A method and system for safe production control of printing press equipment

Abstract

This invention provides a method and system for safe production control of printing press equipment, relating to the field of printing press technology. It includes collecting operating data of a group of printing presses using multimodal sensors, extracting dynamic features using a temporal attention mechanism, and constructing a spatiotemporal correlation matrix to represent the coupling relationships between printing presses. A hierarchical deep learning model is constructed, integrating feature fusion, temporal prediction, and risk assessment to analyze equipment health status, calculate the overall safety risk index, and identify faulty equipment. Based on a deep reinforcement learning-based scheduling decision model, combined with safety thresholds, resource allocation, and production efficiency constraints, an optimal scheduling scheme including equipment start-up and shutdown sequence and load distribution strategies is generated, achieving safe and efficient scheduling of the printing press group. This invention can effectively identify potential risks, avoid safety accidents, and improve production efficiency.

Description

Technical Field

[0001] This invention relates to printing press technology, and more particularly to a method and system for safe production control of printing press equipment. Background Technology

[0002] The ever-expanding scale of printing press fleets places higher demands on their safe and efficient scheduling and management. Traditional printing press scheduling methods rely mainly on manual experience and rules, which are difficult to adapt to complex and ever-changing production environments. With the development of IoT and AI technologies, data-driven intelligent scheduling methods are gradually emerging, providing new ideas for the optimized scheduling of printing press fleets.

[0003] Difficulty in fully perceiving equipment status: Traditional scheduling methods usually rely on monitoring only a few key parameters, which cannot fully reflect the real-time operating status of the printing press group, and can easily lead to safety hazards and efficiency losses.

[0004] Lack of consideration for the interrelationships between equipment: There are complex coupling relationships between printing presses; for example, a failure in one printing press may affect the normal operation of other printing presses. Traditional scheduling methods often ignore this interrelationship, making it difficult to achieve globally optimal scheduling.

[0005] Poor adaptability of scheduling schemes: Changes in the production environment, such as fluctuations in order demand and degradation of equipment performance, can affect the effectiveness of scheduling schemes. Traditional scheduling methods struggle to dynamically adjust scheduling strategies in response to environmental changes, resulting in poor adaptability of scheduling schemes. Summary of the Invention

[0006] This invention provides a method and system for safe production control of printing press equipment, which can solve the problems in the prior art.

[0007] A first aspect of the present invention,

[0008] A method for controlling safe production in printing press equipment is provided, including:

[0009] Real-time operational data is collected by a multimodal sensor network deployed in the printing press cluster. The standardized data stream corresponding to the operational data is then subjected to feature extraction using a temporal attention mechanism to obtain the dynamic feature vector of the printing press cluster. The spatiotemporal correlation matrix of the printing press cluster is then constructed based on the dynamic feature vector.

[0010] Based on the dynamic feature vector and the spatiotemporal correlation matrix, a hierarchical deep learning model is constructed. Historical labeled data is input into the hierarchical deep learning model for training. The trained deep learning model is used to analyze the dynamic feature vector to generate equipment health status assessment results. Based on the equipment health status assessment results and the spatiotemporal correlation matrix, the overall safety risk index and equipment identification results of the printing press group are calculated.

[0011] Based on the overall safety risk index and the equipment identification results, intelligent scheduling constraints for the printing press group are established; a scheduling decision model is constructed based on a deep reinforcement learning algorithm, using safety threshold constraints, resource allocation constraints, and production efficiency constraints as evaluation indicators for the reward function; the optimal scheduling scheme is generated using the scheduling decision model, which includes equipment start-up and shutdown sequence and load distribution strategy; the optimal scheduling scheme is then distributed to the control systems of each printing press for execution.

[0012] Based on the dynamic feature vector and the spatiotemporal correlation matrix, a hierarchical deep learning model is constructed. Training the hierarchical deep learning model by inputting historical labeled data includes:

[0013] The hierarchical deep learning model includes a feature fusion layer, a temporal prediction layer, and a risk assessment layer. In the feature fusion layer, the dynamic feature vector is converted into a query matrix, a key matrix, and a value matrix through a self-attention mechanism, and the inter-feature association weights are calculated through scaled dot product attention. The spatiotemporal association matrix is ​​input into a graph attention network to extract the spatial dependency features between printing presses. The inter-feature association weights and the spatial dependency features are adaptively fused through a gating mechanism to generate a first comprehensive feature representation.

[0014] In the temporal prediction layer, the first comprehensive feature representation is input into a bidirectional long short-term memory network and a multi-scale parallel convolutional neural network, respectively. The bidirectional long short-term memory network extracts temporal features, and the multi-scale parallel convolutional neural network extracts local features. The output features of the two are fused through a residual connection structure to generate a second comprehensive feature representation.

[0015] The second comprehensive feature representation is input into the risk assessment layer to generate a second prediction result;

[0016] Historical labeled data is input into the hierarchical deep learning model for training. The first loss value of the first prediction result is calculated using the mean squared error loss function, and the second loss value of the second prediction result is calculated using the cross-entropy loss function. The first loss value and the second loss value are adaptively balanced using learnable weights to obtain the total loss value. The parameters of the hierarchical deep learning model are optimized based on the total loss value to obtain the trained deep learning model.

[0017] The trained deep learning model is used to analyze the dynamic feature vectors to generate equipment health status assessment results. Based on the equipment health status assessment results and the spatiotemporal correlation matrix, the overall safety risk index and equipment identification results of the printing press group are calculated, including:

[0018] The trained deep learning model is used to analyze the dynamic feature vector, and multiple prediction results are fused through weighted voting to generate a device health status assessment result.

[0019] Based on the equipment health status assessment results and the spatiotemporal correlation matrix, a weighted directed graph is constructed. The nodes of the weighted directed graph represent printing press equipment, and the edge weights represent the risk propagation intensity between equipment. The PageRank algorithm is used to calculate the node importance score.

[0020] The node importance score is combined with a local risk accumulation function, which incorporates a time-dimensional decay factor and a spatial-dimensional graph convolution feature, to calculate the overall safety risk index of the printing press group.

[0021] Based on the weighted directed graph, multi-criteria decision analysis is performed, and equipment is identified through fuzzy comprehensive evaluation method to generate equipment identification results.

[0022] Based on the overall safety risk index and the equipment identification results, intelligent scheduling constraints for the printing press group are established; a scheduling decision model is constructed based on a deep reinforcement learning algorithm, using safety threshold constraints, resource allocation constraints, and production efficiency constraints as evaluation indicators for the reward function, including:

[0023] A risk threshold vector is set according to the overall security risk index. The risk threshold vector is used to divide risk areas that are higher than the preset risk threshold and risk areas that are lower than the preset risk threshold. Corresponding scheduling restriction strategies are set for different risk areas.

[0024] Based on the equipment identification results, resource allocation constraints are constructed. These constraints include an equipment importance scoring matrix and a backup equipment allocation strategy. The equipment importance scoring matrix is ​​used to quantify the importance of equipment in the production system. The scheduling restriction strategy and the resource allocation constraints are combined to form intelligent scheduling constraints for the printing press group.

[0025] A scheduling decision model is constructed based on a deep reinforcement learning algorithm, mapping the intelligent scheduling constraints into a state space and an action space. The state space contains the real-time operating status information of the printing press group, and the action space defines the set of executable scheduling operations. A composite reward function is constructed, using safety threshold constraints, resource allocation constraints, and production efficiency constraints as evaluation indicators for the composite reward function.

[0026] The optimal scheduling scheme is generated using the aforementioned scheduling decision model, and the optimal scheduling scheme includes equipment start-up and shutdown sequence and load distribution strategy; the optimal scheduling scheme is then distributed to each printing press control system for execution, including:

[0027] The optimal scheduling scheme is generated based on the scheduling decision model. A device start-stop state transition matrix is ​​constructed using a dynamic programming algorithm. The load allocation strategy is optimized by combining a mixed integer programming method. The load allocation strategy is based on the remaining capacity of the equipment and the urgency of the task, and a task allocation weight coefficient is established to achieve dynamic balance of the load of the equipment group. The device start-stop state transition matrix and the load allocation strategy are combined to form the optimal scheduling scheme.

[0028] A hierarchical scheduling instruction distribution system is constructed, which parses the optimal scheduling scheme into equipment-level control instructions; a real-time communication mechanism is established between the central scheduler and the local control system of the printing press to send the equipment-level control instructions to each printing press control system.

[0029] The optimal scheduling scheme is generated based on the aforementioned scheduling decision model. A dynamic programming algorithm is used to construct the equipment start-stop state transition matrix. A mixed-integer programming method is combined to optimize the load allocation strategy. This load allocation strategy establishes task allocation weight coefficients based on the equipment's remaining capacity and task urgency, achieving dynamic load balancing for the equipment group. The optimal scheduling scheme is formed by combining the equipment start-stop state transition matrix and the load allocation strategy.

[0030] The equipment start-stop state transition matrix includes the state transition rules, state transition costs, and preheating time constraints of the printing press; the state value function and the state action value function are calculated using a value iteration algorithm, and the optimal start-stop strategy is determined based on the state value function and the state action value function.

[0031] The remaining capacity of the printing press is calculated based on its operating status, and the urgency of the tasks is assessed based on their delivery time. A task allocation weighting coefficient is established using the remaining capacity and the urgency of the tasks, and this coefficient is used to prioritize tasks and allocate resources.

[0032] Based on the task allocation weight coefficients, load balancing constraints are constructed, including equipment capacity limits, task timing relationships, and process parameter requirements. By solving a mixed integer programming model, a task allocation scheme that satisfies the load balancing constraints is obtained, thereby achieving dynamic balance of the printing press group's load.

[0033] A second aspect of the present invention,

[0034] A safety production control system for printing press equipment is provided, comprising:

[0035] The first unit is used to collect operational data in real time through a multimodal sensor network deployed in the printing press cluster, extract features from the standardized data stream corresponding to the operational data using a temporal attention mechanism to obtain the dynamic feature vector of the printing press cluster, and construct the spatiotemporal correlation matrix of the printing press cluster based on the dynamic feature vector.

[0036] The second unit is used to construct a hierarchical deep learning model based on the dynamic feature vector and the spatiotemporal correlation matrix, input historical labeled data into the hierarchical deep learning model for training, use the trained deep learning model to analyze the dynamic feature vector, and generate equipment health status assessment results; based on the equipment health status assessment results and the spatiotemporal correlation matrix, calculate the overall safety risk index and equipment identification results of the printing press group.

[0037] The third unit is used to establish intelligent scheduling constraints for the printing press group based on the overall safety risk index and the equipment identification results; construct a scheduling decision model based on a deep reinforcement learning algorithm, and use safety threshold constraints, resource allocation constraints, and production efficiency constraints as evaluation indicators of the reward function; generate an optimal scheduling scheme using the scheduling decision model, the optimal scheduling scheme including equipment start-up and shutdown sequence and load allocation strategy; and distribute the optimal scheduling scheme to each printing press control system for execution.

[0038] A third aspect of the embodiments of the present invention,

[0039] An electronic device is provided, comprising:

[0040] processor;

[0041] Memory used to store processor-executable instructions;

[0042] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0043] Fourth aspect of the present invention,

[0044] A computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0045] The beneficial effects of this application are as follows:

[0046] 1. Enhance the safety of printing press fleets: By monitoring and analyzing the operating status of printing press fleets in real time, potential safety risks can be effectively identified, and corresponding scheduling measures can be taken in a timely manner to avoid accidents and ensure production safety.

[0047] 2. Optimize resource allocation and improve production efficiency: The intelligent scheduling method based on deep reinforcement learning can dynamically adjust the load distribution and start-up / shutdown sequence of printing presses according to the health status of the equipment and production needs, thereby optimizing resource allocation and improving production efficiency.

[0048] 3. Achieve autonomous learning and continuous optimization of printing press fleets: By feeding back execution performance data into the model training process, the performance of deep learning models and scheduling decision models can be continuously optimized, making the scheduling strategy of printing press fleets more intelligent and adaptive. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating a method for controlling safe production of printing equipment according to an embodiment of the present invention;

[0050] Figure 2 This is a schematic diagram of the structure of a safety production control system for printing equipment according to an embodiment of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0053] Figure 1 This is a flowchart illustrating a method for controlling safe production of printing equipment according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0054] Real-time operational data is collected by a multimodal sensor network deployed in the printing press cluster. The standardized data stream corresponding to the operational data is then subjected to feature extraction using a temporal attention mechanism to obtain the dynamic feature vector of the printing press cluster. The spatiotemporal correlation matrix of the printing press cluster is then constructed based on the dynamic feature vector.

[0055] Based on the dynamic feature vector and the spatiotemporal correlation matrix, a hierarchical deep learning model is constructed. Historical labeled data is input into the hierarchical deep learning model for training. The trained deep learning model is used to analyze the dynamic feature vector to generate equipment health status assessment results. Based on the equipment health status assessment results and the spatiotemporal correlation matrix, the overall safety risk index and equipment identification results of the printing press group are calculated.

[0056] Based on the overall safety risk index and the equipment identification results, intelligent scheduling constraints for the printing press group are established; a scheduling decision model is constructed based on a deep reinforcement learning algorithm, using safety threshold constraints, resource allocation constraints, and production efficiency constraints as evaluation indicators for the reward function; the optimal scheduling scheme is generated using the scheduling decision model, which includes equipment start-up and shutdown sequence and load distribution strategy; the optimal scheduling scheme is then distributed to the control systems of each printing press for execution.

[0057] In one optional implementation, a hierarchical deep learning model is constructed based on the dynamic feature vector and the spatiotemporal correlation matrix, and historical labeled data is input into the hierarchical deep learning model for training, including:

[0058] The hierarchical deep learning model includes a feature fusion layer, a temporal prediction layer, and a risk assessment layer. In the feature fusion layer, the dynamic feature vector is converted into a query matrix, a key matrix, and a value matrix through a self-attention mechanism, and the inter-feature association weights are calculated through scaled dot product attention. The spatiotemporal association matrix is ​​input into a graph attention network to extract the spatial dependency features between printing presses. The inter-feature association weights and the spatial dependency features are adaptively fused through a gating mechanism to generate a first comprehensive feature representation.

[0059] In the temporal prediction layer, the first comprehensive feature representation is input into a bidirectional long short-term memory network and a multi-scale parallel convolutional neural network, respectively. The bidirectional long short-term memory network extracts temporal features, and the multi-scale parallel convolutional neural network extracts local features. The output features of the two are fused through a residual connection structure to generate a second comprehensive feature representation.

[0060] The second comprehensive feature representation is input into the risk assessment layer to generate a second prediction result;

[0061] Historical labeled data is input into the hierarchical deep learning model for training. The first loss value of the first prediction result is calculated using the mean squared error loss function, and the second loss value of the second prediction result is calculated using the cross-entropy loss function. The first loss value and the second loss value are adaptively balanced using learnable weights to obtain the total loss value. The parameters of the hierarchical deep learning model are optimized based on the total loss value to obtain the trained deep learning model.

[0062] Obtain the dynamic feature vectors and spatiotemporal correlation matrices of the devices. The dynamic feature vectors can include the device's operating data and environmental data, such as temperature, pressure, and vibration frequency. The spatiotemporal correlation matrix describes the spatial relationships and mutual influence between different devices. For example, the element values ​​in the matrix can represent the distance between devices, their connection relationships, or the intensity of information transmission. Assume there are 3 devices with dynamic feature vectors [1,2,3], [4,5,6], and [7,8,9], and a spatiotemporal correlation matrix of [[1,0.5,0.2],[0.5,1,0.8],[0.2,0.8,1]].

[0063] A hierarchical deep learning model is constructed, consisting of a feature fusion layer, a temporal prediction layer, and a risk assessment layer. In the feature fusion layer, a self-attention mechanism is first used to convert dynamic feature vectors into query matrices, key matrices, and value matrices. Then, scaled dot product attention is used to calculate the inter-feature association weights, capturing the relationships between different feature dimensions. Simultaneously, the spatiotemporal association matrix is ​​input into a graph attention network to extract spatial dependency features between devices. Finally, a gating mechanism is used to adaptively fuse the inter-feature association weights and spatial dependency features to generate a first comprehensive feature representation. For example, if the inter-feature association weights for device 1 are [0.2, 0.3, 0.5] and the spatial dependency features are [0.6, 0.3, 0.1], the fused first comprehensive feature representation is [0.4, 0.35, 0.3].

[0064] In the temporal prediction layer, the first comprehensive feature representation is input into a bidirectional long short-term memory network and a multi-scale parallel convolutional neural network, respectively. The bidirectional long short-term memory network can capture long-term dependencies in time series data and extract temporal features. The multi-scale parallel convolutional neural network can extract local features at different scales. Then, the output features of the two networks are fused through a residual connection structure to generate a second comprehensive feature representation. For example, the temporal features of device 1 are [0.1, 0.2, 0.7], the local features are [0.3, 0.5, 0.2], and the fused second comprehensive feature representation is [0.2, 0.35, 0.45].

[0065] The second comprehensive feature representation is input into the risk assessment layer. This layer constructs a main branch and an auxiliary branch based on a multi-task learning framework. The main branch uses a multilayer perceptron to predict the device health index and generates a first prediction result. The auxiliary branch uses a conditional random field model to model the risk propagation mechanism between devices and generates a second prediction result. For example, the predicted value of the health index of device 1 is 0.8, and the risk propagation probability is [0.1, 0.2, 0.7].

[0066] The model is trained using historical labeled data. A first loss value is calculated using the mean squared error loss function for the first prediction result, and a second loss value is calculated using the cross-entropy loss function for the second prediction result. The first and second loss values ​​are adaptively balanced using learnable weights to obtain the total loss value. The model parameters are then optimized based on the total loss value to obtain the trained deep learning model. For example, if the true value of the health index of device 1 is 0.9, then the first loss value is 0.01. Assuming the true probability of risk propagation from device 1 is [0.2, 0.1, 0.7], then the second loss value is 0.1. Assuming the learnable weight is 0.5, then the total loss value is 0.055.

[0067] The solution proposed in this application can:

[0068] Improved prediction accuracy. By fusing dynamic feature vectors and spatiotemporal correlation matrices, and combining multiple deep learning models, device status information can be captured more comprehensively, thereby improving prediction accuracy. Enhanced model robustness. The multi-task learning framework and residual connection structure effectively alleviate overfitting problems, enhancing the model's generalization ability and robustness. Improved prediction efficiency. The hierarchical structure and parallel computing mechanism effectively reduce the model's computational complexity and improve prediction efficiency.

[0069] In one optional implementation, the trained deep learning model is used to analyze the dynamic feature vector to generate equipment health status assessment results; based on the equipment health status assessment results and the spatiotemporal correlation matrix, the overall safety risk index and equipment identification results of the printing press group are calculated, including:

[0070] The trained deep learning model is used to analyze the dynamic feature vector, and multiple prediction results are fused through weighted voting to generate a device health status assessment result.

[0071] Based on the equipment health status assessment results and the spatiotemporal correlation matrix, a weighted directed graph is constructed. The nodes of the weighted directed graph represent printing press equipment, and the edge weights represent the risk propagation intensity between equipment. The PageRank algorithm is used to calculate the node importance score.

[0072] The node importance score is combined with a local risk accumulation function, which incorporates a time-dimensional decay factor and a spatial-dimensional graph convolution feature, to calculate the overall safety risk index of the printing press group.

[0073] Based on the weighted directed graph, multi-criteria decision analysis is performed, and equipment is identified through fuzzy comprehensive evaluation method to generate equipment identification results.

[0074] A distributed sensor network collects various operational data from the printing press fleet in real time, such as vibration signals, temperature, current, and pressure. This data undergoes data cleaning and feature engineering to transform it into a high-dimensional dynamic feature vector, which characterizes the real-time operating status of the equipment. For example, the vibration signal of a printing press, after undergoing a Fast Fourier Transform, yields vibration amplitudes across multiple frequency bands. These amplitudes, along with data such as temperature and current, constitute the dynamic feature vector of that printing press.

[0075] To improve the robustness and generalization ability of the model, the dynamic feature vectors are preprocessed. A sliding window mechanism is used to segment the dynamic feature vector sequence into multiple overlapping feature windows, each containing feature data within a certain time period. For example, with a window length of 10 seconds and a sliding step size of 5 seconds, the continuous dynamic feature vector sequence is segmented into multiple overlapping 10-second windows. Then, wavelet transform is applied to each feature window for noise reduction to remove high-frequency noise interference. Finally, the denoised feature sequence is standardized to map the data to a uniform numerical range, eliminating the influence of different feature units. For example, all feature data are normalized to the range of 0 to 1.

[0076] The preprocessed feature sequences are input into multiple heterogeneous sub-models for health status assessment. These sub-models employ different deep learning network structures and parameter configurations, such as convolutional neural networks, recurrent neural networks, and long short-term memory networks. Each sub-model is trained on a large amount of historical data, enabling it to capture subtle changes in equipment operating status from different perspectives. To fully utilize the advantages of different sub-models, the evaluation weights of each sub-model are dynamically adjusted based on historical performance. For example, if a sub-model has a high accuracy rate in predicting equipment failures over a past period, it is given a higher weight. The prediction results of multiple sub-models are merged through weighted voting to generate the final equipment health status assessment result. For example, if three sub-models predict the health status of a printing press as 0.8, 0.9, and 0.7, respectively, with corresponding weights of 0.3, 0.5, and 0.2, the final health status assessment result is 0.8 × 0.3 + 0.9 × 0.5 + 0.7 × 0.2 = 0.83.

[0077] Based on the equipment health status assessment results and a pre-established spatiotemporal correlation matrix, a weighted directed graph is constructed. Nodes in the graph represent printing press equipment, and edge weights represent the risk propagation intensity between equipment, determined by factors such as physical distance and process flow correlation. For example, two adjacent printing presses with closely related processes exhibit higher risk propagation intensity and correspondingly larger edge weights. The PageRank algorithm is used to calculate the importance score of each node, reflecting its influence within the printing press cluster. For instance, a printing press in a core position connecting multiple other machines has a higher importance score. The node importance score is combined with a local risk accumulation function to calculate the overall safety risk index of the printing press cluster. The local risk accumulation function incorporates a time-dimensional decay factor and a spatial-dimensional graph convolution feature, comprehensively considering the impact of the equipment's own risk and the risks of surrounding equipment. The time-dimensional decay factor makes recent risks have a greater impact on the overall risk index, while the spatial-dimensional graph convolution feature considers the propagation effect of risks between equipment.

[0078] Multi-criteria decision analysis is performed based on the constructed weighted directed graph. This analysis comprehensively considers the intrinsic importance of equipment (e.g., equipment value, performance), topological importance (e.g., equipment centrality in the network), and dynamic importance (e.g., equipment real-time health status). A fuzzy comprehensive evaluation method is used to identify the critical equipment that has the greatest impact on the overall security risk of the printing press fleet. For example, a printing press with high intrinsic importance, a critical topological position, and poor health status will be identified as critical equipment. The calculated overall security risk index and equipment identification results are input into the intelligent scheduling decision module to optimize production scheduling schemes, such as adjusting production plans and arranging equipment maintenance, to reduce overall security risk.

[0079] The solution proposed in this application can:

[0080] Improving the accuracy and reliability of equipment health status assessment: Through multi-model fusion and dynamic weight adjustment mechanisms, the limitations of a single model can be effectively reduced, improving the accuracy and reliability of assessment results. Precisely quantifying the overall safety risk of printing press clusters: By constructing a weighted directed graph and introducing a spatiotemporal correlation matrix, the risk propagation effect between equipment can be considered more comprehensively, thereby more accurately quantifying the overall safety risk of the printing press cluster. Achieving intelligent scheduling decisions and risk prevention: By inputting the overall safety risk index and equipment identification results into the intelligent scheduling decision module, production scheduling schemes can be optimized, preventative measures can be taken in advance, safety risks can be reduced, and major accidents can be avoided.

[0081] In one optional implementation, intelligent scheduling constraints for the printing press group are established based on the overall safety risk index and the equipment identification results; a scheduling decision model is constructed based on a deep reinforcement learning algorithm, and safety threshold constraints, resource allocation constraints, and production efficiency constraints are used as evaluation indicators for the reward function, including:

[0082] A risk threshold vector is set according to the overall security risk index. The risk threshold vector is used to divide risk areas that are higher than the preset risk threshold and risk areas that are lower than the preset risk threshold. Corresponding scheduling restriction strategies are set for different risk areas.

[0083] Based on the equipment identification results, resource allocation constraints are constructed. These constraints include an equipment importance scoring matrix and a backup equipment allocation strategy. The equipment importance scoring matrix is ​​used to quantify the importance of equipment in the production system. The scheduling restriction strategy and the resource allocation constraints are combined to form intelligent scheduling constraints for the printing press group.

[0084] A scheduling decision model is constructed based on a deep reinforcement learning algorithm, mapping the intelligent scheduling constraints into a state space and an action space. The state space contains the real-time operating status information of the printing press group, and the action space defines the set of executable scheduling operations. A composite reward function is constructed, using safety threshold constraints, resource allocation constraints, and production efficiency constraints as evaluation indicators for the composite reward function.

[0085] Safety risk assessments and equipment identification are conducted. Real-time operational data from the printing press fleet, such as temperature, pressure, and vibration, as well as equipment operating status and fault records, are collected via a sensor network. Using this data, combined with historical data and expert knowledge, the overall safety risk index for each printing press is calculated. Simultaneously, each piece of equipment is identified, and its type, performance parameters, and maintenance history are recorded. For example, if one printing press has a safety risk index of 0.8 and another has 0.2, the equipment identification results show that the former is an older piece of equipment, while the latter is a newer piece.

[0086] Establish intelligent scheduling constraints. Set a risk threshold vector based on the overall safety risk index. For example, set the risk threshold to 0.5. Areas above 0.5 are high-risk areas, and areas below 0.5 are low-risk areas. Set corresponding scheduling restriction strategies for different risk areas. For example, for printing presses in high-risk areas, restrict them from handling high-load tasks or increase the inspection frequency. For printing presses in low-risk areas, restrictions can be relaxed. Construct resource allocation constraints based on equipment identification results. Construct an equipment importance scoring matrix based on equipment importance, performance parameters, and other information. For example, new equipment scores 10, and old equipment scores 5. At the same time, formulate a backup equipment allocation strategy, such as how to quickly allocate backup equipment to ensure production continuity when an important piece of equipment fails. Combine the scheduling restriction strategies and resource allocation constraints to form intelligent scheduling constraints for the printing press group. For example, a high-risk old piece of equipment will be restricted from handling high-load tasks, and backup equipment will be prioritized to reduce risk.

[0087] A scheduling decision model is constructed. Based on deep reinforcement learning algorithms, intelligent scheduling constraints are mapped into a state space and an action space. The state space contains real-time operating status information of the printing press group, such as the safety risk index, load status, and operating status of each printing press. The action space defines the set of executable scheduling operations, such as assigning a task to a printing press or starting / stopping a printing press. A composite reward function is constructed, using safety threshold constraints, resource allocation constraints, and production efficiency constraints as evaluation indicators. For example, if the scheduling scheme meets the safety threshold constraint, a positive reward is given; if the resource allocation constraint is violated, a negative reward is given; and if production efficiency is improved, a positive reward is given. The scheduling strategy is continuously optimized through deep reinforcement learning algorithms to maximize the composite reward function. For example, through training, it learns to prioritize assigning high-load tasks to low-risk new equipment, thereby improving production efficiency while ensuring safety.

[0088] The trained scheduling decision model is then applied to actual production. Based on the real-time operating status of the printing press group, the model automatically generates the optimal scheduling plan and sends it to the execution layer.

[0089] The solution proposed in this application can:

[0090] Enhancing Safety: By conducting real-time risk assessments of the printing press fleet and combining these assessments with equipment identification results to formulate scheduling strategies, safety risks during production can be effectively reduced, preventing accidents. For example, high-load tasks can be avoided by assigning them to high-risk, older equipment, thus reducing the probability of equipment failure. Optimizing Resource Allocation: By constructing an equipment importance scoring matrix and a backup equipment allocation strategy, resource allocation can be optimized, equipment utilization can be improved, and resource waste can be avoided. For example, tasks can be prioritized for allocation to better-performing and more reliable equipment, and backup equipment can be deployed promptly to cope with emergencies. Improving Production Efficiency: By continuously optimizing scheduling strategies through deep reinforcement learning algorithms, production efficiency can be maximized while meeting safety and resource constraints. For example, through learning, the optimal task allocation scheme can be found, thereby reducing production time and increasing output.

[0091] In one optional implementation, the scheduling decision model is used to generate an optimal scheduling scheme, which includes equipment start-up and shutdown sequence and load distribution strategy; the optimal scheduling scheme is then distributed to each printing press control system for execution, including:

[0092] The optimal scheduling scheme is generated based on the scheduling decision model. A device start-stop state transition matrix is ​​constructed using a dynamic programming algorithm. The load allocation strategy is optimized by combining a mixed integer programming method. The load allocation strategy is based on the remaining capacity of the equipment and the urgency of the task, and a task allocation weight coefficient is established to achieve dynamic balance of the load of the equipment group. The device start-stop state transition matrix and the load allocation strategy are combined to form the optimal scheduling scheme.

[0093] A hierarchical scheduling instruction distribution system is constructed, which parses the optimal scheduling scheme into equipment-level control instructions; a real-time communication mechanism is established between the central scheduler and the local control system of the printing press to send the equipment-level control instructions to each printing press control system.

[0094] In the data acquisition phase, historical operating data of the printing press fleet is collected, including equipment operating status (e.g., start-up and shutdown times, operating speed, energy consumption), task information (e.g., order type, quantity, delivery date), environmental parameters (e.g., temperature, humidity), and quality data during production (e.g., printing accuracy, color deviation). For example, the operating data of a printing press over the past week may be collected, including daily start-up and shutdown times, printing speed, energy consumption, number and type of completed orders, and corresponding quality inspection results.

[0095] A hierarchical deep learning model is trained using collected historical data. This model employs a layered structure, extracting data features layer by layer. The bottom-layer networks learn the underlying operational characteristics of the devices, such as the energy consumption patterns of individual devices; the middle-layer networks learn task characteristics, such as the resource demand patterns of different order types; and the top-layer networks fuse the features extracted from the bottom and middle layers, learning the complex relationships between devices and tasks. For example, the deep learning model can predict the production time and resource consumption of different order types on different printing presses. Suppose an order requires printing 10,000 color brochures; the model predicts that it will take 8 hours and consume 100 kWh of electricity on printing press A, and 10 hours and consume 120 kWh of electricity on printing press B.

[0096] A scheduling decision model is constructed. This model uses a dynamic programming algorithm to build the equipment start-up and shutdown state transition matrix. For example, based on historical data and prediction results, it can be calculated that printing machine A needs 30 minutes to start from a shutdown state and reach normal operation, while printing machine B needs 45 minutes. A mixed-integer programming method is combined to optimize the load allocation strategy. This strategy establishes task allocation weight coefficients based on the remaining capacity of the equipment and the urgency of the task, achieving dynamic balance of the equipment group's load. For example, an urgent order that needs to be completed as soon as possible is assigned a higher weight coefficient and prioritized for production. Assuming that printing machine A has 20 hours of remaining capacity and printing machine B has 15 hours of remaining capacity, the urgent order is allocated to printing machine A based on the weight coefficients and the predicted production time. The equipment start-up and shutdown state transition matrix and the load allocation strategy are combined to form an optimal scheduling scheme, which includes the equipment start-up and shutdown sequence and the load allocation strategy.

[0097] The optimal scheduling scheme is distributed to the control systems of each printing press for execution. A hierarchical scheduling instruction distribution system is constructed to parse the optimal scheduling scheme into equipment-level control instructions. For example, the instruction "Start Printing Press A" is parsed into specific control signals and sent to the control system of Printing Press A. A real-time communication mechanism is established between the central scheduler and the local control systems of the printing presses to distribute equipment-level control instructions to the control systems of each printing press. An instruction execution confirmation mechanism and an emergency response plan are designed to ensure the reliable execution of scheduling instructions. For example, if Printing Press A fails to start successfully after receiving the start instruction, the system will issue an alarm and activate the emergency plan to reassign the task to other available printing presses.

[0098] A multi-dimensional monitoring system is deployed to collect execution performance data. This system records equipment operating status, task execution progress, and product quality data in real time. For example, it monitors the operating speed, temperature, energy consumption, current order completion progress, and quality inspection results of printing press A in real time. Execution performance data is preprocessed using multi-source data fusion methods to establish a data anomaly detection and compensation mechanism. For example, if the printing speed of printing press A is detected to be lower than expected, the system will automatically adjust its operating parameters or issue an alarm to notify maintenance personnel for repair. The preprocessed execution performance data is fed back to the data acquisition stage for continuous optimization of the performance of the hierarchical deep learning model and the scheduling decision model. The feature extraction parameters and network structure weights of the hierarchical deep learning model are updated through incremental learning methods; the state value assessment and policy network parameters of the scheduling decision model are optimized based on online learning strategies, achieving continuous improvement in model performance and dynamic accumulation of knowledge.

[0099] The solution proposed in this application can:

[0100] Improve production efficiency: By optimizing equipment start-up and shutdown sequences and load distribution strategies, idle time and task waiting time can be minimized, thereby increasing production efficiency. For example, intelligent scheduling can reduce daily printing task completion time from 10 hours to 8 hours. Reduce production costs: By optimizing equipment start-up and shutdown strategies, energy consumption can be reduced; by optimizing load distribution strategies, equipment utilization can be improved, thereby reducing production costs. For example, intelligent scheduling can reduce daily energy consumption from 1000 kWh to 800 kWh. Improve product quality: Through real-time monitoring and data analysis, anomalies in the production process can be detected and corrected promptly, ensuring product quality stability. For example, intelligent scheduling can increase the product qualification rate from 95% to 98%.

[0101] In one optional implementation, an optimal scheduling scheme is generated based on the scheduling decision model, and a device start-stop state transition matrix is ​​constructed using a dynamic programming algorithm; a load allocation strategy is optimized by combining a mixed integer programming method, wherein the load allocation strategy establishes task allocation weight coefficients based on the remaining capacity of the equipment and the urgency of the tasks to achieve dynamic balance of the load of the equipment group; combining the device start-stop state transition matrix and the load allocation strategy to form the optimal scheduling scheme includes:

[0102] The equipment start-stop state transition matrix includes the state transition rules, state transition costs, and preheating time constraints of the printing press; the state value function and the state action value function are calculated using a value iteration algorithm, and the optimal start-stop strategy is determined based on the state value function and the state action value function.

[0103] The remaining capacity of the printing press is calculated based on its operating status, and the urgency of the tasks is assessed based on their delivery time. A task allocation weighting coefficient is established using the remaining capacity and the urgency of the tasks, and this coefficient is used to prioritize tasks and allocate resources.

[0104] Based on the task allocation weight coefficients, load balancing constraints are constructed, including equipment capacity limits, task timing relationships, and process parameter requirements. By solving a mixed integer programming model, a task allocation scheme that satisfies the load balancing constraints is obtained, thereby achieving dynamic balance of the printing press group's load.

[0105] Construct a state transition matrix for the printing press. This matrix describes the state changes of the printing press over different time periods, such as power-on, power-off, standby, and preheating. Each element in the matrix represents the cost of transitioning from one state to another, such as energy consumption and time cost. Furthermore, the matrix considers the preheating time constraint of the printing press; for example, a certain preheating time is required to transition from power-off to power-on. For instance, suppose there are three printing presses, with states of power-off, standby, and running. A 3x3xT matrix can be constructed, where T represents the total number of time periods within the scheduling cycle. Each element in the matrix represents the cost of a printing press transitioning from one state to another within a specific time period. For example, the cost for the first printing press to transition from power-off to running in the first time period is 10 (including preheating energy consumption and time cost), while the cost to transition from standby to running is 5 (including only time cost).

[0106] The state-value function and state-action-value function are calculated using a value iteration algorithm. The state-value function represents the long-term value of the printing press in a certain state over a given time period. The state-action-value function represents the long-term value of the printing press after performing a certain action (e.g., turning on or off) over a given time period. Through iterative calculation, the optimal value for each state and action can be obtained. For example, suppose the first printing press is in the off state in the first time period, its state-value function is 50, representing the expected future revenue. If it is turned on, the state-action-value function is 60, indicating that the expected future revenue is higher after turning it on, so turning it on should be chosen.

[0107] Based on the calculated state-value function and state-action-value function, the optimal start-up and shutdown strategy can be determined. This strategy specifies the state in which each printing press should be in each time period to maximize long-term value. For example, based on previous calculations, the first printing press should be turned on in the first time period.

[0108] Optimize load allocation strategies. First, calculate the remaining capacity of the equipment based on the printing press's operating status. For example, if the first printing press can print 1000 copies per hour when it is running, and 500 copies have already been scheduled for printing, then the remaining capacity is 500 copies. Second, assess the urgency of the tasks based on their delivery times. For example, a printing task that needs to be completed within 2 hours is more urgent than a task that needs to be completed within 24 hours.

[0109] A task allocation weighting coefficient is established using the remaining capacity of the equipment and the urgency of the task. This coefficient is used to prioritize tasks and allocate resources. For example, a task with high urgency and large remaining printing press capacity will receive a higher weighting coefficient. Assuming a task has an urgency score of 8 (out of 10) and the target printing press has a remaining capacity of 600 copies, the task allocation weighting coefficient can be calculated as 8 * 600 = 4800.

[0110] Load balancing constraints are constructed based on task allocation weight coefficients. These constraints include equipment capacity limitations, task timing relationships, and process parameter requirements. For example, a printing press cannot be assigned tasks exceeding its capacity within a given time period, certain tasks have a sequential order, and different types of printing tasks require different process parameters.

[0111] By analyzing the load balancing constraints, a task allocation scheme that satisfies all constraints is obtained, achieving dynamic load balancing for the printing press fleet. For example, tasks are prioritized for allocation to printing presses with large remaining capacity and that meet the task timing and process parameter requirements based on weighting coefficients.

[0112] A timing coordination mechanism is constructed by combining and matching the equipment start / stop state transition matrix and the task allocation scheme. This mechanism ensures the consistency between the equipment start / stop timing and the task allocation timing. For example, if a task is assigned to a printing press, the printing press must be turned on before the task starts and, after the task is completed, choose to standby or shut down according to the subsequent task schedule. An optimal scheduling scheme is generated based on the timing coordination results. This scheme specifies the start / stop time of each printing press and the execution time of each task, as well as the assigned printing press.

[0113] The solution proposed in this application can:

[0114] Improve production efficiency: By optimizing task allocation and equipment start-up and shutdown, maximize equipment capacity and reduce idle time, thereby improving overall production efficiency. Reduce energy consumption: By intelligently controlling equipment start-up and shutdown, avoid unnecessary energy waste and reduce production costs. Achieve load balancing: Through task allocation weighting coefficients and load balancing constraints, ensure a more balanced load on the printing press fleet, preventing some equipment from being overloaded while others are idle, and extending equipment lifespan.

[0115] Figure 2 This is a schematic diagram of the structure of a safety production control system for printing equipment according to an embodiment of the present invention, as shown below. Figure 2 As shown, the system includes:

[0116] The first unit is used to collect operational data in real time through a multimodal sensor network deployed in the printing press cluster, extract features from the standardized data stream corresponding to the operational data using a temporal attention mechanism to obtain the dynamic feature vector of the printing press cluster, and construct the spatiotemporal correlation matrix of the printing press cluster based on the dynamic feature vector.

[0117] The second unit is used to construct a hierarchical deep learning model based on the dynamic feature vector and the spatiotemporal correlation matrix, input historical labeled data into the hierarchical deep learning model for training, use the trained deep learning model to analyze the dynamic feature vector, and generate equipment health status assessment results; based on the equipment health status assessment results and the spatiotemporal correlation matrix, calculate the overall safety risk index and equipment identification results of the printing press group.

[0118] The third unit is used to establish intelligent scheduling constraints for the printing press group based on the overall safety risk index and the equipment identification results; construct a scheduling decision model based on a deep reinforcement learning algorithm, and use safety threshold constraints, resource allocation constraints, and production efficiency constraints as evaluation indicators of the reward function; generate an optimal scheduling scheme using the scheduling decision model, the optimal scheduling scheme including equipment start-up and shutdown sequence and load allocation strategy; and distribute the optimal scheduling scheme to each printing press control system for execution.

[0119] A third aspect of the embodiments of the present invention,

[0120] An electronic device is provided, comprising:

[0121] processor;

[0122] Memory used to store processor-executable instructions;

[0123] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0124] Fourth aspect of the present invention,

[0125] A computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0126] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0127] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for safe production control of printing press equipment, characterized in that, include: The printing press cluster collects operational data in real time through a multimodal sensor network deployed in the printing press cluster, and uses a temporal attention mechanism to extract features from the standardized data stream corresponding to the operational data to obtain the dynamic feature vector of the printing press cluster. The spatiotemporal correlation matrix of the printing press group is constructed based on the dynamic feature vectors; Based on the dynamic feature vector and the spatiotemporal correlation matrix, the hierarchical deep learning model is constructed as follows: The hierarchical deep learning model includes a feature fusion layer, a temporal prediction layer, and a risk assessment layer. In the feature fusion layer, the dynamic feature vector is converted into a query matrix, a key matrix, and a value matrix through a self-attention mechanism, and the inter-feature association weights are calculated through scaled dot product attention. The spatiotemporal association matrix is ​​input into a graph attention network to extract the spatial dependency features between printing presses. The inter-feature association weights and the spatial dependency features are adaptively fused through a gating mechanism to generate a first comprehensive feature representation. In the temporal prediction layer, the first comprehensive feature representation is input into a bidirectional long short-term memory network and a multi-scale parallel convolutional neural network, respectively. The bidirectional long short-term memory network extracts temporal features, and the multi-scale parallel convolutional neural network extracts local features. The output features of the two networks are fused through a residual connection structure to generate a second comprehensive feature representation. The second comprehensive feature representation is input into the risk assessment layer to generate a second prediction result. Historical labeled data is input into the hierarchical deep learning model for training. The trained deep learning model is used to analyze the dynamic feature vector to generate an equipment health status assessment result. Based on the equipment health status assessment result and the spatiotemporal correlation matrix, the overall safety risk index and equipment identification result of the printing press group are calculated. Based on the overall safety risk index and the equipment identification results, intelligent scheduling constraints for the printing press group are established; a scheduling decision model is constructed based on a deep reinforcement learning algorithm, using safety threshold constraints, resource allocation constraints, and production efficiency constraints as evaluation indicators for the reward function; the optimal scheduling scheme is generated using the scheduling decision model, including: A dynamic programming algorithm is used to construct the equipment start-stop state transition matrix; a mixed integer programming method is used to optimize the load allocation strategy, which is based on the remaining capacity of the equipment and the urgency of the task, and establishes a task allocation weight coefficient to achieve dynamic balance of the load of the equipment group; the equipment start-stop state transition matrix and the load allocation strategy are combined to form the optimal scheduling scheme; the optimal scheduling scheme is then sent to each printing press control system for execution.

2. The method according to claim 1, characterized in that, Based on the dynamic feature vector and the spatiotemporal correlation matrix, a hierarchical deep learning model is constructed. Training the hierarchical deep learning model by inputting historical labeled data includes: The hierarchical deep learning model includes a feature fusion layer, a temporal prediction layer, and a risk assessment layer. In the feature fusion layer, the dynamic feature vector is converted into a query matrix, a key matrix, and a value matrix through a self-attention mechanism, and the inter-feature association weights are calculated through scaled dot product attention. The spatiotemporal association matrix is ​​input into a graph attention network to extract the spatial dependency features between printing presses. The inter-feature association weights and the spatial dependency features are adaptively fused through a gating mechanism to generate a first comprehensive feature representation. In the temporal prediction layer, the first comprehensive feature representation is input into a bidirectional long short-term memory network and a multi-scale parallel convolutional neural network, respectively. The bidirectional long short-term memory network extracts temporal features, and the multi-scale parallel convolutional neural network extracts local features. The output features of the two are fused through a residual connection structure to generate a second comprehensive feature representation. The second comprehensive feature representation is input into the risk assessment layer. This layer constructs a main branch and a secondary branch based on a multi-task learning framework. The main branch uses a multilayer perceptron to predict the health index of the device and generates a first prediction result. The secondary branch uses a conditional random field model to model the risk propagation mechanism between devices and generates a second prediction result. Historical labeled data is input into the hierarchical deep learning model for training. The first loss value of the first prediction result is calculated using the mean squared error loss function, and the second loss value of the second prediction result is calculated using the cross-entropy loss function. The first loss value and the second loss value are adaptively balanced using learnable weights to obtain the total loss value. The parameters of the hierarchical deep learning model are optimized based on the total loss value to obtain the trained deep learning model.

3. The method according to claim 1, characterized in that, The trained deep learning model is used to analyze the dynamic feature vector to generate equipment health status assessment results; Based on the equipment health status assessment results and the spatiotemporal correlation matrix, the overall safety risk index and equipment identification results of the printing press group are calculated, including: The trained deep learning model is used to analyze the dynamic feature vector, and multiple prediction results are fused through weighted voting to generate a device health status assessment result. Based on the equipment health status assessment results and the spatiotemporal correlation matrix, a weighted directed graph is constructed. The nodes of the weighted directed graph represent printing press equipment, and the edge weights represent the risk propagation intensity between equipment. The PageRank algorithm is used to calculate the node importance score. The node importance score is combined with a local risk accumulation function, which incorporates a time-dimensional decay factor and a spatial-dimensional graph convolution feature, to calculate the overall safety risk index of the printing press group. Based on the weighted directed graph, multi-criteria decision analysis is performed, and equipment is identified through fuzzy comprehensive evaluation method to generate equipment identification results.

4. The method according to claim 1, characterized in that, Based on the overall safety risk index and the equipment identification results, intelligent scheduling constraints for the printing press group are established. A scheduling decision model is constructed based on deep reinforcement learning algorithms, using safety threshold constraints, resource allocation constraints, and production efficiency constraints as evaluation metrics for the reward function. A risk threshold vector is set according to the overall security risk index. The risk threshold vector is used to divide risk areas that are higher than the preset risk threshold and risk areas that are lower than the preset risk threshold. Corresponding scheduling restriction strategies are set for different risk areas. Based on the equipment identification results, resource allocation constraints are constructed. These constraints include an equipment importance scoring matrix and a backup equipment allocation strategy. The equipment importance scoring matrix is ​​used to quantify the importance of equipment in the production system. The scheduling restriction strategy and the resource allocation constraints are combined to form intelligent scheduling constraints for the printing press group. A scheduling decision model is constructed based on a deep reinforcement learning algorithm, mapping the intelligent scheduling constraints into a state space and an action space. The state space contains the real-time operating status information of the printing press group, and the action space defines the set of executable scheduling operations. A composite reward function is constructed, using safety threshold constraints, resource allocation constraints, and production efficiency constraints as evaluation indicators for the composite reward function.

5. The method according to claim 1, characterized in that, The optimal scheduling scheme is generated using the aforementioned scheduling decision model, and the optimal scheduling scheme includes the device start-up and shutdown sequence and the load distribution strategy. The optimal scheduling scheme is distributed to each printing press control system for execution, including: The optimal scheduling scheme is generated based on the scheduling decision model. A device start-stop state transition matrix is ​​constructed using a dynamic programming algorithm. The load allocation strategy is optimized by combining a mixed integer programming method. The load allocation strategy is based on the remaining capacity of the equipment and the urgency of the task, and a task allocation weight coefficient is established to achieve dynamic balance of the load of the equipment group. The device start-stop state transition matrix and the load allocation strategy are combined to form the optimal scheduling scheme. A hierarchical scheduling instruction distribution system is constructed, which parses the optimal scheduling scheme into equipment-level control instructions; a real-time communication mechanism is established between the central scheduler and the local control system of the printing press to send the equipment-level control instructions to each printing press control system.

6. The method according to claim 5, characterized in that, The optimal scheduling scheme is generated based on the scheduling decision model, and the equipment start-stop state transition matrix is ​​constructed through dynamic programming algorithm. The load allocation strategy is optimized by combining mixed integer programming method. The load allocation strategy is based on the remaining capacity of the equipment and the urgency of the task, and a task allocation weight coefficient is established to achieve dynamic balance of the load of the equipment group. Combining the device start / stop state transition matrix and the load distribution strategy to form the optimal scheduling scheme includes: The equipment start-stop state transition matrix includes the state transition rules, state transition costs, and preheating time constraints of the printing press; the state value function and the state action value function are calculated using a value iteration algorithm, and the optimal start-stop strategy is determined based on the state value function and the state action value function. The remaining capacity of the printing press is calculated based on its operating status, and the urgency of the tasks is assessed based on their delivery time. A task allocation weighting coefficient is established using the remaining capacity and the urgency of the tasks, and this coefficient is used to prioritize tasks and allocate resources. Based on the task allocation weight coefficients, load balancing constraints are constructed, including equipment capacity limits, task timing relationships, and process parameter requirements. By solving a mixed integer programming model, a task allocation scheme that satisfies the load balancing constraints is obtained, thereby achieving dynamic balance of the printing press group's load.

7. A safety production control system for printing press equipment, used to implement the method according to any one of claims 1-6, characterized in that, include: The first unit is used to collect operational data in real time through a multimodal sensor network deployed in the printing press cluster, extract features from the standardized data stream corresponding to the operational data using a temporal attention mechanism to obtain the dynamic feature vector of the printing press cluster, and construct the spatiotemporal correlation matrix of the printing press cluster based on the dynamic feature vector. The second unit is used to construct a hierarchical deep learning model based on the dynamic feature vector and the spatiotemporal correlation matrix, input historical labeled data into the hierarchical deep learning model for training, and use the trained deep learning model to analyze the dynamic feature vector to generate equipment health status assessment results. Based on the equipment health status assessment results and the spatiotemporal correlation matrix, the overall safety risk index and equipment identification results of the printing press group are calculated. The third unit is used to establish intelligent scheduling constraints for the printing press group based on the overall security risk index and the equipment identification results. A scheduling decision model is constructed based on a deep reinforcement learning algorithm, using safety threshold constraints, resource allocation constraints, and production efficiency constraints as evaluation indicators for the reward function. The optimal scheduling scheme is generated using the scheduling decision model, which includes the equipment start-up and shutdown sequence and load distribution strategy. The optimal scheduling scheme is then distributed to the control systems of each printing press for execution.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images