Short-distance take-off and landing aircraft pitching moment compensation method based on fuzzy neural network

Abstract

The invention belongs to the technical field of flight control, and relates to a short-distance take-off and landing aircraft pitching moment compensation method based on a fuzzy neural network. According to the method, a reference pitching moment instruction is generated by designing a control law, special fuzzy neural network compensators are constructed for the cruising stage and the vertical landing stage respectively, and pitching moment errors caused by atmospheric turbulence and the ground effect are estimated and compensated online. In the actual flight process, external disturbance is inevitable, so that the pitching moment compensation design of the fuzzy neural network is combined, and the original control strategy is enhanced and optimized. Compared with a traditional single control method, the provided fuzzy neural network composite control method has remarkable anti-interference capability and higher control precision, and shows good engineering application prospects.

Description

Technical Field

[0001] This invention belongs to the field of flight control technology and relates to a method for pitch moment compensation for short takeoff and landing aircraft based on fuzzy neural networks. Background Technology

[0002] Short Take-Off and Vertical Landing (STOVL) aircraft, due to their unique configuration, are capable of taking off and landing in confined spaces, giving them significant military value. However, STOVL aircraft have a wide flight envelope and complex dynamic characteristics, presenting different control challenges in both the cruise and vertical landing phases.

[0003] During the cruise phase, the aircraft flies in fixed-wing mode, primarily subjected to aerodynamic disturbances such as atmospheric turbulence and gusts. These disturbances exhibit significant randomness and uncertainty, directly causing continuous oscillations in the aircraft's attitude, particularly the pitch angle, severely impacting flight stability, the targeting accuracy of mission equipment, and crew comfort. In the more complex vertical descent phase, the aircraft enters and hovers in a planar environment, where it is subjected to strong, nonlinearly changing "ground effects." These factors collectively cause drastic changes in the flow field around the aircraft, resulting in highly nonlinear and time-varying aerodynamic parameters, making it extremely difficult to establish accurate dynamic models. Traditional control methods based on accurate models, such as linear quadratic regulators (LQRs) or classic PID control, will produce significant control errors under such model mismatch conditions, potentially leading to control system performance degradation or even instability, posing a direct threat to aircraft landing safety.

[0004] Existing technologies have explored solutions for near-ground flight and multi-stage disturbance compensation. Patent CN202510649132.5 discloses a "Control Method for Rotary Low-Altitude Aircraft Based on Active Disturbance Rejection (ADDR) for Ground Effect," which employs a multi-channel cooperative architecture combining ADDR and PID control, estimating and compensating for system disturbances through an extended state observer. However, this method primarily addresses near-ground hovering conditions for rotary-wing aircraft. Its control framework and model assumptions do not cover the complete flight envelope of STOVL aircraft from cruise to vertical landing, especially the high-frequency, random disturbances such as atmospheric turbulence encountered during the cruise phase. More importantly, this method fails to design differentiated compensation strategies for the drastically different physical characteristics of disturbance sources during the cruise and landing phases, thus limiting its adaptive compensation capabilities across all stages of STOVL aircraft.

[0005] Patent CN202511676802.9 discloses an "Adversarial Reinforcement Learning Training Method for Robust Control of Fixed-Wing Aircraft," which employs an adversarial reinforcement learning framework to generate perturbations through adversarial agents to improve policy robustness. However, this method relies on large-scale offline training and high-dimensional policy optimization, resulting in high computational complexity and difficulty in guaranteeing real-time performance. For high-frequency control tasks requiring millisecond-level response, such as dynamic compensation of pitch moment in STOVL aircraft, the online application of this method faces significant challenges.

[0006] In recent years, fuzzy logic and neural networks, as representatives of intelligent control methods, have demonstrated unique advantages in handling nonlinearity and uncertainty. Fuzzy logic excels at utilizing expert-language experiential knowledge, independent of precise mathematical models of the object; neural networks possess powerful nonlinear mapping and self-learning approximation capabilities. However, simply applying fuzzy logic or neural networks to the control of STOVL aircraft still has significant shortcomings: the formulation of fuzzy rules heavily relies on expert experience and is difficult to optimize adaptively online; neural networks, due to their "black box" nature, suffer from unclear physical meaning, sometimes limiting their generalization ability beyond training data, and making it difficult to directly embed prior knowledge. Therefore, combining the interpretability and knowledge embedding capabilities of fuzzy systems with the self-learning and adaptive capabilities of neural networks to construct an intelligent compensation architecture that can clearly express physical logic and can be optimized and adjusted online has become crucial for improving the all-stage control performance of STOVL aircraft.

[0007] For STOVL aircraft, the sources and mechanisms of disturbances during cruise and vertical landing phases are fundamentally different. Disturbances during cruise primarily originate from the atmosphere and are related to altitude and airspeed; while disturbances during vertical landing are mainly caused by the moving ground and are closely coupled to altitude. Designing a pitch moment compensation method that can adapt to both cruise and vertical landing phases and possess online adaptive capabilities to effectively address their respective nonlinear disturbances is a critical problem that urgently needs to be solved in current STOVL aircraft flight control technology. Summary of the Invention

[0008] The purpose of this invention is to provide a pitch moment compensation method for short takeoff and landing (STOVL) aircraft based on fuzzy neural networks. The core of this method lies in constructing an adaptive fuzzy neural inference system as an intelligent compensator framework. Dedicated sets of input variables and fuzzy rule bases are configured for the disturbance characteristics of the two typical phases of cruise and vertical landing for STOVL aircraft. Through offline training and online application, precise compensation for specific disturbance moments is achieved. This method can adaptively compensate online for pitch moment errors caused by complex disturbances during the cruise and vertical landing phases of STOVL aircraft, thereby significantly improving the robustness and accuracy of the control system.

[0009] The technical solution of the present invention is as follows:

[0010] A method for pitch moment compensation for short takeoff and landing aircraft based on fuzzy neural networks includes the following steps:

[0011] Step 1: Generate reference pitch moment command

[0012] The pitch attitude of an aircraft directly affects its longitudinal stability and landing trajectory. Nonlinear Dynamic Inversion (NDI) is a control method based on the system's inverse dynamic model. It generates precise control commands for the pitch rate by linearizing a complex nonlinear system in real time. Compared to traditional control methods, NDI is better able to handle nonlinear dynamic characteristics and external disturbances.

[0013] The attitude control law is designed using the NDI method theory. First, the equations for the pitch rate inner loop are given:

[0014] (1)

[0015] In the formula, For pitch rate, The rate of change of pitch angle; For pitching moment, , The pitching moment generated by the aerodynamic control surfaces The pitching moment generated by the power system; This is the pitch moment of inertia.

[0016] Based on the state-space form and the theory of NDI, equation (1) can be rewritten as:

[0017] (2)

[0018] (3)

[0019] In the formula, For the state variables in the pitch rate inner loop equation, The rate of change of this state variable, This is the control input for the loop. and This is a function describing the nonlinear dynamic characteristics of the system in the inner loop equation of the pitch rate. The dynamic characteristics under uncontrolled conditions. The state gain characteristic after implementing control action. When designing the inner-loop control law, in equation (2)... The desired pitch acceleration is calculated using the slow loop, and its angular acceleration command is obtained as follows:

[0020] (4)

[0021] In the formula, This represents the current pitch rate state. To correspond to the rate of change of pitch angle, The bandwidth of the pitch angular rate channel. It is the pitch rate command.

[0022] According to the dynamic inverse method, in order to obtain the desired form of equation (4), the fast-loop dynamic inverse control law expression is:

[0023] (5)

[0024] The slow loop is the attitude angle loop, located outside the inner loop. Its input is the output of the outermost command generator, and its output is the input signal of the inner loop controller. Equation (6) yields the equation set of the slow loop attitude angle loop:

[0025] (6)

[0026] In the formula, The pitch angle, For pitch angular velocity, equation (6) can also be rewritten as:

[0027] (7)

[0028] (8)

[0029] In the formula, These are the state variables in the attitude angle loop equations. The rate of change of this state variable, This is the control input for the loop. and is a function describing the nonlinear dynamic characteristics of the system in the attitude angle loop equations. The dynamic characteristics under uncontrolled conditions. To determine the state gain characteristics after implementing control, when designing the outer-loop control law, The expected value is determined by external instructions. The calculation yields the following expression:

[0030] (9)

[0031] In the formula, This is the current pitch angle. The rate of change of the pitch angle. This is the external pitch angle command. The rate of change of the external pitch angle command. Let be the bandwidth of the pitch angle channel. To obtain the desired form of equation (9), the expression for the slow-loop dynamic inverse control law is:

[0032] (10)

[0033] This led to the time-scale separation of the pitch channel control, the design of inner and outer loop NDI control laws, and their cascading, resulting in the expressions for each level of control law and the generation of the reference pitch moment command.

[0034] In actual flight, aerodynamic parameter perturbations and complex external disturbances (such as turbulence during cruise and ground effects during vertical descent) can lead to model mismatch, making the reference torque command generated by the above methods insufficient to completely offset the actual disturbance torque, thus affecting the control accuracy of pitch attitude. Therefore, based on obtaining accurate reference control, this invention further introduces an intelligent compensation mechanism to adaptively estimate and compensate for the pitch torque error caused by the above factors.

[0035] Step 2: Construction of the Fuzzy Logic System

[0036] The fuzzy logic system is a five-layer feedforward network, including an input layer, a fuzzification layer (using Gaussian membership functions), a rule layer, a normalization layer, and an output layer (linear functions). This framework serves as a general inference mechanism for intelligent compensators, capable of handling nonlinear and uncertain systems, and provides a structural foundation for subsequent staged analysis and rule training.

[0037] Step 2.1: Input Layer: Receives input variables Including ground height Angle of attack ,airspeed Vertical wind speed Wait, and complete the data normalization.

[0038] Step 2.2: Fuzzification Layer: Each input variable corresponds to multiple Gaussian membership functions. The degree to which the input variable belongs to each fuzzy set is calculated using the following formula:

[0039] (11)

[0040] In the formula, The membership function used for fuzzification. For the first The first input variable A fuzzy set, The center point of the fuzzy set The standard deviation of the width of the fuzzy set. Indicates the number of input variables. This represents the number of membership functions for each input.

[0041] Step 2.3: Rule Layer: Fuzzy rules are set based on the 3D mesh partitioning method. The number of rules is the product of the number of membership functions of each input variable. The activation degree of each rule is calculated using the algebraic product operator. The formula is:

[0042] (12)

[0043] In the formula, It is the total number of input variables. Indicates the first Input variables In the The corresponding fuzzy set in the rule The membership degree value on.

[0044] Step 2.4: Normalization layer: Calculate the normalized weights of the activation values ​​of each rule.

[0045] Step 2.5: Output Layer: Output pitch compensation torque through weighted linear combination, the formula is:

[0046] (13)

[0047] In equation (13), For pitch compensation torque; Normalized weights for rule activation; For rule consequent parameters.

[0048] Step 3: Staged Analysis and Strategies Based on Neural Networks

[0049] Based on the fuzzy logic system framework in step 2, staged analysis and strategy design are carried out for the interference characteristics of STOVL aircraft in different flight phases.

[0050] Step 3.1: Cruise Phase: Determine the input variable as flight altitude Angle of attack and vertical wind speed The output is the pitch compensation torque under cruise disturbance, and then the influence of different input states on the output pitch compensation torque is analyzed.

[0051] Step 3.2: Vertical Descent Phase: Determine the input variable as altitude above ground. Angle of attack and airspeed The output is the pitch compensation torque under ground effect disturbance, and then the influence of different input states on the output pitch compensation torque is analyzed.

[0052] Step 4: Design and training of fuzzy neural network rules

[0053] Based on the phased analysis results described in step 3, fuzzy rule bases are designed for both phases, and the fuzzy neural network is trained under supervision using the collected state-torque compensation sample data. By iteratively optimizing the membership function parameters and rule consequent parameters, the network can accurately approximate the nonlinear mapping relationship of the disturbance torque in each phase. After training, high-precision compensation models for two different phases are obtained, which can be directly embedded into the real-time control system to achieve online compensation of dynamic torque errors.

[0054] Step 5: System Integration and Application

[0055] The trained fuzzy neural network model is embedded into the flight control calculation. Based on the current flight phase, the corresponding compensation model is invoked. The compensation model calculates the pitch moment compensation value in real time online and superimposes this value onto the reference moment command generated by the dynamic inverse control law to form the final control command, thereby effectively canceling the inverse error.

[0056] The beneficial effects of this invention are:

[0057] The pitch moment compensation method for short takeoff and landing (STOVL) aircraft proposed in this invention is effective primarily in estimating and compensating for complex external disturbances. This method, through a fuzzy neural network mechanism, can accurately approximate and compensate in real time for nonlinear pitch moment changes caused by atmospheric turbulence during cruise or ground effects during vertical landing. The fusion architecture of fuzzy logic and neural networks enhances the system's tolerance to unmodeled dynamics and parameter perturbations, enabling it to exhibit high robustness and stability when facing complex disturbances. The offline-trained network has a low computational burden in its forward inference process, enabling high-precision calculations to be completed within a 20ms control cycle, ensuring that the relative error between pitch angle tracking and moment compensation is less than 5%. This design combines the physical logic of control laws with the approximation capability of complex systems, providing a theoretically rigorous and engineeringly practical solution for intelligent compensation of flight control systems. Attached Figure Description

[0058] Figure 1 This is a flowchart illustrating the overall implementation of the method of the present invention.

[0059] Figure 2 This is the control law design block diagram for the cruise phase.

[0060] Figure 3 This is a block diagram of the improved control law design for the vertical descent phase.

[0061] Figure 4 It is the effect of the angle of attack on the pitching moment coefficient during the cruise phase.

[0062] Figure 5 This refers to the effect of vertical wind speed on the pitching moment coefficient during the cruise phase.

[0063] Figure 6 It refers to the effect of altitude on dynamic pressure during the cruise phase.

[0064] Figure 7 It is the effect of the vertical descent phase altitude on the scaling factor.

[0065] Figure 8 It is the effect of the angle of attack on the pitching moment coefficient during the vertical descent phase.

[0066] Figure 9 It is the effect of dynamic pressure on pitching moment compensation during the vertical descent phase.

[0067] Figure 10 It is a fuzzy neural network regression analysis during the cruise phase.

[0068] Figure 11 This is a comparison of the prediction results of the fuzzy neural network test during the cruise phase.

[0069] Figure 12 It is a fuzzy neural network regression analysis of the vertical descent phase.

[0070] Figure 13 This is a comparison of the prediction results from the fuzzy neural network test during the vertical descent phase.

[0071] Figure 14 These are simulation results of pitch moment compensation during the cruise phase.

[0072] Figure 15 This is the simulation result of pitch moment compensation during the vertical descent phase. Detailed Implementation

[0073] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0074] The present invention discloses a pitch moment compensation method for short takeoff and landing aircraft based on a fuzzy neural network, the overall process of which is as follows: Figure 1 As shown. The specific steps are as follows:

[0075] Step 1: Generate reference pitch moment command

[0076] Step 1.1: Control Law Design for Cruise Phase

[0077] To address the control requirements during the cruise phase, the control laws for forward speed and altitude are designed separately during flight. The speed loop uses speed error control to determine the throttle thrust, while the altitude loop uses altitude error control to determine the pitch angle deflection. The pitch angle error and pitch rate are then used to derive the pitch rate command, which in turn generates the required pitch moment command. This control method ensures the stability of flight parameters. The specific design is as follows: Figure 2 As shown.

[0078] The speed loop control channel takes the error between the target flight speed and the current flight speed as input, and uses PID control to obtain the throttle position.

[0079] (14)

[0080] In the formula, The throttle position of the tail nozzle. This refers to the throttle position of the exhaust nozzle at the previous moment. This is the current forward speed state. Forward speed command, For time, For the gain parameters of the PID-controlled speed controller, the integral term... The following expression is used:

[0081] (15)

[0082] In the formula, This is the integral term for the previous forward speed state. The time for integration.

[0083] Based on the error between the current flight altitude and the desired flight altitude, PID control is used to first calculate the desired pitch angle. :

[0084] (16)

[0085] In the formula, Current flight altitude status This is a highly anticipated instruction. For time, For the gain parameters of the PID height controller, the integral term Expressed as follows:

[0086] (17)

[0087] In the formula, This is the integral term for the previous altitude state. The time for integration.

[0088] Then, according to the pitch angle command... Current pitch angle Pitch rate and desired pitch rate The pitch moment command is calculated using NDI control. :

[0089] (18)

[0090] Therefore, the control law research during cruise flight is designed to ensure the basic commands for stable tracking of altitude and speed during flight through both altitude loop and speed loop, and to obtain the reference pitch moment command for subsequent fuzzy neural network processing.

[0091] Step 1.2: Design of Control Laws for Vertical Descent Phase

[0092] During vertical descent, the system coordinates the combined application of the tail nozzle and lift fan, using thrust vectoring and throttle control to achieve flight maneuvering. The control law design alters the traditional control logic during cruise, avoiding slow response issues caused by indirect control between inner and outer loops. This step directly controls forward speed or altitude to obtain control commands. Subsequently, the obtained force and torque commands are used as input to a fuzzy neural network to compensate for pitch moment during vertical descent to solve for actuator deflection, thereby ensuring the safety and reliability of the aircraft during flight. The specific design is as follows... Figure 3 As shown.

[0093] Based on the longitudinal flight trajectory of the aircraft, control is performed using the aircraft's altitude error and forward speed error. PID control is employed to obtain the desired forward and longitudinal forces of the aircraft. and as follows:

[0094] (19)

[0095] In the pitch angle control section, the system employs a dynamic inverse control structure. The first layer, through a dynamic inverse controller, generates a pitch rate command based on the error between the pitch angle command and the actual pitch angle. The second layer, another dynamic inverse controller, further processes the pitch rate command and outputs a torque command to adjust the aircraft's attitude. The desired pitch torque input command for subsequent control allocation is obtained based on the pitch angle error control. :

[0096] (20)

[0097] In the formula, The gain parameters of the PID pitch angle controller are used to form a complete control system through multi-level coordination of altitude control, forward speed control, and pitch angle control, combined with a nonlinear dynamic control allocation strategy. This control strategy can improve the system's response performance and stability in complex dynamic environments, laying the foundation for smooth flight and landing of short vertical takeoff and landing aircraft.

[0098] Step 2: Construction of the Fuzzy Logic System

[0099] The aforementioned dynamic inverse control law provides the aircraft with a high-precision reference pitch moment command, forming the basis of attitude control. However, as mentioned earlier, random disturbances such as gusts and ground turbulence still exist, and these factors will cause a dynamically changing error term between the actual required pitch moment and the reference command. If this error is not compensated, it will directly manifest as tracking deviation and oscillation of the pitch angle. To solve this problem, this embodiment designs an intelligent compensator based on a fuzzy neural network on top of the reference control law to estimate and compensate for this moment error online. Specifically, two fuzzy neural network networks are designed and constructed for two typical flight modes of short takeoff and landing aircraft, and their design and implementation are described below.

[0100] Step 2.1: Construction of the Fuzzy Logic System during the Cruise Phase

[0101] Aircraft primarily face atmospheric turbulence and gusts, and their aerodynamic disturbance characteristics are closely related to altitude, angle of attack, and vertical wind speed. This phase aims to design a dedicated fuzzy neural network compensator for online estimation and compensation of disturbance moments during this phase. The inference process consists of five key stages:

[0102] First, we use mathematical methods to describe the input variables. The fuzzification process is as follows:

[0103] (twenty one)

[0104] In the formula, The membership function used for fuzzification. For the first The first input variable A fuzzy set, For the first The center point of a fuzzy set To determine the standard deviation of the fuzzy set width, The number of membership functions for each input.

[0105] The input layer nodes select three key state parameters that characterize aerodynamic disturbances during the cruise phase: flight altitude, flight altitude, and flight level. Angle of attack and vertical wind speed The output of the fuzzy neural network is the cruise pitch compensation torque. It is used to counteract pitch moment errors caused by disturbances.

[0106] Next, we will design the fuzzification of the input variables, using a three-dimensional mesh partitioning. The input variables and fuzzy sets are set as follows:

[0107] Table 1 Input Variable Settings for Cruise Phase

[0108]

[0109] Further, a fuzzy rule base was constructed, using a grid partitioning method to generate a complete rule base, including the total number of rules. Other consequent parameters are learned by a fuzzy neural network.

[0110] Then, the normalized weights of the activation values ​​of each rule are calculated in the normalization layer as follows:

[0111] (twenty two)

[0112] Finally, weighted average deblurring is performed, and the pitch moment compensation is output through a weighted linear combination, as shown in the formula:

[0113] (twenty three)

[0114] Step 2.2: Construction of the Fuzzy Logic System for the Vertical Descent Phase

[0115] During the vertical landing phase, the aircraft is affected by external disturbances caused by the ground effect, negatively impacting system performance. A dedicated fuzzy neural network compensator is designed to estimate and compensate for the disturbance torque during this phase online. The design process is the same as for the cruise phase. In the fuzzification layer, each input variable is defined with three fuzzy sets (e.g., "low", "medium", "high"), using a Gaussian membership function. The fuzzification process is shown in the table below. Fuzzy sets are defined for each input variable, and a Gaussian membership function is used. The parameters and universe of discourse are set as follows:

[0116] Table 2 Input Variable Settings for Vertical Descent Phase

[0117]

[0118] The number of nodes in the rule layer is the product of the number of fuzzy sets of each input variable (e.g., ...). The input variables are fuzzified using a three-dimensional mesh partitioning method; the normalization layer and output layer are constructed according to the standard structure of fuzzy neural networks.

[0119] Step 3: Staged Analysis and Strategies Based on Neural Networks

[0120] Fuzzy neural networks map fuzzy systems into a five-layer neural network structure:

[0121] Table 3 Functions of Fuzzy Neural Network Layers

[0122]

[0123] In the table, Key input parameters representing the current flight status, This represents the maximum value of the input. This represents the minimum value of the input; the other parameters have the same meaning as those in the above-described invention.

[0124] Step 3.1, Analysis of the Influence of Compensation Torque during Cruise Phase

[0125] Aircraft often face external disturbances such as airflow, which can lead to a decrease in attitude control accuracy. This method applies a fuzzy neural network to compensate for wind disturbance torque during cruise.

[0126] Analysis shows that the magnitude of the pitch moment compensation term during the cruise phase is affected by the pitch moment coefficient. Dynamic pressure during flight The magnitude has an impact. Among them, the pitch moment coefficient... Attacked and vertical wind speed disturbance Impact such as Figure 4 , Figure 5 As shown; For dynamic pressure, during the cruise phase, the speed remains constant, atmospheric density is positively correlated with altitude, and consequently, dynamic pressure is also positively correlated with altitude. Figure 6 As shown.

[0127] This allows training of a fuzzy neural network, with flight altitude as the input. Angle of attack Vertical wind speed Three-dimensional key parameters, output pitch moment coefficient With dynamic pressure Then calculate the real-time torque compensation value. .

[0128] Step 3.2, Analysis of the Influence of Compensation Torque During Vertical Descent

[0129] The aircraft is strongly affected by the ground effect, especially as altitude decreases, the ground effect intensifies significantly, and deck motion further introduces nonlinear disturbances, leading to complex changes in the pitch moment. Analysis shows that the magnitude of the pitch moment compensation term is influenced by the ground effect model, and its magnitude is affected by the altitude scaling factor. Pitch moment coefficient and dynamic pressure Influence.

[0130] This allows training a fuzzy neural network, with the input being the ground clearance. Angle of attack ,airspeed Three-dimensional key parameters, outputting the pitching moment coefficient of ground effect. With height ratio coefficient .in, The term is a proportionality coefficient, which is positively correlated with height, such as... Figure 7 ; It is a V-shaped nonlinear function, related to the angle of attack, as follows: Figure 8 ; For dynamic pressure, it is positively correlated with velocity in vertical descent simulations, such as... Figure 9 .

[0131] Step 4: Design and training of fuzzy neural network rules

[0132] Reference Figure 4-6 and Figure 7-9 Rule design was performed on the two analyzed models, cruise and vertical landing.

[0133] Step 4.1: Design of Fuzzy Logic Rules for the Cruise Phase

[0134] Since the effect of altitude on dynamic pressure changes linearly and slowly, altitude has a relatively small impact on the overall pitching moment compensation. Therefore, angle of attack and vertical wind speed disturbance are considered as the main factors affecting the amount of pitching moment compensation.

[0135] When the altitude is low and the angle of attack is small, the angle of attack has a slight effect on the pitching moment coefficient, and the vertical wind speed disturbance dominates the compensation. The design rule is as follows:

[0136] Rule 1: If the altitude is low, the angle of attack is low, and the vertical wind speed disturbance is low, the vertical wind speed has a gentle effect on the pitch moment coefficient, then the pitch moment compensation is large.

[0137] Rule 2: If the altitude is low, the angle of attack is low, and the vertical wind speed disturbance is medium, the vertical wind speed has an increased impact on the pitch moment coefficient, and therefore the pitch moment compensation is large.

[0138] Rule 3: If the altitude is low, the angle of attack is low, and the vertical wind speed disturbance is high, the vertical wind speed has a significantly increased impact on the pitch moment coefficient, then the pitch moment compensation is medium.

[0139] When the altitude is low and the angle of attack is medium, the influence of the angle of attack on the pitching moment coefficient increases. The angle of attack and vertical wind speed have a synergistic effect, and the design rule is as follows:

[0140] Rule 4: If the altitude is low, the angle of attack is medium, and the vertical wind speed disturbance is low, the vertical wind speed has a slight effect on the pitch moment coefficient, then the pitch moment compensation is large.

[0141] Rule 5: If the altitude is low, the angle of attack is medium, and the vertical wind speed disturbance is medium, the vertical wind speed has an increased impact on the pitch moment coefficient, then the pitch moment compensation is medium.

[0142] Rule 6: If the altitude is low, the angle of attack is medium, and the vertical wind speed disturbance is high, the vertical wind speed has a significantly increased impact on the pitch moment coefficient, then the pitch moment compensation is small.

[0143] When the altitude is low and the angle of attack is large, the influence of the angle of attack on the pitching moment coefficient is significantly increased, and the dominant compensation amount of the angle of attack is increased. The design rule is as follows:

[0144] Rule 7: If the altitude is low, the angle of attack is high, and the vertical wind speed disturbance is low, the vertical wind speed has a slight effect on the pitch moment coefficient, then the pitch moment compensation is medium.

[0145] Rule 8: If the altitude is low, the angle of attack is high, and the vertical wind speed disturbance is medium, the vertical wind speed has an increased impact on the pitch moment coefficient, then the pitch moment compensation is medium.

[0146] Rule 9: If the altitude is low, the angle of attack is high, and the vertical wind speed disturbance is high, the vertical wind speed has a significantly increased impact on the pitch moment coefficient, then the pitch moment compensation is small.

[0147] Because the effect of altitude on dynamic pressure changes linearly and slowly, at medium altitudes, the impact on dynamic pressure is moderate, and the impact on compensating for pitching moment is also moderate. At low angles of attack, the angle of attack has a negligible effect on the pitching moment coefficient; vertical wind speed disturbance determines the foundation compensation amount. The design rules are as follows:

[0148] Rule 10: If the altitude is medium, the angle of attack is low, and the vertical wind speed disturbance is low, the vertical wind speed has a slight effect on the pitch moment coefficient, then the pitch moment compensation is large.

[0149] Rule 11: If the altitude is medium, the angle of attack is low, and the vertical wind speed disturbance is medium, the vertical wind speed has an increased impact on the pitch moment coefficient, and therefore the pitch moment compensation is large.

[0150] Rule 12: If the altitude is medium, the angle of attack is low, and the vertical wind speed disturbance is high, the vertical wind speed has a significantly increased impact on the pitch moment coefficient, then the pitch moment compensation is medium.

[0151] When the altitude and angle of attack are both at a moderate level, the influence of the angle of attack on the pitching moment coefficient increases, and the angle of attack and vertical wind speed have a balanced effect. The design rule is as follows:

[0152] Rule 13: If the altitude is medium, the angle of attack is medium, and the vertical wind speed disturbance is low, the vertical wind speed has a slight effect on the pitch moment coefficient, then the pitch moment compensation is large.

[0153] Rule 14: If the altitude, angle of attack, and vertical wind speed disturbance are all moderate, and the vertical wind speed has an increased impact on the pitch moment coefficient, then the pitch moment compensation is moderate.

[0154] Rule 15: If the altitude is medium, the angle of attack is medium, and the vertical wind speed disturbance is high, the vertical wind speed has a significantly increased impact on the pitch moment coefficient, then the pitch moment compensation is small.

[0155] When the altitude is medium and the angle of attack is high, the influence of the angle of attack on the pitching moment coefficient is significantly increased, the dominant compensation amount of the angle of attack is increased, and the design rule is:

[0156] Rule 16: If the altitude is medium, the angle of attack is high, and the vertical wind speed disturbance is low, the vertical wind speed has a slight effect on the pitch moment coefficient, then the pitch moment compensation is medium.

[0157] Rule 17: If the altitude is medium, the angle of attack is high, and the vertical wind speed disturbance is medium, the vertical wind speed has an increased impact on the pitch moment coefficient, then the pitch moment compensation is medium.

[0158] Rule 18: If the altitude is medium, the angle of attack is high, and the vertical wind speed disturbance is high, the vertical wind speed has a significantly increased impact on the pitch moment coefficient, then the pitch moment compensation is small.

[0159] When the altitude is high, the effect of altitude on dynamic pressure changes linearly and slowly, resulting in a relatively small impact on dynamic pressure and consequently a relatively small impact on compensating for pitching moment. At this point, when the angle of attack is small, the angle of attack has a gradual effect on the pitching moment coefficient, but a significant impact on the vertical wind speed disturbance control compensation. The design rule is as follows:

[0160] Rule 19: If the altitude is high, the angle of attack is low, and the vertical wind speed disturbance is low, the vertical wind speed has a gentle effect on the pitch moment coefficient, then the pitch moment compensation is large.

[0161] Rule 20: If the altitude is high, the angle of attack is low, and the vertical wind speed disturbance is medium, the vertical wind speed has an increased impact on the pitch moment coefficient, and therefore the pitch moment compensation is large.

[0162] Rule 21: If the altitude is high, the angle of attack is low, and the vertical wind speed disturbance is high, the vertical wind speed significantly increases the impact on the pitch moment coefficient, then the pitch moment compensation is medium.

[0163] When the altitude is high and the angle of attack is medium, the influence of the angle of attack on the pitching moment coefficient increases. The angle of attack and vertical wind speed are adjusted in tandem, and the design rule is as follows:

[0164] Rule 22: If the altitude is high, the angle of attack is medium, and the vertical wind speed disturbance is low, the vertical wind speed has a slight effect on the pitch moment coefficient, then the pitch moment compensation is large.

[0165] Rule 23: If the altitude is high, the angle of attack is medium, and the vertical wind speed disturbance is medium, the vertical wind speed has an increased impact on the pitch moment coefficient, then the pitch moment compensation is medium.

[0166] Rule 24: If the altitude is high, the angle of attack is medium, and the vertical wind speed disturbance is high, the vertical wind speed has a significantly increased impact on the pitch moment coefficient, then the pitch moment compensation is small.

[0167] When the altitude is high and the angle of attack is large, the influence of the angle of attack on the pitching moment coefficient is significantly increased, and the angle of attack-dominant compensation is maximized. The design rule is as follows:

[0168] Rule 25: If the altitude is high, the angle of attack is high, and the vertical wind speed disturbance is low, the vertical wind speed has a slight effect on the pitch moment coefficient, then the pitch moment compensation is medium.

[0169] Rule 26: If the altitude is high, the angle of attack is high, and the vertical wind speed disturbance is medium, the vertical wind speed has an increased impact on the pitch moment coefficient, then the pitch moment compensation is medium.

[0170] Rule 27: If the altitude is high, the angle of attack is high, and the vertical wind speed disturbance is high, the vertical wind speed has a significantly increased impact on the pitch moment coefficient, then the pitch moment compensation is small.

[0171] Next, based on fuzzy rules, a neural network was trained, with a total of 2000 training sets inputted and 500 training rounds conducted. Validation results include regression results such as... Figure 10 The results showed that the fuzzy neural network used for validation performed well, with the overall trend closely matching the ideal fitting line and no outliers. Subsequent tests will compare the actual values ​​with the predicted values ​​from the validation set. Figure 11 The test set prediction results were good, with a maximum relative error of 1.56%, which is relatively small.

[0172] Step 4.2: Design of Fuzzy Logic Rules for the Vertical Descent Phase

[0173] When the altitude is low, the aircraft is in the strong ground effect zone, and the altitude ratio coefficient in the ground effect model is the largest, playing a dominant role in the compensation magnitude of the pitch moment. The overall compensation amount is adjusted according to airspeed and angle of attack based on this.

[0174] When the altitude and airspeed are low, the dynamic pressure is relatively small, resulting in a lower baseline value for the aerodynamic torque. The compensation amount is mainly affected by changes in the angle of attack. The design rule is as follows:

[0175] Rule 1: If the altitude is low, the airspeed is slow, and the angle of attack is small, the dynamic pressure and aerodynamic efficiency are both low, so the pitching moment compensation is medium.

[0176] Rule 2: If the altitude is low, the airspeed is slow, and the angle of attack is medium, the effect of increasing the angle of attack begins to appear, then the pitch moment compensation is medium.

[0177] Rule 3: If the altitude is low, the airspeed is slow, and the angle of attack is large, the large angle of attack may cause changes in the torque characteristics under this state, so the pitching moment compensation is small.

[0178] When the altitude is low and the airspeed is medium, the dynamic pressure reaches the critical range in this stage, and the aerodynamic torque increases significantly. The compensation amount is determined by the angle of attack and airspeed together. The design rule is as follows:

[0179] Rule 4: If the altitude is low, the airspeed is medium, and the angle of attack is small, the combined effect of large dynamic pressure and ground effect will result in a large pitching moment compensation.

[0180] Rule 5: If the altitude is low, the airspeed is medium, and the angle of attack is medium, the two work together to achieve equilibrium, then the pitch moment compensation is large.

[0181] Rule 6: If the altitude is low, the airspeed is medium, and the angle of attack is large, the increase in the angle of attack causes the aerodynamic characteristics to enter the nonlinear region, then the pitching moment compensation is medium.

[0182] At low altitude and high airspeed, dynamic pressure increases further during this phase, but handling efficiency may approach saturation. The compensation magnitude is primarily adjusted by the angle of attack. The design rule is as follows:

[0183] Rule 7: If the altitude is low, the airspeed is high, and the angle of attack is small, high dynamic pressure and strong ground effect are the main factors, then the pitch moment compensation is large.

[0184] Rule 8: If the altitude is low, the airspeed is high, and the angle of attack is medium, the system is in a high-energy state, then the pitch moment compensation is large.

[0185] Rule 9: If the altitude is low, the airspeed is high, and the angle of attack is large, it is necessary to suppress the possible pitch divergence trend, so the pitch moment compensation should be small.

[0186] When the altitude is medium, the aircraft is in the ground effect transition zone. The altitude proportionality coefficient decreases, the ground effect effect weakens, and the weight of airspeed and angle of attack on the compensation amount increases relatively.

[0187] When the altitude is medium and the airspeed is low, the dynamic pressure level is low, the ground effect support is weakened, and the compensation amount remains at a low level overall. The design rule is as follows:

[0188] Rule 10: If the altitude is medium, the airspeed is slow, and the angle of attack is small, the aerodynamic environment is relatively calm, then the pitch moment compensation is small.

[0189] Rule 11: If the altitude is medium, the airspeed is slow, and the angle of attack is medium, the disturbance effect is limited, and the pitch moment compensation is small.

[0190] Rule 12: If the altitude is medium, the airspeed is slow, and the angle of attack is large, and possible attitude changes need to be addressed, then the pitch moment compensation should be small.

[0191] When the altitude and airspeed are at a medium level, the dynamic pressure enters the effective range, which is the key interval for accurate attitude compensation. The design rules are as follows:

[0192] Rule 13: If the altitude, airspeed, and angle of attack are medium, moderate compensation is required to stabilize the attitude, and the pitch moment compensation is moderate.

[0193] Rule 14: If the altitude, airspeed, and angle of attack are all at a medium level, and the system is in a base compensation state, then the pitch moment compensation is medium.

[0194] Rule 15: If the altitude, airspeed, and angle of attack are medium, the compensation amount needs to be reduced to balance the nonlinear effect, then the pitch moment compensation amount is small.

[0195] When the altitude is medium and the airspeed is relatively high, the dynamic pressure is large, and the compensation system needs to cope with stronger aerodynamic disturbances. The design rules are as follows:

[0196] Rule 16: If the altitude is medium, the airspeed is high, and the angle of attack is small, dynamic pressure is the dominant factor, then the pitch moment compensation is medium.

[0197] Rule 17: If the altitude is medium, the airspeed is high, and the angle of attack is medium, the compensation amount is determined by a combination of factors, and the pitch moment compensation amount is medium.

[0198] Rule 18: If the altitude is medium, the airspeed is high, and the angle of attack is large, the compensation strategy tends to be conservative, and the pitch moment compensation is small.

[0199] At higher altitudes, the aircraft is largely free from ground effect, the altitude ratio coefficient is extremely small and changes gradually, and the pitch moment compensation is entirely determined by airspeed and angle of attack, used to compensate for atmospheric disturbances and trim changes during the approach process.

[0200] When the altitude is high and the airspeed is low, the low dynamic pressure and no ground effect environment, the compensation amount is based on the precise maintenance of the approach attitude, and the design rules are as follows:

[0201] Rule 19: If the altitude is high, the airspeed is slow, and the angle of attack is small, then the pitch moment compensation should be medium in order to maintain a stable descent trajectory.

[0202] Rule 20: If the altitude is high, the airspeed is slow, and the angle of attack is medium, then the pitch moment compensation is small under the baseline compensation condition.

[0203] Rule 21: If the altitude is high, the airspeed is slow, and the angle of attack is large, the pitch moment compensation is medium when compensating for the approach attitude.

[0204] When the altitude is high and the airspeed is moderate, the dynamic pressure is also moderate. The compensation amount is used to correct the attitude and coordinate with the subsequent landing logic. The design rules are as follows:

[0205] Rule 22: If the altitude is high, the airspeed is medium, and the angle of attack is small, to compensate for the subsequent leveling maneuver, then the pitch moment compensation is large.

[0206] Rule 23: If the altitude is high, the airspeed is medium, and the angle of attack is medium, then the pitch moment compensation is medium.

[0207] Rule 24: If the altitude is high, the airspeed is medium, and the angle of attack is large, then the pitching moment compensation is large.

[0208] When the altitude is high and the airspeed is high, the compensation system focuses on suppressing pitch oscillations and velocity disturbances. The design rules are as follows:

[0209] Rule 25: If the altitude is high, the airspeed is high, and the angle of attack is small, the tendency of the nose to drop is suppressed, then the pitching moment compensation is large.

[0210] Rule 26: If the altitude is high, the airspeed is high, and the angle of attack is medium, and the current stable state is maintained, then the pitch moment compensation is medium.

[0211] Rule 27: If the altitude is high, the airspeed is high, and the angle of attack is large, then the pitch moment compensation should be large to provide trim compensation in order to prevent loss of control.

[0212] Next, based on fuzzy rules, the neural network was trained with a total of 114,912 sets of training data and 22,983 sets of validation data, and 1,000 training rounds.

[0213] The validation results include fuzzy neural network regression results, such as... Figure 12 The fuzzy neural network, validated using a dataset of 22,983 samples, showed good fitting performance, with an overall trend close to the ideal fitting line and no outliers. Next, based on the fuzzy neural network's test prediction results, for example... Figure 13 The test set prediction results were good, with a maximum relative error of 4.6%, which is relatively small.

[0214] Step 5: System Integration and Application Simulation Verification

[0215] Step 5.1: Verification of the compensation torque effect during the cruise phase

[0216] During the cruise phase, the fuzzy neural network was used as an online torque compensator for the STOVL aircraft's cruise process. It received altitude, angle of attack, and vertical wind disturbance velocity signals online and output torque compensation values. Simulation results are as follows: Figure 14 .

[0217] During the cruise phase, the aircraft primarily faces atmospheric turbulence and gusts. It can be seen that before the introduction of the fuzzy neural network, aerodynamic disturbances changed significantly with variations in altitude and angle of attack. Simultaneously, unstable wind disturbances caused severe fluctuations in aircraft attitude, further exacerbating the impact of nonlinear factors on aerodynamic loads. Pitch angle control performance deteriorated significantly, attitude maintenance became difficult, and the system response tended to diverge.

[0218] By introducing a fuzzy neural network, the external disturbances caused by wind are quickly and accurately observed and compensated for in real time in the torque command. The compensation amount is dynamically adjusted, enabling the aircraft pitch angle control to converge rapidly to a stable state. Simulation results show that the maximum relative error of torque compensation during the cruise phase is 4.8%, demonstrating significant control effectiveness and effectively improving system robustness.

[0219] Step 5.2: Verification of the compensation torque effect during the vertical descent phase

[0220] During the vertical landing phase, this fuzzy neural network is used as an online torque compensator for the STOVL aircraft's vertical landing process. It receives altitude, angle of attack, and velocity signals online and outputs torque compensation values. The simulation results are as follows: Figure 15 .

[0221] It can be seen that before the introduction of the fuzzy neural network, the ground effect increases with decreasing altitude. Simultaneously, deck motion causes changes in the relative distance between the deck and the aircraft, further amplifying the nonlinear influence of the ground effect. The control performance of the pitch angle deteriorates significantly, eventually tending towards divergence. After introducing the fuzzy neural network, the external disturbances caused by these effects are quickly and effectively observed and compensated for in the torque command, resulting in significantly convergent pitch angle control. The maximum relative error of the torque compensation is 4.3%, demonstrating good performance.

Claims

1. A short take-off and landing aircraft pitch moment compensation method based on fuzzy neural network, characterized in that, Comprising the following steps: Step 1, reference pitch moment command generation The attitude control law is designed by using NDI method. Firstly, the inner loop equation of pitch rate is given as: （1）； wherein is the pitch rate, is the pitch rate change speed; is the pitch moment, , is the pitch moment generated by the aerodynamic surfaces, is the pitch moment generated by the power system; is the pitch moment of inertia; According to the state space form and NDI theory, equation (1) is rewritten as: （2）； （3）； where is a state variable in the pitch rate inner loop equation, is the rate of change of this state variable, is the control input in this loop, and is a function in the pitch rate inner loop equation that describes the nonlinear dynamics of the system, is the dynamics without control, is the state gain characteristic after the control action is implemented; in designing the inner loop control law, is the desired pitch acceleration, which is calculated from the slow loop to give the angular acceleration command, i.e. （4）； wherein is the current pitch rate state, is the corresponding pitch rate change speed, is the bandwidth of the pitch rate channel, is the pitch rate command; According to the dynamic inversion method, in order to get the expected form of equation (4), the fast loop dynamic inversion control law expression is: （5）； The slow loop is the attitude angle loop, which is in the outer layer of the inner loop. Its input is the output of the outer command generator, and its output is the input signal of the inner loop controller. Equation (6) gets the equation group of the slow loop attitude angle loop: （6）； wherein is the pitch angle, is the pitch angular velocity, equation (6) is rewritten as: （7）； （8）； wherein is a state variable in the attitude angle loop equation set, is the change speed of the state variable, is a control input in the loop, and is a function in the attitude angle loop equation set describing the nonlinear dynamic characteristics of the system, is the dynamic characteristic without control, is the state gain characteristic after the control action is implemented, and when designing the outer loop control law, the expected value of is calculated from the external command , and its expression is: （9）； wherein is the current pitch angle state, is the pitch angle state change speed, is the external pitch angle command, is the external pitch angle command change speed, is the pitch angle channel bandwidth; in order to get the desired form of equation (9), the slow loop dynamic inversion control law expression is: （10）； Step 2, fuzzy logic system construction The fuzzy logic system is a five-layer feedforward network, including input layer, fuzzification layer, rule layer, normalization layer and output layer. Step 3, phased analysis and strategy based on neural network Based on the framework of the fuzzy logic system in step 2, the phased analysis and strategy design are carried out according to the interference characteristics of different flight phases of STOVL aircraft. Step 4, fuzzy neural network rule design and training Based on the phased analysis results in step 3, fuzzy rule bases are designed for two phases respectively, and the fuzzy neural network is supervised trained by using the collected state-moment compensation sample data. Through iterative optimization of membership function parameters and rule consequent parameters, the network can accurately approximate the nonlinear mapping relationship of the disturbance moment in each phase. After training, two compensation models for different phases are obtained. Step 5, system integration and application The trained fuzzy neural network model is embedded into the flight control calculation. According to the current flight phase, the corresponding compensation model is called. The compensation model calculates the pitch moment compensation value online in real time, and the value is superimposed on the reference moment command generated by the dynamic inversion control law to form the final control command, so as to realize the effective offset of the inverse error.

2. The short take-off and landing aircraft pitch moment compensation method based on fuzzy neural network according to claim 1, characterized in that, Step 2 is as follows: Step 2.1: Input layer: receives input variables including the height above ground , the angle of attack , the airspeed , the vertical wind speed and the like, performs data normalization; Step 2.2: Fuzzification layer: each input variable corresponds to multiple Gaussian membership functions, which calculate the degree of input variables belonging to each fuzzy set, and the formula is: （11）； wherein, is the membership function used to perform fuzzification, is the membership function used to perform fuzzification, is the membership function used to perform fuzzification, is the membership function used to perform fuzzification, is the center point of the fuzzy set, is the standard deviation of the fuzzy set width, denotes the number of input variables, denotes the number of membership functions per input. Step 2.3: Rule layer: fuzzy rules are set based on the three-dimensional grid partitioning method, the number of rules is the product of the number of membership functions of each input variable, and the algebraic product operator is used to calculate the activation degree of each rule , the formula is: （12）； wherein is the total number of input variables, denotes the membership value on the corresponding fuzzy set of the input variable in the rule Step 2.4: Normalization layer: calculate the normalization weight of each rule activation degree; Step 2.5: Output layer: output the pitch compensation moment by weighted linear combination, and the formula is: （13）； In formula (13), is a pitch compensation moment; is a normalized weight of the rule activation; is a rule consequent parameter.

3. The short take-off and landing aircraft pitch moment compensation method based on fuzzy neural network according to claim 1, characterized in that, Step 3 is as follows: Step 3.1: Cruise phase: Determine input variables as flight altitude , angle of attack , and vertical wind speed , and output as the pitch compensation moment under cruise disturbance, and then analyze the influence of different input states on the output pitch compensation moment; Step 3.2: Vertical Descent Phase: Determine input variables as height above ground , angle of attack , and airspeed , and output as the pitch compensation moment under ground effect disturbance, and then analyze the influence of different input states on the output of the pitch compensation moment.

Images