MES digital collaborative management method and system based on deep learning

Abstract

The invention relates to the technical field of deep learning, and discloses an MES digital collaborative management method and system based on deep learning, and the method comprises the steps: collecting multi-source heterogeneous data, and constructing a dimensionless multi-dimensional data set; analyzing a dynamic association relationship among the data items through node feature embedding and edge relationship learning, and generating a production line operation dependency relationship graph; performing collaborative anomaly detection and root cause positioning in combination with the dynamic association weight between the data items to obtain causal association data between the abnormal event and the data parameters; constructing a digital twinborn simulation environment to simulate an influence path of heterogeneous data parameter intervention production disturbance on collaborative anomaly so as to evaluate an anomaly influence result; constructing an iterative scheduling model according to a deep learning framework, and iteratively generating a multi-objective optimized collaborative scheduling strategy to realize dynamic configuration and exception prevention among heterogeneous data; therefore, collaborative optimization management of production operation parameter configuration and abnormity prevention of multi-source heterogeneous data in large-scale production with high dimension, high complexity and dynamic change can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of deep learning, and more specifically, to a MES digital collaborative management method and system based on deep learning. Background Art

[0002] With the rapid development of Industry 4.0 and intelligent manufacturing, the Manufacturing Execution System (MES), as a key link connecting the Enterprise Resource Planning System (ERP) and the Production Control System (PCS), plays an important role in the digital transformation of small and medium-sized enterprises. Currently, the digital management systems of small and medium-sized enterprises mainly adopt discrete point analysis methods, regarding the production system as a network structure composed of independent nodes, such as production workshops, warehouses, and logistics centers, and business processes; most of these heterogeneous data are used for optimization decisions based on existing production management technologies, and it is difficult to handle large-scale production management work with high dimensions, high complexity, and dynamic changes.

[0003] Therefore, how to achieve collaborative optimization management of production operation parameter configuration and anomaly prevention for multi-source heterogeneous data in large-scale production with high dimensions, high complexity, and dynamic changes has become an urgent problem to be solved. Summary of the Invention

[0004] The present invention provides a MES digital collaborative management method and system based on deep learning, which solves the technical problem of how to achieve collaborative optimization management of production operation parameter configuration and anomaly prevention for multi-source heterogeneous data in large-scale production with high dimensions, high complexity, and dynamic changes in the prior art.

[0005] The present invention provides a MES digital collaborative management method and system based on deep learning, including:

[0006] In a first aspect, a MES digital collaborative management method based on deep learning includes the following steps:

[0007] Collect multi-source heterogeneous data, and through preprocessing and normalization processing, construct a dimensionless multi-dimensional data set;

[0008] Map the dimensionless multi-dimensional data set to graph structure nodes to obtain node features, and define the connection exchange, information flow, and dependency relationships between parameters in the dimensionless multi-dimensional data set as edge relationships;

[0009] Through node feature embedding and edge relationship learning, analyze the dynamic association relationships between data items in the dimensionless multi-dimensional data set, and generate a production line operation dependency relationship map;

[0010] Build a collaborative analysis mechanism for multi-source heterogeneous data to obtain the dynamic correlation weights for collaboration between multi-source heterogeneous data modalities, and combine with the dependency relationship graph for anomaly detection and root cause location to obtain the causal association information between anomaly events and data modalities;

[0011] Based on the causal association information, build a digital twin simulation environment to simulate the influence path of heterogeneous data parameter intervention on production disturbances on collaborative anomalies, so as to evaluate the anomaly influence results of parameter adjustment;

[0012] Build an iterative scheduling model according to the dimensionless multi-dimensional data set, the dependency relationship graph and the causal association information, and combine the anomaly influence evaluation results with the real-time production status to iteratively generate a collaborative scheduling strategy for multi-objective optimization, realizing the dynamic configuration and anomaly prevention between multi-source heterogeneous data.

[0013] Furthermore, collect multi-source heterogeneous data, and through preprocessing and normalization processing, build a dimensionless multi-dimensional data set, including:

[0014] Collect multi-source heterogeneous data during the operation of production and manufacturing in real time and perform data preprocessing to obtain a preprocessed data set;

[0015] Perform standardization transformation on the preprocessed data set and perform data cleaning operations to obtain a clean data set;

[0016] Perform time series alignment processing on the clean data set, and perform normalization processing after collecting at a preset period to obtain a dimensionless multi-dimensional data set.

[0017] Furthermore, through node feature embedding and edge relationship learning, analyze the dynamic association relationship between each data item in the dimensionless multi-dimensional data set, and generate a production line operation dependency relationship graph, including:

[0018] Map the dimensionless multi-dimensional data set to graph structure nodes, and use multi-source heterogeneous data parameters as node features;

[0019] Define the connection exchange, information flow and dependency relationship between multi-source heterogeneous data parameters as edge relationships;

[0020] Use the graph attention network structure for node feature embedding to capture the local correlation of edge relationships;

[0021] Through the edge relationship feature learning mechanism, quantify the interaction intensity between multi-source heterogeneous data modalities;

[0022] According to the local correlation and interaction intensity, perform iterative propagation through a multi-layer graph network to analyze the dynamic association relationship between each data item in the dimensionless multi-dimensional data set, and generate a production line operation dependency relationship graph including the association status between heterogeneous data, production operation parameters and parameter quality indicators.

[0023] Further, the time-series graph convolutional network structure is used to process the dynamically changing multi-source heterogeneous data parameter states, so as to fuse the time-series features and the graph network topological features to obtain a fusion result;

[0024] Based on the fusion result, the static attributes of each data item parameter in the dimensionless multi-dimensional dataset are concatenated and dimension-reduced with the dynamic time-series data to generate a fusion feature vector for a single node, which serves as the basic input for constructing the production line operation dependency relationship graph;

[0025] Explicit edges are defined by the interaction intensity connection, and at the same time, the cosine similarity between the fusion feature vectors of each node is obtained to generate implicit edges. After fusion, a weighted adjacency matrix is formed, and the dynamic edge weights are output together with the node features;

[0026] By learning the dynamic edge weights between node features, aggregating multiple attention results to update the dynamic edge weights, and fusing the neighbor node features to generate updated node features.

[0027] According to the dynamic edge weights and the updated node features, a production line operation dependency relationship graph is jointly constructed to reveal the explicit associations and implicit dynamic dependencies between multi-source heterogeneous data items during production operation, so as to support production line operation decision-making.

[0028] Further, a multi-source heterogeneous data collaborative analysis mechanism is constructed to obtain the dynamic association weights for collaboration between multi-source heterogeneous data modalities, and anomaly detection and root cause localization are performed in combination with the dependency relationship graph to obtain the causal association information between anomaly events and data modalities, including:

[0029] Feature extraction and representation learning are performed on the states between each data item in the dimensionless multi-dimensional dataset to obtain the correlation matrix between multi-source heterogeneous data modalities;

[0030] The contribution degrees between multi-source heterogeneous data modalities are dynamically allocated and adjusted by using the dynamic edge weights to obtain the dynamic association weights for collaboration between multi-source heterogeneous data modalities;

[0031] The association influence degrees between multi-source heterogeneous data modalities are captured according to the dynamic association weights;

[0032] By: , to generate a dynamic association weight matrix;

[0033] Among them, represents the dynamic association weight matrix, Q represents the query matrix, represents the key matrix at the T-th moment, V represents the value matrix, d represents the feature dimension, and softmax represents the normalization function;

[0034] The dynamic association weight matrix is combined with the production line operation dependency relationship graph to construct an enhanced graph structure topology;

[0035] Identifying node features and edge relationships that deviate from the normal pattern on the enhanced graph structure topology;

[0036] By: Performing anomaly detection to obtain anomaly recognition results;

[0037] Wherein, represents the anomaly score, x represents the input feature, represents the reconstructed feature, represents the square of the L2 norm;

[0038] Based on the anomaly recognition results, through backpropagation and subgraph extraction, locate the root cause nodes of the anomaly to obtain root cause location data;

[0039] Using causal inference to quantify the causal association strength between the anomaly recognition results and the root cause location data, generating causal association information, where the causal association strength represents the direct influence degree of the anomaly recognition results on the root cause location data, and is determined by obtaining the difference between the mutual information and the conditional mutual information;

[0040] The causal association strength of the causal association information is:

[0041] ;

[0042] Wherein, C(X - Y) represents the causal association strength of the anomaly recognition result X on the location data Y, I(X, Y) represents the interaction information between the anomaly recognition result X and the location data Y, I(X, Y|Z) represents the conditional interaction information between the anomaly recognition result X and the location data Y under the given constraint condition Z, PA(Y) represents the set of all potential parent nodes of the location data Y, and {X} represents the quantified data of the anomaly recognition result.

[0043] Furthermore, based on the causal association information, construct a digital twin simulation environment to simulate the influence path of heterogeneous data parameter intervention production disturbances on collaborative anomalies, so as to evaluate the anomaly influence results of parameter adjustment, including:

[0044] Map the obtained causal association information into the digital twin simulation environment to establish the corresponding relationship between the physical entity and the digital model;

[0045] Define the interaction rules in the digital twin simulation environment according to the production line operation dependency graph to ensure that the digital twin simulation environment accurately reflects the dynamic characteristic data of actual production operation;

[0046] Design a parameter intervention experimental plan according to the dynamic characteristic data of production operation, and simulate the production disturbance scenario by fixing and adjusting the key parameters in the digital twin simulation environment;

[0047] Based on the production disturbance scenario, trace the propagation path of the production disturbance, record the state changes of the characteristics of each node, and analyze the impact of the production disturbance scenario on the collaborative anomalies between multi-source heterogeneous data modalities;

[0048] Comprehensively evaluate the suppression or exacerbation effects of the production disturbance scenarios simulated by various parameter adjustment schemes to form a quantitative evaluation result of the anomaly impact.

[0049] Furthermore, construct an iterative scheduling model based on the dimensionless multi-dimensional data set, the dependency relationship graph, and the causal association information, including:

[0050] Construct a state representation space based on the dimensionless multi-dimensional data set, and map the current state during production operation into a high-dimensional feature vector, which includes multi-dimensional information data such as equipment operation parameters, quality indicators, energy consumption data, and safety monitoring data;

[0051] Use the production line operation dependency relationship graph to define the constraint condition matrix, and the constraint condition matrix represents the precedence dependencies, resource sharing, and conflict relationships between production units to generate a scheduling relationship matrix that meets the production operation constraint requirements;

[0052] Construct an iterative scheduling model based on the multi-dimensional information data, the scheduling relationship matrix, and the causal association information. The iterative scheduling model specifically includes:

[0053] Generate structural data from the multi-dimensional information data, the scheduling relationship matrix, and the causal association information, and encode the structural data into sequence data;

[0054] Input the sequence data into the iterative scheduling model; the iterative scheduling model includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer, and transfer the intermediate representation data of multiple hidden layers to the output layer, and the output layer iteratively outputs the scheduling strategy prediction results corresponding to the state changes between the multi-source heterogeneous data parameters;

[0055] Input at least two multi-dimensional information data, scheduling relationship matrix, and causal association information data items obtained in real time into the iterative scheduling model, and iteratively output the scheduling strategy prediction results corresponding to the state changes between the multi-source heterogeneous data parameters obtained in real time.

[0056] Furthermore, combine the anomaly impact evaluation results with the real-time production state, and iteratively generate a multi-objective optimized collaborative scheduling strategy to achieve dynamic configuration and anomaly prevention between multi-source heterogeneous data, including:

[0057] Collect key parameters of the real-time production operation status, including equipment operation status, quality inspection data, energy consumption indicators, and safety monitoring information, to form a multi-dimensional real-time status vector; and extract the current topological structure and constraint conditions from the production line operation dependency graph, and combine with the historical abnormal pattern library to capture the abnormal risks of the real-time status;

[0058] Based on the captured abnormal risks, perform multi-step predictions on the future state evolution to identify potential abnormal development trends; for the detected abnormal risk points, generate a design parameter intervention effect scoring matrix according to the abnormal impact assessment results;

[0059] According to the design parameter intervention effect scoring matrix, start a multi-objective optimization decision-making mechanism, set the key indicators of production efficiency, quality stability, energy utilization rate, and safety risk as optimization objectives, and dynamically adjust the weights of each objective according to the current production task priority;

[0060] And search for the optimal scheduling strategy under the constraint conditions, through:

[0061] ;

[0062] To balance the short-term intervention effect and the long-term stable scheduling strategy;

[0063] Among them, J(θ) represents the comprehensive optimization objective, represents the production efficiency objective, represents the quality stability objective, represents the energy utilization rate objective, represents the safety risk objective, 、 、 and respectively represent the dynamic weight coefficients of the four objectives of production efficiency, quality stability, energy utilization rate, and safety risk, and satisfy ;

[0064] Based on the prediction results of the iterative scheduling model, execute the collaborative scheduling strategy deployment, decompose the optimized scheduling instructions into specific execution instructions at the equipment level, process level, and resource level; safely transfer the specific execution instructions to each execution unit through the middleware layer, and at the same time establish an instruction execution confirmation mechanism to achieve dynamic configuration and abnormal prevention among multi-source heterogeneous data;

[0065] Establish a real-time monitoring and feedback mechanism, continuously track the execution effect of the scheduling strategy, and collect data on changes in key performance indicators; when detecting execution deviations or new abnormal events, trigger a quick response mechanism, and select fine-tuning or re-planning strategies according to the degree of deviation.

[0066] Furthermore, the collaborative scheduling strategy includes:

[0067] Decompose the scheduling problem into three levels: strategic, tactical, and operational, and separately handle long-term planning, medium-term scheduling, and real-time response to anomalies among heterogeneous data, forming a complete hierarchical collaborative decision-making framework;

[0068] By realizing the bidirectional integration of constraint scenarios and guiding scenarios, use the differentiable planning layer to unify hard constraints and soft constraints into the same framework, and dynamically adjust the constraint weights according to the real-time production situation to maximize the comprehensive response to production operation requirements;

[0069] According to the enterprise's strategic focus, market demand changes, and production actual situation, dynamically adjust the multi-objective priorities of production efficiency, quality stability, energy utilization rate, and safety risk to achieve dynamic balance among multi-dimensional objectives;

[0070] Design a mechanism that combines anomaly prediction and proactive scheduling. Through the time series prediction model, perform multi-step prediction on the production status and combine it with the digital twin simulation environment for cyclic simulation and evaluation. Proactively adjust the state between heterogeneous data parameters before the anomaly actually occurs, turning passive response into active prevention;

[0071] Establish a combination of knowledge-driven and data-driven, and through a distributed collaborative execution framework, make the collaborative scheduling strategy process transparent and efficiently executed.

[0072] In the second aspect, a deep learning-based MES digital collaborative management system for executing a deep learning-based MES digital collaborative management method, including:

[0073] Data acquisition module: It is used to acquire multi-source heterogeneous data, and through preprocessing and normalization processing, construct a dimensionless multi-dimensional data set;

[0074] Dynamic association module: It is used to map the dimensionless multi-dimensional data set into graph structure nodes to obtain node features, and define the connection exchange, information flow, and dependency relationships among the parameters in the dimensionless multi-dimensional data set as edge relationships; Through node feature embedding and edge relationship learning, analyze the dynamic association relationships among the data items in the dimensionless multi-dimensional data set, and generate a production line operation dependency relationship map;

[0075] Detection and localization module: It is used to construct a multi-source heterogeneous data collaborative analysis mechanism to obtain the dynamic association weights of collaboration among multi-source heterogeneous data modalities, and combine the dependency relationship map for anomaly detection and root cause localization to obtain the causal association information between anomaly events and data modalities;

[0076] Anomaly impact module: It is used to construct a digital twin simulation environment based on the causal association information, simulate the impact path of heterogeneous data parameter intervention on production disturbances on collaborative anomalies, and evaluate the anomaly impact results of parameter adjustment;

[0077] Model training module: It is used to construct an iterative scheduling model based on a dimensionless multi-dimensional data set, a dependency graph, and causal association information;

[0078] Collaborative management module; It is used to combine the abnormal impact assessment results with the real-time production status, iteratively generate a collaborative scheduling strategy for multi-objective optimization, and realize the dynamic configuration and abnormal prevention among multi-source heterogeneous data.

[0079] The beneficial effects of the present invention are as follows: By constructing a deep learning-based MES digital collaborative management method, the present invention effectively solves the problems of data islands, decentralized management of restrictive scenarios, lack of intelligent analysis capabilities, and poor system scalability faced by small and medium-sized enterprises in digital transformation. Moreover, it breaks through the technical bottleneck of collaborative optimization of restrictive scenarios, expresses the dynamic association relationship between data items through a production line operation dependency graph, and constructs a digital twin simulation environment using causal association information to realize a collaborative scheduling strategy for multi-objective optimization. Therefore, it can realize the collaborative optimization management of production operation parameter configuration and abnormal prevention for multi-source heterogeneous data in large-scale production with high dimensions, high complexity, and dynamic changes. And it improves the average level of production efficiency, reduces the product defect rate, reduces the operation cost, reduces safety accidents, and at the same time promotes data value mining to support innovative decision-making. In terms of economic benefits, the investment return period is shortened, the comprehensive economic benefits are improved, the digital transformation threshold of small and medium-sized enterprises is effectively reduced, enabling them to gradually achieve digital upgrading, and significantly enhancing the competitiveness of enterprises. Brief Description of the Drawings

[0080] Figure 1 It is a schematic flow chart of a deep learning-based MES digital collaborative management method provided in an embodiment of the present invention;

[0081] Figure 2 It is a schematic module flow chart of a deep learning-based MES digital collaborative management system provided in an embodiment of the present invention. Detailed Embodiments

[0082] Now, the subject matter described herein will be discussed with reference to example embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein, and the functions and arrangements of the elements discussed can be changed without departing from the protection scope of the content of this specification. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described in some examples can also be combined in other examples.

[0083] At least one embodiment of the present invention discloses a deep learning-based MES digital collaborative management method and system, as Figure 1 shown, including:

[0084] Step 1: Collect multi-source heterogeneous data, and through preprocessing and normalization processing, construct a dimensionless multi-dimensional data set;

[0085] Step 2: Map the dimensionless multi-dimensional data set to graph structure nodes to obtain node features, and define the connection exchange, information flow, and dependency relationships among the parameters in the dimensionless multi-dimensional data set as edge relationships;

[0086] Step 3: Through node feature embedding and edge relationship learning, analyze the dynamic association relationships among the data items in the dimensionless multi-dimensional data set, and generate a production line operation dependency relationship map;

[0087] Step 4: Construct a multi-source heterogeneous data collaborative analysis mechanism to obtain the dynamic association weights for collaboration among multi-source heterogeneous data modalities, and combine the dependency relationship map for anomaly detection and root cause location to obtain the causal association information between anomaly events and data modalities;

[0088] Step 5: Based on the causal association information, construct a digital twin simulation environment to simulate the influence path of heterogeneous data parameter intervention on production disturbances on collaborative anomalies, so as to evaluate the anomaly influence results of parameter adjustment;

[0089] Step 6: According to the dimensionless multi-dimensional data set, dependency relationship map, and causal association information, construct an iterative scheduling model, and combine the anomaly influence evaluation results with the real-time production status to iteratively generate a collaborative scheduling strategy for multi-objective optimization, realizing the dynamic configuration and anomaly prevention among multi-source heterogeneous data.

[0090] In this embodiment, in Step 1, structured or unstructured data is collected from multiple data sources (such as sensors, MES, ERP, PLC, logs), and format cleaning, missing value processing, unit unification, and noise filtering preprocessing steps are performed on the data from different sources. Subsequently, the numerical range differences are unified through normalization (such as Min-Max normalization, Z-score standardization) to construct a dimensionless multi-dimensional data set with a unified structure; realizing the semantic unification and dimension unification of heterogeneous data; reducing the noise and interference of model inputs and improving the accuracy of subsequent analysis; providing a consistent data basis for subsequent graph construction, analysis, and modeling.

[0091] In Step 2, each data parameter in the dimensionless multi-dimensional data set is mapped to a node in the graph, and its connection exchange, signal transmission path, or logical dependency relationship in the actual business / production line is used as an edge to construct a graph structure model. The node contains the original parameter features, and the edge represents the mutual influence relationship between the data; transforming the originally flattened data structure into a graph structure is conducive to expressing complex dependencies; facilitating the capture of the interaction relationships and non-linear transmission paths between data items; laying a structural foundation for subsequent in-depth feature mining using graph neural networks.

[0092] In step 3, the graph embedding method (such as GNN, GraphSAGE, GCN) is used to encode node features, and the dynamic semantic relationships between nodes are obtained through edge weight learning and message passing mechanism, so as to generate a dependency graph that accurately depicts the actual production line operation; realize the visualization and structured expression of the production line data relationship; capture the dynamic dependencies between parameters that change over time during actual operation; lay a foundation for accurate anomaly identification and prediction.

[0093] In step 4, a cross-fusion mechanism between different modalities (such as images, texts, temperatures, pressures) is established, and the dynamic weights between modalities are extracted through collaborative learning of modalities. Combining the previous dependency graph, graph matching and anomaly score calculation (such as GAD, GraphAE) are applied to identify potential anomalies, and causal reasoning techniques (such as Granger causality, Do-Calculus) are combined to locate the root cause; improve the anomaly detection ability in multi-source complex data scenarios; accurately mine the root cause of problems and shorten the troubleshooting time; support starting from causal relationships rather than just correlations to enhance interpretability.

[0094] In step 5, the key causal association information is input into the digital twin system to simulate the system response behavior under various parameter perturbation conditions, and analyze the specific impact of intervention measures on the abnormal propagation path, so as to predict "what consequences will occur if a certain abnormal situation occurs"; conduct virtual simulation experiments without interrupting the real production line; evaluate the parameter adjustment risk in advance and avoid blind intervention; provide a more realistic reference basis for optimization strategies.

[0095] Integrate standardized data, operation dependency graph, and causal information to construct a scheduling optimization model containing multiple constraints and multiple objectives. This model iteratively evolves the scheduling strategy through reinforcement learning, evolutionary algorithms or optimizers, and dynamically adjusts according to the actual production status to achieve closed-loop optimization and anomaly prevention capabilities; realize real-time scheduling optimization from a global data perspective; enhance the robustness and flexible response capabilities of the production line; when an anomaly occurs in the system, it can automatically make scheduling adjustments to improve the automatic recovery ability.

[0096] In a preferred embodiment of this first embodiment, in step 2, multi-source heterogeneous data is collected, and through preprocessing and normalization processing, a dimensionless multi-dimensional data set is constructed, including:

[0097] Step 21: Real-time collect multi-source heterogeneous data during production and manufacturing operations and perform data preprocessing to obtain a preprocessed data set;

[0098] Step 22: Perform standardized conversion on the preprocessed data set and perform data cleaning operations to obtain a clean data set;

[0099] Step 23: Perform time series alignment processing on the clear data set, and implement normalization processing after collecting at a preset period to obtain a dimensionless multi-dimensional data set.

[0100] In this embodiment, step 21 usually relies on sensors, monitoring systems or device interfaces at the production site to collect multi-source heterogeneous data from different production links and different devices. The data types may include temperature, humidity, pressure, device status, production speed, and quality control indicators. Data preprocessing: Since the data comes from multiple sources and has different formats and precisions, the main task of preprocessing is to convert the data from different sources into a unified format and solve common problems such as missing values, noisy data, and outliers. Data cleaning: For example, detect and repair missing values or error values, and remove extreme outliers. Format standardization: Unify heterogeneous data from different sources into a format convenient for subsequent processing (such as timestamp, unit conversion, data type unification); thus ensuring that the data has a unified format when performing the next analysis, facilitating subsequent processing; laying a foundation for subsequent standardization and cleaning, improving data quality and usability; avoiding affecting the analysis results due to data noise or inconsistency.

[0101] The data standardization process in step 22 is to convert the values of different features to the same scale so that the measurement criteria for different data items are consistent. Common standardization methods include: Z-score standardization: By subtracting the mean and dividing by the standard deviation, the data has zero mean and unit variance. Min-Max standardization: Scale the data to a specified interval (for example, [0, 1]), which is suitable for cases where the numerical distribution is uneven. The standardized data ensures that the dimensions of different features (such as temperature and humidity) are consistent, making them have the same influence in subsequent modeling. Data cleaning: Further process problems such as missing values, duplicate data, and error values in the preprocessed data set. For example: Fill in missing values: Use methods such as mean imputation, interpolation, or model-based prediction. De-duplication: Ensure that there are no duplicate records in the data. Outlier processing: Remove or correct possible incorrect data points. Standardization ensures the numerical consistency of the data, avoiding imbalanced effects on the model caused by scale problems for certain features; the cleaning step removes incorrect and inconsistent data, making subsequent data analysis more reliable and reducing interference factors in model training.

[0102] In Step 23, there may be certain time deviations between different production equipment and sensors, or the data collection frequencies may vary. The purpose of time series alignment is to align data from different sources by time points, ensuring that each data set has the same time stamp. For example: Interpolation method: If the data collection time points are inconsistent, use linear interpolation or other interpolation methods to align the data onto a unified time axis. Data synchronization: Adjust the data collection period to ensure that all data points are within a unified time window. Collection at a preset period: Some data may be collected at irregular time intervals. To reduce data redundancy and the risk of overfitting, sample the data at a fixed time period (such as every second or every minute). Periodic resampling: Ensure the uniformity of the data and remove overly frequent data fluctuations.

[0103] Normalization processing: Normalization scales the data proportionally so that different features have the same dimension. Usually, the min-max normalization method is used to scale the data into the range of [0, 1] or [-1, 1]. Dimensionless data set: The normalized data set will not be affected by the original units of the features, enabling subsequent analysis models to more effectively understand the relative relationships between the data. Consistency on the time axis, avoiding the impact of deviations in different data collection periods on subsequent analysis. Periodic collection reduces data redundancy, making the system more efficient and adaptable to the requirements of different collection frequencies.

[0104] Normalization makes the data have no differences in units or scales for each feature, improving the accuracy of subsequent analysis and preventing the model from being overly sensitive to a specific dimension.

[0105] In a preferred embodiment of this Embodiment 1, in Step 3, through node feature embedding and edge relationship learning, the dynamic association relationships between data items in the dimensionless multi-dimensional data set are analyzed to generate a production line operation dependency relationship graph, including:

[0106] Step 31: Map the dimensionless multi-dimensional data set into graph structure nodes, with multi-source heterogeneous data parameters as node features;

[0107] Step 32: Define the connection exchange, information flow, and dependency relationships between multi-source heterogeneous data parameters as edge relationships;

[0108] Step 33: Use the graph attention network structure for node feature embedding to capture the local correlation of edge relationships;

[0109] Step 34: Through the edge relationship feature learning mechanism, quantify the interaction intensity between multi-source heterogeneous data modalities;

[0110] Step 35: Iteratively propagate through a multi-layer graph network based on local relevance and interaction intensity to analyze the dynamic association relationships among data items in a dimensionless multi-dimensional dataset, and generate a production line operation dependency graph that includes the association states between heterogeneous data, production operation parameters, and parameter quality indicators.

[0111] In this embodiment, step 31 first converts the standardized multi-dimensional dataset into a graph structure and maps different data items to the nodes of the graph. These data items can come from different sources (such as sensor data, production parameters, environmental data), so they are called "multi-source heterogeneous data". These heterogeneous data are embedded in the graph structure as node features, that is, each single node carries the corresponding data feature information. This enables a single node to not only represent a data point but also contain the specific parameters or values of the data; by mapping multi-dimensional data into a graph structure and assigning features to individual nodes, it facilitates the application of subsequent graph algorithms, especially the graph neural network (GNN) model. This structure provides a basis for capturing complex relationships between data and lays a solid foundation for information flow and processing in subsequent steps.

[0112] Step 32 defines the connection exchange, information flow, or mutual dependency relationships between different nodes as "edge relationships". These edges reflect the interactions or relationships between nodes. For example, there may be a certain dependency between sensor data and control parameters in a production line, or there is a connection exchange between the operating states of certain devices and environmental parameters. The definition of edge relationships can be based on business logic, physical models, data dependency factors; defining edge relationships can effectively reflect the mutual dependencies and dynamic changes between data into the graph structure. This helps to analyze and learn the interactive impacts between nodes subsequently, and through edge relationships, the physical, temporal, or functional associations between different parameters can be captured, improving the system's adaptability to complex business scenarios.

[0113] Step 33 uses a graph attention network to perform node feature embedding. GAT uses an attention mechanism to calculate the weighted influence of neighboring nodes, making the influence of important nodes on the final representation greater. This mechanism can automatically focus on those adjacent nodes that are more important in the current task, thus effectively capturing the local relevance of edge relationships; the graph attention network can capture the different influence intensities between nodes in edge relationships through a local attention mechanism, which enables the model to flexibly adjust the focus between different nodes and edges according to task requirements, thereby enhancing the modeling ability for complex relationships. This helps to improve the sensitivity of the data model to subtle dependencies between nodes.

[0114] Step 34 learns the features of each edge in the graph through an edge relationship feature learning mechanism to quantify the interaction intensity between different data items (modalities). Different data modalities (such as temperature, humidity, voltage) may have different degrees of impact on the operation of the system. By learning the features of each edge, the model can identify which data items have strong interaction relationships and which have weak ones, thereby quantifying the impact intensity between them; this quantification of edge relationships can provide more accurate input for subsequent decision-making. By identifying and quantifying the interaction intensity between different data items, the model can optimize the resource scheduling and operation strategy of the system, improving the overall operation efficiency and reliability of the system.

[0115] Step 35: Through iterative propagation of a multi-layer graph network based on local correlation and interaction intensity, to analyze the dynamic association relationships between data items in a dimensionless multi-dimensional dataset, and generate a production line operation dependency graph that includes the association status between heterogeneous data, production operation parameters, and parameter quality indicators; Through iterative propagation of a multi-layer graph network, based on the local correlation and interaction intensity of edge relationships learned in the previous steps, analyze the dynamic association relationships between data items in a dimensionless multi-dimensional dataset. This process enables the full integration of the feature information of each node through multi-layer propagation of the graph network, thereby generating a production line operation dependency graph that includes the association status between heterogeneous data, production operation parameters, and quality indicators. This graph can clearly show how each link in the production process interacts with each other and their impact on the overall production process; This graph can not only intuitively display the dynamic associations between different data items, but also reveal potential bottlenecks and optimization spaces in the production process. Through such a graph, production line managers can more effectively monitor the production process and make timely adjustments and optimizations to ensure production efficiency and quality control.

[0116] Specifically, non-numerical attributes such as equipment type (e.g., "Machine Tool A") are mapped to computer-understandable numerical vectors through an embedding layer (for example, converting the category into a 128-dimensional vector).

[0117] Dynamic feature extraction: Use a bidirectional LSTM model for time series data (such as continuous 10-minute temperature records) to capture the forward and backward dependencies of the data and extract dynamic feature vectors (for example, 256-dimensional vectors).

[0118] Feature fusion: Concatenate the static vector and the dynamic vector into a long vector, and then compress and reduce the dimension through a fully connected layer to obtain the initial feature representation of each node. Output data: The fused feature vector of each node, which is used as the input for subsequent graph structure construction.

[0119] Explicit edge definition: According to the physical connection between devices (such as the conveyor belt flow direction) or the process flow chart, directly define the initial edges (for example, the initial weight of the edge from device A to device B is 1).

[0120] Implicit edge discovery: Calculate the cosine similarity between all node features. If the similarity exceeds a threshold (e.g., 0.8), an implicit edge is generated (for example, sensor X and device Y are associated because of similar vibration patterns).

[0121] Adjacency matrix fusion: Combine explicit edges (fixed weights) and implicit edges (cosine similarity as weights) to form an initial weighted adjacency matrix. Output data: The adjacency matrix containing explicit and implicit edges, along with the node feature vectors, are jointly used as the input to the graph attention network.

[0122] For each node and its neighbors, calculate the dynamic weights through the attention mechanism. For example:

[0123] Use multiple attention heads (e.g., 8), and each head independently learns the association strength between nodes.

[0124] The attention weight between node A and neighbor B may represent "the degree of influence of device B on device A during current fluctuations".

[0125] Edge weight update: Aggregate the calculation results of multiple attention heads to update the original edge weights (for example, the weight of device A → device B is increased from 0.8 to 1.2).

[0126] Node feature update: Aggregate neighbor features based on the new edge weights to generate a richer node representation (for example, the features of device A incorporate the dynamic states of upstream and downstream devices). Output data: The updated adjacency matrix (dynamic edge weights) and node features are used to generate the final dependency graph.

[0127] In a preferred embodiment of this first embodiment, a time series graph convolutional network structure is used to process the dynamically changing multi-source heterogeneous data parameter states, so as to fuse the time series features and the graph network topological features to obtain a fusion result;

[0128] Based on the fusion result, splice and reduce the dimensionality of the static attributes of each data item parameter in the dimensionless multi-dimensional dataset and the dynamic time series data to generate a fusion feature vector for a single node, which is used as the basic input for constructing the production line operation dependency graph;

[0129] Define explicit edges through the interaction intensity connection, and at the same time obtain the cosine similarity between the fusion feature vectors of each node to generate implicit edges. After fusion, a weighted adjacency matrix is formed, and the dynamic edge weights are jointly output with the node features;

[0130] By learning the dynamic edge weights between node features, aggregate the results of multiple attentions to update the dynamic edge weights, and fuse the neighbor node features to generate updated node features.

[0131] Construct a production line operation dependency graph based on the dynamic edge weights and the updated node features, revealing the explicit associations and implicit dynamic dependencies among multi-source heterogeneous data items during production operation to support production line operation decision-making.

[0132] In this embodiment, a time series graph convolutional network structure is used to process the dynamically changing multi-source heterogeneous data parameter states. Multi-source heterogeneous data refers to data from different devices, systems, sensors, or data types (such as temperature, speed, pressure, text logs). A time series graph convolutional network (T-GCN) is adopted, combining: time dimension modeling (RNN, GRU, LSTM): capturing the time change trend; graph convolution (GCN): combining the graph structure to model the dependencies among data items.

[0133] At the same time, obtain the dynamic evolution information (time series) and structural relationship information (topology) of the data, laying a foundation for subsequent deep fusion; and can adapt to scenarios where the data is updated and changed in real time.

[0134] Concatenation and dimensionality reduction: static attributes + dynamic time series data, generating a fused feature vector: for each data item (such as a temperature sensor); static attributes (such as device number, location, type) + dynamic time series features (such as temperature changes over a period of time); after concatenation, it is processed by a dimensionality reduction method (such as PCA, AutoEncoder, MLP) to obtain a fused feature vector of a unified dimension. Represent the multi-modal data uniformly, solve the problem of inconsistent dimensions; fuse static and dynamic information, enhancing the semantic expression ability of each node; improving the computational efficiency after dimensionality reduction and reducing redundant features.

[0135] Construct the graph: explicit edges + implicit edges, generating a weighted adjacency matrix:

[0136] Explicit edges: strong connections determined according to business rules or technological processes (such as technological paths, physical connections between devices, etc.);

[0137] Implicit edges: calculate the cosine similarity based on the fused feature vectors of each node;

[0138] Set a threshold or construct k-nearest neighbors to define the similarity edges; after merging the two types of edges, a weighted adjacency matrix is formed.

[0139] Model both explicit physical relationships and implicit data similarities at the same time; more comprehensively reflect the data flow logic and dependency relationships in the production line; facilitate capturing potential impact paths that may exist but are not explicitly modeled.

[0140] Dynamic edge weight learning + multi-attention mechanism to update node features: Through the Graph Attention Network (GAT), learn the different importance of different neighbors to the target node; the multi-head attention mechanism enhances the robustness and expressive power of the model; aggregate the features of neighbor nodes and generate the updated node representation by weighting; at the same time, update the edge weights to reflect the change of the dynamic dependence strength between nodes. More accurately model the dynamic interaction pattern between nodes; adapt to the changing production line status, improve the model's perception of emergencies; dynamically update the graph structure itself to make the graph more timely and real-time.

[0141] Construct a production line operation dependency graph to support operation decisions; the above steps generate:

[0142] The updated node features; dynamic edge weights (explicit + implicit); constitute a complete production line operation dependency graph that evolves over time. Reveal the deep and dynamic associations between multi-source heterogeneous data items; support decision-making tasks such as: fault prediction and diagnosis; process optimization; production line adjustment and scheduling; discovery of abnormal behaviors; construction of a production line-level "digital twin" model.

[0143] In the preferred embodiment of this Example 1, in step 4, construct a multi-source heterogeneous data collaborative analysis mechanism to obtain the dynamic association weights for collaboration between multi-source heterogeneous data modalities, and combine with the dependency graph for anomaly detection and root cause localization to obtain the causal association information between anomaly events and data modalities, including:

[0144] Step 41: Extract features and perform representation learning on the states between data items in the dimensionless multi-dimensional dataset to obtain the correlation matrix between multi-source heterogeneous data modalities;

[0145] Step 42: Dynamically allocate and adjust the contribution degrees between multi-source heterogeneous data modalities using dynamic edge weights to obtain the dynamic association weights for collaboration between multi-source heterogeneous data modalities;

[0146] Step 43: Capture the association influence degree between multi-source heterogeneous data modalities according to the dynamic association weights;

[0147] By: , to generate a dynamic association weight matrix;

[0148] Among them, represents the dynamic association weight matrix, Q represents the query matrix, represents the key matrix at the T-th moment, V represents the value matrix, d represents the feature dimension, and softmax represents the normalization function;

[0149] Step 44: Combine the dynamic association weight matrix with the production line operation dependency graph to construct an enhanced graph structure topology;

[0150] Step 45: Identify node features and edge relationships deviating from the normal pattern on the enhanced graph structure topology;

[0151] Through Perform anomaly detection to obtain anomaly recognition results;

[0152] Among them, represents the anomaly score, x represents the input feature, represents the reconstructed feature, represents the square of the L2 norm;

[0153] Step 46: Based on the anomaly recognition results, locate the root cause nodes of the anomaly through backpropagation and subgraph extraction to obtain root cause location data;

[0154] Step 47: Use causal inference to quantify the causal association strength between the anomaly recognition results and the root cause location data, and generate causal association information. The causal association strength represents the direct influence degree of the anomaly recognition results on the root cause location data, and is determined by obtaining the difference between the mutual information and the conditional mutual information;

[0155] The causal association strength of the causal association information is:

[0156] ;

[0157] Among them, C(X - Y) represents the causal association strength of the anomaly recognition result X on the location data Y, I(X, Y) represents the interaction information between the anomaly recognition result X and the location data Y, I(X, Y|Z) represents the conditional interaction information between the anomaly recognition result X and the location data Y under the given constraint condition Z, PA(Y) represents the set of all potential parent nodes of the location data Y, and {X} represents the quantified data of the anomaly recognition result.

[0158] In this embodiment, in step 41, feature extraction and deep representation learning are performed on dimensionless multi-dimensional data, and models such as autoencoders, graph neural networks (GNNs), and Transformers can be used to capture the hidden relationships between different modal data (such as sensors, logs, videos, etc.). Thus, information alignment and unified representation between different data sources are realized; redundancy is reduced, and the data fusion efficiency is improved; a good foundation is laid for subsequent graph construction and causal analysis.

[0159] Step 42 dynamically adjusts the weights between each modality through dynamic modeling methods (such as attention mechanisms, graph attention networks GAT), reflecting the importance changes of each data modality in different scenarios; overcomes the problem of insufficient response of static modeling to complex system changes; realizes collaborative perception between multiple modalities; and adaptively enhances the data weights of key modalities.

[0160] Step 43 aggregates the dynamic weights obtained in step 42 into a "dynamic association weight matrix" between modalities, which can be regarded as a topological structure of the influence between modalities and is used to express the strong and weak associations between each pair of modalities. It explicitly expresses the influence between modalities, facilitates subsequent integration with the graph structure, and supports downstream anomaly detection and causal analysis.

[0161] Step 44 integrates the dynamic association weight matrix into the existing production line operation dependency graph to form an enhanced graph structure, which can be understood as superimposing the "modal relationship" on the "entity dependency relationship"; constructing a more expressive system structure diagram; while retaining physical / logical dependencies and data associations; and facilitating the mining of implicit abnormal propagation paths.

[0162] Step 45 performs node and edge anomaly detection on the enhanced graph structure. Methods such as graph neural networks and graph autoencoders can be used to identify local substructures that deviate from normal patterns. Efficiently detect abnormal points and abnormal links in the system; improve the accuracy and interpretability of anomaly detection; and accurately locate potential risk areas.

[0163] Based on the anomaly identification result, step 46 tracks the anomaly impact path through back propagation technology, extracts subgraphs at the same time, and locks the source node that causes the anomaly; quickly and accurately locates the root cause of the anomaly; reduces the false alarm rate and missed alarm rate; and provides maintenance personnel with an actionable decision-making basis.

[0164] Step 47 uses causal inference technology to quantify the causal relationship strength between the anomaly and the root cause. The difference between mutual information and conditional mutual information is used to estimate the direct causal impact; strengthen the causal explanation between the anomaly and the root cause; support fault traceability and prevention; and provide data basis for operation and maintenance optimization and strategy adjustment.

[0165] In a preferred embodiment of the first embodiment of the present invention, in step 5, a digital twin simulation environment is constructed based on causal association information to simulate the impact path of heterogeneous data parameter intervention production disturbance on collaborative anomalies to evaluate the abnormal impact results of parameter adjustment, including:

[0166] Step 51: Mapping the acquired causal relationship information into the digital twin simulation environment to establish a corresponding relationship between the physical entity and the digital model;

[0167] Step 52: Define the interaction rules in the digital twin simulation environment according to the production line operation dependency graph to ensure that the digital twin simulation environment accurately reflects the dynamic characteristic data of the actual production operation;

[0168] Step 53: Design a parameter intervention experiment plan based on the dynamic characteristic data of production operation, and simulate the production disturbance scenario by adjusting the key parameters in the digital twin simulation environment at fixed points;

[0169] Step 54: Based on the production disturbance scenario, trace the propagation path of the production disturbance, record the state changes of the characteristics of each node, and analyze the impact of the production disturbance scenario on the collaborative anomaly between multi-source heterogeneous data modalities;

[0170] Step 55: Comprehensively evaluate the suppression or exacerbation effects of the production disturbance scenarios simulated by various parameter adjustment schemes to form a quantitative evaluation result of the anomaly impact.

[0171] In this embodiment, step 51 maps the causal association information obtained from anomaly detection and root cause location (such as the causal relationship between the anomaly node and its cause node) into the digital twin system, corresponding to the synchronization relationship between the physical entity and the digital model.

[0172] Key technologies: Digital twin mapping model; Entity-ID matching and attribute annotation; Spatiotemporal synchronization mechanism (real-time / quasi-real-time). Achieve the consistency between the physical-digital systems; Provide a credible basic model for virtual simulation; Support subsequent disturbance simulation and impact assessment.

[0173] Step 52: Based on the existing production line operation dependency graph, establish the interaction rules between entities in the digital twin system, such as logistics paths, equipment interlocks, process sequences, energy flow / information flow paths; Key technologies: Interaction modeling based on the graph; Dynamic rule engine (such as state machine or event-driven system); Real-time data-driven rule update. Thus, ensure that the simulation environment has real interaction behaviors; Accurately restore the changes in the system operation state; Achieve dynamic feedback and prediction capabilities.

[0174] Step 53: By adjusting the key operating parameters (such as temperature, voltage, speed, feeding rhythm, etc.) in the digital twin system, construct abnormal / disturbance scenarios that may actually occur. Key technologies: Sensitivity analysis to select intervention points; Design of experiments (DoE); Repeatable disturbance simulation mechanism. Quickly simulate various possible operation risks; Conduct risk prediction without affecting the actual production line; Support systematic trial and error and optimal solution screening.

[0175] During the intervention simulation process in step 54, track the propagation path of the disturbance from the source to other nodes in the system, and observe the changes in the states of each node (such as equipment efficiency, data anomaly amplitude, modal coupling strength, etc.). Key technologies: Graph neural network path tracing; Dynamic system response modeling; Multi-modal feature fusion analysis. Thus, reveal the internal abnormal propagation mechanism of the system; Analyze the abnormal co-evolution relationship between modalities; Provide theoretical support for the identification of risk control points.

[0176] Step 55 compares the system's responses to anomalies under different parameter intervention schemes to evaluate which schemes can suppress anomalies and which may exacerbate them, achieving quantitative evaluation. Key technologies: multi-scenario simulation comparison; construction of an anomaly impact index system (such as collaborative anomaly score, response time, and impact range); visualization result analysis. Formulate scientific and effective anomaly response strategies; support the active optimization and preventive maintenance of the system; enhance the overall resilience and reliability of the system.

[0177] In a preferred embodiment of the first embodiment, in step 6, an iterative scheduling model is constructed based on the dimensionless multi-dimensional data set, the dependency relationship graph, and the causal association information, including:

[0178] Step 61: Construct a state representation space based on the dimensionless multi-dimensional data set, and map the current state during production operation into a high-dimensional feature vector, including multi-dimensional information data such as equipment operation parameters, quality indicators, energy consumption data, and safety monitoring data;

[0179] Step 62: Define a constraint condition matrix using the production line operation dependency relationship graph. The constraint condition matrix represents the precedence dependencies, resource sharing, and conflict relationships between production units to generate a scheduling relationship matrix that meets the production operation constraint requirements;

[0180] Step 63: Construct an iterative scheduling model based on the multi-dimensional information data, the scheduling relationship matrix, and the causal association information. The iterative scheduling model specifically includes:

[0181] Step 64: Generate structured data from the multi-dimensional information data, the scheduling relationship matrix, and the causal association information, and encode the structured data into sequence data;

[0182] Step 65: Input the sequence data into the iterative scheduling model; the iterative scheduling model includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer. Transmit the intermediate representation data of multiple hidden layers to the output layer, and the output layer iteratively outputs the scheduling strategy prediction results corresponding to the state changes between multi-source heterogeneous data parameters;

[0183] Step 66: Input at least two multi-dimensional information data, scheduling relationship matrix, and causal association information data items obtained in real time into the iterative scheduling model, and iteratively output the scheduling strategy prediction results corresponding to the state changes between the multi-source heterogeneous data parameters obtained in real time.

[0184] In this embodiment, step 61 standardizes and dimensionlessizes various dynamic operation information (equipment parameters, energy consumption, quality, safety, etc.) involved in the production system, maps them to a unified high-dimensional representation space, and forms feature vectors with engineering semantics, including data: equipment operation parameters (such as rotational speed, temperature, vibration); product quality indicators (such as dimensional error, yield); energy consumption data (such as energy consumption per process, current, voltage, etc.); safety monitoring data (such as number of alarms, risk scores, etc.). By establishing a complete state portrait, it is convenient for the scheduling model to obtain a global view; the dimensionless processing solves the problem of inconsistent unit dimensions and improves the generalization ability of the model; the high-dimensional data fusion realizes the fine-grained characterization of complex systems.

[0185] Step 62 extracts the constraint relationships between each production unit through the production line operation dependency graph, constructs a constraint condition matrix, and clarifies: process sequence (preceding dependency); resource sharing (such as the same robot cannot serve multiple units simultaneously); conflict relationships (such as process concurrency conflicts, physical space conflicts). Ensure that the scheduling model generates feasible solutions and does not violate actual constraints; support complex scheduling dependency modeling in flexible manufacturing systems; avoid problems such as deadlocks and resource conflicts in scheduling.

[0186] Step 63 combines multi-dimensional information data (state features), scheduling constraint matrix, and causal information to establish a deep iterative model for scheduling prediction, and uses historical data to drive the training model to learn the mapping relationship between task states and optimal scheduling strategies. Utilize causal information to enhance the model's perception ability of abnormal propagation paths; improve the adaptability of the scheduling system to complex dynamic scenarios; support online dynamic optimization of scheduling strategies.

[0187] Step 64 converts static structure data (such as graph structure, constraint matrix, feature vector) into time series data that can be processed by a neural network model, facilitating the use of time series modeling structures (such as RNN, Transformer) to capture the evolutionary relationship between history - current - prediction. Use the sequence modeling structure to extract the state evolution trend; enhance the scheduling model's recognition ability of time delay effects; realize the time series prediction ability of scheduling strategies (non-one-time decision).

[0188] Step 65 continuously iteratively learns the complex non-linear relationship between the intermediate state and the scheduling strategy through a neural network structure (typically a multi-layer structure: input layer → multiple hidden layers → output layer), and finally the output layer generates the scheduling strategy prediction result. Model structure:

[0189] Input layer: Receives the state vector + scheduling matrix + causal vector;

[0190] Hidden layer: Each layer abstractly extracts feature information and passes it layer by layer;

[0191] Output layer: Output the predicted scheduling strategy (such as priority list, process-time allocation table, etc.).

[0192] The deep learning structure enhances the non-linear relationship modeling ability; the multi-hidden layer design improves the expression ability of feature interaction; provides interpretable scheduling suggestions or visualization maps.

[0193] In step 66, during the deployment phase, the state information of multiple dimensions (at least two dimensions or more, such as energy consumption + quality) and structural constraints are used as inputs in real time, and the optimal scheduling strategy prediction results matching the current state are iteratively output. Realize real-time dynamic scheduling (Rescheduling); quickly respond to emergencies or production anomalies; support the online optimization mechanism.

[0194] In a preferred embodiment of this embodiment 1, in step 6, the abnormal impact assessment result is combined with the real-time production state, and a collaborative scheduling strategy for multi-objective optimization is iteratively generated to realize the dynamic configuration and anomaly prevention among multi-source heterogeneous data, including:

[0195] Step 67: Collect key parameters of the real-time production operation state, including equipment operation state, quality inspection data, energy consumption indicators, and safety monitoring information, to form a multi-dimensional real-time state vector; and extract the current topological structure and constraint conditions from the production line operation dependency graph, and combine with the historical abnormal pattern library to capture the abnormal risk of the real-time state;

[0196] Step 68: Based on the captured abnormal risks, perform multi-step predictions on the future state evolution to identify potential abnormal development trends; for the detected abnormal risk points, generate a design parameter intervention effect scoring matrix according to the abnormal impact assessment result;

[0197] Step 69: Start the multi-objective optimization decision-making mechanism according to the design parameter intervention effect scoring matrix, set the key indicators of production efficiency, quality stability, energy utilization rate, and safety risk as optimization objectives, and dynamically adjust the weights of each objective according to the current production task priority;

[0198] And search for the optimal scheduling strategy under the constraint conditions, by:

[0199] ;

[0200] To balance the short-term intervention effect and the long-term stable scheduling strategy;

[0201] Among them, J(θ) represents the comprehensive optimization objective, represents the production efficiency objective, represents the quality stability objective, represents the energy utilization rate objective, represents the safety risk objective, , , and respectively represent the dynamic weight coefficients of four objectives of production efficiency, quality stability, energy utilization rate, and safety risk, and satisfy ;

[0202] Step 610: Execute the collaborative scheduling strategy deployment based on the prediction results of the iterative scheduling model, and decompose the optimized scheduling instructions into specific execution instructions at the device level, process level, and resource level; securely transmit the specific execution instructions to each execution unit through the middleware layer, and at the same time establish an instruction execution confirmation mechanism to achieve dynamic configuration and exception prevention among multi-source heterogeneous data;

[0203] Step 611: Establish a real-time monitoring and feedback mechanism to continuously track the execution effect of the scheduling strategy and collect the change data of key performance indicators; when detecting an execution deviation or a new abnormal event, trigger a quick response mechanism, and select fine-tuning or re-planning the strategy according to the degree of deviation.

[0204] In this embodiment, in step 67, the device operation status, quality inspection results, energy consumption data, and safety monitoring information are obtained in real time. State vector construction: These information are uniformly constructed into a high-dimensional state vector to represent the overall operation state of the current system. Structure modeling: Obtain the current production line topology structure and operation constraints through the "production line operation dependency relationship map". Pattern recognition: Combine the historical abnormal pattern library, and judge whether there is a potential abnormal risk currently through pattern matching and abnormal detection algorithms (such as clustering, isolation forest, Bayesian method, etc.). Realize the comprehensive perception of the operation state of complex industrial systems. It can identify potential faults or quality problems in advance and avoid the spread of problems.

[0205] Step 68: Based on the identified abnormal risks, use a time series prediction model (such as LSTM, Transformer, etc.) to predict the future state change trend, construct a "design parameter intervention effect scoring matrix", simulate the effect of different parameter adjustments on abnormal mitigation, and the higher the score, the more effective the intervention; according to the prediction results, clarify which links may deteriorate and need to be processed preferentially. Provide forward-looking early warning and response reference for the scheduling system; screen the most influential control parameters in advance to improve the response efficiency and effect.

[0206] Step 69: Take production efficiency, quality stability, energy utilization rate, and safety risk as the objectives, according to the priority and real-time requirements of production tasks, weight the above objectives, adapt to task changes through evolutionary algorithms, particle swarms, NSGA-II algorithms, search for the optimal scheduling strategy under constraints, taking into account short-term effects and long-term stability; realize personalized, task-oriented scheduling strategies; improve the overall performance and resource utilization efficiency of the system.

[0207] Step 610 refines the overall scheduling plan into executable device-level, process-level, and resource-level tasks; through industrial middleware (such as OPC UA, MQTT, etc.), the instructions are safely and accurately transmitted to different control units; the feedback mechanism ensures that each instruction is correctly received and executed, and can be automatically adjusted when a fault is found; it ensures that the scheduling strategy is implemented without deviating from the design goal; it improves the system's self-healing ability and robustness to execution exceptions.

[0208] Step 611 monitors key performance indicators, such as production rate, yield, and energy consumption; once a deviation in execution or a new anomaly is detected, it triggers fine-tuning (such as PID self-tuning) or re-plans the entire policy process; the new data and feedback results are fed back into the system to further enhance the model's accuracy and adaptability; it constructs a complete "perception - analysis - execution - feedback" closed-loop; it improves the system's adaptability, stability, and self-optimization ability.

[0209] In a preferred embodiment of this Embodiment 1, the collaborative scheduling strategy includes:

[0210] Decompose the scheduling problem into three levels: strategic, tactical, and operational, and separately handle long-term planning, medium-term scheduling, and real-time response to anomalies among heterogeneous data, forming a complete hierarchical collaborative decision-making framework;

[0211] By realizing the bidirectional integration of constraint scenarios and guiding scenarios, using the differentiable planning layer to unify hard constraints and soft constraints into the same framework, and dynamically adjusting the constraint weights according to the real-time production situation to maximize the comprehensive response to production operation requirements;

[0212] According to the enterprise's strategic focus, market demand changes, and production actual situation, dynamically adjust the multi-objective priorities of production efficiency, quality stability, energy utilization rate, and safety risk to achieve dynamic balance among multi-dimensional objectives;

[0213] Design a mechanism that combines anomaly prediction and forward-looking scheduling. Through the time-series prediction model, perform multi-step prediction on the production state and combine it with the digital twin simulation environment for cyclic simulation and evaluation. Actively adjust the state among heterogeneous data parameters before the anomaly actually occurs, transforming passive response into active prevention;

[0214] Establish a combination of knowledge-driven and data-driven, and through the distributed collaborative execution framework, make the collaborative scheduling strategy process transparent and highly efficient.

[0215] In this embodiment, the three-layer decomposition of the scheduling problem: strategic, tactical, operational;

[0216] Strategic layer (long-term planning): Analyze macro trends (such as market demand, enterprise development strategy), formulate long-term resource allocation and production capacity layout plans; provide directional guidance for the enterprise to ensure that the top-level design of the scheduling system is consistent with the enterprise's development.

[0217] Tactical layer (mid-term scheduling): Based on the goals provided by the strategic layer and combined with the production cycle (such as weekly / monthly plans), formulate an optimized production and resource scheduling plan; improve resource utilization rate, balance production load, and enhance mid-term operation efficiency.

[0218] Operational layer (real-time response): Monitor and adjust real-time data during the production process, especially handle abnormal information from heterogeneous systems; enhance the flexibility and robustness of the system, achieve rapid response to emergencies, and ensure production continuity.

[0219] Bidirectional integration of constraint scenarios and guiding scenarios + differentiable programming;

[0220] Bidirectional integration:

[0221] Constraint scenarios: Hard constraints (such as equipment capacity, resource upper limit);

[0222] Guiding scenarios: Soft guidance (such as giving priority to producing high-profit products).

[0223] Integrate the two through a unified differentiable optimization framework, so that the scheduling algorithm not only complies with rigid rules but also retains room for optimization. Achieve "flexibility with principles" and find a better scheduling solution under constraints.

[0224] Differentiable programming layer:

[0225] Introduce a differentiable optimization module (such as deep learning + optimization), embed the scheduling goal into the neural network training process, and dynamically adjust the weights of constraint terms. The model can adapt to production changes, and the optimized solution is more in line with real-time situations, improving the scheduling intelligence level.

[0226] Multi-objective dynamic balance mechanism: According to the enterprise strategy, market demand, and actual production line status at different time points, dynamically adjust the priorities among the following goals: production efficiency; product quality stability; energy utilization rate; safety risk control. Avoid the single-goal deviation of "only pursuing efficiency and ignoring safety"; achieve refined and comprehensive scheduling management; keep the enterprise operation in dynamic balance near the optimal point.

[0227] Abnormal prediction + forward-looking scheduling mechanism:

[0228] Use time series prediction models (such as LSTM, Transformer) to predict the production status in multiple future time steps; simulate possible scenarios in the digital twin environment; if a possible abnormal trend is found, schedule production strategies or resource configurations in advance. Shift from "emergency response after a failure" to "prevention of potential problems"; greatly improve the stability of the system and reduce downtime losses.

[0229] Knowledge-driven + Data-driven Integration, Distributed Collaborative Execution:

[0230] Knowledge-driven: Introduce an expert rule base and an industrial knowledge graph for initial scheduling design;

[0231] Data-driven: Continuously optimize the scheduling model through historical and real-time data;

[0232] Distributed Collaborative Execution Framework: Collaboration among multiple systems (such as ERP, MES, SCADA) to execute scheduling strategies in a distributed manner. Utilize human-machine co-intelligence to enhance the adaptability of the scheduling system; achieve transparency, traceability, and intervenability in the scheduling execution process; improve the overall collaborative efficiency of the enterprise and reduce the problem of information silos.

[0233] Implementation Examples of Multi-scenario Applications: The methods and systems of the present invention can be applied to multiple digital transformation application scenarios. The application cases of each scenario are introduced in detail below in combination with the specific industry characteristics:

[0234] Case 1: Digitalization of the Product Life Cycle.

[0235] 1. Product Design.

[0236] Implementation Scenario: A mechanical manufacturing enterprise faces problems such as a long product design cycle, frequent design changes, and a low first-pass success rate, and needs to improve design efficiency and quality.

[0237] Implementation Plan: Apply this system to realize digital product design. Through a multi-modal deep learning model, analyze historical design data, market feedback, and production data to achieve design optimization. The system collects multi-source information such as CAD design data, material parameters, and historical performance test data, constructs a product knowledge graph, and supports designers in parametric design and simulation verification.

[0238] Implementation Effect: After application, the product design cycle is shortened by 35%, the number of design changes is reduced by 40%, and the first-pass success rate is increased by 28%.

[0239] 2. Process Design.

[0240] Implementation Scenario: The process design process of an auto parts enterprise is complex, the optimization of process parameters depends on experience, and the first-pass qualification rate of products is low, affecting production efficiency.

[0241] Implementation Plan: Use this system to optimize the process design process. The system analyzes the correlation between historical process parameters and product quality through deep learning algorithms, and automatically generates the optimal process route and parameter combination. The system integrates CAD / CAM data, material characteristics, and equipment capacity parameter information to construct a process knowledge base and realize intelligent process design.

[0242] Implementation effect: After application, the process design efficiency has increased by 42%, the accuracy rate of process parameter optimization has reached 93%, and the first-pass yield of products has increased by 25%.

[0243] 3. Marketing management.

[0244] Implementation scenario: A consumer electronics enterprise lacks data support for marketing decisions, fails to accurately grasp customer needs, has high marketing costs and poor results.

[0245] Implementation plan: Apply this system to achieve digital transformation of marketing. The system integrates sales data, customer feedback, social media data and market research information, and analyzes consumer behavior patterns and market trends through a multi-modal Transformer network. The system constructs a customer portrait knowledge graph to support precision marketing and personalized recommendations.

[0246] Implementation effect: After application, the marketing conversion rate has increased by 32%, the customer acquisition cost has decreased by 28%, and the customer satisfaction has increased by 18%.

[0247] 4. After-sales service.

[0248] Implementation scenario: An engineering machinery enterprise has a slow after-sales service response, mostly passive maintenance, and long downtime of equipment due to failures, affecting customer satisfaction.

[0249] Implementation plan: Apply this system to optimize the after-sales service process. The system collects equipment operation data through IoT devices, combines maintenance records and user feedback, and uses a BiLSTM-Attention network to predict equipment failures and maintenance needs. The system constructs an equipment health status assessment model to achieve predictive maintenance and remote diagnosis.

[0250] Implementation effect: After application, the accuracy rate of equipment failure prediction has reached 87%, the unplanned downtime has been reduced by 45%, and the maintenance response time has been shortened by 60%.

[0251] Case 2: Digitalization of production execution.

[0252] 1. Production planning and scheduling

[0253] Implementation scenario: An electronic assembly enterprise has a low production plan execution rate, frequent delivery delays, a long production cycle, and low resource utilization.

[0254] Implementation plan: Apply this system to optimize production planning and scheduling. The system integrates order data, inventory information, equipment status and personnel scheduling information, and generates a multi-objective optimized production plan through a reinforcement learning algorithm. The system constructs a production constraint relationship graph to achieve dynamic scheduling and real-time optimization.

[0255] Implementation effect: After application, the production plan execution rate has increased by 38%, the on-time delivery rate has increased by 45%, and the production cycle has been shortened by 30%.

[0256] 2. Production Control.

[0257] Implementation Scenario: In a precision manufacturing enterprise, the production process monitoring is insufficient, the abnormal response is slow, the production efficiency is low, and the product consistency is poor.

[0258] Implementation Plan: Apply this system to realize intelligent control of the production process. The system collects real-time data of the production line through industrial cameras and sensor networks, and uses the improved YOLOv5 network for product defect detection and production anomaly identification. The system constructs a digital twin simulation environment for the production process to achieve visualization and real-time monitoring of the production process.

[0259] Implementation Effect: After application, the production anomaly response time is shortened by 75%, the production efficiency is increased by 33%, and the product consistency is improved by 29%.

[0260] 3. Quality Management.

[0261] Implementation Scenario: In a medical device enterprise, the quality management process is cumbersome, quality traceability is difficult, the root cause analysis of defects relies on manual experience, and the quality cost is high.

[0262] Implementation Plan: Apply this system to improve the quality management level. The system integrates on-line inspection data, laboratory test results and supplier quality information, and analyzes quality influencing factors and root causes of defects through a multi-modal deep learning model. The system constructs a quality traceability knowledge graph to achieve full-process quality control.

[0263] Implementation Effect: After application, the product defect rate is reduced by 42%, the quality traceability time is shortened by 85%, and the quality cost is reduced by 31%.

[0264] 4. Equipment Management.

[0265] Implementation Scenario: In a steel enterprise, equipment failures occur frequently, maintenance is mainly passive repair, the equipment availability is low, the maintenance cost is high, and the equipment life is short.

[0266] Implementation Plan: Apply this system to optimize the equipment management process. The system collects equipment operation data through vibration sensors and temperature sensors, and uses a temporal graph convolutional network to analyze the equipment health status and predict the risk of failure. The system constructs an equipment knowledge graph to achieve full-life-cycle management of equipment.

[0267] Implementation Effect: After application, the equipment availability is increased by 25%, the maintenance cost is reduced by 35%, and the equipment life is extended by 20%.

[0268] 5. Work Safety.

[0269] Implementation Scenario: In a chemical enterprise, the safety risk is high, the identification of safety hazards relies on manual inspections, the prevention of safety accidents is insufficient, and the emergency response is slow.

[0270] Implementation plan: Apply this system to improve the level of work safety. The system collects safety-related data through gas sensors and video monitoring, and uses a multi-modal Transformer network to identify safety risks and predict potential accidents. The system constructs a safety knowledge graph to achieve risk assessment and intelligent recommendation of emergency plans.

[0271] Implementation effect: After application, the incidence rate of safety accidents decreased by 65%, the recognition rate of safety hazards increased by 78%, and the emergency response time was shortened by 70%.

[0272] 6. Energy consumption management.

[0273] Implementation scenario: A textile enterprise has high energy consumption, low energy utilization efficiency, high energy costs, and lacks effective energy optimization strategies.

[0274] Implementation plan: Apply this system to optimize energy management. The system collects energy consumption data through smart meters and flow meters, and uses a BiLSTM-Attention network to analyze energy consumption patterns and predict energy demand. The system constructs an energy flow knowledge graph to achieve optimal energy distribution and generation of energy-saving strategies.

[0275] Implementation effect: After application, energy consumption decreased by 28%, energy utilization efficiency increased by 32%, and energy costs decreased by 25%.

[0276] Case 3: Supply chain digitization.

[0277] 1. Procurement management.

[0278] Implementation scenario: A household appliance enterprise has high procurement costs, unstable supplier deliveries, long procurement cycles, and lacks a scientific supplier evaluation system.

[0279] Implementation plan: Apply this system to optimize the procurement management process. The system integrates demand forecasting, inventory data, supplier evaluation, and market price information, and optimizes procurement decisions and supplier selection through deep learning algorithms. The system constructs a supplier knowledge graph to achieve intelligent price comparison and risk assessment.

[0280] Implementation effect: After application, procurement costs decreased by 22%, the on-time delivery rate of suppliers increased by 40%, and the procurement cycle was shortened by 35%.

[0281] 2. Warehousing and logistics.

[0282] Implementation scenario: A fast-moving consumer goods enterprise has low warehousing efficiency, high distribution costs, slow inventory turnover, and low utilization rate of logistics resources.

[0283] Implementation Plan: Apply this system to optimize warehouse logistics management. The system collects warehouse logistics information through RFID and AGV status data, and uses reinforcement learning algorithms to optimize warehouse layout and picking paths. The system constructs a logistics network knowledge graph to achieve intelligent inventory allocation and distribution route optimization.

[0284] Implementation Effect: After application, the warehouse efficiency is increased by 45%, the distribution cost is reduced by 30%, and the inventory turnover rate is increased by 38%.

[0285] Case 4: Digitalization of management decision-making.

[0286] 1. Financial management.

[0287] Implementation Scenario: A manufacturing enterprise has low financial analysis efficiency, inaccurate cost control, low capital utilization efficiency, and untimely financial risk warnings.

[0288] Implementation Plan: Apply this system to achieve digital financial management. The system integrates financial data on production costs, sales revenues, and capital flows, and analyzes cost composition and predicts cash flows through deep learning algorithms. The system constructs a financial knowledge graph to achieve cost tracking and financial risk warnings.

[0289] Implementation Effect: After application, the financial analysis efficiency is increased by 65%, the cost control accuracy is improved by 28%, and the capital utilization efficiency is increased by 32%.

[0290] 2. Human resources.

[0291] Implementation Scenario: A large manufacturing enterprise has unreasonable personnel allocation, low employee satisfaction, high human resources costs, and low skill-to-position matching.

[0292] Implementation Plan: Apply this system to optimize human resources management. The system integrates employee performance, training records, and working hours data, and analyzes employee capabilities and predicts talent needs through a multi-modal deep learning model. The system constructs a talent knowledge graph to achieve intelligent shift scheduling and skill matching.

[0293] Implementation Effect: After application, the personnel allocation efficiency is increased by 40%, the employee satisfaction is improved by 25%, and the human resources cost is reduced by 18%.

[0294] 3. Collaborative office.

[0295] Implementation Scenario: A multi-factory manufacturing enterprise has low cross-departmental collaboration efficiency, difficult project progress management, high communication costs, and poor information sharing.

[0296] Implementation Plan: Apply this system to improve collaborative office efficiency. The system integrates project progress, document management, and communication record information, and analyzes collaboration models and optimizes work processes through deep learning algorithms. The system constructs a business process knowledge graph to achieve cross-departmental collaboration and information sharing.

[0297] Implementation effect: After application, the project collaboration efficiency is increased by 50%, the decision-making response time is shortened by 60%, and the communication cost is reduced by 35%.

[0298] 4. Decision support.

[0299] Implementation scenario: A comprehensive manufacturing group relies on experience in decision-making, lacks data support, has a slow market response, fails to adjust strategies in a timely manner, and has low decision-making accuracy.

[0300] Implementation plan: Apply this system to improve the management decision-making level. The system integrates multi-dimensional data of production, sales, finance, and market, and analyzes business relevance and predicts development trends through a multi-modal Transformer network. The system constructs an enterprise operation knowledge graph to realize multi-dimensional data visualization and intelligent decision-making recommendations.

[0301] Implementation effect: After application, the decision-making accuracy rate is increased by 45%, the market response speed is increased by 55%, and the strategic adjustment efficiency is increased by 40%.

[0302] As Figure 2 shown, a MES digital collaborative management system based on deep learning includes:

[0303] Data acquisition module: It is used to acquire multi-source heterogeneous data, and through preprocessing and normalization processing, a dimensionless multi-dimensional data set is constructed;

[0304] Dynamic association module: It is used to map the dimensionless multi-dimensional data set into graph structure nodes to obtain node features, and define the connection exchange, information flow, and dependency relationships between parameters in the dimensionless multi-dimensional data set as edge relationships; through node feature embedding and edge relationship learning, analyze the dynamic association relationships between data items in the dimensionless multi-dimensional data set, and generate a production line operation dependency relationship graph;

[0305] Detection and positioning module: It is used to construct a multi-source heterogeneous data collaborative analysis mechanism to obtain the dynamic association weights of collaboration between multi-source heterogeneous data modalities, and combine with the dependency relationship graph for anomaly detection and root cause location to obtain the causal association information between anomaly events and data modalities;

[0306] Anomaly impact module: It is used to construct a digital twin simulation environment based on the causal association information, simulate the influence path of heterogeneous data parameter intervention on production disturbances on collaborative anomalies, and evaluate the anomaly impact results of parameter adjustment;

[0307] Model training module: It is used to construct an iterative scheduling model according to the dimensionless multi-dimensional data set, the dependency relationship graph, and the causal association information;

[0308] Collaborative management module; it is used to combine the abnormal impact assessment results with the real-time production status, iteratively generate a collaborative scheduling strategy for multi-objective optimization, and achieve dynamic configuration and abnormal prevention among multi-source heterogeneous data.

[0309] The above describes the embodiments of the present invention. However, these embodiments are not limited to the above specific implementation manners. The above specific implementation manners are merely illustrative and not restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make more forms of the same embodiments, all of which fall within the protection scope of this embodiment.

Claims

1. A digital collaborative management method for MES based on deep learning, characterized in that, It includes the following steps: Collect multi-source heterogeneous data, and through preprocessing and normalization processing, construct a dimensionless multi-dimensional data set; Map the dimensionless multi-dimensional data set to graph structure nodes to obtain node features, and define the connection exchange, information flow, and dependency relationships among the parameters in the dimensionless multi-dimensional data set as edge relationships; Through node feature embedding and edge relationship learning, analyze the dynamic association relationships among the data items in the dimensionless multi-dimensional data set, and generate a production line operation dependency relationship map; Construct a multi-source heterogeneous data collaborative analysis mechanism to obtain the dynamic association weights for collaboration among multi-source heterogeneous data modalities, and combine the dependency relationship map for anomaly detection and root cause location to obtain the causal association information between anomaly events and data modalities; Based on the causal association information, construct a digital twin simulation environment to simulate the influence path of heterogeneous data parameter intervention on production disturbances on collaborative anomalies, so as to evaluate the anomaly influence results of parameter adjustment; According to the dimensionless multi-dimensional data set, dependency relationship map, and causal association information, construct an iterative scheduling model, and combine the anomaly influence evaluation results with the real-time production status to iteratively generate a collaborative scheduling strategy for multi-objective optimization, realizing the dynamic configuration and anomaly prevention among multi-source heterogeneous data.

2. The MES digital collaborative management method based on deep learning according to claim 1, characterized in that Collect multi-source heterogeneous data, and through preprocessing and normalization processing, construct a dimensionless multi-dimensional data set, including: Collect multi-source heterogeneous data during the operation of production manufacturing in real time and perform data preprocessing to obtain a preprocessed data set; Perform standardization conversion on the preprocessed data set and perform data cleaning operations to obtain a clean data set; Perform time series alignment processing on the clean data set, and perform normalization processing after collecting it at a preset period to obtain a dimensionless multi-dimensional data set.

3. A MES digital collaborative management method based on deep learning according to claim 1, characterized in that Through node feature embedding and edge relationship learning, analyze the dynamic association relationships among the data items in the dimensionless multi-dimensional data set, and generate a production line operation dependency relationship map, including: Map the dimensionless multi-dimensional data set to graph structure nodes, and use multi-source heterogeneous data parameters as node features; Define the connection exchange, information flow, and dependency relationships among multi-source heterogeneous data parameters as edge relationships; Use the graph attention network structure for node feature embedding to capture the local correlation of edge relationships; Through the edge relationship feature learning mechanism, quantify the interaction intensity among multi-source heterogeneous data modalities; Through multi-layer graph network iterative propagation based on local correlation and interaction intensity, analyze the dynamic association relationships among the data items in the dimensionless multi-dimensional data set, and generate a production line operation dependency relationship map including the association status among heterogeneous data, production operation parameters, and parameter quality indicators.

4. A MES digital collaborative management method based on deep learning according to claim 3, characterized in that, Process the operation status among the dynamically changing multi-source heterogeneous data parameters to make the time series features and graph network topological features fuse with each other to obtain a fusion result; Based on the fusion result, splice and reduce the dimensionality of the static attributes and dynamic time series data of the data item parameters in the dimensionless multi-dimensional data set to generate a fusion feature vector for a single node, which is used as the basic input for constructing the dependency relationship map; Obtain the cosine similarity among the fusion feature vectors of each node to generate implicit edges, and at the same time use the interaction intensity connection to define explicit edges. After fusion, form a weighted adjacency matrix and jointly output dynamic edge weights with the node features; By learning the dynamic edge weights between node features, aggregating the multi-attention results to update the dynamic edge weights, and fusing the neighbor node features to generate updated node features; The production line operation dependency graph is constructed based on the dynamic edge weights and the updated node features, revealing the explicit associations and implicit dynamic dependencies between multi-source heterogeneous data items during production operation to support production line operation decisions.

5. A MES digital collaborative management method based on deep learning according to claim 1, characterized in that, Construct a collaborative analysis mechanism for multi-source heterogeneous data to obtain the dynamic correlation weights of the collaboration between multi-source heterogeneous data modalities, and combine the dependency graph to perform anomaly detection and root cause location to obtain the causal relationship information between abnormal events and data modalities, including: Perform feature extraction and representation learning on the states between data items in dimensionless multidimensional data sets to obtain the correlation matrix between multi-source heterogeneous data modalities; The contribution between multi-source heterogeneous data modalities is adjusted by dynamically allocating dynamic edge weights to obtain the dynamic correlation weights for collaboration between multi-source heterogeneous data modalities. The correlation influence between multi-source heterogeneous data modalities is captured according to the dynamic correlation weights, and a dynamic correlation weight matrix is generated; Combine the dynamic association weight matrix with the production line operation dependency graph to build an enhanced graph structure topology; By identifying node features and edge relationships that deviate from normal patterns on the enhanced graph structure topology, anomaly identification results can be obtained; Based on the anomaly recognition results, the abnormal root node is located through back propagation and subgraph extraction to obtain the root cause location data; Causal inference is used to quantify the causal correlation strength between the anomaly identification results and the root cause location data to generate causal correlation information; the causal correlation strength represents the degree of direct influence of the anomaly identification results on the root cause location data, and is determined by obtaining the difference between the mutual information and the conditional mutual information.

6. A MES digital collaborative management method based on deep learning according to claim 5, characterized in that, A digital twin simulation environment is built based on causal association information to simulate the impact path of heterogeneous data parameter intervention production disturbance on collaborative anomalies, so as to evaluate the abnormal impact results of parameter adjustment, including: Map the acquired causal information into the digital twin simulation environment to establish the corresponding relationship between the physical entity and the digital model; Define the interaction rules in the digital twin simulation environment based on the production line operation dependency graph to ensure that the digital twin simulation environment accurately reflects the dynamic characteristic data of actual production operation; Design parameter intervention experimental schemes based on dynamic characteristic data of production operation, and simulate production disturbance scenarios by adjusting key parameters in the digital twin simulation environment at fixed points; Based on the production disturbance scenario, the propagation path of the production disturbance is tracked, and the state changes of each node feature are recorded to analyze the impact of the production disturbance scenario on the coordinated anomaly between multi-source heterogeneous data modalities; Comprehensively evaluate the suppression or aggravation effects of production disturbance schemes simulated by various parameter adjustment schemes on coordinated anomalies to form quantitative anomaly impact assessment results.

7. A MES digital collaborative management method based on deep learning according to claim 6, characterized in that, Construct an iterative scheduling model based on dimensionless multidimensional data sets, dependency graphs, and causal association information, including: Based on the dimensionless multidimensional data set, the state representation space is constructed to map the current state of production operation into a high-dimensional feature vector, which contains multi-dimensional information data such as equipment operation parameters, quality indicators, energy consumption data and safety monitoring data; Define a constraint condition matrix using the production line operation dependency graph. The constraint condition matrix characterizes the precedence dependencies, resource sharing, and conflict relationships among production units to generate a scheduling relationship matrix that meets the requirements of production operation constraints. Construct an iterative scheduling model based on multi-dimensional information data, the scheduling relationship matrix, and causal association information. The iterative scheduling model specifically includes: Generate structural data from the multi-dimensional information data, the scheduling relationship matrix, and the causal association information, and encode the structural data into sequence data. Input the sequence data into the iterative scheduling model. The iterative scheduling model includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer. Transmit the intermediate representation data of multiple hidden layers to the output layer, and the output layer iteratively outputs the scheduling strategy prediction results corresponding to the state changes among multi-source heterogeneous data parameters. Input at least two multi-dimensional information data, the scheduling relationship matrix, and the causal association information data items obtained in real time into the iterative scheduling model, and iteratively output the scheduling strategy prediction results corresponding to the state changes among the multi-source heterogeneous data parameters obtained in real time.

8. A MES digital collaborative management method based on deep learning according to claim 7, characterized in that Combine the abnormal impact assessment results with the real-time production status, and iteratively generate a collaborative scheduling strategy for multi-objective optimization to achieve dynamic configuration and abnormal prevention among multi-source heterogeneous data, including: Collect key parameters of the real-time production operation status, including equipment operation status, quality inspection data, energy consumption indicators, and safety monitoring information, to form a multi-dimensional real-time status vector. Extract the current topological structure and constraint conditions from the production line operation dependency graph, and combine the historical abnormal pattern library to capture abnormal risks for the real-time status. Perform multi-step prediction on the future state evolution based on the captured abnormal risks to identify potential abnormal development trends. For the detected abnormal risk points, generate a design parameter intervention effect scoring matrix according to the abnormal impact assessment results. Start a multi-objective optimization decision-making mechanism according to the design parameter intervention effect scoring matrix, set the key indicators of production efficiency, quality stability, energy utilization rate, and safety risk as optimization objectives, and dynamically adjust the weights of each objective according to the current production task priority. Search for the optimal scheduling strategy under the constraint conditions to balance the short-term intervention effect and the long-term stable scheduling strategy. Execute the collaborative scheduling strategy deployment based on the prediction results of the iterative scheduling model, and decompose the optimized scheduling instructions into specific execution instructions at the equipment level, process level, and resource level. Safely transmit the specific execution instructions to each execution unit through the middleware layer, and at the same time establish an instruction execution confirmation mechanism to achieve dynamic configuration and abnormal prevention among multi-source heterogeneous data. Establish a real-time monitoring and feedback mechanism to continuously track the execution effect of the scheduling strategy and collect the change data of key performance indicators. When a deviation in execution or a new abnormal event is detected, trigger a quick response mechanism and select to fine-tune or re-plan the strategy according to the degree of deviation.

9. A digital collaborative management method based on deep learning as claimed in claim 8, wherein The collaborative scheduling strategy also includes: Decompose the scheduling problem into three levels: strategic, tactical, and operational, and process the abnormal situations among long-term planning, medium-term scheduling, and real-time response heterogeneous data respectively to form a complete hierarchical collaborative decision-making framework. By realizing the bidirectional integration of constraint scenarios and guiding scenarios, using the differentiable programming layer to unify hard constraints and soft constraints into the same framework, dynamically adjusting the constraint weights according to the real-time production status, and maximizing the comprehensive response to the production operation requirements; Dynamically adjust the multi-objective priorities of production efficiency, quality stability, energy utilization rate, and safety risk according to the enterprise's strategic focus, market demand changes, and actual production conditions, and achieve dynamic balance among multi-dimensional objectives; Design a mechanism that combines anomaly prediction and forward-looking scheduling. Through a time-series prediction model, perform multi-step prediction on the production status and combine it with a digital twin simulation environment for cyclic simulation and evaluation. Actively adjust the state between heterogeneous data parameters before the anomaly actually occurs, and transform passive response into active prevention; Establish a combination of knowledge-driven and data-driven, and through a distributed collaborative execution framework, make the collaborative scheduling strategy process transparent and efficiently executed.

10. A deep learning-based MES digital collaborative management system for implementing a deep learning-based MES digital collaborative management method as described in any one of claims 1-9, characterized in that, Including: Data acquisition module: It is used to collect multi-source heterogeneous data, and through preprocessing and normalization processing, construct a dimensionless multi-dimensional data set; Dynamic association module: It is used to map the dimensionless multi-dimensional data set into graph structure nodes to obtain node features, and define the connection exchange, information flow, and dependency relationships among the parameters in the dimensionless multi-dimensional data set as edge relationships; Through node feature embedding and edge relationship learning, analyze the dynamic association relationships among the data items in the dimensionless multi-dimensional data set, and generate a production line operation dependency relationship map; Detection and location module: It is used to construct a multi-source heterogeneous data collaborative analysis mechanism to obtain the dynamic association weights of collaboration among multi-source heterogeneous data modalities, and combine the dependency relationship map for anomaly detection and root cause location to obtain the causal association information between anomaly events and data modalities; Anomaly impact module: It is used to construct a digital twin simulation environment based on the causal association information, simulate the influence path of heterogeneous data parameter intervention on production disturbances on collaborative anomalies, and evaluate the anomaly impact results of parameter adjustment; Model training module: It is used to construct an iterative scheduling model according to the dimensionless multi-dimensional data set, dependency relationship map, and causal association information; Collaborative management module; It is used to iteratively generate a collaborative scheduling strategy for multi-objective optimization by combining the anomaly impact evaluation results and the real-time production status, and achieve dynamic configuration and anomaly prevention among multi-source heterogeneous data.

Images