Hybrid cooling system multi-target parallel control method and system based on virtual-real migration

Abstract

The invention discloses a mixed cooling system multi-target parallel control method based on virtual-real migration. The method comprises the following steps: coupling characteristic analysis and cooperative control logic construction; a data-driven multi-step rapid prediction model is constructed and corrected online; and virtual-real migration feedback and multi-target parallel control optimization are carried out. Multi-step prediction is realized by constructing a multi-parameter coupled agent model and fusing an LSTM technology; and a virtual-real double-closed-loop collaborative optimization mechanism is constructed under an ACP parallel control framework, so that real-time self-adaption and multi-target rapid adjustment of control logic are realized. The invention further discloses a mixed cooling system multi-target parallel control system based on virtual-real migration. Compared with a traditional method, the technology breaks through the modeling bottleneck and realizes rapid prediction; virtual-real parallel control is initiated, and strong anti-interference and high maneuverability are achieved; carrying out online multi-target collaborative optimization; intelligent cooperative execution control; the application platform is efficiently verified; the dynamic performance is obviously improved; the comprehensive performance is excellent, and the system operation stability and the autonomous control level can be remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ship power plant cooling system control, in particular to a hybrid cooling system multi-objective parallel control method based on virtual-real migration, which covers multi-parameter coupling analysis, prediction model construction and virtual-real coupling control of complex heat exchange systems, and is suitable for intelligent control and optimized operation of hybrid condensers in ship power plants. BACKGROUND

[0002] The hybrid cooling system in the ship power plant has significant advantages in achieving efficient thermal management due to its combination of seawater and freshwater circulation, vacuum condensation and forced spraying, etc. However, the system has a complex structure and variable operating conditions, involving the coupling control of multiple key parameters such as vacuum degree, supercooling degree, spraying liquid film thickness, and condensate liquid level. Traditional first-order mechanism modeling methods based on thermodynamics and fluid mechanics cannot fully and accurately describe its dynamic behavior. The control strategy of existing cooling systems mainly relies on single-objective regulation algorithms such as PID, which is prone to control lag, long steady-state time, system oscillation, and other problems when facing frequent disturbances and multi-objective requirements, making it difficult to meet the high maneuverability and high stability requirements of ships.

[0003] The existing patent application CN117647930A is based on structural acoustic modeling, and constructs an optimization model of the cooling system through acoustic indicators such as vibration velocity and sound power level. Its control logic focuses on rule condition judgment and extraction of static optimal parameter set, mainly dealing with operation optimization under steady-state conditions. It uses an artificially set power-up / power-down control process to adjust parameters using pump-valve combinations to achieve single-time optimal control of vacuum degree and other targets. Although it covers multiple objectives (acoustic performance, economy, etc.), it lacks a systematic trade-off optimization method.

[0004] In recent years, although some research has introduced new concepts such as data-driven modeling, digital twinning, and virtual-real fusion control, there are three main shortcomings: first, the model generalization ability is weak and it is difficult to cover complex conditions; second, the prediction response is slow and it is difficult to support real-time control; third, the virtual-real data migration mechanism is not perfect and the feedback delay is significant.

[0005] Therefore, it is necessary to develop a new cooling system control method to solve the above technical problems. SUMMARY

[0006] The present application provides a parallel control method and system for a hybrid cooling system of a ship power plant, aiming to solve the problems of complex operating characteristics, strong coupling of control parameters, and frequent disturbances, which lead to stability and maneuverability control difficulties.

[0007] The application provides a hybrid cooling system multi-objective parallel control method based on virtual-real migration, a multi-parameter coupled agent model is constructed, and multi-step prediction is realized by fusing an LSTM technology; a virtual-real double closed loop collaborative optimization mechanism is constructed under an ACP parallel control framework, real-time self-adaptation of control logic and rapid multi-objective adjustment are realized.

[0008] The first object of the application provides a hybrid cooling system multi-objective parallel control method based on virtual-real migration, which comprises the following steps:

[0009] (1) Coupling characteristic analysis and collaborative control logic construction:

[0010] As shown in Figure 1 , first, the structure mechanism of the hybrid cooling system is analyzed, then a simulation model is established based on Simscape, finally, real boundary conditions are set for multi-parameter coupling characteristic analysis, and the collaborative control logic of the condensate pump and the nozzle regulating valve group is refined.

[0011] (2) Data-driven multi-step rapid prediction model construction and online correction:

[0012] The specific technical route is shown in Figure 2 , data-driven modeling includes system modeling and online correction: in the modeling stage, rapid multi-step prediction is realized through system decomposition, data preprocessing (PCA) and model training (LSTM); in the correction stage, online monitoring is carried out, and the entropy method and BP neural network are used to update parameters or switch models, so that dynamic correction is realized, and the model accuracy is ensured.

[0013] (3) Virtual-real migration feedback and multi-objective parallel control optimization:

[0014] Virtual-real migration interface: define the standardized data format and real-time communication protocol between the virtual model (agent model and its running environment) and the physical entity system, realize low-delay transmission of bidirectional data flow (state information, control instruction);

[0015] The multi-objective parallel control optimization realizes accurate mapping of the physical system by constructing a virtual "artificial system A" based on the data-driven model; performs "computational experiment C" in A; and finally executes "parallel control P".

[0016] As an embodiment of the application, the coupling characteristic analysis refers to in-depth analysis of the dynamic coupling relationship between multiple parameters such as vacuum degree, supercooling degree, spraying liquid film uniformity and condensate liquid level by using a simulation tool (such as Simscape) based on system structure (condenser, spraying device, condensate pump, hot well, valve, etc.) and thermodynamic and hydraulic principles, to provide basic support for control logic design.

[0017] As an implementation method of the application, the cooperative control logic construction refers to designing the cooperative control logic of the condensate pump rotating speed regulation and the nozzle regulation valve opening degree, clearly defining the cooperation rules of the two under different working conditions and disturbances, avoiding control conflicts, and ensuring the coordinated action of the actuators.

[0018] As an implementation method of the application, the system decomposition and data preprocessing: collect system historical operation data (temperature, pressure, flow, liquid level, valve opening, pump speed, etc.), and use principal component analysis (PCA) for feature dimension reduction and extraction to eliminate redundant information.

[0019] As an implementation method of the application, the model training: based on the processed data, a long short-term memory neural network (LSTM) is used to train and establish a proxy model of the hybrid cooling system. The model can quickly predict the key operating parameters (such as vacuum degree, supercooling degree, liquid level, etc.) at multiple time steps (such as 5-10 steps) in the future according to the current and historical state, and the prediction accuracy is verified by experiments with an average error of less than 5%.

[0020] As an implementation method of the application, the online correction: during system operation, real-time data is continuously collected, the model prediction error is evaluated using the entropy method, and the key parameters of the proxy model are dynamically updated and corrected online through back propagation neural network (BP) or other adaptive algorithms, ensuring that the model prediction result is highly accurate and synchronized with the actual state of the physical system.

[0021] The parallel control framework (ACP) of the application is the core of realizing high stability and high maneuvering control of the ship hybrid cooling system, and its specific framework is as shown in Figure 3 The framework is based on a virtual "artificial system (A)" of a data-driven model (such as LSTM) that accurately maps the physical system.

[0022] As an implementation method of the application, the "computational experiment (C)" is performed in A: the model is used to predict the key states (vacuum degree, supercooling degree, liquid level, etc.) at multiple steps in the future, and an improved multi-objective particle swarm optimization (MOPSO) algorithm is used to quickly optimize the virtual space for the current state and predicted disturbances / working conditions, and solve the optimal control strategy (pump speed and valve opening combination) that meets multiple objectives (stability, accuracy, energy consumption).

[0023] As an implementation method of the present application, the execution "parallel control (P)": the strategy of virtual space optimization is seamlessly migrated to the physical system. The real-time running data of the physical system is fed back to A, the model state is updated and the next round of C is driven, forming a closed-loop control loop of "perception-prediction-optimization-execution-feedback". This virtual-real interaction and continuous evolution of parallel mechanism is the key to cope with strong coupling, multi-disturbance and variable working condition challenges.

[0024] The second object of the present application provides a hybrid cooling system multi-objective parallel control system based on virtual-real migration. The control system provides a unified platform for the implementation of the hybrid cooling system multi-objective parallel control method based on virtual-real migration. The control system includes a parallel control software platform: a special software (such as Figure 4 ) developed based on the QT framework, which integrates model management (loading, training, updating), virtual-real communication interface, parallel control engine (including MOPSO optimizer), human-computer interaction (one-key start-stop, working condition setting, controller switching, disturbance loading), real-time data monitoring and record analysis, etc. Functional modules provide a unified platform for the implementation of the above method.

[0025] Compared with the prior art, the present application has the following beneficial technical effects:

[0026] I. Breakthrough in modeling bottleneck, realize fast prediction: adopt PCA dimension reduction and LSTM network to build high-dimensional state data driven model of ship cooling system, replace complex mechanism model. Realize multi-step prediction of future running state (vacuum degree, supercooling degree, etc. Key parameters) with high precision (measured prediction error <5%) per second, provide reliable forward-looking information for advanced control, and improve controller feedforward capability.

[0027] II. Virtual-real parallel control, strong anti-disturbance and high maneuverability: based on ACP framework, establish the mapping relationship between virtual and real systems, use system sampling data for online model correction and switching, realize seamless migration of model parameters and virtual-real parallel interaction, virtual-real synchronous feedback and dynamic self-adaptive adjustment of parameters. Experimental verification: under seawater temperature disturbance, the supercooling degree control deviation is ≤2℃, and under exhaust flow disturbance, the vacuum degree error is ≤±5kPa, which significantly improves the system stability.

[0028] III. Online multi-objective collaborative optimization: integrate improved MOPSO technology, build a control system based on artificial system, computational experiment and parallel control, and optimize in real time in parallel control. Can optimize multiple conflicting goals such as vacuum degree, supercooling degree, liquid level, etc. Improve the overall performance of the system (system steady-state performance and dynamic response speed, such as reducing energy consumption by about 8%), and find global or approximate optimal solution. The present application supports online correction of the model, improves the adaptability to variable working conditions and nonlinear characteristics, and has stronger adaptability and feedforward capability.

[0029] IV. Intelligent collaborative execution control:

[0030] The coordination logic of the condensate pump and the nozzle valve is constructed to solve the coupling problem of the actuator. It is ensured that the actuator action is coordinated under the working condition switching and disturbance, and the system is quickly responsive and stably operated.

[0031] Five, efficient verification application platform:

[0032] The parallel control software based on QT (the upper computer control software with multi-target switching and disturbance injection capacity) is developed, and the modeling, migration, optimization and testing functions are integrated. The platform successfully realizes one-key start-stop, disturbance loading and multi-controller comparison test, and provides strong support for engineering application. The control platform of the application has the functions of real-time interaction, working condition switching, disturbance injection, etc., and enhances the engineering usability.

[0033] Six, significantly improve the dynamic performance: the key indicators break through: the system stable regulation time under variable working conditions (such as start-up, load switching), which is measured to be shortened by more than 20% compared with the traditional PID controller, greatly improving the system maneuverability and response speed.

[0034] Seven, excellent comprehensive performance: through parallel control, in the complex environment of strong coupling, multiple disturbances and variable working conditions, the overall stability, disturbance resistance (satisfying the above disturbance index) and maneuverability (shortening the regulation time) of the system are significantly and quantitatively improved. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 A technical route for coupling analysis and control logic design of the cooling system is provided.

[0036] Figure 2 A modeling method for a multi-step rapid prediction model of a hybrid cooling system based on data-driven is provided.

[0037] Figure 3 A parallel control framework is provided.

[0038] Figure 4 A QT development interface is provided.

[0039] Figure 5 A data preprocessing experiment graph is provided.

[0040] Figure 6 A comparison graph of the prediction results and the actual results of the key equipment is provided. DETAILED DESCRIPTION

[0041] The application will be described in detail below in combination with the drawings and specific embodiments. The embodiments are implemented on the premise of the technical solution of the application, and detailed implementation methods and specific operation processes are given, but the protection scope of the application is not limited to the following embodiments.

[0042] 1. Model prediction verification test based on data-driven

[0043] (1) System sampling data preprocessing experiment

[0044] The purpose of this experiment: test the data preprocessing effect.

[0045] The steps of this experiment:

[0046] 1) Data sampling on the test bench, the sampling object is all sensor parameters of the cooling system test bench;

[0047] 2) The data obtained by sampling is imported into the data preprocessing algorithm, and the problems of data outliers and data missing are detected to obtain the number and specific position of the problem data, as shown in Figure 5

[0048] 3) The program automatically supplements and replaces the problem data, and checks whether the replaced data solves the data missing and data outliers.

[0049] (2) System key equipment modeling experiment

[0050] The purpose of this experiment: test the accuracy of the key equipment agent model of the hybrid cooling system.

[0051] Experiment steps:

[0052] 1) Data sampling on the cooling system test bench, the sampling object is all input and output data of a specified key equipment (such as GS pump, regulating valve).

[0053] This sampling step is divided into two stages:

[0054] The first stage is the agent model modeling data, which traverses the operation actions of the equipment;

[0055] The second stage is the agent model prediction stage, which can adjust the equipment parameters arbitrarily, collect different working conditions that need to be predicted by the agent model, and combine the two stages to obtain a complete data set.

[0056] 2) Configure the prediction parameters, training parameters and data length for modeling in the program. Use the simulated online verification form, input the collected modeling data set into the program at the real sampling interval, the program will automatically read and store the sampling data, and when the preset modeling data length is reached, the agent model starts modeling.

[0057] 3) Wait for the key equipment agent model to be completed, input the collected prediction data set into the program at the real sampling interval, and the key equipment agent model will output the comparison chart of the prediction result and the actual result of the key equipment (such as Figure 6 ​3) Wait for the completion of the modeling of the surrogate model, input the collected prediction data set into the program at the real sampling interval, and the cooling system surrogate model will output the comparison chart of the prediction results and the actual results of the test bench, and give the real-time prediction error. After the simulation is completed, save the prediction curves of each key parameter, and automatically calculate the prediction error of the prediction data and the actual data recorded by the program, and record it in the table.

[0058] (3) System data-driven model online modeling and key parameter single-step prediction experiment

[0059] Experimental purpose: Test the single-step prediction accuracy of the proxy model of the hybrid cooling system

[0060] Experimental steps:

[0061] 1) Data sampling on the test bench, the sampling object is all sensor parameters of the cooling system test bench.

[0062] This sampling step is divided into two stages:

[0063] The first stage is the data for modeling the surrogate model, which needs to switch the cooling system operating conditions from low to high in turn, and stay at each operating condition for more than 60s, and then switch from high to low in turn;

[0064] The second stage is the prediction stage of the surrogate model. This stage can arbitrarily change the operating conditions, and collect different operating conditions that need to be predicted by the surrogate model. Combine the two stages to obtain a complete data set.

[0065] 2) Configure the prediction parameters, training parameters, and select the data length for modeling in the program. Use the simulated online verification form, input the collected modeling data set into the program at the real sampling interval, the program will automatically read and store the sampling data, after reaching the preset modeling data length, the surrogate model starts modeling.

[0066] 3) Wait for the completion of the modeling of the surrogate model, input the collected prediction data set into the program at the real sampling interval, and the cooling system surrogate model will output the comparison chart of the prediction results and the actual results of the test bench, and give the real-time prediction error. After the simulation is completed, save the prediction curves of each key parameter, and automatically calculate the prediction error of the prediction data and the actual data recorded by the program, and record it in the table.

[0067] (4) System surrogate model online modeling and key parameter multi-step prediction experiment

[0068] The purpose of this experiment is to test the multi-step prediction accuracy of the proxy model of the hybrid cooling system

[0069] Experimental steps:

[0070] 1) Data sampling on the test bench.

[0071] This sampling step is divided into two stages:

[0072] The first stage is to model the data of the proxy model, which needs to switch the working condition of the hybrid cooling system from low to high, and stay at each working condition for more than 60s, and then switch from high to low;

[0073] The second stage is the prediction stage of the proxy model, which can change the working condition at will, collect different working conditions that need to be predicted by the proxy model, and combine the two stages to obtain a complete data set.

[0074] 2) Configure the prediction parameters, training parameters, select the data length for modeling, and select the prediction time length in the program. Use the simulated online verification form, input the collected modeling data set at the real sampling interval into the program, the program will automatically read and store the sampling data, after reaching the preset modeling data length, the proxy model starts modeling.

[0075] 3) Wait for the proxy model to complete modeling, input the collected prediction data set at the real sampling interval into the program, the cooling system proxy model will output the comparison chart of the prediction results and the actual results of the test bench, and give the real-time prediction error, after the simulation is completed, save the prediction curves of each key parameter, automatically calculate the prediction error by the automatically recorded prediction data and actual data, and record it in the table.

[0076] 2、Multi-objective parallel intelligent control test based on virtual-real migration

[0077] (1) One-key start-stop experiment of hybrid cooling system

[0078] Determine the starting working condition and clearly define the temperature, pressure and other boundary conditions of the hybrid cooling system under the starting working condition.

[0079] 1) Prepare the test bench environment of the cooling system's power electrical system, data acquisition and display control system, QT host computer, etc.

[0080] 2) Ensure that all pumps and valves in the cooling system are in the closed state before the system starts;

[0081] 3) Use the Qt software control on the display and control console to perform one-key start-stop experiments according to the designed control logic;

[0082] 4) During the system startup process, pay attention to observe the supercooling degree, vacuum degree, liquid level, pressure, valve opening degree and other key information of the cooling system, and save the numerical values in the corresponding table.

[0083] (2) Steady-state working condition control test experiment of system with disturbance

[0084] This experiment will use seawater temperature disturbance / freshwater temperature / exhaust flow disturbance, and other identifiable upstream and downstream boundary disturbances to the cooling system, use traditional PID controller and virtual-real migration-based multi-objective parallel intelligent controller for steady-state operating condition control, and make comparison.

[0085] 1) Prepare the power electrical system of the cooling system, data acquisition and display control system, QT host computer, etc.

[0086] 2) Use the one-key start-stop function to run the system to steady-state condition 1;

[0087] 3) Use the QT host computer to control the system to load seawater / freshwater / exhaust flow / other upstream and downstream boundary condition disturbances;

[0088] 4) Use QT to load the PID controller, observe the changes in key parameters of the system and save the curves;

[0089] 5) Use QT to load the multi-objective parallel controller based on virtual-real migration, observe the changes in key parameters of the system and save the curves;

[0090] 6) Switch to steady-state conditions 2, 3, 4, etc., repeat steps 3) - 5);

[0091] 7) Summarize the control parameter curves of the traditional PID controller and the multi-objective parallel controller based on virtual-real migration under each steady-state condition with disturbances, and summarize the indicators: under exhaust flow disturbance, the vacuum degree control error is ≤±5 kPa, and under seawater temperature disturbance, the supercooling degree is ≤2℃.

[0092] (3) System multi-condition switching stability and maneuverability control test experiment

[0093] The purpose of this experiment is to test the difference in control effect of the traditional PID controller and the multi-objective parallel controller based on virtual-real migration on system stability and maneuverability when the system is running under variable conditions.

[0094] 1) Prepare the power electrical system of the cooling system, data acquisition and display control system, QT host computer, etc.

[0095] 2) Use the one-key start-stop function to run the system to steady-state condition 1;

[0096] 3) Use QT to load the traditional PID controller, and set the step-by-step upgrade and downgrade condition function, the cooling system will switch according to the conditions 1-conditions 2-...-conditions n-conditions n-1-...-conditions 1, and record the key parameter data and curves of the system;

[0097] 4) Use QT to load the multi-target parallel controller based on virtual-real migration, and set the step-by-step rising and falling working condition function. The cooling system will switch according to the working condition 1-working condition 2-…-working condition n-working condition n-1-…-working condition 1 step by step, and the key parameter data and curves of the system are recorded;

[0098] 5) Summarize the curves obtained in steps 3) and 4), and summarize the indicators: the stable adjustment time of key characteristic parameters during variable working condition operation (compared with the PID controller) is shortened by 20%.

[0099] The above shows and describes the basic principles and main features of the present application and the advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above embodiments, and the above embodiments and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A multi-objective parallel control method for a hybrid cooling system based on virtual-to-physical migration, characterized in that, Comprise: (1) coupling characteristics analysis and collaborative control logic construction: first analyze the structure mechanism of the hybrid cooling system, then establish a simulation model based on Simscape, finally set the real boundary conditions for multi-parameter coupling characteristic analysis, and extract the condensate pump and nozzle regulating valve group's collaborative control logic; (2) data-driven multi-step fast prediction model construction and online correction: data-driven modeling includes system modeling and online correction: modeling stage realizes fast multi-step prediction through system decomposition, data preprocessing and model training; the correction stage carries out online monitoring, and uses entropy method, BP neural network to update parameters or switch models, realizes dynamic correction, ensures the model precision; (3) virtual-real migration feedback and multi-objective parallel control optimization: the virtual migration feedback realizes low-delay transmission of bidirectional data flow by defining standardized data format and real-time communication protocol between virtual model and physical entity system; the multi-objective parallel control optimization realizes accurate mapping of physical system by constructing a virtual "artificial system A" based on data-driven model; "computational experiment C" is carried out in A; finally, "parallel control P" is executed.

2. The multi-objective parallel control method of the hybrid cooling system based on virtual-to-physical migration according to claim 1, wherein The coupling characteristic analysis refers to the dynamic coupling relationship among vacuum degree, supercooling degree, spraying liquid film uniformity and condensate liquid level based on the structure and thermodynamic and hydraulic principles of condenser, spraying device, condensate pump, hot well and valve system, and the in-depth analysis is carried out by using simulation tool Simscape.

3. The method of claim 1, wherein, The collaborative control logic construction refers to the design of the collaborative control logic of condensate pump speed regulation and nozzle regulating valve opening, which clearly defines the coordination rules of the two under different working conditions and disturbances, avoids control conflicts, and ensures the coordinated action of the actuator.

4. The method of claim 1, wherein, The system decomposition and data preprocessing refer to collecting system historical operation data, using principal component analysis for feature dimension reduction and extraction, and eliminating redundant information.

5. The method of claim 1, wherein, The model training refers to training and establishing a proxy model of the hybrid cooling system based on the processed data using long short-term memory neural network; the model can quickly predict the key operating parameters of multiple time steps in the future according to the current and historical state, with an average prediction error of less than 5%.

6. The method of claim 1, wherein, The online correction refers to continuously collecting real-time data during system operation, evaluating the model prediction error using entropy method, and dynamically updating and correcting the key parameters of the proxy model using back propagation neural network BP or other adaptive algorithms, to ensure that the model prediction result is highly accurate and synchronized with the actual state of the physical system.

7. The method of claim 1, wherein, The "computational experiment C" in the artificial system A refers to predicting the future multi-step key state vacuum degree, supercooling degree and liquid level using the model, and combining the improved multi-objective particle swarm optimization MOPSO algorithm to quickly optimize and solve the optimal control strategy that meets the multi-objective stability, precision and energy consumption in the virtual space for the current state and predicted disturbance / working condition.

8. The method of claim 1, wherein, The "parallel control P" refers to seamlessly migrating the optimized strategy in the virtual space to the physical system, synchronously feeding back the real-time operation data of the physical system to A, updating the model state and driving the next round of C, forming a closed-loop control loop of "perception-prediction-optimization-execution-feedback".

9. A multi-objective parallel control system for a hybrid cooling system based on virtual-to-physical migration, characterized in that, The control system provides a unified platform for the implementation of the multi-target parallel control method of the virtual-real migration-based hybrid cooling system according to claim 1, and the control system comprises a parallel control software platform, a special software developed based on a QT framework, integrated model management, a virtual-real communication interface, a parallel control engine, human-computer interaction, a real-time data monitoring and recording analysis function module.

Images