Gas flow regulation system model correction method based on NARX and ELM neural network

Abstract

The invention provides a gas flow regulation system model correction method based on an NARX and ELM neural network, and the method comprises the steps: S1, constructing a gas flow regulation system mechanism model of a solid ramjet, and selecting to-be-compensated parameters of a steady-state model and a dynamic model; s2, based on an NARX neural network, designing a correction method of a solid ramjet gas flow steady-state regulation model, and based on experimental data, performing steady-state model parameter correction; s3, on the basis of the integrated ELM neural network, designing a correction method of the solid ramjet gas flow dynamic adjustment model, and performing dynamic model parameter correction on the basis of experimental data; and S4, in combination with the steps S1 to S3, digital simulation of the solid ramjet gas flow regulation control model after mixed compensation is carried out, and a gas generator response predicted value is obtained. According to the method, the experimental data can be effectively utilized to carry out iterative correction on the mechanism model, and the workload of accurate modeling is greatly reduced while the model precision is improved.

Description

Technical Field

[0001] The invention relates to the technical field of solid ramjet engine gas flow regulation system modeling, and in particular to a gas flow regulation system model correction method based on NARX and ELM neural networks. Background Art

[0002] An accurate system model is the starting point and foundation for controller design. The gas flow control system of a solid fuel injection engine has a high degree of uncertainty. Failure to accurately obtain the system model will result in a mismatch between the designed controller and the controlled system, or the controller must be designed very conservatively to cope with the reduced safety margin caused by model uncertainty. All of the above will result in a loss of system control performance. In order to achieve high-performance gas flow control, the uncertainty of the engine model must be effectively compensated to more effectively guide the controller design and achieve full control performance release.

[0003] However, there are many sources of uncertainty in the solid rocket engine system. Its internal chamber structure, propellant thermodynamic parameters, flow regulation drive mechanism and regulating valve, and pressure sensor all have strong uncertainties. In addition, multiple uncertainties are mutually coupled and have serious nonlinear characteristics, making it difficult to take effective measures for accurate correction and compensation. Summary of the invention

[0004] According to the technical problems raised above, a gas flow regulation system model correction method based on NARX and ELM neural networks is provided. Aiming at the problem of limited accuracy of the mechanism model of the solid-fuel engine due to the uncertainty of model parameters, high temperature, high pressure harsh working environment and multi-physical field coupling, the present invention designs a dynamic model compensation method based on neural network. The method first divides the precise modeling task of the solid-fuel engine into two parts: steady-state and dynamic modeling. For the steady-state mechanism model, the throat area of ​​the gas flow regulating valve is selected as the compensation parameter, and the residual compensation algorithm of the nonlinear autoregressive neural network with exogenous inputs (NARX) is used to compensate the residual of the engine steady-state mechanism model. For the dynamic response rate error of the mechanism model, a pseudo-free volume correction scheme is proposed, and the extreme learning machine (ELM) with error and similarity evolution integration is used as the compensation algorithm to correct the engine free volume and realize dynamic response rate correction.

[0005] The technical means adopted by the present invention are as follows:

[0006] A method for correcting the gas flow regulation system model based on NARX and ELM neural networks, including:

[0007] S1. Construct a mechanism model of the gas flow regulation system of the solid rocket ramjet engine, and select the parameters to be compensated for the steady-state and dynamic models;

[0008] S2. Based on the NARX neural network, design a correction method for the steady-state regulation model of the gas flow of the solid rocket ramjet engine, and correct the steady-state model parameters based on experimental data;

[0009] S3. Based on the integrated ELM neural network, design a correction method for the dynamic regulation model of the gas flow of the solid rocket ramjet engine, and correct the dynamic model parameters based on experimental data;

[0010] S4. Combine steps S1 to S3 to perform digital simulation on the gas flow regulation control model of the solid rocket ramjet engine after hybrid compensation, and obtain the predicted value of the gas generator response.

[0011] Furthermore, step S1 specifically includes:

[0012] S11. Considering the working flight altitude and Mach number conditions of the engine, establish a gas flow regulation model of the solid rocket ramjet engine within its working range, and determine the gas flow regulation range. The gas flow regulation model is as follows:

[0013]

[0014] where p g is the internal pressure of the gas generator, R g is the gas constant, T g is the gas temperature, V is the free volume in the gas generator, ρ b is the propellant density, A b is the propellant combustion area, α and n (0 < n < 1) are the combustion rate coefficient and pressure exponent coefficient of the solid propellant respectively, and the exp[σ b (T - T0)] term is the burn rate temperature correction term, T is the external working temperature of the engine, T0 is the nominal state temperature of the propellant, A t is the gas flow regulation throat area, which is a function of the driving motor rotation angle θ, and C r is the characteristic velocity;

[0015] S12. Select the gas flow regulation throat area A t in the gas flow regulation model as the equivalent parameter to be compensated for the steady-state model, and select the free volume V in the gas generator in the gas flow regulation model as the equivalent compensatory parameter for the dynamic model.

[0016] Furthermore, step S2 specifically includes:

[0017] S21. Collect multi-dimensional time series operation data x of the control system during the operation of the solid-fuel rocket engine. Based on the pressure response change rate of the gas generator of the solid-fuel rocket engine after filtering, extract the steady-state compensation training data x in segments. s :

[0018]

[0019] In the formula, Represents the rate of change of pressure data in multidimensional time series data x, The threshold value representing the rate of change of the pressure data, the purpose of step S21 is to obtain a stable segment in the operating data;

[0020] S22. Construct a steady-state inverse model of the solid ram engine as follows:

[0021] A ts (t k )=ρ bs A bs α s C rs p gs (t k ) n-1

[0022] In the formula, ρ bs , A bs , α s , C rs They represent the model parameters of the steady-state model training data respectively;

[0023] S23, input the acquired steady-state pressure training data into the steady-state inverse model of the solid-state ramjet engine constructed in step S22, and calculate the steady-state control valve input angle θ corresponding to the current mechanism model PBM ,as follows:

[0024]

[0025] In the formula, A ts represents the throat area of ​​the steady-state model, It represents the inverse function of the valve angle and valve throat area model in the nominal valve model;

[0026] S24. Calculate t k The actual control valve input angle θ at the moment c Nominal valve control angle θ of the mechanism model PBM The residual value R θ ,as follows:

[0027] R θ (t k )=θ PBM (t k )-θc (t k )

[0028] S25, R θ (t k ) is the NARX neural network fitting output, and the current flow rate is used to adjust the valve control instruction θ c (t k ), the actual engine pressure response p g (t k ) and the propellant grain area A b (t k ) is the external input of NARX; the first n y The steady-state residual delay output Autoregressive input is used to train the NARX neural network, and the NARX neural network is trained in an open-loop mode until the neural network converges;

[0029] S26, predicting the pressure at the last moment using the designed neural network-mechanism hybrid steady-state model Valve control command θ c (t k-1 ), Area of ​​the Grain Column A b (t k-1 ) and NARX’s delayed prediction output sequence As input, online prediction of valve angle compensation value The valve angle of the computer-data driven steady-state compensation model is as follows:

[0030]

[0031] The compensated valve angle Substitute the PBM mechanism model to obtain the current steady-state predicted compensation pressure In this working mode, the predicted pressure of the steady-state hybrid model is used as the control command input, and the predicted output of the compensation model is used to guide the design of the engine control law and the evaluation of the control characteristics;

[0032] S27, with the actual engine pressure p g (t-1) replaces the model predicted pressure in step S26 It is the NARX neural network control input, and the rest is consistent with step S26. In this mode, the system historical pressure feedback is introduced into the neural network, which is not easy to cause pressure error accumulation and has higher prediction accuracy. This prediction mode is used for engine online health status monitoring and fault diagnosis.

[0033] Further, step S3 specifically includes:

[0034] S31. Collect multi-dimensional time series operation data x of the control system during the actual engine operation process, and extract model dynamic training compensation data x in segments based on the response rate of the engine pressure control d,l ,as follows:

[0035]

[0036] In the formula, x d,l Indicates the extracted first segment of dynamic data, Indicates the pressure change rate threshold;

[0037] S32. Based on the extracted dynamic operation data of each segment, the virtual free volume under each segment of dynamic operation data is calculated as follows:

[0038]

[0039] Where V virtual,l Indicates the free volume of the first segment of data;

[0040] S33, extracting the average value V of each virtual free volume virtual,l-mean , maximum value V virtual,l_max and minimum value V virtual,l-min , the obtained virtual free volume value is compared with the median V of the theoretically calculated free volume of each dynamic segment l Make a difference and obtain the virtual free volume compensation R V,l-mean , R V,l-max and R V,l-min ,as follows:

[0041] R V,l-mean =V virtual,l-mean -V l

[0042] R V,l-max =V virtual,l_max -V l

[0043] R V,l-min =V virtual,l-min -V l

[0044] The dynamic compensation model based on ELM is trained with the theoretical free volume as input and the mean, maximum and minimum values ​​of the virtual residual free volume as predicted output;

[0045] S34, combine the trained ELM free volume correction model with the steady-state compensation model, and sequentially calculate the average relative error R between the predicted response of each engine pressure segment and the actual system e,l and correlation coefficient Based on the calculation results of relative error and correlation coefficient, the calculated ELM model output is adaptively weighted and integrated. The formula for updating each ELM weight is as follows:

[0046]

[0047] W ELM,l-mean =α mean,l / (α mean,l +α max,l +α min,l )

[0048] W ELM,l-max =α max,l / (α mean,l +α max,l +α min,l )

[0049] W ELM,l-min =α min,l / (α mean,l +α max,l +α min,l )

[0050] Where W ELM,l is the weight of the neural network in the free volume state of the first segment of data, w1 and w2 are the weight update coefficients;

[0051] S35. Calculate the average relative error R between the compensation model prediction and the actual experimental data obtained after using different ELM compensations e,l and similarity coefficient as follows:

[0052]

[0053] In the formula, Q represents the total number of samples, y i represents the true value, represents the predicted value, represents the mean of the true value, i represents the i-th segment of trajectory data;

[0054] S36. Construct a linear interpolation table with the free volume observation value of the steady-state compensation model as input and the weights of each ELM under the free volume state of the first segment of data as output, interpolate to obtain the weights of each ELM learning machine under different free volumes, and then perform weighted integration on the outputs of multiple ELMs, and mix the dynamic characteristic compensation model of the integrated ELM with the steady-state characteristic compensation model to realize the prediction of the data-driven hybrid compensation model.

[0055] Further, step S4 specifically includes:

[0056] S41, combining the NARX neural network and the integrated ELM neural network compensator described above with the mechanism model, and calculating the compensation values ​​of the steady-state and dynamic model compensation parameters in real time through the neural network during the simulation process;

[0057] S42, inputting the equivalent value to be compensated into the mechanism model system, co-simulating with the mechanism model, and obtaining the predicted value of the gas generator response.

[0058] Compared with the prior art, the present invention has the following advantages:

[0059] 1. The present invention provides a gas flow regulation system model correction method based on NARX and ELM neural networks, which uses data-driven neural networks to perform model correction and compensation, and can effectively use experimental data to iteratively correct the mechanism model.

[0060] 2. The present invention provides a gas flow regulation system model correction method based on NARX and ELM neural networks, which realizes model correction by equivalent compensation, thereby greatly reducing the workload of accurate modeling while improving model accuracy.

[0061] 3. The present invention provides a gas flow control system model correction method based on NARX and ELM neural networks, which attributes the deviations affecting the steady-state and dynamic uncertainty of the model to its key dynamic parameters. Combined with the neural network method, it realizes high-precision model correction of the gas flow control system of the solid-fuel engine, and ultimately improves the accuracy of the engine control system, which helps to guide the design of the solid-fuel engine control system.

[0062] Based on the above reasons, the present invention can be widely promoted in the field of solid ramjet engine gas flow regulation system modeling and other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0064] Figure 1 It is a schematic diagram of model correction of the gas flow regulation system of the solid ramjet engine of the present invention.

[0065] Figure 2 This is a schematic diagram of the NARX neural network structure provided by the present invention.

[0066] Figure 3 Schematic diagram of the NARX model training and application process provided by the present invention.

[0067] Figure 4 This is a schematic diagram of the structure of the extreme learning machine provided by the present invention.

[0068] Figure 5 Schematic diagram of the ELM model training and application process provided by the present invention. DETAILED DESCRIPTION

[0069] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0070] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0071] In order to improve the model accuracy of the solid ramjet engine, the present invention proposes a gas flow regulation system model correction method based on NARX and ELM neural networks. Figure 1 The figure shows the architecture diagram of the gas flow regulation system model correction method based on NARX and ELM neural network used in the method of the present invention. The specific implementation method of the present invention takes the design of a certain type of aircraft solid ramjet engine as an example, and the specific implementation steps are as follows, including:

[0072] S1. Construct the gas flow regulation system mechanism model of the solid-fuel engine and select the parameters to be compensated for the steady-state and dynamic models;

[0073] S2. Based on the NARX neural network, a correction method for the steady-state regulation model of the gas flow of the solid fuel injection engine is designed, and the steady-state model parameters are corrected based on experimental data;

[0074] S3. Based on the integrated ELM neural network, a correction method for the dynamic regulation model of the gas flow of the solid fuel injection engine is designed, and the dynamic model parameters are corrected based on experimental data;

[0075] S4. Combine steps S1 to S3 to perform digital simulation on the gas flow regulation control model of the solid ramjet engine after hybrid compensation, and obtain the predicted value of the gas generator response.

[0076] During specific implementation, as a preferred implementation manner of the present invention, step S1 specifically includes:

[0077] S11. Select a solid ramjet engine of a certain ballistic model as the research object, consider the working flight altitude and Mach number conditions of the engine, establish a gas flow regulation model of the solid ramjet engine within its working range, and determine the gas flow regulation range. The gas flow regulation model is as follows:

[0078]

[0079] In the formula, p g is the internal pressure of the gas generator, R g is the gas constant, T g is the gas temperature, V is the free volume in the gas generator, ρ b is the propellant density, A b is the propellant combustion area, α and n (0 < n < 1) are respectively the combustion rate coefficient and pressure exponent coefficient of the solid propellant, and the exp[σ b (T - T0)] term is the burning rate temperature correction term, T is the external working temperature of the engine, T0 is the nominal state temperature of the propellant, A t is the gas flow regulation throat area, which is a function of the driving motor rotation angle θ, and C r is the characteristic velocity;

[0080] S12. Select the gas flow regulation throat area A t in the gas flow regulation model as the equivalent parameter to be compensated in the steady-state model, and select the free volume V in the gas generator in the gas flow regulation model as the equivalent compensatory parameter in the dynamic model.

[0081] During specific implementation, as a preferred implementation manner of the present invention, step S2 specifically includes:

[0082] S21. Collect the multi-dimensional time-series operation data x of the control system during the operation of the solid ramjet engine, including the gas generator pressure data, the gas flow regulation valve motor angle, and the propellant grain area. Based on the rate of change of the gas generator pressure response of the solid ramjet engine after filtering, segment and extract the steady-state compensation training data x s :

[0083]

[0084] In the formula, Represents the rate of change of pressure data in multidimensional time series data x, The threshold value representing the rate of change of the pressure data, the purpose of step S21 is to obtain a stable segment in the operating data;

[0085] S22. Construct a steady-state inverse model of the solid ram engine as follows:

[0086] A ts (t k )=ρ bs A bs α s C rs p gs (t k ) n-1

[0087] In the formula, ρ bs , A bs , α s , C rs They represent the model parameters of the steady-state model training data respectively;

[0088] S23, input the acquired steady-state pressure training data into the steady-state inverse model of the solid-state ramjet engine constructed in step S22, and calculate the steady-state control valve input angle θ corresponding to the current mechanism model PBM ,as follows:

[0089]

[0090] In the formula, A ts represents the throat area of ​​the steady-state model, It represents the inverse function of the valve angle and valve throat area model in the nominal valve model;

[0091] S24. Calculate t k The actual control valve input angle θ at the moment c Nominal valve control angle θ of the mechanism model PBM The residual value R θ ,as follows:

[0092] R θ (t k )=θ PBM (t k )-θ c (t k )

[0093] S25, R θ (t k ) is the NARX neural network fitting output, and the current flow rate is used to adjust the valve control instruction θ c (t k ), the actual engine pressure response p g (tk ) and the propellant grain area A b (t k ) is the external input of NARX; the first n y The steady-state residual delay output Autoregressive input is used to train the NARX neural network, and the NARX neural network is trained in an open-loop mode until the neural network converges;

[0094] S26, Online Prediction Mode 1:

[0095] The pressure is predicted at the previous moment using the designed neural network-mechanism hybrid steady-state model Valve control command θ c (t k-1 ), Area of ​​the Grain Column A b (t k-1 ) and NARX's delayed prediction output sequence As input, online prediction of valve angle compensation value The valve angle of the computer-data driven steady-state compensation model is as follows:

[0096]

[0097] The compensated valve angle Substitute the PBM mechanism model to obtain the current steady-state predicted compensation pressure In this working mode, the predicted pressure of the steady-state hybrid model is used as the control command input, and the predicted output of the compensation model is used to guide the design of the engine control law and the evaluation of the control characteristics;

[0098] S27, Online prediction mode 2:

[0099] The actual engine pressure p g (t-1) replaces the model predicted pressure in step S26 is the NARX neural network control input, and the rest is consistent with step S26. In this mode, the system historical pressure feedback is introduced into the neural network, which is not easy to cause pressure error accumulation and has higher prediction accuracy. This prediction mode is used for online engine health status monitoring and fault diagnosis. Figure 2 As shown, a schematic diagram of the NARX neural network structure used in the method of the present invention is shown, including (a) open-loop mode and (b) closed-loop mode; Figure 3 As shown, a schematic diagram of the NARX model training and application process is shown.

[0100] In specific implementation, as a preferred embodiment of the present invention, step S3 specifically includes:

[0101] S31. Collect multi-dimensional time series operation data x (including input data, state data and output data) of the control system during the actual engine operation process, and extract model dynamic training compensation data x in segments based on the response rate of the engine pressure control. d,l ,as follows:

[0102]

[0103] In the formula, x d,l Indicates the extracted first segment of dynamic data, Indicates the pressure change rate threshold;

[0104] S32. Based on the extracted dynamic operation data of each segment, the virtual free volume under each segment of dynamic operation data is calculated as follows:

[0105]

[0106] Where V virtual,l Indicates the free volume of the first segment of data;

[0107] S33, extracting the average value V of each virtual free volume virtual,l-mean , maximum value V virtual,l_max and minimum value V virtual,l-min , the obtained virtual free volume value is compared with the median V of the theoretically calculated free volume of each dynamic segment l Make a difference and obtain the virtual free volume compensation R V,l-mean , R V,l-max and R V,l-min ,as follows:

[0108] R V,l-mean =V virtual,l-mean -V l

[0109] R V,l-max =V virtual,l_max -V l

[0110] R V,l-min =V virtual,l-min -V l

[0111] The dynamic compensation model based on ELM is trained with the theoretical free volume as input and the mean, maximum and minimum values ​​of the virtual residual free volume as predicted output;

[0112] S34, combine the trained ELM free volume correction model with the steady-state compensation model, and sequentially calculate the average relative error R between the predicted response of each engine pressure segment and the actual system e,l and the correlation coefficient R l 2, based on the relative error and correlation coefficient calculation results, the calculated ELM model output is adaptively weighted and integrated, and the formula for updating each ELM weight is as follows:

[0113]

[0114] W ELM,l-mean =α mean,l / (α mean,l +α max,l +α min,l )

[0115] W ELM,l-max =α max,l / (α mean,l +α max,l +α min,l )

[0116] W ELM,l-min =α min,l / (α mean,l +α max,l +α min,l )

[0117] Where W ELM,l is the weight of the neural network in the free volume state of the first segment of data, w1 and w2 are the weight update coefficients;

[0118] S35. Calculate the average relative error R between the compensation model prediction and the actual experimental data obtained after using different ELM compensations e,l and similarity coefficient as follows:

[0119]

[0120] In the formula, Q represents the total number of samples, y i represents the true value, represents the predicted value, represents the mean of the true value, i represents the i-th segment of trajectory data;

[0121] S36, construct a linear interpolation table with the free volume observation value of the steady-state compensation model as input and the weight of each ELM under the free volume state of the first segment of data as output, interpolate to obtain the weight of each ELM learning machine under different free volumes, and then perform weighted integration on the outputs of multiple ELMs, and mix the dynamic characteristic compensation model of the integrated ELM with the steady-state characteristic compensation model to realize the prediction of the data-driven hybrid compensation model. Figure 4 As shown in the figure, a schematic diagram of the extreme learning machine structure used is given, as shown in Figure 5 As shown in the figure, a schematic diagram of the ELM model training and application process is given.

[0122] In specific implementation, as a preferred embodiment of the present invention, step S4 specifically includes:

[0123] S41, combining the NARX neural network and the integrated ELM neural network compensator described above with the mechanism model, and calculating the compensation values ​​of the steady-state and dynamic model compensation parameters in real time through the neural network during the simulation process;

[0124] S42, inputting the equivalent value to be compensated into the mechanism model system, co-simulating with the mechanism model, and obtaining the predicted value of the gas generator response.

[0125] In summary, the present invention fully considers the modeling difficulties of the gas flow regulation model of the solid-fuel engine, and provides a gas flow regulation system model correction method based on NARX and ELM neural networks, which attributes the deviations affecting the steady-state and dynamic uncertainty of the model to its key dynamic parameters. Combined with the neural network method, high-precision model correction of the gas flow regulation system of the solid-fuel engine is achieved, and ultimately the accuracy of the engine control system is improved.

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A gas flow regulation system model correction method based on NARX and ELM neural network, characterized in that: include: S1. Construct the gas flow regulation system mechanism model of the solid-fuel engine and select the parameters to be compensated for the steady-state and dynamic models; S2. Based on the NARX neural network, a correction method for the steady-state regulation model of the gas flow of the solid fuel injection engine is designed, and the steady-state model parameters are corrected based on experimental data; S3. Based on the integrated ELM neural network, a correction method for the dynamic regulation model of the gas flow of the solid fuel injection engine is designed, and the dynamic model parameters are corrected based on experimental data; S4. Combining steps S1 to S3, digital simulation of the gas flow regulation control model of the solid-fuel engine after mixing compensation is performed to obtain a predicted value of the gas generator response.

2. According to claim 1, a gas flow regulation system model correction method based on NARX and ELM neural network is characterized in that: Step S1 specifically includes: S11. Considering the engine operating flight altitude and Mach number conditions, a gas flow regulation model of the solid fuel injection engine within its operating range is established, and the gas flow regulation range is determined. The gas flow regulation model is as follows: where p g is the internal pressure of the gas generator, R g is the gas constant, T g is the gas temperature, V is the free volume in the gas generator, ρ b is the propellant density, A b is the propellant combustion area, α and n (0 < n < 1) are the combustion rate coefficient and pressure exponent coefficient of the solid propellant respectively, the exp[σ b (T - T0)] term is the burning rate temperature correction term, T is the external working temperature of the engine, T0 is the nominal state temperature of the propellant, A t is the gas flow regulation throat area, which is a function of the driving motor rotation angle θ, C r is the characteristic velocity; S12. Select the gas flow regulation throat area A in the gas flow regulation model. t is the equivalent parameter to be compensated for the steady-state model, and the free volume V in the gas generator in the gas flow regulation model is selected as the equivalent compensation parameter of the dynamic model.

3. The gas flow regulation system model correction method based on NARX and ELM neural network according to claim 1 is characterized in that: Step S2 specifically includes: S21. Collect multi-dimensional time series operation data x of the control system during the operation of the solid-fuel rocket engine. Based on the pressure response change rate of the gas generator of the solid-fuel rocket engine after filtering, extract the steady-state compensation training data x in segments. s : In the formula, Represents the rate of change of pressure data in multidimensional time series data x, The threshold value representing the rate of change of the pressure data, the purpose of step S21 is to obtain a stable segment in the operating data; S22. Construct a steady-state inverse model of the solid ram engine as follows: A ts (t k )=ρ bs A bs a s C rs p gs (t k ) n-1 In the formula, ρ bs , A bs , α s , C rs They represent the model parameters of the steady-state model training data respectively; S23, input the acquired steady-state pressure training data into the steady-state inverse model of the solid-state ramjet engine constructed in step S22, and calculate the steady-state control valve input angle θ corresponding to the current mechanism model PBM ,as follows: In the formula, A ts represents the throat area of ​​the steady-state model, It represents the inverse function of the valve angle and valve throat area model in the nominal valve model; S24. Calculate t k The actual control valve input angle θ at the moment c Nominal valve control angle θ of the mechanism model PBM The residual value R θ ,as follows: R θ (t k )=θ PBM (t k )-θ c (t k ) S25, R θ (t k ) is the NARX neural network fitting output, and the current flow rate is used to adjust the valve control instruction θ c (t k ), the actual engine pressure response p g (t k ) and the propellant grain area A b (t k ) is the external input of NARX; the first n y The steady-state residual delay output Autoregressive input is used to train the NARX neural network, and the NARX neural network is trained in an open-loop mode until the neural network converges; S26, predicting the pressure at the last moment using the designed neural network-mechanism hybrid steady-state model Valve control command θ c (t k-1 ), Area of ​​the Grain Column A b (t k-1 ) and NARX’s delayed prediction output sequence As input, online prediction of valve angle compensation value The valve angle of the computer-data driven steady-state compensation model is as follows: The compensated valve angle Substitute the PBM mechanism model to obtain the current steady-state predicted compensation pressure In this working mode, the predicted pressure of the steady-state hybrid model is used as the control command input, and the predicted output of the compensation model is used to guide the design of the engine control law and the evaluation of the control characteristics; S27, with the actual engine pressure p g (t-1) replaces the model predicted pressure in step S26 It is the NARX neural network control input, and the rest is consistent with step S26. In this mode, the system historical pressure feedback is introduced into the neural network, which is not easy to cause pressure error accumulation and has higher prediction accuracy. This prediction mode is used for engine online health status monitoring and fault diagnosis.

4. A gas flow regulation system model correction method based on NARX and ELM neural network according to claim 1, characterized in that: Step S3 specifically includes: S31. Collect multi-dimensional time series operation data x of the control system during the actual engine operation process, and extract model dynamic training compensation data x in segments based on the response rate of the engine pressure control d,l ,as follows: In the formula, x d,l Indicates the extracted first segment of dynamic data, Indicates the pressure change rate threshold; S32. Based on the extracted dynamic operation data of each segment, the virtual free volume under each segment of dynamic operation data is calculated as follows: Where V virtual,l Indicates the free volume of the first segment of data; S33, extracting the average value V of each virtual free volume virtual,l-mean , maximum value V virtual,l_max and minimum value V virtual,l-min , the obtained virtual free volume value is compared with the median V of the theoretically calculated free volume of each dynamic segment l Make a difference and obtain the virtual free volume compensation R V,l-mean , R V,l-max and R V,l-min ,as follows: R V,l-mean =V virtual,l-mean -V l R V,l-max =V virtual,l_max -V l R V,l-min =V virtual,l-min -V l The dynamic compensation model based on ELM is trained with the theoretical free volume as input and the mean, maximum and minimum values ​​of the virtual residual free volume as predicted output; S34, combine the trained ELM free volume correction model with the steady-state compensation model, and sequentially calculate the average relative error R between the predicted response of each engine pressure segment and the actual system e,l and correlation coefficient Based on the calculation results of relative error and correlation coefficient, the calculated ELM model output is adaptively weighted and integrated. The formula for updating each ELM weight is as follows: W ELM,l-mean ＝α mean,l / (α mean,l +α max,l +α min,l ) W ELM,l-max ＝α max,l / (α mean,l +α max,l +α min,l ) W ELM,l-min ＝α min,l / (α mean,l +α max,l +α min,l ) Where W ELM,l is the weight of the neural network in the free volume state of the first segment of data, w1 and w2 are the weight update coefficients; S35. Calculate the average relative error R between the compensation model prediction and the actual experimental data obtained after using different ELM compensations e,l and similarity coefficient as follows: In the formula, Q represents the total number of samples, y i represents the true value, represents the predicted value, represents the mean of the true value, i represents the i-th segment of trajectory data; S36. Construct a linear interpolation table with the free volume observation value of the steady-state compensation model as input and the weights of each ELM under the free volume state of the first segment of data as output, interpolate to obtain the weights of each ELM learning machine under different free volumes, and then perform weighted integration on the outputs of multiple ELMs, and mix the dynamic characteristic compensation model of the integrated ELM with the steady-state characteristic compensation model to realize the prediction of the data-driven hybrid compensation model.

5. The gas flow regulation system model correction method based on NARX and ELM neural network according to claim 1 is characterized in that: Step S4 specifically includes: S41, combining the NARX neural network and the integrated ELM neural network compensator described above with the mechanism model, and calculating the compensation values ​​of the steady-state and dynamic model compensation parameters in real time through the neural network during the simulation process; S42, inputting the equivalent value to be compensated into the mechanism model system, co-simulating with the mechanism model, and obtaining the predicted value of the gas generator response.

Images