An on-line trajectory planning method, system and storage medium for an ascent phase of a combined propulsion aerospace vehicle

Abstract

This application discloses an online trajectory planning method, system, and storage medium for the ascent phase of a combined-propulsion aerospace vehicle, belonging to the field of aircraft guidance and control technology. The method includes: offline generation of a large number of trajectory samples under parameter deviation conditions within the framework of the direct method, forming a dataset for network training; dividing the ascent phase of the combined-propulsion aerospace vehicle into a turbine-based combined cycle propulsion (TBCC) phase and a liquid rocket propulsion (RKT) phase based on the differences in the dynamic characteristics of the engine modes, and designing corresponding state-control networks for each, followed by training to obtain a segmented network; using the segmented network, outputting the control variables required at the current moment, providing the corresponding flight strategy, and realizing online trajectory planning. Numerical simulation results of this application verify that the method can maintain high terminal accuracy and strong robustness under different thrust deviation scenarios while ensuring the accuracy and real-time performance of online applications.

Claims

1. A method for online trajectory planning during the ascent phase of a combined-powered aerospace vehicle, characterized in that, The method includes: (1) Based on the parameter bias model, a large number of trajectory samples under parameter bias conditions are generated offline under the direct method framework to form a sample set for offline network training. (2) Based on the difference in dynamic characteristics of engine modes, the ascending stage of the combined propulsion aerospace vehicle is divided into the turbine-based combined cycle propulsion stage TBCC and the liquid rocket propulsion stage RKT. Corresponding state-control networks are designed for each stage, and the segmented networks are obtained by offline training using the sample set in step (1). (3) Using the trained segmented network, control commands are calculated based on the current flight status, and corresponding flight strategies are given to realize online trajectory planning.

2. The method for online trajectory planning during the ascent phase of a combined-powered aerospace vehicle according to claim 1, characterized in that, Step (1) includes: (11) The trajectory planning problem of the ascent phase of the combined propulsion aerospace vehicle is transformed into a nonlinear programming problem, and the problem is solved to obtain the optimal trajectory of the ascent phase; (12) Considering the aerodynamic uncertainty and thrust deviation during the actual flight process in the ascent phase, a parameter deviation model is established, the aerodynamic coefficient and thrust are adjusted, and a large number of trajectory samples are generated. (13) Network samples are obtained based on a large number of trajectory samples and divided into training and test sets for segmented networks.

3. The method for online trajectory planning during the ascent phase of a combined-powered aerospace vehicle according to claim 2, characterized in that, In step (11), the SQP algorithm is used to solve the nonlinear programming problem; Preferably, the optimal trajectory of the rising segment includes state variable and control variable curves; Preferably, the trajectory optimization mathematical model for the ascent phase trajectory planning problem of the combined-powered aerospace vehicle includes: constructing an ascent phase dynamics model and selecting control variables, constraints, and optimization objectives; Preferably, the construction of the ascending segment dynamic model includes: Considering that the aircraft's ascent phase motion mainly remains in the longitudinal plane, and choosing the launch coordinate system as the reference coordinate system for this phase, the equation of motion of its center of mass in the longitudinal plane can be expressed as: In the formula, h and L r v, θ, m, R e α, g, r, g0, I sp P, D, and L represent the aircraft's flight altitude, flight range, speed, trajectory inclination, aircraft mass, Earth's radius, angle of attack, gravitational acceleration, distance from the Earth's center, gravitational acceleration at sea level, combined engine fuel consumption per second, combined engine thrust, aerodynamic drag, and aerodynamic lift in the longitudinal plane, respectively. Preferably, the formula for calculating the magnitude of gravitational acceleration is: Preferably, the formulas for calculating aerodynamic lift and aerodynamic drag are as follows: In the formula, dynamic pressure q = 0.5ρv is defined. 2 ρ represents atmospheric density, S ref C represents the reference area of ​​the aircraft. L and C D Let represent the lift coefficient and drag coefficient, respectively. Both the lift parameter and the drag coefficient can be expressed as bivariate functions of the angle of attack α and the Mach number Ma, and their expressions are as follows: Preferably, the coefficients in the bivariate function expressions for the angle of attack α and the Mach number Ma are obtained by interpolation, where C L0 C is the first lift coefficient. Lα C is the second lift coefficient. D0 C is the first drag coefficient. Dα This is the second drag coefficient; Preferably, the thrust of the combined-power engine includes the thrust of the TBCC combined-power aerospace vehicle of the first stage in the two-stage-to-orbit flight scheme; Preferably, the thrust model of the first-stage TBCC combined-propellant aerospace vehicle is as follows: In the formula, I represents the rate of change of aircraft mass, and also the rate of fuel consumption. sp Indicates the engine's specific impulse; Preferably, for an air-breathing engine, the combined power engine's fuel consumption per second... It can be represented as: In the formula, C s φ represents the conversion factor, which is a constant. t This represents the engine's throttling coefficient, characterizing fuel control and reflecting the fuel supply rate. The more fuel supplied, the greater the specific impulse. (C) A S is the equivalent intake coefficient. c This refers to the air intake area; Preferably, the engine specific impulse I sp and equivalent intake coefficient C A Let the thrust be expressed as a bivariate function of angle of attack α and Mach number Ma, with both coefficients calculated through interpolation. Then, the complete thrust model of the first-stage TBCC combined-propulsion aerospace vehicle is: P=C s φ t AND sp (Ma,φ t )g0ρνC A (Ma,α)S c ; Preferably, the control variable is: In the formula, the rate of change of angle of attack As a pseudo-control variable, the angle of attack α is considered as an augmented state variable, and φ t Indicates the engine throttle coefficient; Preferably, the constraints include process constraints, state variable constraints, control variable constraints, and boundary condition constraints; Preferably, the process constraints include heat flux density constraints, dynamic pressure constraints, and overload constraints, calculated using the following formulas: In the formula, K represents heat flux density. h ρ represents the heat flux density coefficient, ρ represents the atmospheric density, and v represents the flight speed of the aircraft in the longitudinal plane. q max and n max These represent heat flux density, dynamic pressure, and permissible overload peak values, respectively; α is the angle of attack; and D and L are the aerodynamic drag and aerodynamic lift, respectively. Preferably, the constraints include state variable constraints, wherein the state variables include altitude, range, speed, trajectory inclination, and mass, denoted as x = [h, L]. r ,v,θ,m] T Then the state variable constraint is: x min ≤x≤x max , where x min Let x represent the minimum value of the state variable. max This represents the maximum value of the state variable; Preferably, the constraints further include control variable constraints: taking into account flight performance, control system performance, and engine performance, the control variable constraints are as follows: in min Oh, oh. max Among them, u min u represents the minimum value of the control variable. max This indicates the maximum value of the control variable; Preferably, the constraints further include boundary condition constraints: since the initial state of the aircraft is determined, and the flight mission constrains the terminal state of the ascent phase, the boundary condition constraints are expressed as follows: h(t0)=h0,L r (t0)=L r0 ,ν(t0)=ν0,θ(t0)=θ0,m(t0)=m0 h(t f )=h f ,L r (t f )=L rf ,ν(t f )=n f ,θ(t f )∈[θ f,min ,i f,max ] Where t0 represents the initial time, t f h represents the terminal time, h0 represents the initial height value, and h f L represents the height value at the terminal moment. r0 L represents the distance traveled at the initial moment. rf v represents the distance traveled at the terminal moment, v0 represents the speed at the initial moment, and v f θ represents the velocity value at the terminal moment, θ0 represents the trajectory inclination angle at the initial moment, and θ f The trajectory inclination angle at the terminal moment is represented by m0, and the mass value at the initial moment is represented by θ. f,min θ represents the minimum trajectory inclination angle at the terminal moment. f,max This represents the maximum value of the trajectory inclination angle at the terminal moment; Preferably, the optimization objective is to minimize fuel consumption, and the performance indicators are expressed as follows: J=min(-m(t f )) Where J represents the performance index function, m(t) f This indicates the quality at the terminal moment; Preferably, in step (11), the conversion of the ascent trajectory planning problem of the combined propulsion aerospace vehicle into a nonlinear programming problem is performed using the optimal control toolbox GPOPSII; Preferably, the solution to the nonlinear programming problem is achieved by calling the sparse NLP solver SNOPT based on the SQP algorithm for numerical solution, thereby generating the optimal trajectory of the combined propulsion aerospace vehicle during the ascent phase. The parameter deviation model includes six deviation coefficients: the same deviation coefficients are used for the lift and drag coefficients of first and second stage flight. Each deviation coefficient is reflected in the dynamic equation in a proportional form, and the lift deviation coefficient is n. L The drag deviation coefficient is n D The thrust deviation coefficient is n P ; Preferably, within the Gaussian 3σ probability range, the thrust deviation coefficient n P Within the range of [-10%, 10%]; Preferably, within the Gaussian 3σ probability range, the lift deviation coefficient n D Within the range of [-20%, 20%]; Preferably, within the Gaussian 3σ probability range, the drag deviation coefficient n L Within the range of [-20%, 20%]; Preferably, the generation of the large number of trajectory samples includes: based on the parameter deviation model, randomly generating non-nominal parameter conditions, and generating multiple non-nominal ascending segment trajectories under different parameter deviation conditions to form trajectory samples; Preferably, obtaining network samples from a large number of trajectory samples includes: For the optimal trajectory under nominal conditions and the non-nominal trajectory obtained under randomly generated non-nominal parameter conditions, each trajectory contains 4 state variable curves (h, L). r (v,θ) and two control variable curves (α,φ) t The network sample is obtained by sampling the trajectory of the curve at a certain moment. Each network sample contains a 4-dimensional state variable and a 2-dimensional control variable at a sampling moment, corresponding to the input x of the piecewise network. input =[h,L r ,v,θ] T The output is x output =[α,φ t ] T Among them, h and L r v, θ, α, φ t These represent the aircraft's flight altitude, flight range, speed, trajectory inclination, angle of attack, and engine throttle coefficient in the longitudinal plane, respectively. Preferably, the division of the training and test sets of the segmented network includes: randomly dividing the non-nominal trajectory into the training and test sets according to a given ratio, and assigning an optimal nominal trajectory to the test set.

4. The method for online trajectory planning during the ascent phase of a combined-powered aerospace vehicle according to claim 1, characterized in that, The state-control network includes: a Transformer-BiGRU network designed in the TBCC segment, which combines the structure of Transformer and BiGRU bidirectional gated cyclic network; wherein, BiGRU enhances the GRU's ability to capture contextual information through bidirectional structure, and the Transformer-BiGRU network has higher flexibility and expressive power, used to process the time series data of complex aerodynamic characteristics and thrust coupling in the TBCC segment, and capture long-distance dependencies; Preferably, the Transformer-BiGRU network specifically includes a multi-head attention mechanism module, a multi-layer BiGRU module, and a fully connected layer module arranged sequentially. The multi-head attention mechanism module is used to capture long-distance dependencies in the input sequence; The multi-layer BiGRU module is used to extract bidirectional temporal features; The fully connected layer module is used to map the extracted features to the control variable space; Preferably, the loss function of the state-control network of the TBCC segment is the mean absolute error, the optimizer adopts the Adam algorithm, the initial learning rate is 0.001, and the learning rate is dynamically adjusted by the learning rate scheduler; Preferably, step (2) further includes: before offline training of the model, the input and output data of the TBCC segment are normalized using the minimum-maximum normalization method; Preferably, the state-control network further includes: designing a BiGRU network in the RKT segment; Preferably, the loss function of the RKT segment state-control network is the mean absolute error, and the optimizer uses the Adam algorithm; Preferably, the output of the segmented network includes the angle of attack α and the engine throttle coefficient φ. t .

5. The method for online trajectory planning during the ascent phase of a combined-powered aerospace vehicle according to claim 3, characterized in that, Step (3) includes: generating control variables for the current moment based on the trained segmented network, providing corresponding flight strategies, and realizing online trajectory planning; Preferably, step (3) specifically includes: The Runge-Kutta method is used to simulate online trajectory planning. The output of the segmented network is used as the input command for Runge-Kutta numerical integration. Starting from the initial state, the nonlinear differential equation in the ascending phase dynamic model is recursively integrated. In the next iteration, a new state variable is output, and the new state variable is used as the new input to the network. The network output at this time corresponds to the current control command. This iterative integration is performed to simulate the flight trajectory of the entire ascending phase.

6. An online trajectory planning system for the ascent phase of a combined-powered aerospace vehicle, characterized in that, The system includes: The trajectory sample generation module, based on the parameter deviation model, generates a large number of trajectory samples offline under parameter deviation conditions within the direct method framework, forming a sample set for offline network training. The network training module is used to divide the ascent phase of the aircraft into the TBCC segment and the RKT segment, which serve as the primary and secondary platforms of the combined propulsion aerospace vehicle, respectively. Corresponding state-control networks are designed and the segmented networks are obtained through offline training using sample sets. The online planning module is used to collect flight status data, output control variables through a trained segmented network, and formulate flight strategies online.

7. The online trajectory planning system for the ascent phase of a combined-powered aerospace vehicle according to claim 6, characterized in that, The trajectory sample generation module is used to transform the trajectory planning problem of the ascent phase of a combined-powered aerospace vehicle into a nonlinear programming problem, solve the problem to obtain the optimal trajectory of the ascent phase, establish a parameter deviation model, adjust the aerodynamic coefficient and thrust, and generate a large number of trajectory samples; obtain network samples based on the large number of trajectory samples, and divide them into training and testing sets for segmented networks. Preferably, the trajectory optimization mathematical model for the ascent phase trajectory planning problem of the combined-powered aerospace vehicle includes: constructing an ascent phase dynamics model and selecting control variables, constraints, and optimization objectives; Preferably, the construction of the ascending segment dynamic model includes: Considering that the aircraft's ascent phase motion mainly remains in the longitudinal plane, and choosing the launch coordinate system as the reference coordinate system for this phase, the equation of motion of its center of mass in the longitudinal plane can be expressed as: In the formula, h and L r v, θ, m, R e α, g, r, g0, I sp P, D, and L represent the aircraft's flight altitude, flight range, speed, trajectory inclination, aircraft mass, Earth's radius, angle of attack, gravitational acceleration, distance from the Earth's center, gravitational acceleration at sea level, combined engine fuel consumption per second, combined engine thrust, aerodynamic drag, and aerodynamic lift in the longitudinal plane, respectively. Preferably, the parameter deviation model includes: taking the same deviation coefficients for the lift and drag coefficients of first and second stage flight, for a total of 6 deviation coefficients, each of which is reflected in the dynamic equation in a proportional form, with the lift deviation coefficient being n. L The drag deviation coefficient is n D The thrust deviation coefficient is n P ; Preferably, within the Gaussian 3σ probability range, the thrust deviation coefficient n P Within the range of [-10%, 10%]; Preferably, within the Gaussian 3σ probability range, the lift deviation coefficient n D Within the range of [-20%, 20%]; Preferably, within the Gaussian 3σ probability range, the drag deviation coefficient n L Within the range of [-20%, 20%]; Preferably, when the trajectory sample generation module obtains network samples based on a large number of trajectory samples, it performs the following steps: For the optimal trajectory under nominal conditions and the non-nominal trajectory obtained under randomly generated non-nominal parameter conditions, each trajectory contains 4 state variable curves (h, L). r (v,θ) and two control variable curves (α,φ) t The network sample is obtained by sampling the trajectory of the curve at a certain moment. Each network sample contains a 4-dimensional state variable and a 2-dimensional control variable at a sampling moment, corresponding to the input x of the piecewise network. input =[h,L r ,v,θ] T The output is x output =[α,φ t ] T Among them, h and L r v, θ, α, φ t These represent the aircraft's altitude, range, speed, trajectory angle, angle of attack, and engine throttle coefficient in the longitudinal plane, respectively.

8. The online trajectory planning system for the ascent phase of a combined-powered aerospace vehicle according to claim 6, characterized in that, The network training module includes state-control networks designed for the TBCC and RKT segments respectively. The TBCC segment employs a Transformer-BiGRU network, which combines the structures of Transformer and BiGRU (Bidirectional Gated Recurrent Network). The BiGRU, through its bidirectional structure, enhances the GRU's ability to capture contextual information. The Transformer-BiGRU network offers greater flexibility and expressive power, enabling it to process the complex aerodynamic characteristics and thrust coupling time-series data of the TBCC segment and capture long-range dependencies. Preferably, the Transformer-BiGRU network specifically includes a multi-head attention mechanism module, a multi-layer BiGRU module, and a fully connected layer module arranged sequentially. The multi-head attention mechanism module is used to capture long-distance dependencies in the input sequence; The multi-layer BiGRU module is used to extract bidirectional temporal features; The fully connected layer module is used to map the extracted features to the control variable space; Preferably, the loss function of the state-control network of the TBCC segment is the mean absolute error, the optimizer adopts the Adam algorithm, the initial learning rate is 0.001, and the learning rate is dynamically adjusted by the learning rate scheduler; Preferably, the network training module is further configured to normalize the input and output data of the TBCC segment using a minimum-maximum normalization method before offline training of the model; Preferably, the RKT segment uses a BiGRU network; Preferably, the output of the network training module is a control variable, including the angle of attack α and the engine throttle coefficient φ. t .

9. The online trajectory planning system for the ascent phase of a combined-powered aerospace vehicle according to claim 6, characterized in that, The online planning module is used to request the segmented network trained by the network training module to generate the control variables for the current moment, output the control variables required for the current moment based on the trained segmented network, and provide the corresponding flight strategy to realize online trajectory planning.

10. A storage medium, characterized in that, The storage medium stores computer-executable instructions, which, when loaded and executed by a processor, implement the steps of the online trajectory planning method for the ascent phase of a combined-powered aerospace vehicle as described in any one of claims 1 to 5.

Images