Sensor-fault-resistant multi-mode propulsion control method for low-altitude aircraft

By building a nonlinear model and introducing weight embedding fusion network and residual perception mechanism, the robustness of propulsion control of low-altitude vehicles in complex disturbance environments is solved, intelligent weighted fusion and fault-tolerant parameter adjustment of multimodal propulsion control are realized, and the generalization ability and response flexibility of the aircraft are improved.

CN120161778AActive Publication Date: 2025-06-17XIAMEN UNIV

Patent Information

Application Number
CN202510641776.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-05-19
Publication Date
2025-06-17
Estimated Expiration
2045-05-19

AI Technical Summary

Technical Problem

The prior art is difficult to realize multimodal propulsion control of low-altitude aircraft in complex disturbance environments, especially in sensor failure or abnormal situations, and control strategies are difficult to maintain stability and robustness.

Method used

A multimodal propulsion control method for low-altitude aircraft that resists sensor failure is proposed. By constructing a nonlinear model, collecting flight state and sensor data in real time, introducing weighted embedding fusion network and residual perception mechanism, a weighted control strategy is generated, and intelligent weighted fusion and fault-tolerant parameter adjustment of propulsion control are realized.

Benefits of technology

It improves the generalization ability and response flexibility of the aircraft in the multi-mode and multi-task stages, enhances the adaptability and robustness of control strategies for changes in complex flight environments, and achieves the optimization of energy efficiency of propulsion control.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120161778A_ABST
    Figure CN120161778A_ABST
Patent Text Reader

Abstract

The invention relates to the field of aircraft intelligent control, and discloses a sensor fault resistant low-altitude aircraft multi-mode propulsion control method, which comprises the following steps: constructing a nonlinear model of a short-vertical aircraft hybrid propulsion system, taking deep exploration optimization reinforcement learning as a main control, and combining a fuzzy controller and a model prediction controller to form a strategy fusion group; the method comprises the following steps: introducing a residual perception mechanism and a GIRLS-EKF health estimation module, extracting residual signal features by using CNN and Transform, and carrying out real-time diagnosis and dynamic weight adjustment on an abnormal measurement state of a sensor in combination with generalized residual least square filtering and an extended Kalman structure; the controller fusion module outputs the optimal control quantity in a self-adaptive manner according to the health state and the flight mode, and the fault-tolerant performance and the energy efficiency scheduling capability of the multi-propulsion system under the complex working condition are improved; the method is suitable for low-altitude economy and urban air traffic task scenes, and has good engineering integration and intelligent control application prospects.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of intelligent control of aircraft, and particularly to a multi-modal propulsion control method for low-altitude aircraft resistant to sensor failures. Background Art

[0002] Currently, the low-altitude economy, as a strategically emerging industry strongly supported by the state, is accelerating the birth of new aviation platforms, especially short takeoff and vertical takeoff and landing (STOVL) aircraft, which are widely used in scenarios such as urban air transportation, logistics, and emergency response. Compared with traditional aircraft, STOVL aircraft have the characteristics of multi-mode switching and hybrid operation of multiple types of thrusters, posing higher requirements for intelligent, adaptive, and robust control of the propulsion system.

[0003] During the mission execution of a typical STOVL aircraft, it needs to frequently experience multi-state switching such as vertical takeoff, climb, cruise, hover, transition, descent, and vertical landing. The propulsion system has a complex structure, including electric propulsion, turbofan propulsion, tilting mechanisms, and control surface systems. The control variables are highly coupled, and the dynamic response characteristics vary significantly.

[0004] Traditional propulsion control strategies are mostly single-model controllers, which are difficult to meet the multi-modal flight requirements, have slow responses and low tuning efficiency in complex disturbance environments, and are prone to degradation of control performance. Especially in a multi-sensor system, if there are measurement anomalies or failures, it is easy to cause distorted state estimation, which in turn affects the decision-making of control strategies and poses a great risk of robustness.

[0005] In response to the above challenges, artificial intelligence control algorithms have gradually been introduced into the flight control field. In recent years, the multi-strategy fusion of deep reinforcement learning combined with model predictive control and fuzzy control has become a development trend, but in practical applications, it still faces problems such as unstable strategies, uneven switching, and lack of a robust processing mechanism for sensor anomalies.

[0006] Therefore, the present invention proposes a multi-modal propulsion control method for low-altitude aircraft resistant to sensor failures, a propulsion control method that fuses the output of control strategies, introduces a sensor residual sensing mechanism, and has the ability to tolerate faults and adjust parameters. Summary of the Invention

[0007] The purpose of the present invention is to provide a multi-modal propulsion control method for low-altitude aircraft resistant to sensor failures. To solve the above-mentioned problems in the prior art, the present invention is realized through the following technical solutions: A multi-modal propulsion control method for low-altitude aircraft resistant to sensor failures provided by an embodiment of the present invention specifically includes the following steps: S1: Construct a non-linear model of the propulsion component; S2: Collect the flight state and sensor observation parameters in real time, and input them into the flight state recognition module and the health parameter estimation module to identify the flight mode and the health state of each propulsion component respectively; S3: Introduce a weighted embedding fusion network, input the state residual and the health state, generate the weighted coefficients of the control strategy group, extract the local disturbance features, model the multi-channel state time series dependence, and output the residual weighted vector; S4: Jointly input the recognized state and the health state into the propulsion control optimization of the control strategy group, and calculate the control action output respectively; S5: Weight the control action output according to the weighted coefficients of the control strategy group to generate the final propulsion command, distribute the final propulsion command to each propulsion subsystem of the aircraft, and feedback the signals and sensor data of each propulsion subsystem to the experience pool and the health parameter estimation module, and perform policy update and fault perception accuracy iteration.

[0008] Furthermore, the specific acquisition method of the non-linear model is as follows: Define the control variable vector to describe the control process of the aircraft propulsion system; Define the control variable vector: ; Among them, and represent the output power percentages of two independent electric propulsion modules; represents the deflection angle of the turbofan nozzle; represents the tilting mechanism angle; represents the opening of the tail nozzle of the turbofan or fan thruster; The variables constitute the main control input of the system and participate in the real-time optimization process; Define the system state vector to describe the evolution process of the propulsion system under different control inputs; Define the system state vector: ; Among them, represents the combined thrust of the propulsion system at the current moment; represents the unit thrust energy consumption; represents the aircraft altitude; represents the forward speed of the aircraft; Based on the non-linear coupling between different thrusters, the non-linear coupling includes but is not limited to: the fast dynamic response of the electric thruster, the strong hysteresis of the turbofan system, and the complex coupling of the air flow interference. The system state transition process is non-linearly mapped through the function for non-linear mapping; Further, the method for obtaining the health status of the propulsion component is as follows: Based on the Actor network, update its own parameters according to the reward and punishment value of the action by the Critic network. At this time, the new loss function is:

[0009] The policy gradient update formula adopted by the Actor network is as follows:

[0010] The Critic network updates its parameters through the DQN algorithm, and the gradient update formula is as follows:

[0011] Among them, and are the learning rates of the Actor network and the Critic network respectively; and are the parameters of the Actor network and the Critic network respectively; represents the reward value obtained by the agent executing the action under the environmental state ; represents the discount factor, which is used to measure the importance of future rewards, represents the action value obtained by inputting the action generated by the Actor network at the state into the Critic network, representing the temporal difference error, represents the policy with respect to the parameter at the state ; represents the policy adopted by the Actor network; Identify the flight mode and the health status of each propulsion component based on the reward value obtained from the Actor-Critic model; Further, the method for obtaining the weighting coefficient is as follows: When reaches the maximum number of episodes, initialize a random process for action exploration , obtain the initial state , and calculate the initial residual ; Input it into the residual perception network to obtain the residual weighting coefficient ; Further, the method for obtaining the residual weighting vector is as follows: Randomly initialize the weight parameters of the current Actor network and the weight parameters of the current Critic network ; Initialize the target Actor network and target critic network , their respective network weight parameters are: ; Initialize experience replay pool Store state transfer and control residual information; Initialize the residual perception network and build a CNN + Transformer combined network; The network input is the residual sequence in the current state , which is the difference between the estimated state and the actual state; The CNN module extracts the local perturbation peak, and the Transformer module models the evolution of the residual over time and channels; The network output is the residual weighted vector ; Furthermore, the method for optimizing propulsion control is: By formula

[0012] Rewritten as:

[0013] Where:

[0014] in, is the time derivative of the state variable, is the state matrix, which describes the influence of the state variable itself on its derivative. is the change of the state variable, is the input matrix, describing the effect of the input on the state derivative, is the change in the input variable, is a matrix, reflecting the influence on the state derivative, is the change in the additional variable, is the process noise, is the change in the output variable, is the output matrix, describing the influence of the state on the output, is the direct transfer matrix, which indicates the direct effect of input on output. is a matrix that describes the impact on the output. To measure noise, is the new state variable derivative, is the new state vector; Based on the design of EKF, the formula is:

[0015] Convert to discrete form:

[0016] In the formula, the subscript k represents the value of the variable at the k-th step; Definition:

[0017] Initialization:

[0018] State prediction:

[0019] The forgetting factor in the filter is:

[0020] where is the forgetting factor reference value.

[0021] and The matrix is:

[0022] where

[0023] Measurement prediction:

[0024] The non-linear term The error and noise introduced during linearization Compared with The noise near is negligible; for Perform a first-order Taylor expansion to obtain

[0025] Furthermore; obtain the batch processing regression formula of the EKF:

[0026] where ; The estimation problem in the above formula is rewritten as

[0027] The observation vector is , the regression matrix is , the noise is . The error The covariance matrix of is

[0028] The EKF update state calculated in the correction step , obtained by performing weighted least squares, linear regression:

[0029] Can be transformed into:

[0030] The correction step size of the EKF can be obtained by the following formula:

[0031] where, is the state vector at the k+1 step, is the discrete state transition function, is the process noise at the k step, is the measurement output at the k step, is the discrete measurement function, is the measurement noise at the k step, is the Jacobian matrix of the state transition function with respect to state x, is the Jacobian matrix of the measurement function with respect to state x, is the initial state estimate, is the covariance matrix of the initial state estimate, is the prior state estimate at the k step, is the state transition function at the k step, is the posterior state estimate at the k-1 step, is the prior covariance matrix at the k step, is the forgetting factor at the k step, is the state transition function with respect to state x at the Jacobian matrix at this point, is the posterior covariance matrix at the k-1 step, is the process noise covariance matrix at the k-1 step, is the forgetting factor reference value, is the intermediate matrix for calculating the forgetting factor, is the intermediate matrix, is the matrix the trace of, is the Jacobian matrix of the measurement function h with respect to state x at this point, is the initial covariance, is the measurement noise covariance matrix at the k step, is the Kalman gain matrix at the k step, is the measurement value at the k step, is the posterior state estimate at the k step, is the posterior covariance matrix at the k step, h is the non-linear observation function, is the true state at the k-th step, is the observation noise at the k-th step, is the prior state estimate at the k-th step, is the Jacobian matrix of the observation function h at , and I is the n×n identity matrix, is the prior state estimation error, is the combined observation vector, is the combined regression matrix, is the combined noise vector, is the covariance matrix of the observation noise, is the covariance matrix of the prior state estimate, is the correction term for covariance update; Furthermore, the method for obtaining the control action output is: Change the decision-making process of RE-DEORL from deterministic to stochastic, and add noise to the output action of the policy to achieve exploratory expansion, and finally the action executed by the environment has the expression:

[0032] where, is set as Gaussian white noise, and the mean μ is the output value of the policy network; Based on the continuous progress of the training process, continuously reduce the noise variance to achieve the balance between exploration and exploitation; Furthermore, the method for obtaining the final propulsion instruction is: The thrust error does not exceed the tolerance threshold; the thruster power shall not exceed the limit value; the attitude angle rate of change is restricted to prevent sharp maneuvers; the nozzle area, rotation speed and voltage are restricted; The mathematical form of the objective function is:

[0033] where, J is the objective function, λ is the control penalty factor, is the energy consumed when generating thrust , is the actual thrust generated by the thruster, is the desired target thrust value; Furthermore, the method for performing policy update is: Use the least squares method to calculate the measurement parameters:

[0034] Calculate the residual based on the measurement parameters:

[0035] When all sensors are fault-free, i.e., there are no abnormal measurement points, the residual value is very small, a positive integer close to 0; when some sensors have abnormal measurement points, the residual value is large; Since is an -dimensional ill-conditioned matrix; eliminating its similar rows, i.e., if , then let the elements of the i-th row of be 0, where represents the elements of the i-th row of matrix , and the ill-conditioned matrix tolerance error is a constant greater than 0; Using the above method, the problem of solving equation

[0036] is transformed into the problem of solving a non-homogeneous linear equation system; using elementary row transformations of determinants; the solution is obtained as

[0037] where, is the general solution, the generalized residual is the particular solution, , the general solution term represents a series of combinations of measurement parameters that satisfy a residual of 0, the particular solution term represents a series of combinations of measurement parameters that result in a non-zero residual, and the larger the value of the elements of the particular solution term, the more abnormal the measurement parameters of the corresponding sensor; The new weight function is defined as

[0038] The measurement prediction equation of GIRLS-EKF is

[0039] where, is the weight matrix, , is the measurement parameter estimated by the least squares method, is 's generalized inverse matrix, r is the residual between the observed value and the estimated value, is the i-th row element of the matrix, is the tolerance error constant, is the newly defined weight function, is the residual term, is the threshold constant of the weight function, is the prior state covariance matrix, is the observation noise covariance matrix, is the input variable at the k-th step, which is from Weight matrix composed of weight functions; Furthermore, the method for iterating the fault perception accuracy is as follows: Based on ensuring the consistency of the estimator under clean data, should be fixed as , where is the standard Gaussian distribution, and the scale estimator is solved iteratively by:

[0040] Among them, is the relative standard residual, and the weight function:

[0041] The selected function is:

[0042] The weight function is defined as:

[0043] Among them, represents the expectation of the function, is the standard Gaussian distribution, is the function, is the variance of the scale estimator, m is the number of samples, is the weight function, is the relative standard residual, is the threshold constant, is the state estimation value, is the scale estimation value.

[0044] Advantages of the present invention: 1. The present invention constructs a set of integrated control strategy mechanisms, integrating the reinforcement learning RE-DEORL, model predictive control MPC and fuzzy control methods, combining flight states and system health parameters, realizing the intelligent weighted fusion of control strategy outputs, breaking the limitation that a single strategy in traditional propulsion control is difficult to adapt to multiple flight states, effectively improving the generalization ability and response flexibility of the system under multiple modes and multiple task phases; the RE-DEORL reinforcement learning algorithm adopted combines the Actor-Critic structure and DQN idea, integrates the experience replay mechanism and target network design, improves the stability of policy training and sample utilization efficiency, avoids the policy falling into local optimum, and at the same time improves the adaptability and convergence speed of the control strategy to complex flight environment changes through the dynamic perturbation expansion mechanism; 2. A residual-aware weight network based on CNN and Transformer is introduced. Combining with the GIRLS-EKF filtering structure for multi-channel sensor signal processing and health parameter estimation, it can automatically adjust the control structure in the presence of sensor anomalies, non-Gaussian noise, or thruster performance degradation, realizing the dynamic fault-tolerant adjustment of the system and ensuring the robust performance; adaptively adjusting the output weight of the controller based on flight performance feedback and estimated health status, realizing the "self-evolution" ability of the control strategy; at the same time, the system uses a structured neural network to fuse and decouple control variables, combined with the heterogeneous response characteristics of thrusters, to achieve collaborative drive and optimal power distribution among multiple propulsion components, with good engineering integrability and platform adaptability, suitable for various propulsion configurations such as electric propulsion, turbofan hybrid, and tilt-rotor; realizing the propulsion control robustness and energy efficiency optimization of low-altitude aircraft under complex flight states and sensor anomaly conditions, providing a systematic solution for the intelligent control architecture, and having significant application value and technical promotion prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0046] Figure 1 It is a schematic diagram of the multi-controller fusion mechanism structure for anti-sensor-fault multi-modal propulsion control of a low-altitude aircraft provided in Embodiment 1 of the present invention; Figure 2 It is a flowchart of the Actor-Critic network architecture and training for anti-sensor-fault multi-modal propulsion control of a low-altitude aircraft provided in Embodiment 1 of the present invention; Figure 3 It is a flowchart of the propulsion control optimization based on the RE-DEORL algorithm for anti-sensor-fault multi-modal propulsion control of a low-altitude aircraft provided in Embodiment 1 of the present invention; Figure 4 It is a schematic diagram of the structure of the engine fault-tolerant airborne adaptive model modeling method for anti-sensor-fault multi-modal propulsion control of a low-altitude aircraft provided in Embodiment 1 of the present invention; Figure 5 It is a flowchart of the GIRLS-EKF structure for anti-sensor-fault multi-modal propulsion control of a low-altitude aircraft provided in Embodiment 1 of the present invention; Figure 6 It is a schematic diagram of the GIRLS principle for anti-sensor-fault multi-modal propulsion control of a low-altitude aircraft provided in Embodiment 1 of the present invention; Figure 7 This is the flowchart of the steps of a multi-modal propulsion control method for low-altitude aircraft against sensor failures provided by Embodiment 1 of the present invention. Detailed implementation manners

[0047] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of the present invention.

[0048] Embodiment 1 As Figure 1 and Figure 7 shown, a multi-modal propulsion control method for low-altitude aircraft against sensor failures provided by an embodiment of the present invention specifically includes the following steps: S1: Construct a non-linear model of the propulsion assembly of a short takeoff / vertical landing aircraft; For multiple types of propulsion units such as electric propulsion, turbofan propulsion, and tiltrotor in short takeoff and vertical landing aircraft, construct a unified non-linear modeling and intelligent control architecture, propose a fusion propulsion control strategy based on the RE-DEORL reinforcement learning algorithm master control, and jointly use a health perception mechanism, a residual weighted network, and a controller weight adjustment mechanism to dynamically coordinate the output of the propulsion unit; The control variables include but are not limited to: electric propulsion power distribution ratio, nozzle area, tilt angle, nozzle deflection angle, and mode switching variable; In a specific embodiment, the method for constructing the non-linear model is: Define a control variable vector to describe the control process of the aircraft propulsion system; Specifically, define the control variable vector: ; Among them, and represent the output power percentages of two independent electric propulsion modules, which mainly affect vertical takeoff and landing and low-speed attitude control; represents the turbofan nozzle deflection angle, which controls the fine adjustment of the propulsion direction, especially for attitude correction during takeoff, hover, and descent; represents the tilt mechanism angle, which determines the working direction of the electric thruster or fan and is a key variable for realizing the vertical-horizontal flight transition; Represents the nozzle opening of the turbofan or fan thruster, which affects the outlet air flow velocity and momentum and is an important control parameter for adjusting the balance between horizontal thrust and energy consumption; The variables constitute the main control input of the system and participate in the real-time optimization process; Defines the system state vector to depict the evolution process of the propulsion system under different control inputs; Specifically, define the system state vector: ; Among them, Represents the combined thrust of the propulsion system at the current moment; Represents the energy consumption per unit thrust (such as watts per newton) and represents the energy efficiency performance; Represents the altitude of the aircraft and is used to reflect the vertical flight state; Represents the forward speed of the aircraft and measures the horizontal propulsion output; Due to the existence of non-linear coupling between different thrusters, the non-linear coupling includes but is not limited to: the fast dynamic response of the electric thruster, the strong hysteresis of the turbofan system, and the complex coupling of air flow interference. The system state transition process is non-linearly mapped through the function for non-linear mapping; It should be noted that the function is jointly constructed by the aircraft aerodynamic model, the propulsion component response model, and the power conversion model, and supports iterative optimization during offline simulation and online reinforcement learning training; On the premise of ensuring the stability of the aircraft mission propulsion, minimize the energy consumption per unit thrust, that is, construct the following non-linear constrained optimization model: Performance index: Minimize the energy consumption rate ; Constraints: Flight safety boundaries, including maximum power, upper limit of motor speed, and attitude angle limit, dynamic power balance between multiple thrusters, and thrust output meeting the requirements of different flight states, such as hover thrust and horizontal cruise thrust; The system continuously iteratively updates the policy network through reinforcement learning and jointly outputs with other controllers to achieve fusion regulation; S2: Real-time collect the flight state and sensor observation parameters, and input them into the flight state recognition module and the health parameter estimation module to identify the flight mode and the health state of each propulsion component respectively; In a specific embodiment, real-time collect state variables such as flight altitude, speed, mission phase, inclination angle, thrust demand, etc.; adopt a state classification model based on decision trees to judge the current flight mode, such as vertical takeoff, hover, horizontal cruise, transition, etc.; the output of this module will be used as a reference variable for the policy fusion mechanism; The control strategy group includes the following three types of control algorithm modules: The RE-DEORL main control module is responsible for generating the main strategy and learning control variables; the MPC control module uses the system model to predict control variables and correct trajectories within a short time domain; the fuzzy control module performs fuzzy inference based on empirical rules and state variables to adapt to sudden disturbances and uncertain environments; The fused controller output module takes the outputs of the three controllers as inputs; constructs a weighted fusion mechanism to output the optimal control action; the weights change dynamically and are updated in real time by the self-learning module; The self-learning parameter tuning module uses a lightweight neural network to construct a state-weight mapping function ; dynamically adjusts the output weights of each controller according to the flight mission phase, system performance feedback, and custom reward and punishment mechanisms; realizes the optimal matching and resource utilization of the controller outputs in different scenarios; The execution instruction generation and distribution module receives the fused control quantities, including but not limited to: propulsion power, motor control signals, and nozzle adjustment amounts; converts them into low-level control instructions and distributes them to each propulsion subsystem through the control bus; As Figure 1 shown, the multi-controller fusion mechanism structure for anti-sensor failure forms a multi-layer control system consisting of state recognition → parallel output of the control strategy group → self-learning fusion → final execution, realizing the optimal coupling of the system-level response ability and energy efficiency; It should be noted that in the multi-controller structure, RE-DEORL is the main control core, generating the main control strategy and providing a benchmark for system training and performance drive; the output of the policy network serves as the initial value of the control variable and also as the reference input for fuzzy control and MPC; the training objective of RE-DEORL optimizes the strategy itself, guiding the fusion controller module to learn the optimal combination method in different states; during the training process, the RE-DEORL controller receives the interaction data in the experience pool, updates the value network using the comparison error between the target network and the real-time network, and then feeds it back to the policy network to complete the step-by-step iterative optimization; During the execution of the flight mission, the control structure automatically judges the flight conditions according to the output of the state perception module and proceeds with the following process: Input the current state , for parallel processing by RE-DEORL, MPC, and the fuzzy controller; The three controllers calculate and output respectively , , ; Input the state into the fusion weight network of the fusion device to calculate the fusion weight ; Generate the final control quantity according to the fusion formula ; Send control instructions to the actuator to complete the propulsion control; The system collects execution feedback and stores it in the experience pool for RE-DEORL training and updating; If there is performance degradation or flight mode switching, automatically readjust the fusion weights or trigger the reorganization of the control structure; As Figure 2 shown, the Actor-Critic model obtains a better algorithm structure by combining the policy gradient method and the value function method; In the Actor-Critic model structure, the Actor network is based on the policy gradient algorithm and selects appropriate actions from continuous actions according to the current state; The Critic network is based on value function methods such as DQN, calculates the reward and punishment value brought by the state transition after the execution of this action, and evaluates whether this action is reasonable; The Actor-Critic structure; updates the network parameters step by step to avoid the problem of low model efficiency caused by the episodic update of the policy gradient algorithm; In the specific interaction process, the Actor network obtains the probability value of each action and then selects behaviors based on the probability size; The Critic network is continuously updated to improve the reward and punishment value of selecting each action in each state; Based on the Actor network, it updates its own parameters according to the reward and punishment value of the action by the Critic network. At this time, the new loss function is:

[0049] The policy gradient update formula adopted by the Actor network is as follows:

[0050] The Critic network updates its parameters through the DQN algorithm, and the gradient update formula is as follows:

[0051] Among them, and are the learning rates of the Actor network and the Critic network respectively; and are the parameters of the Actor network and the Critic network respectively; represents the agent in the environmental state executes the action to obtain the size of the reward value; represents the discount factor, which is used to measure the importance of future rewards, represents in the state The action value generated by the Actor network at a certain time and input into the Critic network represents the time difference error. Represents the policy Regarding the parameters In the state The gradient of Represents the policy adopted by the Actor network; Identify the flight mode and the health status of each propulsion component based on the reward value obtained from the Actor-Critic model; S3: Introduce a weight embedding fusion network, input the state residual and the health status, generate the weighted coefficients of the control policy group, extract the local perturbation features, model the multi-channel state time series dependence, and output the residual weighted vector; In a specific embodiment, a residual-aware weight mechanism is introduced based on the original Actor-Critic architecture, which effectively enhances the system's adaptability to non-Gaussian noise and sudden abnormal signals, and incorporates the idea of the DQN algorithm into the Actor-Critic model structure. Its structure includes two parts: a value network and a policy network; Reinforcement learning is usually based on the MDP model, and there is a correlation between data, resulting in unstable training and difficulty in network convergence. The RE-DEORL algorithm uses two methods, an experience replay pool and a target network, to improve the overall performance of the algorithm; When facing sensor signal perturbations, non-Gaussian distributed residuals, and sudden outliers, the Actor-Critic structure is difficult to robustly adjust; The control policy is highly sensitive to noise changes, which may lead to policy divergence or performance degradation. Introduce the residual-aware weight network into the following two core points: State preprocessing stage: Construct a CNN+Transformer network structure Input: State residual, multi-channel sensor residual stream ; CNN module: Extract local perturbation features; Transformer module: Model long-term state dependence and abnormal persistence patterns; Output: Residual weighted vector , as the weighted adjustment parameter of the control variable; Introduce

[0052] The original control quantity Rewrite as:

[0053] Where is the state-sensitive dynamic gain coefficient, which is used to adjust the action amplitude under abnormal disturbances; For the neural network of deep reinforcement learning, a certain amount of sample data is required when updating the weight coefficients of neurons using the gradient descent method; If the online interactive learning method is used, the current data needs to be discarded after the current network update, resulting in a significant reduction in data utilization. The agent needs to interact with the environment more to achieve the final convergence effect; The experience replay technique creates a buffer of a certain size and stores the state transition information ; The state transition sample information enters the buffer in sequence. If the buffer is full, when a new sample enters, the oldest sample is removed from the buffer; It should be noted that as a deterministic policy gradient algorithm, the interaction sequence obtained by the RE-DEORL algorithm based on the policy network is fixed after a given initial state. The agent cannot generate different behaviors to deeply explore the environment, so the policy cannot be improved; As Figure 3 shown, the propulsion control is optimized based on the RE-DEORL algorithm; Specifically, randomly initialize the weight parameters of the current Actor network and the weight parameters of the current Critic network ; Initialize the target Actor network and the target Critic network , and their respective network weight parameters are: ; Initialize the experience replay pool , which is used to store state transition and control residual information; Initialize the residual perception network, construct a CNN + Transformer combined network, which is used to extract the disturbance features and time dependence in the multi-channel state residual sequence; The network input is the residual sequence at the current state, that is, the difference between the estimated state and the true state; The CNN module extracts local disturbance peaks, and the Transformer module models the evolution relationship of the residuals over time and channels; The network output is the residual weighted vector , which is used for dynamic adjustment of control variables; When the maximum number of episodes, initialize a random process for action exploration and obtain the initial state , and calculate the initial residual ; Input it into the residual perception network to obtain the residual weighting coefficient ; Generate a weighted control action, and input the state into the current policy network to output the original action vector ; Use the residual weighting mechanism to obtain the final execution action Send to the environment for execution; The environment feedbacks the new state , reward , and new residual ; Store the state transition sample into the experience replay pool, randomly sample from the experience replay pool samples as the training data for the Actor network and the Critic network, and let , update the parameters of the Critic network by minimizing the loss function

[0054] Among them, is the target value, ; When calculating , represents the discount factor, , the target value network and the target policy network are used, so that the network can remain stable during the training process and is easier to converge; Calculate the gradient of the policy network:

[0055] Update the target networks and :

[0056]

[0057] Among them, and are the parameters of the current policy network and the current value network respectively, and are the parameters of the target policy network and the target value network respectively, is the update coefficient, representing the update step size, which balances between the current network parameters and the target network parameters; ​It should be noted that on the basis of retaining the RE-DEORL training framework, RE-DEORL integrates a state residual self-perception regulation mechanism. By introducing a CNN + Transformer combined perception network and an action weighting structure, it significantly enhances the robustness and adaptive regulation ability of the policy network in a complex perturbation environment; it is particularly suitable for the intelligent control tasks of propulsion systems under the background of multi-sensors, multi-control variables, and non-Gaussian noise. S4: Jointly input the recognition state and the health state into the propulsion control optimization of the control policy group, and calculate the control action output respectively. Such as Figure 4 As shown, construct an adaptive model. The dynamics is strongly non-linear, that is, both the system equation and the measurement equation of the model are non-linear. EKF first performs a Taylor series expansion on the original system and measurement, approximates the system as a first-order linear system, and then makes a linear Kalman filter estimation. The difference from the linear KF is that it uses a non-linear model to calculate the state value and output value to solve the parameter estimation problem of the non-linear system.

[0058] Since the performance degradation of the engine occurs slowly, it is considered to satisfy ; Through the formula

[0059] Rewrite it as:

[0060] In the formula:

[0061] Among them, is the time derivative of the state variable, is the state matrix, depicting the influence of the state variable itself on its derivative, is the change in the state variable, is the input matrix, describing the effect of the input on the state derivative, is the change in the input variable, is the matrix, reflecting the influence on the state derivative, is the change in the additional variable, is the process noise, is the change in the output variable, is the output matrix, describing the influence of the state on the output, is the direct transmission matrix, indicating the direct effect of the input on the output, is the matrix, depicting the influence on the output, is the measurement noise, is the new state variable derivative, is the new state vector; Based on the EKF design, transform the formula:

[0062] into its discrete form:

[0063] In the formula, the subscript k represents the value of the variable at the k-th step; Define:

[0064] Initialization:

[0065] State prediction:

[0066] The forgetting factor in the filter is:

[0067] where, is the forgetting factor reference value.

[0068] and The matrices are:

[0069] where,

[0070] Measurement prediction:

[0071] The non-linear term The error and noise introduced during linearization Compared with The noise near is neglected; perform a first-order Taylor expansion on to obtain

[0072] Furthermore; obtain the batch regression formula of the EKF:

[0073] where, ; Rewrite the estimation problem in the above formula as

[0074] The observation vector is and the regression matrix is and the noise is . The error The covariance matrix is

[0075] The EKF updated state calculated in the correction step , obtained by performing weighted least squares, linear regression:

[0076] Can be transformed into:

[0077] The correction step size of the EKF can be obtained by the following formula:

[0078] Where, is the state vector at the k+1 step, is the discrete state transition function, is the process noise at the k step, is the measurement output at the k step, is the discrete measurement function, is the measurement noise at the k step, is the Jacobian matrix of the state transition function with respect to the state x, is the Jacobian matrix of the measurement function with respect to the state x, is the initial state estimate, is the covariance matrix of the initial state estimate, is the prior state estimate at the k step, is the state transition function at the k step, is the posterior state estimate at the k-1 step, is the prior covariance matrix at the k step, is the forgetting factor at the k step, is the state transition function with respect to the state x at The Jacobian matrix at, is the posterior covariance matrix at the k-1 step, is the process noise covariance matrix at the k-1 step, is the forgetting factor reference value, is the intermediate matrix for calculating the forgetting factor, is the intermediate matrix, is the matrix The trace of, is the Jacobian matrix of the measurement function h with respect to the state x at The Jacobian matrix at, is the initial covariance, is the measurement noise covariance matrix at the k step, is the Kalman gain matrix at the k step, is the measurement value at the k step, is the posterior state estimate at the k-th step, is the posterior covariance matrix at the k-th step, h is the non-linear observation function, is the true state at the k-th step, is the observation noise at the k-th step, is the prior state estimate at the k-th step, is the Jacobian matrix of the observation function h at , I is the n×n identity matrix, is the prior state estimation error, is the combined observation vector, is the combined regression matrix, is the combined noise vector, is the covariance matrix of the observation noise, is the covariance matrix of the prior state estimate, is the correction term for covariance update; such as Figure 5 shown, the improved iteratively reweighted least squares-extended Kalman filter; The GIRLS module calculates weights according to the weight formula. The farther the outlier is from the population, the smaller the assigned weight. When the outlier exceeds a certain limit value c, its weight will become 0; The hybrid adaptive module updates the measurement noise covariance matrix and the forgetting factor based on historical data, thereby continuously optimizing the filter parameters; By using the residual between the measurement value and the model calculated value and the parameters calculated by the filter for iterative operations, the state quantity is estimated; It should be noted that the flight envelope of the engine is large, there are many working modes, and the working environment is harsh. It is easily interfered in a complex environment. In addition, there are a large number of sensors and they are prone to failure, which is likely to generate large measurement outliers; The classical weighted least squares method, that is, the maximum likelihood estimator under Gaussian noise, will be affected by the decrease in accuracy, manifested as an increase in the variance and bias of the estimated value, thus affecting the reliability and estimation accuracy, and having poor robustness; The Huber M estimator is a research method for the abnormal observation values of the filter. By introducing weights to reconstruct the measurement noise covariance matrix, the influence of abnormal observation values on the system is reduced, and the robustness of the filter is improved; The Huber cost function is the most commonly used cost function in M estimation, and its expression is shown in the following formula:

[0079] where γ is the adjustment factor, i is 1, 2,... m, m is the dimension of the observed quantity, is the observation residual; The corrected measurement noise covariance matrix is as follows:

[0080] Wherein, is the weight function, .

[0081] It should be noted that by substituting the above formula for the measurement noise covariance matrix in the EKF, the robust EKF algorithm based on Huber is obtained, which reweights the measurement noise covariance matrix, constructs different weights for observation residuals of different sizes, thereby reconstructing the measurement noise covariance matrix, overcoming the influence of abnormal observation values on the estimation accuracy, and achieving the purpose of improving the robustness of the estimator; According to the formula:

[0082] It should be noted that since information outliers appear in the predicted state , their influence will be propagated to the observation vector on the left side of the above formula through the matrix , thus causing a large estimation error; it has a certain isolation effect on small deviation outliers caused by disturbances such as noise, and improves the robustness and estimation accuracy of the filter to a certain extent; when the sensor fails and there is a large deviation between the measured value and the actual value, it cannot overcome the influence of abnormal measurement values; The weighted least squares estimator is the maximum likelihood estimator under Gaussian noise and is regarded as the estimator that minimizes the regression residual norm, that is satisfies:

[0083] Wherein, is the standard deviation of the residual, that is , the residual vector is , arg min represents the value of the variable when the following formula reaches the minimum; The magnitude of the standard deviation estimate is affected by abnormal measurement values, that is, outliers. In order to minimize the corresponding residual squares, the filter even tends to tilt towards the outliers; As Figure 6 shown, based on the generalized iteratively reweighted least squares, the solid circle in the lower left represents the circle to be estimated, the dots in the lower left represent normal observations, the protruding dot in the upper right represents an abnormal observation, the solid circle in the upper right represents the WLS fitting result, and the dashed circle represents the GIRLS fitting result; The GIRLS method performs iterative reweighting operations on all observations, assigns a weight of 1 to normal observations, assigns a weight less than 1 to abnormal observations, and the farther the abnormal observation is from the normal value, the smaller the weight until it reaches 0; It should be noted that when using the WLS method for fitting, since the weights of the abnormal observations are still 1, there will be a large deviation between the fitting result and the value to be estimated. However, since the GIRLS method isolates the abnormal observations, the fitting effect is much better than that of the WLS method; If the regression minimizes the robust scale of the residuals, the estimation is robust, that is:

[0084] where is a robust scale estimator that minimizes the robust scale estimator; ensuring the isolation of abnormal measurement values, the estimator for estimating the residual scale is defined as follows:

[0085] In the formula, , the function satisfies and is a bounded function, is an even function and is non-decreasing for ; The decision-making process of RE-DEORL is changed from a deterministic process to a stochastic process, and noise is added based on the output action of the policy to achieve exploratory expansion. The action finally executed by the environment is expressed as:

[0086] where is set to Gaussian white noise, and the mean μ is the output value of the policy network; Based on the continuous progress of the training process, the noise variance is continuously reduced to achieve the balance between exploration and exploitation; S5: Weight the control action output according to the weighted coefficients of the control policy group to generate the final propulsion instruction, distribute the final propulsion instruction to each propulsion subsystem of the aircraft, and each propulsion subsystem feeds back signals and sensor data to the experience pool and the health parameter estimation module, and perform policy update and fault perception accuracy iteration; Based on ensuring the consistency of the estimator under clean data, should be fixed to , where is the standard Gaussian distribution, and the scale estimator is solved iteratively as:

[0087] where is the relative standard residual, and the weight function:

[0088] Selected The function is:

[0089] The weight function is defined as:

[0090] It should be noted that the standard residual corresponds to an estimator with an unbounded function, that is and ; where represents the expectation of the function, is the standard Gaussian distribution, is the function, is the variance of the scale estimator, m is the number of samples, is the weight function, is the relative standard residual, is the threshold constant, is the state estimate value, is the scale estimate value; the lack of robustness in this case; understood as using an equal weight for all residuals for weighting, minimizing the robust scale of the residuals gives a highly robust estimator; The GIRLS method first calculates the measurement parameters using the least squares method:

[0091] Calculate the residuals based on the measurement parameters:

[0092] When all sensors are fault-free, i.e., there are no abnormal measurement points, the residual value is very small, a positive integer close to 0; when some sensors have abnormal measurement points, the residual value is large; Since is a dimensional ill-conditioned matrix; eliminate its similar rows, that is, if , then set the i-th row elements of to 0, where represents the i-th row elements of the matrix , and the ill-conditioned matrix tolerance error is a constant greater than 0; Using the above method will

[0093] The problem to be solved is transformed into the problem of solving a non - homogeneous linear equation system; by using the elementary row transformation of the determinant, the solution is obtained.

[0094] Among them, is the general solution, and the generalized residual is the particular solution. , the general solution term represents a series of combinations of measurement parameters that satisfy a residual of 0, and the particular solution term represents a series of combinations of measurement parameters that result in a non - zero residual. The larger the value of the elements in the particular solution term, the more abnormal the corresponding sensor measurement parameter is. The new weight function is defined as

[0095] The measurement prediction equation of GIRLS - EKF is

[0096] Among them, is the weight matrix, , is the measurement parameter estimated by the least - squares method, is the generalized inverse matrix of, r is the residual between the observed value and the estimated value, is the element of the i - th row of the matrix, is the tolerance error constant, is the newly defined weight function, is the residual term, is the threshold constant of the weight function, is the prior state covariance matrix, is the observation noise covariance matrix, is the input variable at the k - th step, which is the weight matrix composed of the weight function; The lowest - energy - consumption control objective is: under different flight modes, while outputting the desired thrust, minimize the energy consumption per unit thrust ; The control objective is used for mission control optimization of the aircraft during cruise, transition, low - altitude inspection, etc.; The control variables include: the electric propulsion power , the nozzle opening , the tilt angle , the nozzle deflection angle .

[0097] The constraint conditions include: the thrust error does not exceed the tolerance threshold; the propulsion power shall not exceed the limit value; the attitude angle change rate is limited to prevent sharp maneuvers; the nozzle area, rotation speed and voltage constraints; The mathematical form of the objective function is:

[0098] where λ is the control penalty factor used to balance the relationship between propulsion performance and energy efficiency, J is the objective function, Thrust generated The energy consumed when is the actual thrust generated by the thruster, is the desired target thrust value; It should be noted that the final problem can be reduced to an optimization problem with non - linear dynamic constraints. During the solution process, the control variable combination is continuously optimized through the RE - DEORL iterative strategy to approximate the optimal thrust - to - energy consumption ratio; The above has described an embodiment of the present invention in detail, but the content described is only a preferred embodiment of the present invention and cannot be considered as limiting the scope of implementation of the present invention; the above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation and historical experience and can be adjusted according to the actual situation; the above is only a preferred embodiment of the present invention and is not used to limit the present invention. All equivalent changes and improvements made according to the scope of the present invention application should still fall within the scope covered by the patent of the present invention.

Claims

1. A multi-mode propulsion control method for a low-altitude aircraft resistant to sensor failure, characterized in that: The following steps are involved: S1: Construct a nonlinear model of the propulsion component; S2: collect flight status and sensor observation parameters in real time, input them into the flight status identification module and health parameter estimation module, and identify the flight mode and health status of each propulsion component respectively; S3: introduce a weight embedding fusion network, input state residuals and health states, generate weight coefficients of the control strategy group, extract local disturbance features, model multi-channel state timing dependencies, and output residual weighted vectors; S4: The identification state and health state are jointly input into the control strategy group to promote control optimization, and the control action outputs are calculated respectively; S5: The control action output is weighted according to the weighted coefficient of the control strategy group to generate the final propulsion instruction, which is distributed to each propulsion subsystem of the aircraft. Each propulsion subsystem feeds back the signal and sensor data to the experience pool and health parameter estimation module, and performs strategy update and fault perception accuracy iteration.

2. A sensor failure-resistant low-altitude aircraft multi-mode propulsion control method according to claim 1, characterized in that: The specific method for obtaining the nonlinear model is: Define the control variable vector to describe the control process of the aircraft propulsion system; Define the control variable vector: ; in, and Represents the output power percentage of the two independent electric propulsion modules; Represents the deflection angle of the turbofan nozzle; Represents the angle of the tilting mechanism; Represents the tail nozzle opening of the turbofan or fan propeller; The variables constitute the main control input of the system and participate in the real-time optimization process; Define the system state vector to characterize the evolution of the propulsion system under different control inputs; Define the system state vector: ; in, Represents the combined thrust of the propulsion system at the current moment; Represents energy consumption per unit thrust; Represents the aircraft altitude; Represents the forward speed of the aircraft; Based on the nonlinear coupling between different propellers, the nonlinear coupling includes but is not limited to: the fast dynamic response of the electric propeller, the strong hysteresis of the turbofan system and the complex coupling of the airflow interference. The system state transfer process is realized through the function Perform non-linear mapping.

3. A sensor failure-resistant low-altitude aircraft multi-mode propulsion control method according to claim 1, characterized in that: The method for obtaining the health status of the propulsion component is: Based on the Actor network, the parameters are updated according to the reward and punishment values ​​of the Critic network for the action. At this time, the new loss function is: The policy gradient update formula used by the Actor network is as follows: The Critic network updates its parameters through the DQN algorithm. The gradient update formula is as follows: in, and They are the learning rates of the Actor network and the Critic network respectively; and They are the parameters of the Actor network and the Critic network respectively; Represents the state of the agent in the environment Next action The size of the reward value obtained; represents the discount factor, which is used to measure the importance of future rewards, Represents in state The action value obtained by the action generated by the Actor network and input to the Critic network represents the time difference error. Representative strategy About parameters In Status The gradient below Represents the strategy adopted by the Actor network; The reward value obtained based on the Actor-Critic model identifies the flight mode and the health status of each propulsion component.

4. A sensor failure-resistant low-altitude aircraft multi-mode propulsion control method according to claim 1, characterized in that: The method for obtaining the weighting coefficient is: when At the maximum number of rounds, a random process is initialized for action exploration , get the initial state , and calculate the initial residual ; Input to the residual perception network to obtain the residual weighted coefficient .

5. The multi-mode propulsion control method for low-altitude aircraft resistant to sensor failure according to claim 1, characterized in that: The method for obtaining the residual weighted vector is: Randomly initialize the current Actor network The weight parameter and the current critic network The weight parameter ; Initialize the target Actor network and target critic network , their respective network weight parameters are: ; Initialize experience replay pool Store state transfer and control residual information; Initialize the residual perception network and build a CNN + Transformer combined network; The network input is the residual sequence in the current state , which is the difference between the estimated state and the actual state; The CNN module extracts the local perturbation peak, and the Transformer module models the evolution of the residual over time and channels; The network output is the residual weight vector .

6. The multi-mode propulsion control method for low-altitude aircraft resistant to sensor failure according to claim 1, characterized in that: The method for propulsion control optimization is: By formula Rewritten as: Where: in, is the time derivative of the state variable, is the state matrix, which describes the influence of the state variable itself on its derivative. is the change of the state variable, is the input matrix, describing the effect of the input on the state derivative, is the change in the input variable, is a matrix, reflecting the influence on the state derivative, is the change in the additional variable, is the process noise, is the change in the output variable, is the output matrix, describing the influence of the state on the output, is the direct transfer matrix, which indicates the direct effect of input on output. is a matrix that describes the impact on the output. To measure noise, is the new state variable derivative, is the new state vector; Based on the design of EKF, the formula is: Convert to discrete form: The subscript k in the formula represents the value of the variable at the kth step; definition: initialization: Status prediction: The forgetting factor in the filter is: in, is the forgetting factor baseline value; and The matrix is: in, Measurement prediction: Nonlinear terms Errors and noise introduced during linearization compared to, Negligible noise in the vicinity; Performing a first-order Taylor expansion yields Then, we get the batch regression formula of EKF: in, ; The estimation problem in the above formula can be rewritten as The observation vector is , the regression matrix is , the noise is ,error The covariance matrix of The EKF update state calculated in the correction step , obtained by performing weighted least squares, linear regression: can be transformed into: The corrected step size of EKF can be obtained as follows: Among them, is the state vector of the k+1th step, is the discrete state transfer function, is the process noise of the k-th step, is the measured output of the kth step, is a discrete measurement function, is the measurement noise at the kth step, is the Jacobian matrix of the state transfer function for state x, is the Jacobian matrix of the measurement function for the state x, is the initial state estimate, is the covariance matrix of the initial state estimate, is the prior state estimate of the k-th step, is the state transition function of the kth step, is the posterior state estimate of the k-1th step, is the prior covariance matrix of the k-th step, is the forgetting factor of the kth step, is the state transition function for state x in The Jacobian matrix at , is the posterior covariance matrix of the k-1th step, is the process noise covariance matrix of the k-1th step, is the forgetting factor baseline value, To calculate the intermediate matrix of the forgetting factor, is the intermediate matrix, For the matrix traces, is the measurement function h for the state x in The Jacobian matrix at , is the initial covariance, is the measurement noise covariance matrix of the k-th step, is the Kalman gain matrix of the kth step, is the measured value at step k, is the posterior state estimate at step k, is the posterior covariance matrix of the kth step, h is the nonlinear observation function, is the true state of the kth step, is the observation noise at the kth step, is the prior state estimate of the k-th step, is the observation function h in The Jacobian matrix at , I is the n×n identity matrix, is the prior state estimation error, is the combined observation vector, is the combined regression matrix, is the combined noise vector, is the covariance matrix of the observation noise, is the covariance matrix of the prior state estimate, is the correction term for the covariance update.

7. The multi-mode propulsion control method for low-altitude aircraft resistant to sensor failure according to claim 1, characterized in that: The method for obtaining the control action output is: The decision-making process of RE-DEORL is changed from a deterministic process to a stochastic process, and noise is added based on the policy output action. Achieve exploratory expansion, and finally perform actions in the environment The expression is: Among them, Set to Gaussian white noise, the mean μ is the output value of the policy network; As the training process continues, the noise variance is continuously reduced to achieve a balance between exploration and utilization.

8. The multi-mode propulsion control method for low-altitude aircraft resistant to sensor failure according to claim 1, characterized in that: The method for obtaining the final advancement instruction is: The thrust error does not exceed the tolerance threshold; the thruster power must not exceed the limit value; the attitude angle change rate is limited to prevent sudden maneuvers; the nozzle area, speed and voltage are constrained; The mathematical form of the objective function is: Among them, J is the objective function, λ is the control penalty factor, Generate thrust The energy consumed is the actual thrust generated by the propeller, is the desired target thrust value.

9. The multi-mode propulsion control method for low-altitude aircraft resistant to sensor failure according to claim 1, characterized in that: The method for updating the strategy is: The measurement parameters are calculated using the least squares method: Calculate residuals based on measured parameters: because is a dimensional ill-conditioned matrix; eliminate its similar rows, that is, if , then let The i-th row element of is 0, where Representation Matrix The i-th row element of the ill-conditioned matrix tolerance error is a constant greater than 0; Using the above method, The problem to be solved is transformed into the problem of solving a system of non-homogeneous linear equations; using elementary row transformation of the determinant; the solution is in, is the general solution, the generalized residual For a special solution, , the general solution term represents the series combination of measurement parameters that satisfies the residual of 0, and the special solution term represents the series combination of measurement parameters that leads to the residual not being 0; The new weight function is defined as The measurement prediction equation of GIRLS-EKF is: in, is the weight matrix, , are the measurement parameters estimated by the least squares method, for The generalized inverse matrix of , r is the residual between the observed value and the estimated value, is the element in the i-th row of the matrix, is the tolerance error constant, is the newly defined weight function, is the residual term, is the threshold constant of the weight function, is the prior state covariance matrix, is the observation noise covariance matrix, is the input variable of step k, which is given by The weight matrix formed by the weight function.

10. The multi-mode propulsion control method for low-altitude aircraft resistant to sensor failure according to claim 1, characterized in that: The method for iterating the fault perception accuracy is: Based on ensuring the consistency of the estimator under clean data, Should be fixed to ,in is a standard Gaussian distribution, the scale estimator Solve by iteration: in, is the relative standard residual, weight function: Selected The function is: The weight function is defined as: in, express The expectation of the function, is a standard Gaussian distribution, for function, is the variance of the scale estimator, m is the number of samples, is the weight function, is the relative standard residual, is the threshold constant, is the estimated value of the state, is an estimated value of the scale.

Citation Information

Patent Citations

  • Minimum fuel consumption control optimization method for aero-engine based on DEORL

    CN117217071A

  • TBCC engine fault-tolerant airborne adaptive model modeling method

    CN119202497A

  • Multi-mode intelligent propulsion control method, device and equipment for short vertical aircraft and storage medium

    CN120143855A

  • Iterative learning control with sparse measurements for insulin injections in people with type 1 diabetes

    WO2021007391A1

  • Deep learning-based thermal management control method and system for electric drive bulldozer battery

    WO2022258079A1

Cited By

  • On-line model updating hybrid test method based on dual adaptive UKF (Unscented Kalman Filter) algorithm

    CN120724748A

  • DMG-MPC algorithm-based logistics AGV formation cooperative control method

    CN120972927A

  • Method for calculating nonlinear flight mechanics fitting modal characteristic parameters of connected-wing aircraft

    CN120974980A

  • Method, device and equipment for estimating disturbance parameters of aircraft, medium and aircraft

    CN122311073A