Control method, device and equipment for industrial production process and medium

Abstract

The invention discloses a control method, device and equipment for an industrial production process and a medium. The method comprises the following steps: acquiring physical entity data and production process data of a target rectifying tower; according to the physical entity data and the production process data, constructing a digital twinborn model of the target rectifying tower; determining a control strategy of the target rectifying tower according to the production process data based on a digital twinborn model; and controlling the production process of the target rectifying tower according to the control strategy. In the technical scheme of the embodiment of the invention, the control strategy conforming to the current actual condition of the target rectifying tower is determined through prediction practice, simulation test and the like in the digital twinborn model, so that the condition of directly and repeatedly adjusting actual production equipment is reduced, the frequency of repeatedly activating constraints of various parameters of the production equipment is reduced, and the production efficiency is improved. And more attempt cost can be saved, so that the energy consumption is reduced, and the production efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of process control technology, and in particular to a method, apparatus, equipment and medium for controlling an industrial production process. Background Technology

[0002] Continuous chemical separation processes are widespread in industry, with coal tar fractionation units being a typical example. These units operate under conditions of strong non-ideal thermodynamics, strong coupling of multiple components, and the coexistence of operating load and hydraulic safety boundaries. Online component analysis typically suffers from minute-level delays and intermittent sampling, leading to state estimation being constrained by information lag and data quality fluctuations.

[0003] In engineering, model predictive control and soft measurement are often used to achieve specification control and energy efficiency optimization. However, when the model and the equipment are mismatched, or when the model enters the area outside the training data coverage under disturbance and material change conditions, the prediction is prone to non-physical states (such as negative composition and out-of-phase points), resulting in frequent constraint activation, increased fluctuations in energy consumption and quality indicators, and poor effect on stable and economical operation. How to reduce the frequency of constraint activation on actual production equipment, how to reduce energy consumption, and how to improve production efficiency have become one of the key research focuses for relevant technical personnel. Summary of the Invention

[0004] This application provides a method, apparatus, equipment, and medium for controlling industrial production processes, thereby reducing the frequency of constraint activation of actual production equipment, reducing energy consumption, and improving production efficiency.

[0005] According to one aspect of this application, a method for controlling an industrial production process is provided, comprising:

[0006] Obtain the physical entity data and production process data of the target distillation column;

[0007] A digital twin model of the target distillation column is constructed based on physical entity data and production process data.

[0008] Based on the digital twin model and production process data, the control strategy for the target distillation column is determined.

[0009] The production process of the target distillation column is controlled according to the control strategy.

[0010] According to another aspect of this application, a control device for an industrial production process is provided, comprising:

[0011] The data acquisition module is used to acquire the physical entity data and production process data of the target distillation column;

[0012] The digital twin module is used to construct a digital twin model of the target distillation column based on physical entity data and production process data;

[0013] The strategy generation module is used to determine the control strategy for the target distillation column based on the digital twin model and production process data.

[0014] The strategy execution module is used to control the production process of the target distillation column according to the control strategy.

[0015] According to another aspect of this application, an electronic device is provided, the electronic device comprising:

[0016] At least one processor; and

[0017] A memory communicatively connected to the at least one processor; wherein,

[0018] The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the industrial production process control method described in any embodiment of this application.

[0019] According to another aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the control method of the industrial production process described in any embodiment of this application.

[0020] According to another aspect of this application, a computer program product is provided, the computer program product including a computer program that, when executed by a processor, implements the control method for an industrial production process according to any embodiment of this application.

[0021] In the technical solution of this application embodiment, a digital twin model of the target distillation column is constructed based on the physical entity data and production process data of the target distillation column. This model provides a digital twin simulation, modeling, and prediction environment for the control of the target distillation column, providing fundamental support for determining a better control strategy. Based on the digital twin model and the production process data, a control strategy for the target distillation column is determined to control the production process of the target distillation column. Through predictive practices and simulation tests in the digital twin model, a control strategy that conforms to the current actual situation of the target distillation column is determined. This reduces the need for repeated adjustments to the actual production equipment and lowers the frequency of repeated activation of constraints on various parameters of the production equipment, thereby saving more trial costs, reducing energy consumption, and improving production efficiency.

[0022] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of a control method for an industrial production process according to Embodiment 1 of this application;

[0025] Figure 2 This is a flowchart of a control method for an industrial production process according to Embodiment 2 of this application;

[0026] Figure 3A This is a schematic diagram of the architecture and data flow of the target distillation column and its control process provided in Embodiment 3 of this application;

[0027] Figure 3B This is a schematic diagram of a thermodynamically feasible manifold and a differentiable projection according to Embodiment 3 of this application;

[0028] Figure 3C This is a schematic diagram of the time-delay augmented moving window state estimation window aligned with the timing, according to Embodiment 3 of this application;

[0029] Figure 3D This is a schematic diagram illustrating the linkage between online optimization within the feasible domain and "coverage mapping to robust margin" according to Embodiment 3 of this application;

[0030] Figure 4 This is a schematic diagram of the structure of a control device for an industrial production process according to Embodiment 4 of this application;

[0031] Figure 5 This is a schematic diagram of the structure of an electronic device that implements the control method for an industrial production process according to the embodiments of this application. Detailed Implementation

[0032] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0033] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0034] Example 1

[0035] Figure 1 This application provides a flowchart of a control method for an industrial production process, as shown in Embodiment 1. This embodiment is applicable to situations involving digital twin simulation and control of a distillation column's production process. The method can be executed by a control device for the industrial production process, which can be implemented in hardware and / or software and can be configured in an electronic device. For example... Figure 1 As shown, the method includes:

[0036] S110. Obtain the physical entity data and production process data of the target distillation column.

[0037] The distillation column is a tower-type gas-liquid contact device used for distillation. Utilizing the different volatility of the components in a mixture—that is, the different vapor pressures of the components at the same temperature—lighter components (low-boiling compounds) in the liquid phase are transferred to the gas phase, while heavier components (high-boiling compounds) in the gas phase are transferred to the liquid phase, thus achieving separation. The target distillation column can be the controlled object in this application, with the aim of guiding its production through digital twin simulation and control. Therefore, the embodiments and implementation methods of this application will be described in terms of the control of the target distillation column.

[0038] Physical entity data can refer to the data related to the various hardware components of the target distillation column itself, i.e., its design and equipment data, including but not limited to the structure, dimensions, and connection methods of each component. Production process data can be various related data generated by the target distillation column during the execution of production tasks, including but not limited to material composition data, thermodynamic data, implementation process data, product quality data, and control process data. It is understood that the relevant physical entity data is already available during the initial design and construction of the distillation column, while production process data can be acquired through different types and functions of sensors deployed at various stages within the target distillation column.

[0039] S120. Based on physical entity data and production process data, construct a digital twin model of the target distillation column.

[0040] Digital twins, in particular, fully utilize data from physical models, sensors, and operational history to integrate multi-disciplinary, multi-physical-quantity, multi-scale, and multi-probabilistic simulation processes, completing mapping in virtual space to reflect the entire lifecycle of the corresponding physical equipment. Digital twins are a concept that transcends reality, and can be viewed as a digital mapping system of one or more important, interdependent equipment systems. In this embodiment, the digital twin model of the target distillation column is constructed and simulated based on the physical entity data and production process data obtained in the aforementioned steps.

[0041] It's important to note that there are three main categories of methods for constructing digital twin models. First, mechanism-driven models, which are based on physicochemical principles such as mass conservation, energy conservation, momentum transfer, and phase equilibrium, establishing rigorous mathematical equations. This category includes building steady-state or dynamic models in specific software or developing customized models. These models have strong extrapolation capabilities and are particularly suitable for analyzing new plants or operating conditions exceeding historical data. Dynamic models are crucial for simulating start-up, shutdown, and major disturbances. Second, data-driven models are very useful when certain key variables in a distillation column (such as product composition) are difficult to measure online. They utilize historical process data (such as temperature, pressure, and flow rate) to build predictive models. For example, dynamic modeling combined with Kalman filtering can quickly predict state variables under non-design conditions. Deep learning models, such as Long Short-Term Memory (LSTM) networks, excel at capturing complex nonlinear dynamic features from data. These methods typically do not require complete physical property data and have short computation cycles, but they rely heavily on data quality and quantity. Third, the hybrid-driven model (a combination of mechanism-driven and data-driven approaches) combines the advantages of both: it uses a mechanistic model to describe the fundamental laws of the process, while simultaneously utilizing real-time data to correct and update the model online. For example, a semi-rigorous mechanistic model can be established as a virtual analyzer, and then actual production data can be used to provide feedback correction. In distillation process optimization, there are also studies that construct digital twins through collaborative modeling of chemical mechanisms and industrial data. This approach allows the model to continuously approximate the operating state of real equipment while maintaining physical meaning. The embodiments of this application preferably employ a hybrid model combining a mechanistic kernel with residual correction.

[0042] The above three methods allow for the integration of models into simulation software or the development of data models using machine learning platforms. Through industry-standard protocols, the model is connected to the supporting control system and real-time database to achieve bidirectional data flow. Historical or real-time data is used to calibrate the model, and its accuracy is verified by comparing predicted and measured values ​​for key parameters (such as product composition and temperature distribution). Finally, the calibrated model is deployed on a cloud platform or local server for continuous operation. This application does not limit the method for constructing the digital twin model, but it preferably employs a hybrid-driven model, that is, simultaneously constructing the digital twin model of the target distillation column through both mechanism-driven and data-driven approaches.

[0043] S130. Based on the digital twin model, determine the control strategy for the target distillation column according to the production process data.

[0044] Digital twin models simulate production processes constrained by physical and / or chemical mechanisms, while production process data reflects the production conditions within the target distillation column. Within the constructed digital twin model, different control strategies are developed based on changes in production conditions. Understandably, these changes in control strategies aim to better match the production status of the target distillation column. For example, analysis of production process data can be performed, and various possible strategies can be predicted or simulated within the digital twin model to determine the optimal control strategy.

[0045] S140. Control the production process of the target distillation column according to the control strategy.

[0046] The control strategy obtained through testing using a digital twin model is used to adjust and guide the production of the target distillation column in reality.

[0047] In the technical solution of this application embodiment, a digital twin model of the target distillation column is constructed based on the physical entity data and production process data of the target distillation column. This model provides a digital twin simulation, modeling, and prediction environment for the control of the target distillation column, providing fundamental support for determining a better control strategy. Based on the digital twin model and the production process data, a control strategy for the target distillation column is determined to control the production process of the target distillation column. Through predictive practices and simulation tests in the digital twin model, a control strategy that conforms to the current actual situation of the target distillation column is determined. This reduces the need for repeated adjustments to the actual production equipment and lowers the frequency of repeated activation of constraints on various parameters of the production equipment, thereby saving more trial costs, reducing energy consumption, and improving production efficiency.

[0048] Example 2

[0049] Figure 2This is a flowchart illustrating a control method for an industrial production process according to Embodiment 2 of this application. This embodiment further refines the process of determining the control strategy for the target distillation column based on the aforementioned embodiments. For example... Figure 2 As shown, the method includes:

[0050] S210. Obtain the physical entity data and production process data of the target distillation column.

[0051] S220. Based on physical entity data and production process data, construct a digital twin model of the target distillation column.

[0052] S230. Based on the digital twin model, the thermodynamically feasible manifold is determined according to the production process data.

[0053] Thermodynamically feasible flow can be a theoretically feasible domain boundary that consists of non-negative components that sum to 1, has bounded phase fractions, and satisfies phase equilibrium consistency (e.g., Rachford-Rice surface) and thermodynamic consistency (e.g., Gibbs–Duhem / Maxwell consistency).

[0054] Production process data can mainly include two categories: fast process quantities, such as temperature, pressure, flow rate, and liquid level in the target distillation column. Flow rates can be measurable online, including feed flow, reflux flow, bottom product flow, and top distillate flow. The other category is slow component quantities, such as gas chromatography and mass spectrometry data. Under the constraints of fast process quantities and slow component quantities, a thermodynamically feasible manifold is formed that at least satisfies the following requirements: composition non-negativity and normalization, bounded phase fraction, phase equilibrium consistency, and thermodynamic consistency.

[0055] For example, in continuous chemical separation devices such as distillation systems containing thermosiphon reboilers and total condensers, the method handles two types of asynchronous measurements: controlling and estimating the rolling cycle Δt. c The acquired fast process quantity, and the quantity with a fixed time delay d a With sampling time Δt a The amount of slow-moving components. Based on the amounts of fast-moving and slow-moving components, a thermodynamically feasible manifold is defined. It must at least satisfy the following conditions: non-negative and normalized composition, bounded phase fraction, consistent phase equilibrium, and thermodynamic consistency. Preferably, it is expressed as:

[0056] in, Represents the liquid phase mole fraction vector. . For gas phase fraction, T represents temperature, and P represents pressure. Represents the vector of feed (or overall) composition. Phase equilibrium partition / volatility coefficient (which can be given by the activity / fugacity relationship and baseline thermodynamics). The activity coefficient, For thermodynamic consistency.

[0057] S240. Based on the preset time-delay augmented state space, perform state estimation on the data of each production process to obtain each candidate state.

[0058] The time-delay augmented state space can be a novel, delay-free, but higher-dimensional state space model that transforms the inherent time delay in the system by introducing new augmented state variables. The construction of this time-delay augmented state space can employ any method from related technologies, and this application does not limit this approach. Candidate states can be the state estimation results obtained by combining fast process quantities and slow component quantities within a sliding window using the time-delay augmented state space. Candidate states can provide information on the current production status of the target distillation column for subsequent control strategy development.

[0059] Furthermore, in an optional implementation, the step of S240, which involves estimating the state of each production process data according to a preset time-delay augmented state space to obtain each candidate state, may include: solving the state trajectory of the production process data within a preset sliding window according to the preset time-delay augmented state space to obtain each candidate state.

[0060] Understandably, time-delay augmented moving window state estimation can provide an estimate of the current state of the entire target distillation column's production system. However, due to errors in various sensor detection methods, this estimate also contains errors, resulting in uncertainties (such as the covariance of the state estimate, the probability distribution of parameters, and the statistical characteristics of measurement noise). Uncertainty propagation can be used to extend this estimate into multiple different samples, which can then be used as candidate states. For example, suppose time-delay augmented moving window state estimation estimates the current coal tar concentration at the top of the target distillation column to be 85% (this is an estimate). However, due to potential errors, this value may fluctuate between 84% and 86%. Based on this, a series of possible coal tar concentration "samples" are generated (e.g., 84.1%, 84.5%, 85.3%, 85.8%, etc.). Then, a mechanistic model is used to predict how these different initial concentrations will evolve in future operations, resulting in a distribution of predicted states. This distribution quantifies the future uncertainty. Of course, the extension of multiple state samples can be generated using methods such as Monte Carlo, unscented transformation, or model ensemble. These samples represent multiple possibilities of the production state in the target distillation column.

[0061] In a specific example, a time-delay augmented state space is constructed, and the estimation optimization is solved jointly using fast and slow measurements within a sliding window. A typical objective function is:

[0062]

[0063] in, It can be a state vector (for example, it can include state variables such as component concentration, temperature, and pressure of each tray in the tower). For manipulated quantities or control inputs (such as reboiling heat load, reflux ratio, etc.). These are rapid process measurements (such as temperature, pressure, and flow rate) that can be acquired at high frequencies by sensors. This refers to the measurement values ​​of slow-moving components acquired at low frequencies through online analytical instruments (such as the concentrations of key components obtained through gas chromatography analysis). This is the set of indexes for component measurements that actually arrive within the window; missing or unarrived component measurements are zeroed out using a masking matrix; resynchronization is performed when a gas chromatograph or gas chromatography-mass spectrometry analyzer restart or clock drift is detected, the timestamp is backdated and a reference frame is inserted within the window; soft measurements can be introduced as bridging observations during the analysis window. This represents a fast measurement function (or observation model) that operates based on the state vector at time j. This function calculates the model predictions for the corresponding rapid process measurements. For example, the function could be a physical or empirical model describing how the temperature of a tray is determined by the liquid composition and pressure of that tray. This represents a slow measurement function (or observation model) that operates based on the state vector at time j. The function calculates the model predictions for the corresponding slow-moving component measurements. For example, this function could be a mechanism or data-driven model describing how the concentration of a product at the top of the column is determined by the state at the top. The overall objective function aims to obtain the optimal estimate of the current system state by minimizing the prediction error of the state trajectory and the deviation between the model and the actual measurements.

[0064] In the above embodiments, by using time-delay augmented state space, the state trajectory of the production process data is solved within a preset sliding window to obtain candidate states. This can provide a predictive estimate of the current production state in the target distillation column, providing a good foundation for formulating control strategies for the current production state and helping to improve the production efficiency of the target distillation column.

[0065] S250. Perform feasibility operations on each candidate state according to the thermodynamic feasible manifold to obtain each state to be processed.

[0066] Understandably, the thermodynamically feasible manifold determined in the preceding steps defines the feasible region of the production state in the target distillation column. However, the samples of each candidate state obtained in the preceding steps may deviate from the feasible region. Through the feasibility operation, all these candidate states are constrained to the thermodynamically feasible manifold. The result after forced feasibility is the state to be processed. These states to be processed, like the candidate states, can exist as state vectors. The feasibility operation can be implemented using a differentiable projection operator.

[0067] In a further optional embodiment, S250 performs a feasibility operation on each candidate state according to the thermodynamically feasible manifold to obtain each state to be processed. This may include: constructing a differentiable projection operator of each candidate state in the differentiable thermodynamically feasible domain within the thermodynamically feasible manifold to perform a feasibility operation to obtain each state to be processed.

[0068] For example, based on the thermodynamically feasible manifolds determined in the foregoing embodiments,

[0069] Constructing differentiable projection operators For any candidate state Solve the closest point problem with simple bounds and consistency with equality:

[0070]

[0071] in, This represents the uncorrected state vector. It is a weighted matrix used to emphasize the relative importance of different components. For example, if the error of a component is more critical than the error of temperature, then the corresponding weight is greater. This weighted matrix can be preset by relevant technical personnel. The results can be obtained using implicit function theorems, etc., and are not limited here. Newton-Lagrange, trust-region, or equivalent constraint solvers are preferred, with numerical backoff (in-bounded least squares) implemented to ensure stability far from the feasible region. Thus, "non-negative composition, in-phase feasibility, and thermodynamic consistency" become the model's structural properties, and any subsequent estimations, predictions, and optimizations are performed within the feasible manifold. The process is carried out within the feasible manifold. By applying a differentiable thermodynamic feasible domain projection to the candidate states, the estimated trajectory is ensured to lie within the feasible manifold. In the above implementation, multiple states in the uncertain information are made feasible, thereby providing a basis for subsequent strategy determination.

[0072] S260. Determine the control strategy based on the status to be processed and the production process data.

[0073] It is understood that the state data to be processed is confined within a thermodynamically feasible manifold. This state data represents the current possible production state of the target distillation column, and the optimal way to control the target distillation column is determined based on various data generated during the production process. For example, a pre-trained machine learning model can be used. This model takes the state vector of the state to be processed and the production process data as input, and outputs a control strategy for the target distillation column. Of course, this machine learning model can be trained using historical state data, production process data, and control strategies. This application does not limit the selected machine learning model or the training process.

[0074] In another optional implementation, the step of determining the control strategy based on the state to be processed and the production process data in S260 may include:

[0075] A1. Generate prediction samples based on each pending status and production process data.

[0076] The predicted samples can be the uncertainty statistics (including mean and variance, etc.) for each state to be processed, as described in the previous example. Furthermore, for uncertainties in parameters, disturbances, and measurement noise, Monte Carlo methods, unscented transformations, or model ensembles are used to generate samples, and these samples are progressively applied in the prediction time domain.

[0077] ;

[0078] in, It is a mechanistic model function.

[0079] All sample states are forced to be feasible at each step, and the mean, variance, quantiles and out-of-limit probabilities of the controlled variables and constraint functions are calculated accordingly. Optionally, the Sobol index is calculated to identify the dominant uncertainty sources, providing a basis for subsequent robustness margin determination.

[0080] A2. Determine the robustness margin based on the state to be processed, the predicted samples, and the preset coverage.

[0081] Here, robustness margin can be the safety boundary corresponding to uncertain information, or it can be understood as the limit that cannot be exceeded in uncertain information. For example, since each state to be processed is a quantified result of uncertainty, these states are all possible. The purpose of this implementation is to transform the operational constraint from a probabilistic problem into a deterministic problem that can be directly processed. For example, taking the temperature in the target distillation column as an example, the problem of "there is a 95% probability that the temperature will not be exceeded" cannot be directly processed; it requires a clear temperature upper limit, such as "it cannot exceed 150°C". The purpose of this implementation is to calculate this clear upper limit.

[0082] Coverage can be a preset confidence level or security level, representing the probability that the desired constraint will be satisfied, for example... =0.99 means that the operation must not violate the constraint in 99% of cases.

[0083] Robustness margin is calculated using the pending state, predicted samples, and preset coverage, specifically as follows, with chance constraints... Transformed into deterministic equivalence:

[0084]

[0085] in, It is the first Constraint functions (e.g., mass / hydraulics / actuators, etc.). For coverage (confidence level, ), It is the confidence coefficient corresponding to the coverage. For the first The robustness margin of a constraint is as follows: A monotonically increasing robust margin function.

[0086] and From The propagation of uncertainty within, among which, The predicted mean on the feasible manifold; During operation, the system adaptively adjusts based on data quality, disturbance intensity, and estimated residuals. and For safety-critical constraints such as hydraulics, a higher basic margin and slower convergence can be set, while a relatively aggressive strategy is allowed for economically oriented quality constraints. When multiple constraints compete, safety constraints should be given priority, and orderly degradation can be achieved through soft constraint relaxation.

[0087] Continuing the previous example, suppose there is an operational constraint that the top temperature of the target distillation column cannot exceed 155°C, and the coverage is set... The robustness margin calculated using the above method tightens the original 155℃ constraint into a deterministic constraint of 151℃. Therefore, it's conceivable that as long as the explicit constraint of "tower top temperature not exceeding 151℃" is followed during production, it can be indirectly guaranteed that the actual tower top temperature will not exceed 155℃ in 99% of cases. In essence, this is a process of adaptively mapping coverage to an operable robustness margin.

[0088] A3. Generate control strategies based on each state to be processed, the predicted samples, and the robustness margin.

[0089] Based on the aforementioned states to be processed, predicted samples, and robustness margins, it is determined how to adjust the various parameters in the target distillation column within the constraints of the robustness margins. For example, a control strategy can be generated using a pre-trained machine learning model. This model takes the aforementioned states to be processed, predicted samples, and robustness margins as input and outputs a control strategy. Of course, the machine learning model can be trained based on historical state vectors, predicted samples, robustness margins, and historical control strategies. This application embodiment does not limit the type of machine learning model or the training process.

[0090] In another alternative implementation, the generation of the control strategy based on each state to be processed, the predicted sample, and the robustness margin, as described in A3, may include:

[0091] B1. Determine the deterministic constraints corresponding to the robustness margin.

[0092] The robustness margin limit can be a deterministic constraint. For example, in the aforementioned implementation, "the top temperature does not exceed 151°C" is a deterministic constraint. Of course, the temperature deterministic constraint is just one of many parameters in the target distillation column; each parameter can correspond to a deterministic constraint. These deterministic constraints are used to determine the subsequent control strategy.

[0093] B2. Based on the state to be processed, the predicted samples, and the deterministic constraints, determine the control strategy within the feasible manifold.

[0094] Based on the pending state, predicted samples, and deterministic constraints, online optimization can be achieved by solving the problem in the rolling prediction time domain, as follows:

[0095]

[0096]

[0097] in, It refers to energy consumption or equivalent economic indicators. This refers to specification deviation (error in achieving product quality targets). , and These are the weighting coefficients of the cost function, which can be preset by relevant personnel. To control the increment, This is the upper limit of the increment.

[0098] To balance multiple objectives, NSGA-III (third-generation non-dominated sorting genetic algorithm) or the reference point method can be used to generate Pareto candidates within the feasible region; all candidates are first evaluated... Forced feasibility: The controller selects the optimal option based on the operating strategy (safety priority / energy efficiency priority) and avoids abrupt control changes through smooth switching. When a set of constraints is detected to be temporarily infeasible, hierarchical relaxation is triggered to maintain hard safety constraints, while slack variables are introduced into quality constraints and penalized with high weights.

[0099] S270. Control the production process of the target distillation column according to the control strategy.

[0100] The control strategy determined in the preceding steps is output from the digital twin model to the control system of the target distillation column, thereby controlling the corresponding adjustments of each link and each piece of equipment in the target distillation column.

[0101] For example, after optimization, including amplitude limiting, anti-saturation, and rate limiting, the output is transmitted via a preset industrial protocol (preferably OPC-UA) at a speed of Δt. c The cycle is sent to the controller; maintaining uniform clock and timestamp consistency. When a communication anomaly, analyzer disconnection, or clock drift exceeding the threshold is detected, it automatically enters a safety mode: increasing coverage and margin, tightening control increments, freezing secondary adjustments and reverting to conservative settings if necessary; when data quality recovers and consistently meets the threshold, it smoothly exits the safety mode using a preset ramp. On the other hand, the entry conditions for full mode include: communication timeout (preferably 0.5–1.5s), online analysis disconnection (preferably greater than 1.2–1.5 × Δt). a If the predicted residual exceeds the threshold (preferably 3–4 times the median absolute deviation in steady state), the coverage is increased to 0.97–0.99 upon entry, the margin is increased by 20%–40%, and the control increment is tightened by 30%–50%. After the data quality is restored, the system exits smoothly with a 5–15 minute ramp.

[0102] In the technical solution of Embodiment 2 of this application, a feasible region is constructed that satisfies the requirements of non-negative and normalized composition, bounded phase fraction, and consistency between phase equilibrium and thermodynamics. The estimated and predicted states are forcibly confined within this feasible region by differentiable projection. Combining the mechanistic model and multiplicative residuals, a time-delay-augmented moving window state estimation is established to uniformly handle the fixed time delay and intermittent sampling of millisecond-level process quantities and minute-level online components. Uncertainty propagation is performed within the feasible region, and the coverage is adaptively mapped to an operable robust margin. This enables online optimization and control of the target distillation column. This method can effectively reduce the risk of exceeding limits and improve energy efficiency and stability, thereby increasing the efficiency of production control based on digital twin models.

[0103] Example 3

[0104] The embodiments in this application are supplementary descriptions based on the foregoing embodiments and implementation methods.

[0105] For the digital twin model involved in this application, a physical information hybrid model (mechanism kernel + residual) is preferably adopted. A mechanism kernel incorporating material / energy balance, phase balance, and equipment constraints is established. The thermodynamic baseline is preferably UNIFAC (Universal Quasi-chemical Functional Group Activity Coefficient) or COSMO-RS (Conductor-like Screening Model for Real Solvents) to provide phase equilibrium partitioning / volatility coefficients. To compensate for the strong non-ideal nature of industrial raw materials and the model-plant mismatch, a multiplicative residual correction (exponential family or lightweight neural network) for the activity coefficient or its logarithm is introduced, and positive and asymptotic boundaries are applied to ensure the physical rationality of the extrapolation region. The mechanism and residuals are explained through... Structural coupling forms an interpretable and extrapolable hybrid model of physical information.

[0106] Figure 3A This is a schematic diagram of the architecture and data flow of the target distillation column and its control process. Sensors and online analyzers connected to the continuous separation unit (distillation column) transmit rapid measurements (period Δt) through an industrial communication gateway. c ) and slow component measurement (beat Δt) a Delay d a The data is fed into the circular buffer and clock synchronization module of the computing platform; the computing platform includes, in sequence: feasible region projection module ( The system includes a delay augmentation module (Moving Horizon Estimation, MHE), an uncertainty propagation module, a coverage mapping to robustness margin mapping module, a feasible domain online optimization module, and an interface and safety mode module. The optimized output is then sent to a DCS (Distributed Control System) or PLC (Programmable Logic Controller) for closed-loop execution after being limited by amplitude / rate.

[0107] Figure 3B For thermodynamically feasible manifolds With differentiable projection A schematic diagram. It is constructed using the simplex and phase fraction intervals. The boundary is superimposed with Rachford–Rice surface and Gibbs–Duhem / Maxwell consistency constraints; any candidate state s is passed through the projection operator. Mapped to The nearest feasible point To avoid negative composition, out-of-phase points, and inconsistent potential functions.

[0108] Figure 3C A schematic diagram illustrating the alignment of the state estimation window with the timing sequence for a time-delay augmented moving window. (Window) Within, the rapid measurement frame is measured according to Δt. c Arrangement; slow component frames by d a Delayed alignment and zeroing of "missing / absent" frames using a masking matrix; when analyzer restart or beat drift is detected, a "resynchronization event" is inserted and the timestamp is backtracked for correction; the consistency between propulsion and measurement is estimated in the projection. Feasible manifolds Completed within the specified time.

[0109] Figure 3D For feasible online optimization and "coverage" Mapped to robust margin A diagram illustrating the linkage. Internal uncertainty propagation output With the probability of exceeding the limit; the margin mapping module maps the coverage. Convert to robust margin Forming deterministic constraints Online optimization generates Pareto candidates within the feasible region and selects the best one. The optimized output is then sent out in a closed loop after being limited / anti-saturation / rate-limited. In case of anomalies, it enters a safe mode. promote, (Increase, tighten Δu), and smoothly exit after data quality is restored.

[0110] In addition to the above Figures 3A to 3D In addition to the illustrated explanations, to enable those skilled in the art to more clearly understand the technical solution of this application, the following will use a specific industrial scenario as an example, such as the optimized control of a 100,000-ton / year coal tar distillation tower in a petrochemical company, to explain in detail the complete implementation process of this application. The main task of this distillation tower is to separate coal tar, increase the yield of the target product light phenolic oil, and reduce the residual effective components in the heavy asphalt at the bottom of the tower, while maintaining the energy consumption of the unit at a low level.

[0111] First, acquire physical entity data and production process data.

[0112] (1) Physical entity data collection: First, collect the design and construction data of the target distillation column, mainly including:

[0113] ① Geometric parameters: The tower is 30 meters high and 2.5 meters in diameter, containing 50 layers of floating valve trays. The feed inlet is located on the 25th tray.

[0114] ② Material and equipment information: The tower body is made of 316L stainless steel, the reboiler is a thermosiphon type, the condenser is a total condensing type, and the model and characteristic data of the pipes, pumps and valves connected to the tower body are also provided.

[0115] (2) Production process data collection: Two types of real-time and historical operational data are acquired through the factory's distributed control system (DCS) and real-time database (PIMS):

[0116] ① Rapid process parameters: These are collected via temperature, pressure, and flow sensors at a high frequency. Specifically, they include: temperature at the top, bottom, and key trays of the column (every 5 seconds), pressure at the top of the column (every 5 seconds), feed flow rate, reflux flow rate, and top product output (every 10 seconds).

[0117] ② Slow-speed component concentrations: These are collected via online analytical instruments, with a low frequency and significant time delay. Specifically, this includes the concentrations of key components (such as phenol and naphthalene) in the light phenolic oil at the top of the tower and the bitumen at the bottom. The data is provided by an online gas chromatograph, updated every 30 minutes, with a transmission and analysis delay of approximately 5 minutes.

[0118] Second, construct a hybrid-driven digital twin model.

[0119] For this coal tar distillation column, a hybrid model combining a mechanistic kernel and data-driven residual correction is constructed.

[0120] (1) Mechanism kernel construction: Based on industrial control simulation software in related technologies, a rigorous mechanism model is established. This model contains the following core equations:

[0121] Mass conservation equation: For each component i on each tray, establish a dynamic mass balance equation:

[0122] ;

[0123] in, Components On the tray The molar liquid holdup on the surface; Represents time; and Representing from the tray The liquid phase molar flow rate and the liquid leaving the tray The liquid phase molar flow rate; and Representing the components respectively On the tray and The liquid phase mole fraction; and Representing from the tray Rising gas phase molar flow rate and exit from the tray Vapor phase molar flow rate; and Representing the components respectively On the tray and The gas phase mole fraction; Representative enters the tray The feed molar flow rate; Representative of feed components The mole fraction.

[0124] For each tray j, the dynamic energy balance equation is established as follows:

[0125]

[0126] in, Refers to trays The total enthalpy value; and Representing from the tray Flowing down and leaving the tray The liquid phase molar enthalpy; and Representing from the tray Ascending and leaving the tower plate e Vapor phase molar enthalpy; Representative enters the tray The feed molar enthalpy; the meanings of other symbols are the same as those defined in the mass conservation equation. The phase equilibrium equation is:

[0127] in: Components On the tray The gas phase mole fraction; Refers to components On the tray The liquid phase mole fraction; Refers to components On the tray The gas-liquid phase equilibrium constant.

[0128] In describing the equilibrium relationship between a gas and a liquid phase, the phase equilibrium constant is... The calculation is crucial. Given the strong non-ideality of the coal tar system, the UNIFAC (Universal Quasi-Chemical Functional Group Activity Coefficient) model is chosen to calculate the liquid phase activity coefficient, thereby obtaining... .

[0129] (2) Data-driven residual correction: To compensate for the mismatch between the mechanism model and the actual plant operation (such as inaccurate physical property parameters, changes in tray efficiency, etc.), a data-driven correction module is introduced.

[0130] A Long Short-Term Memory (LSTM) network was selected. This network has two hidden layers, each containing 64 neurons. The input to the LSTM network is a historical sequence of key states (e.g., temperature distribution) and operational variables (e.g., reflux ratio, heat load) within the tower over a past period (e.g., 60 minutes). Its output is a function for the phase equilibrium constant. Multiplicative correction factor .

[0131] The final phase equilibrium relationship was revised to ;

[0132] By continuously training and updating the LSTM network with real-time data, the digital twin model can continuously approximate the operating state of the real device.

[0133] Third, control strategies are determined based on digital twin models.

[0134] This step is performed periodically on the computing server in the control system backend (e.g., every 5 minutes).

[0135] (1) State estimation and uncertainty propagation:

[0136] Using the aforementioned time-delay augmented moving window state estimation (MHE) method, high-frequency temperature, pressure, and flow rate data are fused with low-frequency, time-delayed component data to obtain the probability distribution of temperature, pressure, and liquid phase composition for each tray in the current column. For example, the system estimates the mean temperature at the top of the column to be 155.2℃ with a standard deviation of 0.5℃.

[0137] The Unscented Transform (UT) method is used to propagate the uncertainty of the current state (represented by the mean and covariance matrices) forward by N prediction steps through a digital twin model (e.g., predicting the system state in the next 30 minutes).

[0138] (2) The calculation of robustness margin includes the following aspects:

[0139] Constraint identification: A key safety constraint is that the tower top temperature must not exceed 156°C to prevent product decomposition and coking. This is a hard constraint.

[0140] Coverage settings: To ensure high security, set the coverage. This means that the constraint must not be violated in 99% of cases.

[0141] Margin calculation: The mean of the predicted trajectory of the future tower top temperature obtained from uncertainty propagation. and standard deviation and coverage (Confidence coefficient corresponding to normal distribution) ), constraining probabilistic chance Transform it into a deterministic constraint: Here This refers to the robustness margin for safe operation.

[0142] (3) Online optimization and control strategy generation:

[0143] Objective function: Establish a multi-objective optimization function. .

[0144] in, , , These are weighting coefficients and can be dynamically adjusted according to the production plan. This represents the total value of the multi-objective optimization function that needs to be minimized; , , These are the weighting coefficients corresponding to the three sub-objectives (reboiler energy consumption, bottom phenol content, and top product purity deviation), used to balance the importance of different optimization objectives.

[0145] Optimization solution: While satisfying the aforementioned calculated deterministic constraints ( Under the premise of other operational constraints (such as valve opening range), the future reflux ratio (R) and reboiler heat load (Q) are used as decision variables, and the non-dominated sorting genetic algorithm (NSGA-III) is used to solve this multi-objective optimization problem. Strategy determination: The algorithm provides a set of Pareto optimal solutions. The controller selects an optimal operating point from these solutions based on the preset "energy efficiency priority" meta-strategy. For example, the calculation results show that the optimal control strategy is: "gradually adjust the reflux ratio from the current 3.5 to 3.2, and at the same time adjust the reboiler steam supply from 10 t / h to 9.5 t / h".

[0146] Fourth, implement control strategies.

[0147] The command is issued, and the control strategy (reflux ratio setpoint) generated above is implemented. Heat load setpoint The data is distributed from the computing server to the factory's DCS system via the OPC-UA protocol.

[0148] Smooth execution: To avoid impacting production, the control logic within the DCS system will ramp up the setpoints. For example, the reflux ratio setpoint will be smoothly transitioned from 3.5 to 3.2 at a rate of 0.05 / minute.

[0149] At the lower level, the PID controller within the DCS automatically adjusts the frequency of the reflux pump inverter and the opening of the steam valve supplying the reboiler based on the new set values, thus completing the final control of the physical equipment.

[0150] Through the complete process of this embodiment, the technical solution of this application achieves precise and robust closed-loop optimization control of the coal tar distillation column. In simulated operation, compared with the traditional control strategy, the yield of light phenolic oil increased by about 1.5%, the overall energy consumption decreased by about 3%, and the risk of overheating at the top of the column due to fluctuations in operating conditions was effectively avoided, demonstrating significant technical advantages and economic benefits.

[0151] To further verify the technical effectiveness of the embodiments of this application, under the same production load and raw material conditions, the robust optimization control method based on digital twins proposed in this application (the embodiment of this application) was compared with the traditional advanced process control (APC) combined with PID control strategy originally used in the distillation column (control group) for a week of simulated operation. The average values ​​of the obtained key performance indicators are shown in Table 1 below.

[0152] Control group description: The control group uses the exact same distillation column physical equipment as the embodiments of this application, but its control system is a relatively mature and advanced process control technology in the industry. This APC controller is based on a step response model for predictive control, but its model update cycle is long and does not consider uncertainties in the process. The safety boundary usually depends on a fixed margin set by the operator.

[0153] Table 1 below compares the control effects of the embodiments of this application with those of the control group. As can be seen from the comparison data in Table 1, the technical solution of the embodiments of this application brings significant beneficial effects:

[0154] Through more accurate model prediction and optimization, the yield of light phenolic oil was significantly increased by 1.54 percentage points, while the energy consumption per unit product was reduced by 3.53%, directly improving the economic benefits of the plant.

[0155] This application completely avoids the risk of overheating at the top of the tower by quantifying uncertainty and dynamically calculating robustness margin, thus ensuring the safety of the production process. Traditional methods, however, still carry the risk of slight overheating due to fluctuations in operating conditions. As can be seen from the variance of the control setpoint, the control strategy of this application adjusts more smoothly, avoiding large and frequent adjustments to the production equipment, which helps extend equipment life and maintain production stability. More stable operation leads to more consistent product quality, further improving the yield rate.

[0156] In summary, compared with the prior art, this application achieves safer, more efficient, and more stable control over industrial production processes by constructing a digital twin model and combining uncertainty propagation and robust optimization, demonstrating the inventiveness and practicality of the invention.

[0157] Table 1

[0158] Performance indicators Control group (based on traditional APC control) Examples of this application Improvement / Reduction Rate Light phenolic oil yield (%) 97.2 98.7 +1.54% Comprehensive energy consumption (GJ / ton of raw material) 0.85 0.82 -3.53% Risk of tower top temperature exceeding the limit (times / week) 1 (Slightly overheated) 0 Risk elimination Control strategy adjustment stability (setpoint variance) 0.28 0.11 Improvement of 60.7% Product quality pass rate (%) 99.1 99.8 +0.71%

[0159] Example 4

[0160] Figure 4 This is a schematic diagram of the structure of a control device for an industrial production process provided in Embodiment 3 of this application. Figure 4 As shown, the device 400 includes:

[0161] The data acquisition module 410 is used to acquire the physical entity data and production process data of the target distillation column;

[0162] The digital twin module 420 is used to construct a digital twin model of the target distillation column based on physical entity data and production process data;

[0163] The strategy generation module 430 is used to determine the control strategy of the target distillation column based on the digital twin model and production process data.

[0164] The strategy execution module 440 is used to control the production process of the target distillation column according to the control strategy.

[0165] In the technical solution of this application embodiment, a digital twin model of the target distillation column is constructed based on the physical entity data and production process data of the target distillation column. This model provides a digital twin simulation, modeling, and prediction environment for the control of the target distillation column, providing fundamental support for determining a better control strategy. Based on the digital twin model and the production process data, a control strategy for the target distillation column is determined to control the production process of the target distillation column. Through predictive practices and simulation tests in the digital twin model, a control strategy that conforms to the current actual situation of the target distillation column is determined. This reduces the need for repeated adjustments to the actual production equipment and lowers the frequency of repeated activation of constraints on various parameters of the production equipment, thereby saving more trial costs, reducing energy consumption, and improving production efficiency.

[0166] In one alternative implementation, the policy generation module may include:

[0167] The feasible region determination unit is used to determine the thermodynamically feasible manifold based on production process data;

[0168] The candidate state determination unit is used to estimate the state of each production process data based on a preset time-delay augmented state space to obtain each candidate state;

[0169] The feasibility processing unit is used to perform feasibility operations on each candidate state according to the thermodynamic feasible manifold to obtain each state to be processed.

[0170] The strategy generation unit is used to determine the control strategy based on the status to be processed and the production process data.

[0171] In one optional implementation, the strategy generation unit may include:

[0172] The sample prediction subunit is used to generate prediction samples based on each state to be processed and production process data;

[0173] The robustness margin determination subunit is used to determine the robustness margin based on the state to be processed, the predicted samples, and the preset coverage.

[0174] The control strategy generation subunit is used to generate control strategies based on each state to be processed, the predicted samples, and the robustness margin.

[0175] In one optional implementation, the control strategy generation subunit may include:

[0176] Constraints are determined from the element and used to determine the deterministic constraints corresponding to the robustness margin;

[0177] The strategy determination unit is used to determine the control strategy within a feasible manifold based on the state to be processed, the predicted samples, and the deterministic constraints.

[0178] In one optional implementation, the candidate state determination unit may be specifically used for:

[0179] Based on the preset time-delay augmented state space, the state trajectory of the production process data is solved within a preset sliding window to obtain each candidate state.

[0180] In one alternative implementation, the feasibility processing unit may be specifically used for:

[0181] Within the thermodynamically feasible manifold, construct differentiable projection operators for each candidate state in the differentiable thermodynamically feasible domain, and perform feasibility operations to obtain each state to be processed.

[0182] The industrial production process control device provided in this application can execute the industrial production process control method provided in any embodiment of this application, and has the corresponding functional modules and beneficial effects for executing the control method of each industrial production process.

[0183] Example 5

[0184] Figure 5A schematic diagram of an electronic device 10, which can be used to implement embodiments of this application, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the application described and / or claimed herein.

[0185] like Figure 5 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0186] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0187] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as control methods for industrial production processes.

[0188] In some embodiments, the industrial production process control method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the industrial production process control method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the industrial production process control method by any other suitable means (e.g., by means of firmware).

[0189] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0190] Computer programs used to implement the methods of this application may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0191] In the context of this application, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium can be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0192] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0193] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0194] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0195] This application also discloses a computer program product, which includes a computer program that, when executed by a processor, implements the industrial production process control method provided in any embodiment of this application. This program product shares the same inventive concept as the industrial production process control methods disclosed in the embodiments of this application, and therefore will not be described in detail here.

[0196] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this application can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this application can be achieved, and this is not limited herein.

[0197] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method of controlling an industrial production process, characterized in that, The method comprises: acquiring physical entity data and production process data of a target distillation column; constructing a digital twin model of the target distillation column according to the physical entity data and the production process data; determining a control strategy of the target distillation column according to the production process data based on the digital twin model; controlling a production process of the target distillation column according to the control strategy.

2. The method of claim 1, wherein, The determining of the control strategy of the target distillation column according to the production process data comprises: determining a thermodynamically feasible flow field according to the production process data; performing state estimation on each of the production process data according to a preset time-delay augmented state space to obtain each candidate state; performing a feasibility operation on each of the candidate states according to the thermodynamically feasible flow field to obtain each to-be-processed state; determining the control strategy according to the to-be-processed state and the production process data.

3. The method of claim 2, wherein, The determining of the control strategy according to the to-be-processed state and the production process data comprises: generating each prediction sample according to each of the to-be-processed state and the production process data; determining a robust margin according to the to-be-processed state, the prediction sample and a preset coverage degree; generating the control strategy according to each of the to-be-processed state, the prediction sample and the robust margin.

4. The method of claim 3, wherein, The generating of the control strategy according to each of the to-be-processed state, the prediction sample and the robust margin comprises: determining a deterministic constraint corresponding to the robust margin; determining the control strategy in the feasible flow field based on the to-be-processed state, the prediction sample and the deterministic constraint.

5. The method of claim 2, wherein, The state estimation on each of the production process data according to a preset time-delay augmented state space to obtain each candidate state comprises: solving a state trajectory of the production process data in a preset sliding window according to the preset time-delay augmented state space to obtain each candidate state.

6. The method of claim 2, wherein, The feasibility operation on each of the candidate states according to the thermodynamically feasible flow field to obtain each to-be-processed state comprises: constructing a differentiable projection operator of each of the candidate states in a differentiable thermodynamically feasible region in the thermodynamically feasible flow field to perform the feasibility operation to obtain each of the to-be-processed state.

7. A control device for an industrial production process, characterized in that The method comprises: acquiring physical entity data and production process data of a target distillation column; constructing a digital twin model of the target distillation column according to the physical entity data and the production process data; determining a control strategy of the target distillation column according to the production process data based on the digital twin model; controlling a production process of the target distillation column according to the control strategy.

8. An electronic device, comprising: The electronic device comprises: at least one processor; and a memory connected with the at least one processor in communication; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the industrial production process control method in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions for causing the processor to implement the control method of the industrial production process according to any one of claims 1-6 when executed.

10. A computer program product, characterised in that, The computer program product comprises a computer program which, when executed by the processor, implements the control method of the industrial production process according to any one of claims 1-6.

Images