Adaptive deep learning prediction control method and device suitable for process industry

Abstract

The invention provides a self-adaptive deep learning prediction control method and device suitable for the process industry. Comprising the following steps: constructing an all-condition dynamic model of a target control variable based on a deep learning technology according to historical control information of the process industry, and performing segmentation processing on a change gain curve of the all-condition dynamic model based on a self-adaptive segmentation algorithm of a threshold value to obtain a segmented linear gain model; a piecewise linear variable constraint of the piecewise linear gain model is determined, and an optimization target of a target control variable is determined based on the piecewise linear variable constraint; and obtaining a step response sequence of the target control variable at the current working condition point based on the piecewise linear gain model, and predicting a future control execution sequence of the target control variable according to the step response sequence and the optimization target. The problem that an existing predictive control method generally constructs a mapping relation between a control variable and a controlled variable based on a linear model, the nonlinear characteristic of the control variable cannot be accurately reflected, and then the predictive control accuracy is low is solved.

Description

Technical Field

[0001] The present application relates to the technical field of adaptive deep learning predictive control for process industries, and more specifically, to an adaptive deep learning predictive control method, device, computer-readable storage medium, and electronic device suitable for process industries. Background Art

[0002] Model Predictive Control (MPC) is a highly effective control method widely used in industrial processes, particularly in the petrochemical and chemical industries, where its effectiveness has been fully demonstrated. By constructing a predictive model of the controlled plant and combining real-time state feedback with rolling optimization in the future time domain, it can effectively coordinate the dynamic balance between multiple variables such as temperature, pressure, and flow, while ensuring safety constraints. This significantly improves the operational efficiency and stability of the plant. However, like many advanced technologies, MPC is not perfect and has exposed some limitations in practical applications. Production equipment in process control industries (such as chemical, petroleum, refining, and pharmaceuticals) often exhibits complex physical, chemical, and thermodynamic processes. For example, the Arrhenius equation for reaction rates in chemical reactors exhibits an exponentially temperature-dependent nonlinear relationship; the separation efficiency of distillation columns exhibits a strongly coupled nonlinear relationship with feed composition and reflux ratio. Furthermore, varying production indicators, such as production load and market demand, may necessitate the use of multiple operating modes for on-site equipment. This makes it difficult for predictive control that relies on models to achieve precise optimization control. Mechanism-based model building is complex and lacks experimental data. In actual construction processes, one or more local linear models are often used for approximation. Therefore, some improved predictive control methods aimed at handling working conditions are called multi-model MPC and have been widely proposed. Their goal is to establish multiple models for each operating mode and design model switching strategies. However, in actual industrial processes, due to the complex environment and the difficulty of accurately distinguishing each mode with offline data, this further reduces the control performance and robustness; and the adaptive predictive control methods in existing technologies usually construct the mapping relationship between the operating variables (independent variables) and the controlled variables (dependent variables) based on linear models, which will cause the model to be unable to accurately reflect the nonlinear characteristics of the process, thereby leading to the problem of low predictive control accuracy. Summary of the Invention

[0003] The main purpose of this application is to provide an adaptive deep learning predictive control method, device, computer-readable storage medium and electronic device suitable for process industry, so as to at least solve the problem that existing predictive control methods usually construct the mapping relationship between operating variables and controlled variables based on linear models, which cannot accurately reflect the nonlinear characteristics of the control variables, thereby leading to low predictive control accuracy.

[0004] To achieve the above-mentioned purpose, according to one aspect of the present application, an adaptive deep learning predictive control method suitable for process industry is provided, comprising: constructing a full-operating-condition dynamic model of a target control variable based on historical control information of the process industry based on deep learning technology, wherein the target control variable consists of an independent variable and a dependent variable, and the independent variable and the dependent variable are in a nonlinear coupling relationship; using a threshold-based adaptive segmentation algorithm, segmenting the change gain curve of the full-operating-condition dynamic model to obtain a piecewise linear gain model; determining the piecewise linear variable constraints of the piecewise linear gain model, and determining the optimization target of the target control variable based on the piecewise linear variable constraints, wherein the optimization target consists of an optimized value of the independent variable and an optimized value of the dependent variable; obtaining a step response sequence of the target control variable at a current operating point based on the piecewise linear gain model, and predicting a future control execution sequence of the target control variable based on the step response sequence and the optimization target, wherein the future control execution sequence is used to control the independent variable of the target control variable.

[0005] According to another aspect of the present application, an adaptive deep learning predictive control device suitable for process industry is provided, comprising: a first construction unit, configured to construct a full-operating-condition dynamic model of a target control variable based on deep learning technology according to historical control information of the process industry, wherein the target control variable consists of an independent variable and a dependent variable, and the independent variable and the dependent variable are in a nonlinear coupling relationship; a segmentation processing unit, configured to segmentally process the change gain curve of the full-operating-condition dynamic model based on a threshold-based adaptive segmentation algorithm to obtain a piecewise linear gain model; a determination unit, configured to determine the piecewise linear variable constraints of the piecewise linear gain model, and determine an optimization target of the target control variable based on the piecewise linear variable constraints, wherein the optimization target consists of an optimized value of the independent variable and an optimized value of the dependent variable; a prediction unit, configured to obtain a step response sequence of the target control variable at a current operating point based on the piecewise linear gain model, and predict a future control execution sequence of the target control variable according to the step response sequence and the optimization target, wherein the future control execution sequence is used to control the independent variable of the target control variable.

[0006] According to another aspect of the present application, a computer-readable storage medium is provided, which includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the adaptive deep learning predictive control methods applicable to the process industry.

[0007] According to a further aspect of the present application, an electronic device is provided, comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs comprise a program for performing any one of the adaptive deep learning predictive control methods suitable for process industry.

[0008] By applying the technical solution of the present application, the target control variable full-condition dynamic model constructed by deep learning technology can accurately capture the complex nonlinear coupling relationship between independent variables and dependent variables, effectively covering the dynamic characteristics under various conditions such as raw material fluctuation, load adjustment, and equipment aging in process industry. Combined with the adaptive segmentation algorithm based on threshold, the nonlinear gain curve is converted into a segmented linear model, which not only simplifies the calculation complexity in model predictive control, but also ensures the applicability and accuracy of the model under different conditions. By determining the segmented linear variable constraint, the independent variables and dependent variables can be optimized specifically, achieving dual improvement of economic benefit and operation safety. In particular, the predictive control framework is constructed based on the real-time step response sequence, which can dynamically adjust the future control execution sequence to ensure that the independent variables of the target control variable always move towards the optimization target, even when facing sudden condition changes. The problem of low prediction control accuracy caused by the fact that existing predictive control methods usually construct the mapping relationship between the operation variable and the controlled variable based on linear models, which cannot accurately reflect the nonlinear characteristics of the control variable, is solved. BRIEF DESCRIPTION OF DRAWINGS

[0009] The accompanying drawings, which form a part of the present description, are included to provide a further understanding of the application and are incorporated in and constitute a part of this application. The embodiments of the present application, and their

[0010] Figure 1 A hardware structure block diagram of a mobile terminal for performing an adaptive deep learning predictive control method suitable for process industry is shown according to an embodiment of the present application;

[0011] Figure 2 A flowchart of an adaptive deep learning predictive control method suitable for process industry is shown according to an embodiment of the present application;

[0012] Figure 3 A flowchart of an adaptive deep learning predictive control method suitable for process industry is shown according to an embodiment of the present application;

[0013] Figure 4 A step response curve of a deep learning model under different conditions is shown according to an embodiment of the present application;

[0014] Figure 5 An algorithm steady-state optimization flow chart provided by an embodiment of the present application is shown.

[0015] Figure 6 A dynamic control algorithm flow chart provided by an embodiment of the present application is shown.

[0016] Figure 7 A structural block diagram of an adaptive deep learning predictive control device for a process industry provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0017] It should be noted that the embodiments and features of the embodiments in the present application can be combined with each other without conflict. The technical solutions in the embodiments of the present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0018] In order 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 described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should fall within the scope of protection of the present application.

[0019] It should be noted that the terms "first", "second" and the like in the specification and claims of the present application and the above-mentioned accompanying drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but can include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0020] For the convenience of description, some nouns or terms related to the embodiments of the present application are described below:

[0021] MPC: Model predictive control, is a special control. Its current control action is obtained by solving a finite time domain open-loop optimal control problem at each sampling instant.

[0022] Deep learning: Deep learning specifically refers to machine learning based on deep neural network models and methods. It develops based on statistical machine learning, artificial neural networks, and other algorithmic models, combined with the development of big data and massive computing power. The most important technical feature of deep learning is its ability to automatically extract features.

[0023] Step response sequence: This refers to the zero-state response of a system after receiving a unit step function input. A zero-state response means that the system is in its initial state before receiving the specified input. This ensures that the system's response is entirely due to the specified input (in this case, a unit step input).

[0024] As introduced in the background technology, existing predictive control methods usually construct the mapping relationship between control variables and controlled variables based on linear models, which cannot accurately reflect the nonlinear characteristics of the control variables, thereby resulting in low predictive control accuracy. In order to solve the problem that existing predictive control methods usually construct the mapping relationship between control variables and controlled variables based on linear models, which cannot accurately reflect the nonlinear characteristics of the control variables, thereby resulting in low predictive control accuracy, the embodiments of the present application provide an adaptive deep learning predictive control method, device, computer-readable storage medium and electronic device suitable for process industry.

[0025] The technical solutions in the embodiments of the present invention will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present invention.

[0026] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 This is a hardware structure block diagram of a mobile terminal for an adaptive deep learning predictive control method for process industry according to an embodiment of the present invention. Figure 1 As shown, the mobile terminal may include one or more ( Figure 1 Only one is shown) a processor 102 (the processor 102 may include but is not limited to a microprocessor MCU or a programmable logic device FPGA and other processing devices) and a memory 104 for storing data, wherein the mobile terminal may also include a transmission device 106 and an input and output device 108 for communication functions. It will be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the mobile terminal. Figure 1 More or fewer components than shown, or with Figure 1 Different configurations shown.

[0027] The memory 104 can be used to store computer programs, such as software programs and modules of application software, such as the computer program corresponding to the adaptive deep learning predictive control method for process industries in the embodiment of the present invention. The processor 102 executes the computer program stored in the memory 104 to perform various functional applications and data processing, thereby implementing the above-mentioned method. The memory 104 may include high-speed random access memory and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely located relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof. The transmission device 106 is used to receive or transmit data via a network. Specific examples of such networks may include a wireless network provided by the mobile terminal's telecommunications provider. In one example, the transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0028] In this embodiment, an adaptive deep learning predictive control method suitable for process industry is provided, which runs on a mobile terminal, a computer terminal or a similar computing device. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0029] Figure 2 Flowchart of the adaptive deep learning predictive control method applicable to process industry according to the embodiment of the present application. Figure 2 As shown, the method includes the following steps:

[0030] Step S201: constructing a full-operation-condition dynamic model of a target control variable based on historical control information of the process industry using deep learning technology, wherein the target control variable consists of an independent variable and a dependent variable, and a nonlinear coupling relationship exists between the independent variable and the dependent variable;

[0031] Process industries, including chemical, petroleum, and electric power, often involve a large number of complex control variables in their production processes. These variables are coupled with nonlinear relationships, making it difficult for traditional linear models to accurately describe these complex relationships. Deep learning technologies, including but not limited to neural networks, can build more accurate dynamic models of all operating conditions by learning from historical control information, such as the relationship between independent variables like temperature, pressure, and flow, and dependent variables like product quality and energy consumption.

[0032] Step S202 , using a threshold-based adaptive segmentation algorithm, segmenting the change gain curve of the full-operation dynamic model to obtain a piecewise linear gain model;

[0033] Specifically, a threshold-based adaptive segmented algorithm is used to segmentally linearize the gain curve of the full-operating-condition dynamic model, significantly improving the model's applicability and the performance of the control system. This technology's effectiveness lies in its ability to decompose complex nonlinear gain curves into multiple simple linear segments, each of which accurately describes system behavior within a specific operating range. This not only simplifies the model, reducing computational complexity and response time, but also improves the model's accuracy and adaptability under different operating conditions. This enables the control system to respond more flexibly and efficiently to dynamic changes in industrial sites, reduces resource waste, and improves the optimization level of the overall control strategy, thereby enhancing system stability and production efficiency.

[0034] Step S203, determining the piecewise linear variable constraints of the piecewise linear gain model, and determining the optimization target of the target control variable based on the piecewise linear variable constraints, wherein the optimization target consists of the optimized value of the independent variable and the optimized value of the dependent variable;

[0035] Specifically, by precisely setting the constraints of the piecewise linear variables, such as the range of each linear interval and the slope limit, we can ensure that the model maintains reasonable linearity and avoids overfitting under different operating conditions. The goal of this optimization is to maximize control efficiency and minimize control error while meeting system stability and performance requirements. Technically, this approach allows the control system to more intelligently adjust its operating parameters to respond to rapidly changing industrial environments. For example, in chemical production, faced with fluctuations in key parameters such as temperature and pressure, the piecewise linear gain model can instantly identify the trend of change and automatically adjust the heater power or valve opening according to the optimization goal, not only maintaining the smooth operation of the process, but also optimizing energy consumption, achieving energy conservation, emission reduction and cost control. In addition, this automatic adjustment based on the optimization goal can also effectively reduce human intervention, avoid operational errors, and further improve the safety and automation level of industrial production.

[0036] Step S204: obtaining a step response sequence of the target control variable at the current operating point based on the piecewise linear gain model, and predicting a future control execution sequence of the target control variable according to the step response sequence and the optimization objective, wherein the future control execution sequence is used to control the independent variable of the target control variable.

[0037] Through this embodiment, the full-condition dynamic model of the target control variable constructed using deep learning technology, applying the aforementioned steps S201, S202, S203, and S204, can accurately capture the complex nonlinear coupling relationship between the independent and dependent variables, effectively covering the dynamic characteristics of process industries under variable operating conditions such as raw material fluctuations, load adjustments, and equipment aging. Combined with a threshold-based adaptive segmentation algorithm, the nonlinear gain curve is converted into a piecewise linear model, which not only simplifies the computational complexity of model predictive control but also ensures the applicability and accuracy of the model under different operating conditions. By determining piecewise linear variable constraints, the independent and dependent variables can be optimized in a targeted manner, achieving both economic benefits and operational safety. In particular, the predictive control framework constructed based on real-time step response sequences can dynamically adjust future control execution sequences, ensuring that the independent variable of the target control variable always moves toward the optimization target, allowing for rapid response even in the face of sudden operating condition changes. This solves the problem that existing predictive control methods, which typically construct mapping relationships between manipulated variables and controlled variables based on linear models, cannot accurately reflect the nonlinear characteristics of the control variables, resulting in low predictive control accuracy.

[0038] In the specific implementation process, the adaptive segmentation algorithm based on the threshold is used to segment the change gain curve of the above-mentioned full-operating-condition dynamic model to obtain a piecewise linear gain model, including: obtaining N segmented intervals by uniformly sampling the operating points of the above-mentioned change gain curve, determining the gain value of each of the above-mentioned segmented intervals, and determining the gain mean of the above-mentioned change gain curve through the above-mentioned gain value of each of the above-mentioned segmented intervals, wherein N>1; calculating the approximate deviation of each of the above-mentioned segmented intervals based on the above-mentioned gain mean and the above-mentioned gain value of each segmented interval based on the mean square error; and separating the above-mentioned segmented intervals whose approximate deviation is less than the deviation threshold from the above-mentioned segmented intervals. The segment interval is marked as a valid segment and stored in the segment set; the cutting step: the target segment interval is divided into a left subinterval and a right subinterval, and the target segment interval is cut by a genetic optimization algorithm with the total approximate deviation of the left subinterval and the right subinterval as the goal to obtain two sub-segment intervals, wherein the target segment interval represents a segment interval or sub-segment interval whose approximate deviation is greater than or equal to the deviation threshold; the cutting step is repeated until the approximate deviations of all the sub-segment intervals are less than the deviation threshold, and the piecewise linear gain model is obtained.

[0039] This method uses a threshold-based adaptive segmentation algorithm that can effectively segment the gain curve, resulting in a more realistic piecewise linear gain model, further improving the optimization effect of the control strategy. In process industries, the gain curve of the control variable is often not linear, but rather changes with the operating point. By uniformly sampling the operating points of the gain curve to obtain N segmented intervals, the complex nonlinear gain curve can be decomposed into multiple linear segments, thereby simplifying the calculation of the control strategy. A genetic optimization algorithm is used to cut the target segmented interval to achieve the goal of minimizing the total approximate deviation, which not only improves the accuracy of the model but also enhances the robustness of the model. Through the coordination of various technical features, namely the combination of the adaptive segmentation algorithm and the genetic optimization algorithm, the technical problem of the difficulty in accurately modeling the nonlinear changes in the gain curve of the control variable in the process industry is solved, and the optimization effect and robustness of the control strategy are improved.

[0040] More specifically, determining the piecewise linear variable constraints of the piecewise linear gain model includes:

[0041] According to the constraint formula:

[0042]

[0043]

[0044]

[0045]

[0046]

[0047]

[0048]

[0049] Determine the piecewise linear variable constraints for the piecewise linear gain model above, where is the optimized action amount of the above independent variables, 、 are the minimum and maximum changes of the above independent variables, 、 are the upper and lower limits of the above independent variable actions, 、 is the minimum and maximum single change of the above dependent variable, M is a constant, i represents the impact on the i-th dependent variable, represents the gain of the k-th operating condition on the i-th dependent variable, is the optimization cost of the ith dependent variable, Optimize increments for independent variables, is the action amount of the above independent variable, is the single change of the dependent variable mentioned above, 、 is the upper and lower operating limits of the dependent variable, To optimize the initial point, It is the segment flag bit, is the gain relationship between the independent variable and the dependent variable, is the kth segmentation point, Optimize variables for piecewise optimization.

[0050] This method determines the piecewise linear variable constraints of the piecewise linear gain model, a key step in ensuring the feasibility of control strategies in practical applications. In process industries, control variable constraints often include, but are not limited to, the minimum and maximum changes of the independent variable, as well as the minimum and maximum single changes of the dependent variable. Setting these constraints prevents the control strategy from exceeding the control range of the actual equipment during execution, thereby ensuring the feasibility of the control strategy. Through the interaction of various technical features, namely the combination of the piecewise linear gain model and the piecewise linear variable constraints, the technical problem of control strategies in process industries potentially exceeding the control range of the equipment in practical applications is resolved, ensuring the feasibility and stability of the control strategy.

[0051] Furthermore, predicting a future control execution sequence of the target control variable according to the step response sequence and the optimization objective includes:

[0052] According to the formula:

[0053]

[0054] Determine the above future control execution sequence, where, is the residual between the above dependent variable and the reference trajectory, is the dependent variable weight, is the algorithm’s future prediction of the dependent variable, is the current time, is the future moment, j represents the jth dependent variable mentioned above, Ncv is the number of dependent variables, P represents the number of steps to be predicted, r j represents the expected reference trajectory of the jth dependent variable above, is the action amount for step i The penalty is M, M is the control time domain, Nmv is the number of the above independent variables, is the action penalty weight of the independent variable, For the above future control execution sequence, For the future action sequence that needs to be optimized, A moment of prediction for the future, is the open-loop prediction sequence, is the step response sequence between the i-th independent variable and the j-th dependent variable at the k-th operating point, is the minimum increment of the independent variable action, is the maximum increment of the independent variable action, For The future predicted value of time, is the amount of action applied at the next time point l, is the lower limit of the independent variable, is the upper limit of the independent variable, is the lower limit of the dependent variable operation, is the upper operating limit of the dependent variable.

[0055] This method can achieve optimal control of control variables in process industries by predicting the future control execution sequence of the target control variable. The step response sequence reflects the dynamic response characteristics of the impact of changes in the independent variable on the dependent variable, while the optimization objective includes the expected control values ​​of the dependent and independent variables. By predicting the future control execution sequence, the control actions of the independent variable can be adjusted in advance to achieve the optimization goal, such as reducing energy consumption or improving product quality. In process industries, the optimization of control strategies often requires considering the combined effects of multiple dependent variables and the penalty costs of the independent variable control actions. Therefore, the calculation of the predicted control execution sequence requires comprehensive consideration of these factors. By interplaying various technical features—namely, the prediction of the step response sequence, the optimization objective, and the future control execution sequence—this method solves the technical problem of the difficulty of adjusting control strategies in advance to achieve optimization goals in process industries, thereby improving the optimization effect and efficiency of control strategies.

[0056] Furthermore, after predicting the future control execution sequence of the target control variable based on the step response sequence and the optimization objective, the method further includes: obtaining real-time control information of the target control variable, and updating the real-time control information to the full-operating-condition dynamic model.

[0057] This method updates the full-condition dynamic model in real time, enabling the control strategy to quickly adapt to the dynamic changes of the process industry and enhancing the robustness and stability of the control system. In practical applications, the production process of process industries often experiences dynamic changes, such as equipment aging and raw material changes, which can affect the relationship between control variables. By acquiring real-time control information of the control variables and updating it to the full-condition dynamic model, these changes can be reflected in a timely manner, allowing the control strategy to be adjusted to ensure control effectiveness. Through the combination of various technical features, namely the acquisition of real-time control information and the updating of the full-condition dynamic model, the technical problem of the control strategy in the process industry having difficulty in quickly adapting to the dynamic changes of the production process is solved, thereby enhancing the robustness and stability of the control system.

[0058] Specifically, in the process of constructing a full-operating-condition dynamic model of the target control variable based on deep learning technology according to the historical control information of the process industry, the above method also includes: constructing a full-operating-condition linear model of other control variables based on the identification algorithm according to the above historical control information of the above process industry, wherein the independent variables and dependent variables of the above other control variables are in a linear coupling relationship, and the above identification algorithm includes an FIR algorithm and a subspace algorithm.

[0059] This method constructs a full-condition linear model of other control variables, which can further improve the control strategy and enhance the control effect. In the process industry, in addition to the target control variable, there are other control variables, and the relationship between these variables and the dependent variable may be a linear coupling relationship. By constructing a full-condition linear model of these control variables based on the identification algorithm, the control process of the process industry can be more comprehensively described, thereby improving the optimization effect of the control strategy. The FIR algorithm and the subspace algorithm can construct a linear model through historical control information, such as input and output data between independent variables and dependent variables. Through the interaction of various technical features, that is, the construction of the identification algorithm and the full-condition linear model, the technical problem of the difficulty in accurately modeling the linear coupling relationship between other control variables and dependent variables in the process industry is solved, the control strategy is improved, and the control effect is improved.

[0060] More specifically, after constructing the full-operating-condition linear model of other control variables based on the identification algorithm, the above method also includes: integrating the above-mentioned full-operating-condition linear model and the above-mentioned full-operating-condition dynamic model as sub-models into the controller model with the control variables as units to form a hybrid controller model, wherein the above-mentioned control variables include the above-mentioned target control variables and other control variables.

[0061] This method integrates the full-condition linear model and the full-condition dynamic model to form a hybrid controller model, which can achieve optimal control of complex control processes in process industries. In practical applications, the control processes in process industries often contain both linear coupling relationships and nonlinear coupling relationships. By integrating the full-condition linear model and the full-condition dynamic model into the controller model, linear and nonlinear relationships can be considered simultaneously, thereby improving the optimization effect of the control strategy. In addition, the construction of the hybrid controller model can also improve the versatility and adaptability of the control strategy, enabling the control strategy to be applied to a wider range of control scenarios. Through the interaction of various technical features, namely the integration of the full-condition linear model, the full-condition dynamic model and the hybrid controller model, the technical problem of the control process in the process industry containing both linear coupling relationships and nonlinear coupling relationships is solved, the optimal control of complex control processes is achieved, and the versatility and adaptability of the control strategy are improved.

[0062] This embodiment also includes the introduction of a meta-learning mechanism to enhance model generalization capabilities. During the training phase of the deep learning prediction model, in addition to traditional supervised learning methods, a meta-learning mechanism is introduced. Meta-learning is a learning method that enables a model to quickly adapt to new tasks with less data. Specifically, the Model-Agnostic Meta-Learning (MAML) framework can be employed. By adjusting the initialization method of the model parameters, the model can be updated through rapid gradient descent when receiving a small amount of new operating condition data, achieving good prediction results.

[0063] During the pre-training phase, we use a large amount of known operating condition data to find a "meta-initialization point" using the MAML method. This initialization point allows the model to be efficiently fine-tuned for unknown operating conditions using only a small amount of data. In actual deployment, when encountering new operating conditions, we extract a portion of the real-time data as a "fast adaptation" dataset for meta-learning. After a few fine-tuning iterations, the model can respond quickly and accurately to the new conditions.

[0064] Consider applying this method to the distillation column control system of a chemical plant. The separation efficiency of a distillation column is not only affected by temperature and pressure but also closely related to the feed composition, which frequently changes during plant operations, creating uncertainty in operating conditions. Using a meta-learning mechanism, our deep learning model can rapidly adjust the prediction model to accommodate sudden changes in feed composition using only a few days of data, maintaining the control system's prediction accuracy within ±3%. This significantly reduces the time and resource consumption required to retrain the model, which can take weeks or even months.

[0065] This embodiment reduces the need for large amounts of data collection and processing to adapt to new working conditions, improves the flexibility of industrial sites in responding to emergencies, and effectively avoids delays and errors in control decisions.

[0066] This embodiment also includes a priority-based dynamic update strategy for control variables. During the dynamic predictive control phase, we consider that real-time updates of certain control variables (such as the opening degree of a safety valve) are crucial to overall operational efficiency and safety. We have developed a priority-based dynamic update strategy that prioritizes control variables based on their importance and response speed. For high-priority variables, a shorter prediction window and a higher update frequency are used; for low-priority variables, a longer prediction window and a lower update frequency are used.

[0067] This approach improves the response speed of key control variables while reducing the use of computing resources by dynamically adjusting the parameter configuration of predictive control. This strategy is particularly important when dealing with large-scale industrial processes.

[0068] Specific implementation plan: In a refinery's catalytic cracking unit, temperature control is central to ensuring chemical reaction efficiency and product quality, while pressure regulation is relatively secondary. By applying a priority-based dynamic update strategy for control variables, the prediction window for temperature control was set to 10 time steps, while the prediction window for pressure control was expanded to 20 time steps. This strategy reduced the temperature control response time by 50%, maintaining process parameters within target ranges even with rapid changes in feedstock composition and market conditions, significantly improving product consistency and unit safety.

[0069] The rapid response of key control variables in this embodiment effectively prevents potential production failures and safety incidents. By rationally allocating computing resources, the computational burden of non-critical variables is reduced, the execution efficiency of the overall control algorithm is improved, and unnecessary over-regulation is reduced, thereby reducing energy consumption and maintenance costs.

[0070] In order to enable those skilled in the art to more clearly understand the technical solution of the present application, the implementation process of the adaptive deep learning predictive control method applicable to the process industry of the present application will be described in detail below with reference to specific embodiments.

[0071] This paper aims to address the model mismatch problem faced by traditional MPC in complex process industries due to its multiple working conditions and time-varying characteristics, especially the defects of multi-model MPC such as fuzzy mode division and rigid switching strategy caused by offline data limitations.

[0072] This embodiment relates to a specific adaptive deep learning predictive control method applicable to process industries. Its core goal is to break through the limitations of fixed models or local linear models and achieve full coverage of dynamic characteristics of complex industrial processes and real-time adaptive control through deep fusion of data-driven modeling and model predictive control. Figure 3 Specifically, it includes the following contents:

[0073] Step 1: Controller modeling: hybrid modeling strategy;

[0074] Construct a hybrid model architecture that includes a linear model and a deep learning model. Use a deep learning prediction model to characterize the nonlinearity, time-varying, and full-condition dynamic characteristics of multiple variables. The specific steps are as follows:

[0075] The dynamic process of general process industry can be expressed by the following formula: ;

[0076] Where F is a linear or nonlinear function, k represents discrete time, N represents the model length of the dynamic process, and Y0 represents the reference measurement value of the controlled variable. U and Y represent the manipulated variable (independent variable) and the controlled variable (dependent variable), respectively.

[0077] First, based on the control scheme and prior knowledge, if it is known that the dynamic response of a CV variable (controlled variable, which is the dependent variable) under the full operating range can be represented by a single linear model, the linear model can be obtained through traditional identification algorithms such as FIR and subspace methods. The linear model structure is as follows: , A(t) is the step response sequence.

[0078] If the dynamic response of the CV variable in the full range of working conditions cannot be represented by a single linear model, a full-condition dynamic model can be established using a deep learning model, and the full-condition model can be obtained by inputting discrete data samples. Specifically, the data collected offline can be set with an appropriate sampling period T s , historical window length T past and prediction window length T fut , the associated input variables u(t) and output variables y(t) are organized in the form of multiple-input single-output (MISO) to form a time series data set and input it into the system step response identification system based on deep learning. The unit step response of each operating point output by its unit step response prediction submodule can be obtained at the same time, such as Figure 4 shown.

[0079] ,in, For deep learning models, represents the operating point at time t, Represents the controlled variable corresponding to the operating point at time t.

[0080] Finally, the traditional linear model and deep learning model are integrated into the controller model as sub-models with the controlled quantity as the unit to form a hybrid controller model : , where i represents the i-th CV variable, is a traditional linear model, For deep learning models.

[0081] Step 2: Global gain piecewise linearization:

[0082] In process industries, predictive control often employs a two-tiered structure: an upper tier for steady-state target calculation or optimization, and a lower tier for dynamic control. This two-tiered structure offers greater assurance of system stability and safety compared to traditional predictive control.

[0083] The control part of this embodiment will continue to use the dual-layer control architecture of steady-state optimization and dynamic control, while the traditional dual-layer predictive control algorithm only supports linear model solution. In order to support the efficient optimization solution of the hybrid model in the steady-state layer, it is necessary to design an adaptive piecewise linear method for the dynamic change gain curve of all working conditions represented by the deep learning model, and convert the steady-state gain relationship of the dynamic change of all working conditions into a resolvable piecewise linear model while retaining the steady-state gain change characteristics of the key working points. The specific process is as follows:

[0084] Based on the deep learning model, a full-condition step response sequence is established, and the steady-state gain is calculated for each operating point. : ;

[0085] Where Us and Ys represent the actual operating conditions. Based on the steady-state gain distribution across all operating conditions, assuming the process is linear within a threshold range or confidence interval, the global gain curve of the deep learning model under all operating conditions can be represented as a linear segmented model. Based on this idea, a threshold-based adaptive segmentation algorithm is designed.

[0086] First, define the entire operating range as , the initial segment set is .

[0087] Then, M operating points are uniformly sampled in the interval. The mean gain in the interval is obtained. : ; Use mean squared error (MSE) to calculate the approximate deviation: ;

[0088] If the MSE is less than the set threshold, the current interval is marked as a valid segment and stored in the segment set; if the deviation exceeds the threshold, a candidate segmentation point l is selected within the interval to divide the interval into a left subinterval and a right subinterval; the optimal cutting point that minimizes the total MSE of the left and right ends after a single cut can be quickly searched using a genetic optimization algorithm. :

[0089] ,in, is the mean square error of the left subinterval, is the mean square error of the right subinterval;

[0090] Until the deviation of all intervals is within the threshold or reaches the preset maximum number of segments. The piecewise linear gain model can be obtained by linear division: ;

[0091] represents the interval indicator function (when When 1 is selected, is the gain of the kth segment.

[0092] Step three, steady-state optimization of mixed model, as Figure 5 shown, specifically includes the following:

[0093] In the two-layer predictive control architecture of process industry, the steady-state optimization layer mainly uses the process steady-state model to solve the steady-state optimal solution that meets the process constraints and control objectives, to provide targets for subsequent dynamic control, and to meet the user's partial benefit demands. The embodiment uses the step two method to approximate the steady-state gain curve of the deep model with a piecewise linear model, and participates in the steady-state optimization calculation.

[0094] Firstly, since the piecewise linear model is introduced, binary variables are defined in the optimization proposition to represent whether the current working condition is in the kth segment. At the same time, the following independent constraints need to be met:

[0095] , representing the kth segment of the current working condition point;

[0096] For each segment k, linearize the influence of the input on the controlled variable: ; Where i represents the influence on the ith controlled variable, represents the gain of the kth segment working condition on the ith controlled variable, and then further improve the interval constraint: ;

[0097] M is a large enough constant, and this constraint, combined with the uniqueness constraint, indicates that only the unique interval constraint will be truly effective. Further, integrate the piecewise linear variable constraint and the linear variable constraint:

[0098] , indicating that the economic target is the action of the operating variable as little as possible.

[0099] , indicating the increment of the operating variable;

[0100] , indicating the position constraint of the operating variable;

[0101] , indicating the increment of the controlled variable;

[0102] , indicating the position constraint of the controlled variable;

[0103] , the uniqueness constraint of the piecewise linear value interval;

[0104] ;

[0105] Where, is the optimal action of the operating variable; and The minimum and maximum changes in the operation amount. and The upper and lower limits of the operation amount; and It is the maximum single change of the controlled variable, which is generally not set; represents the influence of nonlinear model on the controlled quantity, and Represents the influence of the linear model on the controlled variable. By solving the above steady-state optimization problem, the actual optimization target (U, Y) can be obtained.

[0106] Step 4: Dynamic predictive control, such as Figure 6 As shown, specifically including the following:

[0107] After obtaining the global steady-state optimization target, it is necessary to further formulate a dynamic control sequence for the operating variables and the controlled quantity to reach the optimal operating point. The predictive control framework used in this invention is dynamic matrix control, which requires predictive control based on the step response sequence. This algorithm can well meet the requirements of industrial sites for computational efficiency and control accuracy. In order to realize the predictive control calculation of the hybrid model, it is necessary to construct a dynamic matrix calculation by locally linearizing the deep model. The specific contents include the following:

[0108] Get the step response sequence from the current steady-state point to the optimization target. The linear model is a fixed step response sequence, which can be directly obtained by the identification algorithm of step one. The step response process of the deep prediction model, due to the existence of multiple working conditions, nonlinearity and other characteristics, not only depends on the current steady-state position, but also on the subsequent moving direction and steady-state target. Therefore, after obtaining the optimization target, it is necessary to obtain the step response sequence in real time based on the deep prediction model. The specific method is as follows: First, based on the real historical data, construct the future control quantity U of the standard step fut for:

[0109] .

[0110] in, The amount of action required for the current calculation scenario. Inputting the constructed future control amount into the deep prediction model and performing feedforward prediction can obtain a set of prediction quantities. After further processing, the current dynamic sequence can be inferred: ;

[0111] Where A (k) Represents the controlled variable step response sequence of the current operating point, M deep is the deep learning model trained in step 1. Then, based on the step response sequence obtained in real time, the predictive control dynamic programming proposition can be constructed:

[0112] ;

[0113] Among them, the optimization objective consists of two items: is the residual between the controlled quantity and the reference trajectory, where j represents the jth controlled quantity, Ncv is the number of controlled quantities, P represents the number of steps to be predicted, and r j represents the expected reference trajectory of the j-th controlled quantity; is the action amount for step l The penalty is M, the control time domain, and Nmv is the number of manipulated variables. In the constraints, It is an open-loop prediction sequence that can be obtained through real-time prediction of the deep prediction model. (k) ij is the step response sequence between the manipulated variable i and the controlled variable j at the k operating point, is the future input sequence increment that needs to be optimized, and the following three constraints are the upper and lower limits of the manipulated variable, the rate of change, and the upper and lower limits of the controlled variable. By solving the above optimization proposition, a set of future control variables can be obtained , which is the control execution sequence of the control variables for the next P cycles. The algorithm issues the control variable sequence for the first step of execution.

[0114] Step 5: Online Incremental Update: Although deep learning models have achieved dynamic characteristic coverage of all operating conditions through large-scale offline data, in real industrial scenarios, sudden changes in raw material properties (such as coal chemical raw material ash content exceeding a threshold), equipment state drift (such as catalyst activity decay), or extreme operating conditions (such as sudden load fluctuations) may still exceed the training data distribution, causing model deviation. To this end, an online incremental model update mechanism based on real-time data is needed to enhance the algorithm's adaptability. Specifically, a sliding window mechanism is used to store and clean real-time data. Once the data volume reaches the training requirement, the real-time data is input into the deep time series prediction model. It is important to note that to balance new and old knowledge, an elastic weight regularization term is introduced to avoid sudden changes in model parameters. Model accuracy is fine-tuned using online real-time data.

[0115] , where L new is the new loss function, Indicates the parameters that need to be updated. is the regularization term weight. Fi() is a schematic representation of the regularization term expression. Steps 2 through 5 are then repeated. During the control process, data is collected in real time, the deep prediction model is updated, and then, based on the deep prediction model, the model information relied upon during the optimization control process is corrected, thereby achieving stable control under all operating conditions in the process industry.

[0116] The upper layer of the hybrid control model of this embodiment uses a deep time series model to capture the complex dynamic characteristics of the process industry, including multivariable nonlinear coupling, time-varying operating condition migration, and long-period lag effects; the lower layer constructs a predictive control model through a local linearization model based on real-time inference to generate an optimal control action sequence. In addition, this embodiment proposes a model hybrid strategy that can integrate a deep time series prediction model and a linear model. The FIR / subspace model method is used for variables with stable dynamic response characteristics under all operating conditions to obtain an FIR model with a constant step response, while a deep prediction model is used for variables whose dynamic characteristics significantly change with the operating point to establish a nonlinear mapping of the input-output relationship. This embodiment also proposes a predictive control algorithm adapted to the deep time series prediction model. The steady-state optimization layer considers the gain under all operating conditions to generate the optimal operating target under all operating conditions; the dynamic layer infers the local linear sequence in real time to solve the optimal control increment that meets the constraints in the rolling time domain.

[0117] This embodiment specifically achieves the following technical effects:

[0118] 1. A two-layer prediction-control coupling architecture based on deep learning is proposed: the upper layer uses a deep time series prediction model to accurately characterize the nonlinearity, time-varying characteristics, and dynamic characteristics of all working conditions among multiple variables; the lower layer extracts the step response model of the current working point through real-time reasoning and calculates the optimal control action. Compared with the traditional single linear model, this embodiment effectively solves the model mismatch problem caused by changes in the dynamic characteristics of the system after the working condition switches in the industrial process, and significantly improves the control robustness in multiple working condition scenarios. At the same time, compared with directly using deep learning models for control calculations, the output is more interpretable, security is guaranteed, and computational efficiency is greatly improved.

[0119] 2. Fusion of linear and deep learning models: Variable relationships with stable dynamic response characteristics across all operating conditions are modeled as single step responses, while variable relationships with significantly changing dynamic characteristics are modeled using deep neural networks. This hybrid strategy inherits the computational efficiency of traditional linear models while overcoming their limitations in dynamically changing scenarios through deep learning, balancing model accuracy and real-time performance. Furthermore, this fusion strategy is adaptable to a wider range of industrial processes, providing a foundation for large-scale deployment.

[0120] 3. Full-operating-condition nonlinear optimization improves control accuracy and adaptability: Steady-state optimization: Based on the global nonlinear gain curve of the deep model, a nonlinear programming solver is constructed. Compared with traditional linearized steady-state optimization methods, this method can more accurately capture the coupling relationship between variables and optimize the output results more accurately. Dynamic control: By obtaining the step response sequence of the current operating point in real time and dynamically correcting the prediction model parameters, the controller can adapt to complex scenarios such as operating condition drift and time-varying parameters, significantly improving the adaptability and anti-interference performance of multi-step predictive control.

[0121] The embodiment of the present application also provides an adaptive deep learning predictive control device suitable for process industries. It should be noted that the adaptive deep learning predictive control device suitable for process industries in the embodiment of the present application can be used to execute the adaptive deep learning predictive control method suitable for process industries provided in the embodiment of the present application. The device is used to implement the above-mentioned embodiments and preferred implementation modes, and those that have been explained will not be repeated here. As used below, the term "module" can implement a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceived.

[0122] The following introduces the adaptive deep learning predictive control device suitable for process industry provided in the embodiment of the present application.

[0123] Figure 7 Schematic diagram of an adaptive deep learning predictive control device for process industry according to an embodiment of the present application. Figure 7 As shown, the device includes:

[0124] A first construction unit 71 is configured to construct a full-operation-condition dynamic model of a target control variable based on historical control information of the process industry and deep learning technology, wherein the target control variable comprises an independent variable and a dependent variable, and a nonlinear coupling relationship exists between the independent variable and the dependent variable;

[0125] A segmentation processing unit 72 is used to segmentally process the change gain curve of the full-operation dynamic model using a threshold-based adaptive segmentation algorithm to obtain a segmented linear gain model;

[0126] a determination unit 73, configured to determine the piecewise linear variable constraints of the piecewise linear gain model, and determine the optimization target of the target control variable based on the piecewise linear variable constraints, wherein the optimization target is composed of an optimized value of an independent variable and an optimized value of a dependent variable;

[0127] The prediction unit 74 is used to obtain the step response sequence of the above-mentioned target control variable at the current operating point based on the above-mentioned piecewise linear gain model, and predict the future control execution sequence of the above-mentioned target control variable according to the above-mentioned step response sequence and the above-mentioned optimization objective, wherein the above-mentioned future control execution sequence is used to control the above-mentioned independent variable of the above-mentioned target control variable.

[0128] In this embodiment, a first construction unit is used to construct a full-operation dynamic model of the target control variable based on historical control information of the process industry using deep learning technology, wherein the target control variable consists of an independent variable and a dependent variable, and the independent variable and the dependent variable have a nonlinear coupling relationship. A segmentation processing unit is used to segment the change gain curve of the full-operation dynamic model based on a threshold-based adaptive segmentation algorithm to obtain a piecewise linear gain model. A determination unit is used to determine the piecewise linear variable constraints of the piecewise linear gain model and, based on the piecewise linear variable constraints, determine the optimization target of the target control variable, wherein the optimization target consists of the optimized value of the independent variable and the optimized value of the dependent variable. A prediction unit is used to obtain the step response sequence of the target control variable at the current operating point based on the piecewise linear gain model, and predict the future control execution sequence of the target control variable based on the step response sequence and the optimization target, wherein the future control execution sequence is used to control the independent variable of the target control variable. The full-operation dynamic model of the target control variable constructed using deep learning technology can accurately capture the complex nonlinear coupling relationship between the independent variable and the dependent variable, effectively covering the dynamic characteristics of the process industry under variable operating conditions such as raw material fluctuations, load adjustments, and equipment aging. Combined with a threshold-based adaptive segmented algorithm, the nonlinear gain curve is converted into a piecewise linear model, which not only simplifies the computational complexity in model predictive control, but also ensures the applicability and accuracy of the model under different operating conditions. By determining the piecewise linear variable constraints, the independent and dependent variables can be optimized in a targeted manner, achieving a dual improvement in economic benefits and operational safety. In particular, the predictive control framework is constructed based on a real-time step response sequence, which can dynamically adjust the future control execution sequence to ensure that the independent variable of the target control variable always moves in the direction of the optimization target, and can respond quickly even in the face of sudden changes in operating conditions. It solves the problem that existing predictive control methods usually construct the mapping relationship between the operating variable and the controlled variable based on a linear model, which cannot accurately reflect the nonlinear characteristics of the control variable, thereby resulting in low predictive control accuracy.

[0129] As an optional solution, the segmentation processing unit includes a first determination module, a calculation module, a marking module, a cutting module and a repeated execution module; the first determination module is used to obtain N segmentation intervals by uniformly sampling the operating points of the above-mentioned change gain curve, determine the gain value of each of the above-mentioned segmentation intervals, and determine the gain mean of the above-mentioned change gain curve through the above-mentioned gain value of each of the above-mentioned segmentation intervals, wherein N>1; the calculation module is used to calculate the approximate deviation of each of the above-mentioned segmentation intervals based on the mean square error according to the above-mentioned gain mean and the above-mentioned gain value of each segmentation interval; the marking module is used to mark the above-mentioned segmentation intervals whose approximate deviation is less than the deviation threshold as Marked as a valid segment and stored in the segment set; the cutting module is used to perform the cutting step: dividing the target segment interval into a left sub-interval and a right sub-interval, and cutting the target segment interval with the goal of minimizing the total approximate deviation of the left sub-interval and the right sub-interval through a genetic optimization algorithm to obtain two sub-segment intervals, wherein the target segment interval represents a segment interval or sub-segment interval whose approximate deviation is greater than or equal to the deviation threshold; the repeated execution module is used to repeatedly execute the above cutting step until the approximate deviations of all the above sub-segment intervals are less than the above deviation threshold, thereby obtaining the above piecewise linear gain model.

[0130] In an optional solution, the determining unit includes a second determining module, configured to determine the following according to the constraint formula:

[0131]

[0132]

[0133]

[0134]

[0135]

[0136]

[0137]

[0138] Determine the piecewise linear variable constraints for the piecewise linear gain model above, where is the optimized action amount of the above independent variables, 、 are the minimum and maximum changes of the above independent variables, 、 are the upper and lower limits of the above independent variable actions, 、 is the minimum and maximum single change of the above dependent variable, M is a constant, i represents the impact on the i-th dependent variable, represents the gain of the kth segment of the working condition on the ith dependent variable, represents the optimization cost of the ith dependent variable, represents the optimization increment of the independent variable, represents the action amount of the independent variable, represents the single change amount of the dependent variable, 、 represents the operation upper and lower limit range of the dependent variable, represents the optimization initial point, represents the segmentation flag bit, represents the gain relationship between the independent variable and the dependent variable, represents the kth segmentation point, represents the segmented optimization variable.

[0139] An optional scheme, the prediction unit comprises a third determination module, used for determining the future control execution sequence according to the formula:

[0140]

[0141] represents the residual error of the dependent variable and the reference trajectory, represents the dependent variable weight, represents the future predicted value of the algorithm to the dependent variable, represents the current time, represents the future time, j represents the jth dependent variable, Ncv represents the number of dependent variables, P represents the number of steps to be predicted, r j represents the expected reference trajectory of the jth dependent variable, represents the punishment of the ith step action amount , M represents the control time domain, Nmv represents the number of independent variables, represents the action punishment weight of the independent variable, represents the future control execution sequence, represents the future action sequence to be optimized and solved, represents the future predicted time, represents the open-loop prediction sequence, represents the step response sequence between the ith independent variable and the jth dependent variable of the kth working condition point, represents the minimum increment of the independent variable action, represents the maximum increment of the independent variable action, represents the future predicted value of time, represents the action amount applied to the future l time, represents the lower limit of the independent variable, represents the upper limit of the independent variable, represents the lower limit of the dependent variable operation, represents the upper limit of the dependent variable operation.​

[0142] An optional solution, the device further comprises an acquisition unit, configured to acquire real-time control information of the target control variable after predicting a future control execution sequence of the target control variable according to the above-mentioned step response sequence and the above-mentioned optimization target, and update the real-time control information into the full-condition dynamic model.

[0143] An optional solution, the device further comprises a second construction unit, configured to construct a full-condition linear model of other control variables based on a recognition algorithm according to historical control information of the process industry in the process of constructing a full-condition dynamic model of the target control variable based on deep learning technology, wherein the independent variables and dependent variables of the other control variables are in a linear coupling relationship, and the recognition algorithm comprises FIR algorithm and subspace algorithm.

[0144] An optional solution, the device further comprises an integration unit, configured to integrate the full-condition linear model and the full-condition dynamic model into a controller model as a sub-model in units of control variables to form a hybrid controller model after constructing the full-condition linear model of the other control variables based on the recognition algorithm, wherein the control variables include the target control variable and the other control variables.

[0145] The adaptive deep learning predictive control device for process industry comprises a processor and a memory, and the first construction unit, the segmentation processing unit, the determination unit and the prediction unit are stored in the memory as program units, and the corresponding functions are realized by the processor executing the program units stored in the memory. The modules are located in the same processor; alternatively, the modules are located in different processors in any combination.

[0146] The processor comprises a core, and the core retrieves the corresponding program units from the memory. The core can be set to one or more, and the problem that the existing predictive control method generally constructs the mapping relationship between the control variables and the controlled variables based on the linear model, which cannot accurately reflect the nonlinear characteristics of the control variables, and further leads to low prediction accuracy can be solved by adjusting the core parameters.

[0147] The memory can include non-permanent memory in a computer readable medium, random access memory (RAM) and / or non-volatile memory such as read-only memory (ROM) or flash memory (flash RAM), and the memory comprises at least one memory chip.

[0148] An embodiment of the present invention provides a computer-readable storage medium, which includes a stored program. When the program is running, the device where the computer-readable storage medium is located is controlled to execute the adaptive deep learning predictive control method applicable to the process industry.

[0149] An embodiment of the present invention provides a processor, which is used to run a program, wherein the program executes the above-mentioned adaptive deep learning predictive control method applicable to the process industry when running.

[0150] An embodiment of the present invention provides an electronic device comprising a processor, memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements at least the steps of the aforementioned adaptive deep learning predictive control method for process industries. The device herein may be a server, a PC, a PAD, a mobile phone, or the like.

[0151] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program that initializes at least the steps of the above-mentioned adaptive deep learning predictive control method applicable to the process industry.

[0152] Obviously, those skilled in the art will appreciate that the various modules or steps of the present invention described above can be implemented using a general-purpose computing device, can be centralized on a single computing device, or can be distributed across a network of multiple computing devices. They can be implemented using program code executable by the computing device, and thus, can be stored in a storage device and executed by the computing device. In some cases, the steps shown or described herein can be performed in a different order than that shown, or can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0153] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0154] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0155] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0156] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0157] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0158] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0159] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can be implemented using any method or technology for information storage. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change RAM (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media, such as modulated data signals and carrier waves.

[0160] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0161] The above description is merely a preferred embodiment of the present application and is not intended to limit the present application. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

Claims

1. An adaptive deep learning predictive control method suitable for process industry, characterized in that: include: Based on historical control information from the process industry, a full-operation-condition dynamic model of the target control variable is constructed using deep learning technology, wherein the target control variable consists of an independent variable and a dependent variable, and the independent variable and the dependent variable have a nonlinear coupling relationship; Based on a threshold-based adaptive segmentation algorithm, the change gain curve of the full-operation dynamic model is segmented to obtain a segmented linear gain model; Determining piecewise linear variable constraints of the piecewise linear gain model, and determining an optimization target of the target control variable based on the piecewise linear variable constraints, wherein the optimization target consists of an optimized value of an independent variable and an optimized value of a dependent variable; A step response sequence of the target control variable at the current operating point is obtained based on the piecewise linear gain model, and a future control execution sequence of the target control variable is predicted according to the step response sequence and the optimization objective, wherein the future control execution sequence is used to control the independent variable of the target control variable.

2. The method according to claim 1, characterized in that Based on the threshold-based adaptive segmentation algorithm, the change gain curve of the full-operation dynamic model is segmented to obtain a piecewise linear gain model, including: uniformly sampling the operating points of the change gain curve to obtain N segmented intervals, determining a gain value of each segmented interval, and determining a gain mean of the change gain curve according to the gain value of each segmented interval, wherein N>1; Calculating the approximate deviation of each segmented interval according to the gain mean and the gain value of each segmented interval based on the mean square error; Mark the segment intervals where the approximate deviation is less than the deviation threshold as valid segments and store them in the segment set; Cutting step: dividing the target segmented interval into a left subinterval and a right subinterval, and cutting the target segmented interval by a genetic optimization algorithm with the goal of minimizing the total approximate deviation of the left subinterval and the right subinterval to obtain two subsegmented intervals, wherein the target segmented interval represents a segmented interval or subsegmented interval whose approximate deviation is greater than or equal to the deviation threshold; The cutting step is repeatedly performed until the approximate deviations of all the sub-segment intervals are less than the deviation threshold, thereby obtaining the piecewise linear gain model.

3. The method according to claim 1, characterized in that Determining piecewise linear variable constraints of the piecewise linear gain model includes: According to the constraint formula: ； ； ； Determining piecewise linear variable constraints for the piecewise linear gain model, wherein is the optimized action amount of the independent variable, the optimized action amount is between the minimum change amount and the maximum change amount, and the sum of the optimized action amount and the action amount of the independent variable is between the upper and lower limits of the independent variable action, 、 is the minimum and maximum single change of the dependent variable, M is a constant, i represents the impact on the i-th dependent variable, represents the gain of the k-th operating condition on the i-th dependent variable, is the optimization cost of the ith dependent variable, Optimize increments for independent variables, is the action amount of the independent variable, is the single change of the dependent variable, 、 is the upper and lower operating limits of the dependent variable, To optimize the initial point, It is the segment flag.

4. The method according to claim 1, wherein Predicting a future control execution sequence of the target control variable according to the step response sequence and the optimization target includes: According to the formula: ； Determine the future control execution sequence, wherein, is the residual between the dependent variable and the reference trajectory, is the dependent variable weight, is the algorithm’s future prediction of the dependent variable, is the current time, is the future moment, j represents the jth dependent variable, Ncv is the number of dependent variables, P represents the number of steps to be predicted, r j represents the expected reference trajectory of the jth dependent variable, is the action amount for step i The penalty, M is the control time domain, Nmv is the number of independent variables, is the action penalty weight of the independent variable, for said future control execution sequence, For the future action sequence that needs to be optimized, A moment to predict the future.

5. The method according to claim 1, wherein After predicting a future control execution sequence of the target controlled variable according to the step response sequence and the optimization target, the method further includes: Real-time control information of the target control variable is obtained, and the real-time control information is updated into the full-operating-condition dynamic model.

6. The method according to claim 1, characterized in that In the process of constructing a full-operation-condition dynamic model of target control variables based on historical control information of the process industry using deep learning technology, the method further includes: According to the historical control information of the process industry, a full-condition linear model of other control variables is constructed based on an identification algorithm, wherein the independent variables and dependent variables of the other control variables are in a linear coupling relationship, and the identification algorithm includes an FIR algorithm and a subspace algorithm.

7. The method according to claim 6, characterized in that After constructing the full-operating-condition linear models of other control variables based on the identification algorithm, the method further includes: The full-operating-condition linear model and the full-operating-condition dynamic model are integrated into a controller model as sub-models in units of control variables to form a hybrid controller model, wherein the control variables include the target control variables and other control variables.

8. An adaptive deep learning predictive control device suitable for process industry, characterized in that: include: A first construction unit is configured to construct a full-operation-condition dynamic model of a target control variable based on historical control information of the process industry and deep learning technology, wherein the target control variable is composed of an independent variable and a dependent variable, and a nonlinear coupling relationship exists between the independent variable and the dependent variable; A segmentation processing unit is used for segmentally processing the change gain curve of the full-operation-condition dynamic model based on a threshold-based adaptive segmentation algorithm to obtain a segmented linear gain model; a determining unit, configured to determine a piecewise linear variable constraint of the piecewise linear gain model, and determine an optimization target of the target control variable based on the piecewise linear variable constraint, wherein the optimization target consists of an optimized value of an independent variable and an optimized value of a dependent variable; a prediction unit, configured to obtain a step response sequence of the target control variable at a current operating point based on the piecewise linear gain model, and predict a future control execution sequence of the target control variable according to the step response sequence and the optimization objective, wherein the future control execution sequence is used to control the independent variable of the target control variable.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored program, wherein, when the program is running, the device where the computer-readable storage medium is located is controlled to execute the adaptive deep learning predictive control method applicable to the process industry as described in any one of claims 1 to 7.

10. An electronic device, characterized in that: include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include a method for executing the adaptive deep learning predictive control method suitable for process industry as described in any one of claims 1 to 7.

Images