Industrial chain digital collaboration and production factor trading platform

Abstract

The invention relates to the technical field of industrial internet and intelligent manufacturing, and discloses an industrial chain digital collaboration and production element transaction platform, which comprises an edge computing node configured on an industrial equipment side and used for collecting industrial equipment data and performing computing; the data storage is connected with the edge computing node and is used for integrating and storing the data; the private cloud is connected with the data storage, is used for providing services for the large-scale enterprise, and can use a resource matching algorithm model optimized by federated learning, and realizes the coordination between the large-scale enterprise and the public cloud industry ecology by relying on a public cloud industry chain coordination engine and a production element transaction engine; and the public cloud is connected with the data storage and is used for providing digital services for production operation and industrial ecology of small and medium-sized enterprises. According to the method, a distributed cooperative computing architecture of edge computing, private cloud and public cloud is constructed, differentiated services facing different scales of enterprises in an industrial chain are realized, a data barrier is effectively broken through, digital collaboration of the whole chain is promoted, and the service efficiency is improved. And platform support is provided for marketization efficient configuration of production elements such as productivity, materials, computing power, funds, services, carbon emission permit, intellectual property and data, and the operation efficiency and the intelligent level of the whole industrial chain are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of industrial internet and intelligent manufacturing, in particular to an industry chain digitalization collaboration and production factor transaction platform. BACKGROUND

[0002] At present, the wave of industrial internet is driving the manufacturing industry to transform deeply towards intelligence and collaboration. The industry chain is no longer a simple linear upstream and downstream relationship, but has evolved into a complex and dynamic ecological network. How to enable each enterprise on the chain, especially the large number of small and medium-sized enterprises, to participate in digital collaboration at low cost and high efficiency, and to enable production capacity, materials, computing power, funds, services, data, carbon emission rights, intellectual property rights and other production factors to be freely circulated and optimally allocated like commodities, is the core issue that the entire industry is exploring.

[0003] In the prior art, there are some digital tools to promote industry collaboration. Enterprises usually deploy complex enterprise resource planning (ERP) and manufacturing execution systems (MES) to finely manage internal production and sales processes. In external collaboration, by building a supplier relationship management (SRM) platform or using EDI (electronic data interchange), etc., the enterprise interacts with suppliers for order, material and other information. These practices have to some extent opened up local information channels and maintained collaboration in specific business scenarios.

[0004] However, these existing solutions have many shortcomings when building a truly integrated and intelligent industry chain collaboration and factor transaction ecosystem. First, there is an insurmountable "data gap" between different enterprises and different systems. Real-time operation data in the production field, enterprise production capacity and production planning, and other key information are fragmented in information silos, resulting in very low transparency of the entire industry chain. Second, the collaboration mode between enterprises is quite rigid and mostly relies on pre-set processes and manual communication. In the face of market mutations or supply chain abnormalities, this mode is slow to respond and difficult to achieve dynamic resource scheduling and risk sharing. Third, the deeper problem is the lack of "trust". Cooperation between enterprises largely depends on business contracts and long-term relationship maintenance. For the transaction of new production factors such as idle production capacity, computing power, services, data, algorithm models, intellectual property rights, and carbon emission rights, there is a lack of a trusted, efficient and low-cost transaction environment, and their potential value has not been fully activated. Finally, there is a natural contradiction between data sharing and privacy protection. Enterprises have the desire to use data for collaborative innovation, but are strongly concerned about the leakage of their own core data. This dilemma greatly hinders the process of intelligent industry chain. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides an industry chain digitalization collaboration and production factor transaction platform, aiming to solve the problems of low industry chain collaboration efficiency and difficulty in marketization and efficient circulation of production factors caused by information islands, lack of trust and privacy protection difficulties.

[0006] To achieve the above object, the present application is implemented by the following technical solutions: an industry chain digitalization collaboration and production factor transaction platform, comprising:

[0007] The edge computing node arranged on the industrial equipment side can be used to collect data of the industrial equipment, and can also calculate device comprehensive performance and carbon emission performance data in real time based on work orders, plans, quality and collected device running states and running data, and deploy a federated learning model such as a machine vision quality inspection model, a device life prediction model and an advanced algorithm and strategy for advanced process control, so as to realize process optimization and real-time control by means of mathematical models, intelligent algorithms and multivariate control technology; at the same time, the models are coordinated with a public cloud federated learning scheduling hub to generate model parameters through local training.

[0008] The data storage is connected with the edge computing node, the private cloud and the public cloud, and is used to integrate and store data;

[0009] The private cloud is connected with the data storage and is used to provide services to large enterprises; the private cloud can use a resource matching algorithm model optimized by federated learning to respond to early warnings triggered by device comprehensive performance and supply chain abnormal events, carry out advanced planning and scheduling work, rely on an industry chain collaboration engine deployed on the public cloud to send material demand plans and production plans for outsourcing manufacturing to suppliers, form corresponding production orders and purchase orders through a production factor transaction engine deployed on the public cloud after confirmation by the suppliers, and receive and monitor progress information and quality information returned by the suppliers; at the same time, the enterprise private domain data such as device comprehensive performance is called in a federated learning manner to be coordinated with the public cloud federated learning scheduling hub to participate in model training.

[0010] The public cloud is connected with the data storage and is used to provide digital services to small and medium-sized enterprise production and operation and industry ecology; users of the public cloud can use a resource matching algorithm model optimized by federated learning to respond to early warnings triggered by device comprehensive performance and supply chain abnormal events, carry out advanced planning and scheduling work, send material demand plans and production plans for outsourcing manufacturing to suppliers through an industry chain collaboration engine deployed on the public cloud, form corresponding production orders and purchase orders relying on a production factor transaction engine deployed on the public cloud after confirmation by the suppliers, and receive and monitor progress information and quality information returned by the suppliers; at the same time, enterprise private domain data such as device comprehensive performance is called in a federated learning manner to be coordinated with the public cloud federated learning scheduling hub to participate in model training.

[0011] The edge computing node, the private cloud and the public cloud constitute a distributed collaborative computing architecture.

[0012] Preferably, the edge computing node deployed on the industrial equipment side has a communication protocol conversion capability and is deployed with at least one component selected from the group consisting of a federated learning node, a machine vision quality inspection model, a device life prediction model and an advanced process control algorithm model.

[0013] Preferably, the data storage includes:

[0014] a time series database;

[0015] a relational database;

[0016] a non-relational database;

[0017] Preferably, the private cloud is deployed with:

[0018] a digital twin engine;

[0019] an advanced planning and scheduling;

[0020] federated learning.

[0021] Preferably, the public cloud is deployed with:

[0022] a visual model orchestrator;

[0023] a federated learning model scheduling hub;

[0024] an industry chain collaboration engine;

[0025] a production factor transaction engine;

[0026] a resource matching module;

[0027] a multi-tenant manufacturing operation management module;

[0028] a manufacturing knowledge center.

[0029] Preferably, the resource matching module is configured to support and execute one or more optional scheduling strategies and introduce external environmental risk factors and sudden public event factors to optimize the supply chain resilience. When the algorithm scheduling conflicts with human experience, the system can automatically generate multi-solution simulation comparison (such as delivery delay vs. cost change) to assist managers in decision-making. The multi-tenant manufacturing operation management module realizes the manufacturing operation management function based on the multi-tenant cloud architecture and provides manufacturing operation management services for users using the public cloud.

[0030] Preferably, the production factor transaction engine is built based on blockchain technology, wherein the smart contract is further configured to build a dynamic credit model based on the equipment comprehensive performance data and carbon emission performance data.

[0031] Preferably, the dynamic credit model comprises:

[0032] a credit score generation unit configured to generate a credit score based on data in a group consisting of the equipment comprehensive performance data and / or carbon emission performance data and an industry differentiated weight coefficient;

[0033] a smart contract clause matching unit configured to automatically match preset smart contract clauses corresponding to different credit levels according to the credit score.

[0034] Preferably, the production factor comprises at least one of capacity, material, computing power, funds, service, carbon emission right, intellectual property right and data.

[0035] Preferably, the edge computing node and the private cloud are configured as clients of federated learning for training a model locally and generating model parameters.

[0036] The public cloud is deployed with a federated learning model scheduling hub configured as a server of federated learning for updating a global model based on the model parameters uploaded by the clients.

[0037] Preferably, the federated learning model scheduling hub deployed by the public cloud quantifies the contribution of participants based on contribution measurement and incentive distribution rules, and allocates computing power resources or provides transaction fee discounts based on contribution degree, so as to stimulate the willingness of users to participate in model training.

[0038] The present application provides an industrial chain digitalization collaboration and production factor transaction platform.

[0039] 1. The present application can provide differentiated services for enterprises of different sizes in the industrial chain by building a distributed collaborative computing architecture of edge computing, private cloud and public cloud. This flexible distributed design changes the rigid service mode of the prior art platform, which can meet the needs of large enterprises for performance and data security, and can also meet the wishes of small and medium-sized enterprises to transform low-cost digitalization by using the multi-tenant manufacturing operation management of the public cloud and to protect data security, laying a foundation for realizing the smooth flow of digitalization collaboration and production factors in the whole chain.

[0040] 2. The present application adopts a resource matching algorithm model optimized by federated learning, which can more accurately calculate the resource allocation scheme within the industrial chain, which helps to change the inefficient mode of relying on simple information matching between enterprises, and further improves the idle capacity and resource mismatch caused by information asymmetry in traditional collaboration.

[0041] 3、The present application builds a transparent and reliable environment for the value circulation of production factors such as capacity, materials, computing power, funds, services, carbon emission rights, intellectual property rights, and data by means of a production factor transaction engine based on blockchain and smart contracts, which breaks the dependence on traditional transactions on strong intermediaries and cumbersome processes, and provides an effective way to solve the problem of high trust cost of cross-subject transactions and difficulty in marketization of factor value.

[0042] 4、The present application introduces a federated learning technical framework, which allows each participant to collaboratively train a global model without exposing data outside the private domain, which helps to balance the data sharing needs and the privacy protection needs of enterprises in the collaborative development of the industry chain, compared with the existing technology which either sacrifices privacy or sticks to data to form an information island. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 The overall system architecture of the platform of the present application is shown in the figure;

[0044] Figure 2 The resource matching and supply chain resilience optimization flowchart of the present application is shown in the figure;

[0045] Figure 3 The user operation flowchart of the visualization model composer of the present application is shown in the figure;

[0046] Figure 4 The dynamic credit granting and transaction flowchart of the present application is shown in the figure;

[0047] Figure 5 The federated learning collaborative modeling framework of the present application is shown in the figure. DETAILED DESCRIPTION

[0048] The technical solutions in the embodiments of the present application will be described in detail below with reference to the accompanying drawings of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0049] The embodiment of the present application provides an industry chain digitalization collaboration and production factor transaction platform. In a specific implementation, the overall architecture of the platform can be logically divided into edge computing nodes, data storage, private cloud and public cloud.

[0050] Among them, the edge computing nodes, the private cloud and the public cloud constitute a distributed collaborative computing architecture with complementary functions and collaborative work, realizing a complete technical closed loop from device data acquisition, enterprise operation to industry ecological service and production factor transaction.

[0051] It should be noted that the four-part architecture described in the present application, including "edge computing nodes, data storage, private cloud and public cloud", is a preferred and most complete embodiment of the present application. Those skilled in the art can understand that without departing from the core idea of the present application, the platform can also use the following embodiments:

[0052] Optionally, the public cloud and the edge computing node can be used in cooperation.

[0053] In an optional embodiment of the present application, the platform can not include a private cloud, forming a collaborative computing architecture with "edge computing nodes" and "public cloud" as the core.

[0054] Edge computing node: still deployed in the industrial field, responsible for real-time data collection and local completion of core device overall efficiency (OEE), carbon emission performance data calculation, running of federated learning client model, etc.

[0055] Public cloud: directly connected with edge computing nodes and data storage of each enterprise. It receives OEE results and key data uploaded by the edge node, and runs its core industry chain collaboration engine, production factor transaction engine, resource matching module, etc., to provide services to all enterprises.

[0056] In the above mode, the collaboration of the industry chain and the production factor transaction are completed completely within the framework of the public cloud, which has the advantage of utilizing the computing resources on the edge side, reducing the computing load and network delay of the cloud, and improving the response speed.

[0057] Optionally, the public cloud service mode can be used.

[0058] In another optional and more simplified embodiment, the core of the platform can only be the "public cloud" itself.

[0059] Device side: only a lightweight data collection gateway needs to be deployed, and its only task is to collect raw device operation data (such as on-off status, production count, etc.) and upload it safely to the public cloud. Such a data collection gateway is a prior art.

[0060] Public cloud: in this mode, the public cloud not only provides industry chain collaboration and production factor transaction services, but also is responsible for the calculation of device overall efficiency (OEE) and carbon emission performance data in the cloud.

[0061] Optionally, the mixed cloud and data collection gateway can be used in cooperation.

[0062] In another optional embodiment of the present application, the private cloud and the public cloud form a hybrid cloud, and the data collection gateway forms a distributed collaborative computing architecture, realizing device data collection, enterprise operation, industry ecological collaboration and production factor transaction.

[0063] In this mode, the private cloud and the public cloud can perform overall equipment effectiveness (OEE) and carbon emission performance data calculation, and the public cloud provides industry chain synergy and production factor transaction services.

[0064] Optional Embodiment Four: Hybrid Deployment of Public Cloud and Data Collection Gateway and Edge Computing Node

[0065] In another optional embodiment of the present application, the platform is hybridly deployed with data collection gateway and edge computing node, and the public cloud is the core component of the distributed collaborative computing architecture, realizing device data collection, enterprise operation, and industry ecological synergy and production factor transaction.

[0066] In this mode, the public cloud can perform overall equipment effectiveness (OEE) and carbon emission performance data calculation, and the public cloud provides industry chain synergy and production factor transaction services.

[0067] Optional Embodiment Five: Hybrid Deployment Mode

[0068] In another optional embodiment of the present application, the private cloud and the public cloud are hybridly deployed, and the data collection gateway and the edge computing node are hybridly deployed, forming a distributed collaborative computing architecture, realizing device data collection, enterprise operation, and industry ecological synergy and production factor transaction.

[0069] In this mode, the private cloud and the public cloud can perform overall equipment effectiveness (OEE) and carbon emission performance data calculation, and the public cloud provides industry chain synergy and production factor transaction services.

[0070] Optional Embodiment Six: "Private Cloud" Mode

[0071] In another optional embodiment of the present application, the platform can not contain the public cloud, forming a distributed collaborative computing architecture with data collection gateway or edge computing node and "private cloud" as the core.

[0072] The "private cloud" provides the public cloud industry chain synergy and production factor transaction services to the third party through remote procedure call (RPC), virtual private network (VPN), software-defined local area network (SD-LAN), software-defined wide area network (SD-WAN), peer-to-peer network (P2P), intranet penetration, dynamic domain name resolution (DDNS), or similar technologies, and the data collection gateway and / or the edge computing node, realizing device data collection and / or overall equipment effectiveness (OEE) and carbon emission performance data calculation, federated learning model training and use, realizing device data collection, enterprise operation, and industry ecological synergy and production factor transaction.

[0073] In the above embodiments, the focus of the protection of the present application is not the data collection method, but the unique combination of functional modules deployed in the public cloud itself and its cooperative working method, i.e., how it utilizes the received data to provide digital services for the entire industry chain through its built-in resource matching module, dynamic credit model, and industry chain synergy engine, which helps to reduce the deployment and maintenance costs on the user side.

[0074] The edge computing node is a data entry and intelligent outpost for the platform to interact with the physical world, and is configured in the production site of an industrial enterprise. It needs to be specially pointed out that in the present application, the "edge computing node" is a functional collection concept, and the physical implementation method is very flexible. On the one hand, it can be an integrated intelligent edge device integrating data collection, protocol conversion, real-time calculation and model running, etc. The device can be integrated with a trusted execution environment (TEE) and embedded with a hardware-level security module (such as SGX) to ensure that various models such as device life prediction models run in encrypted memory, while realizing trusted verification of results such as machine vision quality inspection, forming a protection strategy against physical side channel attacks.

[0075] On the other hand, it can also be a logical combination of hybrid deployment: that is, one or more data collection gateways responsible for basic data collection and protocol conversion work together with one or more edge computing units responsible for performing advanced tasks such as device overall efficiency (OEE) and carbon emission performance data calculation, running federated learning models, etc. Dynamic identity authentication processes can be carried out through a zero-trust architecture (ZTA) to effectively prevent nodes from being maliciously hijacked.

[0076] In this hybrid deployment mode, the data collection gateway and the edge computing unit can be physically separated, but logically together constitute the functions of the "edge computing node" described in the present application.

[0077] Regardless of the deployment method, the core task is to collect real-time operation data of industrial devices, specifically state, parameter and process data from sensors, actuators, PLCs (programmable logic controllers) and other devices.

[0078] In order to be compatible with the complex and diverse devices and protocols in the industrial field, the edge computing node has multiple communication protocol conversion capabilities. It can parse and convert various industrial fieldbus protocols and industrial Ethernet protocols including but not limited to OPC UA / DA, Modbus, MQTT, etc. It uses device models and point tables conforming to ISA-95 and UNS (Unified Namespace) standards for data collection, laying a foundation for subsequent digital twin, data storage and analysis.

[0079] By processing data on the edge side, 95% of real-time data can be processed locally, reducing cloud load by 40% and strictly controlling response delay within 50ms.

[0080] Further, the edge computing node can evaluate task priority in real time according to power consumption and network bandwidth, automatically migrate high-load tasks (such as machine vision quality inspection) to the cloud, and ensure local execution of critical tasks (such as process control).

[0081] Further, the edge computing node has carbon emission performance calculation capability, and the carbon emission performance calculation work is based on real-time collection of various types of energy consumption data, which comes from metering devices such as electricity meters, water meters, steam meters, and gas meters; at the same time, the measurement of carbon emission performance is realized through two dimensions of carbon emission and carbon intensity improvement rate.

[0082] Carbon emission reduction = current period carbon emission - baseline period carbon emission;

[0083]

[0084] Further, data comparison with government carbon regulatory platforms and third-party auditing agencies (such as SGS) can be realized through API interfaces to ensure the authenticity and credibility of carbon emission reduction; to ensure the non-tamperability of carbon emission performance data, blockchain technology is used to record carbon emission performance logs, and relevant hash values are synchronized to smart contracts; these verified and recorded carbon emission performance data also serve as productivity credit data, providing key input support for subsequent dynamic credit models built in public cloud environments.

[0085] Further, the edge computing node integrates a device overall efficiency (OEE) calculation engine. Based on work orders, plans, quality, and collected device running states (such as startup, shutdown, and failure), production counts, and standard cycle times, the engine can automatically calculate and continuously update device OEE indicators.

[0086] Device overall efficiency is a core indicator for measuring the production efficiency of manufacturing equipment, which is obtained by multiplying three key dimensions: availability, performance efficiency, and quality pass rate. The calculation formula is as follows:

[0087] OEE = Availability × Performance Efficiency × Quality Pass Rate × 100%;

[0088] Among them, the three core dimensions of OEE are:

[0089] Availability: measures the ratio of actual running time to planned production time, reflecting the loss caused by device failure, mold change, debugging, etc.

[0090]

[0091] Performance Efficiency: Measures the gap between actual production speed and theoretical maximum speed, reflecting speed loss caused by idling, deceleration, etc.

[0092]

[0093] Quality Conformance: Measures the ratio of qualified products to total production, reflecting quality loss caused by product defects, rework, etc.

[0094]

[0095] Through the comprehensive calculation of these three dimensions, the platform can accurately quantify the losses of equipment operation efficiency, automatically calculate and continuously update the OEE index, which not only reflects the current health status and production efficiency of the equipment in real time, but also drives delivery early warning and dynamic resource scheduling.

[0096] Historical trend data can also serve as core productivity credit data, providing key input for subsequent dynamic credit models built on public clouds.

[0097] In addition, the edge computing node also has localized data processing and model execution capabilities. According to the specific application scenario, at least one of the following can be flexibly deployed on it:

[0098] Federal Learning Node: As a client of the federal learning system, it directly trains models on the edge side using locally collected data, contributing to the optimization of global models while achieving "data available but invisible" data privacy protection.

[0099] Machine Vision Quality Inspection Model: Can be used for real-time product quality detection on the production line. By deploying on the edge side, it can quickly analyze and determine product images, achieving low-latency and efficient automated quality control.

[0100] Equipment Life Prediction Model: A mathematical tool that estimates the remaining useful life of equipment by analyzing operating data, environmental factors, and historical failure records. Its core value lies in preventive maintenance, cost optimization, safety improvement, and efficiency improvement. It is mainly divided into three categories: statistical theory (such as Weibull distribution, exponential distribution, and Gamma process model), physical failure mechanism (such as Arrhenius model and Paris law model), and data-driven (such as machine learning and deep learning models like random forest, LSTM, CNN, and hybrid models combining physics and data);

[0101] The implementation process includes data collection preprocessing, model selection and training, validation and evaluation, and real-time prediction updating, integrating industrial internet, digital twin, and other technologies to optimize the entire life cycle of equipment.

[0102] Advanced process control: It can be used for complex, nonlinear, and strongly coupled industrial production processes (such as chemical, oil refining, pharmaceutical, and power generation). Through advanced algorithms and strategies beyond conventional PID control, it uses mathematical models, intelligent algorithms, and multivariable control technology to solve dynamic characteristics, constraint conditions, and multi-objective optimization problems that traditional control methods cannot handle. It realizes more precise and efficient process optimization and real-time control. Deploying advanced process control models and algorithms on the edge can ensure fast and reliable control instructions.

[0103] Data storage, as the data hub of the platform, is connected with edge computing nodes, private clouds, and public clouds for data integration and storage.

[0104] Data storage receives data collected and preliminarily processed by edge computing nodes, and private clouds or public clouds further integrate, clean, correlate, and persistently store the data, providing unified, standard, and high-quality data services for distributed collaborative computing.

[0105] To efficiently store and manage different types and structures of data, the data storage internally collaborates with multiple database engines, including:

[0106] Time series database: It is specifically designed for efficiently storing and querying large amounts of time-stamped operational data generated by devices, such as temperature, pressure, vibration frequency, and other time series data.

[0107] Relational database: It is used to store structured business data, such as enterprise order information, bill of materials (BOM), customer information, and production plans.

[0108] Non-relational database: It is used to store semi-structured or unstructured data, such as device operation manuals, process procedure documents, three-dimensional model files, and knowledge graph data.

[0109] Private cloud, connected with data storage, provides services to large enterprises, considering their higher requirements for data security, system performance, business integration, and customized functions. Through private deployment, it can provide a safe, exclusive, and high-performance digital management and operation environment for large enterprises. The private cloud provides a series of deeply customized digital tools, including:

[0110] Digital twin engine: used to build a digital twin body driven by real-time data collected by the data acquisition gateway of the edge computing node deployed in the production site of the industrial enterprise, in accordance with the ISA-95 hierarchical architecture and the standardized interface of the unified namespace (UNS), for the equipment, production line, and even the entire factory of a large enterprise, to realize real-time monitoring, state prediction, process simulation, and optimization decision of the physical entity.

[0111] Specifically, the digital twin body can be built based on Unity3D.

[0112] Advanced planning and scheduling: taking into account complex factors such as orders, capacity, materials, inventory, in-transit, work-in-process, equipment constraints, process path, etc., to make accurate production planning and dynamic production scheduling, through the deployment of an industry chain collaboration engine in the public cloud, to send material demand plans and outsourcing production plans to suppliers, after confirmation by the suppliers, through the deployment of a production factor trading engine in the public cloud, to form corresponding production orders and purchase orders, and to receive and monitor the progress information and quality information returned by the suppliers.

[0113] Federal learning node: the federal learning node deployed in the private cloud can also act as a client in the federal learning, participating in joint modeling across enterprises and training resource scheduling algorithms across enterprises.

[0114] Specifically, the advanced planning and scheduling can use the trained cross-enterprise resource scheduling algorithm, combined with the information mining and intelligent processing tools of the Wiseflow type, using a hybrid method combining the long short-term memory (LSTM) algorithm and the Kalman filter (KF) algorithm, to quantitatively evaluate and predict the supply chain resilience.

[0115] Supply chain resilience refers to the ability of a supply chain to maintain its core functions and quickly recover to the desired state after being disturbed. The platform quantifies it from three dimensions: supply stability, risk response speed, and dynamic adjustment capability. Therefore, the hybrid method of LSTM and Kalman filter used by the platform can effectively balance the advantages of data-driven learning and model-driven optimization, achieving dynamic and robust supply chain state modeling and risk warning.

[0116] LSTM (Long Short-Term Memory Network) captures the nonlinear dynamics and long and short-term dependencies in supply chain time series data through its unique gating mechanism (forgetting gate, input gate, output gate). For example, the periodic fluctuations in supplier delivery delays, the transmission effects of market demand mutations (such as promotional activities), and the cumulative impact of multi-node logistics delays.

[0117] Kalman filtering, based on the state space model, real-time corrects and optimizes the prediction results of LSTM. The specific logic is:

[0118] Define inventory level, logistics timeliness, procurement cost, etc. as key state variables;

[0119] The predicted value output by LSTM is used as an observation input, combined with real-time sensor data such as GPS trajectory, warehouse inventory scanning, and known noise statistical characteristics (such as a ±5% fluctuation in supplier capacity), to correct the state estimation.

[0120] Quantify uncertainty by updating error covariance matrix, output predicted confidence interval, support risk classification warning strategy such as low, medium, high, high risk four levels.

[0121] The typical application scenarios of the above hybrid method include:

[0122] Dynamic inventory optimization (LSTM predicts future demand, Kalman filter dynamically adjusts safety stock threshold combined with real-time sales data), supplier risk assessment (LSTM learns supplier historical performance, Kalman filter integrates the impact weight of sudden policy such as tariff adjustment), logistics resilience enhancement (LSTM models transportation network vulnerability, Kalman filter real-time corrects ETA based on typhoon path and other weather disasters).

[0123] The core value of this method is to realize data-model collaboration, real-time adaptability and precise decision support, which can effectively maintain supply stability, quickly improve risk response capability, and dynamically adjust supply chain collaboration strategy based on real-time data, helping to improve equipment overall efficiency (OEE) by 15% to 20%.

[0124] Public cloud, also connected with data storage, is used to provide digital services to small and medium-sized enterprise production and industry ecology. As the core brain and ecological service hub of the entire platform, public cloud greatly reduces the threshold for small and medium-sized enterprises to participate in industry chain collaboration and realize digital transformation with its openness, elasticity, scalability and cost advantage.

[0125] Public cloud integrates deployment of visual model composer, federated learning model scheduling hub, industry chain collaboration engine, production factor transaction engine, resource matching module, multi-tenant manufacturing operation management module and manufacturing knowledge center.

[0126] Visual model composer: supports small and medium-sized enterprises to combine federated learning components (such as equipment life prediction + quality detection) through drag-and-drop method to automatically generate customized models adapted to local data.

[0127] Federated learning model scheduling hub: receives model parameters or gradients obtained by training models locally with self-owned data (or collaborates through distributed architecture such as blockchain), aggregates parameters from all parties to form a global model, and then distributes the updated model to each party, and iterates until the model converges.

[0128] The federated learning model scheduling center dynamically allocates computing resources or provides transaction fee discounts based on the quantified results of participants' contributions and incentive allocation rules, thereby stimulating participants' willingness to participate. The specific contribution evaluation algorithm is as follows:

[0129] Node contribution = Local data volume × Data quality coefficient × Model improvement rate;

[0130] Supply Chain Collaboration Engine: Responsible for handling and scheduling business processes across enterprises, the supply chain collaboration engine integrates proactive risk monitoring capabilities.

[0131] By continuously subscribing to and analyzing real-time OEE data streams from edge computing nodes, the system enables delivery delay warnings. When the OEE of the equipment involved in a production task is detected to be lower than a preset threshold (e.g., lower than 52%), or shows a continuous downward trend within a specific time window (e.g., a daily decrease of more than 5% for 3 consecutive days), the engine will automatically trigger the delivery delay warning mechanism.

[0132] The early warning information will be pushed to relevant demand-side enterprises, managers of the production task, and logistics service providers in real time through platform notifications, emails, or API interfaces, so that all parties can intervene in advance and take countermeasures such as coordinating resources, adjusting production plans, or activating backup suppliers, thereby eliminating potential delivery risks and transforming passive accident response into proactive risk management.

[0133] Resource matching module:

[0134] By leveraging intelligent algorithms, the system provides optimal matching solutions for resource demanders and suppliers in the industrial chain. Specifically, the core of this module is an intelligent decision engine that supports multiple configurable scheduling strategies. It can dynamically generate multi-level plans based on real-time data and drive their execution. When algorithmic scheduling conflicts with human experience, the system can automatically generate multi-scheme simulation comparisons (such as comparing cost changes with delivery delays) to assist managers in making decisions.

[0135] This module consists of two core parts:

[0136] Optimal solution (core engine):

[0137] Input: Receive multi-dimensional data, including production orders (customer demand), incoming material plans (material arrival time and quantity), process route configuration (process path), production scheduling strategy configuration (optimization objectives, such as lowest cost, delivery priority), resource and equipment configuration (capacity constraints), and work calendar (effective time window).

[0138] Processing logic: The strategy of the resource matching module is essentially resource optimization driven by algorithms, achieving the best balance between production efficiency and cost under complex constraints.

[0139] Enterprises can choose or combine different scheduling strategies based on their business characteristics (such as order mode, resource bottleneck, delivery requirements).

[0140] This module optimizes production planning by integrating enterprise resources (equipment, manpower, materials, time, etc.) while meeting constraint conditions. Its core goals include but are not limited to:

[0141] Minimize delivery delays: Ensure timely delivery of orders;

[0142] Maximize resource utilization: Reduce equipment idle time and waiting time;

[0143] Reduce inventory costs: Balance production rhythm and demand, reduce work-in-process backlog;

[0144] Improve production flexibility: Efficiently handle emergency orders, equipment failures, and other unexpected situations.

[0145] To achieve the above goals, the module can implement one or more of the following scheduling strategies:

[0146] Due Date Driven Priority Scheduling Strategy: This strategy prioritizes orders with urgent delivery dates and dynamically sorts them according to rules such as "earliest delivery date first" or "shortest remaining time first." This strategy is suitable for scenarios where customers have strict contractual constraints on delivery dates, or where order delivery dates fluctuate and require dynamic priority adjustment. It can directly respond to customer demand and significantly reduce the risk of overdue delivery.

[0147] Finite Capacity Scheduling Strategy: This strategy first identifies bottleneck resources (such as critical equipment or high-load work centers) in the production process, and then prioritizes task allocation and optimization scheduling around the capacity of the bottleneck, and coordinates upstream and downstream processes. This strategy is suitable for scenarios where there are obvious capacity bottlenecks in the production process, aiming to maximize the output efficiency of bottleneck resources and thus improve the throughput of the entire production line.

[0148] Cost Optimization Scheduling Strategy: This strategy considers production costs (such as labor, energy consumption, equipment depreciation), inventory holding costs, and potential late penalty costs, and generates a scheduling plan through intelligent algorithms to optimize the overall cost structure. This strategy is suitable for complex scenarios that need to balance cost reduction and on-time delivery, or for long-term (such as quarterly or annual) production planning.

[0149] Dynamic Real-Time Scheduling: This strategy dynamically adjusts the production plan based on real-time data streams from IoT, MES, etc. (such as device state changes, new order insertions, material arrival delays, etc.) using a "rolling scheduling" or "event-driven" mechanism. This strategy is suitable for agile manufacturing scenarios where the production environment is highly uncertain and requires rapid response to market changes, with high flexibility and adaptability.

[0150] Hybrid Scheduling: This strategy allows combining the advantages of the above strategies, for example, using bottleneck resource-based scheduling overall, while applying cost-based optimization to non-bottleneck resources; by setting weights for different goals (such as delivery time, cost, efficiency), the model can generate a comprehensive optimal scheduling scheme to flexibly adapt to complex production scenarios with diversified and multi-dimensional constraints.

[0151] To support the implementation of the above strategies, the invention also integrates the following key technologies:

[0152] Theory of Constraints (TOC): used to guide the identification and management of bottleneck resources.

[0153] Mathematical modeling and optimization algorithms: including but not limited to linear programming (LP), integer programming (IP), genetic algorithms, simulated annealing, etc. heuristic algorithms for solving complex scheduling problems.

[0154] Real-time data integration: ensures that the scheduling model can obtain accurate and timely input data.

[0155] Visualization and interactive interface: facilitates manual intervention, adjustment and confirmation by planning personnel.

[0156] In addition to meeting the constraints built into the above strategies, the model will also fully consider a series of dynamic adjustment constraints when solving.

[0157] The model will fully consider a series of dynamic adjustment constraints when solving: demand satisfaction constraints, capacity limitation constraints, delivery time constraints.

[0158] In addition, supply chain resilience constraints are introduced, for example, based on the analysis results of external environment and sudden public events by tools such as Wiseflow, the total amount of orders allocated to "high risk" rated regions is limited to no more than 20% of the total demand, to forcibly implement risk diversification; when the external environment risk level of a region rises to "high risk", the resource matching module can automatically trigger a series of response actions, including dispersing orders to more than 3 suppliers, reserving 15% safety stock, and starting backup logistics routes, etc.

[0159] At the same time, the model also contains emergency order insertion impact constraints. When a high-priority order insertion occurs, the model will treat it as a hard constraint, recalculate the optimal insertion point and resource allocation scheme, and minimize the impact on the existing plan.

[0160] Output: After solving, five layers of refined execution plans and resource monitoring data are generated.

[0161] Plan output module:

[0162] Order planning: Split customer orders into specific delivery time nodes and priorities.

[0163] Resource planning: Generate detailed time occupation plans for resources such as equipment and manpower for capacity pre-occupation.

[0164] Task planning: Output specific production task work orders (including processes, equipment, time, material requirements) to directly drive the manufacturing execution system.

[0165] Material planning: Accurately calculate material requirement lists based on BOM (Bill of Materials) and production progress, and automatically trigger procurement or material requisition processes.

[0166] Resource load: Real-time monitoring of equipment and labor load status and visual display to provide early warning of capacity bottlenecks.

[0167] Multi-tenant manufacturing operation management module: Based on multi-tenant cloud architecture to realize manufacturing operation management function, providing manufacturing operation management service for users using the public cloud.

[0168] Specifically, the core of multi-tenant manufacturing operation management is to provide production and operation management services for multiple independent users (i.e. "tenants") through a standardized multi-tenant manufacturing operation management software instance. In actual operation, multiple tenants share the system's hardware servers, database clusters, network resources, and production modeling engines, data collection interfaces, and other basic software modules, significantly reducing system deployment and maintenance costs.

[0169] At the same time, with the help of tenant ID marking, permission control, and other logical isolation methods, or independent database partitioning and other physical isolation methods, it ensures that different tenants' production data such as work orders and process parameters, as well as business configurations such as BOM structure and process flow, are not accessed or tampered with by other tenants. In addition to achieving the "one system, multiple tenants sharing, data isolation, and on-demand customization" digital management mode, it also meets the data security and compliance requirements.

[0170] Users using public cloud services can use resource matching algorithm models optimized through federated learning to respond to device comprehensive performance, supply chain abnormal event triggered early warning, carry out advanced planning and scheduling work, rely on the deployment of industry chain collaboration engines in the public cloud to send material demand plans and outsourcing manufacturing production plans to suppliers, after confirmation by the suppliers, form corresponding production orders and purchase orders through the deployment of production factor transaction engines in the public cloud, and receive and monitor the progress information and quality information returned by the suppliers, while calling enterprise private domain data such as device comprehensive performance and public cloud federated learning scheduling center for cooperation to participate in model training.

[0171] Production factor transaction engine: Provide a credible, transparent and efficient market for the circulation and transaction of various production factors in the industry chain.

[0172] Specifically, in order to ensure the security and non-tamperability of the transaction, the engine is built based on blockchain technology, and the rules and execution logic of the transaction are encoded in the smart contract for transaction execution. On this platform, the production factors available for transaction include at least one of capacity, materials, computing power, funds, services, carbon emission rights, intellectual property rights and data.

[0173] In order to solve the trust problem in the transaction and give more opportunities to users with excellent production performance, a dynamic credit model is embedded in the smart contract of the engine;

[0174] The model is designed to automatically and trustlessly call and verify the OEE historical data and carbon emission historical data stored in the data. The dynamic credit model calls the OEE, current carbon emission reduction amount and carbon emission reduction percentage of the data storage through the smart contract, dynamically adjusts the credit limit according to the credit scoring formula and industry differentiated weight coefficient, and the formula is as follows:

[0175] Credit score = OEE mean x δ + performance efficiency x a + quality pass rate x β + carbon emission reduction x γ + carbon intensity improvement rate x η

[0176] In the formula, the credit score is the comprehensive evaluation value of the evaluation of the enterprise's performance and credibility, the OEE mean is the average value of the device comprehensive performance in the evaluation period, the performance efficiency is the ratio of the actual production speed to the theoretical speed, the quality pass rate is the ratio of the number of qualified products to the total production quantity, the carbon emission reduction is the total amount of carbon emission reduction in the current period, which reflects the absolute emission reduction contribution, the carbon intensity improvement rate is the percentage(%) of the year-on-year decline of carbon emission per unit of output value, which reflects the relative emission reduction efficiency, δ, a, β are the weight coefficients corresponding to OEE mean, performance efficiency and quality pass rate respectively, γ, η are the weight coefficients of carbon emission reduction and carbon intensity improvement rate, which are set by industry carbon quota policy and platform rules.

[0177] The dynamic credit granting model can adopt industry-differentiated weight coefficient configuration rules. For example, for high energy consumption industries (such as metallurgy), the equipment comprehensive efficiency (OEE) weight δ coefficient can be configured as 0.2, and the carbon emission reduction weight γ coefficient can be configured as 0.5; for high-precision production industries (such as automobile manufacturing), the OEE weight δ coefficient is configured as 0.4, and the carbon emission reduction weight γ coefficient is configured as 0.3. The carbon emission dynamic adjustment mechanism is realized through intelligent contract to automatically calibrate, when the average carbon intensity of the industry decreases (such as 5%), the η coefficient will be automatically increased (such as 0.02); at the same time, the mechanism is linked with the policy, when the local carbon quota is tightened, the γ coefficient will be proportionally floated, when the quota is reduced (such as 10%), the γ coefficient will be adjusted (such as 1.1 times).

[0178] An enterprise that has maintained a high OEE (such as more than 75%) for a long time will have its intelligent contract address automatically marked as "high credit", and can enjoy lower margin ratio or higher transaction limit when conducting capacity transaction.

[0179] On the contrary, enterprises whose OEE is continuously below the warning line (such as 52%) will have their transaction conditions restricted. This directly converts the actual production capacity of the enterprise into digital credit.

[0180] Manufacturing knowledge center: a knowledge base and model base that gathers industry knowledge, expert experience, best practices, process parameters, etc., which can provide intelligent decision support and problem solving solutions for enterprises through knowledge graph, question and answer system, etc.

[0181] Finally, in order to realize the data collaborative sharing of the industrial chain while strictly protecting the business secrets and data sovereignty of each participating enterprise, the platform adopts an advanced federated learning technology framework, under which the edge computing nodes and private clouds are configured as clients of federated learning, for training models locally and generating model parameters.

[0182] In this process, sensitive raw data always belongs to the user who generates the data and will not leave the boundary of the private domain.

[0183] At the same time, the public cloud is deployed with a federated learning model scheduling hub, which is configured as a server of federated learning.

[0184] The task of the server is not to collect any raw data, but to safely receive model parameters uploaded by each client, which do not contain raw data information.

[0185] Subsequently, the server is used to update the global model based on the model parameters uploaded by the clients, and this updating process is completed through a secure aggregation algorithm to ensure that any single participant cannot infer the private model parameters of any other party from the aggregation result.

[0186] Before uploading the model parameters, the client will also use differential privacy and homomorphic encryption and other technologies for processing to ensure higher security.

[0187] Through this collaborative mode of "data not moving, model moving, data available and invisible", the platform can converge the wisdom of all parties in the industry chain under the premise of protecting data privacy, and jointly build a global intelligent model with better performance and stronger generalization ability, and empower the entire industry ecosystem.

[0188] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. An industry chain digitalization collaboration and production factor transaction platform, characterized in that, The system comprises: an edge computing node arranged at the industrial equipment side, which can be used to collect data of the industrial equipment, and can also calculate the comprehensive performance and carbon emission performance data of the equipment in real time based on work orders, plans, quality, and collected equipment running status and running data, and deploy a federated learning model, such as a machine vision quality inspection model, an equipment life prediction model, and an advanced algorithm and strategy for advanced process control, to realize process optimization and real-time control by means of mathematical models, intelligent algorithms, and multivariate control technology; meanwhile, the model cooperates with a public cloud federated learning scheduling hub to generate model parameters through local training; a data storage connected with the edge computing node, the private cloud, and the public cloud, and used to integrate and store data; a private cloud connected with the data storage, and used to provide services to large enterprises; the private cloud can use a resource matching algorithm model optimized by federated learning to respond to early warnings triggered by equipment comprehensive performance and supply chain abnormal events, carry out advanced planning and scheduling work, send material demand plans and production plans for outsourcing manufacturing to suppliers through an industry chain cooperation engine deployed in the public cloud, form corresponding production orders and purchase orders after confirmation by the suppliers, and receive and monitor progress information and quality information returned by the suppliers; meanwhile, the private cloud cooperates with the public cloud federated learning scheduling hub to participate in model training in a federated learning manner by calling enterprise private domain data such as equipment comprehensive performance data. a public cloud connected with the data storage, and used to provide digital services to small and medium-sized enterprise production and operation and industry ecology; users of the public cloud can use a resource matching algorithm model optimized by federated learning to respond to early warnings triggered by equipment comprehensive performance and supply chain abnormal events, carry out advanced planning and scheduling work, send material demand plans and production plans for outsourcing manufacturing to suppliers through an industry chain cooperation engine deployed in the public cloud, form corresponding production orders and purchase orders after confirmation by the suppliers, and receive and monitor progress information and quality information returned by the suppliers; meanwhile, the public cloud cooperates with the public cloud federated learning scheduling hub to participate in model training in a federated learning manner by calling enterprise private domain data such as equipment comprehensive performance data. The edge computing node arranged at the industrial equipment side has communication protocol conversion capability, and is deployed with at least one component selected from the group consisting of a federated learning node, a machine vision quality inspection model, an equipment life prediction model, and an advanced process control algorithm model.

2. The industry chain digitalization collaboration and production factor transaction platform according to claim 1, characterized in that, The data storage comprises: 3.The industrial chain digitalization collaboration and production factor transaction platform according to claim 1, characterized in that, a time series database; a relational database; a non-relational database. The private cloud is deployed with:

4. The industry chain digitalization collaboration and production factor transaction platform according to claim 1, characterized in that, a digital twin engine; advanced planning and scheduling; federated learning. The public cloud is deployed with:

5. The industry chain digitalization collaboration and production factor transaction platform according to claim 1, characterized in that, a visual model composer; a federated learning model scheduling hub; an industry chain cooperation engine; a production factor transaction engine; a resource matching module; a multi-tenant manufacturing operation management module; a manufacturing knowledge center. ​ 6. The industry chain digitalization collaboration and production factor transaction platform according to claim 5, characterized in that, The resource matching module is configured to support and execute one or more optional scheduling strategies and introduce external environmental risk factors and sudden public event factors to optimize supply chain resilience.

7. The industry chain digitalization collaboration and production factor transaction platform according to claim 5, characterized in that, The production factor transaction engine is constructed based on blockchain technology, wherein the smart contract is further configured to construct a dynamic credit model based on the equipment comprehensive performance data and carbon emission performance data.

8. The industry chain digitalization collaboration and production factor transaction platform according to claim 7, characterized in that, The dynamic credit model comprises: a credit score generation unit configured to generate a credit score based on data in a group consisting of the equipment comprehensive performance data and / or carbon emission performance data and an industry differentiated weight coefficient; a smart contract clause matching unit configured to automatically match preset smart contract clauses corresponding to different credit levels according to the credit score. 9.The industrial chain digitalization collaboration and production factor transaction platform according to claim 5, characterized in that, The production factors include at least one of capacity, materials, computing power, funds, services, carbon emission rights, intellectual property rights, and data.

10. The industry chain digitalization collaboration and production factor transaction platform according to claim 1, characterized in that, The edge computing and the private cloud are configured as clients of federated learning for training a model locally and generating model parameters. The public cloud is deployed with a federated learning model scheduling hub configured as a server of federated learning for updating a global model based on the model parameters uploaded by the clients.

Images