Digital production plan scheduling method and system

Abstract

The invention discloses a digital production plan scheduling method and system, and belongs to the technical field of optimal scheduling, and the method comprises the steps: constructing a distributed storage architecture based on edge computing nodes; a central coordinator is adopted to realize cross-node data synchronization through an improved Raft consensus algorithm, multi-version concurrency control is realized based on a vector clock, and a global consistent data view is established; a visual scheduling platform is built based on a Vue3 framework, and man-machine interaction is realized by adopting a Canvas and WebGL collaborative rendering framework; establishing a dynamic coordinate conversion model based on bilinear interpolation, designing a space mapping function containing distortion compensation, establishing a multi-thread coordinate service based on WebWorker, and realizing submillimeter-level bidirectional mapping of pixel coordinates and physical coordinates; constructing a three-dimensional space-time analysis model fused with the multi-dimensional features; and all the units are subjected to feature fusion through residual connection, and finally a scheduling scheme with a confidence coefficient weight is output. The method and the device have the effect of meeting various scheduling requirements.

Description

Technical Field

[0001] The present application relates to the technical field of optimized scheduling, and in particular to a digital production planning and scheduling method and system. Background Art

[0002] In the modern equipment manufacturing industry, the production process involves numerous steps. Due to the nature of the production process, some larger components cannot be assembled immediately after production, requiring temporary storage at the assembly site. With the increasing diversity and complexity of equipment products, when a large number of components require temporary storage, the rational planning and scheduling of assembly sites becomes crucial.

[0003] Currently, the manufacturing industry is rapidly developing towards digitalization and intelligentization. Digital production planning and scheduling methods are playing an increasingly important role in improving production efficiency, reducing costs, and optimizing resource allocation. Within the production system of equipment companies, the final assembly site is a key area for final product integration. The rational placement of components directly impacts final assembly efficiency, quality, and site utilization. Traditional component placement planning in final assembly sites often relies on manual experience and lacks scientific methods and tools, making it difficult to adapt to the high-variety, small-batch, and highly complex production requirements. Summary of the Invention

[0004] In order to demonstrate strong adaptability and optimization capabilities in complex production scenarios, the present application provides a digital production planning and scheduling method and system.

[0005] The present application provides a digital production planning and scheduling method and system that adopts the following technical solutions:

[0006] In a first aspect, the present application provides a digital production planning and scheduling method, comprising the following steps:

[0007] Build a distributed storage architecture based on edge computing nodes;

[0008] A central coordinator is used to synchronize data across nodes using the improved Raft consensus algorithm. Multi-version concurrency control is implemented based on vector clocks to establish a globally consistent data view.

[0009] Build a visual scheduling platform based on the Vue3 framework, and use Canvas and WebGL collaborative rendering architecture to achieve human-computer interaction;

[0010] A dynamic coordinate transformation model based on bilinear interpolation was established, a spatial mapping function with distortion compensation was designed, and a multi-threaded coordinate service was established based on WebWorker to achieve sub-millimeter bidirectional mapping between pixel coordinates and physical coordinates.

[0011] Construct a three-dimensional spatiotemporal analysis model that integrates multi-dimensional features;

[0012] Each unit performs feature fusion through residual connection and finally outputs a scheduling plan with confidence weights.

[0013] Furthermore, it also includes:

[0014] A multi-dimensional spatiotemporal data engine is built based on PostGIS, using a four-dimensional index structure to store the spatiotemporal topological relationship of the site vector base map, the LOD metadata of the component model, and the timeline layout snapshot with version identification;

[0015] Design a spatiotemporal version controller based on a hybrid index of B+ trees and R trees, supporting incremental storage and merging of multiple timeline branches, and achieving second-level historical state backtracking through version hash chains;

[0016] Deploy an adaptive projection engine, integrate the improved Lambert projection algorithm in GeoServer, and design a dynamic interpolation function:

[0017] Designed an OT conflict resolution algorithm based on time and space priorities, deployed layered WebSocket services, and used Epoll+Redis to achieve 10,000-level concurrent operation synchronization;

[0018] Develop a coordinate coupling module for Mapbox GL and Three.js, and achieve dynamic alignment of 2D / 3D coordinate systems through affine transformation matrices:

[0019] Design a multi-granularity space-time lens model to support time scale switching;

[0020] Develop a time mask renderer based on Compute Shader to achieve progressive visualization of the time dimension through spatiotemporal weight functions.

[0021] Furthermore, it also includes:

[0022] Build a cross-site spatiotemporal federated learning architecture, deploying a Bi-GRU prediction model with a spatiotemporal attention mechanism on each edge node;

[0023] Design a model traceability mechanism to track parameter contributions through gradient fingerprint mapping;

[0024] Develop a clock-driven Delaunay triangulation algorithm, combined with an LSTM buffer capacity prediction model, to generate a topological structure with spatiotemporal constraints. Introduce a spatiotemporal conflict detection matrix to verify the feasibility of layout solutions in real time.

[0025] Build an improved NSGA-III optimization engine and design a three-stage optimization process;

[0026] Integrated constraint processing mechanism, embedding the site spatiotemporal topology as a hard constraint into the optimization process;

[0027] A residual correction module is developed to dynamically adjust the prediction output through the error feedback matrix.

[0028] Furthermore, it also includes:

[0029] Construct a bidirectional conversion channel between EPSG:3857 and the local coordinate system, and design an affine transformation model with distortion compensation;

[0030] Developed a deep learning-based EXIF ​​metadata parser to extract image physical features using a convolutional attention network;

[0031] Developed a layer management system based on CSS Transform Module Level 1, using a versioned differential storage strategy to generate incremental snapshots for each operation;

[0032] Implement non-destructive editing stack and support backtracking to any historical state through snapshot hash chain.

[0033] Furthermore, it also includes:

[0034] In the intelligent decision-making evaluation layer, a dynamic interval-type TOPSIS multi-attribute decision-making model is constructed, and the weight coefficients of the five-dimensional evaluation indicators are dynamically adjusted through the entropy weight-hierarchy analysis fusion algorithm;

[0035] Introducing resource bottleneck index calculation function Where i represents the index variable in the summation symbol; w i represents the weight of the i-th resource factor, all w i The sum is 1; N i Represents the quantitative index value of the i-th resource factor;

[0036] The quantitative evaluation results are passed as input parameters to a two-layer optimization engine. The optimization solution is verified using a Monte Carlo-digital twin hybrid simulator. A capacity fluctuation function is constructed to simulate supply chain disturbance scenarios. A multi-objective particle swarm optimization algorithm is integrated to design a Pareto frontier evolutionary strategy.

[0037] The optimized plan is input into the real-time scheduling execution layer, and the time slots are divided into 15-minute granularity based on the time wheel algorithm. A mixed integer programming model is constructed to generate minute-level work order sequences.

[0038] Furthermore, it also includes:

[0039] Adopting a three-channel input architecture, a multimodal fusion LSTM prediction network is constructed;

[0040] By integrating multi-source information through bidirectional gated recurrent units and attention mechanisms, a probability output function is designed to predict the probability of equipment failure.

[0041] The prediction results are fed into the reinforcement learning optimization engine in real time to design a composite reward function and develop an adaptive exploration strategy;

[0042] Build a blockchain collaboration platform based on Hyperledger Fabric and innovatively design a lightweight PBFT consensus mechanism;

[0043] Deploy NVIDIA Jetson clusters at the edge computing layer to build distributed intelligent nodes. Use the TensorRT engine to optimize production scheduling models. Improve model inference speed through layer fusion and precision calibration technology.

[0044] Developed a model distillation pipeline to compress large LSTM networks trained in the cloud into a lightweight version suitable for edge deployment, maintaining prediction accuracy while reducing computational energy consumption;

[0045] Design an edge-cloud collaborative scheduler to dynamically allocate computing tasks based on queuing theory, and prioritize critical path tasks for processing at edge nodes.

[0046] Furthermore, it also includes:

[0047] A spatiotemporal context-aware RBAC model is constructed, defining a dynamic permission matrix P = [R × (L ⊕ T)], where the role set R includes process engineers and planners, the spatial constraint L generates a geofence based on the physical location of the equipment, and the temporal constraint T uses a sliding time window to limit sensitive operation periods.

[0048] Design an access control system to generate dynamic confidence scores by collecting user behavior characteristics Where i is the index variable in the summation symbol, which is used to traverse from 1 to n different user behavior feature-related factors; n represents the total number of user behavior feature factors involved in calculating the dynamic confidence score; w i is the weight of the i-th user behavior feature factor; f i (x) is a function of the i-th user behavior characteristic factor, and x is an input variable related to the user behavior characteristic; when C < 0.85, multi-factor authentication is triggered to intercept unauthorized operations;

[0049] The permission control system is deeply integrated with the operation log module. The log recording unit adopts an improved CRDT data structure, an operation semantic analyzer is designed to automatically parse instruction types, and the operation timing relationship is marked by a vector clock.

[0050] The collaborative perception engine is enhanced based on the WebRTC architecture. The collaborative perception engine is linked to the permission control system. When unauthorized operations are detected, data desensitization is automatically triggered, and sensitive fields are transmitted using homomorphic encryption.

[0051] In a second aspect, the present application provides a digital production planning and scheduling system, comprising:

[0052] Storage architecture building module, used to build a distributed storage architecture based on edge computing nodes;

[0053] The data synchronization module is used to synchronize data across nodes using a central coordinator and an improved Raft consensus algorithm. It also implements multi-version concurrency control based on vector clocks and establishes a globally consistent data view.

[0054] The human-computer interaction module is used to build a visual scheduling platform based on the Vue3 framework, and uses the Canvas and WebGL collaborative rendering architecture to achieve human-computer interaction;

[0055] The dynamic coordinate conversion module is used to establish a dynamic coordinate conversion model based on bilinear interpolation, design a spatial mapping function including distortion compensation, and establish a multi-threaded coordinate service based on WebWorker to achieve sub-millimeter bidirectional mapping between pixel coordinates and physical coordinates;

[0056] 3D spatiotemporal analysis module, used to build a 3D spatiotemporal analysis model integrating multi-dimensional features;

[0057] The feature fusion module is used to fuse features of each unit through residual connections, and finally outputs a scheduling plan with confidence weights.

[0058] In a third aspect, the present application provides an intelligent terminal comprising a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and execute the above-mentioned digital production planning and scheduling method.

[0059] In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program that can be loaded by a processor and execute the above-mentioned digital production planning and scheduling method.

[0060] In summary, compared with the prior art, the above technical solution has the following beneficial effects:

[0061] The digital production planning and scheduling method and system described in this application builds a distributed storage architecture based on edge computing nodes, which can realize local storage and rapid processing of data, reduce data transmission delays and network congestion. The improved Raft consensus algorithm and vector clock mechanism ensure cross-node data synchronization and multi-version concurrent control, form a globally consistent data view, and improve data accuracy and reliability. The Vue3 framework combines the visual scheduling platform built with Canvas and WebGL to provide a good human-computer interaction experience, which is convenient for operators to intuitively monitor and schedule production. Sub-millimeter bidirectional coordinate mapping improves coordinate conversion accuracy to meet the needs of high-precision positioning in production. The three-dimensional spatiotemporal analysis model that integrates multi-dimensional features, combined with residual connection feature fusion, can comprehensively consider various production factors, output scheduling plans with confidence weights, improve the scientificity and rationality of scheduling, and help enterprises optimize production processes, improve production efficiency, and reduce costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a flow chart of a digital production planning and scheduling method according to an embodiment of the present application. DETAILED DESCRIPTION

[0063] The present application is further described in detail below in conjunction with all the accompanying drawings.

[0064] The present application discloses a digital production planning and scheduling method and system. Figure 1 , a digital production planning and scheduling method includes:

[0065] S101. Construction of distributed storage architecture.

[0066] Specifically, the system deploys edge computing nodes at each production site, and each node is configured with a local database with a hybrid storage structure. Real-time production data (such as sensor readings and equipment status) is stored in a time-series database (such as InfluxDB), which receives data streams through a message queue (such as Kafka). A data retention policy is set to automatically reduce the precision of historical data storage. Device topology relationships (such as hierarchical structures and connection paths) are stored in a graph database (such as Neo4j). A node and relationship model is established, and when physical devices change, the graph structure is updated through the management interface to ensure that the data is consistent with the actual production environment.

[0067] S102: Data synchronization and concurrency control.

[0068] Specifically, the system synchronizes data across nodes using a modified Raft consensus algorithm through a central coordinator cluster. A master node is elected to receive data changes from edge nodes, which are then synchronized to a majority of slave nodes via a write-ahead log before being committed. A vector clock is attached to each data item to record version information. Version numbers are compared during read and write operations, and conflicts are merged or rolled back according to pre-set rules (such as timestamp priority), ensuring a consistent global data view.

[0069] S103. Build a visual scheduling platform.

[0070] Specifically, the system develops a visual scheduling platform based on the Vue3 framework, employing a collaborative rendering architecture of Canvas and WebGL. The Canvas layer implements a quaternion rotation controller, converting user drag operations into quaternion rotations and calculating the rotation matrix. This is applied to the device topology map to achieve gimbal-lock-free 3D operations, while pre-rendering complex structures using off-screen Canvas improves performance. The WebGL layer constructs a scene graph data structure, applying a frustum clipping algorithm to render only visible areas. In combination with instanced rendering technology, it batches similar devices and dynamically switches LOD models based on distance, enabling real-time rendering of millions of production components.

[0071] S104. Implementation of dynamic coordinate conversion model.

[0072] Specifically, the system arranges coordinate control points on the production site, collects the correspondence between physical coordinates and pixel coordinates, and trains a bilinear interpolation model to establish a mapping function. It analyzes the sources of coordinate distortion (such as camera position and perspective effects), constructs a nonlinear distortion compensation model, and corrects deviations in real time during the coordinate conversion process. A multi-threaded coordinate service is established based on WebWorker, assigning compute-intensive tasks to a thread pool. Data is efficiently transferred through shared memory (SharedArrayBuffer), achieving bidirectional mapping between pixel coordinates and physical coordinates with submillimeter accuracy.

[0073] S105. Construction of three-dimensional spatiotemporal analysis model.

[0074] Specifically, the system constructs a three-dimensional spatiotemporal analysis model that integrates multi-dimensional features: first, the local spatiotemporal characteristics of the equipment status are captured by expanding the gated convolutional network, and different expansion rates are set to cover multiple time scales; then, a spatiotemporal graph network based on the multi-head attention mechanism is constructed, the production equipment is abstracted into graph nodes, the process relationships are converted into an adjacency matrix, and the spatiotemporal associations across process nodes are established; finally, an improved DTW algorithm is used to calculate the similarity of production scenarios, and constraints such as equipment capacity and process sequence are added to achieve flexible path matching with constraints.

[0075] S106: Scheduling plan generation and optimization.

[0076] Specifically, the feature vectors output by each analysis unit are fused through residual connections, preserving the original feature information while learning the complex relationships between features. A confidence assessment model is trained, assigning weights to each feature based on factors such as data quality and model stability, and dynamically adjusting the model's reliance on different inputs. The fused features are input into the decision network to generate multiple sets of alternative scheduling plans. Each plan is then feasibility verified and performance evaluated. The optimal plan is selected based on business objectives (e.g., efficiency or cost priority), and a prediction confidence score is provided for decision-making.

[0077] Furthermore, a digital production planning and scheduling method may further include the following steps:

[0078] Integrate the PostGIS extension into the PostgreSQL database to build a multidimensional spatiotemporal data engine. Design a four-dimensional index structure (three dimensions of space and one dimension of time) to store the topological relationships of the site vector basemap by time version, assigning a unique identifier to each spatiotemporal object. Component model LOD metadata is stored in association with spatial location, recording geometric simplification parameters at different levels of detail. Timeline layout snapshots are assigned version identifiers, and indexes are established using timestamps and version numbers, enabling quick queries of layout status at specific points in time. Configure automatic vacuuming and index optimization tasks to ensure efficient data storage and querying.

[0079] Develop a spatiotemporal version controller based on a hybrid index of B+ trees and R-trees. The B+ tree manages timeline branches and version numbers, while the R-tree indexes spatial geometry data. When data changes occur, incremental storage is used to record only the changes, linking previous and next versions via a version hash chain. A branch merge algorithm is designed to handle the integration of changes across multiple timelines. Conflicts are detected during merges and resolved according to pre-set rules (e.g., latest write priority). A historical state retracing function is implemented, quickly locating the data state at a specific point in time via a hash chain and reconstructing a complete view using incremental data.

[0080] An improved Lambert projection algorithm was integrated into GeoServer to optimize projection parameters for the geographic extent of the production site. A dynamic interpolation function was designed to automatically adjust interpolation accuracy based on data density and query scope, improving performance while maintaining accuracy. A projection parameter management interface was configured to support switching projection modes based on site characteristics. A projection conversion service was developed to uniformly convert data from different coordinate systems into an internal storage format and provide a standardized coordinate interface.

[0081] Expand GeoServer's WMS service to add support for the time dimension. Modify the request parser to recognize the timestamp range parameter, allowing data queries to filter data within a specified time period. Develop a spatiotemporal data slice generator to split the spatiotemporal cube into multiple layers based on the time dimension and pre-generate a cache of tiles for commonly used time ranges. Implement dynamic slicing, generating tiles in real time when requesting an uncached time range and caching them for subsequent use. Optimize the tile storage structure, organizing tile files by temporal and spatial indexes to improve access efficiency.

[0082] Design an OT (operation transformation) conflict resolution algorithm based on spatiotemporal priorities, calculating priority based on the operation's timestamp, spatial impact, and business importance. Deploy a layered WebSocket service, using Epoll at the bottom layer for high-performance network I / O and Redis at the top layer to store operation logs and status. When a client operation is received, it is first applied locally and broadcast to other clients, while also being recorded in Redis. Operation synchronization is achieved through Redis's publish-subscribe mechanism. When a conflict is detected, the operation is retained or merged based on priority.

[0083] Develop a coordinate coupling module between Mapbox GL (2D maps) and Three.js (3D scenes). Establish a mapping relationship between the two coordinate systems using an affine transformation matrix, converting geographic coordinates to a uniform world coordinate system. Implement a bidirectional coordinate conversion function that supports calculating the corresponding 3D scene coordinates from a clicked location on the 2D map, as well as projecting the 3D scene location onto the 2D map. Design a dynamic alignment mechanism that automatically adjusts the mapping matrix when the 3D scene is rotated or scaled, maintaining consistency between the 2D and 3D views. Add a coordinate calibration tool that allows users to manually adjust mapping parameters using control points to improve alignment accuracy.

[0084] Integrate Konva.js (2D drawing library) with the WebAssembly acceleration engine to implement vector editing capabilities based on incremental topological analysis. Compute-intensive topological analysis algorithms are compiled into WebAssembly modules to improve processing speed. Design an incremental update mechanism so that when users edit vector graphics, only the affected topological relationships are recalculated, rather than a global reanalysis. Implement layout adjustment capabilities, support dragging, scaling, and other operations on the site layout, and automatically maintain spatial relationships between components. Develop an undo / redo system that records editing operations through command mode and supports multi-step undo and redo.

[0085] Design a multi-granularity spatiotemporal lens model to support flexible switching between different time scales. Implement a time scaling controller to allow users to adjust the time granularity from milliseconds (real-time equipment status) to months (production plan). Develop a time filter to automatically filter and display appropriate data based on the current time scale. Design a spatiotemporal window to highlight data from a specific time period in the view while retaining contextual information. Implement the linkage between the timeline and the spatial view. When the user selects a specific range on the timeline, the spatial view automatically updates to display the corresponding status.

[0086] Develop a time mask renderer based on WebGL Compute Shader. Define a spatiotemporal weight function to calculate rendering weights based on the temporal attributes of the data. Objects closer to the current time have higher weights and clearer displays. Implement progressive visualization of the time dimension, visually displaying the temporal evolution of data through transparency gradients, color mapping, and other methods. Develop a time mask effect, allowing users to select a specific time period and display only the data within that time period or enhance its display effect. Optimize rendering performance and use Compute Shader to parallelize spatiotemporal weight calculations to reduce the burden on the main thread.

[0087] Furthermore, a digital production planning and scheduling method may further include the following steps:

[0088] A spatiotemporal federated learning architecture was constructed, deploying a Bi-GRU prediction model with a spatiotemporal attention mechanism at edge nodes across each production site. A differential privacy mechanism was embedded in the model input layer, and a noise parameter ε ≤ 0.3 was added to protect the privacy of the original data. During model training, the spatiotemporal attention mechanism automatically focused on key spatiotemporal regions, outputting multi-dimensional prediction results such as equipment failure probability distribution, energy consumption fluctuation time series curves, and personnel flow heat maps. A central server aggregated model parameters from each node using a federated learning framework, generating a global model through weighted averaging, which was then distributed to edge nodes for updates. A model compression strategy was designed to reduce the number of transmitted parameters and improve communication efficiency.

[0089] Design a model traceability mechanism and develop a gradient fingerprint system to record the gradient characteristics of each node parameter update during model training. Generate a unique identifier for each gradient update and establish a parameter contribution tracking chain. When model anomalies are detected or node contributions need to be assessed, analyze the impact of each node on the final model by back-tracing the gradient fingerprint. Implement a contribution visualization interface to intuitively display the parameter contribution ratio of different nodes in different time periods, providing a basis for model optimization and resource allocation.

[0090] Dynamic topology generation and verification: A clock-driven Delaunay triangulation algorithm is developed to dynamically generate topology based on the real-time location and status of devices. Incorporating an LSTM buffer capacity prediction model, this allows for proactive prediction of device load changes and adjustments to topology connections. A spatiotemporal conflict detection matrix is ​​introduced to define constraints such as device spacing and process paths, enabling real-time feasibility verification during topology generation. When conflicts are detected, a local replanning algorithm is triggered to adjust connections while satisfying the constraints.

[0091] Implementing an improved NSGA-III optimization engine, we constructed a three-stage optimization process. In the preprocessing phase, we used the Kriging surrogate model to reduce the dimensionality of the high-dimensional design space and screen key parameters to reduce computational complexity. In the core optimization phase, we defined a composite fitness function encompassing metrics such as production efficiency, cost, and energy consumption, and applied the improved NSGA-III algorithm to search for Pareto-optimal solutions. In the postprocessing phase, we used the TOPSIS method to rank the solution set and select the optimal solution based on business requirements. We also designed a parameter sensitivity analysis module to assess the impact of each parameter on the optimization results.

[0092] An integrated constraint processing mechanism converts site spatiotemporal topological information (such as equipment location, process paths, and safety distances) into mathematical constraint expressions and embeds them into the optimization engine. During the optimization process, penalty functions or repair algorithms are used to ensure that the generated solutions meet the hard constraints. A constraint verification module is developed to perform a posteriori checks on the optimization results and automatically trigger local corrections if any constraint violations are found. A dynamic constraint parameter adjustment interface is implemented to support real-time modification of constraints based on production needs.

[0093] Design a multi-scale skip connection architecture consisting of three core modules: the temporal convolution module uses dilated causal convolution to capture dependencies at different time scales through varying dilation rates; the spatial graph network module uses Chebyshev polynomials to approximate graph convolution kernels, efficiently processing spatial relationships between devices; and the cross-modal fusion unit uses a gated attention mechanism to adaptively assign spatiotemporal feature weights based on data characteristics. Skip connections between modules preserve feature information at different scales, improving the model's expressiveness.

[0094] Develop and construct a residual correction module to record the error between the model's predicted values ​​and the actual observed values. Build an error feedback matrix and analyze the distribution of errors in time and space. When the prediction error exceeds a threshold, dynamically adjust the model output, generating the final result through a weighted combination of the original predicted value and the corrected value. Design an adaptive learning rate mechanism to adjust the correction strength based on the error trend to avoid overcorrection or undercorrection. Regularly update the error feedback matrix to ensure the timeliness of the correction module.

[0095] Furthermore, a digital production planning and scheduling method may further include the following steps:

[0096] Construct a bidirectional coordinate conversion channel to implement bidirectional conversion between EPSG:3857 (Web Mercator) and local coordinate systems within the system. Collect the two coordinate values ​​of site control points and train an affine transformation model with distortion compensation. This model can correct for coordinate distortion caused by measurement errors or map projections. Implement a conversion algorithm library, provide high-precision coordinate conversion functions, and support batch conversion operations. Develop a parameter calibration tool that allows users to automatically calculate conversion parameters by inputting control point coordinates, ensuring conversion accuracy.

[0097] Develop a deep learning metadata parser, including an EXIF ​​metadata parser based on a convolutional attention network. Train the model to identify physical features in images (e.g., scale, reference object size). Design a sub-pixel DPI calibration algorithm that calculates the true DPI value by analyzing the pixel ratio of objects of known size in the image, achieving a calibration accuracy of ±0.3px. Build an image preprocessing pipeline to perform noise reduction and enhancement on the input image to improve feature extraction accuracy. Develop a calibration result verification module to verify calibration accuracy by comparing against known standard objects.

[0098] Build a hybrid B+-tree and R-tree index structure: Use an R-tree index in the spatial dimension to partition geographic features by spatial range, accelerating regional queries. Use a skip table in the temporal dimension to achieve O(log n) time slice retrieval. Design a spatiotemporal data storage format that associates spatial geometry with timestamps. Develop a composite query interface to support efficient retrieval based on both spatial and temporal ranges. Implement an automatic index maintenance mechanism to periodically restructure the index structure to maintain query performance.

[0099] Developed a layer management system based on CSS Transform Module Level 1, enabling layer transformations such as translation, rotation, and scaling. A versioned differential storage strategy was adopted, with each layer operation only recording the differences from the previous version, generating incremental snapshots. A layer state machine was designed to manage layer properties such as visibility and transparency. A layer composition tool was developed to combine multiple layers into composite layers for unified management and operation. A layer style editor was implemented, allowing users to customize layer display styles.

[0100] Build a non-destructive editing stack, generating a unique snapshot hash chain for each operation. When a user performs an edit, the system saves the state before the operation and generates a hash value, recording the operation as an incremental change. Implement a historical state retracing function, traversing the hash chain to locate a specific historical state, and reconstructing the complete view using incremental changes. Design an undo / redo manager to support multi-step undo and redo operations. Add branch management capabilities, allowing users to create parallel editing branches and later merge changes from different branches.

[0101] Develop a geometry compression pipeline, integrating the Draco compression algorithm to compress 3D model data. Design preprocessing steps to simplify and resample the model to reduce data size. Implement an adaptive compression parameter adjustment mechanism to select optimal compression parameters based on model complexity and application scenario, ensuring a compression ratio of over 85% while maintaining visual quality. Develop an asynchronous compression module to offload time-consuming compression tasks to a background thread to avoid blocking the main thread. Build a decompression cache system to cache decompression results for frequently used models, improving loading speed.

[0102] Deploy the WebGPU compute pipeline, breaking down rendering tasks into multiple subtasks and allocating execution time through a time-slicing scheduling strategy to ensure stable 60FPS rendering performance. Design a work queue management system to prioritize rendering tasks related to user interaction. Implement a resource preloading mechanism to predict required resources based on scene changes and pre-load them into GPU memory. Develop an adaptive LOD adjustment algorithm to dynamically adjust the model's level of detail based on the object's distance from the camera. Optimize shader code to reduce the GPU's computational burden and improve rendering efficiency.

[0103] A spatial benchmark system was established to provide a unified coordinate transformation benchmark for the data storage engine, ensuring spatial consistency across data from different sources. The hybrid index structure of the spatiotemporal data engine provided efficient spatiotemporal query support for the conflict detection module, enabling rapid location of potential conflict areas. The rendering optimizer dynamically adjusted resource loading strategies based on spatiotemporal query results, prioritizing loading data for visible areas and key time points. Each module maintained state synchronization through a versioned snapshot mechanism. When a module's state changed, related modules were notified of updates through a publish-subscribe model. A cross-module communication protocol was designed to ensure data integrity and consistency during transmission. A central state manager was deployed to coordinate state updates across modules and avoid state inconsistencies.

[0104] Furthermore, a digital production planning and scheduling method may further include the following steps:

[0105] At the intelligent decision-making and evaluation layer, a dynamic interval-based TOPSIS multi-attribute decision-making model was established. Using an entropy-weighted hierarchical analysis fusion algorithm, the weights of five evaluation indicators—equipment utilization, process waiting time, process complexity, maintenance cost, and energy consumption—we were dynamically adjusted. Actual data for each indicator at different time points was collected to form an evaluation vector. This data was regularly updated and weights were recalculated to adapt to dynamic changes in the production process.

[0106] Introducing resource bottleneck index calculation function Where i represents the index variable in the summation symbol; w i represents the weight of the i-th resource factor, all w i The sum is 1. In this embodiment, resource factors include equipment utilization, process waiting time, process complexity, maintenance cost and energy consumption level; N i Represents the quantitative index value of the i-th resource factor; through the fuzzy inference system, the weight w is updated in real time based on real-time data from the production site, such as equipment load, material supply, etc. i The resource bottleneck index is derived according to the calculation function, thereby achieving dynamic perception and quantitative evaluation of production bottlenecks and timely identifying key factors that may affect production efficiency.

[0107] The resource bottleneck index and other results obtained from the quantitative assessment are used as input parameters and passed to the two-tier optimization engine. The upper layer uses the improved stochastic threshold optimization (STOCh) algorithm to construct a constraint model within the DBR (drum-buffer-rope) scheduling framework, combining production capacity, material supply, and other constraints to determine the scheduling strategy for bottleneck equipment. The lower layer designs a hybrid genetic-tabu search algorithm and develops an adaptive mutation operator with time window constraints. Combined with the spatiotemporal topological relationships presented by the process Gantt chart, it performs secondary scheduling of non-bottleneck equipment to optimize the overall production process.

[0108] Optimization solutions were validated using a Monte Carlo-digital twin hybrid simulator. A capacity fluctuation function was constructed to simulate potential supply chain disruptions, such as raw material supply delays and order changes. A multi-objective particle swarm optimization algorithm was integrated into the simulator, using a Pareto frontier evolutionary strategy to comprehensively evaluate optimization solutions based on multiple objectives (such as cost, efficiency, and quality), selecting the most robust options.

[0109] The optimized solution, validated through simulation, is fed into the real-time scheduling execution layer. Using a time-wheel algorithm, time is divided into 15-minute time slots. A mixed integer programming model is constructed, taking into account factors such as equipment status and order priority, to generate minute-by-minute work order sequences to guide the orderly progress of actual production. Work order execution is monitored in real time, and exceptions are promptly addressed.

[0110] In the closed-loop feedback control layer, an LSTM-GRU dual-channel prediction network is deployed to monitor production plan execution in real time. When the weekly plan deviation reaches 5% or more, a local rescheduling mechanism is triggered to adjust the affected production links. When the cumulative deviation reaches 15% or more, a global reoptimization is initiated to re-establish the production scheduling strategy. Through an incremental model update pipeline, the scheduling strategy is continuously optimized based on new data, achieving continuous evolution.

[0111] Data collaboration across all technical layers is achieved through a time-series state bus. The bottleneck index calculated by the decision-making and evaluation layer is fed back to the optimization engine, driving parameter adjustments to better address production bottlenecks. Simulation verification results are fed back to the scheduling execution layer, which is used to update the rule base and optimize actual scheduling rules. Abnormal signals detected by the prediction network trigger a closed-loop control mechanism, enabling the system to promptly respond to changes in the production process and maintain stable and efficient production.

[0112] Furthermore, a digital production planning and scheduling method may further include the following steps:

[0113] At the intelligent prediction and decision-making layer, a three-channel LSTM prediction network is constructed. The time series feature channel receives time series data from sensors such as equipment vibration, temperature, and current, preprocesses it, and then inputs it into the network. The spatial topology channel abstracts the workshop equipment layout into a graph structure and extracts structural features using a graph embedding algorithm. The process constraint channel encodes prior knowledge, such as process connection rules, into feature vectors. The network uses bidirectional gated recurrent units to process time series information, and an attention mechanism automatically assigns weights to each channel. After fusing multi-source information, a probabilistic output function is used to predict equipment failure probabilities.

[0114] The equipment failure probability prediction results are fed into a reinforcement learning optimization engine in real time. A composite reward function is designed that incorporates metrics such as production efficiency, equipment maintenance costs, and energy consumption. An adaptive exploration strategy is developed that dynamically adjusts the exploration rate based on the current system state, ensuring optimization results while avoiding local optima. Through continuous interaction with the environment, the optimal scheduling strategy is learned, achieving dynamic optimization of production resources.

[0115] A blockchain collaboration platform was built based on Hyperledger Fabric, with an innovative lightweight PBFT consensus mechanism designed to optimize messaging and inter-node communication. A sharded storage strategy was employed to store critical process data, and pipeline signature verification technology was used to reduce data upload latency to less than 200ms. A smart contract-driven traceability engine was developed to generate a unique identifier for each product, recording information from raw materials to finished product, supporting traceability at the process, device, and parameter levels. IPFS edge caching nodes were deployed to store non-critical data, reducing the storage burden on the blockchain.

[0116] Deeply integrate the blockchain platform with the digital thread system to establish a virtual-real synchronization pipeline, enabling real-time data exchange between the physical workshop and the virtual model. Design a prioritized timestamp protocol to assign different priorities to different types of data, ensuring that critical data is processed first. By optimizing communication protocols and data processing processes, the latency of the virtual workshop responding to physical system changes is controlled to less than 70ms, ensuring high synchronization between the virtual model and the physical system.

[0117] NVIDIA Jetson clusters were deployed at the edge computing layer to build distributed intelligent nodes. The TensorRT engine was used to optimize the scheduling model. Layer fusion technology was used to reduce the number of network layers. Precision calibration technology maintained model performance while reducing accuracy, improving inference speed and reducing memory usage. A model distillation pipeline was developed to transfer knowledge from large LSTM networks trained in the cloud to lightweight models, maintaining prediction accuracy while reducing computing power. An edge-cloud collaborative scheduler based on queuing theory was designed to dynamically allocate computing tasks, ensuring that critical path tasks are prioritized for processing on edge nodes, improving system responsiveness.

[0118] Each technical component interacts with each other in milliseconds via a high-performance data bus. The predictive network monitors equipment status in real time, issuing warnings when failure probability exceeds a threshold, triggering the optimization engine to make rescheduling decisions. The blockchain platform records the decision-making process and execution results, generating immutable digital certificates. The digital thread system synchronizes optimization plans with the physical execution layer to guide actual production. Edge nodes process sensor data in real time, extract key features, and provide feedback on execution status, forming a closed-loop control system that ensures efficient and stable production operations.

[0119] Furthermore, a digital production planning and scheduling method may further include the following steps:

[0120] At the security access control layer, a spatiotemporal context-aware RBAC (role-based access control) model is constructed. First, the role set R is defined, covering six job categories, such as process engineers and planners, and basic permissions are defined based on the responsibilities of different positions. Based on the physical location information of the equipment, a geographic fence is generated as a spatial constraint L to limit the user's operating permissions within a specific area. A sliding time window is used as a time constraint T to specify the permitted time period for sensitive operations. The three are combined to construct a dynamic permission matrix P = [R × (L ⊕ T)] to ensure that permission allocation is dynamically adjusted over time and space.

[0121] Design an authority control system to collect user behavior characteristics such as operation frequency, path pattern, device fingerprint, etc. and obtain dynamic confidence score Where i is the index variable in the summation symbol, which is used to traverse from 1 to n different user behavior feature-related factors; n represents the total number of user behavior feature factors involved in calculating the dynamic confidence score; wi is the weight of the i-th user behavior feature factor; f i (x) is a function of the i-th user behavior characteristic factor, and x is an input variable related to the user behavior characteristic. When the score C is lower than 0.85, multi-factor authentication is triggered, requiring the user to provide additional authentication information, such as SMS verification code and biometrics, to prevent unauthorized operations and ensure system security.

[0122] The permission control system is deeply integrated with the operation log module. The logging unit uses an improved CRDT (Conflict-Free Replicated Data Type) data structure to ensure the consistency and reliability of logging in a distributed environment. An operational semantic analyzer is developed to automatically parse user operation instruction types, such as add, delete, modify, and query. Vector clocks are used to mark the timing relationship of operations, facilitating subsequent tracing and auditing of operational processes.

[0123] The collaborative perception engine is enhanced based on the WebRTC architecture. A multimodal interaction channel is built, and low-latency media streams are established using STUN / TURN penetration technology. This allows for the simultaneous transmission of multi-dimensional interactive data such as cursor tracks, voice annotations, and gesture annotations, enhancing the richness and real-time nature of collaborative interactions. An intelligent bandwidth allocator is deployed, using LSTM to predict network bandwidth fluctuations and dynamically adjust the resolution and frame rate of the video stream, effectively improving collaborative fluency in weak network environments. A conflict visualization module is developed, using a differential coloring algorithm to highlight conflicting areas of concurrent editing, and combined with an operation traceability map to display operation history, assisting users in making decisions.

[0124] Establish a linkage mechanism between the collaborative perception engine and the permission control system. When the collaborative perception engine detects unauthorized operations, it immediately triggers data desensitization. For sensitive fields, homomorphic encryption technology is used for transmission to ensure data security during transmission, prevent sensitive information leakage, and ensure that legitimate users can perform collaborative operations normally.

[0125] Based on the above method, the embodiment of the present application also discloses a digital production planning and scheduling system. A digital production planning and scheduling system includes:

[0126] The storage architecture building module is used to build a distributed storage architecture based on edge computing nodes. Each production site deploys a local database with a hybrid storage structure, where real-time production data is stored in a time series database and device topology relationships are stored in a graph database.

[0127] The data synchronization module is used by the central coordinator to synchronize data across nodes using the improved Raft consensus algorithm, implement multi-version concurrency control based on vector clocks, and establish a globally consistent data view.

[0128] The scheduling platform construction module is used to build a visual scheduling platform based on the Vue3 framework, and adopts the Canvas and WebGL collaborative rendering architecture to achieve human-computer interaction. Specifically, the Canvas layer integrates the quaternion rotation algorithm to achieve low-latency topology editing functions, and realizes smooth operation without gimbal lock in three-dimensional space by calculating the rotation matrix. The WebGL layer uses frustum clipping and instanced rendering technology, combined with the LOD dynamic grading strategy to support real-time rendering of millions of production components.

[0129] The transformation model module is used to establish a dynamic coordinate transformation model based on bilinear interpolation, design a spatial mapping function including distortion compensation, and establish a multi-threaded coordinate service based on WebWorker to achieve sub-millimeter bidirectional mapping between pixel coordinates and physical coordinates;

[0130] The analysis model module is used to build a three-dimensional spatiotemporal analysis model that integrates multi-dimensional features. This includes: using a dilated gated convolution module to capture the local spatiotemporal features of the equipment state; building a spatiotemporal graph network based on a multi-head attention mechanism, and establishing spatiotemporal associations across process nodes through an adjacency matrix; and using an improved dynamic time warping (DTW) algorithm to calculate the similarity of production scenarios and achieve constrained flexible path matching.

[0131] The feature fusion module is used to fuse features of each unit through residual connections, and finally outputs a scheduling plan with confidence weights.

[0132] An embodiment of the present application further discloses an intelligent terminal, which includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed by a digital production planning and scheduling method as described above.

[0133] The present application also discloses a computer-readable storage medium. The computer-readable storage medium stores a computer program capable of being loaded by a processor and executing the above-described digital production planning and scheduling method. The computer-readable storage medium includes, for example, a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, among other media capable of storing program code.

[0134] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the scope of protection of the invention. Obviously, the embodiments described are only some embodiments of the present invention, rather than all embodiments. Based on these embodiments, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in this field can still combine, add, delete or make other adjustments to the features in the various embodiments of the present invention according to the circumstances without conflict, without making creative work, so as to obtain different other technical solutions that do not deviate from the concept of the present invention in essence, and these technical solutions also fall within the scope of protection of the present invention.

Claims

1. A digital production planning and scheduling method, characterized in that: The following steps are involved: Build a distributed storage architecture based on edge computing nodes; A central coordinator is used to synchronize data across nodes using the improved Raft consensus algorithm. Multi-version concurrency control is implemented based on vector clocks to establish a globally consistent data view. Build a visual scheduling platform based on the Vue3 framework, and use Canvas and WebGL collaborative rendering architecture to achieve human-computer interaction; A dynamic coordinate transformation model based on bilinear interpolation was established, a spatial mapping function with distortion compensation was designed, and a multi-threaded coordinate service was established based on WebWorker to achieve sub-millimeter bidirectional mapping between pixel coordinates and physical coordinates. Construct a three-dimensional spatiotemporal analysis model that integrates multi-dimensional features; Each unit performs feature fusion through residual connection and finally outputs a scheduling plan with confidence weights.

2. A digital production planning and scheduling method according to claim 1, characterized in that: Also includes: A multi-dimensional spatiotemporal data engine is built based on PostGIS, using a four-dimensional index structure to store the spatiotemporal topological relationship of the site vector base map, the LOD metadata of the component model, and the timeline layout snapshot with version identification; Design a spatiotemporal version controller based on a hybrid index of B+ trees and R trees, supporting incremental storage and merging of multiple timeline branches, and achieving second-level historical state backtracking through version hash chains; Deploy an adaptive projection engine, integrate the improved Lambert projection algorithm in GeoServer, and design a dynamic interpolation function: Designed an OT conflict resolution algorithm based on time and space priorities, deployed layered WebSocket services, and used Epoll+Redis to achieve 10,000-level concurrent operation synchronization; Develop a coordinate coupling module for Mapbox GL and Three.js, and achieve dynamic alignment of 2D / 3D coordinate systems through affine transformation matrices: Design a multi-granularity space-time lens model to support time scale switching; Develop a time mask renderer based on Compute Shader to achieve progressive visualization of the time dimension through spatiotemporal weight functions.

3. A digital production planning and scheduling method according to claim 1, characterized in that: Also includes: Build a cross-site spatiotemporal federated learning architecture, deploying a Bi-GRU prediction model with a spatiotemporal attention mechanism on each edge node; Design a model traceability mechanism to track parameter contributions through gradient fingerprint mapping; Develop a clock-driven Delaunay triangulation algorithm, combined with an LSTM buffer capacity prediction model, to generate a topological structure with spatiotemporal constraints. Introduce a spatiotemporal conflict detection matrix to verify the feasibility of layout solutions in real time. Build an improved NSGA-III optimization engine and design a three-stage optimization process; Integrated constraint processing mechanism, embedding the site spatiotemporal topology as a hard constraint into the optimization process; A residual correction module is developed to dynamically adjust the prediction output through the error feedback matrix.

4. A digital production planning and scheduling method according to claim 1, characterized in that: Also includes: Construct a bidirectional conversion channel between EPSG:3857 and the local coordinate system, and design an affine transformation model with distortion compensation; Developed a deep learning-based EXIF ​​metadata parser to extract image physical features using a convolutional attention network; Developed a layer management system based on CSS Transform Module Level 1, using a versioned differential storage strategy to generate incremental snapshots for each operation; Implement non-destructive editing stack and support backtracking to any historical state through snapshot hash chain.

5. A digital production planning and scheduling method according to claim 1, characterized in that: Also includes: In the intelligent decision-making evaluation layer, a dynamic interval-type TOPSIS multi-attribute decision-making model is constructed, and the weight coefficients of the five-dimensional evaluation indicators are dynamically adjusted through the entropy weight-hierarchy analysis fusion algorithm; Introducing resource bottleneck index calculation function Where i represents the index variable in the summation symbol; w i represents the weight of the i-th resource factor, all w i The sum is 1; N i Represents the quantitative index value of the i-th resource factor; The quantitative evaluation results are passed as input parameters to a two-layer optimization engine. The optimization solution is verified using a Monte Carlo-digital twin hybrid simulator. A capacity fluctuation function is constructed to simulate supply chain disturbance scenarios. A multi-objective particle swarm optimization algorithm is integrated to design a Pareto frontier evolutionary strategy. The optimized plan is input into the real-time scheduling execution layer, and the time slots are divided into 15-minute granularity based on the time wheel algorithm. A mixed integer programming model is constructed to generate minute-level work order sequences.

6. A digital production planning and scheduling method according to claim 1, characterized in that: Also includes: Adopting a three-channel input architecture, a multimodal fusion LSTM prediction network is constructed; By integrating multi-source information through bidirectional gated recurrent units and attention mechanisms, a probability output function is designed to predict the probability of equipment failure. The prediction results are fed into the reinforcement learning optimization engine in real time to design a composite reward function and develop an adaptive exploration strategy; Build a blockchain collaboration platform based on Hyperledger Fabric and innovatively design a lightweight PBFT consensus mechanism; Deploy NVIDIA Jetson clusters at the edge computing layer to build distributed intelligent nodes. Use the TensorRT engine to optimize production scheduling models. Improve model inference speed through layer fusion and precision calibration technology. Developed a model distillation pipeline to compress large LSTM networks trained in the cloud into a lightweight version suitable for edge deployment, maintaining prediction accuracy while reducing computational energy consumption; Design an edge-cloud collaborative scheduler to dynamically allocate computing tasks based on queuing theory, and prioritize critical path tasks for processing at edge nodes.

7. A digital production planning and scheduling method according to claim 1, characterized in that: Also includes: A spatiotemporal context-aware RBAC model is constructed, defining a dynamic permission matrix P = [R × (L ⊕ T)], where the role set R includes process engineers and planners, the spatial constraint L generates a geofence based on the physical location of the equipment, and the temporal constraint T uses a sliding time window to limit sensitive operation periods. Design an access control system to generate dynamic confidence scores by collecting user behavior characteristics Where i is the index variable in the summation symbol, which is used to traverse from 1 to n different user behavior feature-related factors; n represents the total number of user behavior feature factors involved in calculating the dynamic confidence score; w i is the weight of the i-th user behavior feature factor; f i (x) is a function of the i-th user behavior characteristic factor, and x is an input variable related to the user behavior characteristic; when C < 0.85, multi-factor authentication is triggered to intercept unauthorized operations; The permission control system is deeply integrated with the operation log module. The log recording unit adopts an improved CRDT data structure, an operation semantic analyzer is designed to automatically parse instruction types, and the operation timing relationship is marked by a vector clock. The collaborative perception engine is enhanced based on the WebRTC architecture. The collaborative perception engine is linked to the permission control system. When unauthorized operations are detected, data desensitization is automatically triggered, and sensitive fields are transmitted using homomorphic encryption.

8. A digital production planning and scheduling system, characterized in that: include: Storage architecture building module, used to build a distributed storage architecture based on edge computing nodes; The data synchronization module is used to synchronize data across nodes using a central coordinator and an improved Raft consensus algorithm. It also implements multi-version concurrency control based on vector clocks and establishes a globally consistent data view. The human-computer interaction module is used to build a visual scheduling platform based on the Vue3 framework, and uses the Canvas and WebGL collaborative rendering architecture to achieve human-computer interaction; The dynamic coordinate conversion module is used to establish a dynamic coordinate conversion model based on bilinear interpolation, design a spatial mapping function including distortion compensation, and establish a multi-threaded coordinate service based on WebWorker to achieve sub-millimeter bidirectional mapping between pixel coordinates and physical coordinates; 3D spatiotemporal analysis module, used to build a 3D spatiotemporal analysis model integrating multi-dimensional features; The feature fusion module is used to fuse features of each unit through residual connections, and finally outputs a scheduling plan with confidence weights.

9. An intelligent terminal, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executes a digital production planning and scheduling method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that A computer program is stored which can be loaded by a processor and executes a digital production planning and scheduling method according to any one of claims 1 to 7.

Images