Data-driven intelligent sensing and optimization scheduling system and method for aluminum processing

Abstract

This invention discloses a data intelligent sensing and optimization scheduling system and method for aluminum processing, belonging to the field of aluminum processing optimization scheduling technology. The method specifically includes: collecting multi-source heterogeneous data on aluminum processing, preprocessing the collected multi-source heterogeneous data, establishing a composite graph, dynamically generating a feature matrix using the correlation between different types of data, constructing a scheduling model for aluminum processing, performing pre-simulation using a virtual production twin, optimally configuring production resources for aluminum processing, applying the optimization scheduling model to aluminum processing in real time, and continuously optimizing the scheduling model using a reinforcement learning strategy for dynamic adjustment. This application can automatically optimize the allocation and scheduling order of resources according to changes in the manufacturing environment, improving production efficiency and reducing energy consumption and waste.

Description

Technical Field

[0001] This invention belongs to the field of aluminum processing optimization scheduling technology, specifically a data intelligent sensing and optimization scheduling system and method for aluminum processing. Background Technology

[0002] Aluminum processing, a crucial link in the metallurgical and metal materials manufacturing field, encompasses multiple production processes including smelting, extrusion, rolling, heat treatment, and finishing. Due to its lightweight, high strength, corrosion resistance, and ease of formability, aluminum is widely used in aerospace, automotive, construction, and electronics industries. With downstream industries continuously raising their demands for the quality and production efficiency of aluminum products, production scheduling and optimal resource allocation in the aluminum processing process have become critical issues.

[0003] Existing aluminum processing scheduling methods mostly rely on manual experience or rule-based optimization models. In practical applications, these methods have the following shortcomings: they lack intelligent sensing capabilities and the optimization objectives are too singular. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a data intelligent sensing and optimization scheduling system and method for aluminum processing. It can comprehensively utilize multi-source heterogeneous data in the aluminum processing process to establish an intelligent sensing model, realize dynamic understanding of complex working conditions, and combine multi-objective optimization and continuous learning mechanisms to configure and dynamically adjust production resources in real time, thereby improving the flexibility and intelligence level of aluminum processing.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] Collect multi-source heterogeneous data on aluminum processing and preprocess the collected multi-source heterogeneous data;

[0007] Establish a composite graph and dynamically generate a feature matrix by utilizing the correlation between different types of data;

[0008] A scheduling model for aluminum processing is constructed, and a virtual production twin is used for pre-simulation to optimize the allocation of production resources for aluminum processing.

[0009] The optimized scheduling model is applied to aluminum processing in real time, and the scheduling model is continuously optimized and dynamically adjusted using reinforcement learning strategies.

[0010] Specifically, the preprocessing of the collected multi-source heterogeneous data includes:

[0011] Let the data collected by the i-th sensor be x. i The collected multi-source heterogeneous data is fused to obtain fused multi-source heterogeneous data;

[0012] Adaptive sparse coding is used to reduce the dimensionality of fused multi-source heterogeneous data.

[0013] Specifically, the establishment of the composite graph, which dynamically generates a feature matrix using the correlation between different types of data, includes:

[0014] The preprocessed multi-source heterogeneous data is hierarchically semantically segmented to construct independent semantic domains;

[0015] Data entity nodes are constructed within each independent semantic domain, and local graph structures are formed with time series, spatial distribution and logical association as edge relationships.

[0016] Interconnect multiple local graph structures to generate a composite graph with cross-modal interaction capabilities;

[0017] An adaptive attention mechanism is introduced into the composite graph so that the connection strength between different modalities can be dynamically reconstructed as the context changes.

[0018] The resulting composite map is iteratively trained to extract and output the feature matrix characterizing aluminum processing.

[0019] Specifically, the interconnection of multiple local graph structures to generate a composite graph with cross-modal interaction capabilities includes:

[0020] The attribute labels of nodes and edges in each local graph structure are re-encoded;

[0021] Based on preset semantic mapping rules, similar nodes in different semantic domains are matched according to the context, and candidate connection relationships are established between the matching results.

[0022] By filtering relationships, candidate connection relationships are dynamically sparsified, retaining only cross-domain connections with preset relevance.

[0023] By introducing intermediate bridging nodes between different local graphs, heterogeneous information can be transferred through these bridging nodes, generating a composite graph with cross-modal interaction capabilities.

[0024] Specifically, the construction of the aluminum processing scheduling model, using a virtual production twin for pre-simulation, and the optimal allocation of aluminum processing production resources, includes:

[0025] A scheduling model for aluminum processing is constructed based on a virtual twin model of composite graphs and feature matrices corresponding to actual aluminum processing production lines.

[0026] The production targets in the aluminum processing process are decomposed, and quantifiable scheduling factors are constructed for each target.

[0027] Generate resource allocation schemes in the virtual production twin, and arrange the schemes layer by layer according to the combination relationship of sub-objectives;

[0028] The process of arrangement involves temporal evolution, allowing each resource allocation scheme to advance gradually in the virtual space, forming different scheduling evolution paths;

[0029] The key nodes of different scheduling evolution paths are compared, and the resource allocation is dynamically corrected according to the set priority rules.

[0030] The scheduling evolution path, which has been retained after multiple rounds of comparison and correction, is connected with the actual aluminum processing production line and used as the optimal resource allocation scheme for execution.

[0031] Specifically, the temporal evolution during the arrangement process allows each resource allocation scheme to advance gradually in the virtual space, forming different scheduling evolution paths, including:

[0032] The resource allocation scheme is divided into multiple evolution stages according to time segments, and an initial scheduling state is set in each evolution stage;

[0033] Set cross-stage dependency constraints in each evolutionary stage so that the resource allocation results of the previous stage serve as the starting point for the evolution of the next stage.

[0034] External disturbance variables are introduced during the stage evolution process to apply random changes to the original scheduling state and simulate production fluctuations;

[0035] At the end of each evolution stage, a snapshot of the resource allocation scheme is stored, and the snapshots of adjacent stages are compared to track the differentiation process of the scheduling path.

[0036] Through phased iterative cycles, the resource allocation scheme gradually evolves into multiple different scheduling paths in the virtual space.

[0037] Specifically, the comparison of key nodes in different scheduling evolution paths and the dynamic correction of resource allocation through set priority rules include:

[0038] The key nodes of each scheduling evolution path are extracted in chronological order and mapped accordingly in a unified comparison coordinate system.

[0039] In the mapped set of key nodes, resource allocation conflicts between nodes are marked one by one according to pre-defined priority rules;

[0040] The marked conflict nodes are corrected hierarchically, and the resource allocations involved in the target are processed in descending order of importance.

[0041] During the correction process, the resource allocation values ​​of conflict nodes are iteratively updated, and the update results are traced back to the corresponding evolution path;

[0042] After multiple rounds of iterative corrections, a new sequence of key nodes is formed, replacing the corresponding nodes in the original path, thus generating the corrected evolution path.

[0043] Specifically, the process of applying the optimized scheduling model to aluminum processing in real time, and continuously optimizing the scheduling model using reinforcement learning strategies for dynamic adjustment, includes:

[0044] During the operation of the aluminum processing production line, real-time status data is periodically extracted and used as environmental feedback input to the scheduling model;

[0045] In the scheduling model, multiple candidate scheduling actions are generated, and each scheduling action interacts with environmental feedback to form a corresponding state transition;

[0046] The state transition results of each scheduling action are stored as experience fragments, and the decision weights of the scheduling model are updated by loop replay.

[0047] Through multiple rounds of online iteration, the updated scheduling model is reapplied to the production line operation to achieve dynamic adjustment of aluminum processing resource allocation.

[0048] A data intelligent sensing and optimization scheduling system for aluminum processing is used to implement the data intelligent sensing and optimization scheduling method for aluminum processing, including: a data acquisition module, a feature extraction module, a resource optimal allocation module, and a continuous optimization module;

[0049] The data acquisition module is used to collect multi-source heterogeneous data of aluminum processing and to preprocess the collected multi-source heterogeneous data.

[0050] The feature extraction module is used to establish a composite graph and dynamically generate a feature matrix by utilizing the correlation between different types of data.

[0051] The resource optimization module is used to construct a scheduling model for aluminum processing, and to perform a pre-simulation using a virtual production twin to optimize the allocation of production resources for aluminum processing.

[0052] The continuous optimization module is used to apply the optimized scheduling model to aluminum processing in real time, and to continuously optimize the scheduling model and make dynamic adjustments using reinforcement learning strategies.

[0053] Specifically, the resource optimal allocation module includes: a scheduling model construction unit, a scheduling evolution path unit, and a resource optimal allocation unit;

[0054] The scheduling model construction unit constructs a scheduling model for aluminum processing based on a virtual twin model of a composite graph and feature matrix corresponding to the actual aluminum processing production line.

[0055] The scheduling evolution path unit is used to generate resource allocation schemes in the virtual production twin and arrange each scheme layer by layer according to the combination relationship of sub-objectives. During the arrangement process, temporal evolution is carried out so that each resource allocation scheme can be gradually advanced in the virtual space to form different scheduling evolution paths.

[0056] The resource optimal allocation unit is used to connect the scheduling evolution path retained after multiple rounds of comparison and correction with the actual aluminum processing production line as the resource optimal allocation scheme to be executed.

[0057] Compared with the prior art, the beneficial effects of the present invention are:

[0058] This invention proposes a data-driven intelligent sensing and optimized scheduling method for aluminum processing. It collects multi-source heterogeneous data generated during aluminum processing and constructs a composite graph to extract the feature matrix of the production process. Based on this, a scheduling model is built, and a virtual production twin is introduced for pre-simulation. Different resource allocation schemes are dynamically simulated and selected. Finally, the scheduling model is continuously iterated, enabling the scheduling scheme to be applied in real-time on the actual production line and continuously corrected based on feedback. Through this method, comprehensive perception and intelligent decision-making for complex aluminum processing conditions are achieved. Under multiple constraints such as energy consumption, production capacity, equipment lifespan, and delivery cycle, flexible allocation and dynamic optimization of production resources are realized, improving the intelligence level and adaptability of the aluminum processing process to complex scenarios. Attached Figure Description

[0059] Figure 1 The flowchart of the data intelligent sensing and optimization scheduling method for aluminum processing provided by the present invention is shown below.

[0060] Figure 2 The flowchart for optimizing scheduling of multi-source heterogeneous data provided by this invention;

[0061] Figure 3 This is a schematic diagram of the evolution path correction and update provided by the present invention;

[0062] Figure 4 This is an architecture diagram of the intelligent data sensing and optimization scheduling system for aluminum processing provided by the present invention. Detailed Implementation

[0063] To facilitate understanding of the technical means, creative features, and achieved objectives and effects of this invention, it should be noted in the description of this invention that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "number one," "number two," and "number three" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. The invention will be further described below in conjunction with specific embodiments.

[0064] Example 1

[0065] Please see Figures 1-3 The present invention provides an embodiment of a data intelligent sensing and optimization scheduling method for aluminum processing, comprising the following specific steps:

[0066] Step S1: Collect multi-source heterogeneous data on aluminum processing through a sensor network, including equipment operation data, process parameters, environmental monitoring data, and operator behavior records, and preprocess the collected multi-source heterogeneous data.

[0067] The specific steps of step S1 are as follows:

[0068] Step S101: Set the data collected by the i-th sensor to x. i The collected multi-source heterogeneous data is fused using the following formula:

[0069] ;

[0070] Where X represents the fused multi-source heterogeneous data, w i Let represent the fusion weight of the i-th sensor data, and m represent the number of sensors. The weights can be dynamically adjusted according to the maximum entropy principle to ensure the maximum amount of information.

[0071] The fused multi-source heterogeneous data X is in matrix form. R represents a real number, n represents the number of heterogeneous data from multiple sources, and d represents the dimension of each heterogeneous data from multiple sources.

[0072] Step S102: Use adaptive sparse coding to reduce the dimensionality of the fused multi-source heterogeneous data.

[0073] Step S2: Establish a composite graph and dynamically generate a feature matrix by utilizing the correlation between different types of data.

[0074] The specific steps of step S2 are as follows:

[0075] Step S201: Perform hierarchical semantic segmentation on the preprocessed multi-source heterogeneous data to construct independent semantic domains.

[0076] In this embodiment, it is first necessary to classify the different types of data generated during the aluminum processing production process, such as equipment operation data, process parameters, environmental monitoring data, and operator behavior records. Since the various types of data have significant differences in dimensionality, temporal sequence, value range, and expression method, the above-mentioned different types of data are defined as independent semantic domains, so that the data in each semantic domain has a relatively consistent semantic expression space.

[0077] Step S202: Construct data entity nodes in each independent semantic domain, and form a local graph structure with time series, spatial distribution and logical association as edge relationships.

[0078] In this embodiment, after hierarchical semantic segmentation, the data within each independent semantic domain is processed into nodes, that is, the original data entries or their aggregation results are abstracted into data entity nodes. For example, in the equipment operation semantic domain, temperature, pressure, and rotational speed records at different time periods are respectively used as nodes; in the process parameter semantic domain, the set parameters at different process stages are used as nodes; and in the environmental monitoring semantic domain, key observation values ​​in the air humidity, noise level, and temperature change curves are used as nodes. In this way, the original raw data is transformed into basic components in a graph structure.

[0079] Step S203: Interconnect multiple local graph structures to generate a composite graph with cross-modal interaction capabilities.

[0080] The specific steps of step S203 are as follows:

[0081] Step S2031: Re-encode the attribute labels of nodes and edges in each local graph structure.

[0082] Step S2032: Based on the preset semantic mapping rules, similar nodes in different semantic domains are matched according to the context, and candidate connection relationships are established between the matching results.

[0083] In this embodiment, during the modeling phase, data nodes within each semantic domain are first assigned label information, such as category labels, timing identifiers, or process stage attributes. By combining these labels with the context, potentially similar nodes between different semantic domains can be identified. For example, the temperature node in the equipment operation semantic domain and the cooling timing node in the process parameter semantic domain both point to heat treatment-related processes in the context, thus indicating that they have a high degree of semantic similarity.

[0084] The establishment of semantic mapping rules is based on the common features and process logic relationships between semantic domains. First, the node features in different semantic domains need to be vectorized to represent them in a unified semantic space. Then, multi-dimensional comparisons are performed based on information such as timestamps, process stages, and spatial locations in the context. If two nodes overlap in the time dimension, are interdependent in the process stage, and have the potential for interaction in the spatial location, they can be considered to meet the similarity matching conditions. Through this process, candidate node pairs can be found between different semantic domains.

[0085] Step S2033: Through relation filtering, the candidate connection relations are dynamically sparsified, and only cross-domain connections with preset correlation are retained.

[0086] In this embodiment, a preset correlation threshold is introduced into the candidate relation set. Each candidate connection is comprehensively scored based on its context matching degree, process logic dependency strength and time consistency, and compared with the threshold. Only when the score of a candidate connection exceeds the threshold is it retained as a cross-domain valid connection, and the rest of the connections are discarded.

[0087] Step S2034: Introduce intermediate bridging nodes between different local graphs to allow some heterogeneous information to be transferred through the bridging nodes, generating a composite graph with cross-modal interaction capabilities.

[0088] In this embodiment, the intermediate bridging node does not directly correspond to a certain type of original data. Instead, it is generated by abstracting a set of nodes with potential connections in multiple semantic domains to create an intermediate entity with neutral semantics. For example, a heat treatment link bridging node can be established between the equipment operation semantic domain and the process parameter semantic domain. This node gathers the correlation information between the equipment temperature curve and the cooling process setting, thereby providing a unified carrying channel for the heterogeneous information exchange between the two types of nodes.

[0089] The establishment of bridging nodes relies on the implicit similarity features between heterogeneous data. First, it is necessary to identify node groups with the same context labels or process attributes in different local graphs. Second, through feature compression and normalization, the multidimensional data in the node groups are mapped to an abstract semantic space. Finally, based on this semantic space, corresponding bridging nodes are generated and connected to the composite graph as new nodes.

[0090] Step S204: Introduce an adaptive attention mechanism in the composite graph to dynamically reconstruct the connection strength between different modal data as the context changes.

[0091] In this embodiment, adjustable weights are assigned to each cross-modal connection, enabling the degree of association to be automatically adjusted under different context conditions. For example, when production is in the heat treatment stage, the weights of the equipment temperature node and the process cooling parameter node are dynamically enhanced, while in the assembly stage, the weights of the personnel operation node and the equipment start-up and shutdown status node are given higher priority. Through this mechanism, the cross-modal connection relationships in the graph can be continuously reconstructed as the scene and working conditions change.

[0092] Step S205: Iteratively train the formed composite map, extract and output the feature matrix characterizing aluminum processing.

[0093] In this embodiment, firstly, during the initial training phase, the node features of each semantic domain are input into the propagation layer of the graph network, and the first aggregation of features is completed through the message passing mechanism. Subsequently, during the iterative training process, multiple rounds of information exchange are continuously performed on each node and its adjacent edges, so that the potential relationship between cross-modalities is gradually strengthened or weakened. After multiple iterations, the feature representation in the composite graph gradually converges, and finally a stable node representation vector, i.e., the feature matrix, is formed.

[0094] Step S3: Construct a scheduling model for aluminum processing, use a virtual production twin for pre-simulation, and optimize the allocation of production resources for aluminum processing.

[0095] The specific steps of step S3 are as follows:

[0096] Step S301: Based on the composite graph and feature matrix and the virtual twin model corresponding to the actual aluminum processing production line, construct the scheduling model for aluminum processing.

[0097] In this embodiment, the physical elements of the production line are first reconstructed in the virtual space, including equipment units, process flow, energy input and material flow, and these elements are mapped one by one with the semantic nodes and relation edges in the feature matrix. Then, the mapping relationship is embedded in the virtual twin so that the virtual twin has a physical structure that matches the actual production line, and a scheduling model is constructed.

[0098] Step S302: Decompose the production targets in the aluminum processing process and construct quantifiable scheduling factors for each.

[0099] In this embodiment, the overall objective is broken down into multiple sub-objectives such as energy consumption control, equipment utilization rate, process connection efficiency, product quality stability, and delivery time limit. Each sub-objective is quantified into a calculable scheduling factor through a data-driven approach. For example, the energy consumption objective is mapped to the power consumption level per unit output, the equipment utilization rate objective is represented as the uptime per unit time, and the process connection efficiency is quantified by the waiting time of the preceding and following processes.

[0100] Step S303: Generate resource allocation schemes in the virtual production twin and arrange the schemes layer by layer according to the combination relationship of sub-objectives.

[0101] In this embodiment, different resource allocation strategies are set in the twin environment, such as equipment scheduling priority, energy allocation ratio, process start timing, and personnel allocation method. Then, multiple feasible resource configuration schemes are automatically generated according to these strategies, and each scheme corresponds to a complete set of scheduling factor values. At the outermost layer, all schemes are grouped based on the primary objective (such as delivery time limit). Then, within each group, the schemes are further refined and arranged layer by layer based on secondary objectives such as energy consumption, quality, and equipment utilization rate.

[0102] Step S304: During the arrangement process, temporal evolution is performed so that each resource allocation scheme is gradually advanced in the virtual space, forming different scheduling evolution paths.

[0103] The specific steps of step S304 are as follows:

[0104] Step S3041: Divide the resource allocation scheme into multiple evolution stages according to time segments, and set the initial scheduling state in each evolution stage.

[0105] In this embodiment, based on the natural rhythm of the production line operation and the process connection rules, the production cycle is broken down into stages such as smelting, forming, heat treatment, testing and assembly, and each stage is regarded as an evolution stage. In each evolution stage, an initial scheduling state is set for the candidate scheme, that is, the equipment availability, material inventory, personnel distribution and energy allocation at the beginning of the stage are specified.

[0106] Step S3042: Set cross-stage dependency constraints in each evolution stage so that the resource allocation result of the previous stage serves as the starting point for the evolution of the next stage.

[0107] In this embodiment, the role of cross-stage dependency constraints is to model the dynamic scheduling problem as a continuous time-series link. Unlike independent stage simulations, it requires that the resource allocation of each stage not only affects the output of the current stage, but also constrains the initial conditions of subsequent stages through the state transfer mechanism.

[0108] Step S3043: In the stage evolution process, external disturbance variables are added to apply random changes to the original scheduling state to simulate production fluctuations.

[0109] In this embodiment, during the operation of each evolution stage, in addition to inheriting the state parameters of the previous stage, multiple disturbance factors are artificially set, such as energy supply fluctuations, equipment performance degradation, personnel operation deviations, or sudden changes in environmental parameters. These disturbance factors are superimposed on the original scheduling state in a random manner, so that the resource allocation and process connection of each stage in the evolution process not only depend on the planned results under ideal conditions, but are also subject to the intervention of random disturbances, which is closer to the real uncertainty in a complex production environment.

[0110] Step S3044: At the end of each evolution stage, a snapshot of the state of the resource allocation scheme is stored, and the snapshots of adjacent stages are compared to track the differentiation process of the scheduling path.

[0111] In this embodiment, the snapshot content includes key parameters such as equipment operating status, energy consumption level, process completion status, material inventory data, and personnel scheduling results. Each stage snapshot is stored in the database of the virtual twin in a structured form, forming a phased state sequence. Subsequently, before entering the next evolution stage, the snapshots of two adjacent stages are compared to identify differences in resource allocation and process connection, and to track the differentiation trajectory of the scheduling path in multi-stage iteration.

[0112] Step S3045: Through phased iterative cycles, the resource allocation scheme gradually evolves into multiple different scheduling paths in the virtual space.

[0113] In this embodiment, the candidate schemes in the initial stage are used as the starting point for iteration, and the output state of the previous stage is used as the input condition for the next stage. Subsequently, perturbation variables and priority correction rules are continuously introduced in the evolution process, so that the same initial scheme gradually produces differentiated results after multiple iterations. After multiple rounds of cyclic iteration, different resource allocation schemes gradually evolve into multiple independent scheduling paths, and each path corresponds to a feasible production operation logic chain.

[0114] like Figure 3 As shown, step S305: compare the key nodes of different scheduling evolution paths, and dynamically correct the resource allocation through the set priority rules.

[0115] exist Figure 3 In the process, multiple scheduling evolution paths are generated and compared in the virtual twin. Each scheduling evolution path consists of multiple key nodes, which are arranged in chronological order to represent the state changes of the resource allocation scheme at different stages. The key nodes of scheduling evolution path 1 and scheduling evolution path 2 are linearly distributed in time, representing the continuous advancement of the evolution state under different schemes.

[0116] In scheduling evolution path 3, a conflict node for resource allocation appeared. This node corresponds to multiple paths competing for the same resource in the same time slice. In order to avoid scheduling failure, in scheduling evolution path 4, the conflict node was first marked, and the resource allocation was hierarchically adjusted according to the priority rules in the correction stage. In the corrected path, the original conflict node was replaced with the processed node, forming an evolution chain with higher consistency.

[0117] The specific steps of step S305 are as follows:

[0118] Step S3051: Extract the key nodes of each scheduling evolution path in chronological order and map them accordingly in a unified comparison coordinate system.

[0119] In this embodiment, in each evolution path, core nodes that can represent changes in production status are selected, such as equipment switching points, energy consumption peak points, inventory critical points, or process completion points. Then, these key nodes are arranged in chronological order to form a time-series node sequence for that path. The node sequences of different paths are uniformly mapped to a comparison coordinate system to ensure that each path corresponds to nodes under the same time reference frame.

[0120] The specific steps are as follows: 1. Identify nodes that have a decisive impact on resource allocation or process connection in each scheduling evolution path and extract them as a set of candidate key nodes; 2. Divide the set of candidate key nodes according to time segments and set anchor points in the segmented time series to ensure that the node order is consistent with the time sequence of the production process; 3. Establish a correspondence between time anchor points between the key node sets of different paths, and fill in or map nodes that cannot be directly matched through interpolation or alignment rules; 4. Generate a cross-path node comparison table in a unified comparison coordinate system, and mark the key nodes of different paths under the same time sequence accordingly; 5. Adjust the comparison table through multiple iterations to keep the mapping relationship of cross-path key nodes stable in the global scope, and use it as the basis input for subsequent difference analysis and correction.

[0121] Step S3052: In the mapped set of key nodes, resource allocation conflicts between nodes are marked one by one according to the pre-set priority rules.

[0122] In the set of key nodes under the comparison coordinates, the nodes are first classified according to resource type (such as energy, equipment, materials, personnel); then, according to the pre-set priority rules, the resource occupancy in the same time slice or adjacent time slices is compared; when two or more nodes are found to have mutually exclusive requirements for the same type of resource, the node relationship is marked as conflict; for example, if a key device is called by multiple paths at the same time, the device node is marked as a conflict node.

[0123] Step S3053: Perform hierarchical correction on the marked conflict nodes, and process the resource allocation involved in the target in descending order.

[0124] In this embodiment, priority levels are set according to the importance of the objectives. For example, safety production-related objectives are placed at the highest level, delivery time-bound objectives are at the second highest level, and objectives such as energy consumption optimization and cost control are in descending order. Subsequently, in the set of conflict nodes, starting from the highest-level objective, the allocation rights of resources are determined one by one. When a resource is locked by a high-level objective, that resource will not be allocated to low-level objective nodes in the same time period. The conflict correction process unfolds step by step under the hierarchical priority logic until all conflict nodes have been corrected.

[0125] Step S3054: During the correction process, the resource allocation values ​​of the conflicting nodes are iteratively updated, and the update results are backtracked to the corresponding evolution path.

[0126] In this embodiment, after each conflict node completes the priority determination, the resource allocation value is adjusted successively to satisfy the total resource constraints and the logical dependencies between nodes. Each iteration update generates a new resource allocation state, which is synchronously written into the corresponding evolution path and serves as the starting condition for subsequent stage deductions.

[0127] Step S3055: After multiple rounds of iterative correction, a new sequence of key nodes is formed, which replaces the corresponding nodes in the original path, generating the corrected evolution path.

[0128] In this embodiment, the corrected nodes are extracted one by one and reordered according to the time coordinates of the original path to ensure that the sequence structure maintains the correspondence with the original path. Then, the corresponding nodes in the original path are gradually replaced by the new key node sequence so that the evolution path no longer contains uncorrected conflicting nodes, but is composed of corrected consistent nodes. After the replacement is completed, the evolution path is updated to a corrected version that can be directly applied to subsequent scheduling optimization.

[0129] Step S306: Connect the scheduling evolution path that has been retained after multiple rounds of comparison and correction with the actual aluminum processing production line and use it as the optimal resource allocation scheme for execution.

[0130] Step S4: Apply the optimized scheduling model to aluminum processing in real time, and continuously optimize the scheduling model using reinforcement learning strategies for dynamic adjustment.

[0131] The specific steps of step S4 are as follows:

[0132] Step S401: During the operation of the aluminum processing production line, real-time status data is periodically extracted and used as environmental feedback input to the scheduling model.

[0133] Step S402: Generate multiple candidate scheduling actions in the scheduling model, and interact with the environment feedback to form corresponding state transitions.

[0134] Step S403: Store the state transition results of each scheduling action as an experience fragment, and update the decision weights of the scheduling model by loop replay.

[0135] Step S404: Through multiple rounds of online iteration, the updated scheduling model is reapplied to the production line operation to achieve dynamic adjustment of aluminum processing resource allocation.

[0136] Example 2

[0137] Please see Figure 4 Another embodiment of the present invention provides a data intelligent sensing and optimization scheduling system for aluminum processing, comprising: a data acquisition module, a feature extraction module, a resource optimal allocation module, and a continuous optimization module;

[0138] The data acquisition module is used to collect multi-source heterogeneous data of aluminum processing and to preprocess the collected multi-source heterogeneous data.

[0139] The feature extraction module is used to establish a composite graph and dynamically generate a feature matrix by utilizing the correlation between different types of data.

[0140] The resource optimization module is used to construct a scheduling model for aluminum processing, and to perform a pre-simulation using a virtual production twin to optimize the allocation of production resources for aluminum processing.

[0141] The continuous optimization module is used to apply the optimized scheduling model to aluminum processing in real time, and to continuously optimize the scheduling model and make dynamic adjustments using reinforcement learning strategies.

[0142] The optimal resource allocation module includes: a scheduling model construction unit, a scheduling evolution path unit, and a optimal resource allocation unit;

[0143] The scheduling model construction unit constructs a scheduling model for aluminum processing based on a virtual twin model of a composite graph and feature matrix corresponding to the actual aluminum processing production line.

[0144] The scheduling evolution path unit is used to generate resource allocation schemes in the virtual production twin and arrange each scheme layer by layer according to the combination relationship of sub-objectives. During the arrangement process, temporal evolution is carried out so that each resource allocation scheme can be gradually advanced in the virtual space to form different scheduling evolution paths.

[0145] The resource optimal allocation unit is used to connect the scheduling evolution path retained after multiple rounds of comparison and correction with the actual aluminum processing production line as the resource optimal allocation scheme to be executed.

[0146] The parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of corresponding technical solutions in the prior art have not been described in detail to avoid excessive elaboration.

[0147] It will be understood by those skilled in the art that the above descriptions are merely preferred examples of the invention and are not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions made within the spirit and principles of the invention should be included within the scope of protection of the invention.

Claims

1. A data intelligent sensing and optimized scheduling method for aluminum processing, characterized in that, include: Collect multi-source heterogeneous data on aluminum processing and preprocess the collected multi-source heterogeneous data; Establish a composite graph and dynamically generate a feature matrix by utilizing the correlation between different types of data; A scheduling model for aluminum processing is constructed, and a virtual production twin is used for pre-simulation to optimize the allocation of production resources for aluminum processing. The optimized scheduling model is applied to aluminum processing in real time, and the scheduling model is continuously optimized and dynamically adjusted using reinforcement learning strategies. The aforementioned aluminum processing scheduling model utilizes a virtual production twin for pre-simulation to optimally allocate aluminum processing production resources, including: A scheduling model for aluminum processing is constructed based on a virtual twin model of composite graphs and feature matrices corresponding to actual aluminum processing production lines. The production targets in the aluminum processing process are decomposed, and quantifiable scheduling factors are constructed for each target. Generate resource allocation schemes in the virtual production twin, and arrange the schemes layer by layer according to the combination relationship of sub-objectives; The process of arrangement involves temporal evolution, allowing each resource allocation scheme to advance gradually in the virtual space, forming different scheduling evolution paths; The key nodes of different scheduling evolution paths are compared, and the resource allocation is dynamically corrected according to the set priority rules. The scheduling evolution path, which has been retained after multiple rounds of comparison and correction, is connected with the actual aluminum processing production line and used as the optimal resource allocation scheme for execution. The process of sequential evolution during the arrangement allows each resource allocation scheme to advance gradually in the virtual space, forming different scheduling evolution paths, including: The resource allocation scheme is divided into multiple evolution stages according to time segments, and an initial scheduling state is set in each evolution stage; Set cross-stage dependency constraints in each evolutionary stage so that the resource allocation results of the previous stage serve as the starting point for the evolution of the next stage. External disturbance variables are introduced during the stage evolution process to apply random changes to the original scheduling state and simulate production fluctuations; At the end of each evolution stage, a snapshot of the resource allocation scheme is stored, and the snapshots of adjacent stages are compared to track the differentiation process of the scheduling path. Through phased iterative cycles, the resource allocation scheme gradually evolves into multiple different scheduling paths in the virtual space; The comparison of key nodes in different scheduling evolution paths and the dynamic correction of resource allocation based on set priority rules include: The key nodes of each scheduling evolution path are extracted in chronological order and mapped accordingly in a unified comparison coordinate system. In the mapped set of key nodes, resource allocation conflicts between nodes are marked one by one according to pre-defined priority rules; The marked conflict nodes are corrected hierarchically, and the resource allocations involved in the target are processed in descending order of importance. During the correction process, the resource allocation values ​​of conflict nodes are iteratively updated, and the update results are traced back to the corresponding evolution path; After multiple rounds of iterative corrections, a new sequence of key nodes is formed, replacing the corresponding nodes in the original path, thus generating the corrected evolution path.

2. The data intelligent sensing and optimization scheduling method for aluminum processing as described in claim 1, characterized in that, The preprocessing of the collected multi-source heterogeneous data includes: Let the data collected by the i-th sensor be x. i The collected multi-source heterogeneous data is fused to obtain fused multi-source heterogeneous data; Adaptive sparse coding is used to reduce the dimensionality of fused multi-source heterogeneous data.

3. The data intelligent sensing and optimization scheduling method for aluminum processing as described in claim 2, characterized in that, The process of establishing a composite graph, which dynamically generates a feature matrix using the correlations between different types of data, includes: The preprocessed multi-source heterogeneous data is hierarchically semantically segmented to construct independent semantic domains; Data entity nodes are constructed within each independent semantic domain, and local graph structures are formed with time series, spatial distribution and logical association as edge relationships. Interconnect multiple local graph structures to generate a composite graph with cross-modal interaction capabilities; An adaptive attention mechanism is introduced into the composite graph so that the connection strength between different modalities can be dynamically reconstructed as the context changes. The resulting composite map is iteratively trained to extract and output the feature matrix characterizing aluminum processing.

4. The data intelligent sensing and optimization scheduling method for aluminum processing as described in claim 3, characterized in that, The step of interconnecting multiple local graph structures to generate a composite graph with cross-modal interaction capabilities includes: The attribute labels of nodes and edges in each local graph structure are re-encoded; Based on preset semantic mapping rules, similar nodes in different semantic domains are matched according to the context, and candidate connection relationships are established between the matching results. By filtering relationships, candidate connection relationships are dynamically sparsified, retaining only cross-domain connections with preset relevance. By introducing intermediate bridging nodes between different local graphs, heterogeneous information can be transferred through these bridging nodes, generating a composite graph with cross-modal interaction capabilities.

5. The data intelligent sensing and optimization scheduling method for aluminum processing as described in claim 4, characterized in that, The process of applying the optimized scheduling model to aluminum processing in real time, and continuously optimizing the scheduling model using reinforcement learning strategies for dynamic adjustment, includes: During the operation of the aluminum processing production line, real-time status data is periodically extracted and used as environmental feedback input to the scheduling model; In the scheduling model, multiple candidate scheduling actions are generated, and each scheduling action interacts with environmental feedback to form a corresponding state transition; The state transition results of each scheduling action are stored as experience fragments, and the decision weights of the scheduling model are updated by loop replay. Through multiple rounds of online iteration, the updated scheduling model is reapplied to the production line operation to achieve dynamic adjustment of aluminum processing resource allocation.

6. A data intelligent sensing and optimization scheduling system for aluminum processing, used to implement the data intelligent sensing and optimization scheduling method for aluminum processing as described in any one of claims 1-5, characterized in that, include: The module includes a data acquisition module, a feature extraction module, a resource optimization module, and a continuous optimization module. The data acquisition module is used to collect multi-source heterogeneous data of aluminum processing and to preprocess the collected multi-source heterogeneous data. The feature extraction module is used to establish a composite graph and dynamically generate a feature matrix by utilizing the correlation between different types of data. The resource optimization module is used to construct a scheduling model for aluminum processing, and to perform a pre-simulation using a virtual production twin to optimize the allocation of production resources for aluminum processing. The continuous optimization module is used to apply the optimized scheduling model to aluminum processing in real time, and to continuously optimize the scheduling model and make dynamic adjustments using reinforcement learning strategies.

7. The data intelligent sensing and optimization scheduling system for aluminum processing as described in claim 6, characterized in that, The optimal resource allocation module includes: a scheduling model construction unit, a scheduling evolution path unit, and a optimal resource allocation unit; The scheduling model construction unit constructs a scheduling model for aluminum processing based on a virtual twin model of a composite graph and feature matrix corresponding to the actual aluminum processing production line. The scheduling evolution path unit is used to generate resource allocation schemes in the virtual production twin and arrange each scheme layer by layer according to the combination relationship of sub-objectives. During the arrangement process, temporal evolution is carried out so that each resource allocation scheme can be gradually advanced in the virtual space to form different scheduling evolution paths. The resource optimal allocation unit is used to connect the scheduling evolution path retained after multiple rounds of comparison and correction with the actual aluminum processing production line as the resource optimal allocation scheme to be executed.

Images