Industrial internet management platform based on artificial intelligence and big data

Abstract

The invention discloses an industrial Internet management platform based on artificial intelligence and big data, and belongs to the technical field of industrial Internet platforms, and the platform comprises an event fingerprint module which is used for monitoring an original event flow, generating an event fingerprint for each event, and outputting an event sequence based on a timestamp, a source device ID and an event type hash value of the event fingerprint; the state mirroring module is used for constructing a state mirroring model of the supply chain, mapping the event sequence into a state vector through a dynamic weight distributor, adjusting the weight by a predetermined emergency degree parameter of the event fingerprint, and outputting a supply chain snapshot; the strategy simulation module is used for starting a strategy simulator, executing strategy simulation through a preloading strategy library by utilizing the supply chain snapshot, and realizing real-time sensing and dynamic state integration of a supply chain event by taking a previous state snapshot as a reference, so that the system can quickly respond to market demand fluctuation and interruption events; and the supply chain decision real-time performance and the information integration efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of industrial internet platform technology, and more specifically, to an industrial internet management platform based on artificial intelligence and big data. Background Technology

[0002] In the context of the Industrial Internet, the efficient operation of the supply chain system is crucial to a company's market competitiveness. However, current technologies have certain shortcomings in dealing with market demand fluctuations or disruptions. Specifically, the supply chain system struggles to respond to such changes in real time, leading to problems such as decision-making lag and inefficiency. Existing forecasting systems mostly rely on batch processing of historical data and cannot dynamically integrate real-time variables such as logistics delays and sudden orders, resulting in information silos at each stage and sluggish decision-making. A deeper analysis of the underlying technical causes reveals that this is due to the rigidity of the system architecture. Traditional optimization models, in pursuit of high-precision predictions, rely on complex calculations and multi-source data integration, a process that introduces significant time delays. At the same time, real-time variables such as equipment status and external events are heterogeneous and dynamic, making it impossible for model updates and decision execution to be synchronized, resulting in a disconnect between computation and response.

[0003] Specifically, in the context of real-time response and dynamic optimization of supply chain disruptions, the pursuit of accuracy in prediction and optimization inevitably leads to delays in response and a deterioration in the real-time nature of decision-making. This is because high-precision prediction requires extensive data cleaning, model retraining, and collaborative verification. Supply chain events, such as raw material shortages and logistics disruptions, require the system to make decisions within seconds. The increased computational complexity directly slows down the response speed, preventing the system from taking timely action before the event worsens. In summary, existing technologies have significant technical deficiencies in real-time response, dynamic optimization, and the balance between accuracy and real-time performance in supply chain systems. Therefore, to address these shortcomings, this invention proposes an industrial internet management platform based on artificial intelligence and big data. Summary of the Invention

[0004] In view of the problems existing in the prior art, the purpose of this invention is to provide an industrial internet management platform based on artificial intelligence and big data, which can realize real-time perception and dynamic status integration of supply chain events, enabling the system to respond quickly to market demand fluctuations and interruption events, and improve the real-time nature of supply chain decision-making and information integration efficiency.

[0005] To solve the above problems, the present invention adopts the following technical solution: An industrial internet management platform based on artificial intelligence and big data includes: The event fingerprint module is used to listen to the raw event stream and generate an event fingerprint for each event. The event fingerprint is based on the timestamp, source device ID, and event type hash value, and outputs an event sequence. The State Mirroring module is used to build a state mirroring model of the supply chain. It maps event sequences into state vectors through a dynamic weight allocator, where the weights are adjusted by a predetermined urgency parameter of the event fingerprint, and outputs a supply chain snapshot. The strategy simulation module is used to start the strategy simulator. It uses supply chain snapshots to perform strategy simulations through a preloaded strategy library. Based on the previous state snapshot, it simulates the state transition probability after applying a strategy perturbation and outputs a strategy effect score. The strategy filtering module is used to import a strategy selector, run a predefined trade-off function to filter strategies based on strategy performance scores, and output strategy instructions. The effect tracking module is used to deploy the effect tracker, monitor the actual effects of the strategy instructions in the supply chain after they are executed, compare the actual execution results with the simulated prediction results to generate a deviation vector, and use the deviation vector to adjust the weight allocation parameters in the state mirror model. The weight adjustment module is used to update the state mirror model using the deviation vector and adjust the model weights through a weight update algorithm. The Autonomous Collaboration Module is used to build an Autonomous Collaboration Engine, which recalibrates the parameters of the state mirror model and the policy simulator at predetermined time intervals, and automatically switches simulation policies and weight settings based on predefined conditions.

[0006] Furthermore, the event fingerprint module is also used for: Perform dynamic causal slicing on the original event stream and output event slices; Extract the joint feature modalities of events within the event slice and map them into a composite fingerprint; The composite fingerprint is compared with the fingerprint activity library within a predetermined time window. If there is a conflict, the fingerprint recalibration procedure is initiated to output the resolved fingerprint. The decomposed fingerprints are sorted and a lightweight temporal dependency graph is constructed. The output is an enhanced event sequence as an extension of the event sequence.

[0007] Furthermore, the state mirroring module is also used for: Analyze the spatiotemporal context of the fingerprint in the enhanced event sequence and calculate the initial time sensitivity coefficient; Event fingerprints are mapped to the supply chain network topology to simulate the disturbance propagation process, and the instantaneous potential energy value of the state vector is calculated based on the event propagation model. Detect potential energy conflicts and introduce a dynamic coordinator to seek potential energy allocation schemes, and output the converged state vector value. Associating state vector sets with event context tags and encapsulating them into an enhanced supply chain snapshot as an extension of the supply chain snapshot.

[0008] Furthermore, the policy simulation module is also used for: Analyze the context graph of strategies for building enhanced supply chain snapshots; The policy is instantiated as a perturbation source, its propagation process in the graph is simulated, and the quantized impact value of the policy on the state vector is recorded. The detection strategy interferes and the quantitative impact values ​​of the conflict are superimposed using a preset nonlinear function to generate a coordinated impact value distribution. The primary and secondary effects are calculated based on the coordinated distribution of impact values, and a multidimensional score of policy effect is generated as an extension of the policy effect score.

[0009] Furthermore, the strategy filtering module is also used for: The strategy effectiveness multidimensional score is mapped to a three-dimensional decision coordinate system and the unit scale is dynamically calibrated. Define a dynamic trade-off window and calculate the weighted distance between the strategy projection point and the predefined ideal decision point; Analyze the complementary characteristics of weighted distance identification strategies, combine the strategy execution sequence, and define the triggering conditions; The policy sequence is bound to the state vector threshold, parameters are pre-encapsulated and adaptive triggering conditions are set, and the output is a self-contained policy instruction package as an extension of the policy instruction.

[0010] Furthermore, the effect tracking module is also used for: Capture the actual execution data of the policy instruction package, establish a logical execution timeline based on the event timestamp sequence and aligned with the snapshot timestamp, and output the anchored execution observation sequence; By comparing the actual observed trajectory with the predicted trajectory, trend deviation and fluctuation deviation are extracted to form a multi-level deviation vector.

[0011] Furthermore, the effect tracking module is also used for: The multi-level deviation vector is back-projected to the policy context graph and the source of contributing events, the deviation distribution pattern is analyzed and the error contribution graph is output through a preset contribution factor decomposition algorithm; Based on the error contribution map, the dominant error source dimension is identified, and the internal parameters of the weight allocator are dynamically adjusted.

[0012] Furthermore, the weight adjustment module is also used for: Based on the error contribution map, a preset frequency domain feature decomposition algorithm is used to process the multi-level deviation vectors, calculate the deviation covariance matrix, and extract the deviation feature vectors. By mapping the deviation feature vectors to the network topology, vulnerable nodes are identified and a vulnerability topology map is constructed. A hierarchical weight update strategy is designed for the vulnerability topology map, employing differentiated update coefficients and introducing propagation constraints; The updated model is backtracked for verification and stability testing, and a reinforced state mirror model is output.

[0013] Furthermore, the autonomous collaboration module is also used for: The output volatility and coefficient of variation of the strategy effect score of the enhanced state mirror model are calculated, and then a stability score is generated. When the stability score exceeds a predetermined threshold, a recalibration process is automatically triggered. Parameter contribution is calculated using historical bias data and error maps to identify highly sensitive parameters and detect policy conflict patterns.

[0014] Furthermore, the autonomous collaboration module is also used for: Parameter adjustment mapping is generated based on sensitivity analysis results, and a strategy switching map is constructed based on conflict detection results. The model parameters are adjusted using a preset incremental method, and the strategy is switched gradually, while the stability score is monitored in real time. Collect updated performance data and compare it with the state before the update to update the engine trigger thresholds and decision rules.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) This solution generates event sequences in real time and constructs enhanced event sequences through the event fingerprint module. Combined with the state mirror module, it dynamically maps the supply chain status, realizes the real-time perception of supply chain events and dynamic integration of status, and enables the system to quickly respond to market demand fluctuations and interruption events, thereby improving the real-time nature of supply chain decision-making and the efficiency of information integration.

[0016] (2) The strategy simulation and screening module of this solution is based on multidimensional scoring and dynamic trade-off function to accurately screen and combine strategies. At the same time, the strategy effect is quantified through strategy context graph and coordination mechanism, which greatly improves the accuracy and adaptability of strategy decision-making, avoids the execution of ineffective strategies, and ensures that the supply chain can efficiently formulate and execute the optimal strategy in complex scenarios.

[0017] (3) The effect tracking and weight adjustment module of this scheme forms a closed-loop optimization mechanism. By back-optimizing the weight of the state mirror model through the deviation vector, and combining frequency domain decomposition and vulnerability analysis to accurately adjust the parameters, the model continuously improves the accuracy of the model in depicting the actual state of the supply chain, reduces prediction deviation, and enhances the reliability and consistency of system decision-making.

[0018] (4) The autonomous collaborative module of this solution automatically calibrates the parameters of the state mirror model and the strategy simulator, monitors the system stability in real time and dynamically switches strategies, realizing the autonomous evolution and stable operation of the system. It can adapt to the dynamic changes of the supply chain without frequent human intervention, reducing the system operation and maintenance costs and ensuring the robustness of long-term operation.

[0019] (5) This solution solves the rigidity problem of traditional systems from the architectural level. Through the collaborative operation of each module, it effectively solves the disconnect between calculation and response while ensuring the accuracy of prediction and optimization. It balances the accuracy and real-time performance of decision-making, enabling the supply chain system to maintain high precision under the requirement of second-level decision-making, and provides technical support for the rapid handling of supply chain interruption events.

[0020] (6) The industrial internet management platform built by this solution covers the entire supply chain process, realizing intelligent management from event perception and state modeling to strategy decision-making and effect feedback. It greatly improves the overall operational efficiency of the supply chain, reduces cost losses caused by decision delays or errors, helps enterprises enhance market competitiveness, and promotes the digital transformation of supply chain management in the industrial internet environment. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0022] Figure 1 This is a flowchart illustrating the various modules of an industrial internet management platform based on artificial intelligence and big data according to the present invention. Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0024] Please see Figure 1 An industrial internet management platform based on artificial intelligence and big data includes: an event fingerprint module, used to monitor the raw event stream and generate an event fingerprint for each event. The event fingerprint is based on a timestamp, source device ID, and event type hash value, and outputs an event sequence. In this embodiment, the event fingerprint module first continuously monitors the raw event streams from various terminal devices and system interfaces. These events include various state changes, instruction transmissions, and anomaly feedback information generated during the supply chain operation. To achieve a unique identifier for each event, the event fingerprint module generates a basic event fingerprint based on three dimensions: timestamp, source device ID, and event type. The key lies in converting multi-dimensional unstructured parameters into feature values ​​in a unified format through hash calculation, ensuring that the fingerprints of different events are distinguishable. This hash calculation logic originates from the abstraction of the core attributes of the event. The timestamp reflects the temporal characteristics of the event, the source device ID defines the source entity of the event, and the event type clarifies the essential attributes of the event. These three together constitute the core identifying elements of the event. Through the mapping relationship of the hash function, heterogeneous raw parameters can be converted into a fixed-length unique identifier. The calculation expression is as follows:

[0025] This formula is based on the characteristics of hash functions. Hash functions can map input data of arbitrary length to output values ​​of fixed length and have a high probability of uniqueness, which perfectly meets the requirement of unique identification for event fingerprints. The timestamp, source device ID, and event type are used as input parameters because these three parameters completely define the characteristics of a single event from three dimensions: time sequence, subject, and attribute. The combination of the three can minimize the collision problem of different events generating the same fingerprint. In the formula, H represents the final generated event fingerprint hash value, T is the precise timestamp of the event, recorded with millisecond precision, D is the unique identifier of the source device that generated the event, which consists of the device's factory number or a unique code assigned by the system, Type is the event classification identifier, such as predefined type codes such as logistics status update, order change, and equipment failure, and hash(·) is the selected asymmetric hash function, such as the SHA-256 cryptographic hash function, used to complete the conversion of multi-dimensional parameters to fixed-length hash values. After the basic event fingerprint is generated, the event fingerprint module organizes the events in chronological order to form an initial event sequence.

[0026] To enhance the relevance and effectiveness of event sequences, the event fingerprint module further performs dynamic causal slicing. This operation, based on the causal relationships between events, analyzes the triggering conditions, data flow, and scope of influence of events to divide the original event stream into several logically related event slices. Each slice reflects relevant events under a specific business scenario or causal chain, thereby achieving a structured decomposition of the event stream and outputting clearly defined event slices. For each event slice, the event fingerprint module extracts the joint feature modalities of all events within it. These feature modalities cover multi-dimensional information such as event attribute parameters, correlation strength, and influence weight. Through feature fusion algorithms, the scattered event features are integrated into a unified composite fingerprint, which can characterize the overall characteristics and correlation patterns of events within the entire slice.

[0027] After the composite fingerprint is generated, the event fingerprint module compares it in real time with the fingerprint activity library within a predetermined time window. The fingerprint activity library stores the valid composite fingerprints that have been processed within this time window and is used to determine whether the newly generated composite fingerprint has duplicates, contradictions, or other conflicts. When a conflict is detected between the composite fingerprint and an existing fingerprint in the activity library, the event fingerprint module immediately starts the fingerprint recalibration procedure. By analyzing the causes of the conflict, such as duplicate event reporting, data transmission errors, logical contradictions, etc., the conflicting fingerprint is processed by parameter correction, weight adjustment, or feature reconstruction. After eliminating the contradictory factors, the resolved fingerprint is output to ensure that the representation of each event slice has consistency and accuracy.

[0028] Finally, the event fingerprint module sorts all the resolved fingerprints according to the chronological order of the events. At the same time, it constructs a lightweight temporal dependency graph based on the dependencies between events, such as the preceding event being the trigger condition for the subsequent event, and different events sharing the same data resources. The temporal dependency graph only retains the association between events, avoiding the increase in computational complexity caused by redundant structures. It clearly presents the temporal logic and dependency paths of events in a graphical way. Based on the sorted resolved fingerprints and the temporal dependency graph, the event fingerprint module finally outputs an enhanced event sequence. The enhanced event sequence not only contains the unique identifier of each event, but also integrates the logical association and temporal information between events, and has richer structured features compared to the initial event sequence.

[0029] In a preferred embodiment of the present invention, the industrial internet management platform further includes: constructing a state mirror model of the supply chain, mapping event sequences into state vectors through a dynamic weight allocator, wherein the weights are adjusted by a predetermined urgency parameter of the event fingerprint, and outputting a supply chain snapshot.

[0030] In this embodiment, to achieve precise binding between event characteristics and supply chain status, the state mirroring module first performs in-depth analysis of the spatiotemporal context of each event fingerprint in the enhanced event sequence. The temporal context covers the specific time when the event occurred, the time interval with other related events, and the temporal position in the entire business process. The spatial context includes the physical deployment location of the event source device, the supply chain node level to which it belongs, and the scope of influence. Based on these analysis results, the state mirroring module calculates the initial time sensitivity coefficient. The initial time sensitivity coefficient is used to quantify the decay or enhancement characteristics of the event's impact over time. Different types of events correspond to different decay or enhancement rules, ensuring that the subsequent weight allocation can fit the temporal impact characteristics of the event.

[0031] After obtaining the initial time sensitivity coefficient, the initial time sensitivity coefficient accurately maps each event fingerprint to the preset supply chain network topology diagram according to its corresponding source device ID and business attributes. The supply chain network topology diagram clearly presents the connection relationship, hierarchical structure and data interaction path of each node in the supply chain. The specific nodes in the supply chain include suppliers, production plants, logistics hubs, and sales terminals. The mapping process of event fingerprints is essentially to determine the initial node in the supply chain network where the event acts as a disturbance source. Subsequently, the state mirroring module simulates the propagation process of the disturbance based on the event propagation model, that is, how the impact of the event on the initial node spreads to other related nodes through the topological connection relationship. This process needs to consider factors such as the correlation strength between nodes, data transmission latency and resource dependence. In order to quantify the state change of each node after the disturbance propagation, the state mirroring module introduces the instantaneous potential energy value calculation. The calculation logic is derived from the analogy of the change in node state caused by the disturbance propagation to the potential energy change caused by energy transfer in a physical system. Combined with the predetermined urgency parameter of the event, the propagation distance and the node correlation weight, the calculation expression is abstracted as follows:

[0032] This formula is based on the propagation attenuation law of event impact and the node association characteristics. The urgency of the event directly determines the initial impact strength; the farther the propagation distance, the weaker the impact. Simultaneously, the association weights between nodes determine the efficiency of impact transmission. The product of these four factors can accurately quantify the degree of state change of a single node under a specific event disturbance. In the formula, This represents the instantaneous potential energy value of the i-th node in the supply chain network. The predefined urgency parameter corresponding to the event fingerprint is predefined by the business scenario and ranges from 0 to 1. The larger the value, the higher the urgency of the event. is the initial state base of the i-th node, reflecting the basic state value when the node is operating normally; k is the propagation attenuation coefficient, which is determined by the topological characteristics of the supply chain network and is fixed as a positive number. The propagation distance from the event disturbance source to the i-th node is quantified by the number of node hops or physical distance. The association weight between the initial node j and the i-th node reflects the degree of resource dependence or data interaction between nodes, ranging from 0 to 1. Through this formula, the state mirror module can calculate the instantaneous potential energy value of each node in the supply chain network, and then integrate them to form a complete state vector.

[0033] Because multiple events may occur simultaneously in the supply chain network, the instantaneous potential energy values ​​corresponding to different events may contradict each other at some nodes, i.e., potential energy conflict. This manifests as the same node changing state in opposite directions or exceeding the node's carrying capacity under the influence of different disturbances. To solve this problem, the state mirroring module introduces a dynamic coordinator. The dynamic coordinator first identifies all nodes with potential energy conflicts and the conflict type. Then, based on the node's operating threshold, business priority, and overall supply chain optimization goals, it seeks the globally optimal potential energy allocation scheme. During the coordination process, the dynamic coordinator adjusts the potential energy values ​​of the conflicts, reduces the impact of conflicts on non-critical nodes, prioritizes the state stability of business nodes, and finally outputs the converged state vector value to ensure that the state vector can truly and consistently reflect the current overall operating state of the supply chain.

[0034] Finally, the state mirroring module integrates the converged state vectors of all nodes into a state vector set, and associates each state vector with its corresponding event context. The associated content includes the event fingerprint that triggered the state change, the disturbance propagation path, and the coordination and processing record. Through this association and marking, the supply chain state not only includes the numerical state of the current node, but also includes the source information of the state change. The state mirroring module encapsulates the associated and marked state vector set to form an enhanced supply chain snapshot. Compared with the basic supply chain snapshot, this enhanced version adds key information such as event source, conflict handling records, and node association relationships.

[0035] In a preferred embodiment of the present invention, the industrial internet management platform further includes: a strategy simulation module, used to start a strategy simulator, perform strategy simulation by preloading a strategy library using a supply chain snapshot, simulate the state transition probability after applying a strategy perturbation based on a previous state snapshot, and output a strategy effect score.

[0036] In this embodiment, to establish a precise correspondence between strategies and supply chain scenarios, the strategy simulation module performs in-depth analysis of the enhanced supply chain snapshot, extracting supply chain node information, inter-node correlation strength, current state vector values, and event context features. Based on this information, a strategy context graph is constructed. The strategy context graph uses supply chain elements as nodes, such as material inventory, production equipment, logistics routes, and order demands, and uses inter-element dependencies, data interaction paths, and resource flow directions as edges. At the same time, various strategies in the pre-loaded strategy library are bound to the graph nodes and their relationships, clarifying the applicable scenarios, target objects, and triggering conditions of each strategy, so that the strategy can accurately match the actual operating state of the supply chain. After the strategy context graph is constructed, the strategy simulation module instantiates each strategy in the preloaded strategy library as a specific disturbance source. Based on the target of the strategy, it locates its initial impact node in the graph. Subsequently, the strategy simulation module simulates the transmission process of the disturbance source based on the transmission characteristics of the supply chain network, i.e., how the impact of the strategy on the initial impact node diffuses to other nodes through the relationships in the graph. This transmission process needs to combine the association weights between nodes, resource flow efficiency, and the sensitivity of the state vector to quantify the degree of change of the strategy on the state vector of each node, forming a quantified impact value. To accurately describe this quantified relationship, a calculation expression is abstracted by combining strategy effectiveness, association strength, and transmission attenuation laws:

[0037] This formula is based on the propagation characteristics of policy influence. The initial influence of a policy at the initial node decays as the propagation path length increases. Simultaneously, the association weights between nodes directly determine the efficiency of influence propagation. The product of these three factors accurately quantifies the actual impact of a single policy on the target node. In the formula, ... Representing the The strategy for the first The quantization impact value of the state vector of each node; For the first The basic performance parameters of a strategy are determined by the strategy's own design goals and implementation intensity, and range from 0 to 1. For the first The initial action node of the first strategy and the second The association weight of each node reflects the degree of resource dependence or data interaction between the two, ranging from 0 to 1; The conduction attenuation coefficient, determined by the topology and resource flow characteristics of the supply chain network, is fixed as a positive number. For the first From the initial action node of the strategy to the first The propagation path length of each node is quantified in terms of node hops.

[0038] When multiple strategies are simulated simultaneously, the quantified impact values ​​of different strategies may interfere with each other at the same node, manifesting as opposite directions of influence, excessive superposition of influence intensity, or conflicting influence ranges. The strategy simulation module detects such policy interference phenomena by monitoring the changes in the quantified impact values ​​of each node in real time, and introduces a preset nonlinear function to superimpose the conflicting quantified impact values. This nonlinear function can dynamically adjust the conflicting impact values ​​according to the node's operating threshold, business priority, and overall supply chain optimization goals, suppressing excessive superposition of influence, coordinating influences in opposite directions, and finally generating a coordinated influence value distribution to ensure that the quantified impact results of each node conform to the actual operating logic.

[0039] Regarding the preset nonlinear function, it is first assumed that multiple strategies produce quantized influence values ​​on the same node i. Where n is the number of strategies applied to this node, a pre-defined nonlinear function is used to sum the impact values ​​of these conflicts, and the final output is the reconciled impact value. Its logic is based on dynamically weighting and constraining the impact value of conflicts according to node operating thresholds and business priorities, suppressing excessive superposition and coordinating directional conflicts. Based on this, the expression can be derived as follows:

[0040] This formula uses the impact value of each strategy. Assign business priority weights Then through the function Introducing node operation thresholds Business Priority The constraints are then superimposed to obtain the coordinated influence value, as shown in the formula. Represented as nodes The final quantified impact value after coordination; n represents the value acting on the node. The total number of strategies; This is represented as the original quantized impact value of the k-th strategy on node i; The business priority weight of the k-th strategy is predefined by the supply chain business rules and has a range of... The higher the priority, the greater the weight; Represented as a nonlinear constraint function, it can be defined as follows: ; in This is the original impact value. For nodes Operating thresholds, such as the maximum and minimum tolerance values ​​of the state vector, For nodes The business priority identifier is used to trigger the determination of threshold constraints.

[0041] Based on the coordinated impact value distribution, the strategy simulation module further calculates the primary and secondary effects brought about by the strategy implementation. The primary effect refers to the state change caused by the strategy directly acting on the target node, which can be directly quantified by the change in the target node's state vector. The secondary effect refers to the chain reaction triggered by the strategy through transmission, that is, the state change of non-directly affected nodes due to the correlation. It needs to be comprehensively quantified by tracking the impact value changes of all nodes on the transmission path. Combining the quantification results of the primary and secondary effects, the strategy simulation module evaluates from multiple dimensions such as response speed, resource utilization, cost control, and risk avoidance, and generates a multi-dimensional score for the strategy effect. Compared with the single-dimensional effect score, this score can more comprehensively reflect the actual application value of the strategy.

[0042] In a preferred embodiment of the present invention, the industrial internet management platform further includes: a strategy selection module, used to introduce a strategy selector, run a predefined trade-off function to select strategies based on strategy effect scores, and output strategy instructions.

[0043] In this embodiment, to establish a precise correlation between strategy scoring and supply chain decision-making scenarios, the strategy filtering module analyzes and enhances the supply chain snapshot, extracting node states, relationships, and event contexts to construct a strategy context graph. This graph binds key supply chain decision dimensions, such as logistics timeliness, inventory levels, and production efficiency, to the target audience and scope of influence of strategies, clarifying the applicability boundaries and collaborative possibilities of different strategies in the current supply chain scenario. Supported by the strategy context graph, the strategy filtering module maps multi-dimensional strategy effectiveness scores to a three-dimensional decision coordinate system. These multi-dimensional scores encompass response speed, resource utilization, and cost control. In this coordinate system, the three dimensions correspond to the objectives of supply chain decision-making. The unit scale of each dimension is dynamically calibrated based on scenario features extracted from enhanced supply chain snapshots, such as node load and resource scarcity, ensuring that the scores of different dimensions are comparable in scale and importance. To quantify the fit between strategies and ideal decision objectives, the strategy selection module defines a dynamic trade-off window and calculates the weighted distance between the projection point of each strategy in the three-dimensional decision coordinate system and the predefined ideal decision point. The calculation of this weighted distance originates from the abstraction of distance, weight, and priority relationships in multi-objective decision-making. Combined with the weight allocation of each decision dimension by supply chain business, the derived formula is as follows:

[0044] The derivation of this formula is based on a multi-dimensional extension of Euclidean distance. Considering the different importance of different decision dimensions to the supply chain, dimensional weights are introduced for weighting, thereby quantifying the overall deviation between the strategy projection point and the ideal point. In the formula, Represents the weighted distance of the s-th strategy; The business weight for the d-th decision dimension is predefined by the supply chain operation rules and ranges from 0 to 1. The larger the weight, the more important the dimension. Let be the projection value of the s-th policy on the d-th dimension. Let d be the coordinates of the predefined ideal decision point in the d-th dimension.

[0045] Based on the weighted distance calculation results, the strategy selection module analyzes the complementary characteristics between different strategies. That is, some strategies perform well in specific decision dimensions and can complement the effects of other strategies. By identifying such characteristics, the strategy selection module combines strategy execution sequences, clarifies the sequential execution logic between strategies, and defines trigger conditions based on the state vector threshold in the enhanced supply chain snapshot, such as initiating a replenishment strategy when the inventory level is below a certain threshold. Finally, the strategy selection module binds the combined strategy sequence with the state vector threshold, pre-encapsulates the parameters required for strategy execution, such as execution intensity and the range of action nodes, and sets adaptive trigger conditions, such as dynamic triggering based on the rate of state change. Finally, it outputs a self-contained strategy instruction package. This instruction package not only contains the execution logic of the strategy but also integrates trigger conditions and parameter adaptive rules, ensuring that the strategy can be accurately adjusted according to the dynamic changes in the supply chain state during actual execution, realizing the fully automated connection from strategy selection to execution.

[0046] In a preferred embodiment of the present invention, the industrial internet management platform further includes: an effect tracking module, used to deploy an effect tracker, monitor the actual effect of the strategy instruction in the supply chain after the strategy instruction is executed, compare the actual execution result with the simulation prediction result to generate a deviation vector, and use the deviation vector to adjust the weight allocation parameters in the state mirror model.

[0047] In this embodiment, the effect tracking module establishes a logical execution timeline based on the event timestamp sequence, aligning the time node of each execution stage with the corresponding supply chain snapshot timestamp. This alignment process eliminates time-dimension bias interference, ensuring the temporal consistency between the observed data and the simulated scenario. Finally, it outputs the anchored execution observation sequence. After obtaining the execution observation sequence, the effect tracking module compares it step-by-step with the predicted trajectory generated during the strategy simulation stage. During the comparison, it focuses on extracting two types of deviations: trend deviation reflects the deviation of the actual execution result from the predicted trajectory in the overall direction of change, while fluctuation deviation reflects the difference between the fluctuation amplitude of local data and the predicted value during execution. The module quantifies and integrates these two types of deviations, constructing a multi-level deviation vector that comprehensively depicts the difference between the actual situation and the prediction, from the overall trend to local fluctuations. To clarify the root causes of the deviations, the effect tracking module back-projects the multi-level deviation vectors onto the strategy context graph and the sources of contributing events. The back-projection process locates the strategy links, supply chain nodes, or original events that contribute significantly to the deviations by associating the deviation vectors with nodes, relationships, and event chains in the graph. Subsequently, the module decomposes the deviations using a preset contribution factor decomposition algorithm, calculates the contribution ratio of each link, node, and event to the overall deviation, and finally outputs an error contribution graph that intuitively presents the distribution pattern and main sources of the deviations.

[0048] Based on the error contribution map, the effect tracking module identifies the dominant error source dimensions. These dimensions may include insufficient simulation accuracy of a specific strategy, unreasonable weight allocation of a certain type of event, or inaccurate characterization of the correlation between supply chain nodes. For these dominant error sources, the effect tracking module dynamically adjusts the internal parameters of the weight allocator in the state mirror model. To accurately quantify the correlation between the magnitude of parameter adjustment and the effect of deviation correction, a calculation expression is abstracted by combining the error contribution ratio and parameter sensitivity:

[0049] This formula is based on the logical relationship between error contribution, parameter sensitivity, and adjustment range. The error contribution ratio determines the priority of parameter adjustment, and parameter sensitivity reflects the degree of influence of the parameter on the deviation. Combining these two factors and calibrating through adjustment coefficients ensures the accuracy and effectiveness of parameter adjustment. In the formula, In the state mirror model, the m-th weight assignment dimension represents the... The adjustment amount of each parameter; This is a global adjustment coefficient, determined by the system stability requirements and convergence speed target, and ranges from 0 to 1; For the m-th dimension The error contribution ratio corresponding to each parameter is derived from the error contribution map, ranging from 0 to 1; For the m-th dimension The sensitivity coefficient of each parameter reflects the degree of influence of parameter changes on the deviation, and is obtained from historical calibration data. Through this parameter adjustment process, the weight allocation of the state mirror model can better fit the actual operating characteristics of the supply chain, reduce the deviation between subsequent simulation predictions and actual execution, and continuously improve the decision-making accuracy and response effectiveness of the entire management platform.

[0050] In a preferred embodiment of the present invention, the industrial internet management platform further includes: a weight adjustment module, used to update the state mirror model using the deviation vector and adjust the model weights through a weight update algorithm.

[0051] In this embodiment, to extract representative deviation patterns from the deviation vectors, the weight adjustment module uses a preset frequency domain eigenvalue decomposition algorithm to process the multi-level deviation vectors. Frequency domain decomposition can transform the deviation signal in the time domain into a superposition of different frequency components, thereby identifying the main fluctuation patterns causing the deviations, such as high-frequency short-term fluctuations and low-frequency long-term trend deviations. Based on the decomposed frequency domain features, the weight adjustment module calculates the deviation covariance matrix. The deviation covariance matrix is ​​used to characterize the linear correlation between different deviation dimensions. Through eigenvalue decomposition of the matrix, deviation feature vectors that reflect the dominant deviation patterns can be extracted. After obtaining the deviation feature vectors, the weight adjustment module maps them to the supply chain network topology. The mapping process associates each element in the feature vector with a node or edge in the topology graph, quantifying the sensitivity of each node in the deviation propagation process. Nodes corresponding to elements with larger values ​​in the feature vectors are more likely to produce significant deviations under that deviation pattern, i.e., they are identified as vulnerable nodes. Based on these vulnerable nodes and their connections in the topology, the module constructs a vulnerability topology map, clearly presenting the distribution of vulnerable nodes and their mutual influence paths. Addressing the differences in node vulnerability revealed by the vulnerability topology map, the weight adjustment module designs a hierarchical weight update strategy. This strategy first divides supply chain nodes into different priority levels based on the node's vulnerability coefficient and error contribution. The vulnerability coefficient is determined by the deviation feature vector mapping result, and the error contribution comes from the error contribution map. Nodes with high vulnerability and large error contribution are assigned to the high-priority level. For nodes at different levels, the weight adjustment module uses differentiated update coefficients, with higher-priority nodes having larger weight adjustment coefficients, which can more significantly correct their sensitivity to deviations. Simultaneously, a propagation constraint coefficient is introduced to limit the propagation range and amplitude of weight adjustments in the network, avoiding state fluctuations in the entire supply chain network caused by large adjustments in local weights. To accurately quantify the weight adjustment amount, a calculation expression is abstracted by combining node vulnerability, error contribution, and propagation constraints.

[0052] This formula is based on the collaborative logic of node characteristics, error impact, and network constraints. It utilizes node vulnerability to determine the necessity of adjustment, error contribution to determine the priority of adjustment, and propagation constraints to ensure the safety of adjustment. These three factors are combined mathematically to achieve precise weight adjustment. In the formula... Representing nodes in the supply chain network To the node The connection weight adjustment amount; The global learning rate is determined by the system's convergence speed and stability requirements, and ranges from 0 to 1. For nodes The vulnerability coefficient reflects the sensitivity of a node to deviations, and is quantified by the deviation feature vector mapping result, ranging from 0 to 1; For nodes The error contribution is derived from the error contribution map and ranges from 0 to 1. For nodes With nodes The shortest path distance in the topology graph; D is the maximum path distance between nodes in the network, used to normalize path distances to generate propagation constraint factors. To ensure that nodes farther away are less affected by the adjustment, the weight adjustment module performs backtracking verification and stability testing on the updated state mirror model after the weight adjustment is completed. Backtracking verification involves re-inputting historical event data into the updated model and comparing the deviation between the model output and the historical actual state to evaluate the effect of weight adjustment on improving model accuracy. Stability testing involves simulating common disturbances in the supply chain, such as order fluctuations and logistics delays, to observe whether the fluctuation range of the model output is within the preset range, ensuring that the weight adjustment will not cause the model to become overly sensitive or unstable. After verification and testing, the weight adjustment module outputs a performance-optimized reinforced state mirror model.

[0053] In a preferred embodiment of the present invention, the industrial internet management platform further includes: an autonomous collaboration module, used to build an autonomous collaboration engine, recalibrate the parameters of the state mirror model and the strategy simulator at predetermined time intervals, and automatically switch the simulation strategy and weight settings based on predefined conditions.

[0054] In this embodiment, the main function of the autonomous collaboration module is to construct an autonomous collaboration engine, which recalibrates the parameters of the state mirror model and the policy simulator at predetermined time intervals, and automatically switches the simulation strategy and weight settings according to predefined conditions to achieve autonomous optimization and stable operation of the system. The engine first calculates the output volatility of the enhanced state mirror model and the coefficient of variation of the policy effect score. The output volatility characterizes the degree of fluctuation of the model output over time, and the coefficient of variation of the policy effect score reflects the degree of dispersion of the effects of different policies. The combination of the two can comprehensively quantify the stability of the system and generate a stability score to determine whether to trigger recalibration. The derivation formula is as follows:

[0055] This formula is based on the logic of evaluating the stability of model output and the stability of policy effect from two dimensions. By assigning weights, the two heterogeneous indicators are integrated into a unified stability score. In the formula, S is the stability score. It is the output volatility of the enhanced state mirror model, which is calculated as the ratio of the standard deviation to the mean of the model's output time series. It is the coefficient of variation of the strategy effectiveness score, which is the ratio of the standard deviation to the mean of the strategy effectiveness score. The weights for output volatility are set according to the system's priority for model stability, ranging from 0 to 1. The weights are the coefficients of variation, and the sum of the two is 1 to ensure that the weights are normalized. When the stability score is lower than the predefined threshold, the recalibration process is triggered.

[0056] Subsequently, the engine uses historical deviation data and error maps to calculate parameter contributions. Historical deviation data records past deviations between model predictions and actual execution, while error maps present the distribution and root causes of errors. By analyzing this data, the contribution of each parameter to the deviation can be quantified, thereby identifying highly sensitive parameters—parameters that significantly affect model output or policy effectiveness. Simultaneously, policy conflict patterns are detected, which are typical scenarios where different policies interfere with each other and cancel each other out. Based on the sensitivity analysis results, the engine generates parameter adjustment maps, clarifying the adjustment direction and magnitude for each highly sensitive parameter. Based on the conflict detection results, a policy switching map is constructed, marking the optimal policy switching path under different policy conflict scenarios. Then, a pre-set incremental method is used to adjust model parameters, avoiding system failure due to sudden parameter changes through small, gradual adjustments. The system is stable; simultaneously, it gradually switches strategies, following the guidance of the strategy switching map, completing the strategy replacement while ensuring system continuity. During this process, the engine monitors the stability score in real time to ensure that parameter adjustments and strategy switching do not lead to a deterioration in system stability. Finally, the engine collects performance data after parameter updates and strategy switching, comprehensively compares it with the state before the update, and analyzes its impact on dimensions such as model accuracy, strategy effectiveness, and system stability. Based on the comparison results, the engine's trigger thresholds, such as the trigger threshold for the stability score, and decision rules, such as the calculation rules for parameter adjustment magnitude and the conditional rules for strategy switching, are dynamically updated. This enables the autonomous collaborative engine to adapt to the dynamic changes in the supply chain, continuously optimize the timing and method of recalibration and strategy switching, and ultimately achieve autonomous collaborative evolution of the industrial internet management platform at the parameter and strategy levels.

[0057] The above description is merely a preferred embodiment of the present invention; however, 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 its improved concepts, should be covered within the scope of protection of the present invention.

Claims

1. An industrial internet management platform based on artificial intelligence and big data, characterized in that, include: The event fingerprint module is used to listen to the raw event stream and generate an event fingerprint for each event. The event fingerprint is based on the timestamp, source device ID, and event type hash value, and outputs an event sequence. The State Mirroring module is used to build a state mirroring model of the supply chain. It maps event sequences into state vectors through a dynamic weight allocator, where the weights are adjusted by a predetermined urgency parameter of the event fingerprint, and outputs a supply chain snapshot. The strategy simulation module is used to start the strategy simulator. It uses supply chain snapshots to perform strategy simulations through a preloaded strategy library. Based on the previous state snapshot, it simulates the state transition probability after applying a strategy perturbation and outputs a strategy effect score. The strategy filtering module is used to import a strategy selector, run a predefined trade-off function to filter strategies based on strategy performance scores, and output strategy instructions. The effect tracking module is used to deploy the effect tracker, monitor the actual effects of the strategy instructions in the supply chain after they are executed, compare the actual execution results with the simulated prediction results to generate a deviation vector, and use the deviation vector to adjust the weight allocation parameters in the state mirror model. The weight adjustment module is used to update the state mirror model using the deviation vector and adjust the model weights through a weight update algorithm. The Autonomous Collaboration Module is used to build an Autonomous Collaboration Engine, which recalibrates the parameters of the state mirror model and the policy simulator at predetermined time intervals, and automatically switches simulation policies and weight settings based on predefined conditions.

2. The industrial internet management platform based on artificial intelligence and big data according to claim 1, characterized in that, The event fingerprint module is also used for: Perform dynamic causal slicing on the original event stream and output event slices; Extract the joint feature modalities of events within the event slice and map them into a composite fingerprint; The composite fingerprint is compared with the fingerprint activity library within a predetermined time window. If there is a conflict, the fingerprint recalibration procedure is initiated to output the resolved fingerprint. The decomposed fingerprints are sorted and a lightweight temporal dependency graph is constructed. The output is an enhanced event sequence as an extension of the event sequence.

3. The industrial internet management platform based on artificial intelligence and big data according to claim 2, characterized in that, The state mirroring module is also used for: Analyze the spatiotemporal context of the fingerprint in the enhanced event sequence and calculate the initial time sensitivity coefficient; Event fingerprints are mapped to the supply chain network topology to simulate the disturbance propagation process, and the instantaneous potential energy value of the state vector is calculated based on the event propagation model. Detect potential energy conflicts and introduce a dynamic coordinator to seek potential energy allocation schemes, and output the converged state vector value. Associating state vector sets with event context tags and encapsulating them into an enhanced supply chain snapshot as an extension of the supply chain snapshot.

4. The industrial internet management platform based on artificial intelligence and big data according to claim 3, characterized in that, The strategy simulation module is also used for: Analyze the context graph of strategies for building enhanced supply chain snapshots; The policy is instantiated as a perturbation source, its propagation process in the graph is simulated, and the quantized impact value of the policy on the state vector is recorded. The detection strategy interferes and the quantitative impact values ​​of the conflict are superimposed using a preset nonlinear function to generate a coordinated impact value distribution. The primary and secondary effects are calculated based on the coordinated distribution of impact values, and a multidimensional score of policy effect is generated as an extension of the policy effect score.

5. The industrial internet management platform based on artificial intelligence and big data according to claim 4, characterized in that, The strategy filtering module is also used for: The strategy effectiveness multidimensional score is mapped to a three-dimensional decision coordinate system and the unit scale is dynamically calibrated. Define a dynamic trade-off window and calculate the weighted distance between the strategy projection point and the predefined ideal decision point; Analyze the complementary characteristics of weighted distance identification strategies, combine the strategy execution sequence, and define the triggering conditions; The policy sequence is bound to the state vector threshold, parameters are pre-encapsulated and adaptive triggering conditions are set, and the output is a self-contained policy instruction package as an extension of the policy instruction.

6. The industrial internet management platform based on artificial intelligence and big data according to claim 5, characterized in that, The effect tracking module is also used for: Capture the actual execution data of the policy instruction package, establish a logical execution timeline based on the event timestamp sequence and aligned with the snapshot timestamp, and output the anchored execution observation sequence; By comparing the actual observed trajectory with the predicted trajectory, trend deviation and fluctuation deviation are extracted to form a multi-level deviation vector.

7. An industrial internet management platform based on artificial intelligence and big data according to claim 6, characterized in that, The effect tracking module is also used for: The multi-level deviation vector is back-projected to the policy context graph and the source of contributing events, the deviation distribution pattern is analyzed and the error contribution graph is output through a preset contribution factor decomposition algorithm; Based on the error contribution map, the dominant error source dimension is identified, and the internal parameters of the weight allocator are dynamically adjusted.

8. The industrial internet management platform based on artificial intelligence and big data according to claim 7, characterized in that, The weight adjustment module is also used for: Based on the error contribution map, a preset frequency domain feature decomposition algorithm is used to process the multi-level deviation vectors, calculate the deviation covariance matrix, and extract the deviation feature vectors. By mapping the deviation feature vectors to the network topology, vulnerable nodes are identified and a vulnerability topology map is constructed. A hierarchical weight update strategy is designed for the vulnerability topology map, employing differentiated update coefficients and introducing propagation constraints; The updated model is backtracked for verification and stability testing, and a reinforced state mirror model is output.

9. An industrial internet management platform based on artificial intelligence and big data as described in claim 8, characterized in that, The autonomous collaboration module is also used for: The output volatility and coefficient of variation of the strategy effect score of the enhanced state mirror model are calculated, and then a stability score is generated. When the stability score exceeds a predetermined threshold, a recalibration process is automatically triggered. Parameter contribution is calculated using historical bias data and error maps to identify highly sensitive parameters and detect policy conflict patterns.

10. An industrial internet management platform based on artificial intelligence and big data according to claim 9, characterized in that, The autonomous collaboration module is also used for: Parameter adjustment mapping is generated based on sensitivity analysis results, and a strategy switching map is constructed based on conflict detection results. The model parameters are adjusted using a preset incremental method, and the strategy is switched gradually, while the stability score is monitored in real time. Collect updated performance data and compare it with the state before the update to update the engine trigger thresholds and decision rules.

Images