Optimal switching control method for aero-engine surge prevention

Abstract

This invention relates to an optimized switching control method for preventing surge in aero-engines, belonging to the fields of aero-engine control, active engine safety protection, model predictive control, and compressor surge prevention. Based on the engine's current state and relevant measurements or estimates of surge precursors, the method uses a surge prediction model to generate a predicted probability of surge precursors and switches between a traditional fuel-speed closed-loop control mode and an MPC risk control mode according to this probability. In the MPC risk control mode, the method constructs a progressive surge margin constraint in the MPC prediction time domain by online adjusting the surge margin safety boundary backoff amount based on the surge risk probability, thereby solving the model predictive control optimization problem and outputting fuel flow and compressor adjustable guide vane angle control quantities. This invention enables proactive intervention in the early stages of surge risk formation, reducing thrust loss caused by solely relying on bleed gas to reduce surge, and improving thrust holding capability while increasing surge margin.

Description

Technical Field

[0001] This invention belongs to the fields of aero-engine control, active engine safety protection, model predictive control and compressor surge prevention technology, and particularly relates to an optimized switching control method for aero-engine surge prevention. Background Technology

[0002] Aero engines undergo various complex operating conditions during flight, including takeoff, climb, cruise, deceleration, and descent. During high-altitude deceleration, rapid throttling, or sudden changes in flight conditions, dynamic mismatches may occur between intake airflow, compressor speed, pressure ratio, and fuel supply, causing the compressor operating point to gradually approach the surge boundary. Surge is a severe form of compressor aerodynamic instability; once it occurs, it can cause drastic thrust fluctuations, compressor flow separation, increased vibration of engine components, and even structural damage. Therefore, maintaining sufficient surge margin in the engine control system is a crucial requirement for ensuring safe operation.

[0003] Existing aero-engine control systems typically rely on fuel flow regulation and closed-loop speed control, supplemented by adjustable guide vanes, bleed valves, limiters, or control schedules for safety protection. This type of control offers advantages such as clear structure, mature engineering, and ease of calibration under normal operating conditions. However, its primary objectives usually revolve around speed tracking and actuator constraints, and the control laws often include a significant safety margin. While this approach can reduce surge risk to some extent, it still suffers from the following shortcomings: First, existing protection logic often uses fixed thresholds or limiters. When the surge risk increases, the fixed constraints may not be tightened enough; when the engine needs to quickly track the target speed while the risk is still controllable, the fixed constraints may be overly conservative, thus failing to fully utilize the engine's optimal operating efficiency.

[0004] Second, traditional control strategies are mostly passive protection methods. The control system only intervenes when pressure fluctuations, abnormal speeds, or abnormal pressure ratios are already quite obvious. At this point, the compressor may already be close to the instability boundary, and it may be necessary to sacrifice engine performance to ensure safe operation.

[0005] Third, the estimation and prediction results of airborne surge margin are affected by model errors, sensor noise, and changes in operating conditions. The original probabilities output by the risk prediction model may also lose their original meaning as the data distribution shifts. If a fixed surge margin backoff amount is used for a long time, it may result in insufficient protection when the risk increases, and unnecessary performance loss when the model is conservative or the risk is low.

[0006] Therefore, there is an urgent need for a closed-loop control method that can predict the probability of surge precursors and coordinate the switching optimization between safety and engine performance during the control process. Summary of the Invention

[0007] The purpose of this invention is to overcome the problems in existing technologies, such as insufficient utilization of surge precursor prediction in traditional fuel-speed closed-loop control, difficulty in adapting to model bias and operating condition drift with fixed surge margin boundaries, and difficulty in directly converting uncalibrated risk probabilities into control safety margins. This invention proposes an optimized switching control method for aero-engine surge prevention. Under normal operating conditions, this method maintains traditional speed closed-loop control, switching to model predictive optimization control when the surge risk probability continuously increases. It constructs a gradual surge margin constraint in the prediction time domain of model predictive control (MPC) by online correction of the safety boundary backoff, achieving synergistic optimization of safety and engine performance.

[0008] The technical solution of the present invention: An optimized switching control method for preventing surge in aero-engines comprises the following steps: Step A: Collect the current engine status, target speed, fuel flow rate, compressor adjustable guide vane angle, and surge precursor related measurements or estimates; Step B: Use the surge prediction model to output the original risk probability, and perform exponential weighted smoothing and trend estimation on the original risk probability to obtain the smoothed risk probability and the discrete estimate of the risk probability change trend. The surge prediction model adopts a stability discrimination model based on pressure ratio and flow characteristics, a prediction model based on pressure fluctuation characteristics, or a surge prediction model based on machine learning, neural networks, or time series classification models.

[0009] The original risk probability is directly output by the surge prediction model, in the current... Always remember Its value range is .

[0010] The smoothed risk probability and the discrete estimate of the risk probability change trend are respectively: , , in, The coefficients are the probability exponential weighted moving average coefficients. For the surge prediction model update cycle, For the present Smoothing risk probability at any given moment This is a discrete estimate of the corresponding probability change trend.

[0011] Step C: Select between the fuel-speed closed-loop control mode and the MPC risk control mode based on the switching conditions; The switching conditions are designed based on smoothed risk probabilities. The engine is in a safe operating range and the risk probability is smoothed. When the value is not higher than 0.5, a fuel-speed closed-loop control mode is adopted; when... If the triggering continues for a preset time, the system will switch to MPC risk control mode; in MPC risk control mode, the following conditions must be met simultaneously: Only when the minimum running time is reached and the relative speed error is within the allowable range will the system switch back to the fuel-speed closed-loop control mode, thereby avoiding frequent switching and ensuring continuous control.

[0012] Step D: Based on the actual occurrence of risk events in the exposure window and the preset risk frequency budget, adjust the surge margin boundary rollback amount online according to the slow update window; The slow update window contains at least one surge prediction model output, meaning the time length is not less than the surge prediction model's running cycle.

[0013] The risk event is used to indicate whether a risk state requiring increased boundary conservatism has occurred within the entire slow update window, not to indicate the specific duration or frequency of the risk event within the slow update window. For the first... Define window event flags for slow-updating windows. If the smoothing risk probability of at least one sample within the slow update window exceeds the probability threshold. If so, it is considered that a risk event has occurred in the slow update window, and then... , Otherwise .

[0014] The exposure window is used to screen valid calibration samples. Define the window exposure flag. If at least one sample exists within the exposure window that satisfies , Or a risk event has already occurred in the exposure window, i.e. Then let , Otherwise .

[0015] in, For surge margin estimation of airborne engine model, To minimize the allowable surge margin, This is the boundary backoff amount currently used for control constraints. Expose bandwidth at the boundary.

[0016] The preset risk frequency budget represents the target proportion of risk events allowed to occur across all exposure windows. When a risk event occurs in an exposure window, it indicates that the current boundary may be insufficiently conservative; when no risk event occurs in an exposure window, it indicates that the current boundary still has a safety margin in this edge-fitting operation. By continuously tracking the occurrence of risk events within each exposure window, the frequency of risk events in the edge-fitting exposure windows will eventually approach the preset budget.

[0017] The boundary rollback adjustment is updated based on the window event flags and a preset risk frequency budget: , in, This indicates a limit operation. A preset risk frequency budget is set to allow risk events to occur at the edge of the exposed window. For online calibration step size, , These are the lower and upper limits of the safe boundary backoff amount, respectively. Indicates the first The amount of boundary rollback for a slow-update window.

[0018] Step E: Add the latest boundary backoff amount in each MPC cycle, and construct the progressive surge margin boundary on the candidate prediction step; The progressive surge margin boundary comprises three parts: basic surge margin, adaptive boundary backoff, and trend preview term. For the candidate prediction time domain, the... Step 1: Define the trend preview item. , in, For adjustment coefficients, To produce a limiting function that generates a non-negative output, This is the engine control cycle. Therefore, the first... The surge margin boundary used in step prediction is expressed as: , .

[0019] Step F: Treat boundary violations as soft constraints and dynamically adjust the thrust reward weights according to the severity of the safety constraint violations; The boundary violation quantity is in the form of: , in, For the first Surge margin estimation of step-predicted output This represents the maximum boundary violation. When... When the boundary conditions are met, the soft constraint penalty term is not activated; conversely, the boundary violation amount increases as the deviation between the two increases.

[0020] The severity of the violation of the security constraint is defined as follows: , in, .

[0021] When the severity of a safety constraint violation is close to zero, it indicates that the system is in a safe state or has only a slight boundary deviation. At this time, the thrust reward weight remains at the basic level. When the severity of a safety constraint violation increases, it indicates that the system is gradually approaching or entering a higher surge risk region. The controller reduces the thrust reward weight accordingly, thereby increasing the penalty for surge risk behavior.

[0022] Step G: Construct a model to predict the control optimization problem; The optimization objective of the model predictive control optimization problem is expressed as: , in, To control the length of the time domain, express The control variables need to be optimized at all times. Objective function. This includes high-voltage rotor speed tracking, control smoothing, surge margin boundary penalty, and thrust reward.

[0023] The solution to the model predictive control optimization problem needs to satisfy engine operating constraints and actuator constraints, including: , , in, This indicates the minimum fuel control quantity. This indicates the maximum value of the fuel control quantity. No. Fuel control amount at any time. This indicates the minimum value of the guide vane angle control. This indicates the maximum value of the guide vane angle control value. Indicates the first The guide vane angle control value at any given time. This indicates the maximum rate of change of the fuel control quantity within each control cycle. Indicates the number of moments relative to the previous moment. The magnitude of change in fuel control quantity at any given time. This represents the maximum rate of change of the guide vane angle control quantity within each control cycle. Indicates the number of moments relative to the previous moment. The range of change in the guide vane angle control quantity at any given time.

[0024] Step H: Solve the model predictive control optimization problem to output the control variables, and repeat steps A to G in the next control cycle.

[0025] The model predictive control optimization problem is solved using a non-gradient optimization algorithm to obtain the optimal control sequence in the future prediction time domain: , In actual execution, only the first optimized control variable is taken, and amplitude limiting and rate limiting are applied to the engine in sequence: .

[0026] Furthermore, in step D, when and At that time, there were: , Increased boundary retreat This causes the subsequent surge margin boundary to retreat in a more conservative direction; when but At that time, there were: , Boundary retreat amount decreased This allows the boundary to gradually release unnecessary conservative margins while remaining securely exposed.

[0027] Furthermore, the objective function in step G is expressed as: , in, To predict the length of the time domain, For the first The high-voltage rotor speed at any given time, The target speed of the high-voltage rotor. , and These correspond to the weighting coefficients for speed tracking, fuel variation, compressor guide vane angle variation, surge margin constraint, and thrust, respectively. Indicates the first The engine model outputs thrust at any given moment.

[0028] Furthermore, to avoid the impact of mode switching, when entering the MPC risk control mode, the closed-loop output of the fuel-speed closed-loop control mode is used as the initial value of the MPC control sequence; when exiting the MPC risk control mode, the integral term of the closed-loop controller of the fuel-speed closed-loop control mode is back-calculated and initialized based on the current control output. Different values ​​are used for the entry and exit thresholds, with the exit threshold being higher than the entry threshold, to avoid frequent switching of control modes near the boundaries.

[0029] Compared with existing technical solutions, the present invention has the following advantages: (1) It can actively switch to MPC optimization control during the surge precursor stage when the risk probability continuously exceeds the threshold, thereby performing multi-variable coordinated intervention before instability occurs and improving the engine's active safety protection capability.

[0030] (2) By using the exposure window close to the red line to adjust the surge margin boundary back-off amount online according to the risk budget, the safety boundary can be adaptively tightened or backed off with the recent risk event rate, model deviation and operating condition changes, reducing the over-conservatism or insufficient protection caused by the fixed conservative margin.

[0031] This method can be implemented as an upper-level switching optimization module in existing full-authority digital electronic control systems. It is suitable for complex operating conditions such as high-altitude deceleration, rapid throttling, mismatch between intake flow and speed, and abnormal pressure ratio changes. It can improve the stability, safety and overall performance of aero engines in transient processes. Attached Figure Description

[0032] Figure 1 This is a system block diagram of the aero-engine switching optimization closed-loop control method of the present invention.

[0033] Figure 2 This is a schematic diagram of the state machine switching between the conventional closed-loop mode and the MPC risk control mode of this invention.

[0034] Figure 3 These are simulation curves of traditional anti-surge methods for aero-engines.

[0035] Figure 4 These are simulation curves of the optimized switching closed-loop control method for aero-engines according to the present invention. Detailed Implementation

[0036] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings and technical solutions.

[0037] See Figure 1 The optimized switching control system for preventing surge in aero-engines according to embodiments of the present invention includes an engine status acquisition module, a conventional fuel-speed closed-loop control module, an engine prediction model module, a surge risk prediction module, a risk probability smoothing and trend estimation module, a surge boundary online correction module, a switching control module, an MPC optimization control module, and a control quantity limiting and speed limiting output module.

[0038] The engine status acquisition module obtains current engine speed, target engine speed, fuel flow rate, compressor adjustable guide vane angle, thrust estimate, and onboard surge margin estimate. The surge risk prediction module outputs the raw risk probability every 100ms; the risk probability smoothing and trend estimation module generates a smoothed risk probability. and the trend of risk probability changes The boundary online correction module updates the rollback amount every 0.5 seconds based on risk events and edge exposure windows. The MPC optimization module follows... Construct a progressive surge margin boundary and output the first variable of the optimal sequence after passing it through a limiting and speed limiting module.

[0039] Step A: Collect the current engine status, target speed, fuel flow rate, compressor adjustable guide vane angle, and surge precursor related measurements or estimates; Step B: Use the surge prediction model to generate the original risk probability with an update period of 100ms, and form a smooth risk probability and a discrete estimate of the risk probability change trend; In specific implementation, the surge prediction model adopts a stability discrimination model based on pressure ratio and flow characteristics, a prediction model based on pressure fluctuation characteristics, or a surge prediction model based on machine learning, neural networks, or time series classification models.

[0040] The original risk probability is directly output by the surge prediction model, in the current... Always remember Its value range is .

[0041] In specific implementation, the smoothed risk probability and the discrete estimation of the risk probability change trend are respectively: , , in, The coefficients are the probability exponential weighted moving average coefficients. To predict the module update cycle, For the present Smoothing risk probability at any given moment This is a discrete estimate of the corresponding probability change trend.

[0042] In practice , .

[0043] Step C: Select between the fuel-speed closed-loop control mode and the MPC risk control mode based on the switching conditions; In practice, the switching conditions are designed based on smoothed risk probabilities. The engine is in a safe operating range and the risk probability is smoothed. When the value is not higher than 0.5, a fuel-speed closed-loop control mode is adopted; when... Furthermore, if the triggering occurs consecutively for three prediction cycles, the system switches to MPC risk control mode; under MPC risk control mode, the following conditions must be met simultaneously: The total running time is not less than 2 seconds, and the relative speed error is... The system maintains the value within the 0.02 range for at least 1 second before switching back to the fuel-speed closed-loop control mode, thereby avoiding frequent switching and ensuring continuous control.

[0044] Step D: Based on the actual occurrence of risk events in the exposure window and the preset risk frequency budget, adjust the surge margin boundary rollback amount online according to the slow update window; In specific implementation, the slow update window contains at least one surge prediction model output, that is, the time length is not less than the surge prediction model running cycle, which is set to be the same as the surge prediction model running cycle, i.e., 100ms.

[0045] In specific implementation, the risk event is used to indicate whether a risk state requiring increased boundary conservatism has occurred within the entire slow update window, rather than indicating the specific duration or number of occurrences of the risk event within the slow update window. For the first... Define window event flags for slow-updating windows. If the smoothing risk probability of at least one sample within the slow update window exceeds the probability threshold. If so, it is considered that a risk event has occurred in the slow update window, and then... , Otherwise .

[0046] In practice, the exposure window is used to screen valid calibration samples. A window exposure flag is defined. If at least one sample exists within the exposure window that satisfies , Or a risk event has already occurred in the exposure window, i.e. Then let , Otherwise .

[0047] in, For surge margin estimation of airborne engine model, To minimize the allowable surge margin, This is the boundary backoff amount currently used for control constraints. Expose bandwidth at the boundary.

[0048] In practice , .

[0049] In practice, the preset risk frequency budget represents the target proportion of risk events allowed to occur across all exposure windows. When a risk event occurs in an exposure window, it indicates that the current boundary may be insufficiently conservative; when no risk event occurs in an exposure window, it indicates that the current boundary still has a safety margin in this edge-fitting operation. By continuously tracking the occurrence of risk events within each exposure window, the frequency of risk events in the edge-fitting exposure windows will eventually approach the preset budget.

[0050] In practice, the boundary backoff amount correction is updated based on the window event flag and the preset risk frequency budget: , in, This indicates a limit operation. A preset risk frequency budget is set to allow risk events to occur in the window that is exposed at the edge. For online calibration step size, , These are the lower and upper limits of the safe boundary backoff amount, respectively. Indicates the first The amount of boundary rollback for a slow-update window.

[0051] In specific implementation, when and At that time, there were: , Increased boundary retreat This causes the subsequent surge margin boundary to retreat in a more conservative direction; when but At that time, there were: , Boundary retreat amount decreased This allows the boundary to gradually release unnecessary conservative margins while remaining securely exposed.

[0052] In practice, the initial value of the boundary backoff amount During the update process , , , The units for values ​​related to surge margin are all percentage points.

[0053] Step E: Add the latest boundary backoff amount in each MPC cycle, and construct the progressive surge margin boundary on the candidate prediction step; In specific implementation, the progressive surge margin boundary includes three parts: basic surge margin, adaptive boundary backoff, and trend preview term. For the candidate prediction time domain, the... Step 1: Define the trend preview item. , in, For adjustment coefficients, To produce a limiting function that generates a non-negative output, This is the engine control cycle. Therefore, the first... The surge margin boundary used in step prediction is expressed as: , .

[0054] In practice , .

[0055] Step F: Treat the boundary violation as a soft constraint and dynamically adjust the thrust reward weight according to the severity of the violation; In specific implementation, the boundary violation quantity takes the form of: , in, For the first Surge margin estimation of step-predicted output This represents the maximum boundary violation. When... When the boundary conditions are met, the soft constraint penalty term is not activated; conversely, the boundary violation amount increases as the deviation between the two increases.

[0056] In specific implementation, the severity of the violation of the security constraint is defined as follows: , in, When the severity of safety constraint violation is close to zero, it indicates that the system is in a safe state or has only slight boundary deviations, and the thrust reward weight remains at the basic level. When the severity of safety constraint violation increases, it indicates that the system is gradually approaching or entering a higher surge risk region, and the controller reduces the thrust reward weight accordingly, thereby increasing the penalty for surge risk behavior.

[0057] In practice That is, assuming the surge margin estimate is below the boundary by 5, which is the worst case where thrust is no longer rewarded and surge risk is fully penalized.

[0058] Step G: Construct a model to predict the control optimization problem; The optimization objective of the model predictive control optimization problem is expressed as: , in, To control the length of the time domain, express The control variables need to be optimized at all times. Objective function. This includes high-voltage rotor speed tracking, control smoothing, surge margin boundary penalty, and thrust reward.

[0059] In practice, the objective function is expressed as: , in, To predict the length of the time domain, For the first The high-voltage rotor speed at any given time, The target speed of the high-voltage rotor. , and These correspond to the weighting coefficients for speed tracking, fuel variation, compressor guide vane angle variation, surge margin constraint, and thrust, respectively. Indicates the first The engine model outputs thrust at any given moment.

[0060] In practice , and Set them to 4000, 50, 5, 1000 and 30 respectively.

[0061] In specific implementation, the solution process for the model predictive control optimization problem needs to satisfy engine operating constraints and actuator constraints, including: , , in, This indicates the minimum fuel control quantity. This indicates the maximum value of the fuel control quantity. No. Fuel control amount at any time. This indicates the minimum value of the guide vane angle control. This indicates the maximum value of the guide vane angle control value. Indicates the first The guide vane angle control value at any given time. This indicates the maximum rate of change of the fuel control quantity within each control cycle. Indicates the number of moments relative to the previous moment. The magnitude of change in fuel control quantity at any given time. This represents the maximum rate of change of the guide vane angle control quantity within each control cycle. Indicates the number of moments relative to the previous moment. The range of change in the guide vane angle control quantity at any given time.

[0062] Step H: Solve the model predictive control optimization problem to output the control variables, and repeat steps A to G in the next control cycle.

[0063] In practice, the model predictive control optimization problem is solved using a non-gradient optimization algorithm to obtain the optimal control sequence in the future prediction time domain. , In actual execution, only the first optimized control variable is taken, and amplitude limiting and rate limiting are applied to the engine in sequence: .

[0064] In practice, to avoid the impact of mode switching, when entering the MPC risk control mode, the traditional closed-loop output of the fuel-speed closed-loop control mode is used as the initial value of the MPC control sequence; when exiting the MPC risk control mode, the integral term of the traditional closed-loop controller of the fuel-speed closed-loop control mode controller is back-calculated and initialized based on the current control output. Different values ​​are used for the entry and exit thresholds, with the exit threshold being higher than the entry threshold, to avoid frequent switching of control modes near the boundary.

[0065] In practical implementation, under the same Mach number and speed variation conditions, a comparative simulation was conducted between the traditional bleed-out surge reduction method and the optimized switching closed-loop control method of this invention. The traditional bleed-out surge reduction method, upon detecting surge risk, reduces the compressor pressure ratio and load by opening the compressor bleed valve, thus moving the operating point away from the surge boundary. This method has simple control logic and direct protection, but bleed-out reduces the effective mainstream mass flow rate into the combustion chamber and turbine, typically accompanied by a decrease in thrust, reduced fuel efficiency, and a slower recovery process. Figure 3 As shown.

[0066] This invention utilizes online adaptive adjustment of surge margin boundary constraints based on risk probability, proactively adjusting fuel flow and compressor adjustable guide vane angles before the risk develops into significant surge. Both methods can prevent the compressor from entering a surge instability state; however, traditional bleed-out surge reduction methods primarily rely on sacrificing effective airflow in the gas path to obtain a safety margin, resulting in faster surge margin recovery but more significant thrust loss. This invention's method, through early intervention and multivariate coordinated control, achieves the following control results: Figure 4 As shown, the minimum surge margin remained consistently between 10 and 15, while the thrust increased from approximately 1121 kg to approximately 1217 kg, an increase of approximately 8.6%. These results demonstrate that the present invention can improve both surge safety margin and thrust retention capability without significantly increasing control fluctuations.

[0067] In summary, this invention proposes an optimized switching control method for preventing surge in aero-engines. This method takes the original risk probability output in a 100ms cycle as input, forms a smoothed risk probability and its trend, and within a slow update window, corrects the surge margin boundary backoff amount online based on the risk budget and edge exposure samples. In the MPC (Multi-Level Control), the backoff amount is combined with the preview term to construct progressive boundary constraints, achieving coordinated optimization of fuel flow and compressor adjustable guide vane angle. Compared with traditional bleed-out surge prevention methods, this invention can proactively intervene in the early stages of surge risk formation, reducing thrust loss caused by solely relying on bleed-out, and improving thrust holding capability while increasing the minimum surge margin. Therefore, this invention can be applied as an upper-level optimized switching control module in aero-engine full-authority digital electronic control systems, possessing good engineering applicability and promotional value.

Claims

1. An optimized switching control method for preventing surge in aero-engines, characterized in that, The steps are as follows: Step A: Collect the current engine status, target speed, fuel flow rate, compressor adjustable guide vane angle, and surge precursor related measurements or estimates; Step B: Use the surge prediction model to output the original risk probability, and perform exponential weighted smoothing and trend estimation on the original risk probability to obtain the smoothed risk probability and the discrete estimate of the risk probability change trend. Step C: Select between the fuel-speed closed-loop control mode and the MPC risk control mode based on the switching conditions; Step D: Based on the actual occurrence of risk events in the exposure window and the preset risk frequency budget, adjust the surge margin boundary rollback amount online according to the slow update window; Step E: Add the latest boundary backoff amount in each MPC cycle, and construct the progressive surge margin boundary on the candidate prediction step; Step F: Treat boundary violations as soft constraints and dynamically adjust the thrust reward weights according to the severity of the safety constraint violations; Step G: Construct a model to predict the control optimization problem; Step H: Solve the model predictive control optimization problem to output the control variables, and repeat steps A to G in the next control cycle.

2. The optimized switching control method for preventing surge in aero-engines according to claim 1, characterized in that, In step B: The surge prediction model adopts a stability discrimination model based on pressure ratio and flow characteristics, a prediction model based on pressure fluctuation characteristics, or a surge prediction model based on machine learning, neural networks, and time series classification models. The original risk probability is directly output by the surge prediction model, in the current... Always remember Its value range is ; The smoothed risk probability and the discrete estimate of the risk probability change trend are respectively: ， ， in, The coefficients are the probability exponential weighted moving average coefficients. For the surge prediction model update cycle, For the present Smoothing risk probability at any given moment This is a discrete estimate of the corresponding probability change trend.

3. The optimized switching control method for surge prevention of aero-engines according to claim 1, characterized in that, In step C: The switching conditions are designed based on smoothed risk probabilities; the engine is in a safe operating range and the risk probability is smoothed. When the value is not higher than 0.5, a fuel-speed closed-loop control mode is adopted; when... If the triggering continues for a preset time, the system will switch to MPC risk control mode; in MPC risk control mode, the following conditions must be met simultaneously: Only when the minimum running time is reached and the relative speed error is within the allowable range will the system switch back to the fuel-speed closed-loop control mode, thereby avoiding frequent switching and ensuring continuous control.

4. The optimized switching control method for preventing surge in aero-engines according to claim 1, characterized in that, In step D: The slow update window contains at least one surge prediction model output, i.e., the time length is not less than the surge prediction model running cycle; The risk event is used to indicate whether a risk state requiring increased boundary conservatism has occurred within the entire slow update window, not to indicate the specific duration or number of occurrences of the risk event within the slow update window; for the first Define window event flags for slow-updating windows. If the smoothing risk probability of at least one sample within the slow update window exceeds the probability threshold. If so, it is considered that a risk event has occurred in the slow update window, and then... ， Otherwise ； The exposure window is used to screen valid calibration samples; a window exposure flag is defined. If at least one sample exists within the exposure window that satisfies ， Or a risk event has already occurred in the exposure window, i.e. Then let ， Otherwise ； in, For surge margin estimation of airborne engine model, To minimize the allowable surge margin, This is the boundary backoff amount currently used for control constraints. Exposing bandwidth at the boundary; The preset risk frequency budget is used to represent the target proportion of risk events allowed to occur in all exposure windows. When a risk event occurs in an exposure window, it indicates that the current boundary may be insufficiently conservative. When no risk event occurs in an exposure window, it indicates that the current boundary still has a safety margin in this edge-fitting operation. By continuously statistically analyzing the occurrence of risk events in each exposure window, the frequency of risk events in the edge-fitting exposure windows will eventually gradually approach the preset budget. The boundary rollback adjustment is updated based on the window event flags and a preset risk frequency budget: ， in, This indicates a limit operation. A preset risk frequency budget is set to allow risk events to occur at the edge of the exposed window. For online calibration step size, , These are the lower and upper limits of the safe boundary backoff amount, respectively. Indicates the first The amount of boundary rollback for a slow-update window.

5. The optimized switching control method for preventing surge in aero-engines according to claim 1, characterized in that, In step E: The progressive surge margin boundary comprises three parts: basic surge margin, adaptive boundary backoff, and trend preview term; for the candidate prediction time domain, the first... Step 1: Define the trend preview item. ， in, For adjustment coefficients, To produce a limiting function that generates a non-negative output, For the engine control cycle; therefore, the first The surge margin boundary used in step prediction is expressed as: ， 。 6. The optimized switching control method for preventing surge in aero-engines according to claim 1, characterized in that, In step F: The boundary violation quantity is in the form of: ， in, For the first Surge margin estimation of step-predicted output For the maximum boundary violation; when When the boundary conditions are met, the soft constraint penalty term is not activated; conversely, the boundary violation amount increases as the deviation between the two increases. The severity of the violation of the security constraint is defined as follows: ， in, ; When the severity of a safety constraint violation is close to zero, it indicates that the system is in a safe state or has only a slight boundary deviation. At this time, the thrust reward weight remains at the basic level. When the severity of a safety constraint violation increases, it indicates that the system is gradually approaching or entering a higher surge risk region. The controller reduces the thrust reward weight accordingly, thereby increasing the penalty for surge risk behavior.

7. The optimized switching control method for surge prevention of aero-engines according to claim 1, characterized in that, In step G: The optimization objective of the model predictive control optimization problem is expressed as: ， in, To control the length of the time domain, express Control variables to be optimized at any given time; objective function This includes high-voltage rotor speed tracking, control smoothing, surge margin boundary penalty, and thrust reward. The solution to the model predictive control optimization problem needs to satisfy engine operating constraints and actuator constraints, including: ， ， in, This indicates the minimum fuel control quantity. This indicates the maximum value of the fuel control quantity. No. Fuel control amount at any time. This indicates the minimum value of the guide vane angle control. This indicates the maximum value of the guide vane angle control value. Indicates the first The guide vane angle control value at any given time. This indicates the maximum rate of change of the fuel control quantity within each control cycle. Indicates the number of moments relative to the previous moment. The magnitude of change in fuel control quantity at any given time. This represents the maximum rate of change of the guide vane angle control quantity within each control cycle. Indicates the number of moments relative to the previous moment. The range of change in the guide vane angle control quantity at any given time.

8. The optimized switching control method for preventing surge in aero-engines according to claim 1, characterized in that, In step H: The model predictive control optimization problem is solved using a non-gradient optimization algorithm to obtain the optimal control sequence in the future prediction time domain: ， In actual execution, only the first optimized control variable is taken, and amplitude limiting and rate limiting are applied to the engine in sequence: ； Furthermore, in step D, when and At that time, there were: ， Increased boundary retreat This causes the subsequent surge margin boundary to retreat in a more conservative direction; when but At that time, there were: ， Boundary retreat amount decreased This allows the boundary to gradually release unnecessary conservative margins while remaining securely exposed.

9. The optimized switching control method for preventing surge in aero-engines according to claim 7, characterized in that, The objective function in step G is expressed as: ， in, To predict the length of the time domain, For the first The high-voltage rotor speed at any given time, The target speed of the high-voltage rotor. , and These correspond to the weighting coefficients for speed tracking, fuel variation, compressor guide vane angle variation, surge margin constraint, and thrust, respectively. Indicates the first The engine model outputs thrust at any given moment.

10. The optimized switching control method for preventing surge in aero-engines according to claim 1, characterized in that, To avoid the impact of mode switching, when entering the MPC risk control mode, the closed-loop output of the fuel-speed closed-loop control mode is used as the initial value of the MPC control sequence; when exiting the MPC risk control mode, the integral term of the closed-loop controller of the fuel-speed closed-loop control mode is back-calculated and initialized based on the current control output; different values ​​are used for the entry threshold and the exit threshold, and the exit threshold is higher than the entry threshold, in order to avoid frequent switching of control modes near the boundary.

Images