Figure 2 It is a structural diagram of an enterprise gene generation system based on enterprise multi-dimensional data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0070] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0071] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0072] Figure 1 FIG. 1 is a flow chart of a method for generating enterprise genes based on enterprise multidimensional data according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0073] S101. Collect enterprise multi-source heterogeneous data, wherein the enterprise multi-source heterogeneous data includes structured data and unstructured data, pre-process the enterprise multi-source heterogeneous data to obtain standard enterprise data, construct an enterprise knowledge graph based on the obtained standard enterprise data, extract structured data features and unstructured data features in the enterprise knowledge graph, and perform feature fusion to obtain multi-dimensional features of the enterprise.

[0074] Among them, structured data can be obtained from enterprise databases, ERP systems, etc., and unstructured data can be obtained by collecting web pages, documents, emails, etc. through Web crawlers or by collecting production operation data through sensors.

[0075] The structured data is cleaned, standardized, and associated and integrated. The unstructured data is segmented and tagged using natural language processing technology, and then information is extracted to obtain standard enterprise data.

[0076] By defining the ontology model of the core concepts and relationships of the enterprise, it is possible to use existing general ontologies or automatically extract based on standard enterprise data, and use NLP technologies such as named entity recognition and relationship extraction to extract entities and relationships from structured and unstructured data. Use a graph database or RDF triples to store the graph, and finally obtain the enterprise knowledge graph.

[0077] Based on the enterprise knowledge graph, we use graph embedding technology to obtain the structured representation vectors of entities and relationships, and use Word2Vec or BERT models to obtain the semantic representation of unstructured data. Through fusion methods such as gated recurrent units or vector concatenation, we fuse structured and unstructured features into a multi-dimensional feature vector of the enterprise.

[0078] In an optional embodiment,

[0079] Based on the obtained standard enterprise data, the enterprise knowledge graph is constructed, the structured data features and unstructured data features in the enterprise knowledge graph are extracted, and the features are fused to obtain the enterprise multi-dimensional features including:

[0080] According to the obtained enterprise knowledge graph, by constructing positive sample pairs and negative sample pairs, the initial embedding representation of the nodes in the enterprise knowledge graph is trained, and the contrastive learning loss function is introduced. The initial embedding representation of the node is updated by minimizing the contrastive learning loss function to obtain the node embedding vector;

[0081] Construct a multi-layer graph attention network. In the first layer of the multi-layer graph attention network, the node embedding vector is used as the initial feature representation of the current node, the attention coefficient between the current node and its neighboring nodes is calculated, and the features of all neighboring nodes are weighted summed based on the attention coefficient to obtain the aggregated feature representation of the current node.

[0082] In each layer of the multi-layer graph attention network, the aggregated feature representations of the nodes output by the previous layer are concatenated, and nonlinear transformation is performed through the feedforward neural network to obtain the target node features, and the target node feature representation is passed to the next layer. When reaching the last layer of the multi-layer graph attention network, the target node features of all nodes are aggregated using a global pooling operation to obtain the feature representation vector of the enterprise knowledge graph;

[0083] According to the obtained feature representation vector of the enterprise knowledge graph, the multi-layer graph attention network is trained end-to-end using a back-propagation algorithm to obtain a trained multi-layer graph attention network model;

[0084] The trained multi-layer graph attention network model is used to extract the structured data features and unstructured data features in the enterprise knowledge graph, and then perform feature fusion to obtain the multi-dimensional features of the enterprise.

[0085] Exemplarily, first, independent encoders (such as MLP or GNN) are used to encode structured data features and unstructured data features to obtain their low-dimensional representations.

[0086] For a corporate knowledge graph sample, we regard its structured features and unstructured features as a positive sample pair, and randomly sample the features of another company as a negative sample. Then, by minimizing the distance between positive sample pairs and maximizing the distance between positive and negative sample pairs, we construct a contrastive learning loss function, and use optimization algorithms such as stochastic gradient descent to minimize the contrastive learning loss function and update the node embedding representation.

[0087] A multi-layer graph attention network is designed, each layer of which contains a multi-head attention mechanism and residual connections. In one of the graph attention layers, the node embedding representation obtained by step pre-training is used as the initial feature representation. The attention coefficient between each node and its neighbor nodes is calculated, and a mask mechanism is used to handle different types of neighbor nodes. Specifically, a mask vector is set according to the type of neighbor nodes (such as homogeneous or heterogeneous). When calculating the attention score, irrelevant neighbor nodes are masked to a very small value, thereby focusing on important neighbor nodes and ignoring irrelevant nodes.

[0088] The attention coefficient is used to perform weighted summation of the features of neighboring nodes to obtain the aggregated feature representation of the node. The output features of the multi-head attention are concatenated and nonlinearly transformed through a feedforward neural network to obtain the updated feature representation of the node. At the top layer of the graph attention network, a global pooling operation is used to aggregate the features of all nodes to obtain the feature representation of the entire knowledge graph.

[0089] In this embodiment, by comparing the design of the learning loss function and the multi-layer graph attention network, the structured representation of nodes can be learned from the enterprise knowledge graph end-to-end, the topological structure and semantic information of the graph data can be fully mined, and high-quality structured feature representation can be obtained. By introducing the graph attention mechanism, the attention distribution of important nodes and their neighboring nodes can be automatically learned, focusing on the relevant nodes and their features that are more important to the semantic representation of the current target node, thereby improving the quality and interpretability of the semantic representation. The hierarchical design of the multi-layer graph attention network enables each layer to perform semantic feature abstraction based on the previous layer, capture higher-order and abstract graph structured knowledge representations layer by layer, and enhance the feature expression capability. It can better extract structured and unstructured heterogeneous data features from the enterprise knowledge graph, obtain high-quality enterprise multi-dimensional semantic feature representations, and provide more valuable feature support for subsequent gene generation and optimization.

[0090] S102. Input the enterprise's multi-dimensional features into a pre-trained enterprise gene generation model to generate initial enterprise genes. Based on the initial enterprise genes, construct an enterprise gene multi-objective optimization model. Solve the enterprise gene multi-objective optimization model through a multi-objective optimization algorithm to obtain optimized enterprise genes.

[0091] In an optional embodiment,

[0092] Input the enterprise's multi-dimensional features into the pre-trained enterprise gene generation model to generate the initial enterprise genes including:

[0093] Build a conditional generative adversarial network model that meets production constraints, including a generator and a discriminator;

[0094] The generator receives random noise and conditional variables as input, generates enterprise genes that meet production constraints through transposed convolution layers and batch normalization layers, and sets a resource constraint layer at the last layer of the generator to map the generated enterprise genes to the resource consumption space, and calculates the difference between the amount of resources consumed by the enterprise genes and the actual amount of resources available to the enterprise as the resource constraint loss;

[0095] The discriminator receives the enterprise gene generated by the generator as input, generates the discrimination result of the enterprise gene through the convolution layer and the batch normalization layer, and at the same time, sets the task constraint discrimination layer in the last layer of the discriminator to predict the probability that the enterprise gene belongs to the preset production task category, and calculates the task constraint loss based on the prediction result and the real label of the preset production task category;

[0096] The comprehensive loss function of the conditional generative adversarial network model is constructed based on the generator loss, discriminator loss and constraint loss. The constraint loss includes resource constraint loss and task constraint loss. By minimizing the comprehensive loss function, the parameters of the conditional generative adversarial network model are updated. The iteration is repeated until the preset termination condition is met, and finally a trained enterprise gene generation model is obtained.

[0097] The extracted multi-dimensional features of the enterprise are mapped into conditional variables using a pre-trained feature encoder, and random noise is sampled from a Gaussian distribution. The obtained conditional variables and random noise are input into the trained enterprise gene generation model to generate the initial enterprise gene.

[0098] Among them, in the process of enterprise gene generation, various constraints in the actual production of the enterprise need to be considered to ensure that the generated enterprise genes are feasible and practical. Therefore, a conditional generative adversarial network model that introduces production constraints is adopted to introduce the enterprise's production resource constraints and production task constraints into the training process of the generative adversarial network to guide the generation of enterprise genes that meet the constraints.

[0099] For example, the production resources of an enterprise, such as manpower, equipment, and raw materials, are the constraints that need to be considered in the process of generating enterprise genes. In order to introduce production resource constraints into the conditional generative adversarial network model, it is first necessary to mathematically model these constraints. Specifically, the production resource constraints are represented as a multidimensional vector. For a given enterprise gene, a resource consumption function can be defined to determine the amount of a certain resource consumed to generate the enterprise gene.

[0100] In order to ensure that the generated enterprise genes meet the production resource constraints, we need to introduce a resource constraint layer (RCL) in the generator of the conditional generative adversarial network model. The role of RCL is to map the enterprise genes output by the generator to the resource consumption space and calculate the difference between it and the actual resource constraints. Optionally, the L2 norm can be used to measure this difference.

[0101] In addition to production resource constraints, enterprise genes also need to meet specific production task requirements, such as product type, quantity, quality, etc. Therefore, these production task constraints can be represented as a set of conditional variables. In order to introduce production task constraints into the conditional generative adversarial network model, we can concatenate the conditional variables with the input noise of the generator to form a conditional input vector. In this way, we can pass the task constraint information to the generator and guide it to generate enterprise genes that meet the task requirements.

[0102] In addition, a task constraint discriminant layer needs to be introduced into the discriminator. The task constraint discriminant layer is used to determine whether the generated enterprise gene meets the given task constraint conditions. The task constraint discriminant layer can be designed as a multi-label classifier to predict which task categories the generated enterprise gene belongs to.

[0103] Combining production resource constraints and production task constraints, a conditional generative adversarial network model that meets production constraints is constructed, which includes a generator and a discriminator. The generator receives random noise and conditional variables as input, and generates enterprise genes that meet the conditional constraints through a series of transposed convolutional layers and batch normalization layers. In the last layer of the generator, the generated enterprise genes are mapped to the resource consumption space through the resource constraint layer, and the resource constraint loss is calculated. The discriminator receives real enterprise genes or generated enterprise genes as input, and outputs a binary discrimination result after a series of convolutional layers and batch normalization layers, indicating whether the input enterprise genes are real or generated. In the last layer of the discriminator, through the task constraint discrimination layer, it is predicted which task categories the input enterprise genes belong to, and the task constraint loss is calculated.

[0104] A comprehensive loss function is constructed based on the generator loss, discriminator loss, and the calculated resource constraint loss and task constraint loss. The generator loss measures the authenticity of the enterprise genes generated by the generator, prompting the generator to generate enterprise genes that are as realistic as possible. The discriminator loss measures the discriminator's ability to distinguish between real and generated enterprise genes, prompting the discriminator to distinguish the authenticity of enterprise genes as accurately as possible. By minimizing the loss function, the final trained enterprise gene generation model can be obtained.

[0105] Using the above-trained enterprise gene generation model, the initial enterprise gene can be generated according to the multi-dimensional characteristics of the enterprise. Specifically, firstly, the corresponding conditional variables are generated according to the extracted enterprise characteristics, and the conditional variables represent the current production resource constraints and production task requirements of the enterprise. A pre-trained enterprise feature encoder can be used to map the feature vector to the conditional variable. The encoder can be a multi-layer perceptron or a convolutional neural network, which is trained by minimizing the feature reconstruction loss and the conditional constraint loss.

[0106] In order to generate diversified corporate genes, it is necessary to sample random noise vectors from a prior distribution (such as a Gaussian distribution). The dimension of the noise vector is usually smaller than that of the corporate gene, so as to capture the essential characteristics of the corporate gene in a low-dimensional space.

[0107] Random noise and conditional variables are input into the trained enterprise gene generation model to generate initial enterprise genes that meet the conditional constraints.

[0108] In an optional embodiment,

[0109] The calculation formula for the comprehensive loss function of the conditional generative adversarial network model constructed based on the generator loss, discriminator loss, and constraint loss is as follows:

[0110] ;

[0111] Among them, L total represents the resource constraint loss, L G represents the generator loss, L D represents the discriminator loss, m represents the number of available resources for an enterprise, n represents the dimension of resource consumption in the generated enterprise gene, and r ij represents the consumption of the i-th resource in the j-th dimension of the generated enterprise gene, g ij represents the value of the i-th resource in the j-th dimension of the generated enterprise gene, R i represents the actual available quantity of the i-th resource, k represents the number of preset production task categories, and y l Indicates l The true category labels of the production tasks, Indicates l is the predicted class label probability of the production task, λ1 represents the hyperparameter for balancing the generator loss and the constraint loss, and λ2 represents the hyperparameter for balancing the discriminator loss and the constraint loss.

[0112] In this embodiment, through the conditional generative adversarial network model, multiple constraints such as the actual available resources of the enterprise and the preset production task categories are integrated into the model training process. By optimizing the resource constraint loss and the task constraint loss, the initial enterprise gene that satisfies both the resource constraint and the task requirement can be generated, ensuring the feasibility and compliance of the generated results. The design of the resource constraint layer and the task constraint discrimination layer enables the model to directly predict resource consumption and task category from the output results, and back-propagate the constraint loss, realizing end-to-end constraint condition generation without the intervention of artificial rules, thereby improving the generation efficiency and quality. The multi-dimensional features of the enterprise extracted a priori are encoded as conditional variables and input into the model, making full use of the structured and unstructured knowledge of the enterprise, making the generated initial genes more in line with the actual situation of the enterprise, and improving the gene quality. Through adversarial training and the introduction of random noise, the unsupervised enterprise gene generation that satisfies multiple constraints is realized, overcoming the bottleneck of large-scale enterprise data annotation difficulties, and efficiently generating high-quality and diversified initial enterprise genes, providing a good initial solution for subsequent gene optimization.

[0113] In an optional embodiment,

[0114] Based on the initial enterprise gene, constructing an enterprise gene multi-objective optimization model includes:

[0115] Construct objective functions, which include minimizing the completion time of production tasks, minimizing production costs, and maximizing resource utilization; construct constraints, which include the delivery time constraints, cost constraints, and resource constraints of production tasks;

[0116] Taking the initial enterprise gene as an optimization variable, constructing an enterprise gene multi-objective optimization model according to the objective function and constraint conditions;

[0117] The calculation formula of the objective function is as follows:

[0118] ;

[0119] Where J represents the objective function, α represents the weight coefficient of completion time, A represents the total number of production tasks, C a represents the completion time of production task a, T a represents the delivery time of production task a, T avg represents the average delivery time of all production tasks, β represents the weight coefficient of production cost, C budget represents the total production cost budget, γ represents the weight coefficient of resource utilization, B represents the number of production resource types, R b represents the actual available quantity of production resource b, and R represents the total available quantity of production resource b.

[0120] For example, after obtaining the initial enterprise gene, it needs to be further optimized to meet the various constraints of the enterprise's production tasks and achieve a balance between multiple objectives such as production efficiency and cost. Therefore, a multi-objective optimization model of enterprise genes is constructed.

[0121] The objective function of the enterprise gene multi-objective optimization model includes minimizing the completion time of production tasks, minimizing production costs, maximizing resource utilization, etc. There may be conflicting relationships between these objectives, which need to be weighed and compromised in the optimization process.

[0122] The constraints of the model include the delivery time constraint, cost constraint and resource constraint of production tasks. The delivery time constraint requires that each production task can be completed within the specified delivery time; the cost constraint requires that the total production cost does not exceed the budget limit; the resource constraint requires that the usage of various production resources does not exceed their available quantity.

[0123] By taking the initial corporate genes as optimization variables and constructing a multi-objective optimization model with objective functions and constraints, we can use multi-objective optimization algorithms to optimize and solve the corporate genes and obtain optimized corporate genes that meet various constraints and achieve a balance between multiple objectives.

[0124] In an optional embodiment,

[0125] The enterprise gene multi-objective optimization model is solved by a multi-objective optimization algorithm, and the optimized enterprise genes include:

[0126] The enterprise gene is divided into multiple submodules, each of which corresponds to a key element of the production plan. Each submodule is encoded, and a set of initial position vectors is generated based on the enterprise's historical data and expert experience to form an initial particle swarm, where each particle represents a potential optimization solution;

[0127] According to the constructed enterprise gene multi-objective optimization model, the objective function value and the degree of satisfaction of the constraint conditions of each particle are calculated. At the same time, the fuzzy measurement mechanism is introduced to combine the objective function value and the degree of satisfaction of the constraint conditions to obtain the comprehensive fitness value of the particle;

[0128] Compare the current comprehensive fitness value of the particle with the historical optimal comprehensive fitness value. If the current comprehensive fitness value is greater than the historical optimal comprehensive fitness value, update the individual optimal solution of the corresponding particle and update the global optimal solution of the particle swarm at the same time.

[0129] Based on the individual optimal solution of each particle and the global optimal solution of the particle swarm, an adaptive mutation and crossover algorithm is used to dynamically adjust the crossover position and crossover probability according to the similarity of the particles to generate new particles;

[0130] A multi-stage constraint processing mechanism is used for the newly generated particles. Particles that violate hard constraints are repaired and particles that violate soft constraints are punished. The global optimal solution of the particle swarm is further optimized through a local search strategy to obtain a new particle swarm.

[0131] Repeat the iteration until the adaptive termination condition is met, terminate the optimization process, and output the optimized enterprise gene non-dominated solution set. Based on the obtained enterprise gene non-dominated solution set, it is visualized and interactive screening and comparison tools are provided for decision makers to select satisfactory enterprise gene solutions as the final optimized enterprise genes based on their preferences and trade-offs.

[0132] Exemplarily, the enterprise gene is divided into multiple submodules, each of which corresponds to a key element of the production plan, such as equipment parameters, process routes, material attributes, etc. Each submodule is real-valued encoded to form a position vector of the enterprise gene. In the initialization stage, a set of initial position vectors that meet the constraints are generated based on the historical data of the enterprise and expert experience to form an initial particle swarm. For multiple optimization objectives, a weighted sum fitness function with adaptive weights is designed. By dynamically adjusting the weight coefficients of each objective, the search focus changes during the optimization process are adapted. At the same time, a fuzzy measurement mechanism is introduced to perform a fuzzy comprehensive evaluation of the degree of satisfaction of the constraints and the objective function value to obtain a comprehensive fitness value.

[0133] When updating the individual optimal solution and the global optimal solution of particles, a particle selection strategy based on non-dominated sorting is adopted. The particle swarm is non-dominated sorted, and excellent particles are selected as the individual optimal solution and the global optimal solution according to the Pareto level and crowding degree of the particles. An external archive is maintained to store the non-dominated solutions in the optimization process, and merge and update them with the current particle swarm.

[0134] In the mutation operation of the evolution strategy, the mutation step length and mutation probability are adaptively adjusted according to the fitness value of the particle and the iteration process. For particles with higher fitness values, the mutation step length is reduced to conduct local fine search; for particles with lower fitness values, the mutation step length is increased to strengthen global exploration. In the crossover operation, multi-point crossover and adaptive crossover probability are used to dynamically adjust the crossover position and crossover probability according to the similarity of the particles.

[0135] For newly generated particles, a multi-stage constraint processing mechanism is adopted. First, the particles that violate the hard constraints are repaired, and the particle position vector is mapped to the feasible solution space through heuristic rules and other methods. Then, the particles that violate the soft constraints are penalized, and a penalty term is introduced in the fitness value to reduce its priority. Finally, the non-dominated solutions under the constraints are further optimized through local search or regeneration. And the adaptive termination condition is designed, which comprehensively considers the number of iterations, the convergence of the fitness value, and the quality of the non-dominated solution. When the number of iterations reaches the upper limit, or the fitness value converges to a certain threshold, or the quality of the non-dominated solution meets the preset requirements, the optimization process is terminated. At the same time, an early stopping mechanism is introduced. When there is no obvious improvement in the set of non-dominated solutions within a certain number of iterations, the optimization is terminated in advance.

[0136] When outputting optimization results, it provides interactive decision support functions. It visualizes the distribution of non-dominated solution sets and provides interactive screening and comparison tools to help decision makers select satisfactory enterprise gene solutions based on preferences and trade-offs.

[0137] In this embodiment, by constructing an objective function, the three key objectives of production task completion time, production cost and resource utilization are comprehensively considered, and a multi-objective optimization algorithm is used to solve it, so that a high-quality enterprise gene solution that balances multiple objectives can be obtained, providing a variety of options for enterprise decision-making. A multi-swarm particle swarm optimization algorithm is adopted, and strategies such as particle encoding, initialization, fitness calculation, and mutation crossover are used to fully explore the solution space and avoid falling into local optimality. At the same time, an adaptive adjustment mechanism is introduced to improve the convergence speed and global search ability of the algorithm. A fuzzy measurement mechanism is introduced to merge the objective function value and the degree of constraint satisfaction into a comprehensive fitness, avoiding the defects of a single fitness function and more comprehensively evaluating the quality of the solution. The introduction of a local search strategy further optimizes the global optimal solution of the particle swarm, effectively avoiding premature convergence and improving the accuracy of the final solution. Visual display and interactive screening tools are provided to provide decision makers with an intuitive and convenient way to view and compare different solutions, supporting flexible decision-making processes for human-machine collaboration.

[0138] S103. Decode the optimized enterprise gene and convert it into an enterprise production plan, and detect production anomalies in combination with real-time production operation data. If a production anomaly is detected, re-optimize the optimized enterprise gene according to the preset production anomaly handling strategy. At the same time, introduce version management and update mechanism to realize iterative optimization and continuous evolution of enterprise genes.

[0139] In an optional embodiment,

[0140] The optimized enterprise gene is decoded and converted into an enterprise production plan, and combined with real-time production operation data, production anomalies are detected. If a production anomaly is detected, the optimized enterprise gene is re-optimized according to the preset production anomaly handling strategy, and a version management and update mechanism is introduced to achieve iterative optimization and continuous evolution of the enterprise gene, including:

[0141] Decode the optimized enterprise genes, extract key information of the enterprise's production tasks, and use the rule-based production plan generation algorithm to convert the extracted key information into an executable enterprise production plan;

[0142] Collect production operation data in real time and perform preprocessing, build a production anomaly detection model based on a machine learning algorithm, and use the production anomaly detection model to identify production anomalies and corresponding production anomaly types;

[0143] Automatically trigger the production plan re-optimization process according to the production exception type and the preset exception handling strategy, use the heuristic optimization algorithm to generate the optimized production plan, and update the enterprise gene according to the optimized production plan, use the evolutionary computing algorithm to search for the optimal enterprise gene, and record the updated version of the current enterprise gene;

[0144] Utilizing a distributed version control system based on blockchain, an enterprise gene version dependency graph is established for each updated version of the current enterprise gene recorded, and smart contracts are designed to control the access rights and update rules of the enterprise gene version. At the same time, a visual enterprise gene management platform is developed to manage and update the enterprise gene version.

[0145] Illustratively, based on predefined enterprise gene encoding rules, the optimized enterprise genes are decoded to extract key information such as enterprise production processes, resource allocation, and task scheduling; using a rule-based production plan generation algorithm, the extracted key information is converted into an executable production plan to form a multi-granularity production plan at the equipment level, workshop level, and factory level; through production plan simulation and evaluation, the converted production plan is feasibility analyzed and optimized to obtain an optimized enterprise production plan.

[0146] Build a multi-source heterogeneous industrial data collection and transmission network to collect production operation data such as equipment status, product quality, material consumption, etc. in real time; use data cleaning, integration and fusion technology to pre-process the collected production operation data to improve data quality; build a production anomaly detection model based on machine learning algorithms to achieve real-time detection and early warning of abnormal situations such as equipment failures, quality defects, and plan deviations, and form a production anomaly report.

[0147] According to the production exception type and the preset exception handling strategy, the corresponding production plan re-optimization process is automatically triggered; the heuristic optimization algorithm and intelligent scheduling technology are used to minimize the impact of production plan adjustments on production activities while meeting the exception handling requirements; the re-optimized production plan is generated and sent to the production execution layer to guide the dynamic adjustment of production activities.

[0148] According to the results of production plan re-optimization, the enterprise gene coding is updated and optimized accordingly; evolutionary computing algorithms, such as genetic algorithms, particle swarm optimization, etc., are used to search for the optimal enterprise gene coding scheme, and a gene quality assessment mechanism is introduced to eliminate low-quality gene coding, retain high-quality gene coding, and improve the adaptability and stability of enterprise genes.

[0149] A distributed version control system based on blockchain is used to generate a new version for each optimization iteration of the enterprise gene, record the optimization time, update content, operator and other information, and establish an enterprise gene version dependency graph to support version tracking, comparison and rollback. A smart contract is designed to control the access rights and update rules of the enterprise gene version to ensure the security and consistency of the enterprise gene optimization process, and a visual enterprise gene management platform is developed to support version query, comparison, analysis and release functions.

[0150] In this embodiment, the optimized enterprise gene is decoded into an executable production plan using a rule-based algorithm, ensuring the rationality and operability of the production plan. At the same time, combined with artificial expert knowledge, the quality and interpretability of plan generation are improved. Through the real-time analysis of production operation data by machine learning models, various abnormal situations and their types occurring in the production process can be identified in a timely manner, providing key support for the dynamic adjustment of production plans. In combination with the production re-optimization plan, an evolutionary algorithm is used to continuously search for better enterprise genes, realizing the iterative optimization and continuous evolution of enterprise genes, ensuring that enterprises always have high-quality production decision genes. The distributed version control system based on blockchain provides safe and reliable technical support for the version management of enterprise genes, establishes a version dependency diagram, and clearly shows the trajectory of gene evolution. At the same time, the gene access rights and update rules are strictly controlled through smart contracts, realizing the dynamic closed-loop management of enterprise production plans and the continuous optimization of enterprise genes.

[0151] Figure 2 FIG. 1 is a schematic diagram of the structure of the enterprise gene generation system based on enterprise multi-dimensional data according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0152] The first unit is used to collect enterprise multi-source heterogeneous data, wherein the enterprise multi-source heterogeneous data includes structured data and unstructured data, pre-process the enterprise multi-source heterogeneous data to obtain standard enterprise data, construct an enterprise knowledge graph based on the obtained standard enterprise data, extract structured data features and unstructured data features in the enterprise knowledge graph, and perform feature fusion to obtain enterprise multi-dimensional features;

[0153] The second unit is used to input the enterprise multi-dimensional features into the pre-trained enterprise gene generation model to generate initial enterprise genes, build an enterprise gene multi-objective optimization model based on the initial enterprise genes, solve the enterprise gene multi-objective optimization model through a multi-objective optimization algorithm, and obtain an optimized enterprise gene;

[0154] The third unit is used to decode the optimized enterprise gene and convert it into an enterprise production plan, and detect production anomalies in combination with real-time production operation data. If a production anomaly is detected, the optimized enterprise gene is re-optimized according to a preset production anomaly handling strategy. At the same time, a version management and update mechanism is introduced to realize iterative optimization and continuous evolution of the enterprise gene.

[0155] According to a third aspect of the embodiments of the present invention,

[0156] An electronic device is provided, comprising:

[0157] processor;

[0158] a memory for storing processor-executable instructions;

[0159] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0160] According to a fourth aspect of the embodiments of the present invention,

[0161] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0162] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0163] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating enterprise genes based on enterprise multidimensional data, characterized in that: include: Collecting enterprise multi-source heterogeneous data, wherein the enterprise multi-source heterogeneous data includes structured data and unstructured data, preprocessing the enterprise multi-source heterogeneous data to obtain standard enterprise data, constructing an enterprise knowledge graph based on the obtained standard enterprise data, extracting structured data features and unstructured data features in the enterprise knowledge graph, and performing feature fusion to obtain enterprise multi-dimensional features; Inputting the enterprise's multi-dimensional features into a pre-trained enterprise gene generation model to generate an initial enterprise gene, constructing an enterprise gene multi-objective optimization model based on the initial enterprise gene, solving the enterprise gene multi-objective optimization model through a multi-objective optimization algorithm, and obtaining an optimized enterprise gene; Decode the optimized enterprise gene and convert it into an enterprise production plan, and detect production anomalies in combination with real-time production operation data. If a production anomaly is detected, re-optimize the optimized enterprise gene according to a preset production anomaly handling strategy, and introduce version management and update mechanisms to achieve iterative optimization and continuous evolution of the enterprise gene; Based on the obtained standard enterprise data, the enterprise knowledge graph is constructed, the structured data features and unstructured data features in the enterprise knowledge graph are extracted, and the features are fused to obtain the enterprise multi-dimensional features including: According to the obtained enterprise knowledge graph, by constructing positive sample pairs and negative sample pairs, the initial embedding representation of the nodes in the enterprise knowledge graph is trained, and the contrastive learning loss function is introduced. The initial embedding representation of the node is updated by minimizing the contrastive learning loss function to obtain the node embedding vector; Construct a multi-layer graph attention network. In the first layer of the multi-layer graph attention network, the node embedding vector is used as the initial feature representation of the current node, the attention coefficient between the current node and its neighboring nodes is calculated, and the features of all neighboring nodes are weighted summed based on the attention coefficient to obtain the aggregated feature representation of the current node. In each layer of the multi-layer graph attention network, the aggregated feature representations of the nodes output by the previous layer are concatenated, and nonlinear transformation is performed through the feedforward neural network to obtain the target node features, and the target node feature representation is passed to the next layer. When reaching the last layer of the multi-layer graph attention network, the target node features of all nodes are aggregated using a global pooling operation to obtain the feature representation vector of the enterprise knowledge graph; According to the obtained feature representation vector of the enterprise knowledge graph, the multi-layer graph attention network is trained end-to-end using a back-propagation algorithm to obtain a trained multi-layer graph attention network model; The trained multi-layer graph attention network model is used to extract the structured data features and unstructured data features in the enterprise knowledge graph, and then perform feature fusion to obtain the multi-dimensional features of the enterprise.

2. The method according to claim 1, characterized in that Input the enterprise's multi-dimensional features into the pre-trained enterprise gene generation model to generate the initial enterprise genes including: Build a conditional generative adversarial network model that meets production constraints, including a generator and a discriminator; The generator receives random noise and conditional variables as input, generates enterprise genes that meet production constraints through transposed convolution layers and batch normalization layers, and sets a resource constraint layer at the last layer of the generator to map the generated enterprise genes to the resource consumption space, and calculates the difference between the amount of resources consumed by the enterprise genes and the actual amount of resources available to the enterprise as the resource constraint loss; The discriminator receives the enterprise gene generated by the generator as input, generates the discrimination result of the enterprise gene through the convolution layer and the batch normalization layer, and at the same time, sets the task constraint discrimination layer in the last layer of the discriminator to predict the probability that the enterprise gene belongs to the preset production task category, and calculates the task constraint loss based on the prediction result and the real label of the preset production task category; The comprehensive loss function of the conditional generative adversarial network model is constructed based on the generator loss, discriminator loss and constraint loss. The constraint loss includes resource constraint loss and task constraint loss. By minimizing the comprehensive loss function, the parameters of the conditional generative adversarial network model are updated. The iteration is repeated until the preset termination condition is met, and finally a trained enterprise gene generation model is obtained. The extracted multi-dimensional features of the enterprise are mapped into conditional variables using a pre-trained feature encoder, and random noise is sampled from a Gaussian distribution. The obtained conditional variables and random noise are input into the trained enterprise gene generation model to generate the initial enterprise gene.

3. The method according to claim 2, characterized in that The calculation formula for the comprehensive loss function of the conditional generative adversarial network model constructed based on the generator loss, discriminator loss, and constraint loss is as follows: ; Among them, L total represents the resource constraint loss, L G represents the generator loss, L D represents the discriminator loss, m represents the number of available resources for an enterprise, n represents the dimension of resource consumption in the generated enterprise gene, and r ij represents the consumption of the i-th resource in the j-th dimension of the generated enterprise gene, g ij represents the value of the generated enterprise gene in the jth dimension, R i represents the actual available quantity of the i-th resource, k represents the number of preset production task categories, and y l Indicates l The true category labels of the production tasks, Indicates l is the predicted class label probability of the production task, λ1 represents the hyperparameter for balancing the generator loss and the constraint loss, and λ2 represents the hyperparameter for balancing the discriminator loss and the constraint loss.

4. The method according to claim 1, characterized in that: Based on the initial enterprise gene, constructing an enterprise gene multi-objective optimization model includes: Construct objective functions, which include minimizing the completion time of production tasks, minimizing production costs, and maximizing resource utilization; construct constraints, which include the delivery time constraints, cost constraints, and resource constraints of production tasks; Taking the initial enterprise gene as an optimization variable, constructing an enterprise gene multi-objective optimization model according to the objective function and constraint conditions; The calculation formula of the objective function is as follows: ; Where J represents the objective function, α represents the weight coefficient of completion time, A represents the total number of production tasks, C a represents the completion time of production task a, T a represents the delivery time of production task a, T avg represents the average delivery time of all production tasks, β represents the weight coefficient of production cost, C budget represents the total production cost budget, γ represents the weight coefficient of resource utilization, B represents the number of production resource types, R b represents the actual available quantity of production resource b, and R represents the total available quantity of production resource b.

5. The method according to claim 4, characterized in that The enterprise gene multi-objective optimization model is solved by a multi-objective optimization algorithm, and the optimized enterprise genes include: The enterprise gene is divided into multiple submodules, each of which corresponds to a key element of the production plan. Each submodule is encoded, and a set of initial position vectors is generated based on the enterprise's historical data and expert experience to form an initial particle swarm, where each particle represents a potential optimization solution; According to the constructed enterprise gene multi-objective optimization model, the objective function value and the degree of satisfaction of the constraint conditions of each particle are calculated. At the same time, the fuzzy measurement mechanism is introduced to combine the objective function value and the degree of satisfaction of the constraint conditions to obtain the comprehensive fitness value of the particle; Compare the current comprehensive fitness value of the particle with the historical optimal comprehensive fitness value. If the current comprehensive fitness value is greater than the historical optimal comprehensive fitness value, update the individual optimal solution of the corresponding particle and update the global optimal solution of the particle swarm at the same time. Based on the individual optimal solution of each particle and the global optimal solution of the particle swarm, an adaptive mutation and crossover algorithm is used to dynamically adjust the crossover position and crossover probability according to the similarity of the particles to generate new particles; A multi-stage constraint processing mechanism is used for the newly generated particles. Particles that violate hard constraints are repaired and particles that violate soft constraints are punished. The global optimal solution of the particle swarm is further optimized through a local search strategy to obtain a new particle swarm. Repeat the iteration until the adaptive termination condition is met, terminate the optimization process, and output the optimized enterprise gene non-dominated solution set. Based on the obtained enterprise gene non-dominated solution set, it is visualized and interactive screening and comparison tools are provided for decision makers to select satisfactory enterprise gene solutions as the final optimized enterprise genes based on their preferences and trade-offs.

6. The method according to claim 1, characterized in that The optimized enterprise gene is decoded and converted into an enterprise production plan, and combined with real-time production operation data, production anomalies are detected. If a production anomaly is detected, the optimized enterprise gene is re-optimized according to the preset production anomaly handling strategy, and a version management and update mechanism is introduced to achieve iterative optimization and continuous evolution of the enterprise gene, including: Decode the optimized enterprise genes, extract key information of the enterprise's production tasks, and use the rule-based production plan generation algorithm to convert the extracted key information into an executable enterprise production plan; Collect production operation data in real time and perform preprocessing, build a production anomaly detection model based on a machine learning algorithm, and use the production anomaly detection model to identify production anomalies and corresponding production anomaly types; Automatically trigger the production plan re-optimization process according to the production exception type and the preset exception handling strategy, use the heuristic optimization algorithm to generate the optimized production plan, and update the enterprise gene according to the optimized production plan, use the evolutionary computing algorithm to search for the optimal enterprise gene, and record the updated version of the current enterprise gene; Utilizing a distributed version control system based on blockchain, an enterprise gene version dependency graph is established for each updated version of the current enterprise gene recorded, and smart contracts are designed to control the access rights and update rules of the enterprise gene version. At the same time, a visual enterprise gene management platform is developed to manage and update the enterprise gene version.

7. An enterprise gene generation system based on enterprise multidimensional data, used to implement the method described in any one of claims 1 to 6, characterized in that: include: The first unit is used to collect enterprise multi-source heterogeneous data, wherein the enterprise multi-source heterogeneous data includes structured data and unstructured data, pre-process the enterprise multi-source heterogeneous data to obtain standard enterprise data, construct an enterprise knowledge graph based on the obtained standard enterprise data, extract structured data features and unstructured data features in the enterprise knowledge graph, and perform feature fusion to obtain enterprise multi-dimensional features; The second unit is used to input the enterprise multi-dimensional features into the pre-trained enterprise gene generation model to generate initial enterprise genes, build an enterprise gene multi-objective optimization model based on the initial enterprise genes, solve the enterprise gene multi-objective optimization model through a multi-objective optimization algorithm, and obtain an optimized enterprise gene; The third unit is used to decode the optimized enterprise gene and convert it into an enterprise production plan, and detect production anomalies in combination with real-time production operation data. If a production anomaly is detected, the optimized enterprise gene is re-optimized according to a preset production anomaly handling strategy. At the same time, a version management and update mechanism is introduced to realize iterative optimization and continuous evolution of the enterprise gene.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.