Intelligent multi-model predictive control method and system based on twin model

Abstract

The application discloses an intelligent multi-model predictive control method and system based on a twin model. The intelligent multi-model predictive control method comprises the following steps: constructing a basic multi-model predictive controller by using multiple working condition basic models and linear interpolation blocks, realizing stable control in the basic multi-model predictive controller, and generating a twin model; constructing an intelligent multi-model predictive controller by using multiple twin models and a selection module; and sequentially performing multi-working condition twin model matching in the intelligent multi-model predictive controller, and finally realizing all-working condition optimal control. The application can realize accurate control in each working condition, and completes the diagnosis and optimization of a controlled system online.

Description

Technical Field

[0001] This invention belongs to the field of industrial process control technology, specifically a smart multi-model predictive control method and system based on twin models. Background Technology

[0002] In industrial process control systems, predictive control systems can effectively solve the optimization problem of process control systems, but they have high requirements for the model. When the system has multiple operating conditions, changes in the model have a significant impact on control performance. Therefore, an intelligent multi-model predictive control method and system that can adapt to all operating conditions is needed.

[0003] Predictive control based on existing models places high demands on the model itself. When the operating conditions change, the characteristics of the controlled object will change significantly, and the predictive control performance of the model will decrease considerably. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to overcome the defects of the prior art and provide an intelligent multi-model predictive control method and system based on twin models, which can perform precise control under each working condition and ultimately achieve full-condition optimized control, thereby completing the diagnosis and optimization of the controlled system online.

[0005] Therefore, the present invention adopts the following technical solution: an intelligent multi-model predictive control method based on a twin model, comprising:

[0006] A basic multi-model predictive controller is constructed using multiple operating condition basic models and linear interpolation blocks. Stable control is achieved and twin models are generated in the basic multi-model predictive controller.

[0007] A smart multi-model predictive controller is constructed using multiple twin models and selection modules;

[0008] In the intelligent multi-model predictive controller, multi-condition twin model matching is performed successively to ultimately achieve full-condition optimized control.

[0009] Furthermore, for a multi-condition controlled system, the intelligent multi-model predictive controller initially contains only an accurate model for one condition for optimized control.

[0010] Furthermore, when the intelligent multi-model predictive controller detects a deterioration in control performance, it switches to basic predictive control. Once the multi-condition controlled system stabilizes, an excitation signal is added to the control command, and closed-loop response data is recorded online. The parameter range of the identifier is set to the parameters of each adjacent model in the basic multi-model predictive controller before online closed-loop model identification is performed. After identification, a model step disturbance simulation experiment is conducted, followed by a step disturbance test on a real object.

[0011] Further, according to the set value and the error amplitude, the system steady state is judged, if the set value is not changed and the error amplitude is in a certain interval, it is considered that the stable state is reached.

[0012] Further, the closed loop identifier can identify the basic model of multiple different working conditions, remember the identification parameters each time, according to the principle of the closest current working condition and each time identification working condition, switch to the latest working condition basic model, which is the basic model identified on line in the corresponding working condition closed loop; the basic predictive control is switched to intelligent predictive control for optimization control, if the working condition changes again, the control performance is poor again, the foregoing excitation and modeling process is repeated, and the corresponding working condition twin model is generated; according to the complex working condition, the corresponding number of twin model identification, simulation and control is set, and when the twin model is generated more than the specified number of times, it is considered that the modeling of the whole working condition is completed.

[0013] Further, the performance judgment module judges whether the control performance is suitable according to multiple amplitude thresholds and delay times of the error.

[0014] Further, after generating the twin model multiple times, accurate control is performed in each working condition, and whole working condition optimization control is realized.

[0015] The application also provides an intelligent multi-model predictive control system based on a twin model, which comprises:

[0016] The basic multi-model predictive controller is constructed by multiple working condition basic models and a linear interpolation block.

[0017] The twin model is realized in the basic multi-model predictive controller to generate a twin model.

[0018] The intelligent multi-model predictive controller is constructed by multiple twin models and a selection module.

[0019] The whole working condition optimization control unit sequentially performs multiple working condition twin model matching in the intelligent multi-model predictive controller, and finally realizes whole working condition optimization control.

[0020] The application has the following beneficial effects: the basic multi-model predictive controller is constructed by multiple working condition basic models and a linear interpolation block, the twin model is realized in the basic multi-model predictive controller to generate a twin model, the intelligent multi-model predictive controller is constructed by the twin model, accurate control can be performed in each working condition, the whole working condition optimization control unit sequentially performs multiple working condition twin model matching in the intelligent multi-model predictive controller, and finally realizes whole working condition optimization control, so that the diagnosis and optimization of the controlled system are completed on line.

[0021] The application will be further described below in combination with the drawings and specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 Flow chart of the intelligent multi-model predictive control method based on twin model of the present application;

[0023] Figure 2 The twin model is established and the control simulation running curve chart (in the figure, 1 is the set value, 2 is the controlled quantity, 3 is the simulation set value, 4 is the simulation model output, 5 is the simulation control quantity, 6 is the actual control quantity, 7 is the identified model parameter, and 8 is the working condition change value) of the present application is shown in the figure;

[0024] Figure 3 The intelligent multi-model predictive controller full working condition running curve chart (in the figure, 1 is the set value, 2 is the controlled quantity, 3 is the control quantity, and 4 is the working condition value) of the present application is shown in the figure. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0026] Embodiment 1

[0027] The present embodiment provides an intelligent multi-model predictive control method based on twin model, as shown in the figure, which constructs a basic multi-model predictive controller with a linear interpolation block and a plurality of working condition basic models (first, each basic working condition is identified offline in an open loop to establish a model, and then the model parameters are formed into a multi-model system through an interpolation method), realizes stable control in the basic multi-model predictive controller and generates a twin model, constructs an intelligent multi-model predictive controller with the twin model and a selection module (the twin model is generated in an online closed loop identification at each working condition point, and the selection module automatically selects the corresponding twin model according to different working conditions), and sequentially performs multi-working condition twin model matching in the intelligent multi-model predictive controller to finally realize full-working condition optimal control. Figure 1

[0028] The intelligent multi-model predictive controller initially only uses one twin model to perform optimal control on the multi-working condition controlled system, and when the system working condition changes, the control system performance deteriorates, and the basic predictive control is switched in. In the figure, the performance judgment module judges whether the control performance is appropriate according to a plurality of amplitude thresholds and delay times of the error, the greater the error judgment threshold, the shorter the delay time, and the easier to trigger the basic predictive control.

[0029] ​The basic multi-model predictive controller is a linear interpolation-based multi-model controller. Since the model is linearly interpolated at typical working points, the model of the basic multi-model predictive controller is relatively accurate, and convergence and stability can be ensured when the basic multi-model predictive controller is used to control a multi-working-point controlled system.

[0030] When it is judged that the basic multi-model predictive controller is stable, an excitation signal is added to the control instruction to excite the basic multi-model predictive controller, and the closed-loop response data are recorded online. The parameter range of the identifier is set to the parameters of the adjacent models in the basic multi-model predictive controller, and then online closed-loop model identification is performed. After the identification is completed, model step disturbance simulation experiment is performed, and then step disturbance experiment is performed on the actual object. In the figure, the system steady state is judged according to the set value and the error amplitude. If the set value does not change and the error amplitude is in a certain interval, it is considered that the steady state is reached.

[0031] The twin model generated by the closed-loop identifier can identify the basic model of multiple different working points, and remember the identification parameters each time. The intelligent multi-model predictive controller switches to the latest working point basic model (i.e., the model identified online in the corresponding working point) according to the principle of closest to the current working point and each time identification working point. The basic predictive control is switched to intelligent predictive control to optimize the control of the multi-working-point controlled system. If the working point changes again, the control performance is poor again, and the above step is repeated to generate the corresponding twin model.

[0032] According to the complex working conditions, the corresponding number of twin model identification, simulation and control can be set. When the number of generated twin models exceeds the specified number, it is considered that the modeling of the whole working condition is completed. After generating the twin model for multiple times, the intelligent multi-model predictive controller can accurately control in each working condition, and realize the optimal control of the whole working condition.

[0033] In the attached Figure 2 In the attached The gain function of the basic multi-model predictive controller corresponding to working condition 1 to working condition 10 is set as (input: 0, 1, 5, 10; output 0, 0.5, 1.5, 2); the inertia time function is (input: 0, 1, 5, 10; output 0, 20, 50, 70); and the delay time function is (input: 0, 1, 5, 10; output 0, 15, 50, 60).

[0034] The gain function of the basic multi-model predictive controller corresponding to working condition 1 to working condition 10 is set as (input: 0, 1, 10; output 0, 0.5, 2); the inertia time function is (input: 0, 1, 10; output 0, 20, 70); and the delay time function is (input: 0, 1, 10; output 0, 15, 60).

[0035] When the operating condition value is 1, the intelligent multi-model predictive controller based on the twin model first identifies the twin model and conducts simulation experiments and disturbance tests on the actual object. When the operating condition value changes to 4, the control performance deteriorates due to the change in the controlled object model. The system repeats the identification, simulation and control process to achieve high-quality control. When the operating condition value changes to 9, the control performance deteriorates again due to the change in the controlled object model. The system identifies the twin model again and achieves high-quality control through simulation and control.

[0036] exist Figure 3 The figure shows the full-condition operation curve of the intelligent multi-model predictive controller. When the condition value changes from 3 to 7, the intelligent multi-model predictive controller automatically switches models, and the control response is stable and the performance is excellent. When the condition value changes from 7 to 1, the intelligent multi-model predictive controller automatically adapts to the model, and the controlled variable stabilizes near the set value after a short period of fluctuation, and the control quality is excellent.

[0037] Example 2

[0038] This embodiment provides an intelligent multi-model predictive control system based on a twin model, which consists of a basic multi-model predictive controller, a twin model, an intelligent multi-model predictive controller, and a full-condition optimization control unit.

[0039] Basic multi-model predictive controller: The basic multi-model predictive controller is constructed using multiple basic operating condition models and linear interpolation blocks (first, offline open-loop identification is performed for each basic operating condition to establish a model, and then the parameters of each model are interpolated to form a multi-model system).

[0040] Twin model: Stabilized control is achieved and a twin model is generated in the basic multi-model predictive controller.

[0041] Intelligent multi-model predictive controller: The intelligent multi-model predictive controller is constructed using multiple twin models and a selection module (the twin models are generated online in closed loop at each operating point, and the selection module automatically selects the corresponding twin model according to different operating conditions).

[0042] Full-condition optimization control unit: Multi-condition twin model matching is performed successively in the intelligent multi-model predictive controller to ultimately achieve full-condition optimization control.

[0043] For a multi-condition controlled system, the intelligent multi-model predictive controller initially contains only an accurate model for one condition for optimized control.

[0044] When the control performance is detected to be poor, the base predictive control is switched to, and when the multi-working-condition controlled system is stable, an excitation signal is added to the control command, the closed-loop response data is recorded on line, the parameter range of the identifier is set to the parameters of the adjacent models in the base multi-model predictive controller, and then the on-line closed-loop model identification is performed; after the identification is completed, the model step disturbance simulation experiment is performed, and then the step disturbance experiment for the actual object is performed. According to the set value and the error amplitude, the system steady state is judged, and if the set value does not change and the error amplitude is in a certain interval, it is considered that the stable state is reached.

[0045] The closed-loop identifier performs the base model identification of multiple different working conditions, remembers the identification parameters each time, and according to the principle of the closest working condition, switches to the latest working condition base model (i.e. the model identified on line in the corresponding working condition); the base predictive control is switched to the intelligent predictive control for optimized control, if the working condition changes again, the control performance is poor again, the excitation and modeling process are repeated, and the twin model corresponding to the working condition is generated; according to the complex situation of the working condition, the corresponding number of twin model identification, simulation and control is set, and when the number of twin models generated exceeds the specified number, it is considered that the modeling of the whole working condition is completed. The performance judgment module judges whether the control performance is suitable according to the multiple amplitude thresholds and delay times of the error.

[0046] After the twin model is generated for multiple times, the precise control is performed in each working condition, and the whole working condition optimized control is realized.

[0047] The above embodiments only exemplarily illustrate the principles and effects of the present application, and are not used to limit the present application. Any person skilled in the art can modify or change the above embodiments without departing from the spirit and scope of the present application. Therefore, all equivalent modifications or changes completed by those skilled in the art without departing from the spirit and technical thought disclosed by the present application should be covered by the claims of the present application.

Claims

1. A smart multi-model predictive control method based on a twin model, characterized in that, include: A basic multi-model predictive controller is constructed using multiple operating condition basic models and linear interpolation blocks. Stable control is achieved and twin models are generated in the basic multi-model predictive controller. A smart multi-model predictive controller is constructed using multiple twin models and selection modules; In the intelligent multi-model predictive controller, multi-condition twin model matching is performed successively to ultimately achieve full-condition optimized control. For a multi-condition controlled system, the intelligent multi-model predictive controller initially contains only an accurate model for one condition for optimized control. When the intelligent multi-model predictive controller detects a deterioration in control performance, it switches to basic predictive control. Once the multi-condition controlled system stabilizes, an excitation signal is added to the control command, and closed-loop response data is recorded online. The parameter range of the identifier is set to the parameters of each adjacent model in the basic multi-model predictive controller before online closed-loop model identification is performed. After identification, a model step disturbance simulation experiment is conducted, followed by a step disturbance test on the actual object. The closed-loop identifier performs basic model identification for multiple different operating conditions, remembers the identification parameters each time, and switches to the latest basic model based on the principle that the current operating condition is closest to the operating conditions identified in each identification. This model is the basic model identified online in the corresponding operating condition. The basic predictive control switches to intelligent predictive control for optimized control. If the operating condition changes again and the control performance deteriorates again, the aforementioned excitation and modeling process is repeated to generate a twin model for the corresponding operating condition. Depending on the complexity of the operating condition, a certain number of twin model identifications, simulations, and controls are set. When a twin model is generated more than the specified number of times, it is considered that the modeling of all operating conditions has been completed.

2. The intelligent multi-model predictive control method based on a twin model according to claim 1, characterized in that, The system steady state is determined based on the set value and the error amplitude. If the set value remains unchanged and the error amplitude is within a certain range, the system is considered to have reached a stable state.

3. The intelligent multi-model predictive control method based on a twin model according to claim 1, characterized in that, The performance evaluation module determines whether the control performance is appropriate based on multiple amplitude thresholds of the error and the delay time.

4. The intelligent multi-model predictive control method based on a twin model according to claim 1, characterized in that, After generating twin models multiple times, precise control is performed under each operating condition to achieve optimized control across all operating conditions.

5. An intelligent multi-model predictive control system based on a twin model, characterized in that, include: Basic multi-model predictive controller: Construct a basic multi-model predictive controller using multiple operating condition basic models and linear interpolation blocks; Twin model: Stabilized control is implemented in the aforementioned basic multi-model predictive controller, and a twin model is generated; Intelligent multi-model predictive controller: Constructs an intelligent multi-model predictive controller using multiple twin models and selection modules; Full-condition optimization control unit: Multi-condition twin model matching is performed successively in the intelligent multi-model predictive controller to finally achieve full-condition optimization control; For a multi-condition controlled system, the intelligent multi-model predictive controller initially contains only an accurate model for one condition for optimized control. When a deterioration in control performance is detected, switch to basic predictive control. Once the multi-condition controlled system stabilizes, add an excitation signal to the control command, record the closed-loop response data online, and set the parameter range of the identifier to the parameters of each adjacent model in the basic multi-model predictive controller before performing online closed-loop model identification. After identification, a model step disturbance simulation experiment is conducted, followed by a step disturbance test on the actual object. The closed-loop identifier performs basic model identification for multiple different operating conditions, remembers the identification parameters each time, and switches to the latest basic model based on the principle that the current operating condition is closest to the operating conditions identified in each identification. This model is the basic model identified online in the corresponding operating condition. The basic predictive control switches to intelligent predictive control for optimized control. If the operating condition changes again and the control performance deteriorates again, the aforementioned excitation and modeling process is repeated to generate a twin model for the corresponding operating condition. Depending on the complexity of the operating condition, a certain number of twin model identifications, simulations, and controls are set. When a twin model is generated more than the specified number of times, it is considered that the modeling of all operating conditions has been completed.

Images