Alcohol production distillation process control system based on digital twin modeling

Abstract

The invention discloses an alcohol production distillation process control system based on digital twin modeling, which relates to the field of alcohol production and comprises a multi-mode sensing and processing unit, a digital twin modeling unit and a collaborative optimization unit. According to the method, traditional, non-intrusive and environmental security data are fused through multi-dimensional sensing, feature extraction optimization is combined, data reliability and effectiveness are improved, a digital twin modeling unit fuses a mechanism and real-time and historical data to construct a model, dynamic adaptability is improved through self-calibration, multi-target optimization and feedforward-feedback cooperative control are achieved, and the reliability and effectiveness of the system are improved. The real-time disturbance compensation module compensates disturbance in advance through a disturbance recognition library and digital twinborn prediction, the product quality is prevented from being affected, the product purity is guaranteed, meanwhile, energy consumption and yield are optimized, disturbance is rapidly dealt with, resource optimization allocation, performance monitoring and fault early warning are achieved through full-life-cycle digital closed-loop management, and the product quality is improved. The problems of disturbance response lag and insufficient compensation are solved, and the stability of product purity is guaranteed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of alcohol production, in particular to an alcohol production distillation process control system based on digital twin modeling. BACKGROUND

[0002] Alcohols (such as methanol, ethanol, ethylene glycol) are basic raw materials in the fields of chemical industry, energy, medicine, etc. In the production process, distillation is the core link to realize the separation and purification of mixtures, and the stability, efficiency and product purity of the distillation process directly affect the economy and environmental protection of alcohol production, so accurate control of the distillation process is crucial.

[0003] For example, a methanol synthesis reactor hybrid modeling method for a digital twin system with publication number CN115220343A, the method comprises: obtaining a first training set for training a reactor network, wherein the first training set includes a plurality of first training samples, each first training sample includes the reactor temperature, reactor pressure and the content of each key component in the actual methanol synthesis reaction, and the reactor network is a multi-layer neural network structure.

[0004] In the prior art, the distillation process of alcohol production involves multi-component mixtures, and is affected by temperature, pressure, flow, reflux ratio and other multivariate coupling. In the face of disturbances, there is significant response lag and insufficient compensation. On the one hand, it is difficult to comprehensively identify the disturbance type, and it is impossible to systematically sort out multiple disturbances and extract the feature vectors of each disturbance, resulting in a blurred disturbance source. On the other hand, it can only rely on passive response of post-measured data, lacks an advance prediction mechanism based on a digital twin model, and cannot predict the impact of disturbances on key indicators, so that measures are taken only when the disturbance has spread to the product quality link. The compensation method only relies on feedback control, which is prone to over-adjustment or insufficient adjustment, resulting in fluctuations in product purity and affecting production stability. SUMMARY

[0005] The present application aims to provide an alcohol production distillation process control system based on digital twin modeling to solve the problem of significant response lag and insufficient compensation when facing disturbances, leading to fluctuations in product purity and affecting production stability as described in the background.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical solution: an alcohol production distillation process control system based on digital twin modeling, comprising a multi-modal perception and processing unit, a digital twin modeling unit and a collaborative optimization unit.

[0007] The multi-modal perception and processing unit is used for multi-dimensional sensing data acquisition, state feature vector extraction of the collected data, unreliable data labeling and sensor self-check triggering; the digital twin modeling unit acquires the collected data of the multi-modal perception and processing unit to construct a digital twin model; the collaborative optimization unit cooperates with the digital twin modeling unit to perform result prediction by using the digital twin model;

[0008] The collaborative optimization unit includes a multi-objective optimization module, a real-time disturbance compensation module, a collaborative control module and an abnormality monitoring module; the multi-objective optimization module is used for constructing an optimization target system and outputting optimization control parameters; based on the prediction capability of the digital twin model, the real-time disturbance compensation module starts compensation before a disturbance affects product quality, customizes compensation strategies for different disturbance types, and cooperates with feedforward control and feedback control; the collaborative control module extends the digital twin model constructed by the digital twin modeling unit to the whole life cycle of the distillation system by linking key equipment of the distillation system; and the abnormality monitoring module is used for labeling an abnormal position in the digital twin model constructed by the digital twin modeling unit and generating a visual report.

[0009] Preferably, the digital twin modeling unit includes a mechanism and data fusion module and a model self-calibration module.

[0010] The mechanism and data fusion module constructs a mechanism model describing the dynamic behavior of the system based on the physical or chemical mechanism of the distillation process, acquires real-time production data and historical data collected in the multi-modal perception and processing unit, corrects the simplification error and parameter uncertainty of the mechanism model, and outputs the digital twin model; the model self-calibration module acquires real-time data uploaded by the edge nodes of the multi-modal perception and processing unit, and dynamically adjusts the parameter weight of the digital twin model based on the Bayesian optimization algorithm.

[0011] Preferably, in the mechanism and data fusion module, after constructing the mechanism model describing the dynamic behavior of the system, the key parameters of the mechanism model are initialized based on the design parameters and historical steady-state data, so that the mechanism model outputs results similar to the physical system under standard working conditions.

[0012] Preferably, in the multi-objective optimization module, when constructing the optimization target system, the alcohol product purity meeting the standard is taken as a constraint, the minimum unit product energy consumption and the maximum hourly output are optimized synchronously, a three-dimensional objective function is constructed, the multi-objective optimization problem is converted into a single-objective optimization problem based on the constraint weighting method, the deep deterministic policy gradient algorithm is adopted, the digital twin model is trained to simulate the working condition, the state data of the digital twin is received in real time, and the optimal control parameters are output.

[0013] Preferably, in the multi-objective optimization module, the formula of the three-dimensional objective function is as follows:

[0014]

[0015] wherein, represents energy consumption weight, represents yield weight, the constraint penalty is 1000 when the purity is less than 99.7%, and the penalty term is 500 if the equipment parameters are out of limits, to avoid safety risks.

[0016] Preferably, in the real-time disturbance compensation module, customizing a compensation strategy for different disturbances includes the following steps:

[0017] S1, constructing a disturbance identification library: combing the typical disturbance types of the alcohol distillation process, extracting the feature vector, and performing similarity matching with the typical features in the disturbance identification library, for unknown disturbances, identifying abnormal patterns through the isolation forest algorithm, marking as a disturbance to be classified, and triggering manual intervention;

[0018] S2, digital twin prediction: based on the digital twin model, predicting the impact of the disturbance on the key indicators of the distillation system in the future 1-2 minutes, inputting the identified disturbance parameters into the digital twin model, correcting the disturbance parameters, performing short-term and medium-term prediction respectively, correcting the model parameters, and controlling the prediction error;

[0019] S3, feedforward-feedback collaborative action: based on the prediction results of the digital twin, combining real-time measured data, and through the collaborative action of feedforward control and feedback control, blocking the spread of the disturbance to product quality, and adjusting.

[0020] Preferably, in step S2, the prediction error control formula is as follows:

[0021]

[0022] wherein, y actual (t) represents the measured key indicator at time t, y pred (t) represents the digital twin prediction value, M represents the number of sampling points within 10 seconds, and when the prediction error e RMSE exceeds the threshold value, triggering online correction of the model parameters.

[0023] Preferably, in step S3, when blocking the spread of the disturbance to product quality, at the initial stage of the disturbance, feedforward control dominates, blocking the spread of the disturbance, and after 30 seconds, feedback control dominates, correcting the deviation, and when the adjustment directions of feedforward and feedback conflict, the product purity priority principle is used to make decisions.

[0024] Preferably, in step S3, the specific formula of the feedforward-feedback collaborative action is as follows:

[0025] Feedforward adjustment of reflux ratio for feed component fluctuation:

[0026]

[0027] wherein, kR represents the proportional coefficient of reflux ratio feedforward adjustment, Δx A represents the feed component deviation, represents the sensitivity of reflux ratio to feed component, x A (t) represents the feed component content at time t, represents the target set value of component A in the feed;

[0028] Reboiler heat feedforward adjustment of feed flow fluctuation:

[0029]

[0030] wherein, k Q represents the safety factor, ΔA represents the feed flow deviation, represents the sensitivity of heat to flow;

[0031] Total feedforward adjustment amount:

[0032]

[0033] wherein, represents the weight coefficient of reflux ratio feedforward adjustment of feed component fluctuation in the total feedforward adjustment amount, represents the weight coefficient of reboiler heat feedforward adjustment of feed flow fluctuation in the total feedforward adjustment amount;

[0034] Feedback fine-tuning based on real-time purity deviation:

[0035]

[0036] e(t) = y actual (t) - y set ;

[0037] wherein, K p represents the proportional coefficient, e(t) represents the purity deviation at time t, K i represents the integral coefficient, represents the deviation cumulative sum from the initial time to time t, K d represents the differential coefficient, represents the change rate of differential deviation, y actual (t) represents the actual purity at time t, y set represents the target purity value;

[0038] Conflict decision priority decision:

[0039]

[0040] wherein, ε represents the purity deviation threshold.

[0041] Preferably, the synergistic control module comprises a global optimization module, a performance evaluation module and a predictive maintenance module;

[0042] The global optimization module acquires digital twin data of multiple distillation systems in the whole plant area based on the digital twin modeling unit, and optimizes resource allocation; the performance evaluation module calculates key KPIs of the system in real time, and compares them with design indicators and historical optimal values to generate a performance decay curve; and the predictive maintenance module predicts equipment failure risk based on the equipment degradation part of the digital twin model, and generates a maintenance scheme in combination with real-time sensing data collected by the multi-modal perception and processing unit.

[0043] Compared with the prior art, the present application has the following beneficial effects:

[0044] The present application fuses traditional, non-invasive and environmentally safe data through multi-dimensional sensing, optimizes feature extraction, improves data reliability and effectiveness, and constructs a model by fusing mechanism and real-time and historical data through the digital twin modeling unit, improves dynamic adaptability through self-calibration, realizes accurate mapping and prediction, multi-objective optimization and feedforward-feedback synergistic control, and the real-time disturbance compensation module compensates for disturbances in advance through disturbance identification library and digital twin prediction, avoids affecting product quality, optimizes energy consumption and yield while ensuring product purity, quickly responds to disturbances, realizes resource optimization allocation, performance monitoring and fault warning through full life cycle digital closed-loop management, reduces unplanned downtime, solves the problems of lagging response to disturbances and insufficient compensation, and ensures stable product purity. BRIEF DESCRIPTION OF DRAWINGS

[0045] Fig. 1 A flowchart of the alcohol production distillation process control system based on digital twin modeling of the present application;

[0046] Fig. 2 A system block diagram of the alcohol production distillation process control system based on digital twin modeling of the present application.

[0047] In the figure: 1, multi-modal perception and processing unit; 2, digital twin modeling unit; 21, mechanism and data fusion module; 22, model self-calibration module; 3, synergistic optimization unit; 31, multi-objective optimization module; 32, real-time disturbance compensation module; 33, synergistic control module; 331, global optimization module; 332, performance evaluation module; 333, predictive maintenance module; 34, abnormality monitoring module. DETAILED DESCRIPTION

[0048] The technical solutions in the embodiments of the present application will be clearly and completely described with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0049] Embodiment one: refer to Figs. 1-2 As shown in the figure: an alcohol production distillation process control system based on digital twin modeling, through the collaborative work of three core units of multi-modal perception and processing, digital twin modeling, collaborative optimization control, precise control, dynamic optimization and full life cycle management of the distillation process are realized.

[0050] I. Core unit and technical details

[0051] (1) Multi-modal perception and processing unit 1 can realize multi-dimensional data acquisition, processing and feature extraction, providing reliable input for subsequent modeling and decision-making.

[0052] 1. Multi-dimensional data acquisition, covering three types of data: traditional parameters, non-invasive sensors, and environmental safety sensors.

[0053] 2. Data processing and feature extraction, through data compression, time domain redundant data is removed, frequency domain wavelet decomposition is performed on high frequency signals, and equipment health feature frequency band is retained; state feature vector extraction includes equipment health features, process performance features, transient disturbance features, and abnormal risk features, providing direct input for digital twin modeling.

[0054] (2) Digital twin modeling unit 2 realizes dynamic prediction and precise simulation by constructing a digital mapping model of the physical distillation process.

[0055] 1. Mechanism and data fusion module 21 first constructs a mechanism model, describes the dynamic behavior of the system based on the physical / chemical mechanism of the distillation process, corrects the model, and uses real-time production data and historical data to correct the simplification error and parameter uncertainty of the mechanism model; finally, based on design parameters and historical steady-state data, the key parameters are initialized to ensure that the model is consistent with the physical system output under standard working conditions.

[0056] 2. Model self-calibration module 22 dynamically adjusts the model parameter weight based on the real-time data uploaded by the edge node through the Bayesian optimization algorithm, improving the adaptability to actual working conditions. When the system experiences a major disturbance, historical similar working condition data from the cloud is called to accelerate the convergence of model parameter optimization.

[0057] (3) Collaborative optimization unit 3 realizes multi-objective optimization, real-time disturbance compensation and full life cycle management based on the prediction results of the digital twin model.

[0058] 1. Multi-objective optimization module 31, which optimizes the target to be the purity of alcohol products as a constraint, simultaneously minimizes the energy consumption per product and maximizes the hourly output, constructs a three-dimensional objective function, converts the multi-objective into a single-objective problem by using a constraint weighting method, uses a deep deterministic policy gradient algorithm, trains an intelligent agent offline through a digital twin model, simulates working conditions, and outputs optimal control parameters.

[0059] 2. Real-time disturbance compensation module 32, which aims to start compensation before the disturbance affects product quality, to avoid purity fluctuations.

[0060] The specific steps include:

[0061] S1, disturbance identification: build a disturbance identification library (including typical disturbances: feed fluctuation, equipment drift, environmental change, etc.), extract feature vectors and match, unknown disturbances are identified by the Isolation Forest algorithm, triggering manual intervention and updating the library;

[0062] S2, digital twin prediction: predict the impact of the disturbance on key indicators (purity, energy consumption) in the next 1-2 minutes based on the model, correct the model through an error control formula to ensure prediction accuracy;

[0063] S3, feedforward-feedback collaborative action: feedforward control dominates the early stage (within 30 seconds), blocking the spread of disturbances (such as adjusting the reflux ratio when the feed component fluctuates, adjusting the reboiler heating capacity when the flow fluctuates), feedback control dominates the later stage, correcting the deviation (based on real-time purity deviation PID fine-tuning), conflict decision, and adjusting the direction of conflict with the principle of prioritizing product purity.

[0064] 3. Collaborative control module 33 can realize digital closed-loop management of the whole life cycle of the distillation system, including global optimization module 331, performance evaluation module 332, and predictive maintenance module 333:

[0065] Global optimization module 331 is based on the digital twin data of multiple distillation systems in the whole plant area, and uses linear programming and genetic algorithm to optimize resource allocation (such as steam / cooling water scheduling);

[0066] Performance evaluation module 332 calculates key KPIs (technical indicators: purity, yield; economic indicators: unit cost; energy consumption indicators: steam / electricity consumption) in real time, compares them with design values and historical optimal values, and generates a performance decay curve;

[0067] Predictive maintenance module 333 combines the equipment degradation model of the digital twin with real-time sensor data to predict fault risks (such as reboiler fouling, pump bearing wear) in advance and generate maintenance solutions.

[0068] 4. Abnormal monitoring module 34 automatically marks abnormal positions (such as temperature abnormal trays, leakage points) in the digital twin model, generates a visual report, and assists in quickly locating the root cause

[0069] II. Core principles

[0070] 1. Digital mapping and prediction: High-precision digital replication of the physical distillation process is achieved through a digital twin model, combined with mechanisms and real-time data to accurately predict the dynamic changes of key indicators such as temperature, pressure, and purity, providing a basis for early intervention.

[0071] 2. Multi-dimensional perception and closed-loop control: Multi-modal sensors cover process, equipment, and environmental data in all dimensions, which are processed and input into the digital twin model. The model prediction results drive the collaborative optimization unit to output control strategies, forming a "perception-modeling-decision-control" closed loop.

[0072] 3. Disturbance active compensation: Through disturbance identification library and digital twin prediction, traditional "after-feedback" is upgraded to "before-prediction + real-time compensation", and feedforward and feedback are coordinated to avoid excessive adjustment, ensuring purity stability while optimizing energy consumption and yield.

[0073] 4. Full life cycle optimization: From design, operation to maintenance and decommissioning, digital twin is used to realize full-process digital management, improve resource utilization, reduce unplanned downtime, and ultimately achieve economic efficiency and stability of alcohol production.

[0074] The present application solves the problems of disturbance response lag and multi-variable coupling difficult control in traditional distillation control by integrating digital twin, multi-modal perception, and intelligent optimization algorithm, providing technical support for efficient and stable operation of alcohol production.

[0075] Example two: reference Figs. 1-2 As shown: an alcohol production distillation process control system based on digital twin modeling, including a multi-modal perception and processing unit 1, a digital twin modeling unit 2, and a collaborative optimization unit 3.

[0076] The multi-modal perception and processing unit 1 is used to cover traditional parameters, non-invasive sensing and environmental safety sensing, to collect multi-dimensional sensing data, extract state feature vectors of the collected data, and mark unreliable data and trigger sensor self-checking.

[0077] Traditional parameters include temperature data collected by fiber grating sensors at each layer of the tower plate, pressure data collected by high-precision piezoelectric sensors, flow data collected by electromagnetic flow meters, liquid level data collected by radar liquid level meters, and component concentration data collected by online gas chromatographs; non-invasive sensing includes real-time monitoring of distillate alcohol purity using near-infrared spectroscopy sensors, detection of abnormal vibration of gas-liquid mass transfer in the tower using acoustic sensors, and monitoring of tower wall temperature field distribution using infrared thermal imaging to predict the risk of fouling; environmental safety sensing includes combustible gas sensors in explosion-proof areas to detect ethanol / methanol leaks, equipment vibration sensors to monitor the bearing state of pumps and compressors, and energy consumption measurement sensors to monitor real-time consumption of steam, electricity, and cooling water.

[0078] In extracting the state feature vector of the collected data, the collected data is first compressed, redundant data in the time domain is removed, and high-frequency vibration signals in the frequency domain are decomposed by wavelets to retain characteristic frequency bands reflecting the health status of the equipment, and the energy values and peak frequencies of each frequency band are retained to replace the original time domain data. For transient signal compression sensing and sparse reconstruction, the state feature vector needs to accurately reflect the health status, process performance, and abnormal risk of the distillation system, providing direct input for digital twin modeling and intelligent decision-making. Specific features include equipment health features, process performance features, transient disturbance features, and abnormal risk features.

[0079] The digital twin modeling unit 2 is configured to construct a digital twin model and achieve accurate mapping and dynamic prediction of the distillation process.

[0080] The digital twin modeling unit 2 includes a mechanism and data fusion module 21 and a model self-calibration module 22. The mechanism and data fusion module 21 constructs a mechanism model describing the dynamic behavior of the system based on the physical or chemical mechanism of the distillation process, uses real-time collected production data and historical data to correct the simplification error and parameter uncertainty of the mechanism model, and improves the dynamic adaptability of the model to actual working conditions. Finally, the digital twin model is output to support subsequent state monitoring, optimization control, and fault warning functions; the model self-calibration module 22 dynamically adjusts the parameter weights of the digital twin model based on the real-time data uploaded by the edge node through the Bayesian optimization algorithm, and when the system experiences a major disturbance, the cloud historical similar working condition data is called to accelerate convergence;

[0081] The core basis of the mechanism model is the physical and chemical nature law of the distillation process, which determines the possibility and efficiency of mixture separation, including phase equilibrium principle, mass transfer and heat transfer principle, material balance principle and energy balance principle. The actual distillation system is abstracted into a modular mathematical unit, each unit corresponds to a specific physical function, describes the variables (such as the temperature, pressure, component concentration, flow rate, etc.) and parameters (such as phase equilibrium constant, mass transfer coefficient, heat transfer coefficient, equipment structure parameter, etc.) of the system state, and establishes mathematical equations for each abstract unit based on phase equilibrium, mass transfer, heat transfer, material / energy balance principle, and realizes the coupling between units through variable association. The association between units is described by mathematical equations to realize the prediction of the system state.

[0082] After constructing the mechanism model describing the dynamic behavior of the system, the key parameters of the mechanism model are initialized based on the design parameters and historical steady-state data, so that the mechanism model outputs results similar to the physical system under standard working conditions.

[0083] The collaborative optimization unit 3 realizes multi-objective dynamic optimization and adaptive control based on the prediction results of the digital twin model; the collaborative optimization unit 3 includes a multi-objective optimization module 31, a real-time disturbance compensation module 32, a collaborative control module 33 and an abnormality monitoring module 34; the multi-objective optimization module 31 is used to build an optimization target system, and to clearly define the optimization direction and boundary of the intelligent agent;

[0084] When building the optimization target system, the alcohol product purity meets the constraints, and the unit product energy consumption is minimized and the hourly production is maximized, a three-dimensional objective function is constructed, the multi-objective problem is converted into a single-objective optimization problem based on the constraint weighting method, the state space fully reflects the real-time running state of the distillation system, including product quality state, process parameter state, energy consumption and production state and disturbance state, the action space is the adjustable control parameter of the intelligent agent, including reflux ratio, reboiler heating steam quantity, feed quantity and tower top take-off valve opening, the digital twin model is trained offline, a high-fidelity virtual distillation environment is constructed by using the digital twin model, the working condition is simulated, the intelligent agent is continuously learned in the virtual environment, the safety risk and cost of physical system trial and error are avoided, the state data of the digital twin is received in real time, and the optimal control parameter is output.

[0085] The formula of the objective function is as follows:

[0086]

[0087] Among them, indicates the energy consumption weight, indicates the production weight, the constraint penalty is 1000 when the purity is less than 99.7%, and the penalty term is 500 if the equipment parameters are out of limit, to avoid safety risks.

[0088] The real-time disturbance compensation module 32 initiates compensation before the disturbance affects product quality based on the prediction capability of the digital twin model, customizes compensation strategies for different disturbance types, cooperates with feedforward control and feedback control, quickly responds and avoids over-adjustment, ensures that the purity of the alcohol product is stable at the target value, and minimizes the impact of the disturbance on energy consumption and yield;

[0089] Customizing compensation strategies for different disturbances includes the following steps:

[0090] S1, Construct a disturbance identification library: Sort out the typical disturbance types of the alcohol distillation process, extract the feature vector, and match the similarity with the typical features in the disturbance identification library. For unknown disturbances, identify abnormal patterns through the isolation forest algorithm, mark them as classified disturbances, and trigger human intervention, while updating the disturbance library to provide a basis for rapid identification.

[0091] The disturbance types include feed-related disturbances, equipment parameter drifts, and external environmental disturbances.

[0092] The feed-related disturbances include feed flow fluctuations, feed component fluctuations, and feed temperature fluctuations. When extracting the feed-related features, the time domain features are the mutation amplitude, duration, and change rate of the flow / component / temperature, and the correlation features are the correlation between the feed parameter fluctuations and the response of the downstream trays. Store them in the library as "feature vector + typical response curve";

[0093] The equipment parameter drifts include reboiler heat transfer efficiency decline, condenser cooling efficiency attenuation, and pump / throttle characteristic drift. The equipment parameter drifts extract features such as performance degradation characteristics, bias accumulation characteristics, and the correlation between equipment drifts and energy consumption.

[0094] The external environmental disturbances include cooling water temperature fluctuations, changes in workshop environmental temperature / humidity, and utility fluctuations. The external environmental disturbance extraction features are the absolute fluctuations of the environmental parameters and the coupling characteristics with the process parameters.

[0095] S2, Digital twin prediction: Based on the digital twin model, predict the impact of the disturbance on the key indicators of the distillation system within 1-2 minutes in the future to provide decision-making basis for feedforward control. Input the identified disturbance parameters into the digital twin model, correct the feed quantity, component, temperature parameters of the model, key parameters of the model, and heat dissipation coefficient of the model, simulate the disturbance diffusion path and impact, and perform short-term prediction and medium-term prediction respectively. The short-term prediction focuses on equipment response, and the medium-term prediction focuses on product quality and energy consumption. The model parameters are corrected every 10 seconds with the measured data of the physical system to control the prediction error.

[0096] The prediction error control formula is as follows:

[0097]

[0098] Where, yactual (t) represents the measured key indicator at time t, y pred (t) represents the digital twin prediction value, M represents the number of sampling points within 10 seconds, and e RMSE When the threshold is exceeded, the model parameters are triggered for online correction.

[0099] S3, feedforward-feedback collaborative action: based on the prediction results of the digital twin, combined with real-time measured data, through the collaborative action of feedforward control and feedback control, the feedforward control adjusts the key control parameters in advance according to the medium-term prediction of the digital twin, blocks the disturbance from spreading to product quality, and avoids excessive adjustment to cause new fluctuations. The feedback control is based on the deviation of real-time sensor data and target value on the basis of feedforward control, and performs fine adjustment.

[0100] At the initial stage of disturbance, feedforward control dominates, quickly blocks disturbance diffusion, and after 30 seconds, feedback control dominates, fine-tunes the deviation. If the adjustment direction of feedforward and feedback conflicts, the product purity priority principle is used to make decisions.

[0101] The specific formula of feedforward-feedback collaborative action is as follows:

[0102] Reflux ratio feedforward adjustment of feed component fluctuation:

[0103]

[0104] Where, k R represents the proportionality coefficient of reflux ratio feedforward adjustment, Δx A represents the feed component deviation, represents the sensitivity of reflux ratio to feed component, x A (t) represents the feed component content at time t, represents the target set value of component A in the feed;

[0105] Reboiler heating capacity feedforward adjustment of feed flow fluctuation:

[0106]

[0107] Where, k Q represents the safety factor, ΔA represents the feed flow deviation, represents the sensitivity of heating capacity to flow;

[0108] Total feedforward adjustment amount:

[0109]

[0110] Where, represents the weight coefficient of reflux ratio feedforward adjustment of feed component fluctuation in the total feedforward adjustment amount, A weight coefficient of reboiler heating amount feedforward adjustment representing feed flow fluctuation in total feedforward adjustment amount;

[0111] Feedback fine-tuning based on real-time purity deviation:

[0112]

[0113] e(t)=y actual (t)-y set ;

[0114] wherein K p represents a proportional coefficient, e(t) represents a purity deviation at time t, K i represents an integral coefficient, represents a deviation cumulative sum from an initial time to time t, K d represents a differential coefficient, represents a change rate of differential deviation, y actual (t) represents an actual purity at time t, y set represents a target purity value;

[0115] Conflict decision priority decision:

[0116]

[0117] wherein ε represents a purity deviation threshold.

[0118] The collaborative control module 33 is used to link key equipment of the distillation system, including distillation column equipment, heat exchange equipment, material conveying equipment, control sensors, actuators and storage equipment, to extend the digital twin to the design, operation, maintenance and decommissioning of the distillation system throughout the life cycle, and to realize digital closed-loop management throughout the process.

[0119] The collaborative control module 33 includes a global optimization module 331, a performance evaluation module 332 and a predictive maintenance module 333; the global optimization module 331 obtains digital twin data of multiple sets of distillation systems in the whole plant area based on the digital twin modeling unit 2, constructs a multi-objective optimization model with the minimum energy consumption and the maximum product yield as the target and the equipment operation constraints, process parameter range, etc. as the constraint conditions based on linear programming, and uses genetic algorithm to iteratively search for the optimal solution to optimize resource allocation.

[0120] The performance evaluation module 332 calculates the key KPIs of the system based on the real-time data of the digital twin model built by the digital twin modeling unit 2, the key KPIs cover the technical, economic and energy consumption performance indicators of the distillation system, the technical indicators include product purity, yield and separation efficiency, the economic indicators include unit product cost and equipment utilization, and the energy consumption indicators include unit product steam consumption and power consumption, the KPI indicators calculated in real time are compared with the indicators set in the design stage, the differences between the actual performance and the design performance of the system are analyzed, the historical data are called to find the historical optimal values of each KPI indicator, the current value is compared with the historical optimal value, the current performance level of the system is evaluated, and the performance decay curve is generated based on linear regression to provide a basis for maintenance.

[0121] The predictive maintenance module 333 predicts the equipment failure risk in advance based on the equipment degradation part of the digital twin model and in combination with real-time sensing data, and generates a maintenance scheme to avoid unplanned shutdown.

[0122] The abnormality monitoring module 34 automatically marks the abnormal position in the digital twin model, generates a visual report, and assists the operator to quickly locate the root cause.

[0123] The present application realizes precise control and optimization of the alcohol production distillation process through the cooperative work of the multi-modal perception and processing unit 1, the digital twin modeling unit 2 and the collaborative optimization unit 3, the multi-modal perception and processing unit 1 first performs multi-dimensional data acquisition, covering traditional parameters, non-invasive sensing and environmental safety sensing, then performs data compression processing, extracts state feature vectors containing equipment health, process performance, etc., and provides input for the subsequent, the digital twin modeling unit 2 builds a model, the mechanism and data fusion module 21 combines physical and chemical mechanisms and real-time and historical data to build and correct the mechanism model, the model self-calibration module 22 adjusts the parameters based on real-time data using Bayesian optimization, calls cloud data to accelerate convergence when major disturbances occur, and realizes precise mapping and dynamic prediction of the distillation process.

[0124] The collaborative optimization unit 3 builds a multi-objective optimization system based on the digital twin model prediction, optimizes energy consumption and yield with purity as a constraint, trains an intelligent agent to output optimal control parameters using a deep deterministic policy gradient algorithm, classifies disturbances through a disturbance identification library, combines digital twin prediction, compensates for disturbances through feedforward-feedback collaborative control, links key equipment, optimizes resource allocation, monitors KPIs and predicts faults based on digital twin data, and realizes full-process digital closed-loop management.

[0125] Although the present application has been described in detail with reference to the foregoing embodiments, the technical solutions recorded in the foregoing embodiments can be modified, or some of the technical features can be replaced by equivalent features, by those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A control system for alcohol production distillation processes based on digital twin modeling, characterized by: The multi-modal perception and processing unit (1), the digital twin modeling unit (2) and the collaborative optimization unit (3) are included. The multi-modal perception and processing unit (1) is used for multi-dimensional sensing data acquisition, state feature vector extraction of the collected data, unreliable data labeling and sensor self-check triggering; the digital twin modeling unit (2) is used for acquiring the collected data of the multi-modal perception and processing unit (1) to construct a digital twin model; and the collaborative optimization unit (3) is used for result prediction in cooperation with the digital twin model and the digital twin modeling unit (2). The collaborative optimization unit (3) includes a multi-objective optimization module (31), a real-time disturbance compensation module (32), a collaborative control module (33) and an abnormality monitoring module (34); the multi-objective optimization module (31) is used for constructing an optimization target system and outputting optimization control parameters; based on the prediction capability of the digital twin model, the real-time disturbance compensation module (32) starts compensation before a disturbance affects product quality, customizes compensation strategies for different disturbance types, and cooperates with feedforward control and feedback control; the collaborative control module (33) extends the digital twin model constructed by the digital twin modeling unit (2) to the whole life cycle of the distillation system by linking key equipment of the distillation system; and the abnormality monitoring module (34) is used for abnormal position labeling in the digital twin model constructed by the digital twin modeling unit (2) and generating a visual report.

2. The alcohol production distillation process control system based on digital twin modeling according to claim 1, characterized in that: The digital twin modeling unit (2) includes a mechanism and data fusion module (21) and a model self-calibration module (22). The mechanism and data fusion module (21) constructs a mechanism model describing the dynamic behavior of the system based on the physical or chemical mechanism of the distillation process, acquires real-time production data and historical data in the multi-modal perception and processing unit (1), corrects the simplification error and parameter uncertainty of the mechanism model, and outputs a digital twin model; and the model self-calibration module (22) acquires real-time data uploaded by the edge nodes of the multi-modal perception and processing unit (1), and dynamically adjusts the parameter weight of the digital twin model based on a Bayesian optimization algorithm.

3. The alcohol production distillation process control system based on digital twin modeling according to claim 2, characterized in that: In the mechanism and data fusion module (21), after constructing a mechanism model describing the dynamic behavior of the system, the key parameters of the mechanism model are initialized based on the design parameters and historical steady-state data, so that the mechanism model outputs results similar to the physical system under standard working conditions.

4. The alcohol production distillation process control system based on digital twin modeling according to claim 2, characterized in that: In the multi-objective optimization module (31), when constructing the optimization target system, the alcohol product purity meeting the standard is taken as a constraint, the minimum unit product energy consumption and the maximum hourly output are optimized synchronously, a three-dimensional objective function is constructed, the multi-objective optimization problem is converted into a single-objective optimization problem based on a constraint weighting method, a deep deterministic policy gradient algorithm is adopted, the digital twin model is trained, the working conditions are simulated, the state data of the digital twin are received in real time, and the optimal control parameters are output.

5. The alcohol production distillation process control system based on digital twin modeling according to claim 4, characterized in that: In the multi-objective optimization module (31), the formula of the three-dimensional objective function is as follows: wherein, represents the energy consumption weight, represents the yield weight, the constraint penalty is 1000 when the purity is < 99.7%, and the penalty term is 500 if the equipment parameters are out of limits.

6. The alcohol production distillation process control system based on digital twin modeling of claim 1, wherein: In the real-time disturbance compensation module (32), customizing compensation strategies for different disturbances includes the following steps: S1, construct a disturbance recognition library: comb the typical disturbance types of alcohol distillation process, extract its feature vector, and match it with the typical features in the disturbance recognition library. For unknown disturbances, identify abnormal patterns through the isolation forest algorithm, mark them as classified disturbances, and trigger manual intervention; S2, digital twin prediction: based on the digital twin model, predict the impact of the disturbance on the key indicators of the distillation system in the next 1-2 minutes. Input the identified disturbance parameters into the digital twin model, correct the disturbance parameters, perform short-term and medium-term prediction, correct the model parameters, and control the prediction error; S3, feedforward-feedback collaborative action: based on the prediction results of the digital twin, combined with real-time measured data, through feedforward control and feedback control collaborative action, block the disturbance from spreading to product quality, and adjust.

7. The alcohol production distillation process control system based on digital twin modeling according to claim 6, characterized in that: In step S2, the prediction error control formula is as follows: Wherein, y actual (t) represents the measured key indicators at time t, y pred (t) represents the digital twin prediction value, M represents the number of sampling points within 10 seconds, and when the prediction error e RMSE exceeds the threshold value, the model parameter online correction is triggered.

8. The alcohol production distillation process control system based on digital twin modeling according to claim 6, characterized in that: In step S3, when blocking the disturbance from spreading to product quality, at the initial stage of disturbance, feedforward control dominates, blocking disturbance spread, after 30 seconds, feedback control dominates, correcting deviation, when the adjustment direction of feedforward and feedback conflicts, product purity priority principle is used for decision-making.

9. The alcohol production distillation process control system based on digital twin modeling of claim 6, wherein: In step S3, the specific formula of feedforward-feedback collaborative action is as follows: Reflux ratio feedforward adjustment of feed component fluctuation: where k R represents the proportional coefficient of the reflux ratio feedforward adjustment, Δx A represents the feed component deviation, represents the sensitivity of the reflux ratio to the feed component, x A (t) represents the feed component content at time t, represents the target set value of component A in the feed; Reboiler heating capacity feedforward adjustment of feed flow fluctuation: where k Q represents the safety factor, ΔA represents the feed flow rate deviation, represents the sensitivity of the heating amount to the flow rate; Total feedforward adjustment amount: wherein, represents a weight coefficient of reflux ratio feed forward adjustment in the total feed forward adjustment amount for feed component fluctuation, represents a weight coefficient of reboiler heat feed forward adjustment in the total feed forward adjustment amount for feed flow fluctuation; Feedback fine-tuning based on real-time purity deviation: e(t) = y actual (t) - y set ; wherein K p represents a proportional coefficient, e(t) represents a purity deviation at time t, K i represents an integral coefficient, represents a deviation cumulative sum from an initial time to time t, K d represents a differential coefficient, represents a change rate of the differential deviation, y actual (t) represents an actual purity at time t, y set represents a target purity value; Conflict decision priority decision: Wherein, ε represents the purity deviation threshold.

10. The alcohol production distillation process control system based on digital twin modeling of claim 1, wherein: The collaborative control module (33) includes a global optimization module (331), a performance evaluation module (332), and a predictive maintenance module (333); The global optimization module (331) obtains digital twin data of multiple distillation systems in the whole plant area based on the digital twin modeling unit (2), and optimizes resource allocation; the performance evaluation module (332) calculates the key KPI of the system in real time, and compares it with the design index and historical optimal value to generate a performance decay curve; the predictive maintenance module (333) predicts the equipment failure risk based on the equipment degradation part of the digital twin model, combined with real-time sensing data collected by the multi-modal perception and processing unit (1), and generates a maintenance scheme.

Images