Intelligent Decision-Making System for Complex Industrial Scenarios Based on Multi-Agent Collaboration

Abstract

This invention relates to the field of intelligent control technology in pulp and paper manufacturing, providing an intelligent decision-making system for complex industrial scenarios based on multi-agent collaboration. The system collects multi-source data on process, equipment, energy, and scheduling, delineates safe operating ranges with equipment load margin as the core, and calibrates process load and upstream / downstream coupling parameters as collaborative benchmarks. Each process agent implements closed-loop control for a single work section and feedforward compensation for adjacent work sections, locally mitigating local disturbances. It coordinates recursive correction of full-link fluctuations, pre-adjustment timing matching of scheduling, and global arbitration of energy quotas to achieve cross-work section collaborative optimization. Collaborative rules are iteratively calibrated based on counterfactual contribution and lightweight reinforcement learning. This invention forms a complete control closed loop from coupling benchmark construction, local deviation mitigation, global collaborative optimization to rule self-iteration, significantly improving the operational stability, energy utilization efficiency, and scheduling smoothness of the entire pulp and paper manufacturing process.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology in pulp and paper making, and in particular to an intelligent decision-making system for complex industrial scenarios based on multi-agent collaboration. Background Technology

[0002] The process parameters of pulping and papermaking are coupled sequentially along the material flow direction, with weak intermediate buffering capacity; at the same time, production is deeply integrated with the water, electricity, and steam energy systems, and fluctuations in production load directly lead to imbalances in the energy supply and demand of the entire plant.

[0003] The existing control system suffers from two major technical bottlenecks: First, the structural contradiction between independent control at the work section level and strong coupling across the entire process. Each work section is independently configured with a PID controller or model predictive controller, focusing solely on its own quality indicators without considering parameter linkages between upstream and downstream work sections. However, fluctuations in pulp hardness in the cooking section alter the refining load in the beating section, and deviations in beating degree are transmitted to the papermaking section, affecting paper basis weight and moisture content, creating a cascading amplification effect along the material flow. Due to the lack of feedforward compensation mechanisms between adjacent work sections and a recursive coordination mechanism across the entire chain, local optimization often leads to quality fluctuations and increased energy consumption in subsequent work sections, failing to achieve global optimization across the entire process.

[0004] Second, there is a mismatch between steady-state control with fixed parameters and highly dynamic operating conditions. Fluctuations in the moisture content and proportion of raw wood chips, changes in production across multiple product types, and temporary equipment maintenance, combined with discrete scheduling tasks, are common occurrences in pulp and paper production. Existing control schemes calibrate PID parameters or feedforward coefficients based on rated steady-state conditions, and control quality drops sharply after the operating conditions deviate from the calibration point. Centralized scheduling systems require full-process recalculation when dealing with discrete scheduling tasks such as production changes, resulting in delayed response and a high proportion of defective products and significant losses in capacity and energy consumption during the production changeover transition phase.

[0005] In summary, existing technologies have not yet solved the technical challenges of unifying the architecture design of distributed work section autonomy and full-process collaborative optimization in strongly coupled, highly dynamic process industrial scenarios, and of pre-matching and coordinating the timing planning of discrete scheduling tasks with the operating parameters of continuous work sections. Therefore, there is an urgent need for a full-process intelligent decision-making system based on multi-agent collaboration to overcome the challenges of fragmented control and disturbance adaptation in the strongly coupled pulp and paper manufacturing process, and to achieve global collaborative optimization. Summary of the Invention

[0006] To overcome the defects and shortcomings of existing technologies, this invention provides an intelligent decision-making system for complex industrial scenarios based on multi-agent collaboration.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides an intelligent decision-making system for complex industrial scenarios based on multi-agent collaboration, comprising the following modules: The full-domain perception module collects production data from the entire pulp and paper production process, and also accesses energy production and consumption data and external production scheduling instructions. Hierarchical decision-making module: Process agents distributed across the core pulp and paper production sections retrieve production data from their respective sections and autonomously adjust the process parameters of their sections; Task agents receive external production scheduling instructions, break down cross-section discrete scheduling tasks, match section execution resources and timing, and coordinate with process agents to pre-adjust operating conditions; Coordination agents summarize energy production and consumption data, break down dynamic energy operation quotas, and arbitrate resource and target conflicts between sections. Collaborative optimization module: It links the intelligent agents of each process along the pulping and papermaking process link to recursively align process control parameters and energy load parameters; it links the task intelligent agents and process intelligent agents to pre-match discrete scheduling task timing and operating condition parameters; it calculates the collaborative contribution of each intelligent agent based on the dynamic energy operation quota completion rate, and iteratively optimizes the collaborative control rules; The closed-loop execution module performs dual-dimensional verification of process quality and equipment safety for all output decision commands. Once the verification is successful, the commands are sent to the field control system for execution.

[0008] According to the above technical solution, the steps of collecting production data from the entire pulp and paper production process and simultaneously accessing energy consumption data and external production scheduling instructions include: S110. Collect production data for each core section of the entire pulp and paper production process, including process control parameters, equipment status parameters, and online quality inspection parameters.

[0009] S120: Collect energy production and consumption data for the entire pulp and paper production process, including electricity supply and consumption parameters and steam production and transportation parameters.

[0010] S130, External production scheduling instructions that connect to the entire pulp and paper production process, including product changeover scheduling instructions and equipment maintenance scheduling instructions.

[0011] According to the above technical solution, the process intelligence agents distributed in each core section of pulp and paper production retrieve production data of the section and autonomously adjust the process parameters of the section; the task intelligence agents receive external production scheduling instructions, break down cross-section discrete scheduling tasks and match the execution resources and timing of the section, and coordinate with the process intelligence agents to pre-adjust the working conditions; the steps of coordinating intelligence agents to summarize energy production and consumption data, break down dynamic energy operation quotas, and arbitrate resource and target conflicts between sections include: S210, the process intelligence agent of the core section retrieves the process control parameters, equipment status parameters and online quality detection parameters of this section; the process status parameters of the pulp and paper making section are limited to a safe operating range; the online quality detection parameters are used as a closed-loop feedback benchmark; and the process control parameters of this section are adjusted in real time according to the deviation of the quality indicators.

[0012] S220: The process agents of adjacent upstream and downstream processes interact in real time to measure the operating deviation of process control parameters and the equipment load margin rate of equipment status parameters; the downstream process agent uses the parameter fluctuation of the upstream process as a feedforward compensation amount to correct the process control parameter set value in advance within the safe operating range of the process parameters of this process, to compensate for the deviation of material transfer and energy load between upstream and downstream, and to reduce the transmission amplitude of local operating condition fluctuations to the downstream process.

[0013] S230: The task agent receives external production scheduling instructions and breaks them down into execution sub-tasks for each related work section. It matches execution resources with the equipment load margin rate of each work section and arranges the task execution sequence according to process connection constraints. Before the task is officially started, the linked process agent uses the target process parameters corresponding to the sub-task as a benchmark and pre-adjusts the process control parameters step by step within the safe operating range of the process parameters of this work section to avoid instantaneous impact and extended transition period when switching working conditions.

[0014] S240, the coordinating agent, summarizes two types of energy production and consumption data: power supply and consumption parameters and steam production and transmission parameters. It combines the load change requirements of the corresponding work sections with the product change scheduling instructions and equipment maintenance scheduling instructions, and generates dynamic energy operation quotas for each work section based on the process load benchmark of each work section. It receives resource adjustment requests from each work section and dynamically adjusts the allocation of work section quotas based on global production priorities and energy consumption constraints.

[0015] According to the above technical solution, the steps of linking intelligent agents of each process along the pulping and papermaking process link to recursively align process control parameters and energy load parameters; pre-matching task intelligent agents and process intelligent agents with discrete scheduling task timing and operating condition parameters; calculating the collaborative contribution of each intelligent agent based on the dynamic energy operation quota completion rate; and iteratively optimizing the collaborative control rules include: S310: Real-time monitoring of the operational deviation amplitude of process control parameters and online quality detection parameters of each section; combined with the trigger signal of external production scheduling instructions; determining whether the production disturbance type is continuous operating condition fluctuation or discrete scheduling task; and determining the disturbance impact range based on the affected sections of the parameter deviation, and triggering the corresponding collaborative control logic.

[0016] S320. In response to continuous operating condition fluctuations, starting from the disturbance source section, the intelligent agents of each process are linked along the material flow direction of the pulping and papermaking process to recursively transmit process parameter deviations and energy load adjustments. Within the safe operating range of process parameters and dynamic energy operation quota constraints of their respective sections, the intelligent agents of each process work together to correct process control parameters and energy load setpoints, reduce the accumulation of parameter deviations throughout the entire process, and suppress the gradual transmission and amplification of fluctuations along the process link.

[0017] S330. For discrete scheduling tasks, the task agent coordinates all related process agents in the entire chain. Combining the task execution sequence, process adjustment cycle of the section, and equipment load margin rate, under the constraint of dynamic energy operation quota, the agent coordinates and matches the pre-adjustment gradient and entry sequence of process parameters of each section to smooth the switching process of the entire process and reduce the disturbance of scheduling actions on the stability of continuous production operation.

[0018] S340 uses online quality compliance rate, dynamic energy operation quota completion rate, and discrete scheduling task timing achievement rate as global evaluation indicators. It compares the global indicator deviation before and after parameter adjustment of each agent and calculates the collaborative contribution of each process agent and task agent to the global operation goal. Correspondingly, it uses a lightweight reinforcement learning parameter optimization method to iteratively optimize the recursive alignment rules of the whole process link parameters, the collaborative matching rules of discrete scheduling task working conditions, and the parameter interaction logic between agents.

[0019] According to the above technical solution, the step of performing a dual-dimensional verification of process quality and equipment safety on all output decision commands, and then sending them to the field control system for execution after the verification is passed, includes: S410. Using the process quality qualification boundary corresponding to the online quality detection parameters and the equipment safety operation limit corresponding to the equipment status parameters as a dual-dimensional verification benchmark, perform pre-constraint verification on the process adjustment command and the scheduling execution command. After determining that the values ​​of the command parameters are all within the constraint boundaries, the command is sent to the field control system for execution.

[0020] Secondly, this application provides an electronic device, including: a processor and a memory, wherein the memory stores a computer program that can be called by the processor, and the processor executes an intelligent decision-making system for complex industrial scenarios based on multi-agent collaboration by calling the computer program stored in the memory.

[0021] Thirdly, this application provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to execute an intelligent decision-making system for complex industrial scenarios based on multi-agent collaboration.

[0022] Compared with the prior art, this application has the following advantages and beneficial effects: 1. By relying on the intelligent agent of the process, the autonomous closed loop of a single work section and the feedforward compensation of adjacent work sections are realized. Upstream operating condition fluctuations do not require global scheduling intervention, and downstream local compensation and correction are carried out in advance, directly suppressing the amplification of disturbances along the process link and avoiding the chain fluctuations in quality and energy consumption of subsequent work sections caused by the optimization of a single work section.

[0023] 2. Through a hierarchical multi-agent architecture, two types of typical disturbances are classified and coordinated: continuous operating condition fluctuations are aligned with process and load parameters segment by segment along the process link to smooth out large-scale operating condition drift across the entire link; discrete scheduling tasks are globally matched to pre-adjust gradients and timings of each section to smooth the switching process of operating conditions such as production changeover and maintenance, shorten the transition period, and reduce defective products and energy consumption during the transition phase.

[0024] 3. The counterfactual baseline method is used to quantify the actual contribution of each cooperative agent. Combined with lightweight reinforcement learning, the cooperative rule parameters are continuously calibrated. It can adapt to dynamic scenarios such as raw material drift and load changes without repeated manual calibration. The cooperative control accuracy is continuously self-optimized as the operation progresses. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the structure of an intelligent decision-making system for complex industrial scenarios based on multi-agent collaboration, provided in an embodiment of this application. Figure 2 This is a flowchart illustrating the hierarchical decision-making module provided in an embodiment of this application; Figure 3 This is a flowchart illustrating the collaborative optimization module provided in an embodiment of this application. Detailed Implementation

[0026] The technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments and specific features in the embodiments are detailed descriptions of the technical solution of this application, rather than limitations thereof. In the absence of conflict, the embodiments and technical features in the embodiments can be combined with each other.

[0027] Please see Figure 1 , Figure 1 This is a schematic diagram of the structure of an intelligent decision-making system for complex industrial scenarios based on multi-agent collaboration, provided in an embodiment of this application. Specifically, it includes the following modules: Full-domain perception module: collects production data from the entire pulp and paper production process, and simultaneously accesses energy production and consumption data and external production scheduling instructions.

[0028] In this embodiment, the collection of production data from the entire pulping and papermaking process, along with the integration of energy consumption data and external production scheduling instructions, includes the following specific details: S110. Collect production data for each core section of the entire pulp and paper production process, including process control parameters, equipment status parameters, and online quality inspection parameters.

[0029] For example, in this embodiment, the three core process segments of the main pulp and paper making process—cooking, pulping, and papermaking—are selected as the data acquisition and intelligent agent deployment objects. The above-mentioned segments are the core process links in pulp and paper production. The process parameters are coupled sequentially along the material flow direction. Energy consumption is concentrated and deeply related to the process operation status. Fluctuations in operating conditions have the characteristic of being transmitted step by step along the process. Through industrial Ethernet and Profinet and Modbus fieldbus, the DCS distributed control system, equipment PLC control unit and online detection instrument of the pulp and paper production line are connected. The operation data of the above core sections are collected synchronously according to a unified sampling period (the default sampling period in this embodiment is 500ms). The raw data is stored in the real-time database after preprocessing by sliding mean filtering and outlier removal. Connect to the field transmitter measurement points of the DCS system of each section to collect in real time the parameters of slurry concentration, material flow rate, process temperature, pipeline pressure, production line speed and slurry-to-mesh ratio of each core section. These parameters are then archived in the section identification zones corresponding to the three core process sections of cooking, pulping and papermaking, and together they constitute the process control parameters of each section, which serve as the direct control objects of the intelligent agents of each process. The sensor units and motor protectors of the continuous digester in the cooking section, the disc mill in the pulping section, the paper machine drive system in the papermaking section, and the drying cylinder group are connected to collect the motor current, operating speed, bearing temperature, equipment vibration amplitude, and real-time load rate parameters of the above core process equipment. These parameters are then archived according to the section identification zones of the three core process sections of cooking, pulping, and papermaking, and together they constitute the equipment status parameters. Based on the rated load value specified on the equipment's nameplate, and combined with the real-time load rate of the equipment collected in real time, the equipment load margin rate is calculated by the difference between the value and the real-time load rate. This rate is used to characterize the proportion of the equipment's remaining load capacity and serves as a safety constraint boundary for adjusting process parameters and a basis for judging task resource matching. The system connects to the online beating degree meter in the pulping section and the QCS quality control system in the papermaking section, and collects pulp beating degree, paper basis weight, paper moisture content and paper ash content parameters in real time for each core process section. These parameters are then archived in the corresponding sections of the three core processes of cooking, pulping and papermaking, forming online quality detection parameters. These parameters serve as the feedback benchmark for closed-loop control of the process sections and the qualification boundary for process quality verification at the execution level.

[0030] S120: Collect energy production and consumption data for the entire pulp and paper production process, including electricity supply and consumption parameters and steam production and transportation parameters.

[0031] Connect to the field monitoring instruments of the energy management system of the pulp and paper plant, power distribution circuit, and steam pipeline network, and synchronously collect real-time energy production and consumption data of the entire plant and each core process section. After cumulative statistics and abnormal jump rejection preprocessing, the data is stored in the real-time database. The total power supply and total power consumption of the entire plant are collected as the benchmark for the overall power supply, which is used to break down the overall dynamic energy operation quota of the entire plant. At the same time, the active power, cumulative power consumption, and overall equipment load ratio of each section are collected according to the three core processes of cooking, pulping, and papermaking. The cumulative power consumption is used to calculate the actual energy consumption of the section and assess the quota completion rate. The active power is used for real-time coordination and alignment of process parameters and energy load. The equipment load ratio is used to dynamically correct the energy quota of the section. The above total power supply and the sub-item data of each section together constitute the power supply and consumption parameters.

[0032] The steam pressure, steam temperature, instantaneous flow rate, and cumulative production and transmission volume of the main steam pipeline network of the entire plant are collected as the global steam total volume benchmark. At the same time, the section identifiers corresponding to the three core process sections of cooking, pulping, and papermaking are used to calculate the steam consumption of each section. The above total data of the entire plant and the section-specific data together constitute the steam production and transmission parameters.

[0033] S130, External production scheduling instructions that connect to the entire pulp and paper production process, including product changeover scheduling instructions and equipment maintenance scheduling instructions.

[0034] The system interfaces with the factory's production execution system to receive and parse external production scheduling instructions throughout the entire process. After verifying the validity of the instructions and aligning them with the execution timestamps, the instructions are stored in the real-time scheduling instruction library. The system receives product changeover scheduling instructions, parses the target product brand, changeover execution time, and target range of associated process parameters corresponding to the instructions, and breaks them down into sub-tasks for each of the three core process sections: cooking, pulping, and papermaking. These sub-tasks serve as the input basis for the task intelligence agent to trigger cross-section changeover coordination. The system receives equipment maintenance and scheduling instructions, analyzes the corresponding maintenance equipment affiliation, maintenance start and end times, work section load reduction requirements, and shutdown scope, and clarifies the task association scope according to the work section identifiers corresponding to the three core process sections of cooking, pulping, and papermaking, providing a basis for load changes to coordinate the intelligent agent to dynamically adjust dynamic energy operation quotas.

[0035] The hierarchical decision-making module consists of process agents distributed across the core pulp and paper production sections, which retrieve production data from their respective sections and autonomously adjust the process parameters of their sections. Task agents receive external production scheduling instructions, break down cross-section discrete scheduling tasks, match the execution resources and timing of each section, and coordinate with process agents to pre-adjust operating conditions. Coordination agents summarize energy production and consumption data, break down dynamic energy operation quotas, and arbitrate resource and target conflicts between sections.

[0036] In this embodiment, as Figure 2As shown, process agents distributed across the core pulping and papermaking processes retrieve production data from their respective sections and autonomously adjust the process parameters of their sections. Task agents receive external production scheduling instructions, break down cross-section discrete scheduling tasks, match section execution resources and timing, and coordinate with process agents to pre-adjust operating conditions. Coordination agents summarize energy production and consumption data, break down dynamic energy operation quotas, and arbitrate resource and target conflicts between sections, including the following specific steps: S210, the process intelligence agent of the core section retrieves the process control parameters, equipment status parameters and online quality detection parameters of this section; the process status parameters of the pulp and paper making section are limited to a safe operating range; the online quality detection parameters are used as a closed-loop feedback benchmark; and the process control parameters of this section are adjusted in real time according to the deviation of the quality indicators.

[0037] The process intelligence agent is an edge control unit deployed locally in each core process section. It is mounted on the DCS system and PLC control unit side of the corresponding process section and has independent data reading, processing and command issuance capabilities. It is the execution subject of autonomous control of a single process section.

[0038] In this embodiment, the process intelligence agent adopts a lightweight three-layer architecture: perception adaptation layer, decision operation layer, and execution output layer. The sensing and adaptation layer interfaces with the industrial fieldbus to enable real-time reading and data format standardization of process control parameters, equipment status parameters, and online quality inspection parameters for this section. The decision-making and computation layer has a built-in PID closed-loop control model with output limiting constraints. It is also configured with a feedforward compensation interface and a collaborative instruction interface. The feedforward compensation interface is used to receive parameter fluctuations transmitted by upstream and downstream adjacent process agents, and the collaborative instruction interface is used to receive upper-level collaborative optimization instructions. The execution output layer converts the target setpoints of the parameters generated by the decision calculation layer into standard control commands, which are then sent to the field actuators to complete the adjustment.

[0039] Each process agent retrieves real-time operational data of process control parameters, equipment status parameters, and online quality inspection parameters from the real-time database, based on the corresponding process identifier, as the input basis for autonomous control of a single process section.

[0040] Using the equipment safety operating limits corresponding to the equipment status parameters collected by S110 as the underlying hard constraints, and combining the equipment's factory rated parameters with the pre-calibrated process load coupling boundary, the process parameter safety operating range of each process control parameter in this section is defined. The process parameter safety operating range is obtained by converting the equipment safety operating limits as the underlying benchmark and the process load coupling boundary coefficient, and is the mapping range of equipment safety constraints on the process adjustment side.

[0041] The process load coupling boundary coefficient and the upstream and downstream process coupling coefficient of S220 are two-level outputs of the same group of process step response calibration work: First, take the core process equipment of a single section as the calibration object, stabilize the equipment to the rated production condition, that is, the normal production load state under the corresponding standard product grade, and the equipment operating parameters and process indicators are in the steady state without fluctuation range, and use this as the calibration benchmark.

[0042] Apply a single standard step disturbance of fixed amplitude to the target process parameters under the reference state, that is, adjust the parameters to the set amplitude in one go and keep them constant, with the disturbance amplitude limited to the range of normal process adjustment; After the equipment reaches a new steady state, the changes in process parameters and the steady-state changes in the load rate of the corresponding core equipment are collected synchronously. The process load coupling boundary coefficient is calculated by the ratio of the two, which represents the change in equipment load corresponding to a unit change in process parameters. Finally, using the upper limit of the rated load of the equipment as a constraint, and combining the current equipment load margin rate and the process load coupling boundary coefficient, the maximum allowable adjustment range of the process parameters is calculated, which is the adjustable upper and lower limit of the safe operating range of the process parameters.

[0043] The quality target value corresponding to the online quality inspection parameters is used as the closed-loop feedback benchmark; the quality target value is divided into two categories according to the production conditions: Under steady-state production conditions, i.e., the continuous and stable production stage without scheduling disturbances, the standard process quality benchmark values ​​corresponding to each product grade are pre-stored in the real-time database. During the scheduling switching operation, i.e. the execution phase of discrete scheduling tasks such as product changeover and equipment maintenance, the target product grade corresponding to the product changeover scheduling instruction in S130 is matched with the quality target value, and the target range of process parameters parsed from the instruction is used as the boundary constraint for process adjustment.

[0044] Execute single-section autonomous closed-loop regulation: Quantify the degree of deviation of quality indicators by using relative deviation calculation method, and take the difference between the real-time detection value and the quality target value as the absolute quality deviation; divide the absolute quality deviation by the quality target value to obtain the relative quality deviation; output the positive and negative direction and deviation magnitude of the relative quality deviation. The target adjustment range of the corresponding process control parameter is calculated based on the deviation range value using the built-in PID closed-loop control model with output limiting constraint, and the target setpoint of the parameter is obtained by superimposing the current value. The calculated target values ​​of the parameters are compared with the safe operating range. Values ​​exceeding the upper and lower limits are automatically truncated to the range boundary values, thus completing the first-level safety constraint verification for a single work section. This level of verification is a real-time and rapid constraint. All output instructions must be executed by the closed-loop module for secondary final inspection before being uniformly issued to the field to drive the corresponding actuators to complete parameter adjustments, so that the deviation of quality indicators continuously converges to the process qualification range.

[0045] The adjustment parameters and feedback benchmarks for each work section are configured according to the work section attributes, and all parameters come from the data set defined in S110: Cooking section: The cooking temperature and material flow rate are used as adjustment parameters, and the control target is to maintain the operating deviation of the outlet slurry temperature, flow rate, and concentration within a preset steady-state deviation threshold. The steady-state deviation threshold is taken from the standard process specification of the corresponding product brand. In this embodiment, it is set as cooking temperature ±1℃, material flow rate ±2%, and slurry concentration ±0.5%, respectively, to ensure the homogeneity and stability of the outlet material of this section and provide consistent feed for the downstream section. Pulping section: The rotation speed of the disc mill and the feed pulp concentration are used as adjustment parameters; the pulp beating degree is used as the closed-loop feedback benchmark, and the quality qualification boundary is consistent with the standard process requirements of the corresponding product grade; Papermaking section: The production line speed and pulp-to-wire ratio are used as adjustment parameters; the paper basis weight and paper moisture content are used as closed-loop feedback benchmarks, and the quality qualification boundary is consistent with the standard process requirements of the corresponding product grade.

[0046] S220: The process agents of adjacent upstream and downstream processes interact in real time to measure the operating deviation of process control parameters and the equipment load margin rate of equipment status parameters; the downstream process agent uses the parameter fluctuation of the upstream process as a feedforward compensation amount to correct the process control parameter set value in advance within the safe operating range of the process parameters of this process, to compensate for the deviation of material transfer and energy load between upstream and downstream, and to reduce the transmission amplitude of local operating condition fluctuations to the downstream process.

[0047] Upstream and downstream adjacent process agents synchronously exchange operation data through the feedforward compensation interface of the decision operation layer according to the unified sampling period set in S110; Based on the material flow direction of the cooking, pulping, and papermaking processes in this embodiment, the upstream and downstream adjacent sections are respectively the cooking section and the pulping section, and the pulping section and the papermaking section. Data interaction is transmitted unidirectionally from upstream to downstream along the material transfer direction.

[0048] The upstream process intelligent agent transmits the operating deviation of the process control parameters and the real-time equipment load margin rate of the adjacent downstream process to the process control parameters of the process. The operating deviation is the difference between the real-time value of the upstream process parameter and the corresponding steady-state process target value. The corresponding steady-state process target value is consistent with the steady-state process reference value used in the S210 single-section closed-loop control and is taken from the standard process procedure of the corresponding product grade.

[0049] The equipment load margin rate is used to dynamically adjust the feedforward compensation strength downstream, avoiding over-adjustment or under-compensation.

[0050] After receiving the parameter deviation from the upstream process, the downstream process agent, combining the baseline upstream-downstream process coupling coefficient and the upstream equipment load margin rate, converts it into the feedforward compensation amount for the corresponding process control parameters of this process. The calculation relationship is as follows: ; ; in, The corrected real-time upstream and downstream process coupling coefficient is used to quantify the actual impact weight of upstream parameter fluctuations on downstream process parameters and is a direct determining variable of the compensation strength. The upstream and downstream process coupling coefficients are used as the benchmark. Together with the process load coupling boundary coefficient of S210, they belong to the two-level output of the same group of process step response calibration work: First, step experiments are carried out independently for the core equipment of each single section to calibrate the correspondence between the process parameters and equipment load of this section and obtain the process load coupling boundary coefficient; then, a linkage step experiment is carried out with the upstream section as the disturbance input end and the downstream section as the response acquisition end to calibrate the transmission relationship between the upstream and downstream process parameters and obtain the benchmark upstream and downstream process coupling coefficients; both use the rated load condition as the experimental benchmark and adopt the standard amplitude process step disturbance to characterize the standard influence of the unit fluctuation of the upstream parameters on the downstream condition under the rated condition, which belongs to the preset model parameters; The real-time load rate of the upstream equipment is taken from the equipment status parameters collected by S110. The higher the real-time load rate, the smaller the adjustment margin of the upstream equipment and the higher the proportion of non-process factors in the fluctuation. The downstream should reduce the compensation response intensity to such fluctuations to avoid misadjustment. Therefore, the actual coupling coefficient decreases as the load rate increases. This is the feedforward compensation amount for the process control parameters of the downstream section. This represents the deviation of the process parameters transmitted from the upstream section. The deviation reference is the steady-state process target value, which is consistent with the steady-state process reference value of the S210 single-section closed-loop control.

[0051] The downstream process agent adds the feedforward compensation to the original output setpoint of the PID closed-loop control model of this section to obtain the pre-corrected parameter target value. The pre-corrected value needs to be verified by the safe operating range constraint of the process parameters of this section. The part exceeding the upper and lower limits is automatically truncated to the range boundary. Finally, the corrected process control parameter setpoint is output and sent for execution to offset the operating condition fluctuation delay caused by the lag in material transfer and energy transmission in advance.

[0052] S230: The task agent receives external production scheduling instructions and breaks them down into execution sub-tasks for each related work section. It matches execution resources with the equipment load margin rate of each work section and arranges the task execution sequence according to process connection constraints. Before the task is officially started, the linked process agent uses the target process parameters corresponding to the sub-task as a benchmark and pre-adjusts the process control parameters step by step within the safe operating range of the process parameters of this work section to avoid instantaneous impact and extended transition period when switching working conditions.

[0053] The task agent is a scheduling-level decision-making unit in the hierarchical decision-making module. It is deployed on the factory production scheduling side server, connecting upwards to the scheduling instruction interface of the factory production and manufacturing execution system, and downwards to the collaborative instruction interface of the agents in each process. Its core functions are instruction parsing and decomposition, load resource verification, execution timing arrangement and pre-tuning and coordination triggering. It is the main body for coordinating and executing cross-section discrete scheduling tasks, and does not directly connect with field sensors and execution mechanisms. The task agent receives external production scheduling instructions parsed by S130 and breaks them down into standardized execution sub-tasks corresponding to the work section according to the instruction type. Each sub-task contains three core types of information: target parameters, execution time, and constraint boundaries. The product changeover scheduling instruction is broken down into process switching sub-tasks for the entire chain of cooking, pulping, and papermaking, and the target process parameter range, target quality index range, and planned official switchover time for each section are extracted. The equipment maintenance scheduling instructions are broken down into load adjustment sub-tasks for the maintenance section, specifying the target load rate of the section, the start and end times of shutdown, and the scope of the affected sections.

[0054] Based on the pre-calculated load margin rate of each section's equipment, perform resource matching verification: when the current equipment load margin rate of a section is greater than the minimum load threshold required for task execution, it is determined that the section has the capacity to undertake the task and can start according to the planned sequence; if the equipment load margin rate is insufficient, the task start time is postponed until the section load drops back to a manageable range before execution is triggered.

[0055] Based on the process connection constraints, the pre-adjustment start-up times of each related process section are arranged, with the material flow transmission lag time as the core calculation basis: Using the planned official switchover time of the downstream papermaking section as the benchmark node, the material transfer lag time between sections is deducted first to obtain the pre-adjustment completion node of the upstream section; the product of the pre-adjustment steps and the unified sampling period of S110 is taken as the total pre-adjustment time of the corresponding section; the total pre-adjustment time of the corresponding section is then deducted from the pre-adjustment completion node of the upstream section, and finally the pre-adjustment start time of the pulping section and the cooking section are determined in sequence. Inter-process material transfer lag time and the coupling coefficient of upstream and downstream processes in S220 They are used for timing calculation and amplitude calculation of cross-section collaboration, respectively: the material transfer lag time is directly used for the backward calculation of the pre-adjustment start time, and the corresponding lag time is deducted from the downstream planned official switch time upstream segment by segment to obtain the pre-adjustment start node of each section. Benchmark upstream and downstream process coupling coefficient The amplitude of the feedforward compensation is directly used in the calculation of the S220 feedforward compensation. The upstream and downstream process coupling coefficients are multiplied by the upstream section parameter deviation to obtain the feedforward compensation amplitude of the downstream section. After the two parameters are initially calculated by process parameter theory, they are simultaneously identified and calibrated in the upstream and downstream section linkage step test. They belong to the graded output results of the same batch of process calibration work as the process load coupling boundary coefficient of S210.

[0056] When the pre-adjusted materials from the upstream section are delivered to the downstream section, the downstream section is simultaneously adjusted to the target operating condition, eliminating the problem of asynchronous switching caused by material transfer delays.

[0057] Before the task officially starts, the task agent issues pre-adjustment instructions to each related process agent, performing step-by-step gradient pre-adjustment based on the target process parameters corresponding to the sub-task; taking the median value of the target process parameter range as the target value, and performing step-by-step gradient pre-adjustment:

[0058] The maximum adjustment amount per step is pre-calibrated based on the stable operation requirements of this section of the process, corresponding to the adjustment rate constraint of the safe operating range; the number of pre-adjustment steps is determined by dividing the total adjustment amount by the maximum adjustment amount per step, and the ratio is rounded up. The total adjustment amount is the difference between the current parameter value and the target value; each process agent adjusts the parameters in the safe operating range of this section of the process parameters one by one according to the unified sampling period set in S110 until the parameters enter the target range.

[0059] In this embodiment, the maximum allowable adjustment of the cooking temperature per step is 1℃. If the total adjustment is 7.5℃, then the number of pre-adjustment steps is 8. The total transition time of the working condition can be directly calculated by combining the sampling period.

[0060] By pre-adjusting in stages, significant changes in process parameters are avoided when the task is officially started, the entire process of switching operating conditions is smoothed, and the operating condition transition cycle corresponding to production changeover and maintenance scheduling is shortened.

[0061] S240, the coordinating agent, summarizes two types of energy production and consumption data: power supply and consumption parameters and steam production and transmission parameters. It combines the load change requirements of the corresponding work sections with the product change scheduling instructions and equipment maintenance scheduling instructions, and generates dynamic energy operation quotas for each work section based on the process load benchmark of each work section. It receives resource adjustment requests from each work section and dynamically adjusts the allocation of work section quotas based on global production priorities and energy consumption constraints.

[0062] The coordinating agent is a global coordination-level decision-making unit in the hierarchical decision-making module. It is deployed on the plant-level energy management side server, connecting upwards to the plant-wide energy management system and downwards to the data interaction interfaces of the process agents and task agents. The core functions of the coordinating agent are global energy quota decomposition, work section resource conflict arbitration and unified constraint output. It does not directly connect to field sensors and execution mechanisms and belongs to the global rule decision-making unit.

[0063] The coordinating agent retrieves total plant-wide electricity and steam energy production and consumption data collected by S120, uses the plant's rated total energy supply as the overall constraint, and combines it with the unit product energy consumption benchmark under rated operating conditions for each section to perform dynamic energy operation quota allocation: The unit product energy consumption benchmark under rated operating conditions for each section is preset according to the product brand. It is calibrated by superimposing the energy consumption value of the process design with historical steady-state operation data. It corresponds one-to-one with the product standard process specifications and is a routine preset parameter for factory energy management, which will not be elaborated here.

[0064] First, the baseline energy quota for each section under steady-state production is calculated. The power quota is allocated based on the proportion of the active power of each section under rated operating conditions to the total active power of the three core sections. The baseline power quota for each section is obtained from the total power supply of the whole plant. Steam quotas are calculated based on the standard for thermodynamic properties of water and steam (IAPWS-IF97). The actual operating enthalpy is calculated from the main steam pressure and main steam temperature collected by S120. The total steam flow of the whole plant is converted into the equivalent flow under a unified energy benchmark by the ratio of the actual enthalpy to the rated operating enthalpy. Then, the benchmark steam quota for each section is allocated according to the load ratio corresponding to the unit product steam energy consumption under the rated operating conditions.

[0065] Combining the production change and maintenance scheduling instructions parsed by S130, the load change range of the corresponding work section is extracted: for production change scenarios, the difference between the rated process load of the target product brand and the current operating load is taken; for maintenance scenarios, the load reduction requirement of the work section is taken from the instructions. The target load value is obtained by superimposing the load change range on the current steady-state reference load (the steady-state operating active power of the work section collected and verified by S120 before the scheduling instruction is triggered). Then, the load correction coefficient is generated by the ratio of the target load value to the steady-state reference load value, and the reference quota is dynamically corrected in a proportional linear manner. That is, the reference quota is multiplied by the load correction coefficient to obtain the dynamic quota, and the dynamic energy operation quota of each work section is obtained, including the two categories of electricity and steam quotas.

[0066] The dismantled quotas are distributed in zones according to the three core work sections of cooking, pulping, and papermaking, serving as the energy constraint boundaries for adjusting the process parameters of each work section; the dynamic quotas are updated in real time when triggered by scheduling instructions, and are calibrated on a rolling basis at fixed cycles when there is no scheduling disturbance.

[0067] The coordinating agent receives energy resource adjustment requests submitted by agents in various processes and tasks, and performs conflict arbitration and quota reallocation based on global production priority rules. When multiple work sections simultaneously apply for increased energy quotas and the total demand exceeds the plant's total supply constraints, the priority is set according to the principle of prioritizing paper quality assurance, papermaking and pulping work sections, and work sections with reduced load. The priority weight of each work section is preset based on the plant's actual production strategy and the tightness of the work section coupling. The lower the weight of the work section, the higher the quota transfer ratio. Then, the adjustment amount is deducted from the quota surplus of the low-priority work sections in reverse according to the weight ratio. All deductions are based on the minimum production guarantee threshold of the work section (the minimum energy consumption required to maintain the stable operation of the work section process, corresponding to the minimum stable operating load of the process) as the lower limit, and the total adjustment amount does not exceed the global gap. The collected allocation quotas will be distributed to the high-priority work sections that applied for additional quotas according to their priority weights, and the total allocation amount will not exceed the plant's total energy supply limit. The final quota allocation results for each work section will be dynamically adjusted to resolve energy resource conflicts between work sections. The adjusted final quotas will be simultaneously distributed to all related intelligent agents as a unified quota benchmark for subsequent full-process collaborative optimization.

[0068] Collaborative optimization module: It recursively aligns process control parameters and energy load parameters with the intelligent agents of each process along the pulping and papermaking process chain; it pre-matches the timing and operating parameters of discrete scheduling tasks with the intelligent agents of the linked tasks and process; it calculates the collaborative contribution of each intelligent agent based on the completion rate of dynamic energy operation quota, and iteratively optimizes the collaborative control rules.

[0069] In this embodiment, as Figure 3 As shown, the process control parameters and energy load parameters of each process agent along the pulping and papermaking process link are recursively aligned; the task agents and process agents are pre-matched with discrete scheduling task timing and operating condition parameters; the collaborative contribution of each agent is calculated based on the dynamic energy operation quota completion rate, and the collaborative control rules are iteratively optimized, including the following specific steps: S310: Real-time monitoring of the operational deviation amplitude of process control parameters and online quality detection parameters of each section; combined with the trigger signal of external production scheduling instructions; determining whether the production disturbance type is continuous operating condition fluctuation or discrete scheduling task; and determining the disturbance impact range based on the affected sections of the parameter deviation, and triggering the corresponding collaborative control logic.

[0070] Quantitative determination of disturbance type: First, calculate the amplitude of the operating deviation of the corresponding parameter for each section, that is, the absolute difference between the real-time detection value of the parameter and the corresponding steady-state process target value. The target value is consistent with the reference value used in the S210 single-section closed-loop control. If an external production scheduling command trigger signal is detected, it is directly determined to be a discrete scheduling task disturbance; if no scheduling command is triggered, and the amplitude of the operational deviation of at least one parameter exceeds the preset process deviation threshold (taken from the standard process specification of the corresponding product brand, which is the abnormal judgment limit for steady-state qualified deviation), it is determined to be a continuous operating condition fluctuation disturbance.

[0071] Quantitative determination of the scope of disturbance impact: The section where the deviation amplitude first exceeds the threshold is taken as the disturbance source section. Along the process material flow direction of cooking, pulping, and papermaking, the deviation status of the corresponding coupled process parameters of each downstream section (which are consistent with the mapping relationship of upstream and downstream parameters corresponding to the real-time upstream and downstream process coupling coefficient) is checked in turn. When the deviation amplitude of the downstream process parameters exceeds the preset correlation judgment threshold, the process is determined to be affected by the disturbance. The preset correlation judgment threshold is obtained by multiplying the upstream process deviation threshold by the benchmark upstream and downstream process coupling coefficient between the corresponding process.

[0072] The total number of affected work sections was counted, and the scope of the disturbance was divided into local disturbance in a single work section and full-link disturbance in multiple work sections. Local disturbance in a single work section refers to the disturbance source work section whose parameters exceed the standard, while full-link disturbance in multiple work sections refers to the disturbance affecting two or more work sections.

[0073] Based on the determined disturbance type and impact range, the corresponding matching level collaborative control logic is triggered: for continuous operating condition fluctuations and disturbances in multiple sections of the entire link, the S320 full-link process parameter recursive collaborative control logic is triggered. Discrete scheduling task disturbances trigger the S330 full-link working condition switching timing coordination control logic, linking the task agent and the related process agents to coordinate and match the pre-adjusted gradient and entry timing.

[0074] S320. In response to continuous operating condition fluctuations, starting from the disturbance source section, the intelligent agents of each process are linked along the material flow direction of the pulping and papermaking process to recursively transmit process parameter deviations and energy load adjustments. Within the safe operating range of process parameters and dynamic energy operation quota constraints of their respective sections, the intelligent agents of each process work together to correct process control parameters and energy load setpoints, reduce the accumulation of parameter deviations throughout the entire process, and suppress the gradual transmission and amplification of fluctuations along the process link.

[0075] For the continuous operating condition fluctuations of the multi-section full-link determined by S310, the collaborative optimization module coordinates and triggers the full-link recursive collaborative control as a global enhancement mechanism for the spontaneous feedforward compensation of adjacent sections of S220: S220 is responsible for the millisecond-level autonomous deviation cancellation between two adjacent sections, and S320 is responsible for the second-level overall correction of the entire process link, covering the entire process of cooking, pulping, and papermaking.

[0076] Starting from the disturbance source section identified by S310, the correction signal is pushed down section by section along the process material flow of cooking, pulping and papermaking: the collaborative correction amount of the process parameters of each downstream section is obtained by multiplying the remaining parameter deviation after correction of the upstream section (the residual deviation that was not completely offset after the upstream section completed its own correction) with the real-time upstream and downstream process coupling coefficient between the corresponding sections. The real-time upstream and downstream process coupling coefficients are calculated using the same logic as those used in the S220 feedforward compensation, i.e., the actual coupling coefficients after correction by the upstream equipment load rate. Synchronously, based on the process load coupling boundary coefficient calibrated by S210 (characterizing the equipment load change corresponding to a unit change in process parameters), the process parameter correction is converted into the corresponding energy load adjustment for the downstream process section, and this adjustment is transmitted synchronously to the downstream process section along with the corrected remaining deviation.

[0077] After receiving the collaborative correction amount, each downstream process intelligent agent adds it to the original parameter setting value of the closed-loop control of its section, and performs dual constraint verification in sequence: first, it compares with the safe operating range of the process parameters of its section defined by S210, and then automatically cuts off parameters that exceed the upper and lower limits to the boundary of the range after correction. Then, by comparing the dynamic energy operation quota for this section issued by S240, the adjustment range of the energy load exceeding the quota limit is automatically reduced to within the quota allowable range. Finally, the verified collaborative correction parameters are output and issued for execution, and the remaining undigested deviations continue to be transmitted to the downstream sections.

[0078] By coordinating and correcting each process segment along the entire chain and gradually absorbing deviations, the accumulation of parameter deviations between each process segment is reduced, local operating condition fluctuations are suppressed from being transmitted and amplified along the process chain, and the stability of the entire process operating conditions and energy consumption is maintained in two dimensions.

[0079] S330. For discrete scheduling tasks, the task agent coordinates all related process agents in the entire chain. Combining the task execution sequence, process adjustment cycle of the section, and equipment load margin rate, under the constraint of dynamic energy operation quota, the agent coordinates and matches the pre-adjustment gradient and entry sequence of process parameters of each section to smooth the switching process of the entire process and reduce the disturbance of scheduling actions on the stability of continuous production operation.

[0080] For the discrete scheduling task disturbances identified by S310, the collaborative optimization module coordinates the task agent and the agents of each related process to perform collaborative control of the entire link working condition switching. The core of S330 is to perform global collaborative matching of the pre-adjustment gradient and the cut-in timing based on the independent pre-adjustment of a single work section, combined with the process coupling relationship and energy quota constraints of the entire link, so as to avoid the problem of load fluctuation and working condition asynchrony caused by the independent adjustment of a single work section.

[0081] Collaborative optimization of pre-adjustment gradient: Combining the real-time equipment load margin rate of each section with the dynamic energy operation quota margin issued by S240, the basic single-step pre-adjustment amount calculated by S230 is corrected. The optimized single-step pre-adjustment amount calculation relationship is as follows: ; in, This refers to the single-step pre-adjustment amount after collaborative optimization; The basic single-step pre-adjustment amount in S230 is obtained by equally dividing the total adjustment amount and the number of pre-adjustment steps; To pre-adjust the gradient correction coefficient, the minimum value between the relative proportion of equipment load margin and the relative proportion of energy quota margin is used. The calculation formula is as follows: ; in, The current real-time equipment load margin rate for the current work section is taken from the calculation results of S110. The rated operating condition reference load margin rate corresponds to the inherent margin level of the equipment under the standard process procedure and when the equipment is at the rated production load. It is obtained by combining the rated parameters of the equipment at the factory with the standard operating condition calibration of the corresponding product brand, and serves as a reference benchmark for the adequacy of the load adjustment space. The remaining balance of the current dynamic energy operation quota for the current work section is obtained by subtracting the real-time energy consumption of the work section collected by S120 from the work section quota value issued by S240. The baseline allowance for rated operating conditions is the fixed allowance after deducting steady-state energy consumption from the baseline energy allowance for the work section under standard production load. It is obtained by combining the factory's energy quota management standard with the standard operating condition energy consumption calibration and serves as a reference benchmark for the adequacy of quota adjustment space.

[0082] The more ample the load margin and quota allowance, the larger the correction coefficient and the higher the single-step adjustment range, thus accelerating the switching speed within the constraints; conversely, the correction coefficient should be reduced and the adjustment cycle lengthened to avoid equipment overload or energy over-quota during the switching process.

[0083] Collaborative optimization of the start-up timing: With the goal of minimizing the total switching time of the entire process and minimizing the fluctuation of operating conditions, the inherent process adjustment cycle of each section (the response time for process parameters to reach a new steady state after the issuance of a single-step pre-adjustment command, which is calibrated in the same batch as the process load coupling boundary coefficient of S210 through step response experiments) and the material transfer lag time between sections (using the same set of linkage step experiment calibration parameters as S220 and S230), the pre-adjustment start-up time obtained by reverse deduction from the S230 basis is globally calibrated. The timing deviation of the pre-adjustment of upstream and downstream sections is checked segment by segment along the material flow direction: the pre-adjustment completion time of the corresponding gradient of the upstream section is superimposed with the material transfer lag time between sections to obtain the time when the adjusted material arrives at the downstream; the difference between this time and the pre-adjustment completion time of the corresponding gradient of the downstream section is obtained to obtain the timing deviation value, which is used to correct the pre-adjustment start time of the downstream section and eliminate the timing deviation caused by the material transfer lag and the mismatch between the section adjustment cycle.

[0084] All the pre-adjusted parameters and timing schemes after collaborative optimization are checked against the dual constraints of the safe operating range of process parameters defined by S210 and the dynamic energy operation quota issued by S240 in each work section. Parameters that exceed the boundary are automatically corrected to the allowable range of constraints and finally issued to each process agent to perform pre-adjustment according to the optimized scheme.

[0085] By coordinating and matching gradients and timing across the entire link, the system smoothly handles the transitions in operating conditions caused by scheduling actions, reducing the disturbance of discrete scheduling tasks to the stability of continuous production operations, and optimizing the operating condition transition cycle under the dual constraints of equipment and energy.

[0086] S340 uses online quality compliance rate, dynamic energy operation quota completion rate, and discrete scheduling task timing achievement rate as global evaluation indicators. It compares the global indicator deviation before and after parameter adjustment of each agent and calculates the collaborative contribution of each process agent and task agent to the global operation goal. Correspondingly, it uses a lightweight reinforcement learning parameter optimization method to iteratively optimize the recursive alignment rules of the whole process link parameters, the collaborative matching rules of discrete scheduling task working conditions, and the parameter interaction logic between agents.

[0087] The statistical period is set to T, and in this embodiment, it is set to 8 hours by default, which corresponds to the duration of a single production shift. This matches the statistical granularity of the regular shift production in the pulp and paper industry, which can cover both the complete steady-state production period and the cycle of a single discrete scheduling task, while also ensuring the timeliness of rule iteration. The three global evaluation indicators are quantified as follows: the online quality compliance rate is the proportion of the number of qualified online quality parameters of the entire process section to the total number of samples within the statistical period. The qualification boundary is consistent with the product standard process quality requirements defined in S210. The data is taken from the online quality detection parameters collected in S110. The dynamic energy operation quota completion rate is the percentage of time during which the actual energy consumption of each section does not exceed the dynamic energy operation quota issued by S240 within the statistical period. The data is taken from the section-level real-time energy consumption data collected by S120 and the quota instructions of S240. The discrete scheduling task timing achievement rate is the percentage of scheduling tasks whose actual completion time and planned time deviation are within the allowable range within the statistical period. The planned time is taken from the task execution timing decomposed by S230, and the actual time is taken from the field execution feedback data.

[0088] The overall comprehensive evaluation index is the weighted sum of three indicators: online quality compliance rate, dynamic energy operation quota completion rate, and discrete scheduling task timing achievement rate. The weight coefficients are preset according to the factory's current production strategy, and the sum of the three weights is 1.

[0089] Pulp and paper mills can dynamically adjust the weighting of core control objectives: In quality-priority scenarios such as producing high-end specialty paper and pilot production of new products, the weighting of online quality compliance rate is increased to prioritize ensuring the stability of finished paper quality; In energy consumption control scenarios such as energy consumption dual control period and peak energy price periods, the weighting of dynamic energy operation quota completion rate is increased to prioritize constraining the total energy consumption of each section; In capacity-priority scenarios such as peak order delivery season and concentrated production changeover, the weighting of discrete scheduling task time sequence achievement rate is increased to prioritize ensuring the timely execution of scheduling tasks.

[0090] For the i-th agent, which includes process agents and task agents, the counterfactual baseline method is used to calculate the collaborative contribution: the global index prediction value after removing the collaborative actions of the agent is used as the counterfactual baseline, and the actual index value after the collaborative control takes effect is compared to quantify its true contribution.

[0091] The counterfactual baseline value is the global comprehensive index value predicted based on the independent closed-loop control logic of a single work section after removing the collaborative function of the i-th agent. It can be obtained by fitting historical steady-state operating data in the non-collaborative mode or by simulating the process mechanism model of a single work section. For the process agent, the S220 feedforward compensation and S320 recursive collaborative functions are removed, and for the task agent, the S230 step-by-step pre-tuning and S330 global timing optimization functions are removed. The actual operating value is the actual global comprehensive index value within the statistical period when all agents are in collaborative mode.

[0092] The absolute contribution of the i-th agent is calculated as follows: ; in, The absolute contribution of the i-th agent represents the marginal improvement of the agent's collaborative action on the global comprehensive evaluation index. A positive value indicates that its collaborative action has a positive gain on the global operating goal, and the larger the value, the stronger the gain. When all intelligent agents work together, the actual global comprehensive evaluation index value within the statistical period is calculated by weighted summation of three items: online quality compliance rate, dynamic energy operation quota completion rate, and discrete scheduling task timing achievement rate. The weighting coefficients are consistent with the preset configuration of the global comprehensive evaluation index. is the counterfactual baseline value corresponding to the i-th agent, that is, the global comprehensive evaluation index value predicted based on the independent closed-loop control logic of a single work section after removing the collaborative function of the agent.

[0093] Normalize the absolute contributions of all agents to obtain the proportion of collaborative contributions of each agent. The calculation relationship is as follows: ; Where n is the total number of agents participating in the collaboration. Let be the collaborative contribution of the i-th agent, representing its proportion of contribution to the global operational goal.

[0094] With the goal of maximizing the overall evaluation index and the reward allocation based on the collaborative contribution of each agent, lightweight reinforcement learning is used to iteratively optimize three types of collaborative rule parameters. The optimization objects include: the recursive alignment rule of parameters in the entire process link, corresponding to the upstream and downstream process coupling coefficients in S220 and S320 (including the benchmark calibration and real-time correction logic); the collaborative matching rule of discrete scheduling task conditions, corresponding to the pre-tuned gradient correction coefficients and timing calibration parameters in S230 and S330; and the parameter interaction logic between agents, corresponding to the trigger thresholds and interaction cycles of deviation transmission and quota application.

[0095] The coordinating agent is a global rule decision-making unit. Its quota decomposition and conflict arbitration logic is based on the physical constraints of the total energy supply of the whole plant and the execution of preset production priority rules. It does not belong to the real-time collaborative control type of agent and is not included in the counterfactual contribution calculation and reinforcement learning parameter optimization scope. Its output dynamic energy operation quota serves as the hard constraint boundary for the iterative optimization of this step. The quota allocation effect is indirectly evaluated through the dynamic energy operation quota completion rate index.

[0096] The parameter iteration update logic is as follows: ; in, For the rule parameter set of the t-th iteration, it covers three types of parameters to be optimized: upstream and downstream process coupling coefficients of S220 / S320, pre-tuned gradient correction coefficients of S230 / S330, and agent interaction trigger threshold. The default learning rate in this embodiment is 0.01, which is a common small step size value for lightweight reinforcement learning. This balances the iteration convergence speed with the stability of production operation and avoids excessive single parameter jumps that could cause fluctuations in operating conditions. The gradient of the global comprehensive index relative to the rule parameters is calculated by differential calculation based on the difference between the index and the parameter adjustment amount between two adjacent statistical periods. This eliminates the need for complex backpropagation calculations and is suitable for lightweight calculation requirements at the edge of industrial sites.

[0097] During the iteration process, all parameter updates are subject to hard constraints based on the safe operating limits of the S210 device and the dynamic energy operating quota of the S240, ensuring that the optimization results do not exceed the safety and energy consumption boundaries. After the iteration is completed, the rule parameters are synchronously updated to the corresponding intelligent agent, serving as the execution benchmark for the next round of collaborative control, thereby achieving continuous self-optimization of the entire process collaborative control logic.

[0098] The closed-loop execution module performs dual-dimensional verification of process quality and equipment safety for all output decision commands. Once the verification is successful, the commands are sent to the field control system for execution.

[0099] In this embodiment, all output decision commands undergo dual-dimensional verification of process quality and equipment safety. After successful verification, the commands are sent to the field control system for execution, including the following specific steps: S410. Using the process quality qualification boundary corresponding to the online quality detection parameters and the equipment safety operation limit corresponding to the equipment status parameters as a dual-dimensional verification benchmark, perform pre-constraint verification on the process adjustment command and the scheduling execution command. After determining that the values ​​of the command parameters are all within the constraint boundaries, the command is sent to the field control system for execution.

[0100] The process quality qualification boundary is based on the process qualification range corresponding to the online quality detection parameters collected by S110, which is completely consistent with the quality qualification boundary adopted by the single-section closed-loop control of S210. During the verification, the quality index value of the section after the execution of the command is predicted based on the current working condition, and it is confirmed that it is within the qualification range to avoid the quality exceeding the limit caused by parameter adjustment.

[0101] The safe operating limits of the equipment are based on the core equipment rated parameter thresholds defined in S110, including the rated current of the motor, the upper limit of the bearing temperature alarm, the upper limit of the vibration amplitude safety limit, and the equipment rated load threshold. This is the underlying benchmark for the safe operating range of process parameters defined in S210, and the boundaries of the two are completely consistent. During verification, the amplitude of the equipment operating parameters corresponding to the execution of the instruction is calculated to confirm that it is within the safe bearing range of the equipment, and to eliminate safety risks such as equipment overload, over-temperature, and over-vibration.

[0102] The verification process compares the command setpoint with the corresponding constraint boundary parameter by parameter. If all parameter values ​​are within the two-dimensional constraint boundary, the command is deemed compliant and directly sent to the field control system for execution. If any parameter exceeds the boundary, the parameter exceeding the limit is automatically truncated to the corresponding constraint boundary value, and the parameter exceeding the limit information and the truncation result are simultaneously fed back to the corresponding decision-making agent for reference when the S340 iteratively optimizes control and collaborative rules.

[0103] By performing dual-dimensional pre-verification, full-scale operational risks are intercepted before instructions are issued, ensuring that the entire system operates within the constraints of quality compliance and equipment safety, and avoiding the risk of exceeding limits caused by collaborative adjustment and scheduling operations.

[0104] Embodiments of the present invention also provide an electronic device, including a memory, a processor, and a communication bus; the memory and the processor are connected via the communication bus. The memory stores an intelligent decision-making system for complex industrial scenarios based on multi-agent collaboration, which can be loaded and executed by the processor, as provided in the above embodiments.

[0105] The memory can be used to store instructions, programs, code, code sets, or instruction sets. The memory may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system, instructions for at least one function, and instructions for implementing the intelligent decision-making system for complex industrial scenarios based on multi-agent collaboration provided in the above embodiments. The data storage area may store data involved in the intelligent decision-making system for complex industrial scenarios based on multi-agent collaboration provided in the above embodiments.

[0106] A processor may include one or more processing cores. The processor executes instructions, programs, code sets, or instruction sets stored in memory, and calls data stored in memory to perform various functions and process data as described in this application. The processor may be at least one of a specific application-specific integrated circuit, a digital signal processor, a digital signal processing device, a programmable logic device, a field-programmable gate array, a central processing unit, a controller, a microcontroller, and a microprocessor. It is understood that, for different devices, the electronic devices used to implement the above-described processor functions may also be other types, and the embodiments of this application do not specifically limit the specific implementation.

[0107] A communication bus may include a pathway for transmitting information between the aforementioned components. The communication bus can be a PCI bus or an EISA bus, etc. Communication buses can be categorized into address buses, data buses, control buses, etc.

[0108] This application provides a computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described in the above embodiments, which is a complex industrial scenario intelligent decision-making system based on multi-agent collaboration.

[0109] In this embodiment, a computer-readable storage medium can be a tangible device that holds and stores instructions used by an instruction execution device. A computer-readable storage medium can be, but is not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof. Specifically, a computer-readable storage medium can be a portable computer disk, a hard disk, a USB flash drive, a random access memory, a read-only memory, an erasable programmable read-only memory, a podium random access memory, a portable compressed disk read-only memory, a digital multifunction disk, a memory stick, a floppy disk, an optical disk, a magnetic disk, a mechanical encoding device, or any combination thereof.

[0110] The terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0111] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the foregoing application concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions claimed in this application.

Claims

1. An intelligent decision-making system for complex industrial scenarios based on multi-agent collaboration, characterized in that, Includes the following modules: The full-domain perception module collects production data from all stages of the pulp and paper making process, and simultaneously accesses energy production and consumption data and external production scheduling instructions; Hierarchical decision-making module: Process agents distributed in each core section of pulp and paper production retrieve production data of the section and autonomously adjust the process parameters of the section; Task agents receive external production scheduling instructions, break down cross-section discrete scheduling tasks and match the execution resources and timing of the section, and coordinate with process agents to adjust the working conditions in advance. The coordinating agent summarizes energy production and consumption data, breaks down dynamic energy operation quotas, and arbitrates resource and objective conflicts between work sections. Collaborative optimization module: Links intelligent agents at each stage of the pulping and papermaking process to recursively align process control parameters and energy load parameters; The intelligent agents for coordinated tasks and the intelligent agents for process steps pre-match discrete scheduling task timing and operating parameters. The collaborative contribution of each intelligent agent is calculated based on the dynamic energy operation quota completion rate, and the collaborative control rules are iteratively optimized. The closed-loop execution module performs dual-dimensional verification of process quality and equipment safety for all output decision commands. Once the verification is successful, the commands are sent to the field control system for execution.

2. The intelligent decision-making system for complex industrial scenarios based on multi-agent collaboration as described in claim 1, characterized in that, Collect production data from all stages of the pulp and paper making process, and simultaneously access energy production and consumption data and external production scheduling instructions, including the following steps: Collect production data for each core section of the entire pulp and paper production process, including process control parameters, equipment status parameters, and online quality inspection parameters. Collect energy production and consumption data for the entire pulp and paper production process, including electricity supply and consumption parameters and steam production and transportation parameters; Access to external production scheduling instructions for the entire pulp and paper production process, including product changeover scheduling instructions and equipment maintenance scheduling instructions.

3. The intelligent decision-making system for complex industrial scenarios based on multi-agent collaboration as described in claim 2, characterized in that, The process intelligence agents distributed across the core sections of pulp and paper production retrieve production data for each section and autonomously adjust the process parameters of that section, including the following steps: The core process intelligent agent retrieves the process control parameters, equipment status parameters, and online quality detection parameters of this process; it uses the equipment status parameters to limit the safe operating range of the process control parameters of the pulping and papermaking process, and uses the online quality detection parameters as a closed-loop feedback benchmark to adjust the process control parameters of this process in real time according to the deviation of the quality indicators. The intelligent agents of adjacent upstream and downstream processes interact in real time to measure the operating deviation of process control parameters and the equipment load margin rate of equipment status parameters. The intelligent agent of downstream processes uses the parameter fluctuation of upstream processes as a feedforward compensation amount to correct the process control parameter set value in advance within the safe operating range of process parameters in this process, to compensate for the deviation of material transfer and energy load between upstream and downstream processes, and to reduce the transmission amplitude of local operating condition fluctuations to downstream processes.

4. The intelligent decision-making system for complex industrial scenarios based on multi-agent collaboration according to claim 3, characterized in that, The task agent receives external production scheduling instructions, breaks down discrete scheduling tasks across work sections, matches work section execution resources and timing, and coordinates with process agents to pre-adjust working conditions, including the following steps: The task agent receives external production scheduling instructions and breaks them down into execution sub-tasks for each related work section. It matches execution resources with the equipment load margin rate of each work section and arranges the task execution sequence according to process connection constraints. Before the task is officially started, the linked process agents use the target process parameters corresponding to the sub-task as a benchmark and pre-adjust the process control parameters step by step within the safe operating range of the process parameters of this work section to avoid instantaneous impact and extended transition period when switching working conditions.

5. The intelligent decision-making system for complex industrial scenarios based on multi-agent collaboration according to claim 4, characterized in that, The coordinating agent aggregates energy production and consumption data, breaks down dynamic energy operation quotas, and arbitrates resource and target conflicts between work sections, including the following steps: The coordinating agent summarizes two types of energy production and consumption data: power supply and consumption parameters and steam production and transmission parameters. It combines the load change requirements of the corresponding work sections with the product change scheduling instructions and equipment maintenance scheduling instructions, and generates dynamic energy operation quotas for each work section based on the process load benchmark of each work section. It receives resource adjustment requests from each work section and dynamically adjusts the allocation of work section quotas based on global production priorities and energy consumption constraints.

6. The intelligent decision-making system for complex industrial scenarios based on multi-agent collaboration as described in claim 5, characterized in that, The process control parameters and energy load parameters are recursively aligned by intelligent agents at each stage of the pulping and papermaking process, including the following steps: The system monitors the amplitude of the operational deviation between the process control parameters and online quality detection parameters of each work section in real time. Combined with the trigger signals of external production scheduling instructions, it determines whether the production disturbance is a continuous operating condition fluctuation or a discrete scheduling task. At the same time, it determines the scope of the disturbance based on the affected work sections of the parameter deviation and triggers the corresponding collaborative control logic.

7. The intelligent decision-making system for complex industrial scenarios based on multi-agent collaboration as described in claim 6, characterized in that, The collaborative task agent and the process agent pre-match discrete scheduling task timing and operating condition parameters, including the following steps: To address continuous operating fluctuations, starting from the disturbance source section, the intelligent agents of each process are linked along the material flow direction of the pulping and papermaking process, and the deviation of process parameters and the adjustment of energy load are transmitted in a step-by-step manner. Within the safe operating range of process parameters and dynamic energy operation quota constraints of their respective sections, the intelligent agents of each process work together to correct the process control parameters and energy load setpoints, reduce the accumulation of parameter deviations throughout the entire process, and suppress the amplification of fluctuations along the process link. For discrete scheduling tasks, the task agent coordinates all related process agents across the entire chain. By combining the task execution sequence, process adjustment cycle of each section, and equipment load margin, and under the constraint of dynamic energy operation quota, the task agent coordinates and matches the pre-adjustment gradient and entry sequence of process parameters of each section, smooths the switching process of the entire process, and reduces the disturbance of scheduling actions on the stability of continuous production operation.

8. The intelligent decision-making system for complex industrial scenarios based on multi-agent collaboration according to claim 7, characterized in that, The collaborative contribution of each agent is calculated based on the dynamic energy operation quota completion rate, and the collaborative control rules are iteratively optimized, including the following steps: Using online quality compliance rate, dynamic energy operation quota completion rate, and discrete scheduling task timing achievement rate as global evaluation indicators, the deviation of global indicators before and after parameter adjustment of each agent is compared, and the collaborative contribution of each process agent and task agent to the global operation goal is calculated. Correspondingly, a lightweight reinforcement learning parameter optimization method is used to iteratively optimize the recursive alignment rules of parameters in the entire process link, the collaborative matching rules of discrete scheduling task working conditions, and the parameter interaction logic between agents.

9. The intelligent decision-making system for complex industrial scenarios based on multi-agent collaboration as described in claim 8, characterized in that, All output decision commands undergo dual-dimensional verification of process quality and equipment safety. Once the verification is successful, the commands are sent to the field control system for execution, including the following steps: Using the process quality qualification boundary corresponding to the online quality detection parameters and the equipment safety operation limit corresponding to the equipment status parameters as dual-dimensional verification benchmarks, pre-constraint verification is performed on process adjustment commands and scheduling execution commands. After determining that the values ​​of the command parameters are all within the constraint boundaries, they are sent to the field control system for execution.

Images