Mechanism constraint dual-time-domain self-evolution optimization control method for air compressor

Abstract

The application belongs to the technical field of fuel cells, and particularly relates to a mechanism constraint double-time-domain self-evolution optimization control method for an air compressor, obtaining multi-working-condition operation data, establishing a coupling simulation model of the compressor, the motor, the air pipeline and the load end, generating a sample data set and constructing a mechanism constraint set, determining a control action feasible region, fusing current state, historical state sequence and mechanism constraint information, extracting mechanism consistent state features, establishing a control model, generating a candidate control action and projecting it to the feasible region to obtain a control instruction, constructing a comprehensive evaluation model combining a short-time domain and a long-time domain based on the control instruction, and updating and adaptively correcting the control model in combination with long-term operation data, outputting a final control instruction and sending it to a high-speed driving motor and variable-geometry adjusting mechanisms at all levels to control the operation of the multi-stage centrifugal air compressor. The application improves the dynamic response performance, operation stability, gas supply matching capability and long-term robustness of the air compressor under wide working conditions.

Description

Technical Field

[0001] This application belongs to the field of fuel cell technology, and specifically relates to a mechanism-constrained dual-time-domain self-evolutionary optimization control method for air compressors. Background Technology

[0002] As proton exchange membrane fuel cell systems evolve towards higher power density, higher dynamic response, and wider operating conditions, the air supply system's impact on stack output performance, oxygen excess coefficient stability, and overall system efficiency has become increasingly prominent, making it one of the key subsystems determining the operational quality of fuel cell systems. As a core component of the air supply system, centrifugal air compressors, due to their compact structure, strong pressurization capacity, and suitability for high-speed operation, have become an important technological approach in the field of fuel cell air supply. Especially in medium- and high-power fuel cell systems, multi-stage centrifugal air compressors combined with variable geometry adjustment mechanisms can, to a certain extent, balance the requirements for higher pressure ratios and wider operating conditions, thus demonstrating promising engineering application prospects.

[0003] Currently, research on centrifugal air compressors for fuel cells mainly focuses on compressor structural design optimization, development of variable geometry adjustment mechanisms, and improvement of control methods. At the control level, existing technologies generally employ PID control, fuzzy control, gain-dispatch control, and empirical rule-based control strategies. Some studies have also introduced intelligent control methods such as model predictive control and reinforcement learning to improve the compressor's response performance and gas supply matching capability under varying operating conditions. These studies have, to some extent, driven the development of fuel cell air compressors from fixed-parameter control to intelligent regulation.

[0004] However, fuel cell multi-stage centrifugal air compressors typically exhibit significant nonlinearity, strong coupling, and multi-constraint characteristics during rapid load changes and wide operating conditions, involving the coordinated regulation of pressure, flow rate, speed, and variable geometric components. Traditional control methods often struggle to achieve multivariate coordinated optimization when faced with these complex dynamic coupling relationships, easily leading to problems such as response lag, regulation oscillation, large air supply deviations, and decreased operating efficiency. Meanwhile, many existing intelligent control methods are simply direct transplants of general algorithms to this scenario, lacking specific design considerations for the surge boundary, aerodynamic stability, actuator dynamic constraints, and fuel cell air demand mechanisms of centrifugal air compressors. This results in shortcomings in physical consistency, engineering interpretability, and real-time implementability.

[0005] Furthermore, most existing control methods focus on performance optimization at a single time scale, typically only considering transient tracking capability or local steady-state efficiency, making it difficult to simultaneously address short-term dynamic response, long-term operating efficiency, and system stability margin maintenance. When the operating environment changes, load patterns shift, or component performance drifts, the online adaptability and long-term robustness of existing control strategies are relatively insufficient, easily leading to a decline in control performance and even causing the system operating point to approach the instability boundary.

[0006] For example, patent document CN201910373642.9 discloses a two-layer predictive control method for a full-power fuel cell air compressor. This method uses an upper-layer predictor to predict vehicle speed in real time, calculates the power required by the fuel cell vehicle under the corresponding operating conditions, and further combines this with a fuel cell cathode flow model to determine the reference air flow rate required for the air compressor output. The lower-layer predictive controller then predicts the air compressor output air flow rate based on the reference flow rate and solves for the control voltage to achieve control of the air compressor output flow rate. This scheme embodies the hierarchical predictive control approach of upper-layer demand prediction and lower-layer execution control. However, this method mainly focuses on the mapping and calculation between vehicle speed, power, and target flow rate. It lacks dedicated modeling and deep embedding of the aerodynamic stability, variable geometric linkage mechanism, and dynamic boundary of the actuator within the multi-stage centrifugal air compressor. Furthermore, it emphasizes the hierarchical transmission relationship between reference quantity generation and execution control, and has not yet coordinated short-term response, long-term efficiency, and stability margin under a unified decision-making framework. Therefore, in wide-condition, highly nonlinear scenarios, it still suffers from insufficient constraint integration and limited control depth.

[0007] For example, patent document CN201911418431.9 discloses a surge control method and system for an air compressor. This method calculates the desired flow rate and pressure based on the power demand command and ideal power generation performance of the fuel cell stack, and applies non-surge limiting processing to both. These values ​​are then input into two PI controllers to obtain the air compressor angular velocity and butterfly valve angle, thereby achieving dual-loop control of outlet flow rate and pressure and improving the surge prevention effect. The characteristic of this scheme is that it combines surge limiting with dual closed-loop control of flow rate and pressure. However, this method is essentially still an external protection approach that generates control objectives first and then applies limiting corrections. That is, stability boundary constraints are not an endogenous component of the control decision generation mechanism, but mainly rely on ex-post constraints to achieve safe control. Such methods struggle to integrate surge margin, efficiency optimization, air supply matching, and actuator smoothness into a single decision framework. Therefore, there is still room for improvement in terms of adaptability to complex operating conditions, long-term performance coordination, and the interpretability of integrated control.

[0008] In summary, existing control technologies for multi-stage centrifugal air compressors used in fuel cells generally suffer from the following problems: First, they struggle to adapt to complex dynamic air supply demands under conditions of rapid load changes, wide operating condition switching, and multi-constraint coupling; second, they lack sufficient integration of surge boundaries, aerodynamic stability, actuator dynamic constraints, and air demand mechanisms, often relying on external corrections for control commands; and third, they struggle to simultaneously achieve short-term dynamic response, long-term operating efficiency, and stability margin maintenance, exhibiting insufficient online adaptability and long-term robustness under environmental changes, load mode shifts, and component performance drift. Therefore, there is an urgent need to propose an intelligent control method for multi-stage centrifugal air compressors used in fuel cells to achieve mechanistic constraint embedding, dynamic safety adjustment, short-term and long-term collaborative optimization, and adaptive control under complex operating conditions. Summary of the Invention

[0009] To address the challenges of existing fuel cell multi-stage centrifugal air compressor control methods in simultaneously achieving mechanistic constraint embedding, short-time dynamic response, long-term operating efficiency, and adaptive capability under complex operating conditions, this application presents a mechanistic constraint dual-time-domain self-evolutionary optimization control method for air compressors. By constructing an integrated control framework that includes mechanistic consistency state coding, control action feasible domain constraint projection, dual-time-domain comprehensive performance evaluation, and operating condition prototype self-evolutionary adjustment, this method achieves coordinated optimization control of the high-speed drive motor and each stage of variable geometric adjustment mechanism. This improves the air compressor's response speed, operational stability, air supply matching capability, and long-term robustness under a wide range of operating conditions.

[0010] This application provides a mechanism-constrained dual-time-domain self-evolutionary optimization control method for air compressors, the method comprising:

[0011] Step 1: Acquire multi-condition operating data representing the compressor operating status, motor operating status, air pipeline operating status, and variable geometry adjustment status;

[0012] Step 2: Based on the multi-condition operating data, establish and calibrate a joint simulation model of the compressor, motor, air piping and load coupling, and generate a sample dataset;

[0013] Step 3: Construct a set of mechanistic constraints based on the sample dataset, and use the motor speed and the opening degree of each stage of the variable geometric adjustment mechanism as control variables to determine the feasible domain of the control action according to the set of mechanistic constraints;

[0014] Step 4: Preprocess the sample dataset, and based on the preprocessed data, fuse and encode the current running state, historical state sequence and corresponding mechanism constraint information to obtain mechanism consistency state features;

[0015] Step 5: Establish a control model based on the consistency state characteristics of the mechanism, and generate candidate control actions based on the control model, including the motor speed adjustment amount and the opening adjustment amount of each stage of variable geometric adjustment mechanism. Project the candidate control actions into the feasible domain of the control actions to obtain the current cycle control command.

[0016] Step 6: Construct a comprehensive evaluation model combining short-time and long-time domains based on the current cycle control command, and update the control model according to the comprehensive evaluation results. During the online operation of the controller, extract typical operating condition features based on long-term operating data, and adaptively correct the local parameters of the control model according to the degree of deviation between the current operating state and the typical operating condition features.

[0017] Step 7: Output the final control command based on the corrected control model, and send the final control command to the high-speed drive motor and each stage of variable geometry adjustment mechanism to control the operation of the multi-stage centrifugal air compressor.

[0018] In a preferred implementation, step 3 further includes:

[0019] Step 3.1: Based on the sample dataset obtained in Step 2, the samples are classified according to the motor speed range, variable geometric opening range, load change rate range and air excess coefficient range. In each classification range, the pressure build-up capability, flow regulation capability, surge margin, actuator dynamic boundary and air supply matching characteristics are extracted and characterized to establish the mechanism constraint characterization relationship under different operating conditions.

[0020] Step 3.2: Based on the mechanism constraint characterization system constructed in Step 3.1, using the motor speed regulation amount and the opening regulation amount of each stage of variable geometric adjustment mechanism as control variables, establish the control action feasible domain that satisfies surge safety margin constraints, actuator dynamic constraints, air excess coefficient constraints and compressor efficiency constraints according to the mechanism constraint characterization system.

[0021] In the preferred implementation, further, in step 3.2, for the current time k, a feasible domain for control actions is established, which is jointly defined by the boundaries of multiple mechanisms. :

[0022] ;

[0023] in, The surge margin function; This represents the current state of the air compressor system. The control action corresponding to the current moment; In the current state and control actions Surge margin function value under the given conditions; For minimum safety margin; This is a function of compressor efficiency; This represents the motor speed adjustment amplitude within the current control cycle. This represents the maximum allowable adjustment range of the motor speed. This refers to the adjustment amplitude of the inlet guide vane opening within the current control cycle. This refers to the maximum allowable adjustment range of the inlet guide vane opening; This represents the amplitude of diffuser opening adjustment within the current control cycle. This represents the maximum allowable adjustment range of the diffuser opening. In the current state and control actions The function value of the excess air coefficient under action; and These are the lower and upper limits of the excess air coefficient, respectively. Current state and control actions The compressor efficiency function value under the action; The minimum allowable efficiency threshold; It should at least characterize one or more of the following: compressor inlet pressure, outlet pressure, inlet temperature, outlet temperature, air mass flow rate, actual motor speed, actual inlet guide vane opening, actual diffuser opening, actuator displacement, load current, load change rate, and air excess coefficient.

[0024] In the preferred implementation, further, in step 4, the mechanism consistency feature vector is:

[0025] ;

[0026] in, This is the mechanism consistency feature vector at the current time k; Mechanism-consistent state encoding mapping; The parameters corresponding to the mechanism consistency state encoding mapping; Let L and L represent the state variables from time k-L+1 to time k, respectively, where L is the historical time sequence length used for state encoding. This refers to the mechanism constraint information corresponding to the current moment.

[0027] In a preferred implementation, step 5 further includes:

[0028] Step 5.1: Based on the mechanism consistency state characteristics obtained in Step 4, candidate control actions containing motor speed regulation and opening regulation of each level of variable geometric adjustment mechanism are generated using the candidate action generation mapping;

[0029] Candidate control actions are:

[0030] ;

[0031] in, Generate mappings for candidate actions; This indicates the initial control action that has not been modified by constraints;

[0032] Step 5.2: Input the candidate control action into the feasible region of control action established in step 3, and perform projection correction on the candidate control action based on the feasible region projection operator to obtain the corrected control action that satisfies the mechanism constraint;

[0033] The revised control action is as follows:

[0034] ;

[0035] in, For feasible region projection operators; The control action is after projection correction.

[0036] In a preferred implementation, step 6 further includes:

[0037] Step 6.1: Based on the final control command, construct short-time domain performance indicators within the short-time domain evaluation window to characterize pressure response speed, flow tracking error, and the effect of regulating oscillation suppression;

[0038] Step 6.2: Based on the final control command, construct long-term performance indicators within the long-term evaluation window to characterize compressor efficiency, power consumption level, stability margin maintenance capability, and long-term smoothness of actuator operation;

[0039] Step 6.3: Construct a comprehensive performance index based on the short-time domain performance index and the long-time domain performance index, and update the control model based on the comprehensive performance index;

[0040] Step 6.4: During the online operation of the controller, extract the operating condition prototype based on long-term operating data, and perform limited-amplitude adaptive correction of the local parameters of the control model according to the degree of deviation between the current operating state and the operating condition prototype.

[0041] In the preferred implementation, further, in step 6.4, the comprehensive performance index is:

[0042]

[0043]

[0044] ;

[0045] in, For short-time domain performance metrics; This refers to the short time-domain evaluation window length; For pressure error; For flow rate error; u represents the change in control action between adjacent control cycles; , , These are the weighting coefficients for the pressure error term, flow error term, and control action change term, respectively. For long-term performance metrics; The length of the long-term evaluation window; For compressor efficiency; This is a stability margin penalty term; P is the power consumption indicator. , , These are the weighting coefficients for the efficiency term, stability margin term, and power consumption term, respectively. The coefficients are the dual-time-domain balance coefficients; k is the current control time. This is the state vector at the current moment; This refers to the revised control action.

[0046] In a preferred implementation, step 2 further includes:

[0047] Step 2.1: Based on the operational data collected in Step 1, establish a joint simulation model including the compressor sub-model, motor sub-model, actuator sub-model, air pipeline sub-model, buffer volume sub-model, and fuel cell stack load sub-model, and realize parameter transfer and boundary coupling between the sub-models through the data interface;

[0048] Step 2.2: Assign and calibrate the compressor characteristic parameters, motor dynamic parameters, actuator response parameters, air pipeline parameters, buffer volume parameters, and fuel cell load requirement parameters of the co-simulation model based on the operating data, so that the steady-state output characteristics and dynamic response characteristics of the co-simulation model match the experimental data;

[0049] Step 2.3: Based on the calibrated co-simulation model, simulation calculations are performed under different load change conditions, different air demand conditions, different motor speed conditions, and different variable geometric adjustment states. Combined with the experimental data obtained in Step 1, a sample dataset covering steady-state conditions, dynamic transition conditions, and boundary conditions is formed.

[0050] In a preferred implementation, further, in step 1, the operating data includes at least pressure, temperature, and flow data characterizing the compressor's operating state; speed, voltage, current, and power data characterizing the motor's operating state; air demand, back pressure, and air supply matching error data characterizing the air supply state; and opening degree, displacement, and response delay data characterizing the variable geometry adjustment state.

[0051] In a preferred implementation, further, in step 7, the final control command includes at least a motor speed adjustment command and an opening adjustment command for each level of variable geometry adjustment mechanism. The motor speed adjustment command is used to determine the target speed or target speed adjustment amount of the high-speed drive motor in the current control cycle, and the opening adjustment command for each level of variable geometry adjustment mechanism is used to determine the target opening or target opening adjustment amount of the corresponding compression stage inlet guide vane, diffuser, or other variable geometry component.

[0052] The beneficial effects of this application are:

[0053] First, the mechanism-constrained dual-time-domain self-evolutionary optimization control method for air compressors proposed in this application establishes and calibrates a joint simulation model coupling the compressor, motor, air pipeline, and load end based on multi-condition operating data. It constructs a set of mechanism constraints and determines the feasible domain of control actions with motor speed and the opening degree of each stage of variable geometric adjustment mechanism as control variables. Then, it fuses and encodes the current operating state, historical state sequence, and corresponding mechanism constraint information to obtain mechanism-consistent state characteristics. Based on these characteristics, candidate control actions are generated and directly projected into the feasible domain of control actions. This allows constraints such as surge boundary, aerodynamic stability, actuator dynamic boundary, air supply matching, and efficiency boundary to be intrinsically embedded in the control decision-making process, eliminating reliance on ex-post constraints. The amplitude correction enables coordinated optimization control of motor speed and variable geometric adjustment mechanisms at all levels under rapid load changes and wide operating conditions. Furthermore, this application constructs a comprehensive evaluation model combining short-term and long-term time domains, and extracts typical operating condition characteristics from long-term operating data to adaptively correct local parameters of the control model. This balances short-term dynamic response speed, long-term operating efficiency, and stability margin maintenance, improving the system's online adaptability and long-term robustness to environmental changes, load mode migration, and component performance drift. Ultimately, this effectively improves the air supply matching accuracy, dynamic response performance, operational stability, energy efficiency, engineering interpretability, and real-time implementability of the fuel cell air supply system.

[0054] Second, in the preferred implementation, step 3 of this application classifies the multi-condition operating data and extracts the pressure build-up capability, flow regulation capability, surge margin, actuator dynamic boundary, and air supply matching characteristics to establish the mechanism constraint characterization relationship under different operating conditions. Furthermore, using motor speed and the opening degree of each stage of the variable geometry adjustment mechanism as control variables, a feasible domain for control actions is constructed, jointly defined by surge safety margin constraints, motor dynamic boundary constraints, actuator dynamic boundary constraints, air excess coefficient constraints, and compressor efficiency constraints. This enables the control command generation process to possess mechanism boundary perception and intrinsic constraint satisfaction capabilities from the source. Simultaneously, by combining the fusion encoding of the current operating state, historical state sequences, and corresponding mechanism constraint information, as well as the comprehensive evaluation combining short-term and long-term domains and the adaptive correction of local parameters based on the deviation degree of typical operating conditions, the coordinated optimization adjustment of the motor and each stage of the variable geometry mechanism can be achieved under conditions of rapid load changes, wide operating condition switching, and component performance drift in the multi-stage centrifugal air compressor of the fuel cell.

[0055] Third, in the preferred implementation, step 4 of this application constructs a mechanism consistency feature vector by inputting the current operating state, historical state sequence, and mechanism constraint information corresponding to the current moment into the mechanism consistency state coding mapping. This allows the originally dispersed multi-source information such as pressure, flow rate, speed, variable geometric opening, and boundary constraints to be uniformly represented as state features that combine temporality, correlation, and physical consistency. This not only avoids the problems of one-sided state expression, noise sensitivity, and insufficient dynamic trend characterization caused by traditional methods that rely solely on single-moment measurements, but also improves the control model's ability to identify the system evolution direction, boundary approximation degree, and multivariable coupling relationship. This is beneficial for improving the response accuracy and control stability of fuel cell multi-stage centrifugal air compressors under rapid load changes, wide operating condition switching, and complex constraint conditions.

[0056] Fourth, in the preferred implementation, step 5 of this application first generates candidate control actions that can characterize the current pressure establishment demand, flow regulation demand, and load change trend based on the mechanism consistency state characteristics. Then, the candidate control actions are projected onto the control action feasible domain jointly defined by mechanism constraints such as surge safety margin, actuator dynamic boundary, air excess coefficient, and compressor efficiency. This makes the control command generation process both demand-oriented and constraint-satisfying. On the one hand, it can more completely retain the pressurization, flow increase, and coordinated regulation intentions reflected in the candidate control actions, avoiding excessive weakening of the control target by the traditional ex-post limiting method. On the other hand, it can quickly complete the constraint correction without deviating from the current regulation direction, reducing the computational burden of directly solving the control command under complex constraint conditions.

[0057] Fifth, in the preferred implementation, step 6 of this application constructs short-time domain performance indicators characterizing pressure response speed, flow tracking error, and regulation oscillation suppression effect based on the final control command, as well as long-time domain performance indicators characterizing compressor efficiency, power consumption level, stability margin maintenance capability, and long-term smoothness of actuator operation. Furthermore, a unified comprehensive performance index is formed to update the control model, enabling the control system to move beyond local optimization at a single time scale. Instead, it ensures timely transient regulation while considering long-term operational economy, stability, and smoothness, effectively avoiding efficiency degradation, increased power consumption, or deteriorated boundary margins caused by solely pursuing rapid response. Simultaneously, by extracting operating condition prototypes from long-term operating data during the controller's online operation and performing limited-amplitude adaptive corrections to the local parameters of the control model based on the deviation between the current operating state and the operating condition prototype, the long-term robustness of the fuel cell multi-stage centrifugal air compressor under wide operating conditions is enhanced.

[0058] Sixth, in the preferred implementation, steps 1 and 2 of this application establish a joint simulation model based on the operational data collected in step 1, including a compressor sub-model, a motor sub-model, an actuator sub-model, an air pipeline sub-model, a buffer volume sub-model, and a fuel cell stack load sub-model. Parameter transfer and boundary coupling between the sub-models are achieved through a data interface. Then, experimental data is used to assign and calibrate the compressor characteristic parameters, motor dynamic parameters, actuator response parameters, air pipeline parameters, buffer volume parameters, and fuel cell stack load requirement parameters. This ensures the model possesses both high steady-state consistency and dynamic response realism, enabling high-reliability simulation calculations under different load variations, air demand conditions, motor speed conditions, and variable geometric adjustment states. A sample dataset covering steady-state, dynamic transition, and boundary conditions is generated. This not only reduces the cost and risk of relying solely on actual machine tests for sample collection and improves the completeness of data acquisition under complex boundary conditions, but also provides more comprehensive data for subsequent mechanism constraint construction, control action feasible domain determination, and control model training.

[0059] Seventh, in the preferred implementation, step 7 of this application explicitly refines the final control command into a motor speed adjustment command and an opening degree adjustment command for each level of variable geometry adjustment mechanism, which respectively correspond to the target control quantities of the high-speed drive motor and each compression stage inlet guide vane, diffuser or other variable geometry components. This enables the output results of the control model to be directly mapped to the specific execution object and specific execution parameters, and enables the motor speed adjustment and the opening degree adjustment of each level of variable geometry components to be executed collaboratively under a unified control objective. Attached Figure Description

[0060] Fig. 1This is a flowchart of the mechanism-constrained dual-time-domain self-evolutionary optimization control method for air compressors in an embodiment of the present invention;

[0061] Fig. 2 This is a flowchart of the mechanism-constrained dual-time-domain self-evolutionary strategy optimization method in the mechanism-constrained dual-time-domain self-evolutionary optimization control method for air compressors in this embodiment of the invention. Detailed Implementation

[0062] To enable those skilled in the art to better understand the technical solutions of this application, the following will provide a more detailed description of this application in conjunction with the accompanying drawings and embodiments.

[0063] The directional terms such as above, below, left, right, front, and back used in this application are based on the positional relationships shown in the attached drawings. Different attached drawings may result in different positional relationships, therefore they should not be interpreted as limitations on the scope of protection.

[0064] In this application, the terms "installation," "connection," "interlocking," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, an integral connection, a mechanical connection, an electrical connection, or a connection that allows communication between components. They can also refer to a direct connection or an indirect connection through an intermediate medium. They can refer to the internal connection of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in this application can be understood according to the specific circumstances.

[0065] This invention describes a mechanism-constrained dual-time-domain self-evolutionary optimization control method for air compressors, aiming to solve the problem that existing fuel cell multi-stage centrifugal air compressor control technologies are unable to adapt to complex dynamic air supply demands under conditions of rapid load changes, wide operating condition switching, and multi-constraint coupling. In particular, existing technologies generally suffer from insufficient integration of surge boundaries, aerodynamic stability, actuator dynamic constraints, and air demand mechanisms. Control commands often rely on external limiting or post-event correction, making it difficult to simultaneously consider short-term dynamic response, long-term operating efficiency, and system stability margin. Furthermore, they lack online adaptability and long-term robustness under environmental changes, load mode migration, and component performance drift. This method addresses the operation of multi-stage centrifugal air compressors in fuel cell air supply systems. It embeds mechanistic constraints such as surge safety boundaries, air supply matching boundaries, actuator dynamic boundaries, and efficiency boundaries into the control decision generation process. By constructing a mechanistic consistency state representation mechanism, a control action feasible domain projection correction mechanism, and a comprehensive evaluation and self-evolutionary update mechanism combining short and long time domains, it implements coordinated optimization control of the high-speed drive motor and each stage of variable geometry adjustment mechanisms. This achieves a comprehensive improvement in pressure build-up speed, flow tracking capability, aerodynamic stability, operating efficiency, and long-term robustness under complex operating conditions. Ultimately, it enhances the dynamic safety adjustment capability, control interpretability, and engineering feasibility of multi-stage centrifugal air compressors in fuel cell systems under a wide range of operating conditions.

[0066] As per the instruction manual Figs. 1-2 This invention provides a mechanism-constrained dual-time-domain self-evolutionary optimization control method for air compressors. This method is applied to a multi-stage centrifugal air compressor in a fuel cell air supply system. The multi-stage centrifugal air compressor is equipped with a high-speed drive motor and various stages of variable geometric adjustment mechanisms. The method includes:

[0067] Step 1: Acquire multi-condition operating data that characterizes the compressor operating status, motor operating status, air pipeline operating status, and variable geometry adjustment status.

[0068] Specifically, an air supply system test platform is constructed, which includes at least: a multi-stage centrifugal air compressor body, a high-speed drive motor, a variable frequency drive, various levels of variable geometry adjustment mechanisms, an intake pipe, an exhaust pipe, a buffer cavity, a throttling loading unit, a fuel cell stack simulation load unit, a central controller, and a multi-parameter synchronous acquisition system.

[0069] The system comprises a multi-stage centrifugal air compressor body for air compression; a high-speed drive motor for rotating the compressor; a frequency converter electrically connected to the motor for adjusting speed; variable geometry adjustment mechanisms at each stage for adjusting flow path openings; intake and exhaust pipes at the compressor inlet and outlet; a buffer chamber in the exhaust pipe to simulate volumetric effects in a real air supply system; a throttling loading unit on the exhaust side for adjusting back pressure and flow conditions; a fuel cell stack simulation load unit to simulate changes in air demand at the fuel cell cathode; a central controller to coordinate the operation of the frequency converter, variable geometry adjustment mechanisms, and test units; and a multi-parameter synchronous acquisition system for synchronously acquiring, recording, and storing key operating parameters during the test.

[0070] In this application, the multi-parameter synchronous acquisition system is used to acquire at least the following parameters in real time: compressor inlet pressure, outlet pressure, inlet temperature, outlet temperature, air mass flow rate, motor speed, inlet guide vane opening at each stage, diffuser opening at each stage, and actuator displacement. Depending on experimental needs, auxiliary parameters such as motor voltage, motor current, drive power, throttle valve opening, ambient temperature, and ambient pressure can also be acquired synchronously. The acquired data from each channel are preferably synchronized using a unified clock to ensure the timing correspondence between different physical quantities, facilitating subsequent analysis of the coupled dynamic process between the air compressor, actuator, and load changes.

[0071] After the test platform was built, the sensors, actuators, and control units were integrated and calibrated. This included: zero-point and range calibration of the pressure sensor, calibration of the temperature sensor, calibration of the flow meter, calibration of the motor speed measurement, and zero-point and stroke calibration of each stage of the variable geometry adjustment mechanism. After calibration, the transmission status between the high-speed drive motor and the air compressor main shaft was checked, confirming the sealing of the intake and exhaust pipes, the connection status of the buffer cavity, and the normal functioning of the throttling loading unit. The central controller then established the test procedure.

[0072] Multi-condition operating data refers to a set of time-series data collected under different load change conditions, different air demand conditions, different motor speed conditions, and different variable geometry adjustment states, which can characterize the operating characteristics of the air supply system. Among them, different load change conditions can include step load increase, step load decrease, ramp load increase, ramp load decrease, and cyclic load change; different air demand conditions can be set by adjusting the demand level and change rate of the fuel cell stack simulated load unit to correspond to different air excess coefficient requirements; different motor speed conditions can include low-speed, medium-speed, and high-speed operating ranges; different variable geometry adjustment states can include changes in the opening degree of each stage inlet guide vane, changes in the opening degree of each stage diffuser, and different combinations of inter-stage opening degrees.

[0073] Preferably, under each set of test conditions, the initial motor speed, initial variable geometry opening, and initial load level are first set, and the system is allowed to run stably for a period of time before a predetermined operating condition disturbance is applied. The operating condition disturbance can be a sudden load change, a change in the motor speed setpoint, a change in the variable geometry opening, or a combination of at least two of the above disturbances. A multi-parameter synchronous acquisition system records data throughout the entire process—before, during, and after the disturbance—to obtain steady-state performance data, transient response data, and boundary instability data of the air compressor under a wide range of operating conditions.

[0074] In this application, step 1 focuses on acquiring and characterizing the following dynamic processes: 1. Pressure build-up process, i.e., the dynamic process of the air compressor outlet pressure changing from an initial value to a target value; 2. Flow regulation process, i.e., the process of air mass flow following the target air demand; 3. Variable geometry response process, i.e., the response process of regulating components such as inlet guide vanes and diffusers from receiving commands to reaching the target opening degree; 4. Motor speed ramp-up and ramp-down process, i.e., the dynamic acceleration and deceleration of the high-speed drive motor under different given values; 5. Coupling and matching process between air supply and fuel cell stack simulated load, i.e., the coordinated response process of the air supply system when the load demand changes; 6. The stability margin change of the air compressor under conditions such as low flow rate, high pressure ratio, and rapid load change, used to identify the operating characteristics when approaching surge or other instability boundaries.

[0075] To ensure that the collected data covers the actual operating range, this embodiment preferably adopts a wide-condition testing scheme. Specifically, multiple test intervals can be divided according to the rated power range, air excess coefficient range, target pressure ratio range, and maximum motor speed range of the target fuel cell system. Steady-state tests and dynamic tests are performed in each interval. Steady-state tests are mainly used to obtain the pressure, flow, and efficiency relationships under different speeds, different variable geometric openings, and different back pressure conditions; dynamic tests are mainly used to obtain the transient response characteristics under load changes and actuator adjustments. Through the above tests, a raw operating dataset covering the normal operating range, transition operating range, and near-boundary operating range is formed.

[0076] In this embodiment, the operational data includes at least pressure, temperature, and flow rate data characterizing the compressor's operating status; speed, voltage, current, and power data characterizing the motor's operating status; air demand, back pressure, and air supply matching error data characterizing the air supply status; and opening degree, displacement, and response delay data characterizing the variable geometry adjustment status. This operational data is stored after time alignment, anomaly checking, and integrity verification, and is used in subsequent steps to establish a joint simulation model of the compressor, motor, air piping, and fuel cell load coupling.

[0077] Step 2: Based on the multi-condition operating data, establish and calibrate a joint simulation model of the compressor, motor, air pipeline and load coupling, and generate a sample dataset.

[0078] Step 2 includes:

[0079] Step 2.1: Based on the operational data collected in Step 1, establish a joint simulation model including the compressor sub-model, motor sub-model, actuator sub-model, air pipeline sub-model, buffer volume sub-model, and fuel cell stack load sub-model, and realize parameter transfer and boundary coupling between the sub-models through the data interface.

[0080] Specifically, the compressor sub-model characterizes the compression process of a multi-stage centrifugal air compressor under different motor speeds, variable geometry openings, and inlet / outlet boundary conditions, reflecting the variation of characteristics such as outlet pressure, air mass flow rate, pressure ratio, and efficiency with operating conditions. The compressor sub-model is established using the characteristic mapping relationship between pressure ratio, flow rate, and efficiency with respect to motor speed, variable geometry opening, and boundary parameters.

[0081]

[0082]

[0083]

[0084]

[0085] in: This refers to the compressor pressure ratio; is the air mass flow rate; is the compressor efficiency; n is the motor speed; α is the opening degree of the variable geometry adjustment mechanism; To alleviate import pressure; To alleviate export pressure; The inlet temperature; This refers to the compressor load torque. is the specific heat capacity at constant pressure; k is the adiabatic index of the gas; This represents the angular velocity of the motor.

[0086] The motor sub-model is used to characterize the dynamic response of a high-speed drive motor under the combined action of drive commands and load torque. The motor sub-model is established using the rotational dynamics equations relating the motor output torque, compressor load torque, and speed.

[0087]

[0088] Where: J is the equivalent moment of inertia; This is the output torque of the motor; B is the compressor load torque; B is the damping coefficient. This represents the angular velocity of the motor.

[0089] The actuator sub-model is used to characterize the opening change process, response delay, and following characteristics of each stage of the variable geometry control mechanism under control commands. The actuator sub-model is established using a dynamic transfer function that includes response delay and inertial elements, while the air piping sub-model is established using pressure loss relationships.

[0090]

[0091] in: The given value is the opening degree; α is the actual opening degree. For opening response gain; The response time constant; The execution delay time.

[0092] The air piping sub-model and the buffer volume sub-model are used to characterize flow transmission, pressure fluctuations, friction losses, and volumetric buffering effects in the intake and exhaust piping. The air piping sub-model is established using pressure loss relationships:

[0093]

[0094] in: This is due to pipeline pressure loss; Air mass flow rate; air density; D is the pipe length; D is the pipe diameter; This is the overall drag coefficient.

[0095] The buffer volume sub-model is established using a pressure dynamic equation based on mass conservation:

[0096]

[0097] in: To buffer the cavity pressure; To buffer the gas temperature inside the cavity; The effective volume of the buffer cavity; The mass flow rate into the buffer cavity; R represents the mass flow rate out of the buffer cavity; R is the gas constant. The fuel cell stack load sub-model is used to characterize the impact of stack load changes on cathode air demand. The fuel cell stack load sub-model is established using the mapping relationship between load demand and target air demand:

[0098]

[0099] in: Target air mass flow rate; Air demand conversion factor; This refers to the output current of the fuel cell stack. This is the excess air coefficient.

[0100] Each sub-model is coupled through boundary variables such as speed, load torque, opening degree, pressure, flow rate, and air demand. Specifically, the actual speed output by the motor sub-model serves as the driving boundary of the compressor sub-model, the load torque output by the compressor sub-model serves as the load boundary of the motor sub-model, the actual opening degree output by the actuator sub-model serves as the geometric boundary of the compressor sub-model, the pressure and flow rate output by the compressor sub-model serve as the flow boundaries of the air pipeline sub-model and the buffer volume sub-model, and the target air demand output by the fuel cell stack load sub-model serves as the load-side boundary, thus forming a joint simulation model that reflects the dynamic response process of the air supply system.

[0101] Step 2.2: Assign and calibrate the compressor characteristic parameters, motor dynamic parameters, actuator response parameters, air pipeline parameters, buffer volume parameters, and fuel cell load requirement parameters of the co-simulation model based on the operating data, so that the steady-state output characteristics and dynamic response characteristics of the co-simulation model match the experimental data.

[0102] First, the parameters of each sub-model in the co-simulation model are initially assigned values. Specifically, based on the multi-condition operating data obtained in step 1, steady-state test data covering low-speed, medium-speed, and high-speed operating ranges, as well as dynamic test data under conditions of load change, motor speed regulation, and variable geometric opening, are selected as the basis for parameter assignment and calibration.

[0103] For the compressor sub-model, based on inlet pressure, outlet pressure, inlet temperature, outlet temperature, and air mass flow rate data collected under different motor speeds, different opening degrees of the variable geometry control mechanism at various stages, and different inlet and outlet boundary conditions, an initial characteristic mapping relationship of the compressor pressure ratio, flow rate, and efficiency with respect to motor speed and variable geometry opening is established. For example, six operating speed points (30000 r / min, 40000 r / min, 50000 r / min, 60000 r / min, 70000 r / min, and 80000 r / min) can be selected, combined with three opening degree operating points (20%, 50%, and 80%), to interpolate or fit the pressure ratio, air mass flow rate, and efficiency at each operating point, thereby obtaining the initial values ​​of the compressor characteristic parameters, pressure ratio parameters, and efficiency parameters.

[0104] For the motor sub-model, based on the motor speed, motor voltage, motor current, and drive power data, initial values ​​are assigned to the motor inertia parameters, damping parameters, torque response parameters, and drive response parameters. For example, the initial values ​​of the motor inertia parameters and torque response parameters can be determined using the least squares identification method, based on the motor speed change curves during the no-load and load acceleration phases, combined with the power input and load torque changes.

[0105] For the actuator sub-model, based on the opening setpoints, actual opening feedback values, and actuator displacement data of each stage of the variable geometric adjustment mechanism, preliminary values ​​are assigned to the actuator response time, lag time, opening gain, and opening change rate limit parameters. For example, the actuator's delay time and first-order inertial time constant can be obtained by fitting the actual opening response curve under a step command.

[0106] For the air pipeline sub-model and the buffer volume sub-model, preliminary values ​​are assigned to the pipeline resistance parameters, local loss parameters, volume parameters, and pressure loss parameters based on the pressure, flow rate, and pressure fluctuation data on the intake and exhaust sides. For example, the pipeline resistance coefficient can be determined based on the relationship between pressure loss and flow rate under steady-state conditions; and the initial values ​​of the buffer volume parameters can be determined based on the pressure change process within the buffer cavity during sudden load changes.

[0107] For the fuel cell stack load sub-model, an initial correspondence between load changes and target air demand is established based on the load power change rate, target excess air coefficient, and air demand change data. For example, piecewise linear fitting or lookup table mapping can be used to establish a model relating load power, excess air coefficient, and target air mass flow rate.

[0108] After initially assigning the above parameters, the joint simulation model is further calibrated using the multi-condition test data from step 1. This calibration includes steady-state calibration and dynamic calibration.

[0109] During steady-state calibration, steady-state test points under different motor speeds, variable geometric openings, and load conditions are selected. The simulation output results of the co-simulation model are compared with the experimental measurement results, and at least the changes in outlet pressure, air mass flow rate, pressure ratio, and temperature are checked. When the error of any steady-state index exceeds the preset range, the characteristic parameters and efficiency parameters in the compressor sub-model and the resistance parameters in the air pipeline sub-model are adjusted until the deviations of each steady-state output index from the experimental data are reduced to within the allowable range. For example, the relative error of outlet pressure can be controlled within ±5%, the relative error of air mass flow rate can be controlled within ±5%, and the pressure ratio error can be controlled within ±5%.

[0110] During dynamic calibration, typical dynamic conditions such as load step changes, motor speed setpoint changes, variable geometry adjustment mechanism opening changes, and sudden changes in air demand are selected. The dynamic response curves of the co-simulation model are compared with the experimental measurement curves. At least the transient boost time, flow regulation time, motor speed climb and fall process, actuator response delay, and air supply matching error are checked. When the error of any dynamic index exceeds the preset range, the inertia parameters and torque response parameters in the motor sub-model, the delay time and time constant parameters in the actuator sub-model, the volume parameters in the buffer volume sub-model, and the air demand mapping parameters in the load sub-model are adjusted until the dynamic response results are basically consistent with the experimental data. For example, the transient boost time error can be controlled within ±10%, the motor speed transition process error within ±8%, the actuator response delay error within ±10%, and the air supply matching error within the preset threshold.

[0111] For example, in a set of dynamic calibration conditions, the initial operating point of the system is set to a motor speed of 50,000 r / min, an opening of 50% for each stage of the variable geometry adjustment mechanism, and a target load of 40% of the rated load. After the system stabilizes, a step disturbance is applied, transitioning from 40% to 70% of the rated load, while keeping the motor speed setpoint constant. The changes in outlet pressure, air mass flow rate, actual motor speed, actual opening of each stage of the variable geometry adjustment mechanism, and air supply matching error are recorded. After comparing the experimental curves with the simulation curves, if the outlet pressure build-up time in the simulation model is faster than the experimental result, the buffer volume parameter is increased or the pipeline resistance parameter is appropriately corrected; if the motor acceleration process in the simulation model is faster than the experimental result, the motor inertia parameter is increased or the torque response parameter is corrected; if the opening response of the variable geometry adjustment mechanism is faster than the experimental result, the actuator delay time is increased. This correction process is repeated until the main output curve of the co-simulation model under this dynamic condition matches the experimental curve.

[0112] When the errors between the co-simulation model and the experimental data in both the main steady-state and main dynamic indices meet the preset requirements, the co-simulation model is considered to have completed calibration, and the calibrated co-simulation model is used for the subsequent generation of wide-condition sample datasets. Preferably, data samples independent of the parameter assignment and calibration stages are used to verify the co-simulation model to improve the model's generalization ability and reliability under unseen conditions.

[0113] Step 2.3: Based on the calibrated co-simulation model, simulation calculations are performed under different load change conditions, different air demand conditions, different motor speed conditions, and different variable geometric adjustment states. Combined with the experimental data obtained in Step 1, a sample dataset covering steady-state conditions, dynamic transition conditions, and boundary conditions is formed.

[0114] Specifically, the simulation operating conditions are first set according to the actual operating range of the target fuel cell air supply system. For example, the motor speed is divided into low-speed, medium-speed, and high-speed zones, corresponding to 30000 r / min-45000 r / min, 45000 r / min-65000 r / min, and 65000 r / min-85000 r / min, respectively. The target air excess coefficient is divided into multiple set ranges, such as 1.8-2.8. Load change conditions are divided into steady-state load, step load increase, step load decrease, ramp load increase, ramp load decrease, and periodic load change. Variable geometry adjustment states are divided into different stages of inlet guide vane opening, different stages of diffuser opening, and different inter-stage combined openings. Multiple sets of simulation operating conditions are generated based on the above variable combinations.

[0115] In a specific example, the initial motor speed is selected as 50,000 r / min, the initial opening of each stage of the variable geometry adjustment mechanism is 50%, the initial target air excess coefficient is 2.2, and the initial system load is 40% of the rated load. After the system reaches the initial steady state, a load step disturbance is applied at simulation time t=2s, increasing the target load from 40% to 70% of the rated load; simultaneously, at t=3s, the target air excess coefficient is adjusted from 2.2 to 2.5; and at t=4s, an opening command is applied to the second-stage inlet guide vane, adjusting its opening from 50% to 65%. The co-simulation model performs iterative calculations according to a unified simulation step size, simultaneously solving the compressor sub-model, motor sub-model, actuator sub-model, air pipeline sub-model, buffer volume sub-model, and fuel cell stack load sub-model at each simulation time, obtaining the outlet pressure, air mass flow rate, actual motor speed, actual opening of each stage of the variable geometry adjustment mechanism, actuator displacement, and the matching results between air supply and fuel cell stack load demand.

[0116] In the simulation, the actual rotational speed output by the motor sub-model is transmitted to the compressor sub-model as the driving boundary; the actual opening degree output by the actuator sub-model is transmitted to the compressor sub-model as the geometric boundary; the outlet pressure and air mass flow rate output by the compressor sub-model are transmitted to the air pipeline sub-model and the buffer volume sub-model as the flow boundary; and the target air demand output by the fuel cell stack load sub-model is transmitted to the air pipeline model as the load boundary. Through the above time-coupled solution, the complete dynamic response curve for each operating condition is obtained.

[0117] During the steady-state operating condition sample generation process, simulation calculations are performed on the steady-state operating points under different motor speeds, different variable geometric openings, and different back pressure conditions. Data such as inlet pressure, outlet pressure, inlet temperature, outlet temperature, air mass flow rate, motor speed, opening of each stage of variable geometric adjustment mechanism, actuator displacement, pressure ratio, efficiency, and air supply matching error are recorded under steady-state conditions to form steady-state operating condition samples.

[0118] During the generation of dynamic transition condition samples, the full-process response under conditions of load step change, motor speed setpoint change, target air excess coefficient change, variable geometric opening change, and their combined disturbances is simulated. Continuous time-series data before, during, and after the disturbance are recorded to form dynamic transition condition samples. Dynamic transition condition samples include at least the pressure build-up process data, flow regulation process data, motor speed ramp-up and ramp-down process data, actuator response process data, and air supply and load demand coupling matching process data.

[0119] During the generation of boundary condition samples, the focus is on extending the simulation to conditions approaching critical operating regions such as low flow rate, high pressure ratio, rapid load changes, and large opening changes, in order to obtain response characteristic data of the system near surge boundary, insufficient gas supply boundary, or actuator saturation boundary. For example, under the condition of maintaining high back pressure and low flow rate, the motor speed is gradually reduced or the variable geometric opening is reduced, and the outlet pressure fluctuation, flow rate fluctuation, gas supply matching error, and actuator limit response are recorded to form boundary condition samples.

[0120] After obtaining the simulation data, the simulation data is aligned with the experimental data obtained in step 1 according to a unified time reference. Specifically, the data of each channel is resampled using a unified sampling period, and time alignment is completed based on timestamps or synchronization trigger markers; for data with missing points, nearest neighbor interpolation, linear interpolation, or spline interpolation is used to fill in the missing points; for obvious outliers and distorted points, they are removed or corrected by combining physical constraints and statistical thresholds.

[0121] Furthermore, the experimental and simulation data are fused according to the operating condition categories. For operating condition ranges that are fully covered by experimental data, experimental data is retained first to maintain the ability of the sample data to represent the real operating process. For operating condition ranges that are difficult to be fully covered by experiments, especially large-scale dynamic disturbance conditions, extreme boundary conditions, and multivariate coupled disturbance conditions, supplementary samples are generated using the calibrated co-simulation model to improve the coverage of the sample dataset for wide operating conditions.

[0122] In one embodiment, the final sample dataset can be constructed according to the following principles: steady-state operating condition samples consist of both experimental data and simulation data, with experimental data accounting for a higher proportion; dynamic transitional operating condition samples consist of a mixture of experimental data and extended simulation data, used to enhance the control model's ability to learn from dynamic changes; boundary operating condition samples are mainly composed of extended simulation data, supplemented by a small amount of experimental verification data, used to enhance the control model's ability to identify critical operating conditions and constraint boundaries.

[0123] After time alignment, outlier handling, missing point correction, operating condition labeling, and unified format storage, a sample dataset covering steady-state, dynamic transition, and boundary conditions is finally formed. The sample dataset includes at least inlet pressure, outlet pressure, inlet temperature, outlet temperature, air mass flow rate, motor speed, opening degree of each stage of variable geometry control mechanism, actuator displacement, load demand, target air excess coefficient, and the matching results between air supply and stack load demand. It may further include operating condition category labels, disturbance type labels, and boundary state labels to provide a data foundation for subsequent mechanism constraint characterization, control action feasible domain construction, and control model training.

[0124] Step 3: Construct a set of mechanistic constraints based on the sample dataset, and use the motor speed and the opening degree of each stage of the variable geometric adjustment mechanism as control variables to determine the feasible region of control action according to the set of mechanistic constraints.

[0125] The set of mechanistic constraints includes: pressure build-up capability, flow regulation capability, surge margin, and compressor efficiency characteristics used to characterize compressor operating constraints; drive-actuator joint dynamic boundary used to characterize motor operating constraints and variable geometric adjustment range constraints; and air supply matching characteristics used to characterize air pipeline operating constraints.

[0126] Step 3 includes:

[0127] Step 3.1: Based on the sample dataset obtained in Step 2, the samples are classified according to the motor speed range, variable geometric opening range, load change rate range, and air excess coefficient range. Within each classification range, pressure build-up capability, flow regulation capability, surge margin, actuator dynamic boundary, and air supply matching characteristics are extracted and characterized to establish the mechanism constraint characterization relationship under different operating conditions.

[0128] Based on the sample dataset obtained in step 2, and considering the wide operating conditions of multi-stage centrifugal air compressors, a mechanistic constraint characterization system is constructed, encompassing pressure build-up capability, flow regulation capability, surge margin, actuator dynamic boundaries, and air supply matching characteristics. To uniformly characterize the operating state of the multi-stage centrifugal air compressor at the current control moment, the system state vector is first defined as:

[0129]

[0130] in, This represents the compressor inlet pressure at time k; This represents the compressor outlet pressure at time k; This indicates the inlet temperature at time k; This represents the outlet temperature at time k; This represents the air mass flow rate at time k; This represents the motor speed at time k; This indicates the opening degree of the inlet guide vane at time k; This indicates the diffuser opening at time k; This represents the displacement of the actuator at time k; Indicates the fuel cell load current; Indicates the rate of change of load; This indicates the excess air coefficient.

[0131] Based on this, using the sample dataset obtained in step 2, construct the state vectors respectively. The relevant relationships are characterized by pressure build-up capability, flow regulation capability, surge margin, actuator dynamic boundary, and air supply matching characteristics. In a specific example, samples within the following range are selected from the sample dataset as a single operating condition category: motor speed 45000 r / min-60000 r / min, inlet guide vane opening 40%-60%, diffuser opening 35%-55%, load change rate 8% / s-12% / s of rated load, and excess air coefficient 2.0-2.4.

[0132] Specifically, regarding pressure build-up capability, based on the compressor outlet pressure variation process under disturbed operating conditions in the sample data, the pressure rise per unit time, the time required to reach the target pressure, and the steady-state pressure recovery deviation are extracted. For example, in a sample where the load jumps from 40% to 70% of the rated load, the time required for the outlet pressure to rise from 150 kPa to 190 kPa is 1.6 s, the pressure rise per unit time is 25 kPa / s, and the steady-state recovery deviation is controlled within ±2 kPa. This establishes the correlation between pressure build-up capability and motor speed, inlet guide vane opening, diffuser opening, and load change rate.

[0133] Regarding flow regulation capability, based on the tracking process of air mass flow rate to target air demand, flow tracking error, flow response time, and steady-state flow deviation are extracted. For example, in a sample where the target air mass flow rate changes from 0.075 kg / s to 0.090 kg / s, the flow regulation time is 1.2 s, and the steady-state tracking error is less than 3%. Based on this, a relationship between flow regulation capability and system state vector is established. The correspondence between them.

[0134] For surge margin, the safe distance of each operating point relative to the surge boundary is determined based on the changes in pressure, flow rate, and pressure ratio under near-boundary operating conditions. For example, in the low flow rate, high pressure ratio sample range, the minimum safety margin threshold is determined by comparing the flow rate difference and pressure ratio difference between the current operating point and the critical surge point. The value is set to 0.10, and a surge margin function is established. The correspondence between the operating status and control actions.

[0135] For the dynamic boundary of the motor, the upper limit of the rate of change of motor speed and the maximum allowable adjustment amplitude are determined based on the motor's acceleration and deceleration processes. For example, under the above operating condition category, the maximum allowable adjustment amplitude of the current motor speed can be determined based on sample data. The speed is set to 3000 r / min, and the boundary condition and system state vector are established. The correspondence between them.

[0136] For the dynamic boundaries of the actuators, the upper limit of the opening change rate, the actuator response delay range, and the displacement boundary of each stage of the variable geometry adjustment mechanism are determined based on the action process of each stage. For example, based on the response curves of the inlet guide vane and diffuser under a step command, the maximum allowable adjustment amplitude of the inlet guide vane under the current operating condition can be determined. The maximum allowable adjustment range of the diffuser is 5%. The response time is 4%, the actuator response delay is no more than 0.15s, and its relationship with the system state vector is established. The correspondence between them.

[0137] For the air supply matching characteristics, based on the deviation between the air supply and the fuel cell stack load demand, a system state vector is established, relating the changes in the excess air coefficient, the amount of insufficient air supply, the amount of excess air supply, and the system state vector. The correspondence between them. For example, within this operating condition category, the allowable range of the excess air coefficient can be set to 1.9 ≤ ≤2.6.

[0138] For compressor efficiency characteristics, a compressor efficiency function is established based on the pressure ratio, flow rate, and drive power data in the sample data. And set a minimum allowable efficiency threshold. For example, under this operating condition category, Set to 0.68.

[0139] Using the above method, system state vectors are formed in each classification interval. The corresponding mechanistic constraint representation relationship is obtained, thus yielding a mechanistic constraint representation system suitable for the subsequent construction of the feasible domain of control actions.

[0140] Step 3.2: Based on the mechanism constraint characterization system constructed in Step 3.1, using the motor speed regulation amount and the opening regulation amount of each stage of variable geometric adjustment mechanism as control variables, establish the control action feasible domain that satisfies surge safety margin constraints, actuator dynamic constraints, air excess coefficient constraints and compressor efficiency constraints according to the mechanism constraint characterization system.

[0141] Specifically, in this application, the control vector is defined as follows:

[0142]

[0143] in, This represents the amount of motor speed adjustment at time k. The amount of adjustment of the inlet guide vane opening at time k; This represents the amount of diffuser opening adjustment at time k.

[0144] For the current time k, a feasible region for control actions is established, which is jointly defined by the boundaries of multiple mechanisms. :

[0145]

[0146] in, The surge margin function; For minimum safety margin; This is a function of compressor efficiency; The minimum allowable efficiency threshold; This represents the maximum allowable adjustment range of the motor speed. This refers to the maximum allowable adjustment range of the inlet guide vane opening; This represents the maximum allowable adjustment range of the diffuser opening. and These are the lower and upper limits of the excess air coefficient, respectively.

[0147] In a specific example, when the system operates under the following conditions: inlet pressure 101 kPa, outlet pressure 180 kPa, inlet temperature 298 K, outlet temperature 395 K, air mass flow rate 0.085 kg / s, motor speed 52000 r / min, inlet guide vane opening 52%, diffuser opening 48%, actuator displacement 12 mm, fuel cell load current 220 A, load change rate 10 A / s, and air excess coefficient 2.15, then the system state vector at that moment is defined by the formula above. .

[0148] In this state, the control model generates the first set of candidate control actions as follows:

[0149]

[0150] Substitute the candidate control action into the control action feasibility region criterion for calculation. If the result is: , , And under the current operating conditions , , ΔαDV,max=4%, λmin=1.9, , Since SMmin = 0.10, the above candidate control action simultaneously satisfies the surge safety margin constraint, motor dynamic boundary constraint, actuator dynamic boundary constraint, air excess coefficient constraint, and compressor efficiency constraint. Therefore, this candidate control action belongs to the feasible region of control actions at the current moment. Accordingly, if the control model generates the second set of candidate control actions as follows:

[0151]

[0152] Because its motor speed regulation exceeds the maximum allowable regulation amplitude ΔN under the current operating conditions. max If the speed is 3000 r / min, then the candidate control action does not meet the motor dynamic boundary constraints and therefore does not belong to the feasible region of control actions at the current moment. .

[0153] For example, if the control model generates a third set of candidate control actions:

[0154]

[0155] Although the candidate control action satisfies the constraints of motor regulation amplitude and opening regulation amplitude, if the calculation yields... Below the lower limit of the excess air coefficient This indicates that the action may lead to insufficient gas supply, therefore this candidate control action also does not belong to the feasible region of control actions. .

[0156] By employing the above method, the surge margin, motor dynamic boundary, actuator dynamic boundary, gas supply matching characteristics, and compressor efficiency characteristics extracted in step 3.1 can be uniformly incorporated into the determination process of the feasible domain of control actions. In other words, only candidate control actions that simultaneously satisfy all mechanistic constraints are considered feasible control actions at the current moment. This approach allows the subsequent control model to consider not only flow and pressure regulation requirements when generating control commands, but also operational safety, execution accessibility, gas supply matching capability, and system efficiency.

[0157] Furthermore, after the feasible domain of control actions is constructed, data within the feasible domain, near the feasible domain boundary, and in the dynamic transition region can be classified and labeled under various speeds, opening combinations, load change modes, and environmental boundaries to form state samples and action samples consistent with mechanistic constraints. This provides a foundation for subsequent mechanistic consistency state encoding, candidate action projection correction, and control model training.

[0158] Step 4: Preprocess the sample dataset, and based on the preprocessed data, fuse and encode the current running state, historical state sequence and corresponding mechanism constraint information to obtain mechanism consistency state features.

[0159] Specifically, based on the sample dataset processed in step 3, preprocessing is performed first. Assuming that both experimental and simulation data are collected according to a uniform sampling period, for example, a sampling period of 10ms, the state, action, constraint, and performance data obtained in steps 2 and 3 are uniformly resampled and aligned according to timestamps.

[0160] In this embodiment, the preprocessing includes at least: state variable normalization, dynamic sequence windowing, outlier removal, noise filtering, explicit labeling of constraint boundaries, working condition category clustering, and separation and reconstruction of short-term dynamic features and long-term performance features.

[0161] State variable normalization is used to standardize the scale of variables with different dimensions, such as pressure, temperature, flow rate, motor speed, variable geometric opening, actuator displacement, air excess coefficient, and load change rate, in order to reduce the impact of differences in the order of magnitude of variables on subsequent training and encoding processes. Normalization can be implemented using interval mapping, linear normalization, or standardization. For example, min-max normalization can be used to map each variable to the [0,1] interval; zero-mean standardization can also be used to standardize the variables. Through normalization, the impact of differences in the order of magnitude of different variables on subsequent encoding and model training is reduced.

[0162] Dynamic sequence windowing is used to divide continuously running data into multiple data windows with time-series information according to a preset time length. Each data window includes at least the current state, a historical state sequence of several consecutive sampling times before the current state, and the control actions, constraint information, and performance indicators corresponding to that time period. Through windowing, the subsequent encoding process can utilize not only the current state information but also the dynamic trend information during the state evolution process. For example, in a dynamic operating condition where the load jumps from 40% to 70% of the rated load, the state sequence can be continuously extracted from 0.2 seconds before the disturbance to several times after the disturbance to reflect the dynamic trends of the pressure build-up process, flow regulation process, motor speed change process, and actuator opening adjustment process.

[0163] Outlier removal identifies and removes non-representative data caused by transient sensor distortion, communication interruptions, actuator jitter, or abnormal sampling at boundary conditions. Noise filtering reduces high-frequency interference components in sampled signals such as pressure, flow rate, speed, and displacement, improving the stability of subsequent state characterization. Outlier identification and filtering can be performed based on physical accessibility, adjacent sampling point deviations, rate of change constraints, and multivariate consistency relationships.

[0164] Explicit constraint boundary labeling is used to map the mechanistic constraint representation system constructed in step 3 onto the sample data. Specifically, for each sample or each time window, its corresponding surge margin state, actuator dynamic margin state, air excess coefficient boundary state, efficiency boundary state, and air supply matching state are labeled, so that the sample contains not only state variables and action variables, but also corresponding mechanistic constraint attribute information. For example, within a certain time window, if the surge margin function value at the current operating point... Approaching minimum safety margin If so, then the sample is marked as "near surge boundary"; if near If the gas supply is insufficient, the sample will be marked as "increased risk of insufficient gas supply"; if the current action margin of the actuator is small, it will be marked as "insufficient dynamic margin of the actuator".

[0165] Operating condition clustering categorizes samples based on characteristics such as load variation patterns, air demand levels, motor speed ranges, variable geometric opening combinations, operating area locations, and stability margin levels. For example, samples can be divided into three categories: steady-state operating conditions, dynamic transient operating conditions, and near-boundary operating conditions. The steady-state operating condition primarily characterizes the steady-state pressure, flow rate, and efficiency relationships under different speeds and opening conditions; the dynamic transient operating condition primarily characterizes the transient response process under conditions of sudden load changes, speed adjustments, and opening changes; and the near-boundary operating condition primarily characterizes the constraint boundary characteristics under conditions of low flow rates, high pressure ratios, rapid load changes, and large opening changes.

[0166] The separation and reconstruction of short-term dynamic characteristics and long-term performance characteristics is used to extract and reconstruct data characterizing the short-term response process and data characterizing the long-term operating effect separately. Short-term dynamic characteristics include at least the pressure change rate, flow tracking deviation, speed response slope, response delay of the variable geometry control mechanism, and regulation oscillation level; long-term performance characteristics include at least compressor efficiency, power consumption level, stability margin maintenance, and long-term smoothness of actuator operation. Through separation and reconstruction, the subsequent control model can learn the transient regulation law and long-term operating law of the system separately. For example, for a set of sample data lasting 10 seconds, the pressure build-up slope, flow error decay rate, and motor speed-up slope in the first second can be used as short-term dynamic characteristics, while the average efficiency, average power consumption, minimum surge margin, and cumulative smoothness of operation over the entire 10-second interval can be used as long-term performance characteristics. Through this separation and reconstruction, the subsequent encoding process can simultaneously perceive short-term dynamic changes and long-term operating effects.

[0167] After the above preprocessing, a unified dataset of state, action, constraint, and performance is formed. Each sample unit in the dataset includes at least: the current running state, the historical state sequence, the corresponding control action, the corresponding mechanism constraint information, and the short-term and long-term performance indicators related to the sample, thus providing standardized input for subsequent mechanism-consistent state coding.

[0168] Furthermore, after completing the sample data preprocessing, based on the preprocessed dataset, the current operating state, historical state sequence, and corresponding mechanistic constraint information are fused and encoded to obtain mechanistic consistency state features. These mechanistic consistency state features characterize the system's operating state at the current control moment, its state evolution trend, and its consistency relationship with the mechanistic constraint boundaries. Let the historical state sequence of length L be... Then, after encoding, the mechanism consistency feature vector is obtained as follows:

[0169]

[0170] in, Mechanism-consistent state encoding mapping; This refers to the mechanism constraint information corresponding to the current moment.

[0171] Specifically, for the current state encoding branch, the current state vector is... As a real-time status input, the current operating point features are extracted through fully connected mapping or other feature transformation units to characterize information such as pressure, flow rate, speed, opening degree, displacement, load current, and air excess coefficient at the current moment.

[0172] For the historical sequence encoding branch, the historical state sequence within the preset window length is... As a dynamic trend input, it is represented through a time-series coding structure. This time-series coding structure can be implemented using unified coding after feature concatenation, fusion coding after branch extraction, or other methods suitable for time-series data processing. For example, in a dynamic operating condition where the load jumps from 40% to 70% of the rated load, the historical state sequence can reflect the process of gradually building up the outlet pressure, gradually following the air mass flow rate, continuously increasing the motor speed, and gradually opening the actuator, thereby extracting information such as the pressure building trend, flow rate change trend, load growth trend, and actuator dynamic trend.

[0173] For the constraint information encoding branch, the mechanism constraint information corresponding to the current moment will be... As a constraint input. This can include surge margin status, actuator dynamic boundary status, air excess coefficient boundary status, efficiency boundary status, and air supply matching status. It can also further include residual margin information related to each constraint boundary. For example, when the surge margin of a sample is low, the actuator dynamic margin is small, and the air excess coefficient is close to the lower limit at a certain moment, the constraint coding branch will highlight that the operating point is in the "near the boundary" state.

[0174] By fusing the current state encoding results, historical sequence encoding results, and constraint information encoding results, we obtain the mechanism consistency state feature vector. The fusion method can be achieved through vector concatenation followed by unified mapping, weighted fusion, or attention fusion. The fused result... It is no longer a simple combination of original state quantities, but a low-dimensional consistent representation that simultaneously includes the current state, short-term dynamic trends, and the relative positional relationships of mechanistic constraint boundaries.

[0175] For example, at the current control time k, the system operating state meets the following conditions: inlet pressure is 101 kPa, outlet pressure is 182 kPa, inlet temperature is 298 K, outlet temperature is 398 K, air mass flow rate is 0.086 kg / s, motor speed is 53000 r / min, inlet guide vane opening is 54%, diffuser opening is 49%, actuator displacement is 12.5 mm, fuel cell load current is 225 A, load change rate is 11 A / s, and air excess coefficient is 2.05. At this time, the state vectors of this time and the previous 20 consecutive sampling times are used to form a historical state sequence. If the mechanistic constraint information at that moment indicates that: the current surge margin function value is close to the minimum safety margin threshold, the actuator opening adjustment margin is close to the upper limit, the air excess coefficient is at the lower edge of the allowable range, and the compressor efficiency meets the minimum threshold requirement, then the above constraint states are uniformly represented as follows: And input it into the mechanism consistency state encoding module. After encoding mapping Then, the mechanism consistency feature vector is obtained. .

[0176] In this embodiment, the mechanism consistency feature vector It can simultaneously characterize: the current operating point is in the operating range of medium-high speed, medium-to-large opening, and rapid load increase; the current system outlet pressure is being established, the air mass flow rate is following the target air demand, and the motor speed is still in the process of increasing; the current operating point is close to the surge boundary and the air supply matching boundary, but has not yet violated the safety constraints and efficiency constraints; the actuator has a certain margin of motion, but continuing to increase the opening by a large margin may approach the dynamic boundary.

[0177] Therefore, compared to directly using the original state variables as the input to the control model, the mechanism-consistent state characteristics... It can more comprehensively reflect the current operating state of the system and its relationship with the mechanistic boundary, thereby improving the pertinence of subsequent candidate control action generation, control action feasible domain projection correction, and dual time domain comprehensive evaluation.

[0178] Step 5: Establish a control model based on the consistency state characteristics of the mechanism, and generate candidate control actions based on the control model, including the motor speed adjustment amount and the opening adjustment amount of each level of variable geometric adjustment mechanism. Project the candidate control actions into the feasible domain of the control actions to obtain the current cycle control command.

[0179] Step 5 includes:

[0180] Step 5.1: Based on the mechanism consistency state characteristics obtained in Step 4, candidate control actions containing motor speed adjustment and opening adjustment of each level of variable geometric adjustment mechanism are generated using the candidate action generation mapping.

[0181] Specifically, based on the mechanism consistency state characteristics obtained in step 4 Construct a candidate action generation module to output candidate control actions at the current control moment:

[0182]

[0183] in, Generate mappings for candidate actions; This indicates the initial control action that has not been modified by constraints.

[0184] In this application, candidate control action It includes at least the motor speed adjustment and the opening adjustment of each level of the variable geometric adjustment mechanism. In other words, the candidate action generation module considers the consistency state characteristics of the mechanism. The information contained in the data, such as pressure build-up trend, flow rate change trend, load growth trend, surge approach degree, actuator dynamic margin, and air supply matching deviation, generates initial control quantities for adjusting motor speed and the opening degree of each stage of variable geometry components.

[0185] Furthermore, the candidate action generation module does not directly output actions based on the original measurement state at a single moment, but rather on the mechanism consistency state features after fusion encoding in step 4. Control candidate variables are generated. Since the mechanism consistency state characteristics have integrated the current operating state, historical state sequence, and corresponding mechanism constraint information, the generated candidate control actions can not only reflect the real-time requirements of the current system, but also reflect the dynamic evolution trend of the system and its relationship with the constraint boundary.

[0186] In this application, the candidate action generation module outputs As the initial result for subsequent control decisions, its focus is on quickly providing the action direction and action amplitude that match the pressure regulation requirements, flow regulation requirements and load change requirements based on the current operating conditions, so as to provide a basis for subsequent constraint correction.

[0187] Step 5.2: Input the candidate control action into the feasible region of control action established in step 3, and perform projection correction on the candidate control action based on the feasible region projection operator to obtain the corrected control action that satisfies the mechanism constraint.

[0188] Specifically, in obtaining candidate control actions Subsequently, to ensure that the control actions always meet the aerodynamic stability requirements and execution constraints under complex operating conditions, the candidate control actions are projected onto the feasible domain of control actions constructed in step 3. The corrected control actions are obtained internally:

[0189]

[0190] in, For feasible region projection operators; The control action is after projection correction.

[0191] In this application, feasible region projection correction refers to: when candidate control actions It is already in the feasible region of control actions. If the candidate control action is within the specified time, then the candidate control action remains unchanged; if the candidate control action is within the specified time, then the candidate control action remains unchanged. Exceeding the feasible domain of control actions When the feasible region projection operator is used, it is mapped to the boundary of the feasible region or the interior of the feasible region that satisfies the constraints, so as to obtain the corrected control action that satisfies the mechanism constraint requirements. .

[0192] Control Action Feasible Domain As defined by the mechanism constraint characterization system in step 3, at least the following constraints must be considered simultaneously during the projection correction process: surge safety margin constraint, actuator dynamic boundary constraint, air excess coefficient constraint, and compressor efficiency constraint. Therefore, the control action after projection correction... It can not only meet the current pressure and flow regulation needs, but also prevent the control action from pushing the system operating point to the surge boundary or other unsafe areas, while ensuring that the actions of each actuator are within the feasible range.

[0193] Unlike traditional control methods that first generate actions and then limit them using additional safety modules, this embodiment projects candidate control actions directly into the feasible domain of control actions, embedding mechanistic constraints as a component of the control decision-making process into the action generation flow. This approach ensures that subsequent output control actions possess mechanistic consistency and constraint satisfaction during the generation stage, thereby improving the interpretability, stability, and implementability of control commands under complex operating conditions.

[0194] Step 5.3: Determine the control action after projection correction as the final control command so that the final control command simultaneously satisfies the surge safety boundary, air supply matching boundary, efficiency boundary and actuator dynamic boundary.

[0195] After projecting and correcting the candidate control actions, the corrected control actions will be... The final control command is determined as the current control moment and used for subsequent control execution. In this application, the final control command includes at least a motor speed adjustment command and opening adjustment commands for each level of the variable geometry adjustment mechanism.

[0196] Specifically, the corrected control actions can be... The component corresponding to the motor speed adjustment is used as the target adjustment amount for the high-speed drive motor. The component corresponding to the opening adjustment amount of each level of the variable geometry adjustment mechanism is used as the target adjustment amount of each level of the variable geometry adjustment mechanism, thereby forming a complete execution instruction for the current control cycle.

[0197] Because the final control command is derived from the candidate control action The control command, obtained through feasible region projection correction, simultaneously satisfies all the mechanistic constraints defined in step 3. In other words, while ensuring pressure build-up and flow regulation capabilities, the final control command also meets the requirements of surge safety boundaries, air supply matching boundaries, efficiency boundaries, and actuator dynamic boundaries. This enables multi-stage centrifugal air compressors to achieve rapid, stable, and constraint-consistent coordinated regulation under a wide range of operating conditions.

[0198] In one specific embodiment, the controlled object is a three-stage centrifugal air compressor used in a fuel cell air supply system. The multi-stage centrifugal air compressor includes a high-speed drive motor, a first-stage adjustable inlet guide vane mechanism, and a second-stage adjustable diffuser mechanism. The controller operates on a rolling basis with a fixed control cycle. At the current control time k, it first collects the real-time operating status information of the system and combines it with the mechanism consistency state characteristics obtained in step 4. Candidate control actions for the current cycle are generated. In this embodiment, the current system operating state can be as follows: compressor inlet pressure is 101 kPa, outlet pressure is 182 kPa, inlet temperature is 298 K, outlet temperature is 398 K, air mass flow rate is 0.086 kg / s, motor speed is 53000 r / min, first-stage inlet guide vane opening is 54%, second-stage diffuser opening is 49%, actuator displacement is 12.5 mm, fuel cell load current is 225 A, load change rate is 11 A / s, and air excess coefficient is 2.05. Combining the current moment and the historical state sequence of multiple consecutive sampling moments, and integrating surge margin, actuator dynamic margin, air excess coefficient boundary state, and efficiency boundary state, the mechanism consistency state characteristics of the current control moment are obtained. Based on the consistent state characteristics of the mechanism In this embodiment, the candidate control actions include motor speed adjustment, first-stage inlet guide vane opening adjustment, and second-stage diffuser opening adjustment. Under the current operating conditions, the candidate action generation module outputs the following candidate control actions based on the characteristics of the outlet pressure being established, the air mass flow not yet fully following, the load continuing to rise, and the current operating point approaching the lower edge of the surge safety boundary: This indicates that the control model intends to increase the target speed of the high-speed drive motor by 3200 r / min, increase the opening of the first-stage inlet guide vane by 4.6%, and increase the opening of the second-stage diffuser by 3.1% within the current control cycle, so as to rapidly improve the system's air supply capacity and pressure build-up speed.

[0199] After obtaining candidate control actions Then, input it into the control action feasible domain established in step 3. Projection correction is performed based on the feasible region projection operator. In this embodiment, the mechanistic constraint under the current operating condition is: minimum surge safety margin. =0.10, maximum allowable adjustment range of the motor =3000 r / min, maximum allowable adjustment amplitude of the first-stage inlet guide vane =5%, maximum allowable adjustment range of the secondary diffuser =4%, the allowable range for excess air coefficient is 1.9≤λ≤2.6, and the minimum allowable efficiency threshold is =0.68. Candidate control actions. After substituting the control action feasible region criterion into the calculation, the prediction shows that the motor adjustment corresponding to this action has exceeded the maximum allowable adjustment amplitude under the current operating condition. Furthermore, if this action is executed directly, the system operating point will further approach the surge boundary, causing the surge margin to decrease to 0.098, which is lower than the minimum safety margin threshold of 0.10. Therefore, this candidate action does not belong to the control action feasible region at the current moment. .

[0200] The controller further performs projection correction on the candidate control action, mapping it to the feasible region that satisfies the constraints, resulting in the corrected control action: After correction, the predicted calculations show that the surge margin recovers to 0.115, the excess air coefficient is 2.03, the compressor efficiency is 0.701, and none of the actuator components exceed the dynamic boundary limits. Therefore, the corrected control action simultaneously satisfies the surge safety margin constraint, the actuator dynamic boundary constraint, the excess air coefficient constraint, and the compressor efficiency constraint.

[0201] After completing the projection correction, the corrected control actions will be... The final control command for the current control cycle is determined. Specifically, it will... The component corresponding to the motor speed adjustment is used as the target adjustment value for the high-speed drive motor, and the components corresponding to the opening adjustment values ​​of the first-stage inlet guide vane and the second-stage diffuser are used as the target adjustment values ​​for the corresponding variable geometry adjustment mechanisms. For example, if the actual motor speed is 53000 r / min, the actual opening of the first-stage inlet guide vane is 54%, and the actual opening of the second-stage diffuser is 49%, then the final control command can be:

[0202]

[0203]

[0204]

[0205] The controller sends the final control command to the variable frequency drive and the corresponding actuator drive unit to complete the coordinated adjustment of the high-speed drive motor and the variable geometry adjustment mechanism at each stage within this control cycle. Since the final control command is obtained by projecting and correcting the candidate control action through the feasible region, it simultaneously satisfies the pressure build-up requirements, flow regulation requirements, and various mechanistic constraint boundary requirements at the output stage.

[0206] Step 6: Construct a comprehensive evaluation model combining short-term and long-term time domains based on the current cycle control command, and update the control model according to the comprehensive evaluation results. During the online operation of the controller, extract typical operating condition features based on long-term operating data, and adaptively correct the local parameters of the control model according to the degree of deviation between the current operating state and the typical operating condition features.

[0207] Step 6 includes:

[0208] Step 6.1: Based on the final control command, construct short-time domain performance indicators within the short-time domain evaluation window to characterize pressure response speed, flow tracking error, and the effect of regulating oscillation suppression.

[0209] Specifically, at the current control moment k, the final control command is output. Then, collect or predict short time-domain windows. The system analyzes the changes in pressure response, flow response, and control actions within the system, and constructs short-time domain performance indicators based on the pressure error, flow error, and control action changes at each sampling time. The short-time domain evaluation primarily measures the pressure response speed, flow tracking error, and regulation oscillation suppression effect to reflect the controller's transient adjustment capability in the face of rapid load changes and sudden changes in air demand.

[0210] Short-time performance metrics can be expressed as:

[0211]

[0212] in, This refers to the short time-domain evaluation window length; For pressure error; For flow rate error; u represents the change in control action between adjacent control cycles; , , These are the weighting coefficients for the pressure error term, flow error term, and control action change term, respectively.

[0213] In this application, the pressure error term is used to evaluate the ability of the pressure build-up process to track the target pressure, the flow error term is used to evaluate the ability of the air supply process to track the target air mass flow rate, and the control action variation term is used to evaluate the smoothness of the control action and the ability to suppress regulation oscillations. Through the above construction, the short-time domain performance indicators... It can comprehensively reflect the speed, accuracy and stability of the control model in short-term dynamic processes.

[0214] Step 6.2: Based on the final control command, construct long-term performance indicators within the long-term evaluation window to characterize compressor efficiency, power consumption level, stability margin maintenance capability, and long-term smoothness of actuator operation.

[0215] Specifically, in the final control command Under the influence of long time-domain windows The compressor efficiency, stability margin, and power consumption are collected or predicted, and long-term performance indicators are constructed. Long-term evaluation is mainly used to reflect the comprehensive performance of the control model under continuous operation conditions across a wide range of operating modes. Long-term performance indicators can be expressed as:

[0216]

[0217] in, The length of the long-term evaluation window; For compressor efficiency; This is a stability margin penalty term; P is the power consumption indicator. , , These are the weighting coefficients for the efficiency term, stability margin term, and power consumption term, respectively.

[0218] In this application, the efficiency term is used to evaluate the compressor's ability to maintain efficiency during long-term operation; the stability margin penalty term is used to evaluate the system's ability to safely maintain its operating point relative to the surge boundary or other stability boundaries; and the power consumption term is used to evaluate the energy consumption level of the motor drive and air supply system under control actions. Through the above construction, long-term performance indicators... It can comprehensively reflect the energy efficiency, safety, and continuous stability of the control model during long-term operation.

[0219] Step 6.3: Construct a comprehensive performance index based on the short-time domain performance index and the long-time domain performance index, and update the control model based on the comprehensive performance index.

[0220] Specifically, the comprehensive performance index can be expressed as:

[0221]

[0222] in, For short-time domain performance metrics; Long-term performance metrics; It is a dual-time-domain balance coefficient used to adjust the relative weights of short-time-domain evaluation and long-time-domain evaluation in the comprehensive performance index.

[0223] In this application, when the system operation emphasizes rapid pressure build-up and flow tracking, the influence of short-time domain performance indicators on the overall performance indicators can be appropriately increased; when the system operation emphasizes efficiency maintenance, stability margin, and long-term smooth adjustment, the dual-time domain balance coefficient can be adjusted. This improves the impact on long-term performance metrics. Consequently, it enhances the overall performance metrics. It can uniformly evaluate the control effect based on control objectives and operational requirements.

[0224] In obtaining comprehensive performance indicators Subsequently, the control model is updated based on comprehensive performance indicators. The update includes adjusting the mapping parameters in the candidate action generation module, the parameters related to mechanism consistency state encoding, and the local optimization parameters related to control decisions. This enables the updated control model to generate better candidate control actions in the next control cycle and achieve better short-time response performance and long-time operating performance while satisfying mechanism constraints.

[0225] Step 6.4: During the online operation of the controller, extract the operating condition prototype based on long-term operating data, and perform limited-amplitude adaptive correction of the local parameters of the control model according to the degree of deviation between the current operating state and the operating condition prototype.

[0226] After completing the dual-time-domain integrated evaluation in steps 6.1-6.3 and updating the control model, a self-evolving adjustment mechanism based on the operating condition prototype is further constructed. This mechanism is used to extract prototypes of typical operating conditions during the online operation of the controller, based on the operating condition distribution characteristics reflected in long-term operating data. Then, based on the degree of deviation between the current operating state and the operating condition prototype, it performs limited-amplitude adaptive corrections to the local parameters of the control model, thereby improving the control model's adaptability to operating condition transitions, environmental changes, actuator hysteresis, and component performance drift.

[0227] Specifically, during continuous operation of the controller, long-term operating data is continuously collected under different load variations, air demand conditions, motor speeds, and variable geometry adjustment states. This long-term operating data includes at least information such as pressure, temperature, air mass flow rate, motor speed, opening degree of each stage of variable geometry adjustment mechanism, actuator displacement, load current, load change rate, air excess coefficient, and dual-time-domain comprehensive evaluation results. Based on this long-term operating data, a typical operating condition distribution formed by the system during long-term operation is statistically analyzed, and multiple operating condition prototypes that can characterize representative operating areas are extracted from it.

[0228] In this application, operating condition prototypes are used to characterize a typical operating state and its corresponding dynamic characteristics, constraint boundary characteristics, and performance characteristics. That is, each operating condition prototype not only reflects the central operating state under a typical load mode, air demand level, and speed opening combination, but also reflects the typical requirements of the control model for pressure build-up, flow regulation, stability margin maintenance, and actuator smoothness under that operating condition. By constructing a set of operating condition prototypes, the relatively dispersed actual operating conditions during long-term operation can be summarized into several representative typical operating condition categories.

[0229] At the current control moment, the current operating state is acquired and compared with each operating condition prototype to calculate the degree of deviation between the current operating state and each prototype. The degree of deviation characterizes how close the current operating state is to the center of a typical operating condition, and can be determined based on multi-dimensional characteristics such as pressure, flow rate, motor speed, variable geometric opening, load change rate, air excess coefficient, and mechanistic constraints. By comparing the degrees of deviation, the target operating condition prototype closest to the current operating state is determined, or it is determined whether the current operating state has deviated from the typical operating area corresponding to an existing operating condition prototype.

[0230] When the deviation between the current operating state and a certain operating condition prototype is within a preset range, the current operating state is considered to belong to the typical operating condition category corresponding to the operating condition prototype, and the local parameters of the control model are specifically modified based on the operating condition prototype. When the deviation between the current operating state and the existing operating condition prototype exceeds the preset range, it is considered that the current operating state has undergone a significant migration or a new operating condition change trend has emerged. At this time, while keeping the main structure of the control model unchanged, the local parameters related to the current operating state can be adaptively modified with a limited amplitude to enhance the adaptability of the control model to the new operating conditions.

[0231] In this application, local parameters are state characterization parameters related to the current operating condition, parameters related to candidate control action generation, local weight parameters in dual-time-domain integrated evaluation, or local parameters related to actuator hysteresis compensation, environmental change compensation, and performance drift compensation. Limited-amplitude adaptive correction means that the adjustment amount of local parameters is limited by a preset correction range, thereby avoiding the impact on the overall stability of the control model due to excessive online adjustment. In this way, continuous adaptation to changes in local operating characteristics can be achieved without changing the main control structure and basic decision-making process of the control model.

[0232] Furthermore, the self-evolutionary adjustment mechanism of the operating condition prototypes can be used in conjunction with the dual-time-domain comprehensive evaluation mechanism constructed in steps 6.1-6.3. This means that the comprehensive performance evaluation results accumulated during long-term operation are used to continuously correct and update each operating condition prototype, enabling the set of operating condition prototypes to evolve continuously with the long-term operation of the system. Therefore, the control model can not only optimize based on the dual-time-domain comprehensive performance indicators within the current control cycle, but also adaptively correct local parameters based on changes in the operating condition prototypes over a longer time scale, thus forming an integrated intelligent control model for wide-condition operation of multi-stage centrifugal air compressors for fuel cells. Through step 6.4, the control model can maintain good dynamic response performance, long-term operating efficiency, and stability margin even when facing complex operating conditions such as load mode migration, ambient temperature changes, actuator hysteresis, and component performance drift.

[0233] In one specific embodiment, the control period is 50ms, and the short time domain evaluation window length is... =20, corresponding to a 1s evaluation time. The target outlet pressure is set at 190kPa, and the target air mass flow rate is 0.090kg / s. Under the final control command, the system outlet pressure increases from 182kPa to 189.2kPa in 0.6s, and the air mass flow rate increases from 0.086kg / s to 0.0893kg / s in 0.8s, with no significant back-and-forth fluctuations in the action changes between adjacent control cycles. Based on this, in the constructed short-time domain performance index, the following is taken: =0.5、 =0.3、 =0.2, and the short-time domain performance index corresponding to the current period is obtained through calculation. =0.084. This result indicates that the control action in this cycle achieves a good balance in terms of pressure build-up speed, flow tracking accuracy, and action smoothness.

[0234] Long-term evaluation window length =120, corresponding to a 6-second continuous operation interval. Within this interval, the following statistics were obtained: the compressor's average efficiency was 0.714, the minimum surge margin remained above 0.112, the average motor power consumption was 7.6kW, and no significant high-frequency oscillations or frequent reverse actions were observed in the first-stage inlet guide vanes and the second-stage diffuser throughout the entire interval. Based on this, the long-term performance indicators constructed were... =0.4、 =0.4、 =0.2, and the long-term performance index corresponding to the current period is calculated. =0.137. This result indicates that the control model maintains a high gas supply capacity without significantly sacrificing long-term efficiency and stability margin.

[0235] Considering the current operating conditions that require both rapid pressure build-up and avoidance of excessive long-term energy consumption, γ = 0.6 is chosen. The controller's comprehensive performance indicators are then further constructed as follows: 0.084 + 0.6 × 0.137 = 0.1662. Compare this comprehensive performance index with the comprehensive performance index corresponding to the control model in the previous stage. Comparing with 0.181, it can be seen that the overall performance has improved. Based on this evaluation result, the controller incrementally updates the control model. For example, it moderately increases the weight related to flow tracking deviation in the candidate action generation module, moderately strengthens the suppression term related to the surge margin approach state, and makes a small correction to the action change penalty term. This allows the candidate control actions generated in the next control cycle to maintain speed while further improving boundary stability and long-term operational smoothness.

[0236] The controller continuously records long-term operating data over the past 24 hours under different load variations, air demand conditions, motor speed conditions, and variable geometric adjustment states. It then extracts several typical operating condition prototypes, such as a low-speed steady-state air supply prototype, a medium-to-high-speed rapid load increase prototype, and a near-boundary protection prototype. After comparing the current operating state with each prototype, it is determined that the deviation from the "medium-to-high-speed rapid load increase prototype" is the smallest, with a deviation of 0.08, less than the preset threshold of 0.12. Therefore, the current operating state is considered to belong to the typical operating range corresponding to this prototype. Based on this, the controller makes limited adjustments to local parameters related to the current operating state. For example, it increases the inlet guide vane sensitivity parameter by 3%, the diffuser opening change suppression parameter by 2%, and corrects the actuator hysteresis compensation coefficient by 4%. The adjustment range for each parameter is limited to within a preset range of ±5% to avoid excessive online adjustments that could affect overall control stability. In this way, the control model can continuously adapt to load mode shifts, environmental changes, and actuator performance drift during long-term operation.

[0237] Step 7: Output the final control command based on the corrected control model, and send the final control command to the high-speed drive motor and each stage of variable geometry adjustment mechanism to control the operation of the multi-stage centrifugal air compressor.

[0238] Specifically, at the current control moment, the real-time operating status information of the multi-stage centrifugal air compressor air supply system is first collected. This real-time operating status information includes at least the following parameters: compressor inlet pressure, outlet pressure, inlet temperature, outlet temperature, air mass flow rate, actual motor speed, actual opening degree of each stage of variable geometric adjustment mechanism, actuator displacement, load current, load change rate, and air excess coefficient. This real-time operating status information is then input into the corrected control model. After mechanistic consistency state coding, candidate control action generation, control action feasible domain projection correction, and dual-time domain comprehensive evaluation correction, the final control command corresponding to the current control cycle is obtained.

[0239] In this application, the final control command includes at least a motor speed adjustment command and opening adjustment commands for each stage of the variable geometry adjustment mechanism. The motor speed adjustment command determines the target speed or target speed adjustment amount of the high-speed drive motor within the current control cycle; the opening adjustment commands for each stage of the variable geometry adjustment mechanism determine the target opening or target opening adjustment amount of the corresponding compression stage inlet guide vane, diffuser, or other variable geometry component. The final control command is the control result after being corrected for the feasible domain constraints of the control action; therefore, it simultaneously satisfies the surge safety boundary, actuator dynamic boundary, air supply matching boundary, and efficiency boundary requirements at the time of output.

[0240] Furthermore, the motor speed adjustment command is sent to the frequency converter driver, which adjusts the drive voltage, drive frequency, or drive current of the high-speed drive motor according to the motor speed adjustment command, so that the high-speed drive motor reaches the target speed or operates according to the target speed change pattern; the opening adjustment commands of each stage of the variable geometry adjustment mechanism are sent to the corresponding actuator drive unit to drive each stage of the variable geometry adjustment mechanism to operate at the target opening. Through the above method, the synchronous adjustment of the speed and the geometric opening of each stage of the multi-stage centrifugal air compressor is achieved.

[0241] In this application, after the final control command is sent, the multi-parameter synchronous acquisition system continues to collect real-time data on the operating status during the control execution process. This data is used to obtain the pressure response, flow response, motor speed response, variable geometry adjustment mechanism response, and the matching results between air supply and fuel cell stack load requirements under the control command. The response results are used to determine the effectiveness of the final control command execution within the current control cycle, and also serve as the basis for the control model input and online updates in the next control cycle.

[0242] Preferably, during the execution of the final control command, the control command can be periodically updated based on the actual response of the high-speed drive motor and the variable geometry adjustment mechanisms at each stage. That is, at the end of each control cycle, the corrected control model is re-inputted based on the latest collected system status to generate the final control command for the next control cycle, so that the multi-stage centrifugal air compressor can maintain continuous, stable and efficient operation under conditions of load changes, air demand changes and environmental boundary changes.

[0243] Step 7 allows the modified control model obtained in the preceding steps to be implemented at the actual execution level, enabling the high-speed drive motor and the variable geometry adjustment mechanisms at each level to work together under a unified control framework. This achieves rapid response, stable operation, and efficient air supply for the multi-stage centrifugal air compressor under a wide range of operating conditions.

[0244] In one specific embodiment, at the next control time k+1, the controller re-acquires the real-time operating status information of the system. After the execution of the control command in the previous cycle, the system status is updated as follows: the compressor outlet pressure increases to 188 kPa, the air mass flow rate increases to 0.089 kg / s, the actual motor speed is 55600 r / min, the actual opening of the first-stage inlet guide vane is 57.6%, the actual opening of the second-stage diffuser is 51.3%, and the air excess coefficient is 2.02. The system is still in the dynamic transition stage after the load increase. At this time, the above real-time status is input into the corrected control model. After mechanism consistency state encoding, candidate control action generation, control action feasible domain projection correction, and dual time domain evaluation correction, the new final control command for the current control cycle is output. The final control command for the current cycle output by the corrected control model is as follows: Therefore, the new execution objective is:

[0245]

[0246]

[0247]

[0248] The controller sends the aforementioned motor speed adjustment command to the frequency converter, which then adjusts the drive frequency and drive current of the high-speed drive motor according to the target speed command. Simultaneously, it sends the target opening adjustment commands for the first-stage inlet guide vanes and the second-stage diffuser to the corresponding actuator drive units, driving each stage of the variable geometry adjustment mechanism to operate at the target opening. Through this method, synchronous adjustment of the high-speed drive motor and each stage of the variable geometry adjustment mechanism is achieved.

[0249] The above descriptions are merely embodiments of this application, and common knowledge regarding specific structures and characteristics in the solutions is not described in detail here. It will be apparent to those skilled in the art that this application is not limited to the details of the above exemplary embodiments, and that this application can be implemented in other specific forms without departing from the spirit or essential characteristics of this application. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of this application is defined by the appended claims rather than the foregoing description. Therefore, it is intended that all variations falling within the meaning and scope of equivalents of the claims be included within this application. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A mechanism-constrained dual-time-domain self-evolutionary optimization control method for air compressors, characterized in that, The method includes: Step 1: Acquire multi-condition operating data that characterizes the compressor operating status, motor operating status, air pipeline operating status, and variable geometry adjustment status; Step 2: Based on the multi-condition operating data, establish and calibrate a joint simulation model of the compressor, motor, air piping and load coupling, and generate a sample dataset; Step 3: Construct a set of mechanistic constraints based on the sample dataset, and use the motor speed and the opening degree of each stage of the variable geometric adjustment mechanism as control variables to determine the feasible domain of the control action according to the set of mechanistic constraints; Step 4: Preprocess the sample dataset, and based on the preprocessed data, fuse and encode the current running state, historical state sequence and corresponding mechanism constraint information to obtain mechanism consistency state features; Step 5: Establish a control model based on the consistency state characteristics of the mechanism, and generate candidate control actions based on the control model, including the motor speed adjustment amount and the opening adjustment amount of each stage of variable geometric adjustment mechanism. Project the candidate control actions into the feasible domain of the control actions to obtain the current cycle control command. Step 6: Construct a comprehensive evaluation model combining short-time and long-time domains based on the current cycle control command, and update the control model according to the comprehensive evaluation results. During the online operation of the controller, extract typical operating condition features based on long-term operating data, and adaptively correct the local parameters of the control model according to the degree of deviation between the current operating state and the typical operating condition features. Step 7: Output the final control command based on the corrected control model, and send the final control command to the high-speed drive motor and each stage of variable geometry adjustment mechanism to control the operation of the multi-stage centrifugal air compressor.

2. The mechanism-constrained dual-time-domain self-evolutionary optimization control method for air compressors according to claim 1, characterized in that, Step 3 includes: Step 3.1: Based on the sample dataset obtained in Step 2, the samples are classified according to the motor speed range, variable geometric opening range, load change rate range and air excess coefficient range. In each classification range, the pressure build-up capability, flow regulation capability, surge margin, actuator dynamic boundary and air supply matching characteristics are extracted and characterized to establish the mechanism constraint characterization relationship under different operating conditions. Step 3.2: Based on the mechanism constraint characterization system constructed in Step 3.1, using the motor speed regulation amount and the opening regulation amount of each stage of variable geometric adjustment mechanism as control variables, establish the control action feasible domain that satisfies surge safety margin constraints, actuator dynamic constraints, air excess coefficient constraints and compressor efficiency constraints according to the mechanism constraint characterization system.

3. The mechanism-constrained dual-time-domain self-evolutionary optimization control method for air compressors according to claim 2, characterized in that, In step 3.2, for the current time k, a feasible region for control actions is established, which is jointly defined by the boundaries of multiple mechanisms. : ； in, This is the surge margin function; This represents the current state of the air compressor system. The control action corresponding to the current moment; In the current state and control actions Surge margin function value under the given conditions; For minimum safety margin; This is a function of compressor efficiency; This represents the motor speed adjustment amplitude within the current control cycle. This represents the maximum allowable adjustment range of the motor speed. This refers to the adjustment amplitude of the inlet guide vane opening within the current control cycle. This refers to the maximum allowable adjustment range of the inlet guide vane opening; This represents the amplitude of diffuser opening adjustment within the current control cycle. This represents the maximum allowable adjustment range of the diffuser opening. In the current state and control actions The function value of the excess air coefficient under action; and These are the lower and upper limits of the excess air coefficient, respectively. Current state and control actions The compressor efficiency function value under the action; The minimum allowable efficiency threshold; It should at least characterize one or more of the following: compressor inlet pressure, outlet pressure, inlet temperature, outlet temperature, air mass flow rate, actual motor speed, actual inlet guide vane opening, actual diffuser opening, actuator displacement, load current, load change rate, and air excess coefficient.

4. The mechanism-constrained dual-time-domain self-evolutionary optimization control method for air compressors according to claim 1, characterized in that, In step 4, the mechanism consistency feature vector is: ； in, This is the mechanism consistency feature vector at the current time k; Mechanism-consistent state encoding mapping; The parameters corresponding to the mechanism consistency state encoding mapping; Let L and L represent the state variables from time k-L+1 to time k, respectively, where L is the historical time sequence length used for state encoding. This refers to the mechanistic constraint information corresponding to the current moment.

5. The mechanism-constrained dual-time-domain self-evolutionary optimization control method for air compressors according to claim 1, characterized in that, Step 5 includes: Step 5.1: Based on the mechanism consistency state characteristics obtained in Step 4, candidate control actions containing motor speed adjustment and opening adjustment of each level of variable geometric adjustment mechanism are generated using the candidate action generation mapping; Candidate control actions are: ； in, Generate mappings for candidate actions; This indicates the initial control action that has not been modified by constraints; Step 5.2: Input the candidate control action into the feasible region of control action established in step 3, and perform projection correction on the candidate control action based on the feasible region projection operator to obtain the corrected control action that satisfies the mechanism constraint; The revised control action is as follows: ； in, For feasible region projection operators; The control action is after projection correction.

6. The mechanism-constrained dual-time-domain self-evolutionary optimization control method for air compressors according to claim 1, characterized in that, Step 6 includes: Step 6.1: Based on the final control command, construct short-time domain performance indicators within the short-time domain evaluation window to characterize pressure response speed, flow tracking error, and the effect of regulating oscillation suppression; Step 6.2: Based on the final control command, construct long-term performance indicators within the long-term evaluation window to characterize compressor efficiency, power consumption level, stability margin maintenance capability, and long-term smoothness of actuator operation; Step 6.3: Construct a comprehensive performance index based on the short-time domain performance index and the long-time domain performance index, and update the control model based on the comprehensive performance index; Step 6.4: During the online operation of the controller, extract the operating condition prototype based on long-term operating data, and perform limited-amplitude adaptive correction of the local parameters of the control model according to the degree of deviation between the current operating state and the operating condition prototype.

7. The mechanism-constrained dual-time-domain self-evolutionary optimization control method for air compressors according to claim 6, characterized in that, In step 6.4, the comprehensive performance indicators are as follows: ； ； ； in, For short-time domain performance metrics; This refers to the short time-domain evaluation window length; For pressure error; For flow rate error; u represents the change in control action between adjacent control cycles; , , These are the weighting coefficients for the pressure error term, flow error term, and control action change term, respectively. For long-term performance metrics; The length of the long-term evaluation window; For compressor efficiency; This is a stability margin penalty term; P is a power consumption indicator. , , These are the weighting coefficients for the efficiency term, stability margin term, and power consumption term, respectively. The coefficients are the dual-time-domain balance coefficients; k is the current control time. This is the state vector at the current moment; This refers to the revised control action.

8. The mechanism-constrained dual-time-domain self-evolutionary optimization control method for air compressors according to claim 1, characterized in that, Step 2 includes: Step 2.1: Based on the operational data collected in Step 1, establish a joint simulation model including the compressor sub-model, motor sub-model, actuator sub-model, air pipeline sub-model, buffer volume sub-model, and fuel cell stack load sub-model, and realize parameter transfer and boundary coupling between the sub-models through the data interface; Step 2.2: Assign and calibrate the compressor characteristic parameters, motor dynamic parameters, actuator response parameters, air pipeline parameters, buffer volume parameters, and fuel cell load requirement parameters of the co-simulation model based on the operating data, so that the steady-state output characteristics and dynamic response characteristics of the co-simulation model match the experimental data; Step 2.3: Based on the calibrated co-simulation model, simulation calculations are performed under different load change conditions, different air demand conditions, different motor speed conditions, and different variable geometric adjustment states. Combined with the experimental data obtained in Step 1, a sample dataset covering steady-state conditions, dynamic transition conditions, and boundary conditions is formed.

9. The mechanism-constrained dual-time-domain self-evolutionary optimization control method for air compressors according to claim 1, characterized in that, In step 1, the operating data includes at least the pressure, temperature and flow data that characterize the compressor's operating status, the speed, voltage, current and power data that characterize the motor's operating status, the air demand, back pressure and air supply matching error data that characterize the air supply status, and the opening degree, displacement and response delay data that characterize the variable geometry adjustment status.

10. The mechanism-constrained dual-time-domain self-evolutionary optimization control method for air compressors according to claim 1, characterized in that, In step 7, the final control command includes at least the motor speed adjustment command and the opening adjustment command of each level of variable geometry adjustment mechanism. The motor speed adjustment command is used to determine the target speed or target speed adjustment amount of the high-speed drive motor in the current control cycle. The opening adjustment command of each level of variable geometry adjustment mechanism is used to determine the target opening or target opening adjustment amount of the corresponding compression stage inlet guide vane, diffuser or other variable geometry component.

Images