Fuel flow prediction method and system based on physical constraints and data driving
By combining physical constraints and data-driven methods, a fuel flow prediction model based on degradation perception and operating condition perception was constructed, which solved the problems of low accuracy in engine thrust estimation and poor model interpretability, and achieved high-precision and stable thrust and fuel flow prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CIVIL AVIATION UNIV OF CHINA
- Filing Date
- 2026-03-17
- Publication Date
- 2026-05-29
AI Technical Summary
In existing technologies, the accuracy of aircraft engine thrust estimation is low and the model interpretability is poor. It is difficult to reflect the degradation effects caused by wear, aging and other factors during long-term operation of the engine. Pure data-driven methods lack physical constraints, resulting in unstable thrust prediction.
A fuel flow prediction method based on physical constraints and data-driven approaches is adopted. By constructing a degradation-aware multiplicative residual correction model and a condition-aware fusion weight function, combined with a physical engine performance model, thrust correction is performed using flight operation data, and constraints are applied through a physical consistency constraint loss function.
It achieves high-precision, low-bias prediction of engine thrust and fuel flow, ensuring the physical consistency and interpretability of the model, and improving the prediction accuracy and stability under complex operating conditions.
Smart Images
Figure CN122113655A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of aircraft performance modeling and flight data analysis technology, specifically to a fuel flow prediction method and system based on physical constraints and data-driven approaches. Background Technology
[0002] Aircraft engine thrust is a key parameter in flight performance analysis, fuel consumption assessment, and operational health management. However, in actual flight, aircraft typically lack sensors to directly measure engine thrust, which can only be obtained through physical models or indirect calculations. Current technologies often use physical models based on engine performance charts or empirical formulas to estimate thrust. These methods rely on ideal engine parameters and fail to reflect the degradation effects caused by wear, aging, and other factors during long-term operation, leading to decreased thrust estimation accuracy. On the other hand, while purely data-driven methods can fit flight operation data, they lack physical constraints, have poor model interpretability, and are prone to instability when operating conditions change significantly or data distribution shifts. Therefore, how to effectively correct the thrust estimation of physical models using flight operation data while ensuring physical consistency and interpretability, thereby achieving high-precision thrust prediction and fuel flow estimation, remains a pressing technical problem in this field. Summary of the Invention
[0003] To achieve the above objectives, the present invention provides the following technical solution:
[0004] A fuel flow prediction method based on physical constraints and data-driven approaches, comprising:
[0005] S1: Acquire the aircraft's flight operation data, which includes at least engine parameters and flight condition parameters;
[0006] S2: Input flight operation data into the physical engine performance model and output the reference thrust corresponding to the flight state;
[0007] S3: Based on flight operation data, construct a degradation-aware multiplicative residual correction model and output a correction factor to characterize the degree of engine degradation. Among them, the correction factor The weight function output by the subsequent working condition confidence modulation network By merging, an effective correction factor can be formed. And the effective correction factor is used as the final correction factor;
[0008] S4: Correct the reference thrust according to the correction factor to obtain the corrected reference thrust;
[0009] S5: Constructing the fusion weight function for operational condition awareness The predicted thrust is obtained by weighted fusion of the physical engine performance model and the degradation-aware multiplicative residual correction model. ;
[0010] S6: Construct a physical consistency constraint loss function based on the longitudinal dynamics of the aircraft to constrain the predicted thrust;
[0011] S7: Based on Predicted Thrust The engine fuel flow rate was calculated.
[0012] Furthermore, the engine parameters include at least: engine low-pressure speed N1, high-pressure speed N2, and exhaust temperature EGT, and the flight operating parameters include at least: flight altitude ALT, Mach number M, throttle lever angle TLA, and total temperature TAT.
[0013] Furthermore, the degradation-aware multiplicative residual correction model is a degradation-condition decoupled dual-channel multiplicative residual neural network model, and its output correction factor is a dimensionless scalar, used to characterize the multiplicative deviation relationship between the reference thrust output by the physical engine performance model and the actual thrust. The degradation-aware multiplicative residual correction model includes: a degradation characterization subnetwork DRN and a condition embedding subnetwork GRN.
[0014] Furthermore, the degradation characterization subnetwork DRN is used to characterize the long-term performance degradation characteristics of the engine, and its input is engine degradation-related features: DRN employs a small residual fully connected network structure, consisting of one hidden layer and one residual connection. The hidden layer has 6 neurons, and the activation function is tanh. The mapping relationship is as follows:
[0015] ;
[0016] ;
[0017] The operational condition embedding subnetwork GRN is used to extract the embedding features of the thrust modulation effect of flight operational conditions. Its input is... Engine operating condition related characteristics: .
[0018] GRN also employs a small residual fully connected structure with 8 neurons in the hidden layer, using tanh as the activation function, and outputting a working condition embedding vector. .
[0019] Among them, ℎ d For the bias vector in the degenerate representation network, 𝑏 d This is the bias vector in the degenerate representation network, used to translate the linear transformation results in fully connected layers to enhance the model's expressive power. dFor degenerate embedding vectors, used to characterize engine health status, 𝑊 d Here, is the learnable weight matrix in the degenerate representation network, used to achieve a linear transformation on the input degenerate feature vector. 𝑟 It is a learnable weight matrix in the residual mapping module, used to implement nonlinear residual transformation of input features.
[0020] Furthermore, the output is a correction factor used to characterize the degree of engine degradation. The basis is: degenerate embedding vector With working condition embedding vector The data is stitched together in the fusion layer and mapped to the original residuals. :
[0021] ;
[0022] The correction factor is obtained through the restricted Sigmoid function:
[0023] ;
[0024] in, Indicates hyperparameters and This is the weight matrix of the fusion layer, used to weight the degenerate embedding vector. With working condition embedding vector The concatenated fused feature vector is linearly mapped to the original degenerate parameters in scalar form. , Let represent the bias vector of the fully connected layer, which is the bias vector of the weight matrix. The corresponding trainable parameters are used to introduce learnable offsets after feature linear transformation, thereby improving the network's ability to express nonlinear mapping relationships.
[0025] Furthermore, a fusion weight function for operational condition perception is constructed. The formula used to characterize the reliability of a physical engine performance model under different operating conditions is as follows:
[0026] ;
[0027] in It is a prior confidence function based on flight conditions, used to describe the empirical estimate of the thrust confidence of a physical baseline model under given altitude and Mach number conditions. `logit` represents the inverse function of the sigmoid function. Indicates the confidence level. For learnable perturbation terms;
[0028] ε is obtained through a lightweight gated residual network:
[0029] ;
[0030] Used to make small adaptive corrections to prior confidence levels. , for Activation function Input for a lightweight gated residual network;
[0031] Forming an effective correction factor The basis is: .
[0032] Specifically, the Degradation Representation Network (DRN), as the first sub-network, is used to extract degradation features that characterize the long-term performance decline and health status changes of the engine, and outputs a degradation embedding vector. The Condition Embedding Subnetwork (GRN), serving as the second subnetwork, is used to extract condition embedding vectors that characterize the impact of changes in operating conditions such as flight altitude, Mach number, and ambient temperature on engine thrust. The degenerate representation vector and the operating condition representation vector are jointly mapped in the fusion layer to generate a dimensionless multiplicative residual correction factor. This is used to characterize the proportional deviation between the physical model's baseline thrust and the actual engine thrust. Based on the first sub-network's degradation representation network, a small residual network structure is used to perform nonlinear mapping on the input degradation-related features N1, N2, and EGT. After passing through a fully connected hidden layer and residual connections, a high-dimensional degradation embedding vector is output. This embedding vector is used to characterize the engine's health state and provides a correction factor for subsequent fusion layers. This enables degradation-aware correction of the physical model's baseline thrust; the hidden layer has 6 units, and the activation function is... To ensure controlled output vector magnitude and smooth gradient, thus improving network training stability; based on the second sub-network, the operational condition embedding network, the flight operational condition-related features ALT, M, TLA, and TAT are input and nonlinearly mapped through a small residual network structure. After passing through a fully connected hidden layer and residual connections, a high-dimensional operational condition embedding vector is output. This is used to characterize the thrust modulation effect of flight conditions and to provide correction factors for subsequent fusion layers. This enables condition-aware modulation of the reference thrust; the hidden layer has 8 units, and the activation function is fixed at 1. This ensures that the magnitude of the output vector is controlled and the gradient is smooth, thereby improving the stability of network training.
[0033] Furthermore, the basis for deriving the corrected baseline thrust is as follows: multiply the final correction factor by the baseline thrust to obtain the corrected baseline thrust; thus, the predicted thrust is obtained. The basis is:
[0034] ;
[0035] ;
[0036] in, Indicates predicted thrust. Indicates the reference thrust. Indicates the confidence level. This represents the correction factor.
[0037] Furthermore, the physical consistency constraint loss function is constructed based on Newton's second law and is used to constrain the dynamic relationship between predicted thrust and aircraft mass, aerodynamic drag, and flight acceleration. The specific formula for calculating the physical consistency constraint loss function is as follows:
[0038] ;
[0039] in:
[0040] ;
[0041] ;
[0042] ;
[0043] In the above formula, Represents the loss function for physical consistency constraints. This represents the mean square error. Indicates predicted thrust. Indicates aerodynamic drag. Indicates the mass of the aircraft. Represents gravitational acceleration. Indicates the flight path angle. Indicates flight acceleration. Indicates the drag coefficient. Indicates dynamic pressure. Indicates wing area. Indicates air density, Indicates vacuum velocity, This indicates the aircraft's ground speed.
[0044] Furthermore, the basis for obtaining engine fuel flow rate is as follows:
[0045] ;
[0046] in, Indicates engine fuel flow rate. Indicates predicted thrust. Specific impulse is the thrust generated per unit weight of fuel flow. It represents the acceleration due to gravity.
[0047] A fuel flow prediction system based on physical constraints and data-driven principles, comprising:
[0048] The data acquisition module is used to acquire the aircraft's flight operation data, which includes at least engine parameters and flight condition parameters.
[0049] The baseline thrust calculation module is used to input flight operation data into the physical engine performance model and output the baseline thrust corresponding to the flight state.
[0050] The degradation correction module, based on flight operation data, constructs a degradation-aware multiplicative residual correction model and outputs correction factors to characterize the degree of engine degradation. Among them, the correction factor The weight function output by the subsequent working condition confidence modulation network By merging, an effective correction factor can be formed. The effective correction factor is used as the final correction factor; and the reference thrust is corrected according to the correction factor to obtain the corrected reference thrust.
[0051] The fusion computing module constructs the fusion weight function for operational condition awareness. The predicted thrust is obtained by weighted fusion of the physical engine performance model and the degradation-aware multiplicative residual correction model. ;
[0052] The physical constraint module constructs a physical consistency constraint loss function based on the longitudinal dynamics of the aircraft to constrain the predicted thrust;
[0053] Fuel flow calculation module, used for predictive thrust The engine fuel flow rate was calculated.
[0054] This invention provides a fuel flow prediction method and system based on physical constraints and data-driven approaches, which has the following advantages: By introducing a physical model as a benchmark thrust, physical constraints on the basic performance of the engine are achieved, avoiding the uninterpretability and distortion problems of purely data-driven models; by using a degradation-aware multiplicative residual model, explicit modeling of the engine's health state and long-term performance degradation is achieved, improving the stability and transferability of thrust prediction; by using a condition-aware fusion weight function, the physical model and the data-driven correction model are adaptively weighted under different flight stages, improving prediction accuracy under complex operating conditions; by introducing physical consistency constraints based on Newton's second law, weakly supervised training is achieved using aircraft flight acceleration in the absence of real thrust labels, significantly improving the reliability and physical interpretability of the model; thus enabling high-precision, low-bias, and strongly physically consistent prediction of engine thrust and fuel flow in actual operating environments. Attached Figure Description
[0055] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0056] Figure 1 A flowchart illustrating an aircraft thrust and fuel flow prediction method based on the fusion of physical models and data-driven approaches, provided for an embodiment of the present invention.
[0057] Figure 2 A schematic diagram of the overall architecture of the OpenAP-based benchmark thrust calculation and degradation-aware multiplicative residual correction model provided in this embodiment of the invention;
[0058] Figure 3 This is a schematic diagram of a dual-channel multiplicative residual network structure for degradation-condition decoupling provided in an embodiment of the present invention;
[0059] Figure 4 This is a schematic diagram of the process for calculating engine fuel flow based on predicted thrust, provided in an embodiment of the present invention. Detailed Implementation
[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0061] I. Overall Process
[0062] S1: Please refer to Figure 1 This embodiment provides a fuel flow prediction method based on physical constraints and data-driven methods, including the following specific steps: S1: Acquire the flight operation data of the aircraft. The flight operation data includes at least engine parameters and flight condition parameters. The engine parameters include: engine low-pressure speed N1, high-pressure speed N2 and exhaust temperature EGT. The flight condition parameters include: flight altitude ALT, Mach number M, throttle lever angle TLA and total temperature TAT.
[0063] S2: Input flight operation data into a pre-built physical engine performance model and output the baseline thrust under the corresponding flight conditions. The physical engine performance model uses a thrust model built with OpenAP to calculate the baseline thrust output under the current conditions based on flight altitude (ALT), Mach number (M), throttle lever angle (TLA), total temperature (TAT), and aircraft parameters.
[0064] S201: In this embodiment of the invention, the "reference thrust corresponding to the flight state" "Obtained by calling the OpenAP thrust calculation model. This baseline thrust is used to characterize the theoretical net engine thrust under current flight conditions and aircraft type, providing a basic input for subsequent thrust correction and fuel flow prediction."
[0065] S202: The input parameters required to calculate the baseline thrust include two categories: flight condition parameters and aircraft type parameters. The flight condition parameters are: ALT: flight altitude; M: Mach number; TAS: supermassive speed; ROC: rate of climb; TLA: throttle lever angle; TAT: total temperature. The aircraft type parameters are: ac: aircraft type; eng: engine type; : Maximum static thrust at sea level; BPR: Bypass ratio of turbofan engine; Rated thrust curve parameters: Nominal thrust parameters during cruise / climb phases. The above aircraft parameters are pre-configured in OpenAP, corresponding one-to-one with specific aircraft and engine models, and are used to support the thrust calculation model.
[0066] S203: Construct a thrust calculation object based on the aircraft type (ac) and engine type (eng), and call the takeoff, climb, or cruise functions to obtain the net thrust value under the corresponding state, depending on the current flight phase. This baseline thrust is used to characterize the theoretical engine thrust under ideal conditions, providing a basic input for subsequent thrust correction and fuel flow prediction models.
[0067] ;
[0068] Takeoff phase:
[0069] ;
[0070] Climbing phase:
[0071] ;
[0072] Cruise phase:
[0073] ;
[0074] in: Indicates the current vacuum velocity. Indicates the current flight altitude; This indicates the current climb rate.
[0075] S3: Based on flight operation data, construct a degradation-aware multiplicative residual correction model and output a correction factor to characterize the degree of engine degradation. Among them, the correction factor The weight function output by the subsequent working condition confidence modulation network By merging, an effective correction factor can be formed. The effective correction factor is used as the final correction factor. It should be noted that the range of the correction factor is constrained to be no greater than 1, so as to reflect the weakening effect of engine degradation on thrust output capability.
[0076] S4: Correct the reference thrust according to the correction factor to obtain the corrected reference thrust. Based on the product between the final correction factor and the reference thrust, the corrected reference thrust is obtained.
[0077] S5: Constructing the fusion weight function for operational condition awareness The predicted thrust is obtained by weighted fusion of the physical engine performance model and the degradation-aware multiplicative residual correction model. : .
[0078] S6: Construct a physical consistency constraint loss function based on the longitudinal dynamics of the aircraft to constrain the predicted thrust;
[0079] S7: Based on Predicted Thrust The engine fuel flow rate was calculated.
[0080] II. Degenerate Perceptual Multiplicative Residual Network
[0081] See appendix Figure 3 , Figure 3 A schematic diagram of the dual-channel multiplicative residual network structure for degradation-condition decoupling is given. In S2, the degradation-aware multiplicative residual correction model includes: a degradation representation sub-network (DRN) and a condition embedding sub-network (GRN). In this embodiment, the degradation-aware multiplicative residual correction model is a dual-channel multiplicative residual neural network model for degradation-condition decoupling, and its output correction factor is a dimensionless scalar, used to characterize the multiplicative deviation relationship between the physical model's reference thrust and the actual thrust.
[0082] The fusion layer corresponding to the degradation-aware multiplicative residual correction model is:
[0083] (1) Degradation Characterization Subnetwork (DRN): The DRN is used to characterize the long-term performance degradation characteristics of the engine. Its input is the engine degradation-related features: DRN employs a small residual fully connected network structure, consisting of one hidden layer and one residual connection. The hidden layer has 6 neurons, and the activation function is... The mapping relationship is as follows:
[0084] ;
[0085] ;
[0086] Among them, ℎ d For the bias vector in the degenerate representation network, 𝑏 d This is the bias vector in the degenerate representation network, used to translate the linear transformation results in fully connected layers to enhance the model's expressive power. d For degenerate embedding vectors, used to characterize engine health status, 𝑊 d Here, is the learnable weight matrix in the degenerate representation network, used to achieve a linear transformation on the input degenerate feature vector. 𝑟 It is a learnable weight matrix in the residual mapping module, used to implement nonlinear residual transformation of input features.
[0087] (2) The operating condition embedding subnetwork GRN is used to extract the embedding features of the flight operating condition on the thrust modulation effect. Its input is Engine operating condition related characteristics: .
[0088] GRN also employs a small residual fully connected structure with 8 neurons in the hidden layer, using tanh as the activation function, and outputting a working condition embedding vector. The same principle applies to the above, so I will not elaborate further here.
[0089] (3) Correction factor generation, and The data is stitched together in the fusion layer and mapped to the original residuals. :
[0090] ;
[0091] The final correction factor is obtained through the restricted sigmoid function:
[0092] ;
[0093] in, Indicates hyperparameters and , This is the weight matrix of the fusion layer, used to weight the degenerate embedding vector. With working condition embedding vector The fused feature vector formed after splicing is linearly mapped to ensure that the correction factor is always less than 1, so as to reflect the decrease in thrust capability caused by engine degradation. Represents the Sigmoid activation function; a correction factor for generating dimensionless multiplicative residuals. It is used to characterize the proportional deviation between the physical model's reference thrust and the actual engine thrust.
[0094] III. Operating Condition Confidence Network
[0095] Constructing a fusion weight function for operational condition awareness The formula used to characterize the reliability of the physical model (i.e., the physical engine performance model) under different operating conditions is as follows:
[0096] ;
[0097] in It is a prior confidence function based on flight conditions, used to describe the empirical estimate of the thrust confidence of a physical baseline model under given altitude and Mach number conditions. `logit` represents the inverse function of the sigmoid function. Indicates the confidence level. For learnable perturbation terms;
[0098] Obtained through a lightweight gated residual network:
[0099] ;
[0100] Used to make small adaptive corrections to prior confidence levels. , for Activation function For input to a lightweight gated residual network.
[0101] IV. Physical Consistency Constraints
[0102] A physical consistency loss is constructed, and its corresponding physical consistency constraint loss function is built based on Newton's second law to constrain the dynamic relationship between predicted thrust and aircraft mass, aerodynamic drag, and flight acceleration. The specific formula for calculating the physical consistency constraint loss function is as follows:
[0103] ;
[0104] in:
[0105] ;
[0106] ;
[0107] ;
[0108] This loss is used to ensure that the thrust output by the neural network matches the actual acceleration of the aircraft. Represents the loss function for physical consistency constraints. This represents the mean square error. Indicates predicted thrust. Indicates aerodynamic drag. Indicates the mass of the aircraft. Represents gravitational acceleration. Indicates the flight path angle. Indicates flight acceleration. Indicates the drag coefficient. Indicates dynamic pressure. Indicates wing area. Indicates air density, Indicates vacuum velocity, This represents the aircraft's ground speed; the predicted thrust was trained by indirectly supervising the aircraft's acceleration in the absence of direct thrust measurement data.
[0109] V. Fuel Flow Calculation
[0110] The engine fuel flow rate is calculated based on the predicted thrust and engine performance parameters;
[0111] The specific formula for calculating the engine fuel flow rate is as follows:
[0112] ;
[0113] in, Indicates engine fuel flow rate. Indicates predicted thrust. Specific impulse is a defined engine performance parameter that represents the thrust generated per unit weight of fuel flow. It represents the acceleration due to gravity, and it represents the ratio between engine thrust and fuel flow rate.
[0114] In this invention, a thrust and fuel flow prediction framework based on a combination of physical models and data-driven approaches is constructed. Using a baseline thrust model as the core, a degradation characterization subnetwork (DRN) and a condition embedding subnetwork (GRN) are integrated to jointly model the thrust and fuel flow of the engine under different flight altitudes, Mach numbers, intake temperatures, and long-term degradation states. This ensures that the prediction results simultaneously satisfy physical consistency and data adaptability, thereby significantly improving the accuracy and stability of thrust and fuel flow prediction. Specifically, an operating condition correction network models external operating conditions such as flight altitude (ALT), Mach number (M), and total air temperature (TAT), enabling continuous modulation of the baseline thrust. Furthermore, a degradation characterization network performs deep feature extraction on key degradation-sensitive parameters such as engine speed (N1, N2) and exhaust gas temperature (EGT) to obtain correction factors. This is used to characterize the evolution of engine performance over its lifespan; based on this, and By introducing a baseline thrust model, the final predicted thrust is obtained. Furthermore, the corresponding predicted engine fuel flow rate is calculated through a physical constraint fuel model. By introducing a loss function based on the joint optimization of physical consistency constraints and data errors, the model can strictly meet the constraints of engine thermodynamics and fuel mechanism while fitting historical data. This ensures that the model maintains high prediction reliability and engineering interpretability even when there are insufficient samples, drastic changes in operating conditions, or significant degradation.
[0115] VI. System Structure
[0116] See appendix Figure 4 The present invention also provides an aircraft thrust prediction system, comprising:
[0117] A fuel flow prediction system based on physical constraints and data-driven methods includes a data acquisition module, a baseline thrust calculation module, a degradation correction module, a fusion calculation module, a physical constraint module, and a fuel flow calculation module that operate sequentially.
[0118] The data acquisition module is used to acquire the aircraft's flight operation data, which includes at least engine parameters and flight condition parameters.
[0119] The baseline thrust calculation module is used to input flight operation data into the physical engine performance model and output the baseline thrust corresponding to the flight state.
[0120] The degradation correction module constructs a degradation-aware multiplicative residual correction model based on flight operation data, and outputs a correction factor to characterize the degree of engine degradation. The correction factor is then fused with a weight function output by the operating condition confidence modulation network to form an effective correction factor, which is used as the final correction factor. The reference thrust is then corrected according to the correction factor to obtain the corrected reference thrust.
[0121] The fusion computing module constructs a fusion weight function based on operating condition perception, and performs weighted fusion of the physical engine performance model and the degradation perception multiplicative residual correction model to obtain the predicted thrust.
[0122] The physical constraint module constructs a physical consistency constraint loss function based on the longitudinal dynamics of the aircraft to constrain the predicted thrust;
[0123] The fuel flow calculation module is used to calculate the engine fuel flow based on the predicted thrust.
[0124] This invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements a fuel flow prediction method based on physical constraints and data-driven methods as described in the method embodiments.
[0125] The present invention provides a computer-readable storage medium that can implement the steps and effects of the fuel flow prediction method based on physical constraints and data-driven in the above-described method embodiments. To avoid repetition, the present invention will not repeat them.
[0126] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.
[0127] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.
[0128] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.
Claims
1. A fuel flow prediction method based on physical constraints and data-driven approaches, characterized in that, The method includes: S1: Acquire the aircraft's flight operation data, which includes at least engine parameters and flight condition parameters; S2: Input flight operation data into the physical engine performance model and output the reference thrust corresponding to the flight state; S3: Based on flight operation data, construct a degradation perception multiplicative residual correction model and output a correction factor to characterize the degree of engine degradation; wherein, the correction factor is fused with the weight function output by the subsequent operating condition confidence modulation network to form an effective correction factor, and the effective correction factor is used as the final correction factor. S4: Correct the reference thrust according to the correction factor to obtain the corrected reference thrust; S5: Construct a fusion weight function for condition perception, and perform weighted fusion of the physical engine performance model and the degradation perception multiplicative residual correction model to obtain the predicted thrust; S6: Construct a physical consistency constraint loss function based on the longitudinal dynamics of the aircraft to constrain the predicted thrust; S7: Engine fuel flow rate is calculated based on predicted thrust.
2. The fuel flow prediction method based on physical constraints and data-driven approach according to claim 1, characterized in that: Engine parameters include at least: engine low-pressure speed N1, high-pressure speed N2, and exhaust temperature EGT. Flight operating parameters include at least: flight altitude ALT, Mach number M, throttle lever angle TLA, and total temperature TAT.
3. The fuel flow prediction method based on physical constraints and data-driven approach according to claim 2, characterized in that: The degradation-aware multiplicative residual correction model is a degradation-condition decoupled dual-channel multiplicative residual neural network model, and its output correction factor is a dimensionless scalar. It is used to characterize the multiplicative deviation relationship between the reference thrust output by the physical engine performance model and the actual thrust. The degradation-aware multiplicative residual correction model includes a degradation characterization subnetwork DRN and a condition embedding subnetwork GRN.
4. The fuel flow prediction method based on physical constraints and data-driven approach according to claim 3, characterized in that: The Degradation Characterization Network (DRN) is used to characterize the long-term performance degradation characteristics of an engine. Its input consists of engine degradation-related features. DRN employs a small residual fully connected network structure, consisting of one hidden layer and one residual connection. The hidden layer has 6 neurons, and the activation function is tanh. The mapping relationship is as follows: ； ； The operational condition embedding subnetwork GRN is used to extract the embedding features of the thrust modulation effect of flight operational conditions. Its input is... Engine operating condition related characteristics: ; GRN also employs a small residual fully connected structure, with 8 neurons in the hidden layer, using tanh as the activation function, and outputting a working condition embedding vector. ; Among them, ℎ d For the state vector in the degenerate representation network, 𝑏 d This is the bias vector in the degenerate representation network, used to translate the linear transformation results in fully connected layers to enhance the model's expressive power. d For degenerate embedding vectors, used to characterize engine health status, 𝑊 d Here, is the learnable weight matrix in the degenerate representation network, used to achieve a linear transformation on the input degenerate feature vector. 𝑟 It is a learnable weight matrix in the residual mapping module, used to implement nonlinear residual transformation of input features.
5. The fuel flow prediction method based on physical constraints and data-driven approach according to claim 4, characterized in that: Output a correction factor to characterize the degree of engine degradation. The basis is: degenerate embedding vector With working condition embedding vector The data is stitched together in the fusion layer and mapped to the original residuals. : ； The correction factor is obtained through the restricted Sigmoid function: ； in, Indicates hyperparameters and This is the weight matrix of the fusion layer, used to weight the degenerate embedding vector. With working condition embedding vector The concatenated fused feature vector is linearly mapped to the original degenerate parameters in scalar form. , Let represent the bias vector of the fully connected layer, which is the bias vector of the weight matrix. The corresponding trainable parameters are used to introduce learnable offsets after the feature linear transformation.
6. The fuel flow prediction method based on physical constraints and data-driven approach according to claim 5, characterized in that: Constructing a fusion weight function for operational condition awareness The formula used to characterize the reliability of a physical engine performance model under different operating conditions is as follows: ； in It is a prior confidence function based on flight conditions, used to describe the empirical estimate of the thrust confidence of a physical baseline model under given altitude and Mach number conditions. `logit` represents the inverse function of the sigmoid function. Indicates the confidence level. For learnable perturbation terms; ε is obtained through a lightweight gated residual network: ； Used to make small adaptive corrections to prior confidence levels. , for Activation function Input for a lightweight gated residual network; Forming an effective correction factor The basis is: .
7. The fuel flow prediction method based on physical constraints and data-driven approach according to claim 6, characterized in that: The basis for deriving the corrected baseline thrust is as follows: multiply the final correction factor by the baseline thrust to obtain the corrected baseline thrust; thus, the predicted thrust is obtained. The basis is: ； ； in, Indicates predicted thrust. Indicates the reference thrust. Indicates the confidence level. This represents the correction factor.
8. The fuel flow prediction method based on physical constraints and data-driven approach according to claim 7, characterized in that: The physical consistency constraint loss function is constructed based on Newton's second law and is used to constrain the dynamic relationship between predicted thrust and aircraft mass, aerodynamic drag, and flight acceleration. The specific formula for calculating the physical consistency constraint loss function is as follows: ； in: ； ； ； In the above formula, Represents the loss function for physical consistency constraints. This represents the mean square error. Indicates predicted thrust. Indicates aerodynamic drag. Indicates the mass of the aircraft. Represents gravitational acceleration. Indicates the flight path angle. Indicates flight acceleration. Indicates the drag coefficient. Indicates dynamic pressure. Indicates wing area. Indicates air density, Indicates vacuum velocity, This indicates the aircraft's ground speed.
9. The fuel flow prediction method based on physical constraints and data-driven approach according to claim 1, characterized in that: The basis for obtaining engine fuel flow rate is: ； in, Indicates engine fuel flow rate. Indicates predicted thrust. Specific impulse is the thrust generated per unit weight of fuel flow. It represents the acceleration due to gravity.
10. A fuel flow prediction system based on physical constraints and data-driven methods, characterized in that, The system includes: The data acquisition module is used to acquire the aircraft's flight operation data, which includes at least engine parameters and flight condition parameters. The baseline thrust calculation module is used to input flight operation data into the physical engine performance model and output the baseline thrust corresponding to the flight state. The degradation correction module constructs a degradation-aware multiplicative residual correction model based on flight operation data, and outputs a correction factor to characterize the degree of engine degradation. The correction factor is then fused with a weight function output by the operating condition confidence modulation network to form an effective correction factor, which is used as the final correction factor. The reference thrust is then corrected according to the correction factor to obtain the corrected reference thrust. The fusion computing module constructs a fusion weight function based on operating condition perception, and performs weighted fusion of the physical engine performance model and the degradation perception multiplicative residual correction model to obtain the predicted thrust. The physical constraint module constructs a physical consistency constraint loss function based on the longitudinal dynamics of the aircraft to constrain the predicted thrust; The fuel flow calculation module is used to calculate the engine fuel flow based on the predicted thrust.