Multivariable performance control method of TBCC engine based on neural network prediction feedback and inverse control

Abstract

The present invention discloses a TBCC engine multivariable performance control method based on neural network prediction feedback and inverse control, the control method includes: a multivariable performance parameter direct controller based on a neural network inverse system; a performance parameter estimator based on a BP neural network; using the multivariable performance intelligent controller, designing a certain type of TBCC engine closed-loop performance control system, and adding a PI compensator on this basis, online correction and update of controller parameters, and studying the tracking performance of the control system. The present invention uses a neural network combined with a derived TBCC multivariable control method, with the goal of minimizing the thrust flow change during the TBCC engine mode conversion process, designs a multivariable controller, calculates the engine control quantity, and controls the fuel flow, afterburner fuel flow and tail nozzle critical area. The control method has a simple structure and has a certain positive role in reducing the workload of the design controller and the design of the multivariable control law.

Description

Technical Field

[0001] The invention belongs to the field of aero-engine simulation and control, and in particular relates to a TBCC engine multivariable performance control method based on neural network prediction feedback and inverse control. Background Art

[0002] For the tandem turbine-based combined cycle engine, in order to meet its wide operating envelope, different operating modes need to be switched under different flight conditions. In order to ensure that the thrust fluctuation is small, the flow is continuous, and the total temperature of the turbine inlet does not exceed the temperature during the mode conversion process, the valve area MSV, the front adjustable area duct ejector FVABI, the rear adjustable area duct ejector RVABI, the main fuel flow Wf, the afterburner fuel flow Wfa, and the nozzle throat area A8 must be selected as the control quantity. It is inevitable to design the TBCC engine control system as a multivariable control system.

[0003] In view of the different working modes of TBCC, people often complete the control by designing different controllers for each mode. The control algorithm often uses multivariable ALQR and single variable PID control. Although this method meets the control requirements of different modes of TBCC engines, the control system is complex and the controller design process is cumbersome. In addition, common controllers use engine speed as the controlled variable, which may cause problems such as mismatch between thrust and speed due to engine failure and degradation.

[0004] The TBCC engine control method based on the neural network inverse control system can determine the engine's operating mode through the actual output of the engine, and the controller outputs the corresponding control quantity without completing the controller switching work; the unmeasurable performance parameters of the TBCC engine can be estimated through the performance parameter estimator based on the neural network; directly using the engine performance parameters as the controlled parameters can avoid changes in the corresponding relationship between thrust and speed and multi-engine mismatch caused by engine failure and degradation, and explore the engine performance potential. Summary of the invention

[0005] The technical problem to be solved by the present invention is to design a TBCC engine multivariable performance control method based on neural network predictive feedback and inverse control in response to the defects of the background technology. Simulation shows that the controller has a good effect on the multivariable control of the TBCC engine.

[0006] The present invention adopts the following technical solutions to solve the above technical problems:

[0007] A multivariable performance control method for a TBCC engine based on neural network prediction feedback and inverse control comprises the following steps:

[0008] Step A), through the neural network inverse system controller, the controller input is the command, including FGr, Wr, T4r, the controller output is the TBCC engine control quantity, and then the deviation between the engine performance parameters estimated by the performance parameter estimator and the input command is obtained by the PI compensator to obtain the control compensation amount, which is superimposed on the original control amount, so as to eliminate the steady-state error during command tracking and realize the direct control of the TBCC engine performance parameters;

[0009] Step B), offline establishing a performance parameter estimator based on a neural network, using the engine measurable parameters measured by sensors to estimate the engine unmeasurable parameters, and feeding back these unmeasurable performance parameters to the controller in real time for control adjustment, establishing a multivariable performance intelligent closed-loop control system for the TBCC engine, realizing the estimation and control of unmeasurable performance parameters, and verifying the modal conversion control performance of the controller;

[0010] Step C), when building the component-level model of the TBCC engine, the condition that the ram duct does not have reverse flow is used to establish the mathematical model of the ram duct, and the modal conversion working point is selected under the condition that the ram duct of the TBCC engine does not have reverse flow; in the modal conversion process, there are six control quantities, including the main fuel flow Wf, the boost fuel flow Wfa, the tail nozzle throat area A8, the mode selection valve area MSV, the front adjustable area duct ejector FVABI, and the rear adjustable area duct ejector RVABI. The three adjustable mechanisms of the TBCC engine, MSV, FVABI, and RVABI are selected for open-loop linear changes to reduce the number of closed-loop control quantities.

[0011] As a multivariable performance control method of a TBCC engine based on a neural network prediction feedback and inverse system of the present invention, the specific steps of step A) are as follows:

[0012] Step A1), establish the neural network inverse system offline, take the output of the TBCC engine as the input data of the neural network, and the input of the TBCC engine as the output data of the neural network. A large amount of data is obtained by running the engine model, and the neural network inverse system is trained and established. The neural network inverse system is expressed as:

[0013] Wf=f1(FGr,Wr,T4r)

[0014] Wfa=f2(FGr,Wr,T4r)

[0015] A8=f3(FGr,Wr,T4r)

[0016] The established neural network inverse system and the TBCC engine together form a pseudo-linear composite system, thereby completing the "linearization" and decoupling of the original nonlinear object;

[0017] Step A2), in order to eliminate the control error generated by the neural network inverse system controller, an additional controller, such as a PI compensator, is added to the linear subsystem after linear decoupling to further improve the stability and robustness of the control system. According to the error between the TBCC real-time performance parameter estimated by the performance parameter estimator and the instruction, the control compensation is calculated through the PI control algorithm. The control problem is expressed as follows:

[0018]

[0019] Where λ1 and λ2 are weight coefficients, FGr is the thrust command value, FG is the thrust response output by the engine, Wr is the flow command value, and W is the flow response output by the engine.

[0020] The PI compensator and the neural network inverse system together form the neural network inverse composite controller of the TBCC engine, which can obtain excellent static and dynamic characteristics and anti-interference ability;

[0021] Step A3), the control amount calculated by the neural network inverse system and the control compensation amount calculated by the PI compensator are superimposed to obtain the final control amount, and transmitted to the actuator.

[0022] As a multivariable performance control method of a TBCC engine based on neural network prediction feedback and inverse control of the present invention, the specific steps of step B) are as follows:

[0023] Step B1), offline training and establishing a performance parameter estimator of the TBCC engine, the neural network input data is the measurable parameters of the TBCC engine, including the flight altitude H, flight Mach number Ma, mode selection valve area MSV, main fuel flow Wf, afterburner fuel flow Wfa, tail nozzle throat area A8, low pressure rotor speed N of the TBCC engine. L and high pressure rotor speed N H The output data of the neural network are the unmeasurable performance parameters of the TBCC engine, including thrust FG, engine inlet flow W and turbine inlet total temperature T4. The TBCC model is run to obtain a large amount of data. The nonlinear function between the measurable parameters and the performance parameters is identified offline through the neural network. The state parameter estimator expression is:

[0024] FG=h1(H,Ma,MSV,Wf,Wfa,A8,N L ,N H )

[0025] W=h2(H,Ma,MSV,Wf,Wfa,A8,N L ,N H )

[0026] T4=h3(H,Ma,MSV,Wf,Wfa,A8,N L ,N H )

[0027] Step B2), inputting the measurable parameters of the engine collected by the sensor into the performance parameter estimator to obtain the unmeasurable performance parameters of the TBCC engine, and feeding the obtained performance parameters back to the PI compensator to form a closed-loop control system, and testing the control effect of the TBCC engine turbofan mode, ramjet mode and mode conversion process;

[0028] As a multivariable performance control method of a TBCC engine based on neural network prediction feedback and inverse control of the present invention, the specific steps of step C) are as follows:

[0029] Step C1), using the condition that the ram duct of the TBCC engine does not have a reverse flow, the static pressure at the ram duct outlet is equal to the static pressure at the fan duct outlet, that is:

[0030] P s,ramduct =P s,fanduct

[0031] According to the static pressure at the ram duct outlet, the ram duct flow rate can be calculated, that is:

[0032]

[0033] Where K is a constant, q(Ma) is the flow function, A is the ram duct outlet area, and T is the total temperature at the ram duct outlet.

[0034] Calculating the ram duct flow in this way reduces the number of common working equations compared to solving the flow balance of the ram duct outlet section as a common working condition. Secondly, compared with the control method that sets the non-reverse flow condition as a performance indicator and uses it as the controlled quantity, when designing the modal conversion working point, the conversion point is selected under the condition that the ram duct does not have reverse flow, and there is no need to consider the performance indicator of non-reverse flow, thereby reducing the number of controlled quantities.

[0035] The essence of the backflow in the ram duct is that the static pressure at the fan outlet is greater than the total pressure at the ram duct inlet. Therefore, the fan pressure ratio cannot be too large. The fan pressure ratio can be expressed as:

[0036]

[0037] Where P fanduct is the total pressure of the fan duct airflow, P s,fanduct is the static pressure of the fan duct airflow, κ is the gas specific heat ratio, and Ma is the airflow Mach number.

[0038] The static pressure at the fan duct outlet cannot be too large, and the airflow Mach number cannot be too small. In order to avoid reverse flow in the ram duct, the minimum Mach number of the fan duct airflow should be greater than 0.82 through CFD model calculation. When the TBCC engine mode conversion working point is selected under the condition that there is no reverse flow in the ram duct, the working point Mach number must be greater than a certain critical value.

[0039] In step C2), the TBCC engine realizes positive mode conversion, that is, converts from turbofan mode to ramjet mode. In order to reduce the number of closed-loop control quantities, characteristic parameters are selected, including MSV, FVABI, and RVABI, for open-loop control. Taking the positive mode conversion as an example, MSV and RVABI change linearly from the minimum state to the maximum state with FVABI.

[0040] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:

[0041] (1) The present invention proposes a TBCC engine multivariable performance control method based on neural network prediction feedback and inverse control, which estimates and directly controls the TBCC performance parameters, so that the engine control quantity directly corresponds to the performance parameters, and fully explores the engine performance potential;

[0042] (2) The present invention proposes a TBCC engine multivariable performance control method based on neural network prediction feedback and inverse control, which can provide control quantities in a timely manner by offline training of the neural network inverse system and performance parameter estimator; the neural network-based inverse system composite controller can provide corresponding inputs according to the engine output, without the need to judge the TBCC engine working mode, and has a simple control structure;

[0043] (3) The present invention proposes a TBCC engine multivariable performance control method based on neural network prediction feedback and inverse control. For the TBCC engine multi-input and multi-output system, three adjustable mechanisms are used as open-loop control to reduce the number of closed-loop control quantities. The neural network inverse system composite controller formed by PI compensation can eliminate the multivariable coupling problem. The multivariable performance intelligent closed-loop control system can eliminate steady-state tracking errors and improve control quality.

[0044] (4) The present invention proposes a multivariable performance control method for a TBCC engine based on neural network predictive feedback and inverse control. The method uses the condition that the TBCC engine ramjet duct does not have reverse flow to establish a ramjet duct mathematical model, thereby reducing the number of common working equations and designing the mode conversion working point under the condition that no reverse flow occurs. There is no need to consider the reverse flow situation during mode conversion control, thereby reducing the number of controlled quantities. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1It is a schematic diagram of the structure of the multivariable performance intelligent controller of the TBCC engine;

[0046] Figure 2 It is a curve chart showing the accuracy change during the training process of the performance parameter estimation neural network;

[0047] Figure 3 It is the simulation diagram of performance parameter control simulation and estimated value verification, (a) is the inlet total flow response and estimated value, (b) is the thrust response and estimated value, (c) is the turbine inlet total temperature response and estimated value, (d) is the performance parameter estimation error;

[0048] Figure 4 It is a schematic diagram of the structure of the components of a tandem turbine-based combined cycle engine;

[0049] Figure 5 It is the simulation diagram of the selected working point of the TBCC engine mode conversion, (a) is the change of the pressure of the TBCC key section with the Mach number, (b) is the change of the thrust performance with the Mach number in a single mode;

[0050] Figure 6 This is the step response diagram of the closed-loop control system of the TBCC engine turbofan mode, (a) is the thrust response, (b) is the flow response, (c) is the total temperature response before the turbine, (d) is the change of afterburner fuel flow, (e) is the change of nozzle throat area, and (f) is the change of main fuel flow;

[0051] Figure 7 It is the multivariable control response and control quantity change diagram of the TBCC engine mode conversion, (a) is the open-loop adjustment of the TBCC characteristic mechanism, (b) is the turbine inlet total temperature command and response, (c) is the thrust command and response, (d) is the inlet flow command and response, and (e) is the closed-loop control quantity change curve. DETAILED DESCRIPTION

[0052] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings.

[0053] The idea of ​​the present invention is to first establish an inverse system composite controller based on a BP neural network for the TBCC engine, a multi-input and multi-output nonlinear system, and establish a neural network inverse system to complete the "linearization" of the original system, eliminate the coupling effect of the multivariable system, add an additional PI compensator, form a feedback closed-loop control system, eliminate tracking errors, and improve control accuracy. In addition, the modal conversion working point is selected under the condition that the TBCC engine ram duct does not have reverse flow. Compared with describing the non-reverse flow as a limiting indicator, the number of controlled quantities is reduced, and the TBCC characteristic components are open-loop controlled to reduce the number of closed-loop control quantities. The TBCC engine multivariable performance intelligent control method based on neural network prediction feedback and inverse control has a simple control method structure, an easy-to-design controller, and a good control effect.

[0054] The specific implementation of the present invention takes the TBCC engine turbofan mode control and mode conversion process control as an example. The TBCC engine multivariable performance control method based on neural network prediction feedback and inverse control includes the following steps:

[0055] Step A), offline training and establishing a TBCC engine neural network inverse system, the TBCC engine control quantity can be obtained by the instruction, and a PI compensator is added to the neural network inverse system to eliminate the steady-state error;

[0056] Step B), offline training and establishing a performance parameter estimator based on a neural network, estimating the unmeasurable performance parameters of the TBCC engine using the measurable parameters measured by the sensor;

[0057] Step C), using the stamping duct no-backflow condition to establish a stamping duct mathematical model, and selecting the mode conversion working point under the stamping duct no-backflow condition;

[0058] Step D), open-loop regulation of TBCC engine characteristic components, including MSV, FVABI, RVABI, and closed-loop regulation of control quantities, including Wf, Wfa, A8, to achieve intelligent control of multivariable performance of turbofan mode and mode conversion.

[0059] The detailed steps of step A) are as follows:

[0060] Step A1), in order to fully train the neural network inverse system, run the TBCC component-level model to obtain training data. In the mode switching flight condition, [Wfa min ,Wfa max ] interval is divided into 100 segments, [A8 min ,A8 max ] interval is divided into 40 segments, and [Wf min ,Wf max ] The interval is divided into 20 segments, and a total of 80,000 sets of data are selected to record the corresponding original system output parameters and original system input parameters. A BP neural network is selected with three input layers, three output layers, two hidden layers, 10 nodes in each hidden layer, and a total of 20 nodes. The target error is 1×10 -6 , the training data is shuffled and normalized, 75% is selected as the training set and 15% as the test set for neural network training. If the test set does not meet the predetermined goal, the iteration is repeated until the training goal is reached.

[0061] Step A2), the mode conversion process neural network inverse system controller is as follows Figure 1 The inverse system of the neural network is shown as follows. The inverse system input y r The instructions include thrust instruction FGr, flow instruction Wr, turbine inlet total temperature instruction T4r, uNN is the reverse system output, engine input, including the main fuel flow Wf NN , afterburner fuel flow Wfa NN , nozzle throat area A8 NN , the controller output can be obtained by the instruction, and the output is:

[0062] Wf=f1(FGr,Wr,T4r)

[0063] Wfa=f2(FGr,Wr,T4r)

[0064] A8=f3(FGr,Wr,T4r)

[0065] The control quantity u obtained by the neural network inverse system NN Add the control quantity u obtained by PI compensator pid Get the control amount after compensation, and set u total It is transmitted to the engine actuator and then acts on the engine model to output the measurable parameter y m It is measured by the sensor, passed through the performance parameter estimator, and fed back to the PI compensator y um , and the command y r Form deviation, realize closed-loop control and eliminate steady-state error.

[0066] The objective function of the control is:

[0067]

[0068] Where λ1 and λ2 are weight coefficients, FGr is the thrust command value, FG is the thrust response output by the engine, Wr is the flow command value, and W is the flow response output by the engine.

[0069] The output increment of the PI compensator is:

[0070] Δu(k)=K P [e(k)-e(k-1)]+K I e(k)

[0071] In the formula, K P is the proportional gain coefficient, K I is the integral gain coefficient, and e is the tracking error.

[0072] The detailed steps of step B) are as follows:

[0073] Step B1), using the TBCC engine model to generate training data in the mode conversion working state according to the control range, the mode selection valve area change range is divided into 20 parts, the afterburner fuel flow and the main fuel flow are divided into 20 parts, and the nozzle throat area is divided into 10 parts. The TBCC engine model is run, and 80,000 sets of data are simulated in the mode conversion working mode as training samples for the neural network. 75% of the data are selected as the training set, 15% of the data as the test set, and 15% of the data as the verification set. The number of validation sample rationality is set to 6, that is, if the validation sample has not decreased for 6 consecutive iterations, the training is stopped, the maximum number of simulation iterations is set to 1000, and the target accuracy is set to 10 -5 , the accuracy change curve of BP network training process is as follows Figure 2 As shown in the figure, the horizontal axis is the number of iterations, and the vertical axis is the mean square error. After several iterations, the minimum mean square error is 4.7607×10 -5 , the correlation coefficient R is 0.99983, and there is no overfitting phenomenon, and the training converges.

[0074] By establishing a good performance parameter estimator, the TBCC engine performance parameters can be estimated:

[0075] FG=h1(H,Ma,MSV,Wf,Wfa,A8,N L ,N H )

[0076] W=h2(H,Ma,MSV,Wf,Wfa,A8,N L ,N H )

[0077] T4=h3(H,Ma,MSV,Wf,Wfa,A8,N L ,N H )

[0078] Figure 3 The simulation experiment diagram of the estimation error of the performance parameter estimator is shown. It can be seen from the figure that the estimation error of the total temperature before the turbine is less than 0.005, the thrust estimation error is less than 0.01, and the flow estimation error is less than 0.02.

[0079] Step B2), the sensor obtains the measurable parameters of the TBCC engine, including the flight altitude H, flight Mach number Ma, mode selection valve area MSV, main fuel flow Wf, afterburner fuel flow Wfa, tail nozzle throat area A8, low pressure rotor speed N of the TBCC engine. L and high pressure rotor speed N H , the thrust FG, inlet flow W and turbine inlet total temperature T4 are obtained through the performance parameter estimator.

[0080] The detailed steps of step C) are as follows:

[0081] The structure of a tandem turbine-based combined cycle engine is as follows: Figure 4 As shown, the ram duct is the section from C2 to C3, and the fan duct is the section from B2 to B3. In order to ensure that the airflow at the ram duct and fan duct outlets can be smoothly mixed without backflow, it is necessary to ensure that the static pressure at the two duct outlets is equal. Further analysis shows that the reverse flow in the ram duct is actually due to the excessive fan pressure ratio, which leads to excessive static pressure at the fan duct outlet. The fan pressure ratio expression is:

[0082]

[0083] Where P fanduct is the total pressure of the fan duct airflow, P s,fanduct is the static pressure at the fan duct airflow outlet, κ is the gas specific heat ratio, and Ma is the airflow Mach number.

[0084] It can be seen from the above formula that in order to reduce the static pressure at the fan outlet, the airflow Mach number needs to be increased, that is, ramjet duct backflow often occurs at low Mach numbers, so the modal conversion Mach number needs to be higher than a certain critical value. Figure 5 To select the simulation diagram of the mode conversion working point, Figure 5 From (a), it can be seen that when the Mach number is greater than 1.5, the total pressure at the inlet of the TBCC characteristic duct is greater than the static pressure at the outlet, and no backflow occurs. When the Mach number is greater than 1.8, the total pressure at the inlet of the ramjet duct is greater than the static pressure at the outlet of the common duct. At the same time, in order to give full play to the performance of the combined engine, Figure 5 From (b), we can see that after the Mach number is greater than 2.0, the performance of the turbofan mode gradually declines, so the mode conversion flight Mach number can be determined to be between 1.8 and 2.0.

[0085] In addition, when establishing the mathematical model of the ramjet duct of the TBCC engine, the condition that the ramjet duct does not have reverse flow can be used, that is:

[0086] P s,ramduct =P s,fanduct

[0087] The mathematical model of the ram duct is established, and the ram duct flow rate is calculated from the static pressure at the ram duct outlet, that is:

[0088]

[0089] Where K is a constant, q(Ma) is the flow function, A is the ram duct outlet area, and T is the total temperature at the ram duct outlet.

[0090] Calculating the ram duct flow in this way reduces the number of common working equations compared to solving the flow balance of the ram duct outlet section as a common working condition. Secondly, compared with the control method that sets the non-reverse flow condition as a performance indicator and uses it as the controlled quantity, when designing the modal conversion working point, the conversion point is selected under the condition that the ram duct does not have reverse flow, and there is no need to consider the performance indicator of non-reverse flow, thereby reducing the number of controlled quantities.

[0091] The detailed steps of step D) are as follows:

[0092] Step D1), in turbofan mode, the engine control quantities are the main combustion chamber fuel Wf, the afterburner fuel Wfa, and the nozzle throat area A8, and the controlled quantities are the engine thrust FG, the total engine flow W, and the total temperature before the turbine T4. Figure 6 The response of the TBCC engine turbofan mode control system under the step command is shown in Figure 2. At the 3rd and 16th seconds, a step of about 5% is added to the thrust, inlet flow, and total temperature before the turbine. It can be seen that the control response is rapid and the steady-state error is zero. Figure 6 Middle (d) to Figure 6 (f) is the control quantity change curve during the step response process. The control quantity, including the afterburner fuel flow, nozzle throat area and main fuel flow, is the control quantity pre-estimated by the neural network inverse system plus the compensation value of the PI compensator. Among them, the control quantity compensation value of the PI compensator is relatively small and is in a guiding position. The control quantity output by the neural network inverse system plays a leading role.

[0093] Step D2), during mode conversion, the three characteristic components of the TBCC engine are open-loop controlled, including MSV, FVABI and RVABI. The open-loop control rule is as follows: Figure 7 As shown in (a), at 2.5s, the mode conversion begins, and MSV, FVABI and RVABI are gradually opened according to the linear open-loop plan. The closed-loop control quantities are the main combustion chamber fuel Wf, the afterburner fuel Wfa, and the nozzle throat area A8. The controlled quantities are the engine thrust FG, the engine total flow W, and the total temperature before the turbine T4. During the conversion of the TBCC engine from turbofan mode to ramjet mode, the turbofan gradually enters the idling state. The main fuel flow is set to reach the design point idling fuel flow value, and the engine is considered to enter the idling state. Figure 7 (b) shows the slow-down command of the combined engine and its response. The total temperature of the turbine inlet drops from 1300K to 1000K. The combined engine turbine base enters the slow-down state and is ready to switch to the ramjet mode. From 0 to 2.5 seconds, the engine is in the turbofan mode. At this time, the mode selection valve is in the closed state. Figure 7 Middle (c), Figure 7As can be seen from (d), the combined engine thrust in the turbofan mode is maintained at 2900 daN, and the total engine flow is maintained at 75 kg / s. At 2.5 s, the mode conversion begins, and the MSV, FVABI, and RVABI are opened according to the linear open-loop plan. The total temperature command before the turbine is Figure 7 At 7.5s, the mode conversion ends and the turbine enters the slow state and waits for shutdown. The change of the control quantity is shown in Figure 7 As shown in (e), the regulation changes continuously. During the entire mode conversion, the engine thrust fluctuation is less than 2%, the inlet total flow fluctuation is less than 4%, and the total temperature before the turbine continues to decrease according to the command, meeting the actual application requirements.

[0094] It should be pointed out that the above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes and substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A multivariable performance control method for a TBCC engine based on neural network prediction feedback and inverse control, characterized in that: The following steps are involved: Step A), through a neural network inverse system controller, the controller input is the engine command, the controller output is the TBCC engine control quantity, and then the deviation between the engine output performance parameter and the input command is calculated, and the control compensation amount is obtained by the PI compensator; Step B), using BP neural network to identify the nonlinear function of the measurable parameters in the multivariable control law offline to estimate the unmeasurable parameters; using the identified neural network to estimate the unmeasurable performance parameters of the engine from the measurable parameters obtained by the sensor, and feeding back the performance parameters to establish a multivariable performance intelligent closed-loop control system for the TBCC engine, and verifying the modal conversion control performance of the controller; Step C), when building a component-level model of the TBCC engine, a ramjet duct mathematical model is established using the condition that the ramjet duct does not have reverse flow, and the mode conversion working point is selected under the condition that the ramjet duct of the TBCC engine does not have reverse flow; the TBCC engine changes from a turbofan mode to a ramjet mode; The specific steps of step A) are as follows: Step A1), first establish the TBCC engine inverse system, select the TBCC engine thrust FGr, the total inlet flow Wr, the turbine inlet total temperature T4r, that is, the output of the original system as the input of the neural network inverse system, select the main fuel flow Wf, the afterburner fuel flow Wfa, the tail nozzle throat area A8, that is, the control quantity of the original system as the output of the neural network inverse system, and establish the inverse system expression as follows: Wf=f1(FGr,Wr,T4r) Wfa=f2(FGr,Wr,T4r) A8=f3(FGr,Wr,T4r) Using the TBCC engine model to generate a large amount of corresponding data, the neural network inverse system of the TBCC engine multi-input and multi-output nonlinear continuous system is trained and established; Step A2), the engine control quantity obtained by the neural network inverse system is transmitted to the actuator and acts on the TBCC engine. The sensor measures the measurable parameters of the TBCC engine, including the flight altitude H, the flight Mach number Ma, the TBCC engine mode selection valve area MSV, the main fuel flow Wf, the afterburner fuel flow Wfa, the tail nozzle throat area A8, the low pressure rotor speed N L , high pressure rotor speed N H , the performance parameter estimator is established to estimate the performance parameter from the measurable parameter, and the deviation between it and the instruction is fed back to the PI compensator. The output increment of the PI compensator is: Δu(k)=K P [e(k)-e(k-1)]+K I e(k) In the formula, K P is the proportional gain coefficient, K I is the integral gain coefficient, e is the tracking error; The control compensation is obtained by adjusting the PI algorithm, and the control amount of the neural network inverse system is compensated to obtain the control amount, which is transmitted to the engine actuator to form a closed-loop feedback control system. The control goal is to reduce the deviation between the output of the control object and the instruction. The control problem is expressed as follows: The specific steps of step B) are as follows: Step B1), firstly, establish a TBCC engine performance parameter estimator based on BP neural network, select the flight altitude, flight Mach number, mode selection valve area, main fuel flow, afterburner fuel flow, tail nozzle throat area, low pressure rotor speed and high pressure rotor speed of the TBCC engine, which are measurable output parameters of the engine as inputs of the neural network; select the TBCC engine thrust, inlet total flow and turbine inlet total temperature, which are unmeasurable performance parameters, as outputs of the neural network; the performance parameter estimator expression is as follows: FG=h1(H,Ma,MSV,Wf,Wfa,A8,N L ,N H ) W=h2(H,Ma,MSV,Wf,Wfa,A8,N L ,N H ) <h2 style=";text-align:left;direction:ltr">T4 = h3(H,Ma,MSV,Wf,Wfa,A8,N<h2 style=";text-align:left;direction:ltr"> L <h2 style=";text-align:left;direction:ltr"> ,N<h2 style=";text-align:left;direction:ltr"> H <h2 style=";text-align:left;direction:ltr"> ) The TBCC engine model is used to generate a large number of corresponding measurable and unmeasurable parameters, and the nonlinear functions are identified offline through neural networks to establish the performance parameter estimator of the TBCC engine. Step B2), the estimated performance parameter is fed back to the PI compensator, the PI compensator obtains the compensation control amount from the deviation between the feedback value and the command, and the compensated TBCC engine control amount is input to the engine actuator to form a closed-loop control system; After the design of the TBCC engine multivariable performance intelligent controller is completed, the tracking performance of the controlled quantity in the TBCC engine turbofan mode, ramjet mode and mode conversion is tested to observe and verify the control effect of the controller; The specific steps of step C) are as follows: Step C1), the TBCC engine establishes a ram duct model taking into account the situation that no reverse flow occurs when the ram duct is started, and uses the condition that no reverse flow occurs, that is, the static pressure at the ram duct outlet is equal to the static pressure at the fan duct outlet, combined with the cross-sectional area of ​​the ram duct outlet, to solve the ram duct flow rate and reduce the number of common working equations; the TBCC engine mode conversion working point is selected under the condition that no reverse flow occurs in the ram duct; In step C2), the TBCC engine is converted from the ramjet mode to the turbofan mode, and the characteristic cross sections MSV, FVABI, and RVABI are open-loop controlled. In the process of converting from the turbofan mode to the ramjet mode, MSV and RVABI increase linearly and proportionally with FVABI.

Images