A data governance platform for smart park construction

Abstract

The application discloses a data management platform for smart park construction and belongs to the technical field of smart park management, and specifically comprises the following steps: collecting resource demand data and resource supply data of each enterprise in the park, performing demand feature extraction operation on the collected resource demand data to generate a structured demand description, performing supply feature extraction operation on the resource supply data to generate a structured supply description, calculating the matching degree between demand and supply based on the structured demand description and the structured supply description and generating an initial matching scheme, performing constraint condition checking operation on the initial matching scheme according to the park operation strategy to identify matching pairs that violate the constraint conditions, performing scheme adjustment operation on the matching pairs that violate the constraint conditions to generate an optimized matching scheme, and finally scheduling the park resources to complete resource data sharing between enterprises; and the application realizes intelligent management and efficient sharing of resource data in the park, and improves the overall resource utilization efficiency of the park.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart park management, and particularly relates to a data governance platform for smart park construction. BACKGROUND

[0002] With the continuous advancement of smart city construction, as the core carrier of the integration of industrial agglomeration and urban functions, smart parks have become a key scenario for improving regional economic vitality and operational efficiency. At present, the number of enterprises in the park is increasing, and the demand for resource interaction between enterprises, such as equipment rental, technical cooperation, and service sharing, is increasing. At the same time, the digital transformation of the park has led to an explosive growth of multi-source heterogeneous data such as equipment operation, service load, and enterprise demand. How to achieve precise matching of resource supply and demand through data integration and efficient use has become a core problem that needs to be solved in smart park construction, and is also an important direction for promoting green and low-carbon operation of the park and reducing enterprise operating costs.

[0003] At present, some data management related technologies have appeared in the field of smart parks. Some systems can realize the basic collection and storage of enterprise resource data. Some simple resource matching tools try to connect resource demanders and suppliers through data classification and keyword retrieval, breaking down the information barriers between enterprises to a certain extent, providing preliminary support for resource sharing, and achieving a small increase in resource utilization efficiency in specific scenarios.

[0004] However, the existing technology mainly stays in the surface classification of resource supply and demand data, and fails to deeply mine the key information such as space-time characteristics and specification parameters in the data, resulting in low matching accuracy of supply and demand, and easy occurrence of resource type mismatch or time window conflict. In the matching process, the overall operation strategy of the park is not fully integrated, and the constraints such as resource allocation priority and sharing restrictions are not considered, which may cause contradictions such as local resource overuse and imbalance of overall operation order. SUMMARY

[0005] The purpose of the present application is to provide a data governance platform for smart park construction, which solves the following technical problems:

[0006] The existing technology mainly stays in the surface classification of resource supply and demand data, and fails to deeply mine the characteristics of resource supply and demand data, and only matches based on surface classification. In the matching process, the park operation strategy is not effectively integrated, and the dynamic consideration of constraints such as resource allocation priority and sharing restrictions is lacking.

[0007] The purpose of the present application can be achieved by the following technical solutions:

[0008] A data governance platform for smart park construction, comprising:

[0009] A data collection module is configured to collect resource demand data and resource supply data of enterprises in the park, wherein the resource demand data comprises equipment usage demand and service cooperation demand, and the resource supply data comprises idle equipment information and idle service capability information;

[0010] A feature structuring processing module is configured to perform demand feature extraction on the resource demand data to generate a structured demand description, and perform supply feature extraction on the resource supply data to generate a structured supply description;

[0011] A multi-modal matching calculation module is configured to calculate a matching degree between the demand and the supply based on the structured demand description and the structured supply description, and generate an initial matching scheme;

[0012] A multi-level constraint checking module is configured to perform constraint condition checking on the initial matching scheme according to a park operation strategy, and identify matching pairs that violate the constraint condition;

[0013] A resource scheduling execution module is configured to perform scheme adjustment on the matching pairs that violate the constraint condition, generate an optimized matching scheme, and schedule park resources to complete resource data sharing between enterprises.

[0014] As a further scheme of the present application, in the data collection module, the specific process of collecting the resource demand data and the resource supply data is as follows:

[0015] Through a park data interface, operation data of equipment management subsystems and service management subsystems of the enterprises are synchronized, equipment usage plans containing time intervals and performance requirements are extracted from the equipment management subsystems, and service requests containing service types and quality requirements are obtained from the service management subsystems;

[0016] Meanwhile, enterprise resource states are monitored, idle equipment is identified and its available time window is recorded, service load states are analyzed and idle service capability is counted, and the equipment usage plans and the service requests are integrated into a resource demand data set, and the idle equipment information and the idle service capability are integrated into a resource supply data set.

[0017] As a further scheme of the present application, in the feature structuring processing module, the specific process of the demand feature extraction and the supply feature extraction is as follows:

[0018] A dynamic demand vector containing space-time constraint conditions is constructed, the dynamic demand vector fuses resource type features, time window features and geographic location features, a supply description vector with capacity elasticity is generated, and the supply description vector contains resource specification parameters, available time patterns and location coordinate information;

[0019] The spatio-temporal dimension decomposition is performed on the dynamic demand vector, and three components of a resource type dimension, a time dimension and a space dimension are separated; the multi-scale feature extraction is performed on the supply description vector, and resource specification features, time availability features and location accessibility features are obtained;

[0020] The cross-modal association model of the demand vector and the supply vector is established, the association weights of the demand dimensions and the supply features are calculated through the attention mechanism, the joint feature representation of the demand and the supply is generated based on the association weights, and the structured demand description and the structured supply description are formed.

[0021] As a further scheme of the application, in the multi-modal matching calculation module, the specific process of calculating the matching degree is:

[0022] The multi-modal matching model is constructed, the multi-modal matching model simultaneously processes the resource specification features, the time availability features and the location accessibility features; the structured demand description is input into the query encoder of the multi-modal matching model to generate a query representation; and the structured supply description is input into the key-value encoder of the multi-modal matching model to generate a key-value representation.

[0023] The interaction weights of the query representation and the key-value representation in each feature dimension are calculated through the cross-attention mechanism, the resource specification matching degree, the time availability matching degree and the location accessibility matching degree are fused based on the interaction weights, the contribution proportion of each matching degree component is dynamically adjusted by using the gated fusion network, and the final matching degree score of the demand and the supply is output according to the fusion result.

[0024] As a further scheme of the application, in the multi-modal matching calculation module, the specific process of generating the initial matching scheme is:

[0025] The descending order sorting operation is performed on the matching degree score to form a candidate matching pair sequence, the effective matching pairs in the candidate matching pair sequence are screened by setting a matching degree threshold, and the demands below the matching degree threshold are marked as unmatching states;

[0026] The time windows of each effective matching pair are checked one by one, the time arrangements with overlap are identified, the resource allocation of each effective matching pair is analyzed, the difference between the total resource demand and the available amount is calculated, for the matching pairs with time overlap, the start time or the end time of the time window is adjusted, for the matching pairs with resource demand exceeding the available amount, the supplier is reselected or the resource amount is segmented, the adjusted matching pair information is recorded, and the initial matching scheme containing all the effective matching pairs is generated.

[0027] As a further scheme of the application, in the multi-level constraint verification module, the specific process of the constraint condition verification operation is:

[0028] The multi-level check engine includes a rule analysis layer, a condition matching layer and a conflict detection layer, constraint conditions in the park operation strategy are analyzed in the rule analysis layer, and an executable check rule set is generated, matching pairs in an initial matching scheme are matched with the check rule set in the condition matching layer, and matching pairs satisfying the constraint conditions are identified;

[0029] In the conflict detection layer, the mutual influence relationship between the matching pairs is analyzed, the matching pairs with resource conflicts or time conflicts are detected, a constraint violation degree evaluation model is established, the constraint violation degree evaluation model calculates a violation degree score based on the severity and influence range of the violated rules, the matching pairs violating the constraint conditions are marked according to the violation degree score, and a constraint check report containing the marking result is generated.

[0030] As a further scheme of the application, the specific working process of the rule analysis layer is:

[0031] The park operation strategy document is loaded, the park operation strategy document contains resource allocation rules and sharing limit conditions in a text format, semantic analysis technology is used to extract key constraint elements in the strategy document, the key constraint elements include constraint subjects, constraint conditions and constraint strengths;

[0032] The key constraint elements are converted into machine-readable rule expressions, the rule expressions include premise conditions and execution actions, logical normalization processing is performed on the rule expressions, redundancies and contradictions between the rules are eliminated, a rule priority system is established, the rule priority system is determined based on the constraint strengths and execution sequences, and the normalized rule expressions are stored in an executable rule library.

[0033] As a further scheme of the application, in the resource scheduling execution module, the specific process of the execution scheme adjustment operation is:

[0034] An adjustment model based on a conflict propagation network is constructed, the nodes of the conflict propagation network represent matching pairs, the edges represent the conflict relationship between the matching pairs, the hierarchical marking results in the constraint check report are analyzed, the matching pairs violating the constraint conditions that need to be processed preferentially are determined, the core conflict nodes are located in the conflict propagation network, and the core conflict nodes are matching pairs with high violation degree scores and large influence ranges;

[0035] A local re-matching algorithm is used to adjust the matching relationship of the core conflict nodes, the local re-matching algorithm finds alternative schemes while keeping other matching pairs unchanged, the node states in the conflict propagation network are updated according to the local re-matching results, the conflict detection and re-matching processes are iteratively executed until all the matching pairs violating the constraint conditions are processed, and the matching relationship change history in the adjustment process is recorded.

[0036] As a further scheme of the application, the construction process of the conflict propagation network is:

[0037] With the matching pair in the initial matching scheme as the network node, the resource dependency relationship and the time dependency relationship between the matching pairs are analyzed, the node connection edge is established, the conflict propagation strength of each edge is calculated, the conflict propagation strength is determined based on the shared resource amount and the time overlap degree, when adjustment occurs in a certain matching pair, the state of the adjacent node is updated along the conflict propagation edge, the topology structure of the conflict propagation network is dynamically maintained, the conflict state change of each node in the network is recorded, and the priority of the matching pair that needs to be adjusted is updated in real time according to the network state change.

[0038] The beneficial effects of the present application are:

[0039] The present application solves the problem of low matching accuracy caused by the surface processing of data in the prior art by constructing a dynamic demand vector and a supply description vector fused with resource type, time window and geographical location characteristics, and deeply mining key information of supply and demand data through spatiotemporal dimension decomposition, multi-scale feature extraction and cross-modal association model, significantly improving the matching degree of supply and demand by dynamically adjusting the contribution proportion of resource specifications, time availability and location accessibility matching degree in combination with a multi-modal matching model and a gated fusion network; the multi-level verification engine analyzes the park operation strategy, generates a machine-readable verification rule set and establishes a constraint violation degree evaluation model, and integrates constraint conditions such as resource allocation priority and sharing restrictions to solve the problem of local resource overuse caused by neglecting the overall operation strategy of the park in the prior art, and to ensure the operation order of the park; the conflict propagation network locates the core conflict nodes with high violation degree and large influence range, and iteratively optimizes the scheme on the premise of maintaining the stability of other matching relationships in combination with a local rematching algorithm to solve the problem of inefficient conflict adjustment mechanism in the prior art, realize dynamic and efficient scheme optimization, and finally greatly improve the resource allocation efficiency of the smart park, reduce the operating cost of enterprises, and help the green and low-carbon operation of the park. BRIEF DESCRIPTION OF DRAWINGS

[0040] The present application will be further described below in conjunction with the accompanying drawings.

[0041] Figure 1 is a module schematic diagram of the present application. DETAILED DESCRIPTION

[0042] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0043] Please refer to Figure 1 The present application is a data governance platform for smart park construction, which comprises:

[0044] a data collection module configured to collect resource demand data and resource supply data of enterprises in the park, the resource demand data including equipment usage demand and service collaboration demand, and the resource supply data including idle equipment information and idle service capability information;

[0045] a feature structuring processing module configured to perform demand feature extraction on the resource demand data to generate a structured demand description, and perform supply feature extraction on the resource supply data to generate a structured supply description;

[0046] a multi-modal matching calculation module configured to calculate a matching degree between the demand and the supply based on the structured demand description and the structured supply description, and generate an initial matching scheme;

[0047] a multi-level constraint checking module configured to perform constraint condition checking on the initial matching scheme according to a park operation strategy, and identify matching pairs that violate the constraint condition;

[0048] a resource scheduling execution module configured to perform scheme adjustment on the matching pairs that violate the constraint condition, generate an optimized matching scheme, and schedule park resources to complete resource data sharing between enterprises.

[0049] In the data collection module, the specific process of collecting the resource demand data and the resource supply data is as follows:

[0050] The data collection module realizes unified access of data of enterprise subsystems through a standardized park data interface, and adopts RESTful API or MQTT protocol to adapt to communication requirements of different manufacturers' equipment management subsystems and service management subsystems, so as to ensure compatibility and real-time performance of cross-system data transmission. In the data synchronization stage, the module initiates a data request to each enterprise equipment management subsystem at a preset period (for example, 5 minutes / time), and synchronously obtains equipment operation data, which includes basic information such as equipment unique identifier (equipment ID), equipment model, real-time running state (online / offline), and historical use record; when extracting the equipment usage plan from the synchronous data, fields containing time interval and performance requirement are mainly selected, wherein the time interval records the starting time and the ending time of the equipment demand in the format of time stamp, and the performance requirement clearly indicates the required computing power (in the unit of floating point operation number per second FLOPS), storage capacity (in the unit of GB), and running stability (such as continuous fault-free duration requirement) of the equipment.

[0051] For the service management subsystem, the module obtains service request data in real time through a subscription mechanism. Service requests are divided into categories such as technical support services, logistics collaboration services, and R&D resource sharing services according to service types. Each service request is associated with service quality requirements, including service response time (e.g., within 1 hour), service completion rate (e.g., no less than 95%), and service result acceptance standards (e.g., technical document delivery integrity).

[0052] In the enterprise resource state monitoring link, the module uses a combination of real-time sampling and timed polling to dynamically monitor the state of enterprise equipment: when the equipment has not produced usage records for a continuous period of time exceeding a preset threshold (e.g., 24 hours), it is determined to be idle equipment, and the available time window of the equipment is recorded in the format of "start timestamp-end timestamp" to clearly indicate the specific period when the equipment can be scheduled. The analysis of service load state is based on the ratio of the current occupied resources to the total resources to calculate the load rate. When the load rate is below a preset threshold (e.g., 30%), the remaining resource capacity is counted as idle service capacity. The statistical dimension is quantified by unit time (e.g., per hour), and the specific performance is the additional service time or service number that can be provided per hour.

[0053] Finally, the module integrates the equipment usage plan and service request of the same enterprise into a resource demand dataset with enterprise ID as the core index. The dataset is stored in JSON format, and each data contains enterprise ID, demand type (equipment / service), demand parameters, time window, and geographic location. At the same time, the idle equipment information (equipment ID, model, available time window, performance parameters) and idle service capacity (service type, idle resource amount, unit time service capacity) are integrated into a resource supply dataset, also indexed by enterprise ID to ensure the correlation and traceability of supply and demand data, providing a structured data source for subsequent feature processing.

[0054] In the feature structured processing module described in the present application, the specific process of the demand feature extraction operation and the supply feature extraction operation is as follows:

[0055] The core of the feature structuring processing module is to convert unstructured or semi-structured supply and demand data into machine-identifiable structured features. First, the vector is constructed: for resource demand data, a dynamic demand vector containing space-time constraint conditions is constructed, which is composed of resource type features, time window features, and geographic location features. Among them, the resource type feature is classified and coded according to the equipment class (such as server, printer), service class (such as technical support, logistics), and the category information is converted into a numerical vector by using one-hot encoding or embedding vector (Embedding); the time window feature converts the time interval of demand into a time vector through a time quantization algorithm, including the starting time (numerical value in hours), the duration (hours), and the time flexibility coefficient (such as the fixed time demand coefficient is 0, and the adjustable demand coefficient is 0.5); the geographic location feature uses the internal coordinate system of the park (such as a two-dimensional rectangular coordinate system with the park entrance as the origin), converts the physical location associated with the demand (such as factory A-1 floor, office building B-3 floor) into (x, y) coordinate values, forms a spatial feature vector, and the three parts of features are fused into a unified dimension dynamic demand vector through vector splicing algorithm.

[0056] For resource supply data, a supply description vector with capacity elasticity is generated, which contains resource specification parameters, available time mode, and location coordinate information. The resource specification parameters are refined according to the resource type, such as device class resources including computing power (FLOPS), storage capacity (GB), and interface type, service class resources including service response speed (seconds / time), service personnel number, and service coverage; the available time mode distinguishes between periodic and non-periodic, the periodic mode (such as idle from 9:00 to 17:00 every day) is converted into a time series vector by using periodic encoding, and the non-periodic mode (such as random idle period) generates a probability vector through time distribution statistics; the location coordinate information is associated with the physical area number of the park (such as area 1, area 2), and is converted into coordinate values consistent with the dynamic demand vector, and after standardization processing (such as normalizing numerical parameters to the 0-1 interval), the supply description vector is spliced.

[0057] In the feature decomposition and extraction stage, the spatio-temporal dimension decomposition is performed on the dynamic demand vector, the vector is split into three independent components by using tensor decomposition algorithm: resource type component (only containing resource type feature vector), time component (only containing time window feature time series data), and space component (only containing geographical location feature coordinate vector), each component is stored separately and marked with the associated demand ID, facilitating subsequent processing according to the dimension; the supply description vector is subjected to multi-scale feature extraction, and a sliding window algorithm combined with convolution operation is used to extract features from three levels: micro scale (specification parameters of single device / single service), meso scale (cluster features of the same type of resources in the same enterprise, such as the average computing power of all idle servers in a certain enterprise), and macro scale (overall supply distribution of a certain type of resources in the park, such as the number distribution of idle devices in the eastern area of the park), and finally the resource specification features (fine-grained parameter vector), time availability features (multi-scale time distribution vector), and location accessibility features (spatial distance cost vector, such as the estimated time for moving the device to the demand location) are obtained.

[0058] In order to establish the association between supply and demand features, the module constructs a cross-modal association model for demand vector and supply vector, which is based on attention mechanism: first, calculate the association similarity between each component of dynamic demand vector (resource type, time, space) and each feature of supply description vector (resource specification, time availability, location accessibility), use cosine similarity algorithm to quantify the correlation between the two, and the similarity value is the association weight; based on the association weight, the features of demand and supply are weighted and fused, for example, when the weight of resource type component and resource specification feature is high, the two features are preferentially fused to generate a joint feature representation of demand and supply, which is a high-dimensional feature vector in a unified dimension; finally, the joint feature representation is converted into structured demand description and structured supply description according to the preset format, the description document contains feature name, feature value, association weight, feature source (demand / supply) and other fields, ensuring that the multi-modal matching module can directly call the feature data.

[0059] In the multi-modal matching calculation module of the present application, the specific process of calculating the matching degree is as follows:

[0060] The multi-modal matching calculation module first constructs a multi-modal matching model, which adopts an encoder-decoder architecture based on Transformer, and specially designs a multi-modal feature processing layer to simultaneously process resource specification features, time availability features, and location accessibility features. Independent feature mapping units are set in the layer to convert different types of features into feature vectors of a unified dimension (such as 512 dimensions), avoiding matching bias caused by differences in feature dimensions.

[0061] In the feature encoding stage, the model inputs the structured demand description into a query encoder, which consists of 3 layers of Transformer encoder layers, each containing a multi-head self-attention mechanism and a feedforward neural network: the self-attention mechanism models the association between each feature in the structured demand description, capturing the dependency between the features within the demand (such as the association between resource type and time window); the feedforward neural network performs a nonlinear transformation on the attention output, ultimately generating a query vector (QueryVector) containing the core features of the demand, with a vector dimension consistent with the output dimension of the multi-modal feature processing layer. At the same time, the structured supply description is input into a key-value encoder, which also uses 3 layers of Transformer encoder layers, but its output is divided into a key vector (KeyVector) and a value vector (ValueVector): the key vector is used to calculate the interaction weight with the query vector in the subsequent step, containing the index information of the supply features; the value vector contains detailed parameter information of the supply features, which is used for feature comparison in the matching degree calculation.

[0062] The cross-attention mechanism is the core step of matching degree calculation. This mechanism takes the query vector and the key vector as input, calculates the original interaction weight between them in each feature dimension (resource specification, time availability, location accessibility) through dot product operation, and then performs Softmax normalization to obtain the normalized interaction weight. The weight value ranges from 0 to 1, and a higher value indicates that the feature dimension has a greater impact on supply-demand matching. Based on the interaction weight, the module calculates the matching degree of each feature dimension: the resource specification matching degree is obtained by comparing the similarity of demand resource specification parameters and supply resource specification parameters (such as the inverse of the proportion of the difference between demand computing power and supply computing power to demand computing power); the time availability matching degree is obtained by calculating the overlap rate of demand time window and supply available time window (the ratio of overlapping duration to total demand duration); the location accessibility matching degree is obtained by the inverse of the distance or reachable time between demand location and supply location (the closer the distance, the shorter the reachable time, the higher the matching degree).

[0063] To realize the dynamic fusion of each matching degree component, the module uses a gated fusion network, which includes a gating unit and a fusion unit: the gating unit generates a gating coefficient for each matching degree component through a Sigmoid activation function, and the coefficient value is dynamically adjusted based on the interaction weight (for example, if the interaction weight of a certain feature dimension is high, the gating coefficient of the corresponding matching degree component will also be large); the fusion unit multiplies each matching degree component by the corresponding gating coefficient and sums them up to obtain the final fusion result. After normalization, the result is output as the final matching score between demand and supply, with a score range of 0-1. The higher the score, the higher the degree of supply-demand matching.

[0064] In the multi-modal matching calculation module described in the present application, the specific process of generating an initial matching scheme is as follows:

[0065] The first step of the initial matching scheme generation is the sorting and screening of candidate matching pairs: the module performs a descending order sorting on the final matching score of all demand-supply pairs, and the sorting algorithm uses quicksort to ensure high efficiency when the data volume is large (such as thousands of supply-demand pairs). After sorting, a candidate matching pair sequence is formed, and each element in the sequence contains the demand ID, supply ID, and matching score. Then a matching score threshold is set, and the threshold determination logic is based on the lowest matching score of historical matching success cases in the park and the expected matching accuracy of the current operation (for example, if the lowest score of historical success cases is 0.6 and the expected accuracy requirement is 90%, the threshold is set to 0.6). The threshold can be dynamically adjusted according to the supply and demand relationship in the park. The module marks the matching pairs with scores higher than the threshold in the candidate matching pair sequence as valid matching pairs, and marks the demands with scores lower than the threshold as unmatched, storing the unmatched demands in an unmatched demand list with associated demand ID and demand parameters for subsequent re-initiation of matching.

[0066] The conflict check of valid matching pairs includes time window conflict check and resource allocation conflict check: in the time window conflict check section, the module extracts the demand time window and supply available time window from the valid matching pairs one by one, and uses the interval intersection algorithm to determine whether there is overlap (if the intersection length of two time windows is greater than 0, it is determined that there is time overlap); for matching pairs with time overlap, the module first queries the time flexibility coefficient of the demand side, if the coefficient is greater than 0 (indicating that the time can be adjusted), the starting time or ending time of the time window is adjusted within the allowed time range of the demand, the adjustment amplitude does not exceed 10% of the total time length of the demand, ensuring that the adjusted time window does not overlap with other matching pairs' time windows, while meeting the basic time requirements of the demand; if the time flexibility coefficient is 0 (indicating that the time cannot be adjusted), mark the matching pair as a conflict pair to be processed temporarily and process it uniformly after the resource allocation conflict check.

[0067] In the resource allocation conflict check section, the module takes the supply side as the unit, counts the total resource demand of all valid matching pairs of this supply side (for example, a supply side has 3 equipment usage demands, with a total demand of 150GB storage capacity), and extracts the resource available amount of this supply side (for example, the total storage capacity of idle equipment is 120GB), calculates the difference between the total resource demand and the available amount; if the difference is less than or equal to 0 (resource sufficient), the matching pair is retained; if the difference is greater than 0 (resource insufficient), two adjustment strategies are adopted: one is to select a new supply, and select a supply with the second highest matching score (and score higher than the threshold) and sufficient resources from the supply data set to replace the original supply; the second is to split the resource amount, if the supply resource supports splitting (such as idle service capacity can be split by time), the resource amount is split into multiple parts according to the demand priority (such as the time of demand submission), and allocated to different demand sides, ensuring that each demand side obtains not less than the minimum resource amount.

[0068] After all the adjustments are completed, the module records the adjusted matching pair information, including the supplier ID before and after the adjustment, the time window parameter, the resource allocation amount, the adjustment reason (time overlap / resource shortage), and integrates all valid matching pairs with the supplier as the dimension to generate an initial matching scheme. The scheme adopts a structured document format, including the scheme number, generation time, valid matching pair list associated with each supplier (demand ID, resource type, time window, resource allocation amount), unmatched demand list, ensuring the completeness and executability of the scheme, and providing basic data for subsequent constraint condition verification.

[0069] In the multi-level constraint verification module of the application, the specific process of the constraint condition verification operation is as follows:

[0070] I. Overall architecture of multi-level verification engine:

[0071] The multi-level verification engine adopts a layered decoupling design, and realizes constraint condition verification through the collaborative work of the rule analysis layer, the condition matching layer, and the conflict detection layer. The layers communicate with each other through standardized data interfaces to ensure the consistency and traceability of data flow. The engine integrates a distributed computing framework at the bottom layer to support parallel verification of large-scale initial matching schemes (such as thousands of matching pairs), improving verification efficiency. At the same time, it has a built-in log recording module to record key operations (such as rule analysis results, matching results, and conflict detection results) during the processing of each layer in real time, providing data support for subsequent problem troubleshooting and scheme optimization.

[0072] II. Technical implementation of rule analysis layer:

[0073] The rule analysis layer first loads the park operation strategy document through the document analysis interface. The document format supports PDF, Word, and TXT. During the loading process, the unstructured document (such as PDF) is converted into editable text through a format conversion tool to ensure the extractability of the text content. For resource allocation rules (such as "Park key support enterprises are given priority to obtain computing power resources") and sharing restriction conditions (such as "Dangerous goods transportation services must not share logistics equipment with ordinary goods transportation services"), the layer uses a pre-trained language model (such as BERT) combined with a park domain dictionary (containing terms such as "key support enterprises", "computing power resources", "dangerous goods transportation", etc.) to perform semantic analysis. Through named entity recognition technology, key constraint elements are extracted: constraint subjects that clearly define the objects of the rules (such as "key support enterprises", "dangerous goods transportation services", "logistics equipment"), constraint conditions that define the restrictions that the subject must meet (such as "priority to obtain", "must not share", "resource usage upper limit"), and constraint strength divided into high, medium, and low three levels according to the park strategy (such as "safety-related restrictions" as high strength, "ordinary resource allocation priority" as medium strength).

[0074] After the key constraint elements are extracted, they are converted into machine-readable rule expressions using an IF-THEN logical structure: the IF part is the premise condition (e.g., "the demand side in the matching pair belongs to the key support enterprise"), and the THEN part is the execution action (e.g., "the matching pair has priority in resource allocation"). To eliminate rule redundancy and contradictions, a logical reasoning engine is introduced in the layer to perform normalization processing on the rule expressions: duplicate rules (with identical premise conditions and execution actions) are removed; for contradictory rules (e.g., "rule A requires A enterprise to have priority in resource allocation, and rule B restricts A enterprise from obtaining resources"), the rule with higher strength is retained, and if the strengths are the same, the context of the park operation strategy document is used to determine the retained rule.

[0075] In the rule priority system establishment stage, the constraint strength is the primary basis (high strength rule priority > medium strength > low strength), and rules with the same strength are further sorted according to the execution order in the park operation strategy document (rules earlier in the document have higher priority), resulting in an ordered rule list. The normalized rule expressions are written to the executable rule library through a structured storage interface, and the rule library uses a relational database (e.g., MySQL) for storage. Each rule record contains fields such as rule ID, constraint subject, premise condition, execution action, constraint strength, and priority, supporting fast queries by rule ID, constraint subject, and other dimensions.

[0076] III. Technical implementation of the condition matching layer:

[0077] The condition matching layer takes the initial matching scheme and the executable rule library as input and uses the Rete rule matching algorithm to achieve efficient matching. First, the attributes of each matching pair in the initial matching scheme are disassembled, and the demand side attributes (enterprise ID, enterprise type, demand type, resource demand), supply side attributes (enterprise ID, supply resource type, resource availability), and matching attributes (time window, geographic location) are extracted, forming a structured matching pair attribute set.

[0078] During the matching process, the algorithm converts the rule expressions in the rule library into a matching network, where the network nodes correspond to attribute judgments in the rule premise conditions (e.g., "whether the demand side is a key support enterprise" or "whether the resource type is computing power resources"). The matching pair attribute set passes through the judgments in the order of the nodes: if the matching pair attributes satisfy all the premise conditions of a certain rule, it is determined that the matching pair "meets the constraint conditions" and is marked as a "compliant matching pair"; if it does not satisfy any premise condition or only satisfies part of the premise conditions, it is marked as a "to-be-conflict-detected matching pair" and records the rule ID and reason (e.g., "does not meet the 'demand side is a key support enterprise' premise condition").

[0079] To improve the matching efficiency, a cache mechanism is introduced in the layer to cache the frequently accessed rules (such as general rules applied to most matching pairs) and the matched attribute judgment results to the memory, thereby reducing repeated queries and calculations; meanwhile, incremental matching is supported, and when the initial matching scheme is locally adjusted, only the matching pairs after the adjustment are re-matched, without the need for full verification, thereby reducing the consumption of computing resources. After the matching is completed, the matching result report is generated in the layer, which includes the number of compliant matching pairs, the number of matching pairs to be conflict detected, the number of matching hits of each rule, and the like, and the report is transmitted to the conflict detection layer in JSON format.

[0080] IV. Technical implementation of the conflict detection layer:

[0081] The conflict detection layer first constructs a matching pair dependency graph, taking the matching pairs to be conflict detected as nodes, establishing connection edges between the nodes based on resource dependency (sharing the same supply resource) and time dependency (overlapping time windows), and forming a non-directed graph structure. The mutual influence relationship of the nodes in the graph is analyzed through a topological sorting algorithm: if two nodes share the same supply resource and the total demand quantity of the resource exceeds the available quantity of the supplier, it is determined as a "resource conflict"; if the time windows of two nodes overlap and share the same non-parallel usable resource (such as a dedicated device), it is determined as a "time conflict", and the conflict type is marked by the node attribute.

[0082] In the constraint violation degree evaluation model construction stage, the model input includes the constraint strength of the violated rule (high strength rule weight > medium strength > low strength) and the influence range (the number of other matching pairs associated with the conflict node, the larger the associated number, the greater the influence range), and the violation degree score (the score range is 0-10, the higher the score, the more serious the violation degree) is calculated through weighted summation logic. For example, the matching pair that violates the high strength rule and affects more than 10 other matching pairs has a violation degree score of 8-10; the matching pair that violates the low strength rule and only affects 1-2 other matching pairs has a violation degree score of 1-3.

[0083] Based on the violation degree score, the matching pairs that violate the constraint condition are marked in stages: the violation degree score of 8-10 is marked as "first-class conflict" and needs the highest priority processing; the score of 4-7 is marked as "second-class conflict" and needs medium priority processing; the score of 1-3 is marked as "third-class conflict" and needs low priority processing. The final constraint verification report includes the conflict matching pair ID, the conflict type, the violated rule ID, the violation degree score, the hierarchical marking result, the influence range statistics, and the like, and the report is stored in a structured document format and pushed to the resource scheduling execution module through an interface.

[0084] In the resource scheduling execution module described in the application, the specific process of the execution scheme adjustment operation is as follows:

[0085] I. Construction and execution of the conflict propagation network adjustment model:

[0086] The conflict propagation network adjustment model is based on graph theory. First, the constraint verification report is loaded, the hierarchical label results and the influence range information of the matching pairs that violate the constraint conditions are extracted through the report analysis module, the priority processing order is determined in the order of "first-level conflict > second-level conflict > third-level conflict", and the matching pairs that have been labeled as "compliant" are filtered, and only the conflict matching pairs are retained to participate in network construction.

[0087] The network node is defined as a conflict matching pair, and each node is associated with attributes including matching pair ID, violation degree score, influence range (associated matching pair number), and current state (to be adjusted / in adjustment / adjusted); the edge between nodes is defined as a conflict relationship, and the edge connection is established only between conflict matching pairs with resource dependency or time dependency. By traversing the influence range data in the constraint verification report, the core conflict node is located in the conflict propagation network: the core conflict node must meet two conditions, one is that the violation degree score is ≥8 (first-level conflict), and the other is that the influence range is ≥5 associated matching pairs. The core node is distinguished from other nodes by red marking to ensure priority processing.

[0088] The model has a built-in state updating mechanism to synchronize the node adjustment progress in real time: when the core conflict node enters the adjustment process, the state is updated to "in adjustment"; if the constraint condition is met again after adjustment, the state is updated to "adjusted"; if there is still a conflict after adjustment, the state is rolled back to "to be adjusted" and the violation degree score is re-evaluated. State change information is recorded in real time through the log module to facilitate tracking of the adjustment process.

[0089] II. Implementation of local rematching algorithm:

[0090] The local rematching algorithm adjusts the matching relationship for the core conflict node. The algorithm strictly follows the "locality principle" during execution, i.e., only the supply side of the core conflict node is reselected, and the matching relationship of other matching pairs (including associated matching pairs) is maintained unchanged, avoiding the expansion of the adjustment range and causing new conflicts.

[0091] The first step of the algorithm is supply side screening: extract the supply side that matches the demand attributes (resource type, performance requirement, time window) of the core conflict node from the resource supply data set. The screening conditions are that the matching degree score is higher than the preset threshold (consistent with the initial matching scheme threshold) and the supply side has not been labeled as "resource insufficient", forming a candidate supply side list; the list is sorted in descending order of matching degree score, and the supply side with the highest score is selected first.

[0092] The second step is compliance verification: a new matching pair is formed by the core conflict node and the candidate supplier, the lightweight interface of the multi-level constraint verification module is called to quickly verify whether the new matching pair meets the park operation policy constraint conditions (only the rules directly related to the matching pair are verified); if the constraint conditions are met and there is no resource conflict or time conflict, it is determined as the final alternative; if not, the next supplier is verified in sequence according to the candidate supplier list until a compliant solution is found.

[0093] The third step is result application: after determining the alternative, the supplier information of the core conflict node is updated, and the edges of the conflict propagation network are connected to notify the nodes where the associated matching pairs are located to update the resource dependency and time dependency state (such as the associated nodes of the original shared resource need to recalculate the resource available amount); if a compliant solution is still not found after traversing the candidate supplier list, the core conflict node is marked as "to be globally adjusted" and the alarm mechanism is triggered to notify the park management personnel to intervene.

[0094] After the algorithm is executed, the matching relationship change history is recorded, the history data includes adjustment time, original matching pair information (demand ID, original supply ID, conflict type), new matching pair information (new supply ID, matching degree score), adjustment reason, verification result and other fields, the change history is stored with timestamp as index, and can be queried by time period or matching pair ID.

[0095] III. Conflict propagation network construction technology implementation:

[0096] The conflict propagation network construction is based on the initial matching scheme, first extracts all matching pairs (including compliant and conflict matching pairs) in the scheme as network nodes, each node is assigned a unique node ID, and the node attributes also include demand ID, supply ID, resource type, resource demand, time window (start / end timestamp), enterprise ID, etc. The attribute data is directly read from the structured document of the initial matching scheme.

[0097] The establishment of node connection edges is based on two dependency relationship analyses: resource dependency relationship analysis compares the supply IDs of different nodes, if the supply IDs of two nodes are the same (share the same supply resource), a resource dependency edge is established; time dependency relationship analysis calculates the overlap degree of the time windows of two nodes, if the overlap degree > 0 (there is time overlap) and the resource type is "non-parallel use resource" (such as dedicated detection equipment), a time dependency edge is established. Each edge is associated with a conflict propagation intensity parameter, which is determined based on the shared resource amount and the time overlap degree: the larger the shared resource amount (such as two nodes requiring 100GB of storage, accounting for 80% of the total capacity of the supplier), the higher the conflict propagation intensity; the higher the time overlap degree (such as the overlap duration accounting for 90% of the total duration of the demand), the higher the conflict propagation intensity, the intensity value is normalized to 0-1.

[0098] The dynamic maintenance of network topology is achieved by an event-driven mechanism: when a certain node has a matching relationship adjustment, an "adjustment event" is triggered, the event spreads to adjacent nodes along the conflict propagation edge, and the adjacent nodes update their resource dependency state or time dependency state (such as recalculating the remaining available amount of shared resources) after receiving the event; if the state change of the adjacent node leads to a new conflict (such as the resource available amount being lower than the demand amount), the node is marked as a "potential conflict node" and the priority is re-evaluated.

[0099] The network state change record adopts an incremental storage method, only records the changed attribute fields (such as resource available amount, conflict state), and does not store complete node information, thereby reducing the storage resource occupation; the priority update mechanism is based on the node state and the violation degree score, and adjusts the node processing order in real time, so as to ensure that the high-priority nodes (core conflict nodes, potential conflict nodes) enter the adjustment process in priority.

[0100] The above describes one embodiment of the present application in detail, but the content described is only a preferred embodiment of the present application and cannot be considered as limiting the scope of the present application. Any equivalent changes and improvements made according to the scope of the present application should still belong to the patent coverage of the present application.

Claims

1. A data governance platform for smart park construction, characterized in that, The method comprises the following steps: a data collection module is configured to collect resource demand data and resource supply data of enterprises in the park, wherein the resource demand data comprises equipment usage demand and service cooperation demand, and the resource supply data comprises idle equipment information and idle service capability information; a feature structuring processing module is configured to perform demand feature extraction on the resource demand data to generate a structured demand description, and perform supply feature extraction on the resource supply data to generate a structured supply description; a multi-modal matching calculation module is configured to calculate a matching degree between the demand and the supply based on the structured demand description and the structured supply description, and generate an initial matching scheme; a multi-level constraint verification module is configured to perform constraint condition verification on the initial matching scheme according to a park operation strategy, and identify matching pairs that violate the constraint conditions; a resource scheduling execution module is configured to perform scheme adjustment on the matching pairs that violate the constraint conditions, generate an optimized matching scheme, and schedule park resources to complete resource data sharing between enterprises; in the feature structuring processing module, the specific process of the demand feature extraction and the supply feature extraction is as follows: a dynamic demand vector containing a space-time constraint condition is constructed, and the dynamic demand vector integrates resource type features, time window features and geographical location features; a supply description vector with capacity elasticity is generated, and the supply description vector contains resource specification parameters, available time patterns and location coordinate information; a space-time dimension decomposition is performed on the dynamic demand vector to separate three components of resource type dimension, time dimension and space dimension; multi-scale feature extraction is performed on the supply description vector to obtain resource specification features, time availability features and location accessibility features; a cross-modal association model of the demand vector and the supply vector is established, the association weight of each dimension of the demand and each feature of the supply is calculated through an attention mechanism, the joint feature representation of the demand and the supply is generated based on the association weight, and the structured demand description and the structured supply description are formed; in the multi-modal matching calculation module, the specific process of calculating the matching degree is as follows: a multi-modal matching model is constructed, the multi-modal matching model simultaneously processes resource specification features, time availability features and location accessibility features; a query encoder of the multi-modal matching model is input with the structured demand description to generate a query representation; a key-value encoder of the multi-modal matching model is input with the structured supply description to generate a key-value representation; the interaction weight of the query representation and the key-value representation in each feature dimension is calculated through a cross-attention mechanism, the resource specification matching degree, the time availability matching degree and the location accessibility matching degree are fused based on the interaction weight, the contribution proportion of each matching degree component is dynamically adjusted by using a gated fusion network, and the matching degree between the demand and the supply is output according to the fusion result; in the resource scheduling execution module, the specific process of performing the scheme adjustment is as follows: an adjustment model based on a conflict propagation network is constructed, a node of the conflict propagation network represents a matching pair, an edge of the conflict propagation network represents a conflict relationship between the matching pairs, a hierarchical marking result in a constraint verification report is analyzed, a matching pair that needs to be processed preferentially is determined, a core conflict node is located in the conflict propagation network, and the core conflict node is a matching pair with a high violation degree score and a large influence range. The local re-matching algorithm is used to adjust the matching relationship of the core conflict node, the local re-matching algorithm finds a substitute solution while keeping other matching pairs unchanged, the node state in the conflict propagation network is updated according to the local re-matching result, the conflict detection and re-matching process is iteratively executed until all matching pairs that violate the constraint condition are processed, and the matching relationship change history in the adjustment process is recorded; The construction process of the conflict propagation network is as follows: The matching pairs in the initial matching scheme are taken as network nodes, the resource dependency relationship and the time dependency relationship between the matching pairs are analyzed, the node connection edges are established, the conflict propagation strength of each edge is calculated, the conflict propagation strength is determined based on the shared resource amount and the time overlap degree, when a certain matching pair is adjusted, the state of the adjacent node is updated along the conflict propagation edge, the topology structure of the conflict propagation network is dynamically maintained, the conflict state change of each node in the network is recorded, and the matching pair priority to be adjusted is updated in real time according to the network state change.

2. The data governance platform for smart park construction of claim 1, wherein, In the data acquisition module, the specific process of acquiring resource demand data and resource supply data is as follows: The operation data of the equipment management subsystem and the service management subsystem of each enterprise is synchronized through the park data interface, the equipment use plan containing the time interval and the performance requirement is extracted from the equipment management subsystem, and the service request containing the service type and the quality requirement is obtained from the service management subsystem; Meanwhile, the enterprise resource state is monitored, the idle equipment is identified and the available time window is recorded, the service load state is analyzed and the idle service capacity is counted, the equipment use plan and the service request are integrated into a resource demand data set, and the idle equipment information and the idle service capacity are integrated into a resource supply data set.

3. The data governance platform for smart park construction of claim 1, wherein, In the multi-modal matching calculation module, the specific process of generating the initial matching scheme is as follows: The descending order sorting operation is performed on the matching degree score to form a candidate matching pair sequence, the matching degree threshold is set to screen the effective matching pairs in the candidate matching pair sequence, and the demand below the matching degree threshold is marked as an unmatched state; The time window of each effective matching pair is checked one by one, the time arrangement with overlap is identified, the resource allocation of each effective matching pair is analyzed, the difference between the total resource demand and the available amount is calculated, for the matching pairs with time overlap, the start time or the end time of the time window is adjusted, for the matching pairs with resource demand exceeding the available amount, the supplier is reselected or the resource amount is split, the adjusted matching pair information is recorded, and the initial matching scheme containing all the effective matching pairs is generated.

4. The data governance platform for smart park construction of claim 1, wherein, In the multi-level constraint verification module, the specific process of the constraint condition verification operation is as follows: A multi-level verification engine is constructed, the multi-level verification engine includes a rule analysis layer, a condition matching layer and a conflict detection layer, the constraint conditions in the park operation strategy are analyzed in the rule analysis layer to generate an executable verification rule set, the matching pairs in the initial matching scheme are matched with the verification rule set in the condition matching layer, and the matching pairs satisfying the constraint conditions are identified; The conflict detection layer analyzes the mutual influence relationship between the matching pairs, detects matching pairs with resource conflicts or time conflicts, establishes a constraint violation degree evaluation model, calculates a violation degree score based on the severity and influence range of the violation rule, marks the matching pairs that violate the constraint conditions according to the violation degree score, and generates a constraint verification report containing the marking results.

5. The data governance platform for smart park construction of claim 4, wherein, The specific working process of the rule analysis layer is as follows: Load the park operation strategy document, which contains resource allocation rules and sharing restriction conditions in text format, extract key constraint elements from the strategy document using semantic analysis technology, including constraint subjects, constraint conditions and constraint strengths; Convert the key constraint elements into machine-readable rule expressions, which include premise conditions and execution actions, perform logical normalization processing on the rule expressions to eliminate redundancy and contradictions between rules, establish a rule priority system based on constraint strength and execution order, and store the normalized rule expressions in an executable rule library.

Images