Industrial configuration-factory affair management system and method based on data fusion and closed-loop feedback

Abstract

The invention discloses an industrial configuration-factory affair management system and method based on data fusion and closed-loop feedback, and the system comprises a multi-source data fusion layer, a closed-loop collaborative optimization layer, and a configuration control and management execution layer. The multi-source data fusion layer associates data sources of different sources through a protocol conversion and data cleaning module and a feature extraction and semantic association module, establishes a data model and constructs a knowledge graph, and the closed-loop collaborative optimization layer performs model optimization on the constructed unified data model and knowledge graph through a multi-target optimization module. Then verification is carried out through an analogue simulation and execution verification module, an optimization strategy is generated, and the configuration control and management execution layer feeds back an execution result to the multi-target optimization module after execution so as to form a feedback learning closed-loop module; according to the method, the problem of real-time data fusion of multi-protocol equipment is solved, cross-level closed-loop cooperation of control instructions and service decisions is realized, and the problem of multi-target optimization conflicts in a dynamic environment is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial Internet of Things and industrial plant management, and relates to an industrial configuration-plant management system and method based on data fusion and closed-loop feedback. Background Art

[0002] With the continuous development of industrial automation and informatization, the scale of factories is expanding day by day. On the one hand, the types and quantities of industrial equipment involved are increasing continuously, and the production process has become more complex; on the other hand, there are a variety of supporting plant management systems. Traditional management systems only have a single function, such as warehousing management systems, machine monitoring systems, personnel management systems, etc. These systems cannot form collaborative operations and are difficult to meet the needs of modern industrial production.

[0003] Specifically, the industrial configuration system only collects data at the device layer, while the industrial plant management platform focuses on business management, and the protocols of the industrial configuration system and the plant system are heterogeneous, resulting in data fusion delays and serious data islands. Moreover, the existing industrial configuration system can only execute preset instructions for one-way data transfer and control, and cannot achieve real-time closed-loop feedback according to the business changes of the plant management system. For example, when the industrial configuration system device alarms the plant management system, it cannot dynamically adjust the device parameters according to the energy consumption strategy. It can be seen that the existing industrial configuration parameters and business indicators lack a unified mapping model, which is likely to cause insufficient collaborative capabilities. In addition, in traditional solutions, configuration control and management are disjointed, control instructions need to be manually approved, the response time is extremely long, and there is a problem of control lag. Summary of the Invention

[0004] The purpose of the present invention is to solve the semantic conflicts of multi-source data, and a collaborative industrial configuration-plant management system for data acquisition and fusion-dynamic collaborative optimization-closed-loop execution is constructed to achieve real-time linkage between device configuration control and plant management decision-making.

[0005] The technical solution adopted by the present invention is to provide an industrial configuration-plant management system based on data fusion and closed-loop feedback. The key lies in that the above-mentioned industrial configuration-plant management system includes a multi-source data fusion layer, a closed-loop collaborative optimization layer, and a configuration control and management execution layer. The above-mentioned multi-source data fusion layer associates data sources from different sources through a protocol conversion and data cleaning module and a feature extraction and semantic association module, establishes a data model and constructs a knowledge graph. The above-mentioned closed-loop collaborative optimization layer optimizes the constructed unified data model and knowledge graph through a multi-objective optimization module, and then verifies it through a simulation and execution verification module to generate an optimization strategy. The configuration control and management execution layer executes and feeds back the execution result to the multi-objective optimization module to form a feedback learning closed-loop module.

[0006] Specifically, the data sources from different above-mentioned sources include industrial configuration data sources and plant management business data sources.

[0007] Furthermore, the method of the above-mentioned industrial configuration - plant management system specifically includes:

[0008] S1. Access of data sources: The industrial configuration data source and the plant management business data source are accessed to the management system through industrial protocols;

[0009] S2. Protocol conversion and data cleaning: Through a dynamic protocol parsing engine, protocol conversion is carried out using a rule-driven protocol adaptation algorithm; the above-mentioned data cleaning is to eliminate data noise, missing values, and outliers, and output heterogeneous data sources;

[0010] S3. Feature extraction and semantic association: Key data features are extracted from the above-mentioned heterogeneous data sources, and cross-domain semantic associations are established;

[0011] S4. Establish a data model and construct a knowledge graph: Through semantic association and the knowledge graph, the real-time operation data of the configuration system is associated with the data source of the plant management system business process flow to establish an association model;

[0012] S5. Target optimization: It includes establishing a target, modeling, and optimization solving, and outputting an optimization strategy;

[0013] S6. Simulation and execution verification: Using digital twin technology, a virtual operation environment is established to verify the feasibility and effectiveness of the multi-objective optimization strategy, and the verified optimization strategy is issued;

[0014] S7. Form a feedback learning closed loop: The execution results after the verified optimization strategy is issued are collected in real time, and the model parameters and optimization target weights are dynamically adjusted through reinforcement learning to form a closed-loop self-evolution mechanism of optimization - execution - feedback - iteration;

[0015] S8. Industrial configuration control execution: Convert the optimization strategy into PLC-executable instructions to adjust the device operation parameters in real time;

[0016] S9. Execution of plant management business processes: Automatically trigger plant management business processes.

[0017] Specifically, the industrial protocol in step S1 includes any one of the Modbus protocol, OPC UA protocol, MQTT protocol, and HTTP protocol.

[0018] Specifically, the data cleaning in step S2 includes:

[0019] S2-2-1. Data pre-cleaning: The data is pre-cleaned by sliding window noise reduction and invalid value removal;

[0020] S2-2-2, Rule Verification and Cleaning: Establish cleaning rules through the built-in expert rule library to perform rule verification and cleaning on the data;

[0021] S2-2-3, Enhanced Cleaning: Analyze latent anomalies in the data by establishing an anomaly detection model;

[0022] S2-2-4, Data Standardization Processing and Output: Align and unify the cleaned data and output heterogeneous data sources.

[0023] Furthermore, the feature extraction in step S3 is to extract feature data features with high information density and low redundancy; semantic association is to establish a semantic relationship network across devices and systems, establish an expression mapping of data - event - knowledge, and feedback the semantic association results to the feature extraction module to dynamically adjust the feature weights.

[0024] Even further, in step S5, the target establishment is to establish targets in the dimensions of energy efficiency, production efficiency, equipment status, and cost in the industrial configuration and plant management scenarios; the modeling is to construct an objective function based on the key features output by the feature extraction layer.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] The present invention solves the problem of real-time data fusion of multi-protocol devices, realizes the cross-level closed-loop collaboration of control instructions and business decisions, and solves the problem of multi-objective optimization conflicts in a dynamic environment. Through the constructed "data acquisition and fusion - dynamic collaborative optimization - closed-loop execution" industrial configuration - plant management collaborative system, the real-time linkage between device configuration control and plant management decision-making is realized. Brief Description of the Drawings

[0027] Figure 1 is a framework diagram of the industrial configuration - plant management system of the present invention. Detailed Embodiments

[0028] The technical solutions in the embodiments of the present invention will be described clearly and completely below. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0029] Embodiment 1

[0030] The industrial configuration - plant management system in this embodiment includes a multi-source data fusion layer, a closed-loop collaborative optimization layer, and a configuration control and management execution layer.

[0031] Among them, the multi-source data fusion layer includes a protocol conversion and data cleaning module, a feature extraction and semantic association module, a unified data model, and a knowledge graph; the closed-loop collaborative optimization layer includes a multi-objective optimization module for model optimization and a simulation and execution verification module, which outputs optimization strategies to the configuration control and management execution layer; the configuration control and management execution layer includes an industrial configuration execution module and a plant management execution module, and after execution, it feeds back the execution results to the multi-objective optimization module to form a feedback learning closed-loop module.

[0032] The specific method for managing using the industrial configuration-plant management system of this embodiment is as follows:

[0033] S1. Access of data sources:

[0034] Industrial configuration data sources and plant management business data sources are accessed to the management system through their respective different industrial protocols. Among them, the industrial protocols include any one of Modbus protocol, OPC UA protocol, MQTT protocol, and HTTP protocol;

[0035] Industrial configuration data sources may include device operation parameters collected in real time by controllers such as PLC / DDC, and sensor data of various types such as temperature, pressure, flow rate, liquid level, current, voltage, power, and frequency;

[0036] Plant management business data may include various plant management-related information such as asset information, point inspection information, maintenance or repair information, work order information, energy information, production information, and performance information.

[0037] S2. Protocol conversion and data cleaning:

[0038] In order to achieve seamless communication and data quality of heterogeneous devices and systems, and thus solve the problem of mixing various protocols such as Modbus protocol, OPC UA protocol, MQTT protocol, and HTTP protocol in the industrial field, protocol conversion and data cleaning are required.

[0039] S2-1. Protocol conversion is carried out through a dynamic protocol parsing engine, using a rule-driven protocol adaptation algorithm for protocol conversion, such as: Modbus register address mapping table, OPC UA node tree structure, etc., and at the same time, protocol extension can be customized; a real-time protocol adaptation mechanism is adopted, using a rule-driven protocol adaptation algorithm to achieve dynamic protocol switching.

[0040] S2-2. Data cleaning is to eliminate data noise, missing values, and outliers, and output heterogeneous data sources to ensure that the data quality meets high-precision requirements;

[0041] Specifically, it includes:

[0042] S2-2-1, Data pre-cleaning: Pre-clean the data by means of sliding window noise reduction and invalid value elimination;

[0043] S2-2-2, Rule verification cleaning: Establish cleaning rules through the built-in expert rule library to perform rule verification cleaning on the data to ensure the reliability of the data source;

[0044] S2-2-3, Enhanced cleaning: Analyze hidden anomalies in the data by establishing an anomaly detection model;

[0045] S2-2-4, Data standardization processing and output: Align and unify the cleaned data and output heterogeneous data sources.

[0046] S3, Feature extraction and semantic association: In this step, extract the key features of the data from the above-mentioned heterogeneous data sources and establish cross-domain semantic associations.

[0047] S3-1, Feature extraction:

[0048] Extract feature data features with high information density and low redundancy to adapt to the requirements of different business scenarios; for high-density feature data, build a sensitivity factor library, and use the data of the change factors that can significantly affect a certain indicator as high-density feature data. For example, among the factors affecting equipment health, the vibration frequency has a higher data density than the ambient temperature; low redundancy means that the same type of sensitive factors should not be recorded repeatedly. For example, total energy consumption and sub-item energy consumption do not need to be extracted simultaneously.

[0049] S3-2, Semantic association: Establish a semantic relationship network across devices and systems, establish a unified expression mapping of data - event - knowledge, and synchronously feedback the semantic association results to the feature extraction module to dynamically adjust the feature weights.

[0050] S4, Establish a data model and construct a knowledge graph: Through semantic association and knowledge graph, associate the real-time operation data of the configuration system with the business process management data source of the plant service system, and establish a multi-dimensional association model, such as the device - energy - fault - work order association model;

[0051] S5, Target optimization: Include establishing goals, modeling, and optimization solving, and output optimization strategies.

[0052] S5-1, Goal establishment: In the industrial configuration and plant service management scenarios, establish goals in the energy efficiency dimension, production efficiency dimension, equipment status dimension, and cost dimension, such as goals in different dimensions such as energy efficiency goals, production efficiency, equipment health, and cost goals;

[0053] S5-2, Modeling: Based on the key features output by the feature extraction layer, such as equipment energy efficiency index and health score, construct an objective function;

[0054] S5-3. Optimized solution: Output the multi-objective optimal strategy set.

[0055] S6. Simulation and execution verification:

[0056] Adopt digital twin technology to establish a virtual operation environment, verify the feasibility and effectiveness of the multi-objective optimization strategy, reduce the trial-and-error risk of the physical system, ensure the safe and stable operation of industrial configuration and plant management, and issue the verified optimization strategy.

[0057] S7. Form a feedback learning closed loop:

[0058] Real-time collect the execution results after the verified optimization strategy is issued, and dynamically adjust the model parameters and optimization target weights through reinforcement learning to form a closed-loop self-evolution mechanism of optimization - execution - feedback - iteration, continuously improving the system performance.

[0059] S8. Industrial configuration control execution: Convert the optimization strategy into PLC-executable instructions to adjust the device operation parameters in real time;

[0060] S9. Execution of plant management business processes: Automatically trigger plant management business processes, such as energy scheduling instructions, preventive maintenance work orders, etc.

[0061] Embodiment 2

[0062] Taking a certain lithium battery factory project as an example, further illustrate how to use the industrial configuration - plant management system of the present invention for multi-objective optimization of the factory air compressor group.

[0063] S1. Access to data sources:

[0064] The industrial configuration data sources in this embodiment include sensor data such as the real-time air supply volume (m 3 / min), air supply pressure (Mpa), current (A), outlet temperature (°C), vibration frequency (Hz), etc. of the air compressor group. The protocols involved are Modbus RTU protocol, OPC UA protocol, and MQTT protocol.

[0065] The plant management business data includes information such as peak-valley electricity prices (yuan / kWh), energy consumption quota indicators (kWh), and production line air pressure requirements (Mpa). The protocols involved are HTTP protocol and OPC protocol.

[0066] S2. Protocol conversion and data cleaning:

[0067] S2-1. Protocol conversion is carried out through a dynamic protocol parsing engine, using a rule-driven protocol adaptation algorithm for protocol conversion. In this embodiment, when the network bandwidth < 50 Mbps, the protocol engine automatically switches to the Modbus RTU mode; when vibration spectrum data needs to be transmitted, it switches to the OPC UA protocol.

[0068] S2-2. Data cleaning is to eliminate data noise, missing values, and outliers, output heterogeneous data sources, and ensure that the data quality meets high-precision requirements;

[0069] Specifically, it includes:

[0070] S2-2-1. Data pre-cleaning: Pre-clean the data by means of sliding window noise reduction and invalid value elimination. In this embodiment, median filtering is used for the air compressor gas supply, supply pressure, outlet temperature, operating current, and vibration frequency to identify data that significantly deviates, and eliminate short-term data fluctuations and outliers;

[0071] S2-2-2. Rule verification and cleaning: Establish cleaning rules through a built-in expert rule library to perform rule verification and cleaning on the data to ensure the reliability of the data source. In this embodiment, an alarm threshold rule library is set. For example, when the current exceeds 110% of the rated value and lasts for more than 10 seconds, it is marked as an overload warning;

[0072] S2-2-3. Enhanced cleaning: Establish an anomaly detection model to perform latent anomaly analysis on the data. In this embodiment, an exhaust pressure LSTM prediction model is established to roll-predict the pressure value within the next 5 seconds. When the predicted pressure value exceeds the pressure demand range, it is marked as a latent anomaly to trigger a latent alarm;

[0073] S2-2-4. Data standardization processing and output: Align and unify the cleaned data, and output heterogeneous data sources. In this embodiment, an interpolation method is used to align the sampling data of the sensors.

[0074] S3. Feature extraction and semantic association: In this step, key data features are extracted from the above-mentioned heterogeneous data sources to establish cross-domain semantic associations.

[0075] S3-1. Feature extraction:

[0076] Extract feature data features with high information density and low redundancy to adapt to different business scenario requirements; In this embodiment, 3 important features of the air compressor system are extracted: energy efficiency features (such as: air compressor power Kw / (m 3 / min), unit gas production energy consumption (kWh) / m 3 ); health features (such as: vibration frequency overlimit value); business features (such as: electricity cost, production line gas demand fluctuation).

[0077] S3-2, Semantic Association: Establish a semantic relationship network across devices and systems, establish a unified expression mapping of data - event - knowledge, and synchronously feedback the semantic association results to the feature extraction module to dynamically adjust the feature weights. In this embodiment, an association network of air compressor - energy consumption quota - production line pressure demand - electricity price is established, and the weights of features are dynamically set. For example, when the electricity price is at a peak, the weight coefficient of the energy consumption feature is increased.

[0078] S4, Establish a data model and construct a knowledge graph: Through semantic association and the knowledge graph, associate the real-time operation data of the configuration system with the data sources of the plant facility system business process management, and establish a multi-dimensional association model; In this embodiment, an association model of air compressor equipment - energy - fault - production line demand - maintenance work order - electricity price strategy is established;

[0079] S5, Target Optimization: Include establishing goals, modeling, and optimization solving, and output optimization strategies.

[0080] S5-1, Goal Establishment: In the industrial configuration and plant facility management scenarios, establish goals in the energy efficiency dimension, production efficiency dimension, equipment status dimension, and cost dimension simultaneously;

[0081] S5-2, Modeling: Based on key features such as the equipment energy efficiency index and health score output by the feature extraction layer, construct an objective function; In this embodiment, a three-dimensional objective function of the energy efficiency goal, equipment health, and production line demand of the air compressor is established, and the weight coefficients are dynamically set. For example, the weight coefficient of the energy efficiency goal is 0.5, the weight coefficient of equipment health is 0.3, and the weight coefficient of production line demand is 0.2.

[0082] S5-3, Optimization Solving: Output a multi-objective optimal strategy set. In this embodiment, deep reinforcement learning is used to generate strategies: According to the multi-dimensional objective weights in S5-2, calculate the optimal set value of the air supply pressure. Supplying air at this pressure can achieve the strategy of giving priority to energy conservation.

[0083] S6, Simulation and Execution Verification:

[0084] Adopt digital twin technology to establish a virtual operation environment, verify the feasibility and effectiveness of the multi-objective optimization strategy, reduce the risk of trial and error in the physical system, ensure the safety and stability of industrial configuration and plant facility management, and issue the verified optimization strategy; In this embodiment, a three-dimensional configuration model of the air compressor system is established and parameter matching is performed. According to the air supply pressure value calculated in S5-3, simulate the system operation to verify the feasibility of the strategy.

[0085] S7, Form a feedback learning loop:

[0086] Collect the execution results after the optimized strategy is verified in real time, and dynamically adjust the model parameters and the weights of the optimization objectives through reinforcement learning to form a closed-loop self-evolution mechanism of optimization-execution-feedback-iteration, continuously improving the system performance; in this embodiment, according to the operation data and the energy-saving deviation value fed back by the air compressor system in real time, it is fed back to the optimization model in real time for model parameter adjustment.

[0087] S8. Industrial configuration control execution: Convert the optimized strategy into PLC-executable instructions to adjust the device operation parameters in real time; in this embodiment, the optimized air supply pressure strategy is sent to the PLC for execution.

[0088] S9. Execution of the plant management business process: Automatically trigger the plant management business process, such as energy scheduling instructions, preventive maintenance work orders, etc.; in this embodiment, the work order process is automatically triggered. When the issued strategy triggers the low limit of the device health score, the plant management system generates a maintenance work order and assigns it to the operation and maintenance personnel set in the system.

Claims

1. An industrial configuration - plant management system based on data fusion and closed - loop feedback, characterized in that The industrial configuration - plant management system described above includes a multi-source data fusion layer, a closed-loop collaborative optimization layer, and a configuration control and management execution layer. The multi-source data fusion layer associates data sources from different origins through a protocol conversion and data cleaning module and a feature extraction and semantic association module, establishes a data model, and constructs a knowledge graph. The closed-loop collaborative optimization layer optimizes the constructed unified data model and knowledge graph through a multi-objective optimization module, and then verifies it through a simulation and execution verification module to generate an optimization strategy. The configuration control and management execution layer feeds back the execution result to the multi-objective optimization module after execution to form a feedback learning closed-loop module.

2. The industrial configuration - plant facility management system based on data fusion and closed - loop feedback according to claim 1, wherein The data sources from different origins include industrial configuration data sources and plant management business data sources.

3. The method of the industrial configuration - plant management system based on data fusion and closed - loop feedback according to claim 1, characterized in that, The method of the industrial configuration - plant management system specifically includes: S1. Access of data sources: Industrial configuration data sources and plant management business data sources are accessed to the management system through industrial protocols. S2. Protocol conversion and data cleaning: Through a dynamic protocol parsing engine, protocol conversion is performed using a rule-driven protocol adaptation algorithm; the data cleaning is to eliminate data noise, missing values, and outliers, and output heterogeneous data sources. S3. Feature extraction and semantic association: Key data features are extracted from the heterogeneous data sources to establish cross-domain semantic associations. S4. Establish a data model and construct a knowledge graph: Through semantic associations and knowledge graphs, the real-time operation data of the configuration system is associated with the data sources of the plant management system business process management to establish an association model. S5. Objective optimization: Includes establishing objectives, modeling, and optimization solving, and outputting an optimization strategy. S6. Simulation and execution verification: Using digital twin technology, a virtual operation environment is established to verify the feasibility and effectiveness of the multi-objective optimization strategy, and the verified optimization strategy is issued. S7. Form a feedback learning closed-loop: The execution results after the verified optimization strategy is issued are collected in real time, and the model parameters and optimization objective weights are dynamically adjusted through reinforcement learning to form a closed-loop self-evolution mechanism of optimization - execution - feedback - iteration. S8. Industrial configuration control execution: Convert the optimization strategy into PLC-executable instructions to adjust the device operation parameters in real time. S9. Execution of plant management business processes: Automatically trigger plant management business processes.

4. The method of the industrial configuration - plant management system based on data fusion and closed - loop feedback according to claim 3, wherein, The industrial protocol in step S1 includes any one of Modbus protocol, OPC UA protocol, MQTT protocol, and HTTP protocol.

5. The method of the industrial configuration - plant management system based on data fusion and closed - loop feedback according to claim 3, wherein The data cleaning in step S2 includes: S2-2-1. Data pre-cleaning: The data is pre-cleaned through sliding window noise reduction and invalid value elimination. S2-2-2. Rule verification cleaning: Through a built-in expert rule library, cleaning rules are established to perform rule verification cleaning on the data. S2-2-3. Enhanced cleaning: Through establishing an anomaly detection model, latent anomaly analysis is performed on the data. S2-2-4. Data standardization processing and output: The cleaned data is aligned and unified, and heterogeneous data sources are output.

6. The method of the industrial configuration-plant management system based on data fusion and closed-loop feedback according to claim 3, characterized in that, The feature extraction in step S3 is to extract feature data features with high information density and low redundancy. Semantic association is to establish a semantic relationship network across devices and systems, establish an expression mapping of data-events-knowledge, and feedback the semantic association results to the feature extraction module to dynamically adjust the feature weights.

7. The method of the industrial configuration - plant management system based on data fusion and closed - loop feedback according to claim 3, wherein In step S5, target establishment is to establish targets in the dimensions of energy efficiency, production efficiency, equipment status, and cost in the industrial configuration and plant management scenarios simultaneously. Modeling is to construct an objective function based on the key features output by the feature extraction layer.

Images