A Human-Machine Interactive Maintenance System and Equipment for a Model Predictive Controller

Abstract

The present invention relates to a human-computer interactive maintenance system and device for a model predictive controller. The maintenance system includes: a matching judgment module for judging whether the currently running prediction model matches the process system; a warehousing module for saving the historical prediction model to a model library and saving the historical controller parameters to a parameter library; a model fitting module for fitting the historical prediction models in the model library to obtain candidate prediction models; a model identification module for performing model identification based on the identification data range input by an operator to generate a test model; a test module for obtaining test parameters adapted to the test model, performing open-loop simulation calculations to obtain an open-loop output, and based on the selected operation of the operator, using the corresponding test model and test parameters as the prediction model and controller parameters that match the process system. The system of the present invention can form a deep interaction with the operator to obtain a better model predictive controller.

Description

Technical Field

[0001] The present invention relates to the technical field of predictive control in the process industry, and particularly to a human-computer interactive maintenance system and device for a model predictive controller. Background Art

[0002] Model predictive control is a control method based on a predictive model. According to the current state and parameters of the system, the optimal control means for a period of time in the future is calculated. That is, its current control action is obtained by solving a finite-time open-loop optimal control problem at each sampling instant. Compared with traditional PID control, model predictive control is not only applicable to difficult-to-control controlled objects such as multi-coupling and large time delay, but also can model constraints such as process conditions and economic indicators, and cooperate with optimization techniques to complete the control task. Therefore, it has broad application prospects in the process industry. However, in actual production, with various fluctuating factors such as changes in production objectives, changes in the ratio of production raw materials, equipment aging, and insufficient catalyst activity, the load and working conditions of the process system will continuously change. On the one hand, the change in the dynamic characteristics of the process system easily leads to the mismatch of the predictive model; on the other hand, considering the stability and control effect of the model predictive controller, when the working conditions of the process system change, the parameter scheme of the model predictive controller usually needs to be adjusted accordingly. Therefore, when the load and working conditions of the process system change, it is necessary to timely maintain the model predictive controller and adjust its predictive model and controller parameters.

[0003] In response to the above maintenance requirements of the model predictive controller, the prior art usually tends to use adaptive algorithms and deep learning-based algorithms to select or adjust the predictive model or controller parameters, while minimizing the participation of operators to make the model predictive controller more intelligent. However, for the process system, due to the prominent coupling phenomenon between its sections, between equipment, and between controller parameters, the adjustment or perturbation of a certain controller parameter often causes changes in other controller parameters and the product quality of subsequent sections, and it is difficult to accurately describe and judge directly through algorithms or formulas. Therefore, the maintenance method based on algorithms generally has a worse effect in actual applications than the theoretical effect. Industrial field operators usually have rich process knowledge and operation experience, can evaluate the state and potential trends of the entire system, and make up for the deficiencies of the algorithm, but lack experience in model identification and parameter debugging. However, the existing interaction between the model predictive controller and the operator still stays at the level of direct input and display of various operation variables and controlled variables of the process system, and cannot enable the operator to deeply participate in the maintenance process of the model predictive controller, and the process knowledge and operation experience of the operator cannot be well utilized and exerted. Summary of the Invention

[0004] (I) Technical Problems to be Solved

[0005] In view of the above-mentioned disadvantages and deficiencies of the prior art, the present invention provides a human-computer interactive maintenance system and device for a model predictive controller, which solves the technical problem that the existing model predictive controller has insufficient interaction with operators and cannot utilize the process knowledge and operation experience of operators.

[0006] (II) Technical solution

[0007] In order to achieve the above object, the main technical solutions adopted by the present invention include:

[0008] In a first aspect, an embodiment of the present invention provides a human-computer interactive maintenance system for a model predictive controller, which is used to maintain the model predictive controller of a process system. The maintenance system includes:

[0009] A matching and judging module, configured to obtain real-time process data of the process system, obtain controller parameters of the model predictive controller, and judge whether the currently running prediction model matches the process system based on a matching index preset by the operator; if so, generate a warehousing signal; if not, generate an alarm signal;

[0010] A warehousing module, configured to, when receiving the warehousing signal, save the currently running prediction model as a historical prediction model to a model library, and save the controller parameters corresponding to the current prediction model as historical controller parameters to a parameter library;

[0011] A model fitting module, configured to, when receiving a model fitting request from the operator, fit the historical prediction models in the model library based on the fitting data range input by the operator, obtain at least one historical prediction model whose fitting degree meets the requirements within the fitting data range as candidate prediction models; and, based on the operator's selection operation on a certain candidate prediction model, output the candidate prediction model as a prediction model matching the process system, or as a test model;

[0012] A model identification module, configured to, when receiving a model identification request from the operator, perform model identification based on the identification data range input by the operator to generate a test model;

[0013] A test module, configured to obtain test parameters adapted to the test model based on the test model; and, based on the test model and the test parameters, perform open-loop simulation calculations according to the test data range input by the operator and the test parameter modification operation to obtain an open-loop output, and based on the operator's selection operation on the open-loop output, use the test model and test parameters corresponding to the open-loop output as a prediction model and controller parameters matching the process system.

[0014] The maintenance system proposed in the embodiment of the present invention stores historical prediction models and historical controller parameters that match the process system through the warehousing module for use by the model fitting module, model identification module, and testing module. The model fitting module, model identification module, and testing module of the maintenance system interact with the operator, perform model fitting, model identification, or testing based on the warehoused prediction models and controller parameters according to the data range input by the operator, and output the results, and perform subsequent operations according to the decisions made by the operator based on the above results. Therefore, while reasonably utilising the operator's process knowledge and operation experience, the maintenance process of the model predictive controller is modularised. While ensuring the openness of the maintenance system, the operation difficulty of the operator in debugging the model predictive controller is reduced, enabling a deep interaction between the maintenance system and the operator. As a result, the maintenance system can perform calculations, simulations, and adjustments according to the data range input by the operator and output corresponding models, controller parameters, or intermediate results, while the operator can judge and select the quality of the corresponding models, controller parameters, or intermediate results based on their own process knowledge and operation experience to obtain a better model predictive controller, thereby achieving the purpose of optimising the control effect of the model predictive controller on the process system.

[0015] Optionally, in the matching judgment module, the controller parameters include: the set value of the controlled variable; the real-time process data includes: the measured value of the controlled variable, the measured value of the manipulated variable;

[0016] The matching indicators include: window length h, standard deviation threshold s_thrd, delay alarm duration t1, delay warehousing duration t2; and h < t1, h < t2, and both t1 and t2 are integer multiples of h;

[0017] The window length h represents the time length of the measured value of the controlled variable that is used each time to judge whether the currently running prediction model matches the process system;

[0018] The standard deviation threshold s_thrd represents the range index that allows the measured value of the controlled variable to fluctuate;

[0019] The delay alarm duration t1 represents the duration during which the fluctuation of the measured value of the controlled variable is allowed to exceed the standard deviation threshold s_thrd without an alarm;

[0020] The delay warehousing duration t2 represents the duration during which the measured value of the controlled variable is allowed to fluctuate within the standard deviation threshold s_thrd without being warehoused.

[0021] Optionally, in the matching judgment module, the step of judging whether the currently running prediction model matches according to the real-time process data based on the matching indicators preset by the operator includes:

[0022] Based on the real-time process data of the process system, obtain the measured values of the controlled variables at each window length h.

[0023] According to formula (1), calculate the deviation e(i,t) between the measured value pv(i,t) and the set value sp(i,t) of the i-th controlled variable at the t-th moment, where i is a positive integer.

[0024] Based on e(i,t), calculate the standard deviation s(i,t) of the i-th controlled variable within the window length h at the t-th moment according to formula (2).

[0025] Judge whether the values of s(i,t), s(i,t - h), s(i,t - 2h), ……, s(i,t - t1) are all not within the standard deviation threshold s_thrd within the continuous t1 duration pushed forward from the t-th moment; if so, it is determined that the current prediction model does not match the process system.

[0026] Judge whether the values of s(i,t), s(i,t - h), s(i,t - 2h), ……, s(i,t - t2) are all within the standard deviation threshold s_thrd within the continuous t2 duration pushed forward from the t-th moment; if so, it is determined that the current prediction model matches the process system.

[0027] The formula (1) is:

[0028] e(i,t) = pv(i,t) - sp(i,t) (1)

[0029] In formula (1), pv(i,t) represents the measured value of the i-th controlled variable at the t-th moment, and sp(i,t) represents the set value of the i-th controlled variable at the t-th moment.

[0030] The formula (2) is:

[0031]

[0032] In formula (2), e(i,t - j) represents the deviation between the measured value and the set value of the i-th controlled variable at the (t - j)-th moment.

[0033] Optionally, the prediction model includes model parameters.

[0034] When the model is a second-order model, the model parameters include: gain, time constant one, time constant two, time delay, integral flag bit.

[0035] Optionally, in the model fitting module, the fitting data range includes: the first numerical range of the model parameters of the target prediction model, the first historical process data for fitting.

[0036] Fitting the historical prediction models in the model library to obtain at least one historical prediction model with a fitting degree meeting the requirements within the fitting data range, including:

[0037] Preselecting the historical prediction models in the model library based on the first numerical range, and taking the historical prediction models with model parameters falling within the first numerical range as the initial prediction models;

[0038] Based on the initial prediction models and the corresponding controller parameters of the initial prediction models, constructing a fitting model prediction controller, and driving the fitting model prediction controller to operate based on the first historical process data to obtain the predicted value pv1 of the controlled variable of the fitting model prediction controller on the first historical process data pre (i,t),

[0039] Based on the predicted value pv1 of the controlled variable pre (i,t), calculating the fitting degree index fit(i) of each initial prediction model according to formula (3), and outputting the top N initial prediction models with the highest fitting degree index fit(i) as the candidate prediction models; where N is a positive integer;

[0040] The formula (3) is:

[0041]

[0042] In formula (3), fit(i) represents the fitting degree evaluation index of the i-th controlled variable;

[0043] pv(i,t) represents the measured value of the i-th controlled variable at the t-th moment in the first historical process data;

[0044] pv1 pre (i,t) represents the predicted value of the i-th controlled variable at the t-th moment obtained by driving the fitting model prediction controller based on the first historical process data;

[0045] pv mean (i) represents the average value of the measured values of the i-th controlled variable in the first historical process data;

[0046] max(a,b) represents the operation of taking the maximum value of a and b;

[0047] min(a,b) represents the operation of taking the minimum value of a and b.

[0048] Optionally, in the model identification module, the identification data range includes: the second numerical range of the model parameters of the target prediction model, and the second historical process data for model identification;

[0049] Performing model identification based on the identified data range input by the operator to generate a test model, including:

[0050] Based on the grid search method, search within the second numerical range according to the evaluation function J, and output the model parameters with the optimal evaluation as the test model;

[0051] The evaluation function J is represented by formula (4), and the formula (4) is:

[0052]

[0053] In formula (4), pv(i,t) represents the measured value of the i-th controlled variable at the t-th moment based on the historical process data within the second time range;

[0054] T represents the length of the time range corresponding to the second historical process data;

[0055] pv2 pre (i,t) represents the predicted value of the i-th controlled variable at the t-th moment, which is obtained through formula (5);

[0056] The formula (5) is:

[0057]

[0058] In formula (5), u(j,t-l) represents the measured value of the j-th manipulated variable at the t-l moment in the second historical process data;

[0059] g(i,j,l) represents the impulse response coefficient between the j-th manipulated variable and the i-th controlled variable, and l, i, and j are all positive integers;

[0060] L ij is the model length corresponding to the current model parameters.

[0061] Optionally, in the test module, the test data range includes: the third historical process data for testing;

[0062] The test module includes:

[0063] A parameter matching unit for calculating the distance by traversing the historical prediction models in the model library based on the test model, obtaining the historical prediction model closest to the test model, and calling the historical controller parameters corresponding to the historical prediction model from the parameter library as the test parameters adapted to the test model;

[0064] A virtual test unit for calculating, based on the test model and test parameters and the third process data, an open-loop output of the test model based on the test parameters in a virtual environment; and, when receiving a test parameter modification operation from an operator, updating the test parameters according to the modification operation, and recalculating in the virtual environment the open-loop output of the test model based on the new test parameters, until receiving a selection operation from the operator for the current test model and test parameters, outputting the test model and the controller test parameters as a prediction model and controller parameters matching the process system.

[0065] Optionally, in the parameter matching unit, the traversing the historical prediction models in the model library for distance calculation includes:

[0066] Receiving the model gain weight w1, time constant weight w2, and time delay weight w3 input by the operator, traversing the historical prediction models in the model library, and calculating the distance D between the historical prediction model and the test model according to formula (6), where formula (6) is:

[0067]

[0068] In formula (6), K(i,j), T1(i,j), T2(i,j), T d (i,j) respectively represent the gain, time constant one, time constant two, and time delay between the jth manipulated variable and the ith controlled variable in the test model; K c (i,j), T 1c (i,j), T 2c (i,j), T dc (i,j) respectively represent the gain, time constant one, time constant two, and time delay between the jth manipulated variable and the ith controlled variable in the historical prediction model.

[0069] Optionally, in the virtual test unit, the calculating, based on the third process data, the open-loop output of the test model based on the test parameters in the virtual environment includes:

[0070] Constructing a test model predictive controller in the simulation environment according to the test parameters and the test model, and driving the test model predictive controller to run based on the third process data to obtain the manipulated variables continuously output by the test model predictive controller in the open-loop state, forming an open-loop manipulated variable sequence, and taking the open-loop manipulated variable sequence as the open-loop output of the test model based on the test parameters.

[0071] In a second aspect, an embodiment of the present invention provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the computer program is executed by the processor, it implements the architecture of the human-computer interactive maintenance system of the model predictive controller described in the first aspect.

[0072] (III) Beneficial Effects

[0073] The maintenance system proposed in the embodiment of the present invention stores historical prediction models and historical controller parameters matching the process system through the warehousing module for use by the model fitting module, the model identification module, and the testing module. The model fitting module, the model identification module, and the testing module of the maintenance system interact with the operator. Based on the warehoused prediction models and controller parameters, they perform model fitting, model identification, or testing according to the data range input by the operator and output the results, and execute subsequent operations according to the decisions made by the operator based on the above results. Therefore, while reasonably utilizing the operator's process knowledge and operation experience, the maintenance process of the model predictive controller is modularized. While ensuring the openness of the maintenance system, the operation difficulty of the operator in debugging the model predictive controller is reduced, enabling a deep interaction between the maintenance system and the operator. As a result, the maintenance system can perform calculations, simulations, and adjustments according to the data range input by the operator and output corresponding models, controller parameters, or intermediate results, while the operator can judge and select the advantages and disadvantages of the corresponding models, controller parameters, or intermediate results based on their own process knowledge and operation experience to obtain a better model predictive controller, thereby achieving the purpose of optimizing the control effect of the model predictive controller on the process system. Description of the Drawings

[0074] Figure 1 It is a schematic diagram of the architecture of a human-computer interactive maintenance system of a model predictive controller provided in the embodiment;

[0075] Figure 2 It is a schematic diagram of the process of the matching judgment module in a maintenance system provided in the embodiment. Detailed Embodiments

[0076] To better explain the present invention for easy understanding, the present invention will be described in detail below with reference to the drawings through specific embodiments.

[0077] The maintenance system provided in the embodiment of the present invention aims to combine the respective advantages of the maintenance system and the operator to establish a platform for deep interaction with the operator. Based on the relatively strong computing power of the maintenance system itself, it can effectively utilize the process knowledge and operation experience of industrial field operators and obtain the support of the operator at key decision-making nodes of the maintenance system, thereby obtaining a prediction model and controller parameters with a better control effect on the process system.

[0078] To better understand the above technical solution, the exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present invention can be understood more clearly and thoroughly, and the scope of the present invention can be fully conveyed to those skilled in the art.

[0079] Embodiment 1

[0080] As Figure 1 shown in the schematic architecture diagram, this embodiment provides a human-computer interactive maintenance system for a model predictive controller, which is used to maintain the model predictive controller of a process system. The maintenance system can be implemented on a computer device or other electronic devices with computing capabilities. The maintenance system includes: a matching judgment module, a warehousing module, a model fitting module, a model identification module, and a testing module. Specifically as follows:

[0081] The matching judgment module is used to obtain the real-time process data of the process system, obtain the controller parameters of the model predictive controller, and judge whether the currently running prediction model matches the process system based on the matching metrics preset by the operator; if so, it indicates that the control effect of the current model predictive controller is good, generates a warehousing signal, and calls the warehousing module to save the current prediction model and controller parameters for future use or reference; if not, generates an alarm signal to remind the operator that there is a mismatch between the current prediction model and the process system. The alarm signal can be displayed through the HMI (Human Machine Interface) screen or other display terminals.

[0082] Specifically, the controller parameters include: the set value of the controlled variable; the real-time process data includes: the measured value of the controlled variable, the measured value of the manipulated variable. The matching metrics are used to judge whether the currently running prediction model matches the process system. In an implementation manner of this embodiment, the matching metrics include: window length h, standard deviation threshold s_thrd, delay alarm duration t1, delay warehousing duration t2; and h < t1, h < t2, and both t1 and t2 are integer multiples of h.

[0083] The window length h represents the time length of the measured values of the controlled variables that are each time used to determine whether the currently running prediction model matches the process system. The standard deviation threshold s_thrd represents the range index that allows the measured values of the controlled variables to fluctuate. The delayed alarm duration t1 represents the duration during which the measured values of the controlled variables are allowed to fluctuate beyond the standard deviation threshold s_thrd without triggering an alarm. The delayed storage duration t2 represents the duration during which the measured values of the controlled variables are allowed to fluctuate within the standard deviation threshold s_thrd without being stored. The specific numerical ranges of the above window length h, standard deviation threshold s_thrd, delayed alarm duration t1, and delayed storage duration t2 are set by the operator according to the specific process system, based on their own process knowledge and operation experience.

[0084] In the matching judgment module, the alarm signal is used to prompt the operator that there is a mismatch between the current prediction model and the process system. When the operator receives the above alarm signal, they can choose to ignore it according to their own experience, or choose to maintain the model predictive controller. If the operator chooses to maintain the model predictive controller, the maintenance system calls the model fitting module to perform subsequent processing.

[0085] The storage module is used to, when receiving the storage signal, save the currently running prediction model as a historical prediction model to the model library; and save the controller parameters corresponding to the current prediction model as historical controller parameters to the parameter library.

[0086] Specifically, when the operator receives the alarm signal and chooses to maintain the model predictive controller, the maintenance system calls the model fitting module to, in cooperation with the operator's model fitting request, directly obtain a prediction model that matches the process system, or obtain a test model for subsequent processing, so as to finally obtain a prediction model that matches the process system. Specifically as follows:

[0087] The model fitting module is used to, when receiving the operator's model fitting request, based on the fitting data range input by the operator, fit the historical prediction models in the model library to obtain at least one historical prediction model whose fitting degree meets the requirements within the fitting data range as the candidate prediction models; and, based on the operator's selection operation on a certain candidate prediction model, output the candidate prediction model as the prediction model that matches the process system, or output it as a test model.

[0088] That is, based on the fitting data range input by the operator, the model fitting module obtains one or more candidate models that meet the requirements for the operator to select. The operator can directly select one of the candidate prediction models as the new prediction model that matches the process system; or, if the operator needs further specific information to make a judgment, the selected candidate prediction model by the operator is output as a test model, and the test module is called for testing; or, if the operator is not satisfied with any of the candidate prediction models, the maintenance system can, according to the model identification request triggered by the user, call the model identification module to generate a test model and call the test module for testing.

[0089] The model identification module is used to perform model identification based on the identification data range input by the operator when receiving the model identification request from the operator and generate a test model.

[0090] The test module is used to obtain test parameters adapted to the test model based on the test model; and, based on the test model and test parameters, perform open-loop simulation calculations according to the test data range and test parameter modification operations input by the operator to obtain an open-loop output, and based on the selected operation of the operator on the open-loop output, use the test model and test parameters corresponding to the open-loop output as the prediction model and controller parameters that match the process system.

[0091] Specifically, after the test module performs open-loop simulation calculations according to the test data range and test parameter modification operations input by the operator to obtain an open-loop output, the operator can determine whether the current prediction model and controller parameters are appropriate based on the open-loop output. If not, the operator can repeatedly modify the test parameters. Each time the test module receives the test parameter modification operation from the operator, it re-performs the open-loop simulation calculation until the obtained open-loop output is a result satisfactory to the operator. Based on the selected operation of the operator on the open-loop output, the corresponding test model and test parameters are used as the prediction model and controller parameters that match the process system.

[0092] To better understand the human-computer interaction logic of the maintenance system in the present invention, the interaction process between the maintenance system of the present invention and the operator is described below from the perspective of the operator.

[0093] When receiving the alarm signal generated by the maintenance system, the operator can judge whether the model predictive controller needs to be maintained according to various states and data of the process system and combined with his own experience. If the model predictive controller needs to be maintained, a model fitting request is sent, and the fitting data range is input, so that the maintenance system searches for candidate predictive models with a fitting degree meeting the requirements in the fitting data range from the model library through the model fitting module. For these candidate predictive models, the operator can make a selection: directly select the candidate predictive model considered suitable for the process system as the new predictive model in the model predictive controller; or if it is considered that the selected candidate predictive model needs to be tested for further judgment, then use the selected candidate predictive model as a test model for the subsequent test module for testing; or if it is considered that all the current candidate predictive models do not match the process system, a model identification request is sent, the identification data range is input, and a test model is generated through the model identification module. For the test models generated by the model fitting module and / or the model identification module, the operator inputs the test data range for testing, and judges whether the current test model matches the process system according to the open-loop output of the test model. If it is considered not to match, the test parameters are repeatedly updated through the test parameter modification operation until it is judged based on the open-loop output that the current test model matches the process system, and the test model and test parameters corresponding to the current open-loop output are selected as the predictive model and controller parameters matching the process system.

[0094] It should be noted that the specific implementation manner for the maintenance system of the present invention to receive data such as the fitting data range, identification data range, and test data range input by the operator, as well as operation contents such as the model fitting request, model identification request, test parameter modification operation, and selection operation sent by the operator, is the prior art and will not be elaborated in the present invention.

[0095] The maintenance system proposed in the embodiments of the present invention stores historical prediction models and historical controller parameters that match the process system through an incoming library module for use by a model fitting module, a model identification module, and a testing module. The model fitting module, the model identification module, and the testing module of the maintenance system interact with the operator, perform model fitting, model identification, or testing based on the incoming prediction models and controller parameters according to the data range input by the operator, and output the results, and perform subsequent operations according to the decisions made by the operator based on the above results. Therefore, while reasonably utilizing the operator's process knowledge and operation experience, the maintenance process of the model predictive controller is modularized. While ensuring the openness of the maintenance system, the operation difficulty of the operator in debugging the model predictive controller is reduced, enabling a deep interaction between the model predictive controller and the operator, enabling the maintenance system to perform calculations, simulations, and adjustments according to the data range input by the operator, and outputting corresponding models, controller parameters, or intermediate results, while the operator can judge and select the advantages and disadvantages of the corresponding models, controller parameters, or intermediate results based on their own process knowledge and operation experience to obtain a better model predictive controller, thereby achieving the purpose of optimizing the control effect of the model predictive controller on the process system.

[0096] Embodiment 2

[0097] To better understand Embodiment 1, this embodiment will be described in detail in combination with the specific architecture of the maintenance system.

[0098] The embodiments of the present invention provide a human-computer interactive maintenance system for a model predictive controller, including: a matching judgment module, an incoming library module, a model fitting module, a model identification module, and a testing module. Each module will be described in detail below.

[0099] The matching judgment module is used to obtain the real-time process data of the process system, obtain the controller parameters of the model predictive controller, and judge whether the currently running prediction model matches the process system based on the matching indicators preset by the operator; if so, generate an incoming library signal; if not, generate an alarm signal.

[0100] Specifically, the matching indicators include: window length h, standard deviation threshold s_thrd, delay alarm duration t1, and delay incoming library duration t2; and h < t1, h < t2, and both t1 and t2 are integer multiples of h.

[0101] The process by which the matching judgment module judges whether the currently running prediction model matches based on the above matching indicators includes:

[0102] A1. Based on the real-time process data of the process system, obtain the measured value pv(i,t) of the controlled variable at each window length h.

[0103] A2. Calculate the deviation e(i,t) between the measured value pv(i,t) and the set value sp(i,t) of the i-th controlled variable at the t-th moment according to formula (1), where i is a positive integer.

[0104] A3. Based on e(i,t), calculate the standard deviation s(i,t) of the i-th controlled variable within the window length h at the t-th moment according to formula (2).

[0105] A4. Determine whether the values of s(i,t), s(i,t - h), s(i,t - 2h), ……, s(i,t - t1) within the continuous t1 duration pushed forward from the t-th moment are all not within the standard deviation threshold s_thrd; if so, determine that the current prediction model does not match the process system, generate an alarm signal, and jump to A1; if not, jump to A1.

[0106] A5. Determine whether the values of s(i,t), s(i,t - h), s(i,t - 2h), ……, s(i,t - t2) within the continuous t2 duration pushed forward from the t-th moment are all within the standard deviation threshold s_thrd; if so, determine that the current prediction model matches the process system, generate a warehousing signal, and jump to A1; if not, jump to A1.

[0107] Among them, the formula (1) is:

[0108] e(i,t) = pv(i,t) - sp(i,t) (1)

[0109] In formula (1), pv(i,t) represents the measured value of the i-th controlled variable at the t-th moment, and sp(i,t) represents the set value of the i-th controlled variable at the t-th moment;

[0110] The formula (2) is:

[0111]

[0112] In formula (2), e(i,t - j) represents the deviation between the measured value and the set value of the i-th controlled variable at the t - j moment.

[0113] In the matching judgment module, in this embodiment, the influence of process fluctuations in the process system on the judgment result is eliminated by setting a delay alarm duration t1 and a delay warehousing duration t2. In actual production, the process system often has fluctuations in its controlled variables due to input fluctuations or external disturbances, which deteriorate the control effect of the model predictive controller within a limited period of time. However, this does not represent the actual control effect of the model predictive controller. Therefore, the present invention improves the tolerance of the matching judgment module to process fluctuations by setting a delay alarm duration t1 in the matching index. An alarm signal is generated only when the large fluctuation duration of the controlled variable in the process system exceeds the delay alarm duration t1. Similarly, since the controlled variable is stable within a limited period of time due to accidental reasons, it does not mean that the model predictive controller has a good control effect. A more accurate determination of the model predictive controller needs to be made on a longer time scale. Therefore, the present invention improves the judgment criterion of the matching judgment module by setting a delay warehousing duration t2 in the matching index. When the controlled variable in the process system has good stability on a longer time scale, it is determined that the current prediction model matches the process system.

[0114] The schematic diagram of the working process of the matching judgment module is as Figure 2 shown.

[0115] The warehousing module is used to save the currently running prediction model as a historical prediction model to the model library when receiving a warehousing signal; and save the controller parameters corresponding to the current prediction model as historical controller parameters to the parameter library.

[0116] Specifically, the prediction model includes model parameters; more specifically, when the model is a second-order model, the model parameters include: gain, time constant one, time constant two, time delay, and integral flag bit.

[0117] Saving the currently running prediction model as a historical prediction model to the model library means saving the model parameters of the historical prediction model to the model library. In addition, for the convenience of managing and classifying and searching the model library and the parameter library, the number of controlled variables, the number of manipulated variables, the sampling time, and the warehousing time corresponding to the historical prediction model or the historical controller parameters can also be saved accordingly.

[0118] The model fitting module is used to, when receiving a model fitting request from an operator, fit the historical prediction models in the model library based on the fitting data range input by the operator, obtain at least one historical prediction model with a fitting degree meeting the requirements within the fitting data range as the candidate prediction model; and output the candidate prediction model as the prediction model matching the process system or as a test model based on the operator's selection operation on a certain candidate prediction model.

[0119] Specifically, the fitting data range includes: the first numerical range of the model parameters of the target prediction model, and the first historical process data for fitting. In particular, the first numerical range can be the value range of all model parameters of the target prediction model, or it can be the value range of several model parameters selected by the operator. For example, if the operator is only interested in two model parameters, namely time constant one and time constant two, then only the value ranges of these two model parameters can be set, and the value ranges of the remaining three model parameters are not limited, and the maintenance system defaults the value ranges of the remaining three model parameters to (-∞, +∞).

[0120] The process of fitting the historical prediction models in the model library to obtain at least one historical prediction model with a fitting degree meeting the requirements within the fitting data range specifically includes:

[0121] B1. Preselect the historical prediction models in the model library based on the first numerical range, and regard the historical prediction models whose model parameters fall within the first numerical range as the initial prediction models. Based on step B1, the model fitting module can first screen out some historical prediction models that do not meet the requirements, which can reduce the calculation amount on the one hand and prevent overfitting problems in the subsequent steps B2 - B3 on the other hand.

[0122] B2. Based on the initial prediction model and the controller parameters corresponding to the initial prediction model, construct a fitting model predictive controller, and drive the fitting model predictive controller to run based on the first historical process data to obtain the predicted value pv1 pre (i,t) of the controlled variable of the fitting model predictive controller on the first historical process data.

[0123] B3. Based on the predicted value pv1 pre (i,t) of the controlled variable, calculate the fitting degree index fit(i) of each initial prediction model according to formula (3), and output the top N initial prediction models with the highest fitting degree index fit(i) as the candidate prediction models. Where N is a positive integer.

[0124] The formula (3) is:

[0125]

[0126] In formula (3), fit(i) represents the fitting degree evaluation index of the i-th controlled variable;

[0127] pv(i,t) represents the measured value of the i-th controlled variable at the t-th moment in the first historical process data;

[0128] pv1 pre(i, t) represents the predicted value of the i-th controlled variable at the t-th moment obtained by running the fitting model prediction controller based on the drive of the first historical process data;

[0129] pv mean (i) represents the average value of the measured values of the i-th controlled variable in the first historical process data;

[0130] max(a, b) represents the operation of taking the maximum value of a and b;

[0131] min(a, b) represents the operation of taking the minimum value of a and b.

[0132] The model identification module is used to perform model identification based on the identification data range input by the operator when receiving the model identification request from the operator, and generate a test model.

[0133] Specifically, the identification data range includes: the second numerical range of the model parameters of the target prediction model, and the second historical process data for model identification.

[0134] Performing model identification based on the identification data range input by the operator to generate a test model includes:

[0135] Based on the grid search method, search within the second numerical range according to the evaluation function J, and output the model parameters with the best evaluation as the test model. Specifically, in formula (4), the smaller the value of J, the better the evaluation of the corresponding model parameters.

[0136] The evaluation function J is represented by formula (4), and formula (4) is:

[0137]

[0138] In formula (4), pv(i, t) represents the measured value of the i-th controlled variable at the t-th moment based on the historical process data within the second time range;

[0139] T represents the length of the time range corresponding to the second historical process data;

[0140] pv2 pre (i, t) represents the predicted value of the i-th controlled variable at the t-th moment, which is obtained by formula (5);

[0141] Formula (5) is:

[0142]

[0143] In formula (5), u(j, t - l) represents the measured value of the j-th manipulated variable at the (t - l)-th moment in the second historical process data;

[0144] g(i, j, l) represents the impulse response coefficient between the j-th manipulated variable and the i-th controlled variable, where l, i, and j are all positive integers;

[0145] L ij is the model length corresponding to the current model parameters.

[0146] A test module, configured to obtain test parameters adapted to the test model based on the test model; and, based on the test model and the test parameters, perform open-loop simulation calculations according to the test data range and test parameter modification operations input by the operator, obtain an open-loop output, and based on the selected operation of the operator on the open-loop output, use the test model and test parameters corresponding to the open-loop output as the prediction model and controller parameters that match the process system.

[0147] Specifically, the test data range includes: the third historical process data for testing.

[0148] The test module includes a parameter matching unit and a virtual test unit, specifically as follows:

[0149] The parameter matching unit is configured to, based on the test model, traverse the historical prediction models in the model library to calculate distances, obtain the historical prediction model with the closest distance to the test model, and call the historical controller parameters corresponding to the historical prediction model from the parameter library as the test parameters adapted to the test model.

[0150] In the above parameter matching unit, the traversing the historical prediction models in the model library to calculate distances further includes the following steps:

[0151] Receive the model gain weight w1, time constant weight w2, and time delay weight w3 input by the operator, traverse the historical prediction models in the model library, and calculate the distance D between the historical prediction model and the test model according to formula (6). The smaller the value of D, the closer the distance.

[0152] The formula (6) is:

[0153]

[0154] In formula (6), K(i, j), T1(i, j), T2(i, j), T d (i, j) respectively represent the gain, time constant one, time constant two, and time delay between the j-th manipulated variable and the i-th controlled variable in the test model; K c (i, j), T 1c (i, j), T 2c (i, j), T dc(i, j) respectively represent the gain, time constant one, time constant two, and time delay between the j-th manipulated variable and the i-th controlled variable in the historical prediction model.

[0155] A virtual test unit, configured to calculate, based on the test model and test parameters and the third process data, the open-loop output of the test model based on the test parameters in a virtual environment; and, when receiving a test parameter modification operation from an operator, update the test parameters according to the test parameter modification operation, and recalculate the open-loop output of the test model based on the new test parameters in the virtual environment until receiving a selection operation from the operator for the current test model and test parameters, and output the test model and the controller test parameters as the prediction model and controller parameters matching the process system.

[0156] In the above virtual test unit, the step of calculating, based on the third process data, the open-loop output of the test model based on the test parameters in a virtual environment further includes the following steps:

[0157] Construct a test model prediction controller in a simulation environment according to the test parameters and the test model, and drive the test model prediction controller to run based on the third process data to obtain the manipulated variables continuously output by the test model prediction controller in an open-loop state, forming an open-loop manipulated variable sequence, and use the open-loop manipulated variable sequence as the open-loop output of the test model based on the test parameters.

[0158] It should be noted that the above first process data, second process data, and third process data are data reflecting the operating state of the process system continuously collected and saved by the process system through existing technical means, specifically including but not limited to temperature data, pressure data, flow data, liquid level data, etc. collected by various devices in the process system based on sensors. Since the process data is continuously recorded, in actual operation, the above first process data, second process data, and third process data can be limited by specifying the corresponding time range.

[0159] Embodiment III

[0160] The embodiment of the present invention further provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the computer program is executed by the processor, it implements the architecture of the human-computer interactive maintenance system of the model predictive controller in Embodiment I or II.

[0161] Since the system / apparatus described in the above embodiments of the present invention is the system / apparatus adopted for implementing the method of the above embodiments of the present invention, those skilled in the art can understand the specific structure and variations of the system / apparatus based on the method described in the above embodiments of the present invention, and thus will not be elaborated herein. Any system / apparatus adopted by the method of the above embodiments of the present invention falls within the scope of protection of the present invention.

[0162] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system or a computer program product. Therefore, the present invention can be implemented in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can be implemented in the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0163] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems) and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be realized by computer program instructions.

[0164] It should be noted that in the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in the claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present invention can be implemented by means of hardware including several different elements and by means of a suitably programmed computer. In the claims listing several apparatuses, several of these apparatuses can be embodied by the same hardware. The use of the words first, second, third, etc. is only for convenience of expression and does not denote any order. These words can be construed as part of the element name.

[0165] In addition, it should be noted that in the description of this specification, the description of terms such as "an embodiment", "some embodiments", "embodiments", "examples", "specific examples" or "some examples" means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0166] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications after learning the basic creative concept. Therefore, the claims should be construed to cover the preferred embodiments as well as all changes and modifications falling within the scope of the present invention.

[0167] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and its equivalent technologies, the present invention should also cover these modifications and variations.

Claims

1. A human-computer interactive maintenance system for a model predictive controller, which is used to maintain the model predictive controller of a process system, characterized in that, The maintenance system includes: A matching and judging module, which is used to obtain the real-time process data of the process system, obtain the controller parameters of the model predictive controller, and judge whether the currently running prediction model matches the process system based on the matching index preset by the operator; if so, generate a warehousing signal; if not, generate an alarm signal; A warehousing module, which is used to save the currently running prediction model as a historical prediction model to the model library when receiving the warehousing signal, and save the controller parameters corresponding to the current prediction model as historical controller parameters to the parameter library; A model fitting module, which is used to, when receiving the model fitting request of the operator, fit the historical prediction models in the model library based on the fitting data range input by the operator, and obtain at least one historical prediction model with a fitting degree meeting the requirements within the fitting data range as the candidate prediction models; and, based on the selection operation of the operator on a certain candidate prediction model, output the candidate prediction model as the prediction model matching the process system, or output it as a test model; A model identification module, which is used to, when receiving the model identification request of the operator, perform model identification based on the identification data range input by the operator and generate a test model; A test module, which is used to obtain test parameters adapted to the test model based on the test model; and, based on the test model and the test parameters, perform open-loop simulation calculation according to the test data range and test parameter modification operation input by the operator to obtain an open-loop output, and based on the selection operation of the operator on the open-loop output, use the test model and test parameters corresponding to the open-loop output as the prediction model and controller parameters matching the process system; In the model fitting module, the fitting data range includes: the first numerical range of the model parameters of the target prediction model, and the first historical process data for fitting; The fitting of the historical prediction models in the model library to obtain at least one historical prediction model with a fitting degree meeting the requirements within the fitting data range includes: Preselecting the historical prediction models in the model library based on the first numerical range, and taking the historical prediction models whose model parameters fall within the first numerical range as the initial prediction models; Based on the initial prediction model and the controller parameters corresponding to the initial prediction model, a fitting model predictive controller is constructed, and the fitting model predictive controller is driven to operate based on the first historical process data, so as to obtain the predicted value pv1 of the controlled variable of the fitting model predictive controller on the first historical process data pre (i,t), Based on the predicted value pv1 of the controlled variable pre (i, t), calculate the goodness-of-fit index fit(i) of each initial prediction model according to formula (3), and output the top N initial prediction models with the highest goodness-of-fit index fit(i) as candidate prediction models; where N is a positive integer; The formula (3) is: In formula (3), fit(i) represents the fitting degree evaluation index of the i-th controlled variable; pv(i,t) represents the measured value of the i-th controlled variable at the t-th moment in the first historical process data; pv1 pre (i, t) represents the predicted value of the i-th controlled variable obtained by the fitting model predictive controller at the t-th moment based on the drive of the first historical process data; pv mean (i) represents the average value of the measured value of the i-th controlled variable in the first historical process data; max(a,b) represents the operation of taking the maximum value of a and b; min(a,b) represents the operation of taking the minimum value of a and b; In the model identification module, the identification data range includes: the second numerical range of the model parameters of the target prediction model, and the second historical process data for model identification; The generating of the test model by performing model identification based on the identification data range input by the operator includes: Searching within the second numerical range based on the grid search method according to the evaluation function J, and outputting the model parameters with the optimal evaluation as the test model; The evaluation function J is represented by formula (4), and the formula (4) is: In formula (4), pv(i,t) represents the measured value of the i-th controlled variable at the t-th moment based on the historical process data within the second time range; T represents the length of the time range corresponding to the second historical process data; pv2 pre (i, t) represents the predicted value of the i-th controlled variable at the t-th moment, which is obtained through formula (5); The formula (5) is as follows: In formula (5), u(j,t-l) represents the measured value of the j-th manipulated variable at the (t-l)-th moment in the second historical process data; g(i,j,l) represents the impulse response coefficient between the j-th manipulated variable and the i-th controlled variable, where l, i, and j are all positive integers; L ij is the model length corresponding to the current model parameters.

2. The maintenance system according to claim 1, characterized in that In the matching judgment module, the controller parameters include: the set value of the controlled variable; the real-time process data includes: the measured value of the controlled variable and the measured value of the manipulated variable; The matching indicators include: the window length h, the standard deviation threshold s_thrd, the delayed alarm duration t1, and the delayed storage duration t2; and h < t1, h < t2, and both t1 and t2 are integer multiples of h; The window length h represents the time length of the measured values of the controlled variables used each time to judge whether the currently running prediction model matches the process system; The standard deviation threshold s_thrd represents the range index allowing the fluctuation of the measured values of the controlled variables; The delayed alarm duration t1 represents the duration allowing the fluctuation of the measured values of the controlled variables to exceed the standard deviation threshold s_thrd without alarm; The delayed storage duration t2 represents the duration allowing the measured values of the controlled variables to fluctuate within the standard deviation threshold s_thrd without being stored in the database.

3. The maintenance system according to claim 2, wherein In the matching judgment module, the judgment of whether the currently running prediction model matches according to the real-time process data based on the matching indicators preset by the operator includes: Based on the real-time process data of the process system, obtain the measured values of the controlled variables at each window length h; According to formula (1), calculate the deviation e(i,t) between the measured value pv(i,t) and the set value sp(i,t) of the i-th controlled variable at the t-th moment, where i is a positive integer; Based on e(i,t), calculate the standard deviation s(i,t) of the i-th controlled variable within the window length h at the t-th moment according to formula (2); Judge whether the values of s(i,t), s(i,t-h), s(i,t-2h), ……, s(i,t-t1) are all not within the standard deviation threshold s_thrd within the consecutive t1 duration pushed forward from the t-th moment; if so, determine that the current prediction model does not match the process system; Judge whether the values of s(i,t), s(i,t-h), s(i,t-2h), ……, s(i,t-t2) are all within the standard deviation threshold s_thrd within the consecutive t2 duration pushed forward from the t-th moment; if so, determine that the current prediction model matches the process system; The formula (1) is as follows: e(i,t) = pv(i,t) - sp(i,t) (1) In formula (1), pv(i,t) represents the measured value of the i-th controlled variable at the t-th moment, and sp(i,t) represents the set value of the i-th controlled variable at the t-th moment; The formula (2) is as follows: In formula (2), e(i, t-j) represents the deviation between the measured value and the set value of the i-th controlled variable at the (t-j)-th moment.

4. The maintenance system according to claim 1, characterized in that The prediction model includes model parameters; When the model is a second-order model, the model parameters include: gain, time constant one, time constant two, time delay, and integral flag bit.

5. The maintenance system according to claim 4, characterized in that, In the test module, the test data range includes: the third historical process data for testing; The test module includes: A parameter matching unit, configured to calculate distances by traversing historical prediction models in the model library based on the test model, obtain the historical prediction model closest to the test model, and call the corresponding historical controller parameters of the historical prediction model from the parameter library as the test parameters adapted to the test model; A virtual test unit, configured to calculate the open-loop output of the test model based on the test parameters in a virtual environment based on the test model and test parameters and the third process data; and, when receiving a test parameter modification operation from an operator, update the test parameters according to the modification operation and recalculate the open-loop output of the test model based on the new test parameters in the virtual environment until receiving a selection operation from the operator for the current test model and test parameters, and output the test model and controller test parameters as the prediction model and controller parameters matching the process system.

6. The maintenance system according to claim 5, characterized in that In the parameter matching unit, the calculating distances by traversing historical prediction models in the model library includes: Receiving the model gain weight w1, time constant weight w2, and time delay weight w3 input by the operator, traversing the historical prediction models in the model library, and calculating the distance D between the historical prediction model and the test model according to formula (6), where formula (6) is: In formula (6), K(i,j), T1(i,j), T2(i,j), and T d (i,j) respectively represent the gain, time constant one, time constant two, and time delay between the j-th manipulated variable and the i-th controlled variable in the test model; K c (i,j), T 1c (i,j), T 2c (i,j), T dc (i,j) respectively represent the gain, time constant one, time constant two, and time delay between the j-th manipulated variable and the i-th controlled variable in the historical prediction model.

7. The maintenance system according to claim 5, characterized in that In the virtual test unit, the calculating the open-loop output of the test model based on the test parameters in a virtual environment based on the third process data includes: Constructing a test model prediction controller in a simulation environment according to the test parameters and the test model, and driving the test model prediction controller to run based on the third process data to obtain the operating variables continuously output by the test model prediction controller in an open-loop state, forming an open-loop operating variable sequence, and using the open-loop operating variable sequence as the open-loop output of the test model based on the test parameters.

8. A computer device, characterized in that, Includes: A memory, a processor, and a computer program stored on the memory and executable on the processor, where when the computer program is executed by the processor, it implements the architecture of the human-computer interactive maintenance system of the model predictive controller according to any one of claims 1 to 7.

Images