Method and system for arranging and constructing industrial chain based on semantics

Abstract

The invention discloses a method and a system for arranging and constructing an industrial chain based on semantics. The method comprises the following steps: collecting enterprise, supply, demand, market and policy data, cleaning and standardizing to generate a semantic map and embedding representation; calculating semantic signal vectors of the candidate links or node clusters; selecting chaotic mapping and setting disturbance intensity according to the signal vector, and generating an initial candidate solution set; coding the chromosomes with topology maintenance, and checking and repairing non-compliant individuals to obtain a feasible solution set; evolutionary optimization is carried out by using a chaos sequence, and an elite solution set is screened according to fitness; updating a semantic coverage file based on the elite solution set; if drifting or stagnation is detected, self-adaptive adjustment and restart are executed, and finally the industry chain topology and the alternative path are output and written back to the atlas. According to the method, semantic map modeling is combined with chaotic mapping and evolutionary optimization, so that intelligent generation, compliance repair and dynamic adaptive optimization of the industry chain topology are realized.

Description

Technical Field

[0001] This invention relates to the field of supply chain construction and optimization, and in particular to a method and system for constructing supply chains based on semantic orchestration. Background Technology

[0002] With the development of global economic integration and regional industrial clustering, the construction and optimization of industrial chains have become important ways to enhance enterprise competitiveness and promote coordinated industrial development. Existing methods for constructing industrial chains mostly rely on manual experience or rule matching based on static databases, which often suffer from problems such as lagging data updates, limited coverage, and difficulty in adapting to complex constraints.

[0003] In recent years, semantic graph and knowledge graph technologies have been increasingly applied to supply chain analysis. By modeling multi-dimensional data such as enterprises, supply, demand, markets, and policies, relationship networks can be formed and semantic queries can be performed. However, the application of existing semantic graph methods in supply chain construction still faces two major challenges: First, the lack of dynamic perception of semantic uncertainty, constraint tension, and external environment drift leads to insufficient robustness of the generated supply chain solutions; second, existing methods often only focus on link matching or local optimization, making it difficult to achieve global topology optimization and multi-objective trade-offs, resulting in insufficient flexibility and sustainability of the supply chain.

[0004] To overcome these shortcomings, academia and industry have proposed some explorations based on intelligent optimization algorithms, such as genetic algorithms and particle swarm optimization. However, these methods are prone to getting trapped in local optima under high-dimensional and complex constraints, and lack consideration for both compliance and diversity within the industrial chain. In particular, traditional methods lack adaptive adjustment mechanisms when facing external policy changes, market fluctuations, and supply and demand structure adjustments.

[0005] Therefore, how to provide a method and system for semantic orchestration and building an industry chain is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a method and system for constructing industry chains based on semantic orchestration. This invention uses semantic graphs and embedded representations to uniformly model enterprise, supply, demand, market, and policy data. It combines chaotic mapping and evolutionary optimization to generate and filter industry chain topology schemes, utilizes a semantic repair mechanism to ensure compliance, and maintains global convergence and diversity through elite solution sets and semantic coverage archives. When external drift or optimization stagnation is detected, adaptive adjustments and local restarts are triggered to achieve continuous optimization of the dynamic environment. This method can efficiently output industry chain topologies and alternative path sets that conform to semantic logic and policy constraints, improving the intelligence and robustness of industry chain construction.

[0007] A method for semantic orchestration and industry chain construction according to an embodiment of the present invention includes the following steps:

[0008] Collect data on enterprises, supply, demand, markets, and policies, clean and standardize them, and generate semantic graphs and their corresponding semantic embedding representations;

[0009] Based on the semantic graph and semantic embedding representation, calculate the semantic signal vector of the candidate link or node cluster;

[0010] Based on the semantic signal vector, a chaotic mapping strategy is selected and the perturbation intensity is set to generate an initial candidate solution set;

[0011] The initial candidate solution set is encoded as a topology-preserving industry chain chromosome. It is then verified based on the semantic arrangement and compliance constraints contained in the semantic graph. Individuals that do not meet the constraints are repaired in situ using a semantic repairer to obtain a feasible candidate solution set.

[0012] The feasible candidate solution set is evolved by using the chaotic mapping strategy and the sequence modulation mutation and selection generated by the perturbation intensity, and evaluated and screened according to the fitness function to obtain the current elite solution set;

[0013] The semantic coverage file is updated and output based on the elite solution set;

[0014] When it is determined that there is external drift based on the semantic signal vector, or when the elite solution set is not improved within a preset number of algebras, adaptive adjustment and local restart are performed, and after the iteration is completed, the target industry chain topology and alternative path set are output, and the running feedback is written into the semantic graph.

[0015] Furthermore, the generation of the semantic embedding representation includes:

[0016] Entity identification and attribute extraction are performed on the collected enterprise data, supply data, demand data, market data, and policy data;

[0017] Based on supply and demand, upstream and downstream relationships, and policy constraints, establish multiple types of relationship edges and construct a semantic graph;

[0018] Duplicate entities in the semantic graph are merged, attribute values ​​are formatted with a uniform time format, unit of measurement and encoding rules, and incomplete attributes are supplemented by data from the same source.

[0019] Based on the semantic graph, a graph neural network model is used to train the nodes and relationships into vectors to obtain low-dimensional embedding vectors. The embedding vectors simultaneously represent the node attribute features and structural relationship features.

[0020] Furthermore, the step of calculating the semantic signal vector includes:

[0021] The entropy value of each candidate link or node cluster is statistically analyzed in the probability distribution of the semantic embedding representation, and the semantic uncertainty is determined by the entropy value.

[0022] Calculate the embedding vector distance between candidate links or node clusters, and normalize the distance to convert it into a similarity score, which is then used as the semantic compatibility.

[0023] The constraints related to qualifications, quotas and time sequence in the semantic graph are numerically processed, the constraint occupancy rate is statistically calculated, and the occupancy rate is used as the constraint tension.

[0024] The semantic embedding distribution generated from current market, supply and demand, and policy data is compared with the historical semantic embedding distribution to measure the difference, and the external drift is obtained based on the result of the difference measurement.

[0025] The semantic uncertainty, semantic compatibility, constraint tension, and external drift are then combined into a semantic signal vector.

[0026] Furthermore, the step of selecting a chaotic mapping strategy includes:

[0027] Receive the semantic signal vector and extract semantic uncertainty, semantic compatibility, constraint tension, and external drift.

[0028] If the semantic uncertainty exceeds the first threshold, or the external drift exceeds the second threshold, then triangulation is used as the mapping strategy.

[0029] If the semantic compatibility is higher than the third threshold, the constraint tension is lower than the fourth threshold, and the external drift is lower than the second threshold, then the Logistic mapping strategy is adopted.

[0030] If none of the above conditions are met, the Chebyshev mapping strategy is adopted.

[0031] Furthermore, the step of generating the initial candidate solution set includes:

[0032] The chaos intensity coefficient is calculated based on the semantic signal vector, including summing the semantic uncertainty, constraint tension and external drift according to preset weights, taking the complement of the semantic compatibility and multiplying it by the corresponding weight, summing and normalizing the result to obtain the chaos intensity coefficient.

[0033] The chaos intensity coefficient is then combined with the selected mapping strategy to generate a perturbation sequence;

[0034] Iterate over the selected mapping to generate a chaotic sequence of length N, and convert the chaotic sequence into a sampling threshold and a perturbation amplitude, respectively;

[0035] Enterprise nodes, resource nodes, and connection relationships are extracted from the node pool and path pool of the semantic graph according to the sampling threshold, and several topology drafts are formed by splicing them together according to the preset origin and destination roles.

[0036] The parameters of the topology draft are perturbed according to the perturbation magnitude, and an initial candidate solution set is obtained and output.

[0037] Furthermore, the step of converting the chaotic sequence into a sampling threshold and a perturbation amplitude includes:

[0038] Generate a chaotic sequence of length N, and normalize each value in the sequence to between zero and one;

[0039] The normalized values ​​are scaled according to the chaos intensity coefficient to obtain the perturbation factor sequence.

[0040] The difference between each term in the perturbation factor sequence and 0.5 is taken as the perturbation amplitude;

[0041] Arrange the values ​​in the perturbation factor sequence in ascending order and select the median as the sampling threshold;

[0042] Using the sampling threshold as a boundary, the node pool and path pool in the semantic graph are extracted to obtain a preliminary topology draft.

[0043] Furthermore, the steps for generating the feasible candidate solution set include:

[0044] The enterprise nodes, resource nodes, demand nodes, and policy nodes included in the initial candidate solution set are numbered according to the supply and demand direction and time sequence relationship;

[0045] The nodes and their connections are encoded into a topology-preserving chromosome structure according to the numbering order, and constraint information is recorded.

[0046] Check whether the chromosome structure violates the constraints in the semantic map, including qualification restrictions, quota restrictions, admission conditions, and timing requirements;

[0047] For individuals that do not meet the constraints, repair is performed sequentially according to the constraint priority. During repair, alternative nodes or connections that meet the constraints are selected from the node pool or path pool.

[0048] The repaired individuals are re-encoded and added to the candidate solution set to form a feasible candidate solution set, which is then output.

[0049] Furthermore, the generation of the elite solution set includes:

[0050] The feasible candidate solution set is copied into the current population, and the number of evolutionary iterations is set;

[0051] In each generation, the sequence generated by the selected chaotic map is invoked to perform mutations on the candidate solutions of the current population. The mutations include node replacements and path adjustments to the topology.

[0052] Perform a crossover operation on the mutated candidate solutions to generate the next generation of candidate solution set;

[0053] For each candidate solution, a fitness value is calculated based on cost, timeliness, risk, and constraint satisfaction.

[0054] Sort the fitness values ​​and select the top-ranked individuals to form the elite solution set.

[0055] The elite solution set is incorporated into the next generation candidate solution set, and the process continues until the preset number of generations is reached, at which point the elite solution set is output.

[0056] Furthermore, the generation of the target industry chain topology and the set of alternative paths includes:

[0057] When the external drift exceeds the threshold or the elite solution set is not improved within a preset algebra, adjust the type of chaotic mapping strategy and reset the perturbation intensity.

[0058] Select a subset of individuals from the relevant subpopulations, generate a new chaotic sequence, reinitialize the individuals, and form a restart solution set;

[0059] The restarted solution set is incorporated into the ongoing evolutionary process, and together with the unrestarted individuals, it enters the next iteration;

[0060] After the iteration is completed, the updated industry chain topology and set of alternative paths are output.

[0061] The output results and feedback information from the iteration process are written into the semantic graph.

[0062] A system for semantic orchestration and building an industry chain according to an embodiment of the present invention is characterized by comprising the following modules:

[0063] The data processing module is used to collect enterprise, supply, demand, market and policy data, perform cleaning and standardization, and generate semantic graphs and their semantic embedding representations;

[0064] The signal calculation module is used to calculate the semantic signal vector of candidate links or node clusters based on semantic graphs and semantic embedding representations.

[0065] The candidate solution generation module is used to select a chaotic mapping strategy and set the perturbation intensity based on the semantic signal vector, generate a chaotic sequence and convert it into a sampling threshold and perturbation amplitude to form an initial candidate solution set;

[0066] The constraint verification module is used to encode the initial candidate solution set into a topology-preserving industry chain chromosome, and to verify it based on semantic arrangement and compliance constraints. Individuals that do not meet the constraints are repaired to obtain a feasible candidate solution set.

[0067] The evolutionary optimization module is used to evolve the feasible candidate solution set using chaotic sequences, and to generate an elite solution set based on the fitness function evaluation and selection.

[0068] The archive maintenance module is used to update semantically covered archives based on the elite solution set;

[0069] The adjustment and output module is used to perform adaptive adjustment and local restart when external drift or elite solution set stagnation is detected, and outputs the target industry chain topology and alternative path set after iteration, while writing the operation feedback into the semantic graph.

[0070] The beneficial effects of this invention are:

[0071] This invention, by introducing a semantic orchestration mechanism and semantic graph embedding representation, can comprehensively characterize the semantic relationships between nodes and links in the industrial chain under a unified modeling of multi-dimensional data such as enterprises, supply, demand, market, and policies. Compared with existing methods based on static rule matching, this invention can dynamically perceive semantic uncertainty, constraint tension, and external environment drift, achieving real-time optimization and adaptive adjustment of the industrial chain topology.

[0072] This invention utilizes a combined control of chaotic mapping strategy and perturbation intensity to introduce a diversity adjustment mechanism during the initial candidate solution generation and evolutionary optimization process, effectively avoiding the problem of traditional optimization algorithms easily getting trapped in local optima. Simultaneously, a semantic repairer is used to in-situ repair candidate solutions that do not meet compliance constraints, ensuring the feasibility and compliance of the industry chain solutions under constraints such as policies, qualifications, and quotas.

[0073] This invention further achieves global convergence and diversity balance in the supply chain optimization process by maintaining elite solution sets and semantic coverage archives. In the event of sudden changes in the external environment or prolonged periods without improvement, it implements adaptive adjustment and local restart mechanisms, enhancing the method's robustness and continuous optimization capabilities in dynamic environments. In summary, this invention can efficiently generate supply chain topologies and alternative path sets that conform to semantic logic and compliance requirements in complex dynamic environments, providing strong technical support for the collaborative construction and optimization of supply chains. Attached Figure Description

[0074] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0075] Figure 1This is a flowchart of a method for constructing an industry chain based on semantic orchestration proposed in this invention;

[0076] Figure 2 This diagram illustrates the strategy determination and restart mechanism of a method for constructing an industry chain based on semantic orchestration proposed in this invention.

[0077] Figure 3 This is a structural diagram of a system based on semantic orchestration and building an industry chain, as proposed in this invention. Detailed Implementation

[0078] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0079] refer to Figure 1-3 A method based on semantic orchestration and industry chain construction includes the following steps:

[0080] Collect data on enterprises, supply, demand, markets, and policies, clean and standardize them, and generate semantic graphs and their corresponding semantic embedding representations;

[0081] Based on the semantic graph and semantic embedding representation, calculate the semantic signal vector of the candidate link or node cluster;

[0082] Based on the semantic signal vector, a chaotic mapping strategy is selected and the perturbation intensity is set to generate an initial candidate solution set;

[0083] The initial candidate solution set is encoded as a topology-preserving industry chain chromosome. It is then verified based on the semantic arrangement and compliance constraints contained in the semantic graph. Individuals that do not meet the constraints are repaired in situ using a semantic repairer to obtain a feasible candidate solution set.

[0084] The feasible candidate solution set is evolved by using the chaotic mapping strategy and the sequence modulation mutation and selection generated by the perturbation intensity, and evaluated and screened according to the fitness function to obtain the current elite solution set;

[0085] The semantic coverage file is updated and output based on the elite solution set;

[0086] When it is determined that there is external drift based on the semantic signal vector, or when the elite solution set is not improved within a preset number of algebras, adaptive adjustment and local restart are performed, and after the iteration is completed, the target industry chain topology and alternative path set are output, and the running feedback is written into the semantic graph.

[0087] In this embodiment, the generation of the semantic embedding representation includes:

[0088] Entity identification and attribute extraction are performed on the collected enterprise data, supply data, demand data, market data, and policy data. Entity identification is accomplished through a combination of rule-based and machine learning methods, extracting key terms from enterprise names, product categories, demand indicators, price information, and policy clauses, specifically including:

[0089] For structured data in tables and databases, directly parse the header fields and primary key fields, such as "Company Name", "Product Code", and "Policy Number", and label them as the corresponding entities;

[0090] For text-based data, a keyword table containing industry names, product categories, demand indicators, and policy terms is pre-built. The text is then matched word by word to extract the entities corresponding to the keyword table.

[0091] In market and policy documents, rules are set to identify information in specific formats. For example, data ending in "yuan / ton" is marked as a price entity, and data ending in "year", "month", or "day" is marked as a time entity.

[0092] The identified company names, product categories, and policy clauses are compared with existing industry classification tables and policy standard entries to confirm the entity categories and standardize their representations;

[0093] The identified synonyms, abbreviations, or acronyms are uniformly mapped to standardized names. For example, "steel enterprise" and "steel company" are merged into the same corporate entity.

[0094] A unique identifier is generated for each entity to ensure that the same entity is not repeated during subsequent graph construction.

[0095] For corporate entities, we extract their industry category, production capacity, and credit rating; for supply entities, we extract the types of raw materials, origin, and transportation methods; for demand entities, we extract the quantity, delivery cycle, and priority; and for policy entities, we extract the applicable regions, restrictive clauses, and incentive conditions.

[0096] Based on supply and demand relationships, upstream and downstream relationships, and policy constraints, multiple types of relationship edges are established and a semantic graph is constructed. For enterprises and supply entities, "supply" relationship edges are established; for enterprises and demand entities, "demand" relationship edges are established; and for enterprises and policy entities, "constraint" or "applicability" relationship edges are established. Relationship edges are recorded in the form of triples, forming a directed graph containing nodes and edges.

[0097] Duplicate entities in the semantic graph are merged, attribute values ​​are formatted with a uniform time format, unit of measurement and encoding rules, incomplete attributes are supplemented by data from the same source, and missing attributes are supplemented by data from the same source enterprise or similar node.

[0098] A graph neural network model is trained based on a standardized semantic graph. The model takes node attribute vectors and adjacency matrix as input and outputs low-dimensional embedding vectors of the nodes. During training, by minimizing graph structure reconstruction error and adjacency prediction error, the node embeddings can simultaneously represent attribute features and structural relationship features. After training, the resulting embedding vectors serve as the semantic embedding representation in this invention, providing input for subsequent semantic signal vector calculations.

[0099] In this embodiment, the step of calculating the semantic signal vector includes:

[0100] The entropy values ​​of each candidate link or node cluster are statistically analyzed in the probability distribution of the semantic embedding representation, and the semantic uncertainty is determined by the entropy values. The embedding vector of each candidate link or node cluster is mapped to a probability density distribution. The entropy value of the distribution is calculated. The magnitude of the entropy value is used as the semantic uncertainty. The higher the entropy value, the greater the uncertainty of the link or node cluster in semantic representation.

[0101] Calculate the embedding vector distance between candidate links or node clusters, and normalize the distance to convert it into a similarity score, which is used as the semantic compatibility. Take out the embedding vectors of two candidate links or node clusters in turn and calculate the cosine distance. Normalize the obtained distance value to obtain a similarity score between zero and one. Use the similarity score as the semantic compatibility to reflect the degree of proximity of different nodes in the semantic space.

[0102] The constraints related to qualifications, quotas, and time series in the semantic graph are numerically processed, and the constraint occupancy rate is calculated and used as the constraint tension. Qualification requirements are converted into binary variables, with one for compliance and zero for non-compliance. Quota conditions are converted into proportional values, and the ratio of current occupancy to total quota is calculated. Time series constraints are converted into time occupancy rates, and the ratio of allocated time to allowed time windows is calculated. The above values ​​are summarized into the constraint occupancy rate, which is used as the constraint tension.

[0103] The semantic embedding distribution generated from current market, supply and demand, and policy data is compared with the historical semantic embedding distribution to measure the difference, and the external drift is obtained based on the difference measurement result. A new semantic embedding distribution is generated using current market, supply and demand, and policy data as input. The semantic embedding distribution obtained from historical data is used as a benchmark. The difference between the current distribution and the historical distribution is compared using a distribution difference measurement method, such as the relative entropy or the maximum mean difference between the distributions. The obtained difference value is used as the external drift.

[0104] The semantic uncertainty, semantic compatibility, constraint tension, and external drift are combined into a semantic signal vector. This semantic signal vector serves as the input for subsequent chaotic mapping strategy selection and evolutionary processes, ensuring the adaptability of supply chain optimization under different semantic environments.

[0105] In this embodiment, the step of selecting a chaotic mapping strategy includes:

[0106] Receive the semantic signal vector and extract semantic uncertainty, semantic compatibility, constraint tension, and external drift.

[0107] If the semantic uncertainty exceeds the first threshold, or the external drift exceeds the second threshold, then triangular mapping is used as the mapping strategy. When the semantic uncertainty is high or the external drift is large, it indicates that the system environment fluctuates significantly, and a single local search is prone to getting stuck in invalid solutions. Therefore, triangular mapping with stronger ergodicity must be selected to maintain the diversity of the solution space.

[0108] If the semantic compatibility is higher than the third threshold, the constraint tension is lower than the fourth threshold, and the external drift is lower than the second threshold, then the Logistic mapping strategy is adopted. When the semantic compatibility is high and the constraint pressure is low, it indicates that the current environment is stable, and the focus of the search is to fine-tune the existing solutions. In this case, using Logistic mapping for local convergence is more effective.

[0109] When none of the above conditions are met, the Chebyshev mapping strategy is adopted. When the environment is in an intermediate state and the first two extreme conditions are not met, the Chebyshev mapping strategy is adopted to balance exploration and convergence.

[0110] In this embodiment, the step of generating the initial candidate solution set includes:

[0111] The chaos intensity coefficient is calculated based on the semantic signal vector, including summing the semantic uncertainty, constraint tension and external drift according to preset weights, taking the complement of the semantic compatibility and multiplying it by the corresponding weight, summing and normalizing the result to obtain the chaos intensity coefficient.

[0112] The chaotic intensity coefficient is then combined with the selected mapping strategy to generate a perturbation sequence. When the chaotic intensity coefficient is large, the perturbation range is increased and the iteration step size is increased. When the chaotic intensity coefficient is small, the perturbation range is reduced and the iteration step size is decreased.

[0113] Iterate over the selected mapping to generate a chaotic sequence of length N, and convert the chaotic sequence into a sampling threshold and a perturbation amplitude, respectively;

[0114] Enterprise nodes, resource nodes, and connection relationships are extracted from the node pool and path pool of the semantic graph according to the sampling threshold, and several topology drafts are formed by splicing them together according to the preset origin and destination roles.

[0115] The parameters of the topology draft are perturbed according to the perturbation magnitude, and an initial candidate solution set is obtained and output.

[0116] In this embodiment, the step of converting the chaotic sequence into a sampling threshold and a perturbation amplitude includes:

[0117] Generate a chaotic sequence of length N, and normalize each value in the sequence to between zero and one;

[0118] The normalized values ​​are scaled according to the chaos intensity coefficient to obtain the perturbation factor sequence.

[0119] The difference between each term in the perturbation factor sequence and 0.5 is taken as the perturbation amplitude;

[0120] Arrange the values ​​in the perturbation factor sequence in ascending order and select the median as the sampling threshold;

[0121] Using the sampling threshold as a boundary, the node pool and path pool in the semantic graph are extracted to obtain a preliminary topology draft. Enterprise entities are extracted from the enterprise node pool; raw material or component entities are extracted from the resource node pool; demand entities are extracted from the demand node pool; and relevant constraint entities are extracted from the policy node pool. Several topology drafts are then formed by splicing them together according to preset starting and ending roles.

[0122] In this embodiment, the steps for generating the feasible candidate solution set include:

[0123] The enterprise nodes, resource nodes, demand nodes, and policy nodes included in the initial candidate solution set are numbered according to the supply and demand direction and time sequence relationship;

[0124] The nodes and their connections are encoded into a topology-preserving chromosome structure according to the numbering order, and constraint information is recorded.

[0125] Check whether the chromosome structure violates the constraints in the semantic map, including qualification restrictions, quota restrictions, admission conditions, and timing requirements;

[0126] For individuals that do not meet the constraints, repair is performed sequentially according to the constraint priority. During repair, alternative nodes or connections that meet the constraints are selected from the node pool or path pool; each item is checked in the order of priority of qualification, quota, admission and time sequence; alternative nodes or connections that meet the current constraints are searched in the node pool or path pool; and the replacement result is replaced to the corresponding position of the original chromosome structure.

[0127] The repaired individuals are re-encoded and added to the candidate solution set to form a feasible candidate solution set, which is then output.

[0128] In this embodiment, the generation of the elite solution set includes:

[0129] The feasible candidate solution set is copied into the current population, and the number of evolutionary iterations is set;

[0130] In each generation, the sequence generated by the selected chaotic map is invoked to perform mutation on the candidate solutions of the current population. The mutation includes node replacement and path adjustment of the topology. When replacing a node, the position of a node in the candidate solution is first determined according to the chaotic sequence, the role type of the node is identified, and a set of replacement nodes matching the role type is searched in the semantic graph. Nodes that do not meet the qualification, quota, admission, or timing requirements are excluded. A replacement node is selected from the set to replace the original node, maintaining the connection relationship with the nodes before and after. The local structure after replacement is constrained and verified. If a conflict is detected, the next candidate node is tried. If none of them meet the requirements, the replacement operation is rolled back. When adjusting a path, a connection relationship in the candidate solution is located according to the chaotic sequence. Other feasible adjacent nodes of the starting node are searched. Nodes that do not meet the semantic constraints are excluded. A new node is selected to replace the original connection relationship to form a new path segment. The adjusted path is subjected to loop detection and duplication check. If a problem is found, the conflicting part is deleted and a new path is selected. Compliance verification is performed on the final structure. If the requirements are not met, the adjustment is rolled back.

[0131] Perform a crossover operation on the mutated candidate solutions to generate the next generation of candidate solutions. In the crossover operation, select two parent candidate solutions from the population, determine a split position, divide the two candidate solutions into a front segment and a back segment, concatenate the front segment of one parent with the back segment of the other parent to generate a new offspring, and concatenate in reverse to generate another offspring. Check the offspring structure. If there are duplicate nodes or conflicts, delete or replace them with feasible nodes of the same type in turn. If necessary, adjust the path to ensure connectivity from the starting node to the ending node. Finally, perform compliance verification. If it cannot be repaired, discard the offspring and keep the parent to enter the next generation.

[0132] For each candidate solution, a fitness value is calculated based on cost, timeliness, risk, and constraint satisfaction.

[0133] Sort the fitness values ​​and select the top-ranked individuals to form the elite solution set.

[0134] The elite solution set is incorporated into the next generation candidate solution set, and the process continues until the preset number of generations is reached, at which point the elite solution set is output.

[0135] In this implementation, the fitness value of each candidate solution is obtained by combining four indicators: cost, timeliness, risk, and constraint satisfaction. Each indicator is first calculated with a uniform standard, then normalized and directionally aligned, and finally aggregated into a single score according to a preset weight. A decision order is set for cases with equal weights.

[0136] The cost index is based on the total cost of candidate solutions, covering components such as raw material procurement, production and processing, work-in-process and finished goods inventory, warehousing and loading / unloading, trunk and branch line transportation, customs and taxes. For sub-items with different unit definitions, a unified unit of measurement and currency conversion are performed to obtain the total cost per unit output. To eliminate dimensional differences between different generations, extreme value normalization is performed using the minimum and maximum values ​​within the current evaluation window as boundaries, and the trend of "lower cost, higher score" is reversed to obtain a dimensionless cost score.

[0137] The timeliness index is based on the end-to-end cycle from the upstream starting point to the terminal node, accumulating the procurement cycle, production cycle time, transit time, and planned waiting time in sequence according to the process flow, with the critical path used as the baseline duration for the candidate solution. For cases with parallel segments, the longest of the parallel segments is included in the critical path. After calculation, extreme value normalization is performed within the same evaluation window, and the principle of "shorter cycle, higher score" is reversed to obtain the timeliness score.

[0138] The risk indicator collects stability-related data from both the node and route levels, including historical default records and frequency of quality anomalies at supply nodes, credit ratings and financial health of enterprises, probability and volatility of transportation route disruptions, and compliance sensitivity of policy-related nodes. Each item is weighted according to its source reliability and impact scope and then aggregated into a risk exposure value. To align with other indicators, the risk exposure value is normalized to a range and then inverted to obtain a risk score, where lower risk corresponds to a higher score.

[0139] Constraint satisfaction is determined based on hard and soft constraints in the knowledge graph. Hard constraints include qualification requirements, access lists, and prohibited pairings. If any hard constraint is not satisfied, the candidate solution is marked as infeasible and directly enters the repair process or is eliminated in this round of evaluation. Soft constraints include quota usage and time window tension. Each constraint item is assessed for satisfaction, the proportion of satisfied items is calculated, and severity coefficients are assigned to different items to obtain the overall satisfaction rate. The overall satisfaction rate, after normalization, becomes the constraint satisfaction score.

[0140] After obtaining four dimensionless scores, a single fitness value is generated based on pre-configured weight coefficients. The weights remain constant throughout the run and are set by the business scenario before execution. To avoid a single indicator dominating the results, interval pruning and outlier correction are performed on the four scores before fitness value generation to ensure that extreme values ​​are not unreasonably amplified. For tied candidate solutions, the order is determined by comparing the constraint satisfaction score, cost score, timeliness score, and risk score in that order; if still tied, the candidate solution with the larger semantic gap in the semantic coverage file is given priority. The fitness value, along with intermediate process data, is recorded after calculation and used as input for subsequent selection of elite solutions and evolutionary iterations.

[0141] In this implementation, when the external drift exceeds a threshold, or when the elite solution set fails to improve over several consecutive generations, the mapping strategy is modified and a local restart phase begins. The modification of the mapping strategy is based on the components of the semantic signal vector: when the semantic uncertainty continues to increase while the compatibility decreases, it indicates that the distribution of candidate solutions in the semantic space is diverging. In this case, the mapping strategy is switched to triangular mapping to enhance the leap of the search. When the external drift is large and the constraint tension is at a medium to high level, it indicates that environmental changes have caused the original path constraints to fail. In this case, Chebyshev mapping is used to expand the perturbation range and add new candidate paths. When the compatibility is high and the uncertainty remains in a low range, it indicates that the candidate solutions tend to be concentrated and the constraints are easy to satisfy. In this case, the Logistic mapping is maintained or switched to maintain stable convergence. The perturbation intensity is adjusted by weighting the semantic uncertainty, constraint tension, and external drift. If the combined value of the three is higher than the baseline level, the perturbation amplitude is increased and the iteration step size is increased to make a significant adjustment to the population structure. If the combined value is lower than the baseline level, the perturbation amplitude is reduced and the step size is shortened to reduce the destruction of existing high-quality solutions.

[0142] After revising the mapping strategy and setting the perturbation intensity, a subset of individuals is randomly selected from the existing subpopulation to perform a local restart. The restart operation regenerates the chaotic sequence and constructs new candidate solutions using updated sampling thresholds and perturbation amplitudes. This ensures that these individuals are structurally consistent with the original solutions, but introduce new differences in node combinations or path parameters, thus supplementing the population's diversity. After the local restart is complete, the restarted individuals are sent to the next generation of evolution along with the unrestarted individuals.

[0143] When the evolutionary iteration reaches the termination condition, the optimal solution with stable structure and satisfying semantic constraints is selected from the final population as the target industry chain topology. Simultaneously, multiple solutions with similar fitness values ​​are retained as a candidate path set. While outputting the topology and path set, the semantic signal vectors, mapping strategy adjustment records, constraint verification status, and iteration trajectory recorded during the process are written into the semantic graph to update the semantic embedding representation and signal vectors. This allows the system to regenerate and evolve candidate solutions based on the new semantic environment in the next execution.

[0144] A system based on semantic orchestration and building an industry chain includes the following modules:

[0145] The data processing module is used to collect enterprise, supply, demand, market and policy data, perform cleaning and standardization, and generate semantic graphs and their semantic embedding representations;

[0146] The signal calculation module is used to calculate the semantic signal vector of candidate links or node clusters based on semantic graphs and semantic embedding representations.

[0147] The candidate solution generation module is used to select a chaotic mapping strategy and set the perturbation intensity based on the semantic signal vector, generate a chaotic sequence and convert it into a sampling threshold and perturbation amplitude to form an initial candidate solution set;

[0148] The constraint verification module is used to encode the initial candidate solution set into a topology-preserving industry chain chromosome, and to verify it based on semantic arrangement and compliance constraints. Individuals that do not meet the constraints are repaired to obtain a feasible candidate solution set.

[0149] The evolutionary optimization module is used to evolve the feasible candidate solution set using chaotic sequences, and to generate an elite solution set based on the fitness function evaluation and selection.

[0150] The archive maintenance module is used to update semantically covered archives based on the elite solution set;

[0151] The adjustment and output module is used to perform adaptive adjustment and local restart when external drift or elite solution set stagnation is detected, and outputs the target industry chain topology and alternative path set after iteration, while writing the operation feedback into the semantic graph.

[0152] Example 1:

[0153] To verify the feasibility of this invention in practice, it was applied to the construction and optimization of the new energy vehicle power battery (LFP cathode material) industry chain in a certain East China region. The original data covers enterprise, supply, demand, market, and policy information from the second quarter of 2025, involving 4 upstream suppliers of iron ore and lithium salts, 2 cathode material manufacturers, 2 regional distributors, and 2 local energy consumption and environmental quota policies. The following presents a directly implementable process and results.

[0154] The system collects data on the supply side, including lithium carbonate and iron ore prices, monthly availability, quality inspection pass rates, and historical default rates; on the manufacturing side, it collects capacity, yield, and unit energy consumption; on the demand side, it collects monthly order volume and delivery windows from OEMs; on the market side, it collects trunk / branch line freight rates and delivery times; and on the policy side, it collects energy intensity and emission quota thresholds. Nodes are categorized into four types: "enterprise—resource—demand—policy," and three types of relationship edges are established: "supply," "demand," and "applicable / constraint," with unified currency, units of measurement, and timeframes. After cleaning and merging of identical nodes, the graph model is trained to obtain semantic embedding vectors for each node, embedded in a 64-dimensional space, while adjacency information is saved for subsequent constraint verification.

[0155] Taking the candidate link clusters of "upstream → manufacturing → distribution → OEM" as the object, the entropy value of the embedding distribution is statistically analyzed to determine the semantic uncertainty (the average value of this batch of candidates is 0.71); the embedding similarity between each pair of link clusters is calculated and normalized to obtain the semantic compatibility (average 0.66); qualifications, quotas, and time series are converted into numerical occupancy rates to obtain the constraint tension (average 0.58); with 2025Q1 as the historical benchmark, the difference in the current embedding distribution is compared to obtain the external drift (0.27).

[0156] Thresholds were set as follows: uncertainty threshold 0.65, drift threshold 0.20, compatibility threshold 0.70, and stress threshold 0.50. Since both uncertainty and drift exceeded the thresholds simultaneously, a triangular mapping was chosen as the initial strategy. The system was weighted positively according to "high uncertainty, stress, and drift," and the "compatibility" threshold was weighted with its complement value, resulting in a normalized chaos intensity coefficient of 0.68. Based on this, the perturbation amplitude was relaxed and the iteration step size was increased to prepare for generating the initial candidate solution set.

[0157] Iterative mapping yields a chaotic sequence of length N=100, which is then converted to a sampling threshold of 0.53 and an average perturbation amplitude of 0.12. Five draft topologies are extracted from the node pool and path pool using these thresholds and concatenated. Each draft undergoes sequential verification based on qualifications, quotas, access requirements, and timing. Two drafts were found to trigger "distribution qualification deficiency" and "policy quota overflow," respectively. After replacing the distribution node and reselecting the upstream supply path, the two drafts passed the verification, ultimately resulting in four feasible initial candidates.

[0158] Four candidate replicas were used as the current population, with 12 generations of iteration. Each generation used a chaotic sequence to drive node replacement and path adjustment, and performed crossover. Fitness was calculated by summing the normalized scores of cost, timeliness, risk, and constraint satisfaction according to weights (0.35 / 0.25 / 0.25 / 0.15); in case of a tie, the order was "constraint → cost → timeliness → risk". In the 7th generation, an increase in trunk line freight rates and tightening local energy consumption policies were detected, causing the external drift to briefly rise to 0.33 and the stress level to 0.61. Based on the correction rules, the mapping was switched to a Chebyshev mapping, and 30% of the subpopulation was restarted, continuing evolution until convergence.

[0159] Table 1 Key Indicators of Candidate Solutions in the Examples

[0160]

[0161]

[0162] As shown in Table 1, Scheme A, after convergence, becomes the final output path. Its cost decreases by approximately 7.1% (9.8 million → 9.1 million RMB / batch), the cycle is shortened by 15.4% (26 → 22 days), the risk is reduced by 18.4% (0.38 → 0.31), the constraint satisfaction increases from 0.86 to 0.93, and the fitness increases from 0.62 to 0.78. This improvement mainly comes from two structural adjustments: the upstream switch from S3 to S2 to avoid implicit delays caused by tight energy consumption quotas; the distribution switch from D2 to D1 to reduce timeliness fluctuations caused by trunk line congestion, and policy node adaptation ensures that the energy consumption red line is no longer triggered. When external drift intensifies in the 7th generation, this invention switches the mapping to Chebyshev mapping according to the correction rules and performs a partial restart. In the subsequent 5 generations, diversified path segments are quickly introduced, enabling Scheme A to achieve simultaneous improvement in cost and timeliness without compromising compliance. Alternative options B and C have convergence metrics similar to A, but are slightly inferior in terms of timeliness or constraint satisfaction. They are included in the alternative path set as emergency replacement paths to enhance resilience. Overall, semantic repair ensures that structural adjustments do not exceed the bottom lines of qualifications, quotas, access, and timing. Semantic coverage archives suppress premature maturation, and the dynamic correction and partial restart of the chaos strategy effectively address the phased drift of policies and freight rates.

[0163] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for constructing an industry chain based on semantic orchestration, characterized in that, Includes the following steps: Collect data on enterprises, supply, demand, markets, and policies, clean and standardize them, and generate semantic graphs and their corresponding semantic embedding representations; Based on the semantic graph and semantic embedding representation, calculate the semantic signal vector of the candidate link or node cluster; Based on the semantic signal vector, a chaotic mapping strategy is selected and the perturbation intensity is set to generate an initial candidate solution set; The initial candidate solution set is encoded as a topology-preserving industry chain chromosome. It is then verified based on the semantic arrangement and compliance constraints contained in the semantic graph. Individuals that do not meet the constraints are repaired in situ using a semantic repairer to obtain a feasible candidate solution set. The feasible candidate solution set is evolved by using the chaotic mapping strategy and the sequence modulation mutation and selection generated by the perturbation intensity, and evaluated and screened according to the fitness function to obtain the current elite solution set; The semantic coverage file is updated and output based on the elite solution set; When it is determined that there is external drift based on the semantic signal vector, or when the elite solution set is not improved within a preset number of algebras, adaptive adjustment and local restart are performed, and after the iteration is completed, the target industry chain topology and alternative path set are output, and the running feedback is written into the semantic graph.

2. The method for semantic orchestration and industry chain construction according to claim 1, characterized in that, The generation of the semantic embedding representation includes: Entity identification and attribute extraction are performed on the collected enterprise data, supply data, demand data, market data, and policy data; Based on supply and demand, upstream and downstream relationships, and policy constraints, establish multiple types of relationship edges and construct a semantic graph; Duplicate entities in the semantic graph are merged, attribute values ​​are formatted with a uniform time format, unit of measurement and encoding rules, and incomplete attributes are supplemented by data from the same source. Based on the semantic graph, a graph neural network model is used to train the nodes and relationships into vectors to obtain low-dimensional embedding vectors. The embedding vectors simultaneously represent the node attribute features and structural relationship features.

3. The method for semantic orchestration and industry chain construction according to claim 1, characterized in that, The steps for calculating the semantic signal vector include: The entropy value of each candidate link or node cluster is statistically analyzed in the probability distribution of the semantic embedding representation, and the semantic uncertainty is determined by the entropy value. Calculate the embedding vector distance between candidate links or node clusters, and normalize the distance to convert it into a similarity score, which is then used as the semantic compatibility. The constraints related to qualifications, quotas and time sequence in the semantic graph are numerically processed, the constraint occupancy rate is statistically calculated, and the occupancy rate is used as the constraint tension. The semantic embedding distribution generated from current market, supply and demand, and policy data is compared with the historical semantic embedding distribution to measure the difference, and the external drift is obtained based on the result of the difference measurement. The semantic uncertainty, semantic compatibility, constraint tension, and external drift are then combined into a semantic signal vector.

4. The method for semantic orchestration and industry chain construction according to claim 1, characterized in that, The steps for selecting a chaotic mapping strategy include: Receive the semantic signal vector and extract semantic uncertainty, semantic compatibility, constraint tension, and external drift. If the semantic uncertainty exceeds the first threshold, or the external drift exceeds the second threshold, then triangulation is used as the mapping strategy. If the semantic compatibility is higher than the third threshold, the constraint tension is lower than the fourth threshold, and the external drift is lower than the second threshold, then the Logistic mapping strategy is adopted. If none of the above conditions are met, the Chebyshev mapping strategy is adopted.

5. The method for semantic orchestration and industry chain construction according to claim 1, characterized in that, The step of generating the initial candidate solution set includes: The chaos intensity coefficient is calculated based on the semantic signal vector, including summing the semantic uncertainty, constraint tension and external drift according to preset weights, taking the complement of the semantic compatibility and multiplying it by the corresponding weight, summing and normalizing the result to obtain the chaos intensity coefficient. The chaos intensity coefficient is then combined with the selected mapping strategy to generate a perturbation sequence; Iterate over the selected mapping to generate a chaotic sequence of length N, and convert the chaotic sequence into a sampling threshold and a perturbation amplitude, respectively; Enterprise nodes, resource nodes, and connection relationships are extracted from the node pool and path pool of the semantic graph according to the sampling threshold, and several topology drafts are formed by splicing them together according to the preset origin and destination roles. The parameters of the topology draft are perturbed according to the perturbation magnitude, and an initial candidate solution set is obtained and output.

6. The method for semantic orchestration and industry chain construction according to claim 5, characterized in that, The step of converting the chaotic sequence into a sampling threshold and a perturbation amplitude includes: Generate a chaotic sequence of length N, and normalize each value in the sequence to between zero and one; The normalized values ​​are scaled according to the chaos intensity coefficient to obtain the perturbation factor sequence. The difference between each term in the perturbation factor sequence and 0.5 is taken as the perturbation amplitude; Arrange the values ​​in the perturbation factor sequence in ascending order and select the median as the sampling threshold; Using the sampling threshold as a boundary, the node pool and path pool in the semantic graph are extracted to obtain a preliminary topology draft.

7. The method for semantic orchestration and industry chain construction according to claim 1, characterized in that, The steps for generating a feasible candidate solution set include: The enterprise nodes, resource nodes, demand nodes, and policy nodes included in the initial candidate solution set are numbered according to the supply and demand direction and time sequence relationship; The nodes and their connections are encoded into a topology-preserving chromosome structure according to the numbering order, and constraint information is recorded. Check whether the chromosome structure violates the constraints in the semantic map, including qualification restrictions, quota restrictions, admission conditions, and timing requirements; For individuals that do not meet the constraints, repair is performed sequentially according to the constraint priority. During repair, alternative nodes or connections that meet the constraints are selected from the node pool or path pool. The repaired individuals are re-encoded and added to the candidate solution set to form a feasible candidate solution set, which is then output.

8. The method for semantic orchestration and industry chain construction according to claim 1, characterized in that, The generation of the elite solution set includes: The feasible candidate solution set is copied into the current population, and the number of evolutionary iterations is set; In each generation, the sequence generated by the selected chaotic map is invoked to perform mutations on the candidate solutions of the current population. The mutations include node replacements and path adjustments to the topology. Perform a crossover operation on the mutated candidate solutions to generate the next generation of candidate solution set; For each candidate solution, a fitness value is calculated based on cost, timeliness, risk, and constraint satisfaction. Sort the fitness values ​​and select the top-ranked individuals to form the elite solution set. The elite solution set is incorporated into the next generation candidate solution set, and the process continues until the preset number of generations is reached, at which point the elite solution set is output.

9. The method for semantic orchestration and industry chain construction according to claim 1, characterized in that, The generation of the target industry chain topology and the set of alternative paths includes: When the external drift exceeds the threshold or the elite solution set is not improved within a preset algebra, adjust the type of chaotic mapping strategy and reset the perturbation intensity. Select a subset of individuals from the relevant subpopulations, generate a new chaotic sequence, reinitialize the individuals, and form a restart solution set; The restarted solution set is incorporated into the ongoing evolutionary process, and together with the unrestarted individuals, it enters the next iteration; After the iteration is completed, the updated industry chain topology and set of alternative paths are output. The output results and feedback information from the iteration process are written into the semantic graph.

10. A system for semantic orchestration and industry chain construction, comprising the method for semantic orchestration and industry chain construction as described in any one of claims 1 to 9, characterized in that, Includes the following modules: The data processing module is used to collect enterprise, supply, demand, market and policy data, perform cleaning and standardization, and generate semantic graphs and their semantic embedding representations; The signal calculation module is used to calculate the semantic signal vector of candidate links or node clusters based on semantic graphs and semantic embedding representations. The candidate solution generation module is used to select a chaotic mapping strategy and set the perturbation intensity based on the semantic signal vector, generate a chaotic sequence and convert it into a sampling threshold and perturbation amplitude to form an initial candidate solution set; The constraint verification module is used to encode the initial candidate solution set into a topology-preserving industry chain chromosome, and to verify it based on semantic arrangement and compliance constraints. Individuals that do not meet the constraints are repaired to obtain a feasible candidate solution set. The evolutionary optimization module is used to evolve the feasible candidate solution set using chaotic sequences, and to generate an elite solution set based on the fitness function evaluation and selection. The archive maintenance module is used to update semantically covered archives based on the elite solution set; The adjustment and output module is used to perform adaptive adjustment and local restart when external drift or elite solution set stagnation is detected, and outputs the target industry chain topology and alternative path set after iteration, while writing the operation feedback into the semantic graph.

Images