Civil aircraft flight control system dynamic reconstruction method fusing uncertainty quantization and toughness decision

Abstract

The invention provides a civil aircraft flight control system dynamic reconstruction method fusing uncertainty quantification and toughness decision, and belongs to the field of aviation electromechanical product reliability engineering, and the method comprises the steps: S1, multi-source uncertainty modeling and optimization; s2, running state monitoring and risk prediction; s3, evaluating the toughness margin of the flight control system; S4, generating a candidate toughness response strategy; s5, multi-attribute utility evaluation based on online model prediction; s6, selecting an optimal robust toughness decision; s7, executing an optimal robust toughness decision; and S8, empirical learning and model adaptive optimization are carried out. The invention provides a novel operation safety guarantee theory and method which deeply integrates uncertainty quantification, a toughness engineering principle and a multi-criterion robust decision-making aiming at severe uncertainty faced by a dynamic reconfigurable civil aircraft flight control system in a complex operation environment and challenge on flight safety. And the safety decision quality and the overall operation toughness of the system under uncertainty can be obviously improved.

Description

Technical Field

[0001] The present invention belongs to the field of aviation electromechanical product reliability engineering, and specifically relates to a method for dynamic reconstruction of a civil aircraft flight control system that integrates uncertainty quantification and resilience decision-making. Background Art

[0002] Civil aircraft flight control systems (FCSs) are undergoing a profound transformation from traditional fixed architectures to intelligent, autonomous, and dynamically reconfigurable ones. Dynamic reconfiguration technology provides FCSs with unprecedented flexibility, enabling real-time adjustments to control laws, sensor fusion strategies, and actuator allocation schemes based on complex and changing route conditions, unexpected airport conditions, component aging or failure, and even refined fuel efficiency optimization requirements. This significantly improves flight performance, mission adaptability, and fault tolerance. However, the flip side of this technological advancement is a dramatic increase in operational uncertainty. Dynamic reconfiguration itself introduces new configuration spaces and transition processes. Combined with the inherent meteorological uncertainty, fluctuating air traffic flow, variability in crew status and decision-making, and potential hazards introduced by maintenance activities, these factors create a complex and dynamically evolving uncertainty network. The coupling and propagation of these uncertainties pose a significant challenge to traditional safety assurance methods that rely on deterministic failure assumptions and static analysis, such as fault tree analysis (FTA) and failure mode and effects analysis (FMEA), making it difficult to effectively address the emerging risks brought about by dynamic reconfiguration.

[0003] Against this backdrop, resilience engineering (RE), as a proactive safety management paradigm, offers a new theoretical perspective for addressing these challenges. The core concept of RE is to acknowledge that the inherent uncertainty and variability of complex systems are the norm rather than the exception, emphasizing that systems should be able to not only avoid catastrophic failures under both expected and unexpected disturbances, but also maintain their core functions, proactively adapt to environmental changes, and learn from experience. Applying RE principles to the operational safety assurance of dynamically reconfigurable civil aircraft FCSs, and exploring new theories and methods that can quantify uncertainty, assess resilience levels, and support dynamic optimization decisions, is of vital theoretical significance and urgent engineering application value for fully realizing the potential of dynamic reconfigurable technology while meeting the extremely high safety standards of civil aviation, which typically require a probability of catastrophic events below 10-9 per flight hour.

[0004] In recent years, scholars at home and abroad have conducted numerous beneficial explorations in related fields. However, existing research still has significant shortcomings in the following areas: First, there is a lack of systematic, integrated modeling and quantitative propagation analysis of the multi-source heterogeneous uncertainty of dynamically reconfigurable civil aircraft FCSs during the operational phase; second, research on translating abstract resilience engineering principles into concrete, operational, and quantifiable engineering methods and technical indicators is still in-depth; third, existing decision support methods often struggle to balance robustness and real-time requirements under dynamic uncertainty; and finally, research results that deeply integrate uncertainty quantification, resilience engineering concepts, and complex decision-making theory to form a complete closed-loop operational safety assurance solution are particularly scarce.

[0005] Therefore, there is an urgent need to provide a new method to ensure the operational safety of dynamically reconfigurable civil aircraft FCS. Summary of the Invention

[0006] In order to address the deficiencies of the above-mentioned prior art, the present invention proposes a dynamic reconstruction method for civil aircraft flight control systems that integrates uncertainty quantification and resilience decision-making. It constructs a dynamic safety assurance closed-loop process with resilience capability as the core element and runs through the entire FCS operation process. Resilience capability includes anticipation, monitoring, response, and learning. Combined with a robust resilience decision-making method based on online model prediction and multi-attribute utility optimization, it can support the adaptive reconstruction and response of FCS under uncertainty.

[0007] Specifically, on one hand, the present invention provides a method for dynamic reconstruction of a civil aircraft flight control system that integrates uncertainty quantification and resilience decision-making, which includes the following steps: S1. Multi-source uncertainty modeling and optimization: A dynamic Bayesian network that can describe multi-source uncertainty is used to construct a dynamic influence diagram model, and the dynamic influence diagram model is optimized and updated using real-time observation data; S2. Operational status monitoring and risk prediction: After obtaining a dynamic understanding of the current uncertainty state, continuously monitor the performance of key systems and predict future short-term risks using the dynamic expected failure probability based on the dynamic influence diagram model; S3. Evaluate the resilience margin of the flight control system: Evaluate the current resilience margin of the flight control system and determine whether to perform a resilient response on the flight control system based on the resilience margin. If so, proceed to step S4; otherwise, return to step S2. S4. Generate candidate resilience response strategies: Generate a set of candidate resilience response strategies The response strategy includes one or more of the following: dynamic reconfiguration of the FCS, adjustment of flight parameters, modification of mission objectives, and coordinated operations of the crew. They are the 1st, 2nd, ..., Nth strategies respectively; S5. Multi-attribute utility evaluation based on online model prediction: Using the optimized dynamic influence graph model, each candidate strategy Conduct forward prediction and multi-dimensional assessment of potential consequences after implementation; S6. Select the optimal robust and resilient decision: Introduce robustness considerations and select the decision with the highest score under uncertainty disturbance as the optimal robust and resilient decision. ; S7, execute the optimal robust resilience decision: the optimal robust resilience decision selected in step S6 Converted into specific control instructions or operation suggestions, and executed by the flight control system; S8. Experience learning and model adaptive optimization: Through a continuous feedback learning loop, the dynamic influence diagram model and decision logic are iteratively optimized using actual operation data, and the dynamic influence diagram model is continuously fed back for learning and parameter update.

[0008] Preferably, step S1 includes the following sub-steps: S11. A dynamic Bayesian network is used to describe the uncertainty in the dynamic evolution of the system by defining the instantaneous dependency relationship between variables within a time slice and the state transition relationship between variables between adjacent time slices, thereby obtaining a dynamic influence diagram model. S12. Define the identified uncertainty factors as chance nodes in the dynamic Bayesian network; S13. Define the probability law of node state evolution over time and construct a state transition model of dynamic Bayesian network; S14. Online estimation of the posterior probability distribution of all latent variables in the dynamic Bayesian network using real-time observation data streams provided by onboard sensors ,The dynamic influence diagram model is optimized based on the posterior probability distribution.

[0009] Preferably, step S14 specifically includes the following sub-steps: S141, initialization, t=0: from the prior distribution Extract particles , and assign initial weights ; S142, prediction: For each particle , according to the system state transition model , from the state transition probability Sampling new particle states; S143, Update: Based on the latest observation data and observation model , calculate each predicted particle Importance weight of S144, resampling: when the number of effective particles When the value is below a certain threshold, resampling is performed.

[0010] Preferably, in step S4, candidate resilience response strategies are generated based on a predefined rule base, case-based reasoning, model-based optimization search, or human-computer collaboration.

[0011] Preferably, typical evaluation attributes of the multi-attribute utility evaluation in step S5 include safety attributes, mission performance attributes, economy / efficiency attributes, passenger / crew experience attributes, and system resilience enhancement attributes.

[0012] Preferably, the continuous feedback learning method in step S8 includes dynamic Bayesian network parameter and structure learning, utility function calibration and preference learning.

[0013] Preferably, step S2 specifically includes the following sub-steps: S21. Operation status monitoring: S211. Key performance indicator monitoring: Key performance indicators include flight path tracking error, attitude control accuracy, stability margin, fuel consumption rate, and ride comfort; S212, Safety Boundary Monitoring: Define multi-dimensional and dynamically changing safety operating boundaries based on aircraft design specifications, airworthiness provisions and flight manuals ; S22. Short-term risk prediction: Using the updated As the initial state, through the state transition model Perform forward recursion to predict the system state in one or more time steps in the future The probability distribution of the system state, and then, the calculation system state violates the predefined safety boundary Probability of: ; ; Among them, the computing system state does not violate the predefined security boundary Probability For a given current state Under this condition, for the next ΔT / δ time steps, each time step j is from t+1 to t+ΔT / δ, the system state Not in the unsafe set under nominal control The expectation of the product of the probabilities .

[0014] Preferably, step S3 specifically includes the following sub-steps: S31. Calculation of toughness margin : ; in: is the weight coefficient, ; , reflecting the current short-term security expectations; It measures the richness and expected effects of the available reconstruction options or adjustment strategies when the flight control system encounters new disturbances or uncertain changes in its current state. It is a measure of the expected speed and degree to which a system can recover to an acceptable performance level after a performance degradation or deviation has occurred. S32. Determine whether to perform a resilient response on the flight control system based on the resiliency margin: when Below the preset dynamic threshold , or short-term DEPoF Higher than When the current toughness is judged to be insufficient, the process goes to step S4 to perform a toughness response on the flight control system.

[0015] Preferably, step S5 specifically includes the following sub-steps: S51. Extend the dynamic influence diagram model for forward simulation and result prediction: For each candidate strategy , instantiate it as a decision node in the dynamic influence diagram model, and then, from the current moment The state posterior probability distribution Starting from, combined with the state transition model of dynamic Bayesian network , for each particle Perform forward propagation to predict the entire prediction time domain System Status The probability distribution of the final possible outcome The probability distribution of ; S52. Construct a multi-attribute utility function to evaluate the comprehensive effects of different strategies. The multi-attribute utility function is as follows: ; in: The final result state from the prediction and / or the entire state trajectory The extracted The value of the evaluation attribute; It is Single attribute utility function of attributes; It is The weight coefficient of each attribute indicates the relative importance of the attribute in the overall decision. ; S53. Calculate the expected utility of each strategy: When using particle filtering for state prediction, the current state By particle set For each candidate strategy , whose expected utility is approximated by applying the strategy to each particle and performing a forward simulation.

[0016] Preferably, the method for introducing robustness considerations in step S6 is robust decision-making based on the worst prospect, a chance-constrained programming method, or a multi-objective optimization and Pareto frontier analysis method.

[0017] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention is the first to deeply and systematically integrate uncertainty quantification theory, especially dynamic probability graphical models, the core principles of resilience engineering (emphasizing dynamic adaptation and learning), and robust multi-attribute decision-making theory, and applies it to the specific field of dynamic reconfigurable civil aircraft FCS operational safety assurance, forming a new theoretical perspective.

[0018] (2) The present invention designs a complete and operational closed-loop operation safety assurance process, which clearly defines the key links and their internal connections from uncertainty modeling and perception, risk anticipation, resilience assessment, decision triggering, strategy generation, utility evaluation, robust selection, to execution monitoring and learning optimization.

[0019] (3) This paper proposes a series of quantitative resilience indicators for operational decision-making, such as dynamic expected probability of failure DEPoF and resilience margin RM, and constructs a multi-attribute utility optimization decision-making model based on online model prediction and considering the worst prospect, which significantly enhances the scientific nature and reliability of decision-making.

[0020] (4) The core of the method of the present invention is to evaluate and select different system configurations (reconfiguration modes) and response strategies. It is naturally suitable for the security assurance requirements of dynamically reconfigurable systems and surpasses traditional security analysis methods for fixed architecture systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a schematic diagram of the workflow of the present invention; Figure 2 Schematic diagram of the structure of a civil aircraft flight control system in an embodiment of the present invention; Figure 3 Schematic diagram of the dynamic Bayesian network model of the present invention; Figure 4 This is a schematic diagram of the state transition of the dynamic Bayesian network model of the present invention; Figure 5 A radar chart comparing key performance indicator predictions under different decision strategies in an implementation case of the present invention; Figure 6 is the robust utility value of different response strategies in the implementation case of the present invention Comparison bar chart; Figure 7 Schematic diagram comparing the expected total utility and robustness penalty term of different response strategies in the implementation case of the present invention. DETAILED DESCRIPTION

[0022] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

[0023] The present invention provides a theory and method for ensuring the operation safety of a dynamically reconfigurable civil aircraft flight control system that integrates uncertainty quantification and resilience decision-making. Figure 1 As shown, it includes the following steps: S1. Multi-Source Uncertainty Modeling and Optimization: A dynamic Bayesian network capable of describing multi-source uncertainty is used to construct a dynamic influence diagram model, which is then optimized and updated using real-time observation data. The core of this step is to construct a probabilistic graphical model that dynamically reflects the various uncertainties and their interactions within the civil aircraft FCS operating environment and to update it using real-time observation data. Figure 2 This is a schematic diagram of the structure of a civil aircraft flight control system in an embodiment of the present invention. This step specifically includes the following steps: S11. Model Selection and Construction: Dynamic Bayesian Networks (DBNs) are used as the core modeling tool. DBNs describe the uncertainty in the dynamic evolution of a system by defining the instantaneous dependencies between variables within a time slice and the state transitions between adjacent time slices, resulting in a dynamic influence diagram model. For the decision-making and utility components, DBNs naturally extend to dynamic influence diagrams (DIDs). Figure 3 Schematic diagram of the dynamic Bayesian network model of the present invention. Figure 4 Schematic diagram of state transition of the dynamic Bayesian network model of the present invention.

[0024] S12. Define the identified uncertainty factors as chance nodes in the dynamic Bayesian network: Define the identified environmental uncertainty factors as , uncertainty factors of the system's internal state , human uncertainty factors , and the uncertainties associated with the reconstruction process , are defined as chance nodes in DBN. These nodes can be discrete states, such as the ice level is divided into "none, light, medium, heavy", or they can be obtained by reasonable discretization of continuous variables. Wind shear intensity ,visibility , icing level , the uncertainty factor of the system internal state Such as sensors Health Factors , hydraulic system pressure , controls the remaining computing power of the computer , human uncertainty factors Such as the unit's compliance with the current automation level , the crew's situational awareness level of complex situations , uncertainty factors related to the reconstruction process Such as specific reconstruction mode The probability of successful switching , the expected performance of the new configuration after reconstruction )wait.

[0025] S13. Define the probability law of node state evolution over time and construct the state transition model of dynamic Bayesian network: define the probability law of node state evolution over time, that is, ,in Represents at the moment The actions taken by the system include control inputs from the FCS, operating instructions from the crew, and active reconfiguration decisions. For example, sensor health factors The evolution of may follow a physics-based degradation model such as: ; This is usually a discretized degradation model with a conditional probability table adjusted for intensity of use and environmental stress. is continuous, its evolution can be expressed as: ; in, is the degradation function, is the process noise. Similarly, the evolution of meteorological conditions It can be parameterized based on Markov chain models from meteorology or the output of more complex numerical weather prediction models.

[0026] S14. Online estimation of the posterior probability distribution of all latent variables in the dynamic Bayesian network using real-time observation data streams provided by onboard sensors , optimize the dynamic influence diagram model based on the posterior probability distribution: use the real-time observation data stream provided by onboard sensors such as atmospheric data computer, inertial measurement unit, engine monitoring unit, cockpit voice / operation recorder, etc. , through the sequential Monte Carlo method such as particle filtering or its variants such as unscented particle filtering UPF, regularized particle filtering RPF, etc., the posterior probability distribution of all latent variables in DBN, that is, the uncertainty factors that are not directly observed, is estimated online Particle filtering approximates the posterior distribution by maintaining a set of weighted samples, or particles, and is effective in handling nonlinear, non-Gaussian dynamic systems. Each particle represents a possible hypothesis about the system state, and its weight is updated based on its consistency with the observed data.

[0027] Particle filter mathematical principle: The goal is to recursively estimate the state vector The posterior probability density function (PDF) of the set of all latent variables in the DBN The particle filter uses a set of weighted random samples, namely particles to approximate this posterior PDF, where is the number of particles, It is Particles at time status, is its corresponding weight, and The posterior PDF can be approximated as: ; in, is the Dirac delta function.

[0028] It includes the following sub-steps: S141, Initialization (t=0): From the prior distribution Extract particles , and assign initial weights .

[0029] S142, prediction: For each particle , according to the system state transition model , from the importance proposal distribution is usually chosen as the state transition probability Sample the new particle state: ; For DBN, this means that for each particle state , according to its parent node in Status and , and CPTs, sampled of each component.

[0030] S143, Update: Based on the latest observation data and observation model , calculate each predicted particle Importance weight: ; The weights are then normalized: ; S144, Resampling: In order to avoid the particle degradation problem that a few particles have most of the weight, when the number of effective particles Below a certain threshold, e.g. ), resampling is performed. Resampling is based on the current weight From the particle set Extract with Replacement A new particle , and reset its weight to Commonly used resampling methods include polynomial resampling, systematic resampling, stratified resampling, etc.

[0031] By iteratively performing prediction, update, and necessary resampling steps, the particle filter is able to continuously track the posterior probability distribution of the system state.

[0032] S2. Operational Status Monitoring and Risk Prediction: After gaining a dynamic understanding of the current uncertainty state, continuously monitor key system performance and predict future short-term risks based on a dynamic influence diagram model. After gaining a dynamic understanding of the current uncertainty state, continuously monitor key system performance and anticipate potential risks in the short term. This step proposes using the dynamic expected probability of failure (DEPoF), a quantitative resilience metric for operational decision-making, as a key parameter.

[0033] S21. Operation status monitoring: Key performance indicators (KPIs) and safe operating envelope (SOE) monitoring, including the following sub-steps: S211. Key performance indicator monitoring: such as flight path tracking error, attitude control accuracy, stability margin (such as structural stress margin under gust loads, stall margin, roll oscillation margin), fuel consumption rate, and ride comfort (such as vertical overload root mean square value).

[0034] S212, Safety Boundary Monitoring: Define multi-dimensional, dynamically changing safety operating boundaries based on aircraft design specifications, airworthiness provisions such as CCAR / FAR Part 25, and flight manuals (AFM / FCOM). For example, the maximum allowable overload at different altitudes and Mach numbers, the minimum control speed at different icing levels, etc.

[0035] S22、Short-term risk expectation based on DBN: Using the updated As the initial state, through the DBN state transition model Forward recursion is usually performed, assuming that no active decision is taken within the short-term prediction window, that is, For nominal control, one or more time steps into the future are predicted, e.g. System status within the time window Then, the computation system state violates the predefined safety boundary. Probability of: ; ; Among them, the computing system state does not violate the predefined security boundary Probability For a given current state Under this condition, for the next ΔT / δ time steps, each time step j is from t+1 to t+ΔT / δ, the system state Not in the unsafe set under nominal control The expectation of the product of the probabilities .

[0036] This probability is an important component of the dynamic expected failure probability DEPoF in the short term. In actual calculations, Monte Carlo simulation is usually used: The initial state of the sample, It is represented by a particle set of particle filtering, and then a forward simulation is performed to count the proportion of paths that violate the boundary.

[0037] S3. Evaluate the resilience margin of the flight control system: Evaluate the current resilience margin of the flight control system, and determine whether to perform a resilient response to the flight control system based on the resilience margin. If so, proceed to step S4, otherwise return to step S2. In order to determine whether the system needs to initiate an active resilience response including reconstruction, define and evaluate the current resilience margin of the system. This step proposes the resilience margin, a quantitative resilience indicator for operational decision-making, which can better judge the decision trigger. Specifically includes the following sub-steps: S31. Calculate toughness margin using the quantitative model of toughness margin : Not only should current security, such as 1-DEPoF, be considered, but also the system’s potential to cope with future uncertainties. A more comprehensive The definition can be: ; in: is the weight coefficient, .

[0038] , reflecting current security expectations in the short term.

[0039] : Measures the richness and expected effectiveness of the available reconstruction options or adjustment strategies for a system in its current state if it encounters new disturbances or uncertainties. For example, it can be defined as the number of effective reconstruction strategies that can reduce the DEPoF to below a certain safety level under current conditions, or its average expected utility gain.

[0040] : Measures the expected speed and degree to which a system can recover to acceptable performance levels after a performance degradation or deviation has occurred. For example, the expected recovery time or performance recovery percentage after adopting an optimal recovery strategy can be estimated based on historical data or simulations.

[0041] S32. Determine whether to perform a resilient response on the flight control system based on the resiliency margin: when Below the preset dynamic threshold , or short-term DEPoF Higher than When the system considers the current resilience insufficient, it needs to initiate a proactive resilience decision-making process. This threshold can be dynamically adjusted based on factors such as flight phases such as takeoff, cruise, approach and landing, and mission criticality.

[0042] S4. Generate candidate resilience response strategies: Generate a set of candidate resilience response strategies The response strategy includes one or more of the following: dynamic reconfiguration of the FCS, adjustment of flight parameters, modification of mission objectives, and coordinated operations of the crew. These strategies are item 1, item 2, ..., item N. These strategies include not only the dynamic reconfiguration of the FCS, but may also include adjustments to flight parameters, modifications to mission objectives, and even collaborative operations with the crew. Figure 5 A radar chart comparing key performance indicator predictions under different decision strategies is shown. Strategies can be generated based on: S41, Predefined Rule-Based System, RBS: Rules are usually in the form of: IF (condition 1 AND condition 2 AND ...) THEN (recommended strategy ). For example: IF ( = "High Turbulence" AND = "Exception" AND <"critical threshold") THEN (consider strategy : "Activate Bump Mode" AND : "Isolate the suspicious angle of attack sensor"). The conditional part of the rule is usually a judgment of the state of certain nodes in the DBN, which are determined by given.

[0043] S42. Case-Based Reasoning (CBR): The core idea of ​​CBR is to use past experience to solve new problems. The process usually includes: S421, Retrieve: According to the current question (such as A feature vector describing the system state and risk , retrieve the most similar historical cases from the case library Similarity It can be calculated by weighted feature distance: ; in, It is a feature The weight of It is a feature The local similarity function between them is the inverse of the distance for numerical features and the matching degree for symbolic features.

[0044] S422, Reuse: The most similar case retrieved The solution is the historical response strategy As an initial candidate strategy for the current problem.

[0045] S423, Revise: Adjust and optimize the candidate strategy based on the differences between the current problem and the search case. This may require additional domain knowledge or rules. For example, if the current turbulence intensity is higher than the case, the parameters of the turbulence mode may need to be enhanced.

[0046] S424, Retain: When a new problem and its solution are verified to be effective, they are stored as a new case in the case library to enable learning. The new case includes a description of the problem, the adopted strategy, and the actual results.

[0047] S43. Model-Based Optimization Search: If there is a simplified evaluable function To quickly evaluate strategies In the current state For short-term effects such as short-term risk reduction or simplification utility, a heuristic search algorithm can be used. For example, a genetic algorithm (GA): S431, Encoding and Population Initialization: The strategy Encoded as chromosomes such as binary strings or parameter vectors. Randomly generate a set of candidate strategies (chromosomes) as the initial population .

[0048] S432, fitness evaluation: population Each strategy ,use Or a simplified utility function to calculate its fitness .

[0049] S433, selection: according to fitness Select excellent strategies to enter the next generation, such as roulette selection and tournament selection, to form the parent population .

[0050] S434, Cross: The strategy in the probability Perform crossover operations such as single-point crossover and multi-point crossover to generate new offspring strategies.

[0051] S435, mutation: each gene in the offspring strategy is mutated with probability Perform mutation operations to introduce new diversity.

[0052] S436, forming a new generation of population Repeat steps 2-6 until the termination condition is met, such as reaching the maximum number of iterations. Or find a satisfactory solution. Finally, the excellent individuals in the population constitute the candidate strategy set .

[0053] S44. Human-Machine Collaborative Generation: For complex or unexpected scenarios, allow crew members to participate in the development and selection of strategies. Generated strategies should be diverse, covering different risk preferences, ranging from conservative strategies (such as immediately entering a safe mode) to aggressive strategies (such as attempting to maintain mission performance through detailed reconstruction).

[0054] S5. Multi-attribute utility evaluation based on online model prediction: Using the optimized dynamic influence graph model, each candidate strategy The core of this step is to use the DBN / DID model that has been built and updated in real time to predict the potential consequences of each candidate strategy. The potential consequences after execution are forward-lookingly predicted and evaluated in multiple dimensions.

[0055] S51. Extend DID for forward simulation and result prediction: For each candidate strategy , instantiate it as a decision node in DID. Then, from the current moment The state posterior probability distribution (by particle set Representation), combined with the state transition model of DBN (Here As an action or initial action that lasts in the prediction time domain), through sequential Monte Carlo simulation, that is, for each particle Perform forward propagation or analytical / approximate reasoning to predict one or more key time nodes in the future or the entire forecast time domain System Status The probability distribution of the final possible outcome Such as the probability distribution of safe landing, mission completion, accident occurrence, etc. .

[0056] S52. Construction and Quantification of Multi-Attribute Utility Function (MAUF): To evaluate the combined effects of different strategies, a MAUF is constructed that reflects the trade-off preferences of decision makers, such as airlines, pilots, and regulators, among multiple objectives, such as safety, mission performance, economy, and passenger comfort. This function typically takes a weighted sum or multiplicative form. A general weighted sum form is as follows: (4); in: The final result state from the prediction and / or the entire state trajectory The extracted The value of an evaluation attribute.

[0057] It is The single attribute utility function or value function of an attribute maps the physical measurement value of the attribute, such as fuel consumption, flight time, and overload peak, to [0,1] or a standardized utility scale, reflecting the different levels of satisfaction with the attribute. For example, can be a decreasing function of the predicted DEPoF, It can be an increasing function of the probability of task completion.

[0058] It is The weight coefficient of each attribute indicates the relative importance of the attribute in the overall decision. These weights can be adjusted dynamically based on the flight phase, such as takeoff, cruise, approach, mission criticality, and the current risk situation. For example, in an emergency, the weight of safety Will improve significantly.

[0059] Typical evaluation attributes This can include: 1) Safety attributes: such as the maximum DEPoF within the prediction time domain, the cumulative probability of violating the safety margin, and the minimum margin to key safety thresholds such as the stall boundary and the structural load limit.

[0060] 2) Mission performance attributes: such as the probability of successful mission completion, the degree of achievement of key mission objectives, and the degree of deviation from the flight profile.

[0061] 3) Economic / efficiency attributes: such as expected fuel consumption, flight time, and component life loss. The component life loss refers to the failure to adopt certain high-load fault-tolerant strategies.

[0062] 4) Passenger / crew experience attributes: such as predicted ride comfort (based on overload, vibration, etc.) and crew workload level.

[0063] 5) System resilience enhancement attributes: After executing this strategy, the system's remaining resource margin, such as energy, computing power, available redundant components, and the ability to cope with subsequent potential disturbances.

[0064] S53. Calculate the expected utility of each strategy: When using particle filtering for state prediction, the current state By particle set For each candidate strategy , its expected utility can be approximately calculated by applying the strategy to each particle and performing a forward simulation: S531, for each particle , in strategy Next, use the DBN state transition model conduct Forward Monte Carlo simulation of the step. For simplicity, it is assumed that each particle only generates one predicted trajectory and the final result .

[0065] S532: Calculate the utility of the predicted path / result .

[0066] S533, Strategy The total expected utility of is the weighted average of the utilities produced by all particles: .

[0067] If for each particle conduct Independent random forward simulations yield Trajectory and utility ,but .

[0068] S6. Select the optimal robust and resilient decision: Introduce robustness considerations and select the decision with the highest score under uncertainty disturbance as the optimal robust and resilient decision. Due to the uncertainty of the model itself (such as the imprecision of CPT parameters) and the unpredictability of the future environment, simply maximizing expected utility may lead to underestimation of potential risks. Therefore, it is necessary to introduce robustness considerations and select decisions that can still perform well under uncertain perturbations.

[0069] S61. Robust decision-making based on the worst-case scenario: A common approach is to consider the statistical properties of utility distribution, such as choosing a strategy that maximizes "expected utility minus a certain multiple of the standard deviation," as in formula ( ) as shown: ; in, is a risk aversion coefficient, The larger it is, the more decision makers tend to avoid large fluctuations in utility, that is, to pursue a more robust strategy. Given current information and taking a strategy Under the condition of , the standard deviation of the future total utility and its square variance can be calculated by the particle set: ; .

[0070] S62. Introducing Chance Constrained Programming: Another approach is to meet certain security constraints, such as the probability that DEPoF is below a certain minimum threshold is not less than Under the premise of maximizing the expected utility related to task performance.

[0071] ; ; in, is a small probability value representing the tolerable violation probability of the safety constraint. It can be estimated by counting the proportion of particles that meet this condition in the forward simulation.

[0072] S63. Multi-objective optimization and Pareto frontier analysis: Taking safety, performance, cost, etc. as multiple conflicting objectives, a multi-objective optimization algorithm such as NSGA-II is used to find a set of Pareto optimal solutions for decision makers or higher-level autonomous decision logic to make the final choice based on the specific situation.

[0073] Through step S6, the system selects decisions that can still perform well under uncertain disturbances, and then proceeds to step S7 to closely and continuously monitor the key state parameters of the system, the execution progress of the decision, and environmental changes. Then, it proceeds to step S8 to form a continuous feedback learning loop, and continuously optimizes its internal model and decision logic using actual operation data.

[0074] S7, execute the optimal robust resilience decision: the optimal robust resilience decision selected in step S6 Converted into specific control instructions or operation suggestions and executed by the flight control system. These are translated into specific control instructions or operational recommendations, which are then executed by the FCS's actuators, such as the autopilot, flight management system, or the flight crew. During this execution, critical system status parameters, the progress of decision execution, and environmental changes must be closely and continuously monitored to promptly detect deviations in execution or new, unexpected conditions.

[0075] S8. Experience Learning and Model Adaptive Optimization: Through a continuous feedback learning loop, actual operation data is used to iteratively optimize the dynamic influence diagram model and decision logic. The DBN / DID model is continuously fed back and updated with learning parameters to ensure the system's self-evolution. Its ability to recognize uncertainty, the accuracy of risk prediction, and the effectiveness of resilient decision-making will continue to improve with the accumulation of operational experience. Through a continuous feedback learning loop, actual operation data is used to iteratively optimize the dynamic influence diagram model and decision logic. The DBN / DID model is continuously fed back and updated with learning parameters to ensure the system's self-evolution. Its ability to recognize uncertainty, the accuracy of risk prediction, and the effectiveness of resilient decision-making will continue to improve with the accumulation of operational experience.

[0076] S81, DBN parameter and structure learning: 1) Parameter learning: Compare the actual observation results after each decision execution, such as sensor readings, component state changes, environmental evolution, reconstruction success or failure, actual performance, etc., with the prior predictions of the DBN model. If there is a systematic deviation, Bayesian parameter update methods such as Dirichlet-Multinomial conjugate prior update for discrete variables or Kalman filter-type parameter estimation algorithms for continuous variables are used to adjust the CPT parameters of the relevant nodes in the DBN online or offline. For example, if the node The parent node is , its CPT is , observed Second-rate At the same time, in the Dirichlet prior The posterior parameters are: .

[0077] 2) Structural Learning: When conditions permit and there is sufficient data support: If long-term observations indicate that the model structure, i.e., the dependencies between variables, are significantly inconsistent with the actual system behavior, consider using DBN structural learning methods based on score search, such as BIC scoring, MDL scoring combined with hill climbing algorithms, simulated annealing algorithms, or constraint-based methods, such as PC algorithms, to optimize the model structure locally or globally. This can help discover potential causal relationships that were previously unrecognized.

[0078] S82. Utility Function Calibration and Preference Learning: 1) Weight adjustment: By analyzing the choices made by decision makers such as pilots in simulators in historical decision-making cases, or the actual preferences of experts for different outcomes in the subsequent evaluation of automated decisions, the weight coefficients of each attribute in the MAUF can be adjusted in reverse. , making it more consistent with the value orientation in actual operation. This can be achieved through Inverse Reinforcement Learning (IRL) or Preference Learning techniques.

[0079] 2) Single-attribute utility function Morphological learning: If it is found that the satisfaction with a certain attribute, such as ride comfort, is not as linear or simply nonlinear as the preset relationship, a large number of "attribute value-satisfaction score" data pairs can be collected and non-parametric regression or machine learning methods such as Gaussian process regression can be used to fit a more accurate single-attribute utility function curve.

[0080] Through the above-mentioned learning mechanism, the entire safety assurance system can achieve self-evolution, and its ability to recognize uncertainty, the accuracy of risk prediction, and the effectiveness of resilient decision-making will continue to improve with the accumulation of operating experience.

[0081] To better understand the present invention, the following uses a specific application scenario—"Ensuring the Resilience and Safety of a Certain Civil Aircraft FCS Under Sudden Severe Weather and Sensor Performance Degradation"—with reference to the accompanying figures to illustrate the implementation steps of the present system and method. Based on scientifically sound model parameters and flight data assumptions, this case study will focus on uncertainty reasoning, calculation of key resilience indicators, and comparative risk-benefit analysis of different reconstruction / response strategies. The case study background is as follows: (1) Research object: This case study selected a typical dual-channel fly-by-wire long-range wide-body passenger aircraft as the research object. The FCS of this type of aircraft has a high degree of automation and a certain degree of fault-tolerant reconstruction capability, including multiple control law modes such as Normal Law, Direct Law, Turbulence Damping Mode, dynamic adjustment of sensor data source selection and fusion logic such as isolating faulty sensors, switching to backup sources or estimated values, and redundant switching capabilities of some actuators. The core of the FCS is two primary flight control computers (PFCCs) and two backup flight control computers (SFCCs), which can process data from three independent air data systems (ADSs), three inertial reference systems (IRSs), and two GPSs.

[0082] (2) Operational scenario: The aircraft was in the high-altitude cruising phase of the North Atlantic route with parameters of FL370 and Mach number 0.82, and was scheduled to fly from point A to point B. According to the en route weather forecast, a moderate turbulence area was expected in the middle of the flight path. However, during the actual flight, the aircraft encountered a strong and sudden atmospheric turbulence (Clear Air Turbulence, CAT) that was far beyond expectations. At the same time, a key angle of attack sensor in the main atmospheric data system ADS-1, Alpha Vane 1, AoA1, began to show intermittent data jumps and signs of decreased accuracy. This composite disturbance scenario of "bad weather plus key sensor performance degradation" posed a severe test to the stability of the FCS and the crew's control.

[0083] (3) Candidate reconstruction / response strategies available for FCS ( ): In this specific scenario, the candidate strategies that can be considered by the system or unit with system assistance mainly include: 1) Maintain the current cruise mode, using the Normal Law strategy, using ADS-1's AoA1 as the primary source. The control law and sensor fusion logic remain unchanged, relying on the robustness of the FCS.

[0084] 2) Activates Turbulence Damping Mode. This mode enhances the ability to suppress high-frequency wind gust disturbances by adjusting control gains and filtering characteristics, but this may come at the expense of some ride comfort (more frequent active control) and a slight increase in fuel consumption.

[0085] 3) ADS Refactoring: Isolate the suspect AoA1 sensor and adjust the data fusion logic. Reduce or eliminate the weight of ADS-1 AoA1 readings in state estimation and control law calculations, relying more on ADS-2 AoA2 (assuming it is functioning properly, but with slightly lower baseline accuracy than AoA1) or other estimates (such as AoA2 estimates based on GPS / INS data and aerodynamic models, which are less accurate). This may reduce the overall accuracy of AoA information, impacting flight performance and the accuracy of some protective functions.

[0086] 4) : Combined strategy: Activate the "bump mitigation" control law mode and perform ADS reconstruction ( ).

[0087] 5) Request ATC clearance to change altitude or heading to avoid areas of severe turbulence. This strategy is the most direct way to reduce environmental disturbances, but it can significantly impact flight plans, increase flight time and fuel consumption, and may conflict with other air traffic.

[0088] The present invention provides a theory and method for ensuring the operation safety of a dynamically reconfigurable civil aircraft flight control system that integrates uncertainty quantification and resilience decision-making. Figure 1 As shown, it includes the following steps: S1. Uncertainty modeling and parameter assumptions (based on scientific rationality): To simplify the demonstration, we construct a DBN / DID model that includes core uncertainty factors. minute.

[0089] S11. Key uncertainty nodes (chance nodes) and their states and (partial) CPT assumptions: S111, (Turbulence intensity): {weak (W), moderate (M), strong (S), very strong (VS)}.

[0090] (1) Initial observation: =M (based on forecast). When an emergency occurs, it is monitored by onboard sensors (such as accelerometers) and the posterior is updated by particle filtering. =VS.

[0091] (2) State transfer (Example, simplified Markov chain, with one transfer per minute): Table 1 Overview of turbulence intensity state transitions S112, (Main angle of attack sensor 1 health status): {Normal (N), Intermittent abnormal (I), Complete failure (F)}.

[0092] (1) Initial: = N. When an emergency occurs, the built-in test equipment (BITE) and data consistency check are performed, and the posterior is updated by particle filtering. =I.

[0093] (2) State transfer (Example, accelerated degradation under high stress): P( =I | =N, =VS) = 0.15; P( =F | =I, =VS) = 0.20; P( =I | =I, =VS) = 0.70 (maintain intermittent abnormality); (Other state transition probabilities are omitted).

[0094] S113, (Comprehensive angle of attack information accuracy): {High (H), Medium (M), Low (L)}. Its CPT depends on , backup angle of attack sensor (This example assumes =N always holds) and whether (Adjust data fusion) strategy.

[0095] (1) If no (Right now ): P( =H | =N, =No) = 0.98; P(M)=0.02; P(L)=0; P( =H | =I, =No) = 0.20; P(M)=0.50; P(L)=0.30; P( =H | =F, =No) = 0.00; P(M)=0.20; P(L)=0.80.

[0096] (2) If (Right now ): (Isolate AoA1, use AoA2 or combined estimate): P( =H | Any, =Yes) = 0.10 (because AoA2 accuracy is slightly lower or the estimate is more inaccurate); P( =M | Any, =yes) = 0.70; P( =L | Any, =yes) = 0.20.

[0097] S114, (Pitch stability margin): {Sufficient( 15%), General (10-15%), Critical (5-10%), Insufficient ( 5%)}. Its CPT depends on , , and the current control law mode ( Normal, for Turbulence Damping).

[0098] (1) Example: =VS, =L, ControlMode= (Normal): P(inadequate)=0.60; P(critical)=0.30; P(general)=0.10; P(sufficient)=0.00.

[0099] (2) Example: =VS, =M, ControlMode= (Turbulence Damping): P(inadequate)=0.05; P(critical)=0.20; P(general)=0.60; P(sufficient)=0.15.

[0100] (3) (Other CPT items are set based on engineering experience and simulation data); S115, (Ride comfort index, 0-10, the higher the better): Discretized into {excellent (8-10), good (6-7), medium (4-5), poor (<4)}. Depends on and control law models.

[0101] =VS, ControlMode= : P(poor)=0.9; P(medium)=0.1; =VS, ControlMode= (or ): P(poor)=0.5; P(fair)=0.4; P(good)=0.1 (the bump pattern improved but still poor); S116, (Fuel consumption factor per unit time, >1 indicates an increase relative to the nominal value): : 1.00; : 1.05; : 1.00; : 1.05; : 1.25 (significant increase in fuel consumption due to rerouting and altitude / speed changes) S12, decision node :That is, the aforementioned .

[0102] S13. Utility Function The attributes and weights of (Formula 8) ) (Cruise phase, safety first): (1) : Safety utility. If the margin is "insufficient", the utility is -100; "critical" is -20; "average" is 5; "sufficient" is 10. Weight .

[0103] (2) : Comfort utility. If the comfort is "poor", the utility is -10; "medium" is 0; "good" is 5; "excellent" is 10. Weight .

[0104] (3) : Fuel economy utility. Defined as , that is, relative to the nominal fuel consumption, for every 5% increase in the factor, the utility decreases by 0.5. Weight .

[0105] (4) : Flight plan deviation utility. If you select Changing the route will produce a significant negative utility of -50, and other strategies will be 0. .

[0106] S2. Initial state and disturbance occurrence: The aircraft is in nominal mode Down flight ( =M, =N). Incident: Airborne sensors detect severe turbulence, and the BITE system detects anomalies in the AoA1 data. Particle filtering updates the DBN evidence, and the posterior probability of the current state is highly concentrated at: =VS (very strong turbulence), =I (intermittent abnormality of main angle of attack sensor).

[0107] S3, DBN / DID online reasoning and risk / resilience assessment (with Take the nominal strategy as an example, forecasting the next 5 minutes): S31. Evidence update: =VS, =I Enter DBN as currently known.

[0108] S32, state prediction (in No action is taken): S321. First, according to =I and (Right now = No), confirm the current Distribution: P(H)=0.20, P(M)=0.50, P(L)=0.30.

[0109] S322, using DBN to perform forward Monte Carlo simulation ( times, for example ), predict the evolution of each state in the next 5 minutes. : Evolves according to its Markov chain transition probability.

[0110] According to its The evolution of the transition probability of influence.

[0111] according to and The CPT is determined.

[0112] according to , , and (Normal Mode) CPT determination.

[0113] S323, statistics in the next 5 minutes, The path ratio is Assume that the simulation results are: ; ; but .

[0114] S324, .

[0115] S325, suppose that at this time (The system has multiple alternative reconstruction strategies), (If you can recover from the current predicament, the effect is average).

[0116] S326, usage weight (Safety first, while taking into account adaptation and recovery).

[0117] S327, .

[0118] S328, Assumption . because , the current resilience margin is insufficient, triggering the proactive decision-making process.

[0119] S4. Candidate reconstruction / response strategy generation: rule / case / optimization-based strategy sets .

[0120] S5. Robust utility evaluation of candidate strategies (assuming risk aversion coefficient :For each , forward Monte Carlo simulation via DBN / DID ( times), predict the average state distribution in the next 5 minutes, and calculate the utility of each single attribute based on this Expected value , and then get the total expected utility (Formula 5) and the standard deviation of utility (formula ).

[0121] by For example, the specific derivation and The process is as follows: : Based on the average probability within 5 minutes .

[0122] .

[0123] (Assume Continuous VS, ControlMode= ): .

[0124] .

[0125] : .

[0126] : .

[0127] .

[0128] : Assume that by The simulation results The sample statistics show that the standard deviation is 15.0.

[0129] Robust utility ( ) = .

[0130] Similarly, for other strategies Perform the same simulation and calculation. Assume that the simulation and calculation results are summarized as follows: Table 2 Comparison of robust utility evaluation results of different response strategies (detailed derivation and assumed data) Note: is the cumulative probability that the predicted stability margin is lower than the "normal" state. The utility value is calculated based on the aforementioned assumed CPT, single-attribute utility function, and weights. StdDev is the assumed standard deviation of the simulation results.

[0131] Status description of each mode: for (Maintain status quo): Calculated as detailed above. Safety utility is low due to high stability margin risk caused by strong turbulence and angle of attack sensor anomalies. Comfort is poor.

[0132] for (Bump Mode): Significantly improves stability ( significantly reduced to 0.25), Improved. Comfort slightly improved to due to more active control (Assuming that in bumpy mode, P(bad)=0.5, P(medium)=0.4, P(good)=0.1). Fuel efficiency is slightly reduced. (Corresponding to FuelFactor=1.05.) The standard deviation decreases as the system responds more proactively.

[0133] for (Adjust ADS fusion): Due to the isolation of AoA1, relying on less accurate AoA2 or estimation, the improvement in stability is not as good as obvious( is 0.45). Fuel efficiency and same.

[0134] for (Combination strategy): The most significant improvement in stability ( down to 0.10), Higher. Comfort and fuel economy affect similar.

[0135] for (Avoidance route): Assuming a successful avoidance, turbulence is reduced to a weak level, and the sensor issue persists but has a reduced impact. Safety and comfort are optimally restored. Although the negative utility of fuel and plan deviation is significant, the total expected utility and robust utility are still the highest. The standard deviation is minimized, indicating a more certain outcome.

[0136] S6. Decision selection and analysis: According to the robust utility value in Table 2 ( ),Strategy The robust utility value of (avoidance route) is the highest (2.535). And its expected total utility (4.035) is also the highest among all strategies. While effectively improving safety ( Only 0.02, the lowest among the system adjustments), and the uncertainty of the results is also small ( ). This shows that after accounting for outcome volatility and risk aversion, is a more robust option. If the decision maker is extremely risk averse ( value is larger), The advantages will be more obvious.

[0137] In this case, the system will recommend an execution strategy .

[0138] To draw a radar chart, each indicator needs to be positively skewed and normalized to the range of [0,1].

[0139] Saf ( ): .

[0140] Comf ( ): . .

[0141] Fuel ( ): . Corresponding to maximum fuel consumption (such as -2.5), .

[0142] Plan ( ): . .

[0143] Table 3 Indicators in each state mode The radar chart will visually compare the prediction performance of different decision strategies (focusing on d1, d2, d4, and d5) in multiple key dimensions such as safety, comfort, fuel efficiency, and plan adherence after positive and normalization under the current uncertainty conditions. At the expense of a certain degree of fuel efficiency and plan adherence, safety and comfort are significantly improved, showing good overall resilience.

[0144] In order to more intuitively compare the final evaluation results of each strategy, Figure 6 The robust utility values ​​of different response strategies are shown ( ) bar chart.

[0145] from Figure 6 It can be seen that the robust utility values ​​of each candidate strategy are calculated according to formula (3'). The bar corresponding to (avoidance route) is the highest, indicating that it is the optimal robust choice in the current situation.

[0146] Figure 7 The composition of the robust utility value is further explained, and the expected total utility of each strategy is compared through grouped bar charts ( ) and the “robustness penalty” due to uncertainty ( ). The figure shows the expected total utility (blue bars, assumption) and robust penalty term (orange bars, assumption, γ=0.5) for each strategy. Robust utility value U robust This is the value of the blue bar minus the value of the orange bar. For example, strategy d5 has the highest expected total utility (4.035) and a relatively small uncertainty penalty, giving it a leading robust utility. Strategy d1, while having an expected utility of -16.655, suffers from a large uncertainty penalty, which further deteriorates its robust utility.

[0147] S7. Response Execution and Effect Monitoring: The selected optimal robust and resilient decision d5 is converted into specific control instructions or operational recommendations, which are executed by the flight control system's actuators (such as the autopilot, flight management system) or the crew. During execution, key system state parameters, the decision's execution progress, and environmental changes must be closely and continuously monitored.

[0148] S8, learning and optimization: If you perform After the strategy is implemented, the actual fuel consumption rate continues to be higher than the model prediction value (for example, by comparing the actual fuel consumption recorded by QAR data with the model-based The learning module will use this data to update the Bayesian parameters and increase the The CPT value of the relevant fuel consumption rate node (or adjust distribution parameters), or adjusting the fuel economy in the utility function Scoring criteria / weight , so that future decisions can more accurately reflect the actual impact of fuel consumption, thereby improving the overall effectiveness of decision-making.

[0149] This case study uses a civil aircraft operation scenario to demonstrate how the proposed method can integrate uncertainty information to make multi-attribute, robust resilience decisions. By quantifying the risks of different reconstruction / response strategies (e.g. From the strategy The 0.63 is reduced to the strategy 0.02) and returns (strategy (The robust utility value is 2.535, which is better than other system internal adjustment strategies). This method can provide powerful, data-driven decision support for the flight control system or crew, thereby improving flight safety and operational resilience in a systematic and quantitative manner under complex disturbance environments.

[0150] The embodiments described above are merely descriptions of preferred implementations of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should fall within the scope of protection determined by the claims of the present invention.

Claims

1. A method for dynamic reconfiguration of a civil aircraft flight control system that integrates uncertainty quantification and resilient decision-making, characterized by: It includes the following steps: S1. Multi-source uncertainty modeling and optimization: A dynamic Bayesian network that can describe multi-source uncertainty is used to construct a dynamic influence diagram model, and the dynamic influence diagram model is optimized and updated using real-time observation data; S2. Operational status monitoring and risk prediction: After obtaining a dynamic understanding of the current uncertainty state, continuously monitor the performance of key systems and predict future short-term risks using the dynamic expected failure probability based on the dynamic influence diagram model; S3. Evaluate the resilience margin of the flight control system: Evaluate the current resilience margin of the flight control system and determine whether to perform a resilient response to the flight control system based on the resilience margin. If so, proceed to step S4; otherwise, return to step S2. S4. Generate candidate resilience response strategies: Generate a set of candidate resilience response strategies The response strategy includes one or more of the following: dynamic reconfiguration of the FCS, adjustment of flight parameters, modification of mission objectives, and coordinated operations of the crew. They are the 1st, 2nd, ..., Nth strategies respectively; S5. Multi-attribute utility evaluation based on online model prediction: Using the optimized dynamic influence graph model, each candidate strategy Conduct forward prediction and multi-dimensional assessment of potential consequences after implementation; S6. Select the optimal robust and resilient decision: Introduce robustness considerations and select the decision with the highest score under uncertainty disturbance as the optimal robust and resilient decision. ; S7, execute the optimal robust resilience decision: the optimal robust resilience decision selected in step S6 Converted into specific control instructions or operation suggestions, and executed by the flight control system; S8. Experience learning and model adaptive optimization: Through a continuous feedback learning loop, the dynamic influence diagram model and decision logic are iteratively optimized using actual operation data, and the dynamic influence diagram model is continuously fed back for learning and parameter update.

2. The method for dynamic reconfiguration of a civil aircraft flight control system integrating uncertainty quantification and resilient decision-making according to claim 1 is characterized by: Step S1 includes the following sub-steps: S11. A dynamic Bayesian network is used to describe the uncertainty in the dynamic evolution of the system by defining the instantaneous dependency relationship between variables within a time slice and the state transition relationship between variables between adjacent time slices, thereby obtaining a dynamic influence diagram model. S12. Define the identified uncertainty factors as chance nodes in a dynamic Bayesian network; S13. Define the probability law of node state evolution over time and construct a state transition model of dynamic Bayesian network; S14. Online estimation of the posterior probability distribution of all latent variables in the dynamic Bayesian network using real-time observation data streams provided by onboard sensors ,The dynamic influence diagram model is optimized based on the posterior probability distribution.

3. The method for dynamic reconfiguration of a civil aircraft flight control system integrating uncertainty quantification and resilient decision-making according to claim 2 is characterized by: Step S14 specifically includes the following sub-steps: S141, initialization, t=0: from the prior distribution Extract particles , and assign initial weights ; S142, prediction: For each particle , according to the system state transition model , from the state transition probability Sampling new particle states; S143, Update: Based on the latest observation data and observation model , calculate each predicted particle Importance weight of S144, resampling: when the number of effective particles When the value is below a certain threshold, resampling is performed.

4. The method for dynamic reconfiguration of a civil aircraft flight control system integrating uncertainty quantification and resilient decision-making according to claim 1 is characterized by: In step S4, candidate resilience response strategies are generated based on a predefined rule base, case-based reasoning, model-based optimization search, or human-computer collaboration.

5. The method for dynamic reconfiguration of a civil aircraft flight control system integrating uncertainty quantification and resilient decision-making according to claim 1 is characterized by: Typical evaluation attributes of the multi-attribute utility evaluation in step S5 include safety attributes, mission performance attributes, economy / efficiency attributes, passenger / crew experience attributes, and system resilience enhancement attributes.

6. The method for dynamic reconfiguration of a civil aircraft flight control system integrating uncertainty quantification and resilient decision-making according to claim 1 is characterized by: The continuous feedback learning method in step S8 includes dynamic Bayesian network parameter and structure learning, utility function calibration and preference learning.

7. The method for dynamic reconfiguration of a civil aircraft flight control system integrating uncertainty quantification and resilient decision-making according to claim 1 is characterized by: Step S2 specifically includes the following sub-steps: S21. Operation status monitoring: S211. Key performance indicator monitoring: Key performance indicators include flight path tracking error, attitude control accuracy, stability margin, fuel consumption rate, and ride comfort; S212, Safety Boundary Monitoring: Define multi-dimensional and dynamically changing safety operating boundaries based on aircraft design specifications, airworthiness provisions and flight manuals ; S22. Short-term risk prediction: Using the updated As the initial state, through the state transition model Perform forward recursion to predict the system state in one or more time steps in the future The probability distribution of the system state, and then, the calculation system state violates the predefined safety boundary Probability of: ； ； Among them, the computing system state does not violate the predefined security boundary Probability For a given current state Under this condition, for the next ΔT / δ time steps, each time step j is from t+1 to t+ΔT / δ, the system state Not in the unsafe set under nominal control The expectation of the product of the probabilities .

8. The method for dynamic reconfiguration of a civil aircraft flight control system integrating uncertainty quantification and resilient decision-making according to claim 1 is characterized by: Step S3 specifically includes the following sub-steps: S31. Calculation of toughness margin : ； in: is the weight coefficient, ; , reflecting the current short-term security expectations; It measures the richness and expected effects of the available reconstruction options or adjustment strategies when the flight control system encounters new disturbances or uncertain changes in its current state. It is a measure of the expected speed and degree to which a system can recover to an acceptable performance level after a performance degradation or deviation has occurred. S32. Determine whether to perform a resilient response on the flight control system based on the resiliency margin: when Below the preset dynamic threshold , or the short-term dynamic expected failure probability Higher than When the current toughness is judged to be insufficient, the process goes to step S4 to perform a toughness response on the flight control system.

9. The method for dynamic reconfiguration of a civil aircraft flight control system integrating uncertainty quantification and resilient decision-making according to claim 1 is characterized by: Step S5 specifically includes the following sub-steps: S51. Extend the dynamic influence diagram model for forward simulation and result prediction: For each candidate strategy , instantiate it as a decision node in the dynamic influence diagram model, and then, from the current moment The state posterior probability distribution Starting from, combined with the state transition model of dynamic Bayesian network , for each particle Perform forward propagation to predict the entire prediction time domain System Status The probability distribution of the final possible outcome The probability distribution of ; S52. Construct a multi-attribute utility function to evaluate the comprehensive effects of different strategies. The multi-attribute utility function is as follows: ； in: The final result state from the prediction and / or the entire state trajectory The extracted The value of the evaluation attribute; It is Single attribute utility function of attributes; It is The weight coefficient of each attribute indicates the relative importance of the attribute in the overall decision. ; S53. Calculate the expected utility of each strategy: When using particle filtering for state prediction, the current state By particle set For each candidate strategy , whose expected utility is approximated by applying the strategy to each particle and performing a forward simulation.

10. The method for dynamic reconfiguration of a civil aircraft flight control system integrating uncertainty quantification and resilience decision-making according to claim 1 is characterized by: The method for introducing robustness considerations in step S6 is robust decision-making based on the worst prospect, chance-constrained programming method, or multi-objective optimization and Pareto frontier analysis method.

Images