Flight trajectory control method and system based on operational state awareness and command learning

Abstract

This invention belongs to the field of data processing technology and relates to a flight trajectory control method and system based on operational state perception and command learning. It aims to solve the problems of difficult deviation correction, weak dynamic adaptability, and insufficient fusion of command and state perception in traditional methods under complex disturbances. The invention includes: acquiring flight data; constructing a joint input tensor and obtaining a hidden state sequence through an encoder; obtaining an operational probability distribution based on a prediction model, and then determining the predicted roll angle command; obtaining the basic trajectory and disturbance compensation amount based on the roll angle command and wind disturbance sequence, respectively, and correcting them through a residual correction network to obtain the lateral offset trajectory; determining the roll angle command correction amount under safety envelope constraints to correct the lateral motion state; and performing safety boundary detection after correction. This invention, through real-time perception of operational state and fusion of command learning, can accurately capture disturbances and characteristic changes, effectively improving control efficiency, adaptability, and intelligence.

Description

Technical Field

[0001] This invention belongs to the field of flight control technology, specifically relating to a flight trajectory control method and system based on operational state perception and command learning. Background Technology

[0002] In the aviation safety system, and specifically in the aerospace field, precise control of flight trajectory is one of the core technologies for ensuring the safe and efficient operation of a vehicle. As flight missions become increasingly complex, vehicles face various dynamic disturbances during actual flight, including environmental disturbances such as atmospheric turbulence and gusts, as well as inherent changes in characteristics such as aging of the vehicle's power system and delays in actuators. The combined effect of these factors can cause deviations between the actual flight trajectory and the preset commands. If these deviations are not corrected in time, they may lead to excessive trajectory deviations, increased energy consumption, or even safety accidents.

[0003] Traditional flight trajectory control methods largely rely on pre-set mathematical models, establishing control equations with fixed parameters to achieve trajectory planning and correction. However, such methods require extremely high model accuracy. When faced with complex and changing operational states (such as sudden command adjustments or extreme weather conditions), fixed models struggle to adaptively capture dynamic changes, easily leading to control lag or overshoot problems. Furthermore, traditional methods have limited learning capabilities for operator input commands, failing to effectively correlate historical commands with trajectory responses, resulting in insufficient control robustness under multi-task switching or atypical operating conditions.

[0004] In recent years, with the development of intelligent sensing and machine learning technologies, data-driven control methods have gradually become a research hotspot. However, in existing technologies, most solutions only focus on state perception or command optimization, failing to achieve a deep integration of dynamic perception of operational state and command learning, resulting in a difficulty in simultaneously achieving real-time performance and accuracy in trajectory control.

[0005] Therefore, how to construct a technical solution that can accurately sense the flight operation status and achieve trajectory stability control under dynamic disturbances has become a key problem that urgently needs to be solved in this field. Summary of the Invention

[0006] To address the aforementioned problems in existing technologies, namely, the difficulty in correcting trajectory deviations under complex disturbances in flight control methods, the weak dynamic adaptability of models, and the insufficient fusion of commands and state perception, the first aspect of this invention proposes a flight trajectory control method based on operational state perception and command learning, the method comprising the following steps:

[0007] Acquire flight data, which includes wind disturbance sequences, operation sequences, and reference roll angle command sequences;

[0008] A joint input tensor is constructed based on the operation sequence and the reference roll angle command sequence. After being processed by the encoder, a hidden state sequence is obtained. The hidden state sequence is then input into a preset prediction model to obtain the operation probability distribution, and then the predicted roll angle command is determined.

[0009] A kinematic model is constructed based on the roll angle command to obtain the basic trajectory, a wind disturbance model is constructed based on the wind disturbance sequence to obtain the disturbance compensation amount, and the synthesis result of the basic trajectory and the disturbance compensation amount is corrected by the residual correction network to obtain the lateral offset trajectory.

[0010] The range of the safety envelope is adjusted according to the lateral offset trajectory, and under the constraint of the safety envelope, the roll angle command correction amount is determined based on the lateral offset trajectory. The lateral motion state of the carrier is corrected according to the roll angle command correction amount.

[0011] After correction, a safety boundary check is performed. If the check result does not fall within the safety envelope, the correction amount is recalculated and the correction operation is repeated. If the safety envelope requirements are met, monitoring continues.

[0012] In some preferred embodiments, the method for constructing a joint input tensor based on the operation sequence and the reference roll angle command sequence, and obtaining the hidden state sequence after encoder processing, is as follows:

[0013] The operation sequence and the reference roll angle instruction sequence are processed using learnable embedding functions respectively, mapped to the same semantic space, and the two types of embedding vectors are concatenated in dimension to form a joint input tensor.

[0014] The joint input tensor is used as the input sequence. After being processed by an encoder for extracting temporal features, a hidden state sequence is obtained. The hidden state sequence contains contextual information and temporal dependencies of the input sequence.

[0015] In some preferred embodiments, the predicted roll angle command is determined by:

[0016] Based on the hidden state sequence, the operation probability distribution is obtained through a prediction model;

[0017] The operation type with the highest probability under the operation probability distribution is taken as the predicted editing operation type for the current time step;

[0018] Based on the predicted editing operation type and combined with a preset set of operations, the roll angle instruction is determined.

[0019] In some preferred embodiments, the roll angle instruction is determined based on the predicted editing operation type and a preset set of operations, using the following method:

[0020] Construct an operation set based on historical data:

[0021] ;

[0022] in, The first step is to obtain the correction amount from the clustering historical data. k Roll angle historical correction amount , , , These represent different operation types. This indicates that the reference roll angle command remains unchanged. This indicates that a second instruction is added based on the baseline instruction. k Class of corrections, This means reducing the type k correction amount based on the baseline instruction. This indicates that the reference instruction is directly replaced. k =1...K, where K is the number of clusters;

[0023] Based on the correspondence between the measurement and editing operation types and the operation set, the roll angle instruction is determined. :

[0024] ;

[0025] ;

[0026] in, As the reference roll angle command, Offset compensation for REPLACE operation type In hidden state, To map high-dimensional hidden features to linear weights of roll angle commands, To predict the type of editing operation, For probability distribution, To find the type of editing operation with the highest probability under the operation probability distribution.

[0027] In some preferred embodiments, the method for obtaining the lateral offset trajectory is as follows:

[0028] The kinematic model is used as input, and the basic trajectory of the lateral displacement of the carrier under undisturbed conditions is calculated through preset kinematic equations. ;

[0029] ;

[0030] The wind disturbance model is based on the temporal characteristics of the wind disturbance sequence and outputs the disturbance compensation amount for the lateral displacement of the carrier by fitting the wind field action law. ;

[0031] ;

[0032] The residual correction network obtains residual correction by learning the model error patterns in historical trajectory data. ;

[0033] ;

[0034] ;

[0035] Based on the aforementioned base trajectory, disturbance compensation amount, and residual correction, the corrected lateral offset trajectory is obtained:

[0036] ;

[0037] in, Let airspeed be constant. For time step, Wind direction angle The cosine of the wind direction angle. The predicted roll angle instruction for the current time step is sinusoidalized. The predicted lateral offset for the previous time step. The predicted roll angle instruction for the current time step. The instantaneous wind speed and volume of the wind disturbance at the current time step. Output the hidden layer dimensions to the encoder. The input feature vector is the residual correction network, and MLP is a multilayer perceptron. These are learnable parameters in an MLP. In hidden state, It is the set of real numbers.

[0038] In some preferred embodiments, the flight data further includes the lateral offset trajectory of the carrier at the current moment, flight speed, offset safety threshold, and envelope boundary slope; the method for determining the roll angle command correction based on the lateral offset trajectory under the safety envelope constraint is as follows:

[0039] Based on the lateral offset trajectory, the roll angle command correction sequence is obtained by reverse calculation;

[0040] Based on the genetic algorithm, the roll angle command correction sequence is encoded as an individual of the genetic algorithm. The population is iterated by combining the fitness function, and the roll angle command correction sequence corresponding to the best individual is selected as the candidate optimal solution.

[0041] Based on the Lagrange multiplier method, with the objective function of minimizing the pilot's operational workload and the safety envelope as the constraint, the optimal correction value among the candidate optimal solutions is used as the roll angle command correction value.

[0042] A second aspect of this invention proposes a flight trajectory control system based on operational state perception and command learning, the system comprising:

[0043] The data acquisition module is configured to acquire flight data in real time. The flight data includes the current lateral offset trajectory, flight speed, offset safety threshold, envelope boundary slope, wind disturbance sequence, operation sequence, and reference roll angle command sequence.

[0044] The preprocessing module is configured to construct a joint input tensor based on the operation sequence and the reference roll angle instruction sequence, and obtain the hidden state sequence after being processed by the encoder;

[0045] The prediction module is configured to input the hidden state sequence into a preset prediction model to obtain the operation probability distribution, and then determine the roll angle command based on the operation probability distribution.

[0046] The offset analysis module is configured to construct a kinematic model based on the roll angle command to obtain the basic trajectory, construct a wind disturbance model based on the wind disturbance sequence to obtain the disturbance compensation amount, and correct the synthesis result of the basic trajectory and the disturbance compensation amount through a residual correction network to obtain the lateral offset trajectory.

[0047] The safety envelope adjustment module is configured to analyze the offset trend based on the lateral offset trajectory and adjust the range of the safety envelope.

[0048] The correction module is configured to determine the roll angle command correction amount based on the lateral offset trajectory under the safety envelope constraint, and correct the lateral motion state of the carrier according to the roll angle command correction amount;

[0049] The feedback module is configured to perform safety boundary detection after correction. If the detection result does not fall within the safety envelope, it returns to recalculate the correction amount and repeats the correction operation. If it meets the safety envelope requirements, it continues to monitor.

[0050] The beneficial effects of this invention are:

[0051] 1. The pilot's stick operation and the reference roll angle command are embedded and fused to form a joint input tensor, which is then encoded using an encoder. The command parsing and trajectory correction strategies are continuously optimized through a learning mechanism, reducing the reliance on manually preset parameters and achieving a deep integration of dynamic perception of operation status and command learning.

[0052] 2. An instruction editing learning mechanism was constructed, defining an operation set containing multiple editing operations. By predicting the operation type at each time step, the baseline roll angle instruction was precisely edited and adjusted, achieving optimized instruction generation and improving the flexibility and adaptability of flight control.

[0053] 3. By combining physical kinematic equations and residual correction networks, and introducing wind disturbance modeling, the trajectory prediction can not only reflect physical laws but also adapt to environmental changes, thereby improving anti-interference capabilities and ensuring the safe flight of aircraft under complex weather conditions.

[0054] 4. A dynamic weighted joint training strategy is proposed. By dynamically adjusting the loss weights of tasks such as editing operation, trajectory prediction and wind disturbance handling, the problem of imbalance of loss among tasks in multi-task learning is solved, the overall performance of the model is improved, and effective balance and collaborative optimization among different tasks are achieved.

[0055] 5. By using genetic algorithms and the Lagrange multiplier method to optimize the roll angle command correction under the safety envelope constraint, the lateral deviation trajectory of the aircraft is accurately controlled within the safe range, thus realizing real-time protection and dynamic adjustment of flight safety. Attached Figure Description

[0056] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0057] Figure 1 This is a flowchart of a flight trajectory control method based on operation state perception and instruction learning in an embodiment of the present invention. Detailed Implementation

[0058] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0059] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0060] This invention provides a flight trajectory control method based on operational state perception and instruction learning. By sensing operational state in real time and fusing instruction learning, this method can accurately capture disturbances and characteristic changes to reduce trajectory deviation and improve dynamic response speed. It breaks through the limitations of fixed models, adapts to complex working conditions, and enhances robustness. While improving control efficiency, adaptability, and intelligence, it provides support for the efficient and safe execution of flight missions.

[0061] The present invention provides a flight trajectory control method based on operation state perception and command learning, comprising the following steps:

[0062] Acquire flight data, which includes wind disturbance sequences, operation sequences, and reference roll angle command sequences;

[0063] A joint input tensor is constructed based on the operation sequence and the reference roll angle command sequence. After being processed by the encoder, a hidden state sequence is obtained. The hidden state sequence is then input into a preset prediction model to obtain the operation probability distribution, and then the predicted roll angle command is determined.

[0064] A kinematic model is constructed based on the roll angle command to obtain the basic trajectory, a wind disturbance model is constructed based on the wind disturbance sequence to obtain the disturbance compensation amount, and the synthesis result of the basic trajectory and the disturbance compensation amount is corrected by the residual correction network to obtain the lateral offset trajectory.

[0065] The range of the safety envelope is adjusted according to the lateral offset trajectory, and under the constraint of the safety envelope, the roll angle command correction amount is determined based on the lateral offset trajectory. The lateral motion state of the carrier is corrected according to the roll angle command correction amount.

[0066] After correction, a safety boundary check is performed. If the check result does not fall within the safety envelope, the correction amount is recalculated and the correction operation is repeated. If the safety envelope requirements are met, monitoring continues.

[0067] To more clearly explain the flight trajectory control method based on operation state perception and command learning of this invention, the following will be combined with... Figure 1 The steps in the embodiments of the present invention will be described in detail below.

[0068] The flight trajectory control method based on operation state perception and command learning according to the first embodiment of the present invention includes steps S1-S5, each of which is described in detail below:

[0069] S1. Acquire flight data, including wind disturbance sequences. , operation sequence and reference roll angle command sequence ;

[0070] in, The variable representing the reference roll angle command sequence indicates the time step. t The roll angle command generated based on the basic flight control logic serves as the pilot's initial expectation of the aircraft's roll angle. The dimension is T, which is the reference roll angle command value corresponding to T time steps.

[0071] Preferably, the flight data also includes the lateral offset trajectory, flight speed, offset safety threshold, and envelope boundary slope at the current moment.

[0072] In this embodiment, the operation sequence is acquired through a flight parameter recorder and a data glove worn by the pilot; the wind disturbance sequence is acquired through a weather radar and an airspeed tube, and only the lateral (perpendicular to the runway direction) component is considered; the carrier is an aircraft.

[0073] In this embodiment, the reference roll angle command sequence is obtained through the PD controller:

[0074] ;

[0075] in, This represents the current lateral offset error. This represents the rate of change of the current lateral offset error. These are the proportional and derivative parameters of the controller, respectively.

[0076] This application innovatively embeds and integrates the pilot's stick operation with the reference roll angle command to form a joint input tensor. It uses the Transformer architecture for encoding to achieve dynamic perception and modeling of the pilot's operational intentions, providing a foundation for subsequent command editing.

[0077] S2. Construct a joint input tensor based on the operation sequence and the reference roll angle command sequence, and obtain the hidden state sequence after encoder processing; input the hidden state sequence into a preset prediction model to obtain the operation probability distribution, and then determine the predicted roll angle command. .

[0078] Preferably, the method for constructing a joint input tensor based on the operation sequence and the reference roll angle command sequence, and obtaining the hidden state sequence after encoder processing, is as follows:

[0079] The operation sequence and the reference roll angle command sequence are processed using learnable embedding functions, mapped to the same semantic space, and the two types of embedding vectors are concatenated in dimensionality to form a joint input tensor. ;

[0080]

[0081] in, For learnable embedding functions, d The embedding dimension is the vector dimension resulting from the transformation of the original pilot stick input and reference roll angle commands through an embedding function. It is used to map different types of input data to the same semantic space. T ×2 d This represents the dimension of the joint input tensor. 2d The dimension of the concatenated vector;

[0082] The joint input tensor is used as the input sequence, and after being processed by a Transformer encoder for extracting temporal features, the hidden state sequence is obtained. The hidden state sequence contains contextual information and temporal dependencies of the input sequence.

[0083] More preferably, the location encoding uses relative time encoding, the method of which is as follows:

[0084] ;

[0085] ;

[0086]

[0087] in, A hidden state sequence containing contextual information and temporal dependencies of the input sequence, with dimension . ; For the hidden state sequence at time step t The corresponding hidden state is an element in the encoder's output sequence; d h It is the output hidden layer dimension of the Transformer encoder. The positional encoding values ​​at different time steps and dimensions are calculated using sine and cosine functions. It is used to provide the model with temporal sequence information and help the model understand the temporal relationship in the input sequence.

[0088] Relative time encoding plays a role in the model's processing of input sequences (such as a joint input tensor consisting of pilot operation sequences and reference roll angle command sequences), ensuring that the model can grasp the sequential logic of input information in the time dimension, and providing temporal contextual support for subsequent tasks such as editing operation prediction.

[0089] Preferably, the method for determining the predicted roll angle command is as follows:

[0090] Construct an operation set based on historical data:

[0091] ;

[0092] in, To obtain the historical roll angle correction value by clustering historical data, an action set is obtained by enumerating typical correction values. , , , These represent different operation types, used to define the operations that the model can perform on the baseline roll angle command at each time step. k =1....K;

[0093] Based on the hidden state sequence, the predicted value at the time step is obtained through a prediction model. t Operation probability distribution ;

[0094] ;

[0095] These are the learnable parameters of the model. This is the weight matrix. This is a bias term used to adjust the hidden state output by the encoder. The probability of conversion into an operation has a distribution; It is the dimension identifier of the real vector space, and the output probability vector. have Each dimension corresponds to a probability of a certain operation type;

[0096] The operation type with the highest probability under the operation probability distribution is taken as the predicted editing operation type for the current time step. ;

[0097] ;

[0098] Based on the predicted editing operation type and a preset set of operations, the roll angle instruction is determined according to the correspondence between the predicted editing operation type and the set of operations. :

[0099] ;

[0100] in, As the reference roll angle command, Offset compensation for REPLACE operation type In hidden state, To map high-dimensional hidden features to linear weights of roll angle commands, To predict the type of editing operation, It represents a probability distribution.

[0101] In this embodiment, it is obtained by clustering historical data. At its core, define the executable instruction editing actions of the model at each time step:

[0102] This indicates that the baseline roll angle command remains unchanged (without correction). This indicates that a type k correction is added to the base instruction. This means reducing the type k correction amount based on the baseline instruction. This indicates that the baseline instruction is directly replaced (suitable for scenarios requiring significant adjustments).

[0103] More preferably, the predicted editing operation type is subject to domain knowledge constraints, and the probability distribution is renormalized:

[0104] .

[0105] in, tLet T be the time step and T be the total number of time steps. Let the operation distribution of the last 5 time steps of the final sequence be 0. Define an operation set containing various editing operations. By predicting the operation type of each time step, the reference roll angle command is precisely edited and adjusted to achieve optimized command generation and improve the flexibility and adaptability of flight control.

[0106] S3. Construct a kinematic model based on the roll angle command to obtain the basic trajectory, construct a wind disturbance model based on the wind disturbance sequence to obtain the disturbance compensation amount, and correct the synthesis result of the basic trajectory and the disturbance compensation amount through a residual correction network to obtain the lateral offset trajectory. .

[0107] The method for obtaining the lateral offset trajectory is as follows:

[0108] The kinematic model takes the roll angle command as input and calculates the basic lateral offset trajectory of the carrier under undisturbed conditions using preset kinematic equations. ;

[0109] ;

[0110] The wind disturbance model is based on the temporal characteristics of the wind disturbance sequence and outputs the disturbance compensation amount for the lateral displacement of the carrier by fitting the wind field action law. ;

[0111] ;

[0112] Define the input feature vector, which includes the predicted roll angle command for the current time step, the predicted lateral offset for the previous time step, the wind disturbance for the current time step, and the hidden state of the encoder output;

[0113] ;

[0114] The residual correction network obtains residual correction by learning the model error patterns in historical trajectory data. ;

[0115] ;

[0116] Based on the aforementioned base trajectory, disturbance compensation amount, and residual correction, the corrected lateral offset trajectory is obtained:

[0117] ;

[0118] in, Let airspeed be constant. For time step, Wind direction angle The cosine of the wind direction angle. The predicted roll angle instruction for the current time step is sinusoidalized. The predicted lateral offset for the previous time step. The predicted roll angle instruction for the current time step. The instantaneous wind speed and volume of the wind disturbance at the current time step. Output the hidden layer dimensions to the encoder. The input feature vector is the residual correction network, and MLP is a multilayer perceptron. In hidden state, It is the set of real numbers.

[0119] By combining physical kinematic equations and residual correction networks, and introducing wind disturbance modeling, trajectory prediction can reflect physical laws and adapt to environmental changes, thereby improving anti-interference capabilities and ensuring safe flight of aircraft under complex weather conditions.

[0120] Preferably, the loss function of the lateral offset trajectory for:

[0121] ;

[0122] ;

[0123] ;

[0124] ;

[0125] in, To avoid loss of editing time, For trajectory loss, For wind disturbance consistency regularization; Weighting for wind disturbance impact, These are the weighting coefficients. i= 1, 2, 3; For actual editing operations, The total number of time steps. For probability distribution, For time t The lateral offset trajectory, For time t The actual offset trajectory, For time t The ideal lateral offset trajectory under windless conditions.

[0126] More preferably, the weights of the loss function are dynamically weighted:

[0127] ;

[0128] in, These are the task uncertainty parameters, where i=1, 2, and 3 correspond to... , , The is used to balance the proportions of editing operation loss, trajectory loss, and wind disturbance consistency regularization term in the total loss function. The task uncertainty parameter reflects the model's uncertainty estimate for different tasks (such as editing operation prediction, trajectory prediction, and wind disturbance processing). It is used to dynamically adjust the weight of the corresponding loss in the total loss. When the model has a large uncertainty for a certain task, the corresponding weight will decrease, and vice versa.

[0129] By dynamically adjusting the loss weights of tasks such as editing operations, trajectory prediction, and wind disturbance handling, the problem of imbalanced losses among tasks in multi-task learning is solved, improving the overall performance of the model and achieving effective balance and synergistic optimization among different tasks. Simultaneously, by decoupling the physical kinematics and residual correction network and deeply integrating wind disturbance modeling, the "pure control error" is accurately extracted, allowing the model to focus on improving its own strategy rather than the interference of wind disturbances during optimization. This achieves a synergy between physical accuracy and disturbance resistance flexibility, making trajectory prediction more accurate and stable under complex weather conditions.

[0130] S4. Adjust the range of the safety envelope according to the lateral offset trajectory, and under the constraint of the safety envelope, determine the roll angle command correction amount based on the lateral offset trajectory, and correct the lateral motion state of the carrier according to the roll angle command correction amount.

[0131] Preferably, the range of the safety envelope is adjusted according to the lateral offset trajectory, and the method is as follows:

[0132] The sensitivity and safety requirements for lateral drift differ at different flight phases (such as takeoff, climb, cruise, descent, and landing). For example, during takeoff and landing, the aircraft has lower tolerance for lateral drift, therefore the safety control envelope is more stringent. Based on the initial heading drift trajectory, the drift trend and acceleration are determined, and potential hazards are identified using the predicted lateral drift trajectory.

[0133] A safety envelope is constructed based on the offset trend, acceleration, wind speed and corresponding flight stage, combined with a preset offset safety threshold.

[0134] The predicted heading deviation trajectory is obtained based on the real-time received data, and the safety envelope is readjusted based on the predicted heading deviation trajectory. The predicted lateral deviation trajectory already includes the influence of wind on the lateral motion of the aircraft. Therefore, the safety envelope can be adjusted according to environmental conditions, such as appropriately reducing the safety envelope under strong crosswind conditions.

[0135] More preferably, in this embodiment, the lateral offset safety threshold is a key parameter for evaluating the pilot's safe control envelope, and is determined based on runway width, flight mission type, etc. For example, if the runway width is... The safety threshold can then be set at half the runway width. ;

[0136] Based on the predicted lateral offset trajectory, identify the time points that exceed the lateral offset safety threshold. That is, satisfying These points, known as potential danger points, are used to quantify risks, trigger corrective mechanisms, and proactively improve safety indices.

[0137] More preferably, the velocity and acceleration of the lateral offset trajectory are calculated to analyze the offset trend, and the method is as follows:

[0138] ; ;

[0139] in, For offset rate, The lateral drift acceleration is the velocity of the aircraft. If the lateral drift velocity and acceleration are large, it may mean that the aircraft's lateral drift trend is intensifying, and the safety envelope needs to be considered more carefully.

[0140] In this embodiment, the safety envelope is represented graphically, with time or spatial location as the horizontal axis and lateral offset as the vertical axis, plotting the upper and lower limits of the safety envelope. The area within this envelope is the safe zone, and the area outside this zone is the danger zone.

[0141] Preferably, the method for determining the roll angle command correction based on the lateral offset trajectory under the safety envelope constraint is as follows:

[0142] The flight data also includes the carrier's current lateral offset trajectory, flight speed, offset safety threshold, and envelope boundary slope;

[0143] Based on the lateral offset trajectory, the roll angle command correction sequence is obtained by reverse calculation. And initialize the parameters;

[0144] Based on the genetic algorithm, the roll angle command correction sequence is encoded as an individual of the genetic algorithm. The population is iterated by combining the fitness function, and the roll angle command correction sequence corresponding to the best individual is selected as the candidate optimal solution.

[0145] Based on the Lagrange multiplier method, with the objective function of minimizing the pilot's operational workload and the safety envelope as the constraint, the optimal correction value among the candidate optimal solutions is used as the roll angle command correction value.

[0146] Preferably, the roll angle command correction sequence is obtained by reverse calculation based on the lateral offset trajectory. The method is as follows:

[0147] Establish the dynamic relationship between lateral offset and roll angle, define the offset error, calculate the correction amount using a PID controller, discretize the result to obtain the correction amount sequence, and then iteratively optimize the parameters.

[0148] More preferably, the correction sequence is dynamically expanded:

[0149] ;

[0150] in, The step size coefficient is used to divide the "maximum-minimum difference interval" into N parts, which is the maximum / minimum difference value in the instruction correction sequence. When the frequency of use exceeds the threshold, a new correction value is activated.

[0151] Constructing discrete adjustment values ​​through "range segmentation" prepares for subsequent selection / generation of specific reference roll angle commands—a crucial step in the command generation process of "defining candidate command values." This becomes a discrete set containing "original deviation + uniform segmentation step size". Values ​​can then be selected from this set to generate discrete baseline roll adjustment commands (such as adjusting command values ​​by a fixed step size, which meets the engineering requirement that "commands must be discrete and controllable").

[0152] More preferably, the parameter initialization includes setting initial parameters for the genetic algorithm based on the aircraft model and flight stage, such as the population size (e.g., 100-200 individuals), maximum number of iterations (e.g., 50-100 generations), crossover probability (0.6-0.9), and mutation probability (0.01-0.1). Simultaneously, the initial iteration precision (e.g., 1e-6) and maximum number of iterations (e.g., 1000 times) of the Lagrange multiplier method are determined.

[0153] More preferably, candidate optimal solutions are selected based on genetic algorithms, and the method is as follows:

[0154] 1) Individual Encoding and Population Initialization: The roll angle command correction sequence is encoded into individuals using a genetic algorithm, with each individual representing a possible correction strategy; during population initialization, multiple different correction sequences are randomly generated as initial individuals to ensure population diversity.

[0155] 2) Fitness function calculation: For each individual, simulate the change in lateral offset trajectory after the corresponding roll angle command correction is applied to the aircraft, calculate the deviation of the trajectory from the safety envelope, the pilot's operation amount (such as the sum of the absolute values ​​of the correction amount, operation frequency, etc.), and the flight state stability (such as the rate of change of lateral offset, the rate of change of roll angle command, etc.), and construct the fitness function by combining these indicators.

[0156] 3) Selection operation: Based on individual fitness, select individuals with higher fitness as parents, so that they have a higher probability of passing on their genes to the next generation;

[0157] 4) Crossover and mutation operations: Crossover operations are performed on the selected parent individuals to randomly exchange some gene segments to generate new individuals; at the same time, the genes of the individuals are mutated with a certain probability to randomly change the values ​​of some correction factors, thereby increasing the population diversity and exploration ability.

[0158] 5) Iterative update and termination condition judgment: Repeat selection, crossover and mutation operations to continuously update the population; stop iterating when the maximum number of iterations is reached or the fitness of individuals in the population converges (e.g., the change in the optimal fitness of the population is less than a certain threshold for several consecutive generations); output the roll angle instruction correction sequence corresponding to the best individual at this time as a candidate optimal solution.

[0159] Preferably, the genetic algorithm is optimized:

[0160] The comparison results between the current lateral offset trajectory and the safety envelope (whether it exceeds the limit, the duration of exceeding the limit / offset) are used as constraints and encoded into the fitness function of the genetic algorithm (the over-limit strategy is directly marked as an invalid individual and eliminated); the potential risk level of the subsequent short-term lateral offset trend (such as the predicted rapid expansion of offset under strong winds) is used as the optimization objective to drive the genetic algorithm to search for control strategies that "can suppress the spread of risk" (the higher the risk, the greater the fitness weight of the corresponding strategy).

[0161] Alternatively, in this embodiment, the fitness function is:

[0162] ;

[0163] in, This is a weighting coefficient, determined based on flight mission and safety requirements.

[0164] As an alternative, in this embodiment, methods such as roulette wheel selection and tournament selection are used to select individuals with high fitness as parents.

[0165] The effectiveness and robustness of the search are evaluated using the population evolution characteristics of the genetic algorithm.

[0166] Population diversity: Monitor the degree of difference between individuals (control strategies) during iteration (such as the range of roll angle correction and adjustment frequency) to avoid the algorithm converging to a local optimum too early (such as focusing only on "not going out of bounds" but causing drastic fluctuations in the control quantity).

[0167] Fitness convergence: Track the fitness changes of the best individual (combining "safety envelope fit" and "risk suppression effect") to determine whether the algorithm stably approaches the global optimum (if the fitness continues to improve and fluctuates little, it indicates that the search is effective).

[0168] More preferably, the optimal correction is solved based on the Lagrange multiplier method, the method of which is as follows:

[0169] 1) Construction of objective function and constraints: The objective function is to minimize the pilot's operational inputs (e.g., minimizing the sum of squares of roll angle command corrections).

[0170] ;

[0171] At the same time, the constrained lateral offset trajectory must be within the safety envelope, that is, for each time point... t ,satisfy ,in To predict lateral offset, The lateral offset change caused by the correction amount;

[0172] Introducing Lagrange multipliers Constructing the Lagrange function:

[0173] ;

[0174] 2) Solving for the optimal solution: Differentiate the Lagrange function to derive the system of optimality condition equations; solve the system of equations using numerical methods to obtain the optimal roll angle command correction sequence that satisfies the constraints, ensuring that the aircraft's lateral drift is precisely controlled within the safety envelope;

[0175] 3) Verification and Application of Correction Quantity: The obtained optimal correction quantity is applied to the current roll angle command to simulate and verify whether the aircraft's lateral deviation trajectory is stable within the safety envelope. If the requirements are met, the corrected roll angle command is sent to the flight control system for execution; if not, the constraints or optimization parameters can be adjusted appropriately, and the solution and verification can be repeated until a feasible solution is obtained.

[0176] Genetic algorithms provide global exploration capabilities, while the Lagrange multiplier method ensures constraint satisfaction and accuracy. Furthermore, it can dynamically adjust the correction amount through real-time feedback. By utilizing genetic algorithms and the Lagrange multiplier method to optimize the roll angle command correction amount under the safety envelope constraint, it ensures that the aircraft's lateral deviation trajectory is accurately controlled within the safe range. This overcomes the limitations of traditional single methods in dynamic constraint response, multi-constraint coordination, and adaptation to complex working conditions, achieving stable trajectory control within the safety envelope.

[0177] S5. After correction, perform safety boundary detection. If the detection result does not fall within the safety envelope, return to recalculate the correction amount and repeat the correction operation. If it meets the safety envelope requirements, continue monitoring.

[0178] Continuously monitor the aircraft's actual lateral deviation trajectory, flight attitude, and other state parameters to evaluate the control effect; if the actual trajectory still deviates from the expected one or new interference factors appear, promptly feed the new data back to the genetic algorithm and Lagrange multiplier method modules, restart the optimization process, and dynamically adjust the roll angle command correction amount to ensure that the aircraft remains stable within the safety envelope.

[0179] Preferably, the correction effect is evaluated based on performance evaluation metrics, which include:

[0180] Instruction generation efficiency:

[0181] ;

[0182] Track safety deviation:

[0183] ;

[0184] Wind disturbance suppression rate:

[0185] ;

[0186] in, The ideal trajectory is one without wind; SDI>1 indicates crossing the boundary.

[0187] Although the steps in the above embodiments are described in the above order, those skilled in the art will understand that in order to achieve the effect of this embodiment, different steps do not need to be executed in such an order. They can be executed simultaneously (in parallel) or in a reverse order. These simple variations are all within the protection scope of this invention.

[0188] A second embodiment of the present invention provides a flight trajectory control system based on operational state perception and command learning, the system comprising:

[0189] The data acquisition module is configured to acquire flight data in real time. The flight data includes the current lateral offset trajectory, flight speed, offset safety threshold, envelope boundary slope, wind disturbance sequence, operation sequence, and reference roll angle command sequence.

[0190] The preprocessing module is configured to construct a joint input tensor based on the operation sequence and the reference roll angle instruction sequence, and obtain the hidden state sequence after being processed by the encoder;

[0191] The prediction module is configured to input the hidden state sequence into a preset prediction model to obtain the operation probability distribution, and then determine the roll angle command based on the operation probability distribution.

[0192] The offset analysis module is configured to construct a kinematic model based on the roll angle command to obtain the basic trajectory, construct a wind disturbance model based on the wind disturbance sequence to obtain the disturbance compensation amount, and correct the synthesis result of the basic trajectory and the disturbance compensation amount through a residual correction network to obtain the lateral offset trajectory.

[0193] The safety envelope adjustment module is configured to analyze the offset trend based on the lateral offset trajectory and adjust the range of the safety envelope.

[0194] The correction module is configured to determine the roll angle command correction amount based on the lateral offset trajectory under the safety envelope constraint, and correct the lateral motion state of the carrier according to the roll angle command correction amount;

[0195] The feedback module is configured to perform safety boundary detection after correction. If the detection result does not fall within the safety envelope, it returns to recalculate the correction amount and repeats the correction operation. If it meets the safety envelope requirements, it continues to monitor.

[0196] It should be noted that the flight trajectory control system based on operation state perception and command learning provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be further decomposed or combined. For example, the modules in the above embodiments can be merged into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the various modules or steps and are not considered as an improper limitation of the present invention.

[0197] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related descriptions of the system described above can be found in the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0198] An electronic device according to a third embodiment of the present invention includes:

[0199] At least one processor; and

[0200] A memory communicatively connected to at least one of the processors; wherein,

[0201] The memory stores instructions that can be executed by the processor to implement the flight trajectory control method based on operation state perception and instruction learning described above.

[0202] A computer-readable storage medium according to a fourth embodiment of the present invention stores computer instructions, which are executed by a computer to implement the above-described flight trajectory control method based on operation state perception and instruction learning.

[0203] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related descriptions of the electronic device and computer-readable storage medium described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0204] Those skilled in the art will recognize that the modules and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. The programs corresponding to the software modules and method steps can be placed in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art. To clearly illustrate the interchangeability of electronic hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the invention.

[0205] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0206] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0207] The terms “first”, “second”, etc., are used to distinguish similar objects, not to describe or indicate a specific order or sequence.

[0208] The term "comprising" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent in such process, method, article, or apparatus / device.

[0209] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A flight trajectory control method based on operational state perception and command learning, characterized in that, The method includes the following steps: Acquire flight data, which includes wind disturbance sequences, operation sequences, and reference roll angle command sequences; A joint input tensor is constructed based on the operation sequence and the reference roll angle command sequence. After being processed by the encoder, the hidden state sequence is obtained. The hidden state sequence is input into the preset prediction model to obtain the operation probability distribution, and then the predicted roll angle command is determined. A kinematic model is constructed based on the roll angle command to obtain the basic trajectory, a wind disturbance model is constructed based on the wind disturbance sequence to obtain the disturbance compensation amount, and the synthesis result of the basic trajectory and the disturbance compensation amount is corrected by the residual correction network to obtain the lateral offset trajectory. The range of the safety envelope is adjusted according to the lateral offset trajectory, and under the constraint of the safety envelope, the roll angle command correction amount is determined based on the lateral offset trajectory. The lateral motion state of the carrier is corrected according to the roll angle command correction amount. After correction, a safety boundary check is performed. If the check result does not fall within the safety envelope, the correction amount is recalculated and the correction operation is repeated. If the safety envelope requirements are met, monitoring continues.

2. The flight trajectory control method based on operation state perception and command learning according to claim 1, characterized in that, The method for constructing a joint input tensor based on the operation sequence and the reference roll angle command sequence, and obtaining the hidden state sequence after encoder processing, is as follows: The operation sequence and the reference roll angle instruction sequence are processed using learnable embedding functions respectively, mapped to the same semantic space, and the two types of embedding vectors are concatenated in dimension to form a joint input tensor. The joint input tensor is used as the input sequence. After being processed by an encoder for extracting temporal features, a hidden state sequence is obtained. The hidden state sequence contains contextual information and temporal dependencies of the input sequence.

3. The flight trajectory control method based on operation state perception and command learning according to claim 1, characterized in that, The method for determining the predicted roll angle command is as follows: Based on the hidden state sequence, the operation probability distribution is obtained through a prediction model; The operation type with the highest probability under the operation probability distribution is taken as the predicted editing operation type for the current time step; Based on the predicted editing operation type and combined with a preset set of operations, the roll angle instruction is determined.

4. The flight trajectory control method based on operation state perception and command learning according to claim 3, characterized in that, Based on the predicted editing operation type and combined with a preset set of operations, the roll angle instruction is determined using the following method: Construct an operation set based on historical data: ； in, The first step is to obtain the correction amount from the clustering historical data. k Roll angle historical correction amount , , , These represent different operation types. This indicates that the reference roll angle command remains unchanged. This indicates that a second instruction is added based on the baseline instruction. k Class of corrections, This means reducing the type k correction amount based on the baseline instruction. This indicates that the reference instruction is directly replaced. k =1...K, where K is the number of clusters; Based on the correspondence between the measurement and editing operation types and the operation set, the roll angle instruction is determined. : ； ； in, As the reference roll angle command, Offset compensation for REPLACE operation type In hidden state, To map high-dimensional hidden features to linear weights of roll angle commands, To predict the type of editing operation, For probability distribution, To find the type of editing operation with the highest probability under the operation probability distribution.

5. The flight trajectory control method based on operation state perception and command learning according to claim 4, characterized in that, Domain knowledge constraints are applied to the predicted editing operation types: ； in, t Let T be the time step, and T be the total number of time steps. The last 5 time steps of the final sequence have an operation distribution of 0. For probability distribution, This indicates that the base instruction is directly replaced.

6. The flight trajectory control method based on operation state perception and command learning according to claim 1, characterized in that, The method for obtaining the lateral offset trajectory is as follows: The kinematic model takes the roll angle command as input and calculates the basic lateral offset trajectory of the carrier under undisturbed conditions using preset kinematic equations. ; ； The wind disturbance model is based on the temporal characteristics of the wind disturbance sequence and outputs the disturbance compensation amount for the lateral displacement of the carrier by fitting the wind field action law. ; ； The residual correction network obtains residual correction by learning the model error patterns in historical trajectory data. ; ； ； Based on the aforementioned base trajectory, disturbance compensation amount, and residual correction, the corrected lateral offset trajectory is obtained. : ； in, Let airspeed be constant. For time step, Wind direction angle The cosine of the wind direction angle. The predicted roll angle instruction for the current time step is sinusoidalized. The predicted lateral offset for the previous time step. The predicted roll angle instruction for the current time step. The instantaneous wind speed and volume of the wind disturbance at the current time step. Output the hidden layer dimensions to the encoder. The input feature vector is the residual correction network; MLP is a multilayer perceptron. These are learnable parameters in an MLP. In hidden state, It is the set of real numbers.

7. The flight trajectory control method based on operation state perception and command learning according to claim 6, characterized in that, The loss function of the lateral offset trajectory for: ； ； ； ； in, To avoid loss of editing time, For trajectory loss, For wind disturbance consistency regularization; Weighting for wind disturbance impact, The instantaneous wind speed and volume of the wind disturbance at the current time step. These are the weighting coefficients. i =1, 2, 3; For actual editing operations, t For time step, The total number of time steps. For probability distribution, For residual correction, For time t The lateral offset trajectory, For time t The actual offset trajectory, For time t The ideal lateral offset trajectory under windless conditions.

8. The flight trajectory control method based on operation state perception and command learning according to claim 6, characterized in that, The method for adjusting the range of the safety envelope based on the lateral offset trajectory is as follows: Based on the initial heading deviation trajectory, determine the deviation trend and acceleration; A safety envelope is constructed based on the offset trend, acceleration, wind speed and corresponding flight stage, combined with a preset offset safety threshold. The predicted heading deviation trajectory is obtained based on the real-time received data, and the safety envelope is readjusted based on the predicted heading deviation trajectory.

9. The flight trajectory control method based on operation state perception and command learning according to claim 8, characterized in that, The flight data also includes the carrier's current lateral offset trajectory, flight speed, offset safety threshold, and envelope boundary slope; under the safety envelope constraint, the roll angle command correction is determined based on the lateral offset trajectory, using the following method: Based on the lateral offset trajectory, the roll angle command correction sequence is obtained by reverse calculation; Based on the genetic algorithm, the roll angle command correction sequence is encoded as an individual of the genetic algorithm. The population is iterated by combining the fitness function, and the roll angle command correction sequence corresponding to the best individual is selected as the candidate optimal solution. Based on the Lagrange multiplier method, with the objective function of minimizing the pilot's operational workload and the safety envelope as the constraint, the optimal correction value among the candidate optimal solutions is used as the roll angle command correction value.

10. A flight trajectory control system based on operational state perception and command learning, characterized in that the system comprises: The data acquisition module is configured to acquire flight data in real time. The flight data includes the current lateral offset trajectory, flight speed, offset safety threshold, envelope boundary slope, wind disturbance sequence, operation sequence, and reference roll angle command sequence. The preprocessing module is configured to construct a joint input tensor based on the operation sequence and the reference roll angle instruction sequence, and obtain the hidden state sequence after being processed by the encoder; The prediction module is configured to input the hidden state sequence into a preset prediction model to obtain the operation probability distribution, and then determine the roll angle command based on the operation probability distribution. The offset analysis module is configured to construct a kinematic model based on the roll angle command to obtain the basic trajectory, construct a wind disturbance model based on the wind disturbance sequence to obtain the disturbance compensation amount, and correct the synthesis result of the basic trajectory and the disturbance compensation amount through a residual correction network to obtain the lateral offset trajectory. The safety envelope adjustment module is configured to analyze the offset trend based on the lateral offset trajectory and adjust the range of the safety envelope. The correction module is configured to determine the roll angle command correction amount based on the lateral offset trajectory under the safety envelope constraint, and correct the lateral motion state of the carrier according to the roll angle command correction amount; The feedback module is configured to perform safety boundary detection after correction. If the detection result does not fall within the safety envelope, it returns to recalculate the correction amount and repeats the correction operation. If it meets the safety envelope requirements, it continues to monitor.

Images