Self-adaptive production scheduling system based on artificial intelligence

Abstract

The invention relates to the technical field of intelligent manufacturing and production management, in particular to a self-adaptive production scheduling system based on artificial intelligence, which comprises a data acquisition and reference construction module for analyzing process data to construct a directed acyclic graph representing a non-interference state as a reference map; the theoretical disturbance simulation module is used for converting the interference rule into a graph change instruction, generating a theoretical damaged state graph and obtaining a theoretical difference feature vector; the theoretical difference feature vector comprises, but is not limited to, a vector form obtained after a difference matrix is expanded according to rows or columns in terms of mathematical representation; the real deviation extraction module is used for collecting real-time state data to construct a real-time operation state diagram and calculating a real difference feature vector; a double-domain coupling decision module; an adaptive scheduling execution module; according to the method, the causal relationship is verified by comparing the form of theoretical deduction and actual observation, non-systematic noise is effectively filtered, and accurate response to real faults is realized while the stability of the production rhythm is maintained.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and production management technology, specifically to an adaptive production scheduling system based on artificial intelligence. Background Technology

[0002] In the application scenario of adaptive production scheduling in discrete manufacturing workshops, the production system relies on precise resource allocation and timing logic to ensure the timely delivery of orders and the stability of production rhythm. The scheduling management system usually needs to combine enterprise resource planning data and workshop feedback to perceive the execution progress of production flow in real time. For responding to and rescheduling production disruptions, existing solutions generally adopt a time-threshold-based triggering architecture. This involves acquiring the start and finish times of processes through data acquisition terminals deployed on the production line, calculating the deviation between actual and planned progress using statistical analysis methods, and directly using events where time delays exceed a preset threshold as trigger conditions to initiate the rescheduling algorithm. While this approach is feasible to some extent in a single, linear production environment, its over-reliance on single statistical data in the time dimension and lack of logical backtracking to the physical mechanisms of failures makes it prone to misidentifying non-systematic noise caused by sensor signal jitter, minor random fluctuations in manual operation, or network latency as valid systematic disturbances. This leads to the issuance of incorrect scheduling instructions. Furthermore, this extensive management approach, which focuses only on delay duration while ignoring the topology of the disturbance in the production network, struggles to eliminate local fluctuations that do not affect final delivery. This causes the production line to frequently fall into meaningless scheduling jitter, disrupting the continuity and stability of production operations. Therefore, how to establish a closed-loop control mechanism with causal relationship verification capabilities, effectively filter out non-systematic environmental noise, and accurately identify the real fault modes based on the dual constraints of physics and logic, thereby improving the robustness and accuracy of scheduling decisions, has become an urgent technical problem to be solved. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides an adaptive production scheduling system based on artificial intelligence. Specifically, the technical solution of this invention includes: The data acquisition and benchmark construction module is used to acquire product bill of materials and standard process route data, analyze physical operation cycle time and logical dependency relationship, and construct a directed acyclic graph structure representing the allocation of production resources and the time sequence flow under interference-free conditions as a benchmark graph. The theoretical disturbance simulation module is used to call a pre-stored disturbance rule database, convert the generated disturbance events into corresponding graph structure change instructions or weight modification parameters, apply the graph structure change instructions or weight modification parameters to the baseline graph, generate a theoretical damaged state graph, and calculate the difference between the adjacency matrix or eigenvector of the theoretical damaged state graph and the baseline graph to obtain a theoretical difference feature vector; the theoretical difference feature vector is mathematically represented, including but not limited to, the vector form obtained by expanding the difference matrix by rows or columns. The reality deviation extraction module is used to collect real-time status data of each process in the production site, construct the current real-time operation status map, and calculate the reality difference feature vector of the real-time operation status map relative to the benchmark map. The dual-domain coupled decision module is used to map the real difference feature vector and the theoretical difference feature vector to the same feature space, calculate the similarity value between the two, and determine whether the current deviation is a valid system disturbance based on the similarity value. An adaptive scheduling execution module is used to generate a new scheduling scheme with the corrected production operation sequence as the target, in response to the result of the dual-domain coupled decision module determining that the system disturbance is valid.

[0004] Preferably, the data acquisition and benchmark construction module includes: The logic parsing unit is used to identify the process node IDs in the production process and the predecessor and successor relationships between the process nodes based on the product bill of materials and the standard process route data. The graph generation unit is used to construct the directed acyclic graph as a reference graph, wherein the nodes of the reference graph represent processes, the edges represent the material and information flow paths between processes, the weight values ​​of the edges represent the theoretical shortest operation time, and the reference graph serves as the initial state for deviation calculation.

[0005] Preferably, the theoretical perturbation simulation module includes: The parameter conversion unit is used to query the interference rule database and extract the node ID and edge weight modification rule corresponding to the interference event. The shadow simulation unit is used to select the corresponding node ID and edge weight modification rules based on the current risk prediction information and apply them to the copy of the baseline graph in an independent simulation process to generate the theoretical damage state graph. The theoretical difference unit is used to compare the theoretical damaged state map with the baseline map by using a graph editing distance algorithm or matrix subtraction, extract the structural changes or weight changes caused by the disturbance, and generate the theoretical difference feature vector.

[0006] Preferably, the parameter conversion unit is used to perform the following operations: For interference events that cause extended processing time, the edge weight modification rule is defined as multiplying the original weight value of the corresponding edge by a time-based wear coefficient to obtain a new weight value. For material shortage interference events that cause process interruption, the map structure change instruction is defined as deleting the connecting edges between specified process nodes in the baseline map, thereby breaking the map connectivity.

[0007] Preferably, the reality deviation extraction module includes: The state reconstruction unit is used to match the real-time state data of each process in the production site to the corresponding nodes in the graph structure, update the weight values ​​of the edges generated by actual execution and the connection status of the edges, and generate the real-time operation state graph containing real-time time and actual execution order. The reality difference unit is used to calculate the difference in edge weight values ​​and edge existence between the adjacency matrix of the real-time operation state graph and the adjacency matrix of the reference graph, and generate the reality difference feature vector.

[0008] Preferably, the dual-domain coupled decision module includes: The feature embedding unit is used to encode the real-time operation state map and the theoretical damage state map using a graph neural network model, and extract high-dimensional feature embeddings of the real difference feature vector and the theoretical difference feature vector; A similarity calculation unit is used to calculate the cosine similarity value between the actual difference feature vector and the theoretical difference feature vector; The isomorphism determination unit is used to preset a similarity threshold and determine whether the cosine similarity value is greater than the similarity threshold; if so, it determines that the current deviation has causal transmission characteristics and generates a signal that is confirmed as a valid system disturbance.

[0009] Preferably, the isomorphism determination unit is further used for: If the cosine similarity value is less than or equal to the similarity threshold, then the current deviation is determined to be a non-systematic random deviation. The adaptive scheduling execution module is also used to maintain the current scheduling scheme and not trigger a rescheduling operation in response to the result determined to be a non-systematic random deviation.

[0010] Preferably, the adaptive scheduling execution module includes: The decision generation unit is used to calculate and output the corrected production operation sequence and resource allocation scheme based on a multi-objective optimization function after confirming the effective system disturbance. The instruction issuing unit is used to convert the revised production operation sequence into control instructions and issue them to the production site.

[0011] Compared with the prior art, the present invention has the following beneficial effects: 1. This system constructs a closed-loop control loop based on a synthetic analysis method, effectively solving the technical problem that traditional scheduling schemes cannot distinguish between physical random noise and systematic causal disturbances. Unlike the extensive mode that triggers production rescheduling solely based on the length of production delays, this scheme uses a dual-domain coupled decision module to compare theoretically deduced fault patterns with observed deviation patterns, and verifies causal relationships using the similarity values ​​of the feature space. This mechanism can effectively filter out non-systematic noise caused by sensor jitter or minor fluctuations in worker operations, ensuring that scheduling commands are issued only for changes in real physical constraints, thereby achieving accurate responses to real faults while maintaining the stability of production rhythm.

[0012] 2. This system establishes a high-purity physical reference system based on a directed acyclic graph, providing an absolute zero-point scale for deviation calculation. By strictly defining the edge weights of the reference graph as the theoretical shortest operating time derived from equipment physical parameters or standard operating procedures, rather than using historical averages that include management inefficiencies, this method reconstructs the ideal physical form of the production line in the digital space. This design eliminates data pollution caused by general management habits, enabling the calculated deviation vector to purely reflect physical anomalies, laying a solid data foundation for high-precision disturbance identification.

[0013] 3. This system introduces a mechanism-based theoretical perturbation simulation and parameter transformation mechanism, realizing the transformation from qualitative expert knowledge to quantitative mathematical operators. The system does not rely on historical statistical data, but uses an exponential decay function and matrix connectivity cut-off operation to pre-simulate the distortion effect of interference events such as tool wear or material shortage on the topology of the graph in shadow memory. By generating a theoretical difference matrix with physical semantics, this method upgrades the subsequent matching process to mechanism-based pattern recognition, giving the system the ability to predict fault patterns.

[0014] 4. This system employs feature embedding and isomorphic judgment techniques based on graph convolutional neural networks to construct a fault filter capable of identifying spatiotemporal distortions in the production flow. By performing high-dimensional feature encoding and cosine similarity calculation on the difference matrix between reality and theory, the system can not only capture delays in a single process but also quantify the spatial distribution characteristics of delays in the network. This judgment logic based on graph isomorphism avoids scheduling jitter caused by overreacting to random fluctuations, ensuring that production rescheduling is triggered only when the topological form of the actual deviation is highly isomorphic to the theoretical simulation, thus guaranteeing the seriousness and continuity of production instructions. Attached Figure Description

[0015] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. Example 1:

[0017] Please see Figure 1 An adaptive production scheduling system based on artificial intelligence includes: The data acquisition and benchmark construction module is used to acquire product bill of materials and standard process route data, analyze physical operation cycle time and logical dependency relationship, and construct a directed acyclic graph structure representing the allocation of production resources and the time sequence flow under interference-free conditions as a benchmark graph. The theoretical disturbance simulation module is used to call the pre-stored disturbance rule database, convert the production disturbance events into corresponding graph structure change instructions or weight modification parameters, apply the graph structure change instructions or weight modification parameters to the baseline graph, generate a theoretical damaged state graph, and calculate the difference between the adjacency matrix or eigenvector of the theoretical damaged state graph and the baseline graph, so as to obtain the theoretical difference eigenvector; the theoretical difference eigenvector is mathematically represented, including but not limited to the vector form obtained by expanding the difference matrix by rows or columns. The reality deviation extraction module is used to collect real-time status data of each process in the production site, construct the current real-time operation status map, and calculate the reality difference feature vector of the real-time operation status map relative to the baseline map. The dual-domain coupled decision module is used to map the real difference feature vector and the theoretical difference feature vector to the same feature space, calculate the similarity value between the two, and determine whether the current deviation is a valid system perturbation based on the similarity value. The adaptive scheduling execution module is used to generate a new scheduling scheme with the corrected production operation sequence as the target, in response to the result of the dual-domain coupled decision module determining that the system disturbance is valid.

[0018] This embodiment details the overall architecture and execution logic of an AI-based adaptive production scheduling system. This system aims to solve the technical challenge of traditional scheduling schemes failing to distinguish between physical random noise and systematic causal disturbances. The system initiates a data acquisition and benchmark construction module. This module acquires product bill of materials and standard process route data through an enterprise resource planning (ERP) or manufacturing execution system (MES) interface, analyzes physical operation cycles and logical dependencies, and constructs a directed acyclic graph (DAG) structure representing the allocation and temporal flow of production resources under interference-free conditions as a benchmark graph. This graph is not only a static dataset but also an absolute zero-point reference system for the production process. A theoretical disturbance simulation module then intervenes. This module does not rely on historical statistical data but instead calls a pre-stored disturbance rule database to convert production disturbance events into corresponding graph structure change instructions or weight modification parameters. These instructions or parameters are then applied to the benchmark graph to generate a theoretically damaged state graph. Based on this, the system calculates the difference between the adjacency matrix or eigenvector of the theoretical damaged state map and the baseline map, and obtains the theoretical difference eigenvector. The theoretical difference eigenvector is mathematically represented, including but not limited to the vector form obtained by expanding the difference matrix by rows or columns. At the same time, the actual deviation extraction module collects real-time status data of each process in the production site through IoT gateways deployed in the production site, constructs the current real-time operation status map, and calculates the actual difference eigenvector of the real-time operation status map relative to the baseline map. The dual-domain coupled decision module executes the core judgment logic, maps the actual difference eigenvector and the theoretical difference eigenvector to the same feature space, calculates the similarity value between the two, and determines whether the current deviation is a valid system disturbance based on the similarity value. In response to the result of the dual-domain coupled decision module determining that it is a valid system disturbance, the adaptive scheduling execution module generates a new scheduling scheme with the corrected production operation sequence as the target. This embodiment constructs a closed-loop control loop based on a synthetic analysis method. In complex discrete manufacturing workshop scenarios, the system no longer triggers rescheduling solely based on the duration of production delays. Instead, it verifies causal relationships by comparing theoretically deduced fault patterns with observed deviation patterns. This mechanism effectively filters out non-systematic noise caused by sensor jitter, minor fluctuations in worker operations, etc., ensuring that scheduling commands are issued only for changes in real physical constraints. Thus, while maintaining the stability of the production rhythm, it achieves accurate responses to real faults. Example 2:

[0019] The data acquisition and benchmark building module includes: The logic parsing unit is used to identify the process node IDs and the predecessor and successor relationships between process nodes in the production process based on the product bill of materials and standard process route data. The graph generation unit is used to construct a directed acyclic graph as a baseline graph. In the baseline graph, the nodes represent processes, the edges represent the flow paths of logistics and information between processes, the weight values ​​of the edges represent the theoretical shortest operation time, and the baseline graph serves as the initial state for deviation calculation.

[0020] In this embodiment, the reference map is mathematically represented in the form of a weighted adjacency matrix, where the non-zero element values ​​are... This represents the theoretical shortest operation time corresponding to the edge weight; This embodiment is a detailed description of the data acquisition and benchmark construction module in Embodiment 1, focusing on how to establish a high-purity physical reference system. The logic parsing unit in this module, based on the product bill of materials and standard process route data, executes a deep parsing algorithm to identify the process node IDs and the predecessor-successor relationships between process nodes, thereby establishing the topological skeleton of the task. The graph generation unit constructs a directed acyclic graph as the benchmark graph, where nodes represent processes, and edges represent the material and information flow paths between processes. The edge weights are strictly defined as representing the theoretical shortest operation time, derived from the extreme values ​​of equipment physical parameters or standard operating procedures (SOPs), rather than historical averages, aiming to eliminate data pollution caused by inefficient management. The benchmark graph serves as the initial state for deviation calculation, providing a frictionless ideal benchmark for subsequent differential calculations. This embodiment reconstructs the ideal physical form of the production line in the digital space by constructing a benchmark map that eliminates historical redundancy and loss. This design ensures that the subsequently calculated deviation vector purely reflects the physical anomalies rather than the continuation of management habits, thus laying a data foundation for high-precision disturbance identification. Example 3:

[0021] The theoretical perturbation simulation module includes: The parameter conversion unit is used to query the interference rule database and extract the node ID and edge weight modification rules corresponding to the interference event. The shadow simulation unit is used to select the corresponding node ID and edge weight modification rules based on the current risk prediction information and apply them to a copy of the baseline graph in an independent simulation process to generate a theoretical damage state graph. The theoretical difference unit is used to compare the theoretical damaged state map with the baseline map by using graph editing distance algorithm or matrix subtraction, extract the structural changes or weight changes caused by the disturbance, and generate the theoretical difference feature vector.

[0022] The graph editing distance algorithm quantifies the degree of topological shift by calculating the weighted algebraic sum of the minimum sequence of atomic operations required to transform a theoretically damaged state graph into a baseline graph. The atomic operations include adding / deleting nodes, adding / deleting edges, and modifying the weight values ​​of edges. Risk prediction information is pushed in real time by the external equipment health management system through the API interface, or it is generated by this system based on the historical completion time series of each process through the Long Short-Term Memory (LSTM) network. The parameter conversion unit is used to perform the following operations: For interference events that cause extended processing time, the edge weight modification rule is defined as multiplying the original weight value of the corresponding edge by a time-based wear coefficient to obtain a new weight value. To address material shortage interference events that cause process interruptions, the diagram structure change instruction is defined as deleting the connecting edges between specified process nodes in the baseline diagram, thus breaking the diagram connectivity.

[0023] This embodiment further deepens the theoretical disturbance simulation module in Embodiment 1, particularly regarding the specific execution of the simulation mechanism described in Embodiment 1 and the parameterization rules of this embodiment. The parameter conversion unit in this module queries the disturbance rule database to extract the node IDs and edge weight modification rules corresponding to the disturbance events. For disturbance events that lead to extended processing time, such as tool wear or equipment aging, the parameter conversion unit does not rely on vague qualitative descriptions but instead uses an exponential wear function based on performance degradation leading to increased processing time to define the edge weight modification rules as follows: ; in, This is expressed as the calculated theoretical working time weight under the damaged state; Represented as standard operating hours as defined in the baseline map; This is expressed as the cumulative operating time of the equipment; Expressed as the inherent wear coefficient of the equipment; Represented as the aging rate factor; Its dimension is the reciprocal of the unit of time, that is... To ensure the exponent term The value is dimensionless. In response to material shortage interference events that cause process interruptions, the parameter conversion unit defines the graph structure change instruction as changing the corresponding element in the adjacency matrix of the baseline graph. The value is set to 0, thus mathematically severing connectivity; the shadow simulation unit, in an independent simulation process, generates the adjacency matrix of the theoretical damaged state diagram based on the current risk prediction information and applying the above rules. Theoretical difference units perform matrix subtraction operations: ,in, Using the adjacency matrix of the baseline graph, a theoretical difference matrix quantifying the degree of topological distortion is obtained. This matrix provides the basic structured data for subsequent feature embedding; The non-zero elements in the table correspond to the process nodes or logical edges that have been disturbed, and their values ​​reflect the degree of deviation of the physical properties. This embodiment introduces a specific exponential decay model and matrix difference operations to transform qualitative domain expert knowledge into quantitative mathematical operators, enabling the system to predict fault patterns. By pre-simulating the distortion effect of faults on the topological structure of the graph in shadow memory, the system can generate structured features with physical semantics, which makes the subsequent matching process no longer a blind data fitting, but a mechanism-based pattern recognition. Example 4:

[0024] The reality deviation extraction module includes: The state reconstruction unit is used to match the real-time state data of each process in the production site to the corresponding nodes in the graph structure, update the weight values ​​of the edges generated by actual execution and the connection status of the edges, and generate a real-time operation state graph containing real-time time and actual execution order. The real-time difference unit is used to calculate the difference in edge weights and edge existence between the adjacency matrix of the real-time operation status graph and the adjacency matrix of the baseline graph, and generate a real-time difference feature vector.

[0025] This embodiment further specifies the reality deviation extraction module in Embodiment 1, detailing the mapping process from the physical world to the digital twin. The state reconstruction unit matches the real-time status data of each process in the production site, including but not limited to the start / completion timestamps and photoelectric gate trigger signals uploaded by the PLC, to the corresponding nodes in the graph structure. During this process, the unit updates the weight values ​​of the edges generated by actual execution and the connection status of the edges, generating a real-time operation status graph that includes real-time time and actual execution order. The reality difference unit calculates the difference between the two in terms of edge weight values ​​and edge existence by comparing the adjacency matrix of the real-time operation status graph with the adjacency matrix of the reference graph, generating a reality difference matrix. This matrix not only records the delay of a single process, but also captures the spatial distribution characteristics of the delay in the network through the matrix structure. This embodiment achieves the capture of holographic images of the production site by reconstructing the operation status diagram in real time. By using the difference calculation of the adjacency matrix, the system can accurately extract the distortion of the production flow in both time and space dimensions. Whether it is a hidden process skip or an explicit equipment stoppage, it can be transformed into a unified mathematical expression, providing a standardized input for subsequent dual-domain coupling. Example 5:

[0026] The dual-domain coupled decision module includes: The feature embedding unit is used to encode the real-time operation state map and the theoretical damage state map using a graph neural network model, and extract high-dimensional feature embeddings of the real difference feature vector and the theoretical difference feature vector. The similarity calculation unit is used to calculate the cosine similarity between the actual difference feature vector and the theoretical difference feature vector; The isomorphism determination unit is used to preset a similarity threshold and determine whether the cosine similarity value is greater than the similarity threshold; if so, it determines that the current deviation has causal transmission characteristics and generates a signal that is confirmed as a valid system disturbance.

[0027] The isomorphism determination unit is also used for: If the cosine similarity value is less than or equal to the similarity threshold, the current deviation is determined to be a non-systematic random deviation. The adaptive scheduling execution module is also used to maintain the current scheduling scheme and not trigger a rescheduling operation in response to results determined to be non-systematic random deviations.

[0028] This embodiment is a logical deepening of the dual-domain coupled decision module in Embodiment 1, covering the similarity determination and negative decision logic of the embodiment; the feature embedding unit uses a graph convolutional neural network (GCN) model to encode the real difference matrix and the theoretical difference matrix respectively, and the specific inter-layer propagation formula is defined as: ; in, The ReLU activation function is used. The graph convolutional neural network model needs to be trained offline before application: various fault samples from historical production processes are collected as the training set, and the classification label is determined by whether the fault is a systemic disturbance. The cross-entropy loss function is used to minimize the deviation between the predicted similarity and the actual causal label, thereby iteratively optimizing the weight matrix. ; To incorporate the difference adjacency matrix with self-loops, The input is either the actual difference matrix or the theoretical difference matrix. It is the identity matrix; for The degree matrix; For the first The feature matrix of the layer; For the first The trainable weight matrix of the layer; For the first Output of the graph feature matrix of the layer; Based on this propagation logic, the node embedding matrix is ​​obtained through aggregation calculation using two GCN layers, and then a global average pooling layer is used to compress and aggregate the node dimensions, thereby extracting the reality difference feature vectors. Eigenvectors of theoretical differences Full-image high-dimensional feature embedding; similarity calculation unit calculates the cosine similarity value between the two. : ; in, The cosine similarity value between the calculated real-world difference and the theoretical difference in the feature space; The isomorphism determination unit presets a similarity threshold. Similarity threshold The preferred value range is... to Specifically, sensitivity calibration is performed based on the noise level of the workshop sensors. The specific calibration process is as follows: acquire historical operating data under fault-free conditions in the workshop, calculate the corresponding cosine similarity distribution, and then apply this distribution... The upper bound of the confidence interval is determined as the initial similarity threshold. The higher the noise level, the higher the similarity threshold. Adjust accordingly to enhance the filtering effect; respond to the cosine similarity value. The system determines that the current deviation has causal transmission characteristics, meaning that the topological form of the actual deviation is highly isomorphic to the theoretically predicted fault form, and generates a signal confirming it as a valid system disturbance; conversely, in response to the judgment... The system determines that the current deviation is a non-systematic random deviation; in this case, the adaptive scheduling execution module maintains the current scheduling plan and does not trigger a rescheduling operation. The core of this embodiment lies in constructing a fault filter based on graph isomorphism; the mathematical process of feature extraction is clarified through the GCN formula, and the degree of matching between chaos and fault is quantified through the cosine similarity formula, which avoids scheduling jitter caused by overreacting to random fluctuations and ensures the seriousness and continuity of production instructions. Example 6:

[0029] The adaptive scheduling execution module includes: The decision generation unit is used to calculate and output the corrected production operation sequence and resource allocation scheme based on a multi-objective optimization function after confirming the effective system disturbance. The instruction issuing unit is used to convert the revised production operation sequence into control instructions and issue them to the production site.

[0030] This embodiment is a concretization of the execution level of the adaptive scheduling execution module in Embodiment 1; the decision generation unit is used to calculate a new scheduling scheme based on a multi-objective optimization function after confirming a valid system disturbance. The definition is as follows: ; The fitness function is directly defined as the multi-objective optimization function described above. The reciprocal of, that is .

[0031] in, This represents the minimum value of the optimization objective; This is represented as the maximum completion time, Makespan. Represented as workpiece The actual completion time; Represented as workpiece Delivery time; Represented as weighting coefficients, satisfying ; This indicates the penalty for total delay time; The algorithm employs a genetic algorithm for solving the problem, using a process index sequence encoding method. Chromosomes represent a feasible job sequence; the fitness function is directly defined as the aforementioned multi-objective optimization function. The reciprocal of, that is The objective function is achieved by maximizing the fitness value. The solution minimizes the problem; through selection, crossover, and mutation operations, using the confirmed perturbation information as constraints, the output is... The modified production operation sequence and resource allocation scheme minimize the value; the instruction issuing unit converts the modified production operation sequence into control instructions and issues them to the production site to complete the execution of actions in the physical world; This embodiment realizes an automated closed loop from perception to execution; by directly converting the optimized scheduling scheme into low-level control commands, the system eliminates the information gap between the decision-making layer and the execution layer, ensuring that after confirming systemic risks, the production line can switch to the optimal risk-avoidance operation mode at a speed of milliseconds.

[0032] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An adaptive production scheduling system based on artificial intelligence, characterized in that, include: The data acquisition and benchmark construction module is used to acquire product bill of materials and standard process route data, analyze physical operation cycle time and logical dependency relationship, and construct a directed acyclic graph structure representing the allocation of production resources and the time sequence flow under interference-free conditions as a benchmark graph. The theoretical disturbance simulation module is used to call a pre-stored disturbance rule database, convert the generated disturbance events into corresponding graph structure change instructions or weight modification parameters, apply the graph structure change instructions or weight modification parameters to the baseline graph, generate a theoretical damaged state graph, and calculate the difference between the adjacency matrix or eigenvector of the theoretical damaged state graph and the baseline graph to obtain a theoretical difference feature vector; the theoretical difference feature vector is mathematically represented, including but not limited to, the vector form obtained by expanding the difference matrix by rows or columns. The reality deviation extraction module is used to collect real-time status data of each process in the production site, construct the current real-time operation status map, and calculate the reality difference feature vector of the real-time operation status map relative to the benchmark map. The dual-domain coupled decision module is used to map the real difference feature vector and the theoretical difference feature vector to the same feature space, calculate the similarity value between the two, and determine whether the current deviation is a valid system disturbance based on the similarity value. An adaptive scheduling execution module is used to generate a new scheduling scheme with the corrected production operation sequence as the target, in response to the result of the dual-domain coupled decision module determining that the system disturbance is valid.

2. The adaptive production scheduling system based on artificial intelligence according to claim 1, characterized in that: The data acquisition and benchmark construction module includes: The logic parsing unit is used to identify the process node IDs in the production process and the predecessor and successor relationships between the process nodes based on the product bill of materials and the standard process route data. The graph generation unit is used to construct the directed acyclic graph as a reference graph, wherein the nodes of the reference graph represent processes, the edges represent the material and information flow paths between processes, the weight values ​​of the edges represent the theoretical shortest operation time, and the reference graph serves as the initial state for deviation calculation.

3. The adaptive production scheduling system based on artificial intelligence according to claim 1, characterized in that: The theoretical perturbation simulation module includes: The parameter conversion unit is used to query the interference rule database and extract the node ID and edge weight modification rule corresponding to the interference event. The shadow simulation unit is used to select the corresponding node ID and edge weight modification rules based on the current risk prediction information and apply them to the copy of the baseline graph in an independent simulation process to generate the theoretical damage state graph. The theoretical difference unit is used to compare the theoretical damaged state map with the baseline map by using a graph editing distance algorithm or matrix subtraction, extract the structural changes or weight changes caused by the disturbance, and generate the theoretical difference feature vector.

4. The adaptive production scheduling system based on artificial intelligence according to claim 3, characterized in that: The parameter conversion unit is used to perform the following operations: For interference events that cause extended processing time, the edge weight modification rule is defined as multiplying the original weight value of the corresponding edge by a time-based wear coefficient to obtain a new weight value. For material shortage interference events that cause process interruption, the map structure change instruction is defined as deleting the connecting edges between specified process nodes in the baseline map, thereby breaking the map connectivity.

5. The adaptive production scheduling system based on artificial intelligence according to claim 1, characterized in that: The reality deviation extraction module includes: The state reconstruction unit is used to match the real-time state data of each process in the production site to the corresponding node in the graph structure, update the weight value of the edge generated by the actual execution and the connection state of the edge, and generate the real-time operation state graph containing the real-time time and the actual execution order. The reality difference unit is used to calculate the difference between the adjacency matrix of the real-time operation state graph and the adjacency matrix of the reference graph in terms of edge weight values ​​and edge existence, and generate the reality difference feature vector.

6. The adaptive production scheduling system based on artificial intelligence according to claim 1, characterized in that: The dual-domain coupled decision module includes: The feature embedding unit is used to encode the real-time operation state map and the theoretical damage state map using a graph neural network model, and extract high-dimensional feature embeddings of the real difference feature vector and the theoretical difference feature vector; A similarity calculation unit is used to calculate the cosine similarity value between the actual difference feature vector and the theoretical difference feature vector; The isomorphism determination unit is used to preset a similarity threshold and determine whether the cosine similarity value is greater than the similarity threshold; if so, it determines that the current deviation has causal transmission characteristics and generates a signal that is confirmed as a valid system disturbance.

7. The adaptive production scheduling system based on artificial intelligence according to claim 6, characterized in that: The isomorphism determination unit is also used for: If the cosine similarity value is less than or equal to the similarity threshold, then the current deviation is determined to be a non-systematic random deviation. The adaptive scheduling execution module is also used to maintain the current scheduling scheme and not trigger a rescheduling operation in response to the result determined to be a non-systematic random deviation.

8. The adaptive production scheduling system based on artificial intelligence according to claim 1, characterized in that: The adaptive scheduling execution module includes: The decision generation unit is used to calculate and output the corrected production operation sequence and resource allocation scheme based on a multi-objective optimization function after confirming the effective system disturbance. The instruction issuing unit is used to convert the revised production operation sequence into control instructions and issue them to the production site.

Images