Industrial control method and system based on anthropomorphized thought chains and hybrid expert models

Abstract

This invention provides an industrial control method and system based on anthropomorphic thinking chain and hybrid expert model, including: Step S1: Based on the operator's thinking decision points and decision chain when controlling the equipment, construct a decision model based on anthropomorphic thinking chain and multi-layer hybrid expert model for each process segment in the complete production chain; Step S2: Train the decision model of each process segment using historical data to obtain the trained decision model; Step S3: Use the trained decision model to obtain the control setpoint based on the real-time sensing data of the corresponding process segment.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to an industrial control method and system based on anthropomorphic thought chains and hybrid expert models, and more specifically to a method for constructing a real-time control intelligent agent for industrial production equipment based on anthropomorphic thought chains and multi-layer hybrid expert models. Background Technology

[0002] In the field of industrial automation control, traditional control methods often rely on fixed algorithms or single models, making it difficult to accurately simulate the thought and decision-making processes of human operators and adapt to the complex and ever-changing production process requirements. Existing technologies often lack a deep understanding of operator decision-making logic and operational procedures, resulting in insufficient control precision, poor adaptability to changing operating conditions, and inadequate collaborative control across multiple process stages. These shortcomings make it difficult to meet the demands of modern industrial production for efficient and precise control. Therefore, there is an urgent need for an industrial control method that integrates anthropomorphic thinking with efficient modeling techniques.

[0003] Patent document CN117112231A (application number: 202311227940.X) discloses a multi-model collaborative processing method and apparatus. The method includes creating a front-end model tree in response to a user operation; processing the front-end model tree to obtain a logical model tree; processing the logical model tree to obtain a multi-level model execution queue; and sequentially executing the model execution queues in the multi-level model execution queue to obtain a collaborative processing result. This prior art provides a multi-model scheduling execution method, which aims to make full use of computing resources and accelerate model execution speed. This is different from the goal and effect of multi-level multi-expert model collaboration used in this invention to achieve real-time automatic control comparable to high-level and highly complex manual operations.

[0004] Patent document CN116866381A (application number: 202310815908.7) discloses a method for collaborative operation between industrial equipment, applied to an edge cloud server. The method includes: receiving a collaborative operation request from upstream industrial equipment, the request including control requirements; establishing network communication with downstream industrial equipment based on the collaborative operation request to obtain data uploaded by the downstream industrial equipment; and configuring and starting a target industrial equipment that meets the control requirements based on the business process and the data uploaded by the downstream industrial equipment, thereby achieving collaborative operation between the upstream, downstream, and target industrial equipment. This prior art provides a generalized method for collaboration between industrial equipment in a production process, but it is not a standard collaborative framework, nor does it reflect anthropomorphic modeling of industrial control decision-making elements and thought processes. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the purpose of this invention is to provide an industrial control method and system based on anthropomorphic thinking chain and hybrid expert model.

[0006] An industrial control method based on anthropomorphic thought chain and hybrid expert model provided by the present invention includes: Step S1: Based on the operator's thought decision points and decision chain when controlling the equipment, construct a decision model based on anthropomorphic thought chain and multi-layered hybrid experts for each process segment in the complete production chain; Step S2: Use historical data to train the decision model for each process segment to obtain the trained decision model; Step S3: Using the trained decision model, obtain the control setpoint based on the real-time sensing data of the corresponding process section.

[0007] Preferably, the decision-making model based on anthropomorphic thinking chain and multi-layered hybrid experts is a multi-layered network, and the composition level of the multi-layered network is determined based on the macro-decision logic and overall operation process of human operators in industrial control. Based on the micro-level thinking and decision-making sequence and specific logical connections of human operators in specific process control, the topology of each network layer is determined.

[0008] Preferably, the multi-layer network includes one or more of the following: a preprocessing layer, a pre-constraint layer, a control pre-computation layer, a comprehensive optimization layer, and a real-time output layer; The preprocessing layer is used to preprocess the real-time sensing data of the process section. The pre-constraint layer is used to fine-tune the objectives and constraints under certain process control requirements; The control pre-computation layer is used to calculate the optimal setpoint for each control point based on the preprocessed sensing data and the fine-tuned targets and constraints. The comprehensive optimization layer is used to comprehensively process the calculated optimal setpoints of each control point so that the setpoints of each control point meet the overall goal of the current process segment. The real-time output layer is used to send the set values ​​of each control point after comprehensive processing to the corresponding device or preset interface.

[0009] Preferably, each layer of the network is a model network with a serial, parallel, or nested topology composed of multiple sub-models; the topology of each layer of the network maps to the operator's decision-making link. Each layer of the network is a directed acyclic graph (DAG) structure. The DAG structure is used to allow data to flow through different paths within the current layer according to requirements, thereby achieving adaptive processing for different working conditions.

[0010] Preferably, the sensing data in step S3 includes: production plan data, material specification data, material status data, relevant sensor data, relevant production equipment status data, process specifications, and production line design data; The control settings include any one or more of the following: speed, rotational speed, temperature, position, tension, pressure, power, frequency, flow rate, flow volume, (valve and other devices) opening degree, mode, etc.

[0011] An industrial control system based on anthropomorphic thought chain and hybrid expert model provided by the present invention includes: Module M1: Based on the operator's thought decision points and decision chain when controlling equipment, a decision model based on anthropomorphic thought chain and multi-layered hybrid experts is built for each process segment in the complete production chain; Module M2: Use historical data to train the decision model for each process segment to obtain the trained decision model; Module M3: Utilizes the trained decision model to obtain control setpoints based on real-time sensing data from the corresponding process section.

[0012] Preferably, the decision-making model based on anthropomorphic thinking chain and multi-layered hybrid experts is a multi-layered network, and the composition level of the multi-layered network is determined based on the macro-decision logic and overall operation process of human operators in industrial control. Based on the micro-level thinking and decision-making sequence and specific logical connections of human operators in specific process control, the topology of each network layer is determined.

[0013] Preferably, the multi-layer network includes one or more of the following: a preprocessing layer, a pre-constraint layer, a control pre-computation layer, a comprehensive optimization layer, and a real-time output layer; The preprocessing layer is used to preprocess the real-time sensing data of the process section. The pre-constraint layer is used to fine-tune the objectives and constraints under certain process control requirements; The control pre-computation layer is used to calculate the optimal setpoint for each control point based on the preprocessed sensing data and the fine-tuned targets and constraints. The comprehensive optimization layer is used to comprehensively process the calculated optimal setpoints of each control point so that the setpoints of each control point meet the overall goal of the current process segment. The real-time output layer is used to send the set values ​​of each control point after comprehensive processing to the corresponding device or preset interface.

[0014] Preferably, each layer of the network is a model network with a serial, parallel, or nested topology composed of multiple sub-models; the topology of each layer of the network maps to the operator's decision-making link. Each layer of the network is a directed acyclic graph (DAG) structure. The DAG structure is used to allow data to flow through different paths within the current layer according to requirements, thereby achieving adaptive processing for different working conditions.

[0015] Preferably, the sensing data in module M3 includes: production plan data, material specification data, material status data, relevant sensor data, relevant production equipment status data, process specifications, and production line design data; The control settings include any one or more of the following: speed, rotational speed, temperature, position, tension, pressure, power, frequency, flow rate, flow volume, (valve and other devices) opening degree, mode, etc.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention uses a multi-layer network design for the decision-making model to determine the network hierarchy based on the macro-decision logic and overall operation process of human operators, and combines the micro-thinking decision-making sequence and logical association method to determine the topology within the layer. This enables the model to accurately map the operator's decision-making link, simulate the flexible decision-making thinking of human operators, effectively solve the problems of insufficient control accuracy and poor adaptability to changing working conditions in existing models, and adapt to the personalized control needs of different process sections. 2. This invention breaks through the limitations of the existing technology that simply superimposes anthropomorphic thinking and hybrid expert models. It integrates the anthropomorphic thinking chain throughout the entire model building process, and the multi-layered hybrid expert model provides efficient modeling support. It retains the rich control experience and decision-making logic of human operators, while giving full play to the model's efficient computing and autonomous learning capabilities, thus taking into account both the flexibility and efficiency of control. 3. This invention constructs a dedicated decision model for each process segment in the complete production chain, and trains the model by combining real-time sensing data and historical data of each process segment. It can accurately output control setpoints that are adapted to each process segment, avoid the limitation of a single model adapting to multiple process segments, and improve the collaborative control level of the entire production chain. 4. The decision-making model of this invention can autonomously output control setpoints based on real-time sensing data, simulate the decision-making process of human operators, reduce human intervention, reduce dependence on experienced operators, simplify control operation procedures, improve the automation level and stability of industrial control, and reduce human error in the production process. 5. The multi-layer network of this invention can flexibly select combinations of layers such as preprocessing layers and pre-constraint layers. The topology within the layers can adopt various forms such as serial, parallel, and nested, which can adapt to the control needs of different industrial fields and different process complexities. It has a wide range of application scenarios and promotion value. Moreover, the model can be continuously trained and optimized through historical data, and the control effect can be further improved in long-term use. Attached Figure Description

[0017] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 A schematic diagram of the construction system for a real-time control intelligent agent for industrial production equipment.

[0018] Figure 2 This is a schematic diagram of an anthropomorphic thought chain and a multi-layered hybrid expert model structure. Detailed Implementation

[0019] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0020] Example 1 An industrial control method based on anthropomorphic thought chain and hybrid expert model provided by the present invention includes: Step S1: Based on the operator's thought decision points and decision chain when controlling the equipment, construct a decision model based on anthropomorphic thought chain and multi-layered hybrid experts for each process segment in the complete production chain; Specifically, the decision-making model based on anthropomorphic thinking chain and multi-layered hybrid experts is a multi-layered network. The composition level of the multi-layered network is determined based on the macro-decision logic and overall operation process of human operators in industrial control. Based on the micro-thinking decision-making sequence and specific logical connections of human operators in specific process control, the topology of each network layer is determined. The multi-layer network includes one or more of the following: a preprocessing layer, a pre-constraint layer, a control pre-computation layer, a comprehensive optimization layer, and a real-time output layer; The preprocessing layer is used to preprocess the real-time sensing data of the process section. The pre-constraint layer is used to fine-tune the objectives and constraints under certain process control requirements; The control pre-computation layer is used to calculate the optimal setpoint for each control point based on the preprocessed sensing data and the fine-tuned targets and constraints. The comprehensive optimization layer is used to comprehensively process the calculated optimal setpoints of each control point so that the setpoints of each control point meet the overall goal of the current process segment. The real-time output layer is used to send the set values ​​of each control point after comprehensive processing to the corresponding device or preset interface.

[0021] Each layer of the network is a model network with a serial, parallel, or nested topology composed of multiple sub-models; the topology of each layer of the network maps to the operator's decision-making link.

[0022] Step S2: Use historical data to train the decision model for each process segment to obtain the trained decision model; Step S3: Using the trained decision model, obtain the control setpoint based on the real-time sensing data of the corresponding process section.

[0023] The sensing data includes: production plan data, material specification data, material status data, relevant sensor data, relevant production equipment status data, process specifications, and production line design data. The control settings include any one or more of the following: speed, rotational speed, temperature, position, tension, pressure, power, frequency, flow rate, flow volume, (valve and other devices) opening degree, mode, etc.

[0024] The present invention also provides an industrial control system based on anthropomorphic thinking chain and hybrid expert model. The industrial control system based on anthropomorphic thinking chain and hybrid expert model can be implemented by executing the process steps of the industrial control method based on anthropomorphic thinking chain and hybrid expert model. That is, those skilled in the art can understand the industrial control method based on anthropomorphic thinking chain and hybrid expert model as a preferred embodiment of the industrial control system based on anthropomorphic thinking chain and hybrid expert model.

[0025] This embodiment provides a method for real-time control of industrial production equipment. It simulates the decision-making process of a human operator and utilizes a multi-layered hybrid expert decision-making model. The perception layer consists of the digital systems and knowledge that the human operator needs to monitor and understand; the decision layer is a multi-layered routing hybrid expert model network; and the control layer consists of the industrial equipment and system points that the human operator needs to digitally control. This embodiment can completely replace human operator operation in the designed scenario, achieving "Level 4 autonomous driving" in industrial control scenarios.

[0026] Example 2 Example 2 is a preferred example of Example 1. According to the embodiment of this invention, an industrial control method and system based on anthropomorphic thinking chain and hybrid expert model is provided, including: taking the control thinking and decision-making process of human operators as a reference, collecting and sensing all key human decision-making elements of the real-time control scenario of production equipment, and using anthropomorphic thinking chain and multi-layer hybrid expert model to perform real-time calculation and decision-making on the aforementioned decision-making elements, so as to obtain a decision-making level that is comparable to or even exceeds that of excellent human operators for the control settings of related equipment, thereby realizing high-quality real-time control of related equipment in normal working conditions, replacing human operators.

[0027] This embodiment will take the control scenario of a continuous annealing furnace for cold rolling of carbon steel as an example to illustrate the specific steps of the method in detail. It should be noted that the continuous annealing furnace is a key piece of equipment used to perform complex heat treatment on multiple coils of steel strip welded together in series to produce cold-rolled steel strips (coils) of different grades. Its control objective is to maximize production efficiency (including maximizing capacity, reducing material waste, and reducing the defect rate) while ensuring the quality of each coil of steel strip (such as annealing temperature, material strength, etc.) and production safety.

[0028] The industrial control method based on anthropomorphic thought chains and hybrid expert models includes: Step 1: Determine the list of decision-making factors that human operators rely on during the production process. These typically include, but are not limited to, production planning data, material specification data, material status data (e.g., temperature, speed, location), relevant sensor data, relevant production equipment status data, process procedures, production line design data, etc., and digitally collect and store these factors. Through in-depth interviews with experienced operators, analysis of standard operating procedures, and long-term observation of their actual operations, it can be determined that the decision-making data list in the continuous annealing furnace control scenario includes at least the following types: Production planning data typically originates from Manufacturing Execution Systems (MES) or Enterprise Resource Planning (ERP) systems, and includes production sequence and material specifications. Specifically, this may include: production sequence: including the sequence of steel coils currently being produced on the production line, and the planned sequence of steel coils to be produced; material specification data, such as: strip entry / exit coil numbers; length, thickness, and width of the entry coil; length, thickness, and width of the contracted coil; contract number; annealing curve type; element content; tapping mark; steel grade, etc.

[0029] Material location data is the real-time status of the steel strip's position tracking after it enters the production line. Specifically, it can include: the position of the steel strip joint and the steel strip's position tracking information.

[0030] Sensor data is the most direct source for operators to perceive the internal status of the equipment. It is usually provided by a distributed control system or a data acquisition and monitoring control system. Specifically, it may include: temperature setting parameters for multiple furnace sections inside the continuous annealing furnace, such as: temperature setpoints for each section, actual temperatures for each section, actual furnace temperatures for each section, temperatures of each radiant tube, temperature setting modes for each section, combustion power for each section, fan speed, fan on / off status, etc.; the running speed of the steel strip inside the continuous annealing furnace; tension setting parameters for each section of the steel strip, such as: tension setpoints for each section, tension adjustment amounts for each section, actual tension values ​​for each section, tension setting modes for each section, etc.; and parameters related to the straightening rollers, such as: the position of the hydraulic cylinders and the position of the steel strip for each straightening roller.

[0031] Process specifications and production line design data are relatively static but extremely important information, forming the basic framework of operations. For example, the annealing curve type specified in the process specification describes the upper and lower temperature limits of the strip in each furnace section, as well as the target temperature. Production line design data includes physical parameters such as the width of the continuous annealing furnace, the strip length in each furnace section, the location of each piece of equipment in the entire production line, and the upper and lower limits of each set value specification.

[0032] All the decision-making data mentioned above can be collected in real-time or periodically from systems such as Manufacturing Execution Systems (MES), Distributed Control Systems (DCS), and Laboratory Information Management Systems (LISS) by deploying a data acquisition gateway and using an open platform communication architecture and standard industrial protocols such as Modbus. The collected data is assigned a precise timestamp and stored in a structured manner in one or more databases, providing a data source for subsequent model training and real-time inference.

[0033] Step 2: Determine the list of control settings that human operators input through the human-machine interface (HMI or MMI) during the control process. These settings typically include, but are not limited to, speed, temperature, position, tension, pressure, and mode. The values ​​of these control settings are then digitally acquired and stored. In the current continuous annealing furnace scenario, the control point setpoint data input by the operator through the human-machine interface mainly includes: Speed ​​setting: The running speed of the steel strip in the continuous annealing furnace (unit: mpm).

[0034] Temperature setting: Temperature setting for each furnace section (unit: °C).

[0035] Tension setting value: The tension of the steel strip in each furnace section (unit: kN).

[0036] Tension adjustment amount: The amount of adjustment (unit: %) based on the tension set value.

[0037] Each change to these control point settings is captured by the data acquisition system and aligned with the decision element data with the same timestamp collected in step S1 to form an (input, output) dataset. Each data record can be understood as "at a certain moment, when the operator observes a series of decision elements (inputs), he makes a specific set of control point settings (outputs)".

[0038] Step 3: Referring to the thought processes and decision-making chains of human operators when controlling the equipment, map them into a decision-making model based on anthropomorphic thought chains and multi-layered hybrid experts. The input of this model network is the data from Step 1, and the output is the data from Step 2. The model consists of multiple sub-models forming a multi-layered network. The topology of this network maps to the decision-making chains of human operators, and its structure can be standardized into 5 layers: (1) preprocessing layer, (2) pre-constraint layer, (3) control pre-computation layer, (4) comprehensive optimization layer, and (5) real-time output layer. The reasoning interaction logic and data flow between each layer are as follows: Figure 1 As shown, the model receives real-time collected decision element data as initial input, and after a series of internal hierarchical processing steps, finally outputs control point setpoints. Each node in the model network is a sub-model, whose function can be either "routing" or "computation". When its function is "routing", its calculation results determine the activity status and data flow direction of the data connection between its upstream and downstream sub-models. When its function is "computation", its calculation results are transmitted to the downstream sub-model for consumption via the data connection between sub-models, such as... Figure 2 As shown, Figure 2 The network topology shown is for illustrative purposes only and does not represent a limitation to this topology in actual design and application. Within a model layer, such as a pre-constrained layer, multiple sub-models are organized into a directed acyclic graph structure. This structure allows data to flow through different paths within the layer as needed, thereby enabling adaptive processing for different operating conditions.

[0039] In this embodiment, the model network can be considered a local network within the pre-constraint layer, which dynamically activates the sub-models that need to participate in the calculation based on real-time operating conditions. Taking the current application scenario as an example, when the production plan requires a change in the target steel strip grade, the "routing" node routes the input data to the local sub-network used to control the production of the corresponding grade based on this information. Each sub-model in this sub-network is responsible for sequentially calculating the reasonable range of parameters such as temperature, tension, and speed for the operating condition, which serves as the output of the pre-constraint layer.

[0040] More specifically, step 3 includes: Step 3.1: The preprocessing layer simulates the instinctive behavior of human operators when filtering, organizing, and correcting information while observing dashboards and data reports. Its function is to perform data cleaning, deduplication, missing value imputation, unit conversion, and data merging on the input raw decision-making data. For example, sensor data may contain transient spike noise or missing values ​​due to communication interruptions. The preprocessing layer uses algorithms such as moving average filtering or Kalman filtering to smooth signals like temperature and pressure; for transient missing values, it can use the effective values ​​from the previous time step or linear interpolation to impute them. Furthermore, this layer precisely aligns and merges data from different systems (such as production plans from a manufacturing execution system and real-time temperature from a distributed control system) based on timestamps, forming a unified and clean feature vector for use by subsequent layers. This layer ensures that the model's decisions are based on high-quality, conflict-free data.

[0041] Step 3.2: For the pre-constraint layer, this layer simulates the "feel" decision-making of a human operator who dynamically fine-tunes the standard process procedure based on rich experience and perception of the equipment's current "temperament." The goals and constraints given in the standard process procedure are usually for ideal operating conditions, while expert operators will flexibly adjust them according to the actual situation. This is the function of the pre-constraint layer; it dynamically adjusts the process control goals and constraints based on the input decision-making data, especially production plan data and real-time operating condition data. In this example of a continuous annealing furnace, suppose the production plan requires the production of a specific grade of steel strip, whose standard process requires an annealing temperature of 800°C in a certain furnace section. However, the pre-constraint layer receives feedback from the material specification data and finds that the thickness of the current batch of steel strip is 0.5% higher than the standard value. An experienced operator would anticipate that the slightly higher thickness might lead to a slower heating rate. Therefore, he might fine-tune the target operating temperature to 805°C to compensate for the extra thickness. Accordingly, a sub-model in the pre-constraint layer (e.g., a shallow neural network trained on historical data or a set of fuzzy logic rules) performs this calculation, outputting a dynamically adjusted target temperature of 805℃, instead of a static 800℃. It should be noted that the implementation of this layer can be very flexible, typically containing multiple sub-models arranged in a serial, parallel, nested, or other topological network. Each sub-model is responsible for the dynamic adjustment of a specific aspect (e.g., quality target adjustment, safety constraint adjustment, energy consumption target adjustment). These sub-models can employ mechanistic rules (e.g., adjustments based on materials mechanics, metallurgy, and heat transfer formulas), big data analytics techniques (including but not limited to probability statistics, ANOVA, chi-square tests, correlation analysis, etc.), or machine learning algorithms (including but not limited to machine learning, reinforcement learning, decision optimization algorithms such as SVM, XGBoost, DNN, LSTM, Transformer, DQN, genetic algorithms, linear optimization, etc.). Under different production plans and operating conditions, the sub-models that need to participate in the calculation may differ. The network described above can dynamically adjust the network structure and data links, activating only the sub-models that actually need to participate in the calculation, thereby improving computational efficiency. In particular, if the production plan of the control scenario remains unchanged, this layer can be omitted, and the output of the preprocessing layer can be directly output to the control pre-computation layer.

[0042] Step 3.3: For the control pre-calculation layer, this layer aims to simulate the preliminary, independent thinking of a human operator regarding the appropriate setpoint values ​​for each independent control setpoint after determining the overall goals and constraints. Specifically, the function of the control pre-calculation layer is to independently calculate the setpoint values ​​for each of multiple control points. This layer typically consists of a set of parallel sub-models, each corresponding to one or a small group of closely related control points. For example: Sub-model group A receives information such as the furnace correction roller cylinder, strip position, tension of each section, strip speed in the furnace, and strip width, and calculates whether strip correction is needed and the specific parameter settings; Sub-model group B calculates the strip speed setpoint in the furnace based on the current strip temperature, strip specifications, and current strip speed; Sub-model group C, based on the dynamic target temperature output from the pre-constraint layer and the real-time temperature of the current continuous annealing furnace, calculates the preliminary strip temperature setpoints for each furnace section through a small model similar to that in the pre-constraint layer; Sub-model group D calculates the tension setpoints for each furnace section based on the strip specifications and current tension of each section. At this stage, the calculations of each sub-model are relatively independent, with little or no consideration of the coupling effects between them. This simulates the human brain's ability to decompose complex problems and solve sub-problems separately.

[0043] Step 3.4: For the comprehensive optimization layer, after initially calculating the setpoints for each control point, skilled experts will conduct a global review and final weighing of options, as complex couplings and conflicts often exist between the various control points. Examples include coupling between speed and temperature, coupling between correction and speed / tension, and coupling between temperature and tension. The function of the comprehensive optimization layer is to perform this comprehensive processing and optimization. This layer receives the initial setpoints output from the control pre-calculation layer and coordinates these setpoints based on the actual state of the industrial equipment and production constraints obtained from the decision element data to resolve potential conflicts between the control points. This layer can employ multi-objective optimization algorithms, such as genetic algorithms or particle swarm optimization algorithms, to find the optimal solution set, i.e., a set of final control point setpoints, such that no other objective can be optimized without sacrificing any other objective (e.g., safety, quality, efficiency, cost). For any setpoint that does not meet the objectives of this layer, a corresponding alarm message will be generated.

[0044] Step 3.5: This layer corresponds to the physical action of the human operator inputting the final decision into the control system through the human-machine interface. Its main function is to combine and encapsulate the final setpoints output by the comprehensive optimization layer according to the control interface specifications and communication protocol specifications of the industrial equipment. This layer ensures that the agent's decisions can be accurately received and executed by the underlying control system.

[0045] Step 4: Based on the data stored in Step 1 and Step 2, train the decision-making model based on anthropomorphic thinking chain and multi-layered hybrid experts in Step 3.

[0046] In this embodiment, for each sub-model in the model network in step 3, modeling is performed using mechanistic models, big data analysis techniques (including but not limited to probability statistics, ANOVA, chi-square tests, correlation analysis, etc.) or machine learning algorithms (including but not limited to machine learning, reinforcement learning, decision optimization algorithms, such as generalized linear models, tree models, SVM, XGBoost, DNN, LSTM, Transformer, DQN, genetic algorithms, linear optimization, etc.). The corresponding input and output variable sets are determined based on the mechanism and experience, representative historical data are selected, and the models are trained using the conventional training methods corresponding to each model.

[0047] Step 5: Deploy the well-trained model to the actual production environment. This model receives decision-making data in real time and connects its output control commands to the control interface of industrial equipment, thereby replacing human operators in performing real-time, closed-loop automated control. Specifically, once the model has been trained and its performance verified through offline testing, it can be deployed to the production environment. Deployment is typically done on edge computing devices or industrial servers connected to a distributed control system or PLC network.

[0048] Those skilled in the art will understand that, in addition to implementing the system, apparatus, and their modules provided by this invention in purely computer-readable program code, the same program can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system, apparatus, and their modules provided by this invention can be considered a hardware component, and the modules included therein for implementing various programs can also be considered structures within the hardware component; alternatively, modules for implementing various functions can be considered both software programs implementing the method and structures within the hardware component.

[0049] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. An industrial control method based on anthropomorphic thought chain and hybrid expert model, characterized in that, include: Step S1: Based on the operator's thought decision points and decision chain when controlling the equipment, construct a decision model based on anthropomorphic thought chain and multi-layered hybrid experts for each process segment in the complete production chain; Step S2: Use historical data to train the decision model for each process segment to obtain the trained decision model; Step S3: Using the trained decision model, obtain the control setpoint based on the real-time sensing data of the corresponding process section.

2. The industrial control method based on anthropomorphic thought chain and hybrid expert model according to claim 1, characterized in that, The decision-making model based on anthropomorphic thinking chain and multi-layered hybrid experts is a multi-layered network. The composition level of the multi-layered network is determined based on the macro-decision logic and overall operation process of human operators in industrial control. Based on the micro-level thinking and decision-making sequence and specific logical connections of human operators in specific process control, the topology of each network layer is determined.

3. The industrial control method based on anthropomorphic thought chain and hybrid expert model according to claim 2, characterized in that, The multi-layer network includes one or more of the following: a preprocessing layer, a pre-constraint layer, a control pre-computation layer, a comprehensive optimization layer, and a real-time output layer; The preprocessing layer is used to preprocess the real-time sensing data of the process section. The pre-constraint layer is used to fine-tune the objectives and constraints under certain process control requirements; The control pre-computation layer is used to calculate the optimal setpoint for each control point based on the preprocessed sensing data and the fine-tuned targets and constraints. The comprehensive optimization layer is used to comprehensively process the calculated optimal setpoints of each control point so that the setpoints of each control point meet the overall goal of the current process segment. The real-time output layer is used to send the set values ​​of each control point after comprehensive processing to the corresponding device or preset interface.

4. The industrial control method based on anthropomorphic thought chain and hybrid expert model according to claim 2, characterized in that, Each layer of the network is a model network with a serial, parallel, or nested topology composed of multiple sub-models; the topology of each layer of the network maps to the operator's decision-making link. Each layer of the network is a directed acyclic graph (DAG) structure. The DAG structure is used to allow data to flow through different paths within the current layer according to requirements, thereby achieving adaptive processing for different working conditions.

5. The industrial control method based on anthropomorphic thought chain and hybrid expert model according to claim 1, characterized in that, The sensing data in step S3 includes: production plan data, material specification data, material status data, relevant sensor data, relevant production equipment status data, process specifications, and production line design data. The control settings include any one or more of the following: speed, rotational speed, temperature, position, tension, pressure, power, frequency, flow rate, flow volume, opening degree, and mode setting.

6. An industrial control system based on anthropomorphic thought chain and hybrid expert model, characterized in that, include: Module M1: Based on the operator's thought decision points and decision chain when controlling equipment, a decision model based on anthropomorphic thought chain and multi-layered hybrid experts is built for each process segment in the complete production chain; Module M2: Use historical data to train the decision model for each process segment to obtain the trained decision model; Module M3: Utilizes the trained decision model to obtain control setpoints based on real-time sensing data from the corresponding process section.

7. The industrial control system based on anthropomorphic thought chain and hybrid expert model according to claim 6, characterized in that, The decision-making model based on anthropomorphic thinking chain and multi-layered hybrid experts is a multi-layered network. The composition level of the multi-layered network is determined based on the macro-decision logic and overall operation process of human operators in industrial control. Based on the micro-level thinking and decision-making sequence and specific logical connections of human operators in specific process control, the topology of each network layer is determined.

8. The industrial control system based on anthropomorphic thought chain and hybrid expert model according to claim 7, characterized in that, The multi-layer network includes one or more of the following: a preprocessing layer, a pre-constraint layer, a control pre-computation layer, a comprehensive optimization layer, and a real-time output layer; The preprocessing layer is used to preprocess the real-time sensing data of the process section. The pre-constraint layer is used to fine-tune the objectives and constraints under certain process control requirements; The control pre-computation layer is used to calculate the optimal setpoint for each control point based on the preprocessed sensing data and the fine-tuned targets and constraints. The comprehensive optimization layer is used to comprehensively process the calculated optimal setpoints of each control point so that the setpoints of each control point meet the overall goal of the current process segment. The real-time output layer is used to send the set values ​​of each control point after comprehensive processing to the corresponding device or preset interface.

9. The industrial control system based on anthropomorphic thought chain and hybrid expert model according to claim 7, characterized in that, Each layer of the network is a model network with a serial, parallel, or nested topology composed of multiple sub-models; the topology of each layer of the network maps to the operator's decision-making link. Each layer of the network is a directed acyclic graph (DAG) structure. The DAG structure is used to allow data to flow through different paths within the current layer according to requirements, thereby achieving adaptive processing for different working conditions.

10. The industrial control system based on anthropomorphic thought chain and hybrid expert model according to claim 6, characterized in that, The sensing data in module M3 includes: production plan data, material specification data, material status data, relevant sensor data, relevant production equipment status data, process specifications, and production line design data. The control settings include any one or more of the following: speed, rotational speed, temperature, position, tension, pressure, power, frequency, flow rate, flow volume, opening degree, and mode setting.

Images