An intelligent flight operation generation method, device, equipment and medium

Abstract

This invention relates to a method, apparatus, device, and medium for generating intelligent flight operations. The method includes the following steps: S1. Calibrate script settings before flight begins; S2. After flight begins, collect real-time response data of the aircraft under pilot control and define its features; S3. Standardize the collected data; S4. Construct a model based on the standardized data; S5. Calculate the prediction difference using a loss function on the model; S6. Train and optimize model parameters based on the difference. This invention trains individual models based on actual pilot control data, adaptively fitting multi-dimensional parameters such as operation amount, operation speed, and response latency. Compared with the fixed mapping of the average standard, this significantly reduces individual understanding bias, causing the command-control bias to converge to a set threshold after calibration, thus improving the consistency and predictability of human-machine interaction.

Description

Technical Field

[0001] This invention belongs to the field of assisted flight decision-making technology, specifically relating to an intelligent flight operation generation method, device, equipment, and medium. Background Technology

[0002] Artificial intelligence-assisted flight decision-making systems are a significant trend in the future development of aircraft avionics systems. However, the development of such systems has long struggled to address the issue of direct communication between humans and intelligent systems. Humans are accustomed to communicating in natural language, which often uses non-quantitative and vague terms to describe the physical characteristics of things, such as: "slowly increasing the throttle means the speed is a bit low." Almost all artificial intelligence and automation systems require the quantitative reception and transmission of control commands, such as: "set the throttle to 90% within 10 seconds to accelerate to 275 knots." When humans communicate with intelligent systems, they tend to use natural language, necessitating a mechanism to map natural language into quantitative commands that machines can recognize. Taking throttle changes as an example, existing methods generally operate on two principles: 1. Manual pre-calibration is performed. For example, a 30% increase in throttle within 10 seconds is defined as a slow increase; a 30% increase within 5 seconds is defined as a uniform increase; and a 30% increase within 1 second is defined as a rapid increase. When natural language is received, keywords such as slow, uniform, and rapid are detected, and then a mapping is performed between the keywords and specific numerical values, thereby enabling the intelligent system to understand the natural language.

[0003] 2. Calibration using big data. Using historical flight data, professionals categorize different throttle changes into several levels, such as slow, constant speed, and fast, creating a dataset. Machine learning methods are then used for training to obtain the mapping relationship between specific throttle changes and their corresponding levels in natural language. Method 1 is the simplest implementation; the intelligent system can only convert natural language to commands using fixed preset values. However, for aircraft flight, the handling characteristics differ significantly under different flight states. A single preset level classification is clearly insufficient to meet the needs of complex flight states, and creating preset levels for each flight state is both labor-intensive and impractical. Method 2 uses machine learning to extract command mapping relationships for different flight states from a large amount of historical data. However, this method relies on professionals building a massive training dataset, which is also very labor-intensive. Both methods require leveraging professional experience to manually or automatically construct a standard mapping mechanism. This mechanism tends to use the average standard recognized by most people to map natural language and machine commands. However, each individual pilot has distinct operational characteristics. For example, a calm and collected pilot might perceive a 30% increase in throttle within 5 seconds as a rapid increase, while a more aggressive pilot might view it as a slow increase. If the standard mapping mechanism interprets this as a uniform increase, then the pilot's understanding differs from the standard mechanism, leading to poor communication between the pilot and the intelligent system and consequently, reduced system efficiency. Therefore, existing technologies suffer from the following drawbacks: manual pre-calibration methods can only set fixed mappings for limited flight conditions, failing to cover complex and varied real-world flight scenarios, resulting in limited effectiveness in converting natural language to commands under different operating conditions. Both manual pre-calibration and big data calibration methods heavily rely on the involvement of professionals and the accumulation of extensive historical data, leading to significant workload and high costs in data preparation and calibration. Furthermore, average standards do not consider the subjective differences in pilots' understanding of natural language commands, easily causing misunderstandings between pilots and the intelligent system, affecting communication efficiency and system performance. Traditional methods are difficult to continuously and quickly adjust parameters according to the actual flight process, lack dynamic adaptive capabilities, and cannot accurately match the pilot's current operating habits. Summary of the Invention

[0004] In order to overcome the problems existing in the prior art, the present invention provides an intelligent flight operation generation method, apparatus, device and medium to overcome the current defects.

[0005] A method for generating intelligent flight operations, the method comprising the steps of: S1. Before the flight begins, calibrate the script settings; S2. Collect and define the characteristics of the aircraft's response data under the pilot's control in real time after the start of flight; S3. Standardize the collected data; S4. Building models based on standardized data; S5. Calculate the prediction difference using a loss function on the model; S6. Train and optimize model parameters based on differences.

[0006] In addition to the aspects and any possible implementations described above, an implementation is further provided, wherein S1 specifically includes: S11. Design natural language instruction scripts for pilot flight operations; S12. Determine the action level of the pilot's flight operations; S13. Define the initial physical parameter range for pilot flight operations; S14. The pilot performs a flight mission by flying.

[0007] In addition to the aspects described above and any possible implementation, an implementation is further provided in which the physical parameters include at least one or a combination of several of the following: action level, operation quantity, operation amplitude, operation speed, and response delay.

[0008] In addition to the aspects and any possible implementations described above, a further implementation is provided in which the aircraft control response data includes pilot control data and aircraft response data, and the aircraft control response data is constructed as a feature vector of the pilot and each natural voice command response.

[0009] In addition to the aspects and any possible implementations described above, a further implementation is provided, wherein the model in S4 is: The model in S4 is: ,in, Let F be the predicted value, and F be the machine learning model. The standardized feature vector, These are the model parameters.

[0010] In addition to the aspects described above and any possible implementations, a further implementation is provided in which the expression for the loss function is: ,in, Let N be the loss function, and M be the total number of samples and the number of control parameters to be regressed, respectively. i and m are independent variables, taking positive integer values. This represents the true value of the m-th manipulation parameter in the i-th sample; Let m be the predicted value of the control parameter in the i-th sample. The weight of the m-th control parameter.

[0011] In addition to the aspects and any possible implementations described above, a further implementation is provided in which the training and optimization in S6 are achieved by iteratively updating the model parameters θ, and the update formula is: ,in: For the first t+ Model parameters in the first iteration For the first t Model parameters at the next iteration η For learning rate, This represents the gradient of the current loss function with respect to the parameters.

[0012] The present invention also provides an intelligent flight operation generation device, the device being used to implement the method, comprising: a setting module, used to set a calibration script before the start of flight; The data acquisition module is used to collect and define the characteristics of the aircraft's response data in real time after the start of flight. The standardization module is used to standardize the collected data. Modules for building models based on standardized data; The prediction module is used to calculate the prediction discrepancies by applying a loss function to the model. The optimization module is used to train and optimize model parameters based on differences.

[0013] The present invention also provides an electronic device, the electronic device comprising: Memory, which stores executable instructions; A processor that executes the executable instructions in the memory to implement the method.

[0014] The present invention also provides a computer storage medium storing a computer program, the computer program being executed by a processor to implement the method described.

[0015] Beneficial effects of the present invention The intelligent flight operation generation method of the present invention includes the following steps: S1. Calibrate script settings before flight begins; S2. After flight begins, collect real-time response data of the aircraft under pilot control and define features; S3. Standardize the collected data; S4. Construct a model based on the standardized data; S5. Calculate the prediction difference using a loss function on the model; S6. Train and optimize model parameters based on the difference. The beneficial effects of the present invention are as follows: 1) Based on the pilot's actual operation data, a model belonging to each individual is trained. The model adaptively fits multi-dimensional parameters such as operation amount, operation speed and response time delay. Compared with the fixed mapping of the average standard, it significantly reduces the individual understanding bias and makes the command-control bias converge to the set threshold after calibration, thereby improving the consistency and predictability of human-computer interaction.

[0016] 2) Introduce aircraft state characteristics (speed, attitude, rate of change of attitude angle, position, etc.) and automatically adjust the quantitative output with natural language during takeoff, cruise, approach or landing phases to ensure the consistency of command semantics and operation across flight phases and overcome the failure of fixed intervals under complex conditions.

[0017] 3) Initial individualized training is completed by collecting data in one go through standardized calibration scripts, without the need for large-scale manual annotation and expert tuning. It supports rapid individualization with a small number of samples and subsequent lightweight incremental updates, which significantly reduces the cost of data preparation and deployment.

[0018] 4) Supports online learning or periodic recalibration, continuously updates model parameters as pilots' experience and habits change, and combines convergence criteria and learning rate control to avoid oscillations and forgetting, maintaining stable and high-precision mapping performance over the long term. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a flowchart of the algorithm of the present invention. Detailed Implementation

[0020] To better understand the technical solution of this invention, the content of this invention includes, but is not limited to, the specific embodiments described below. Similar technologies and methods should be considered within the scope of protection of this invention. To make the technical problems to be solved, the technical solutions, and advantages of this invention clearer, a detailed description will be provided below in conjunction with the accompanying drawings and specific embodiments.

[0021] It should be understood that the embodiments described in this invention are merely some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention.

[0022] like Figure 1 As shown, the present invention provides an intelligent flight operation generation method, the method comprising the following steps: S1. Before the flight begins, calibrate the script settings; S2. Collect and define the characteristics of the aircraft's response data under the pilot's control in real time after the start of flight; S3. Standardize the collected data; S4. Building models based on standardized data; S5. Calculate the prediction difference using a loss function on the model; S6. Train and optimize model parameters based on differences.

[0023] Preferably, S1 specifically includes: S11. Design natural language instruction scripts for pilot flight operations; S12. Determine the action level of the pilot's flight operations; S13. Define the initial physical parameter range for pilot flight operations; S14. The pilot performs a flight mission by flying.

[0024] Specifically, the process of this invention is as follows: Addressing the problem that traditional flight assistance systems struggle to adapt to individual pilot operating styles, this invention proposes a personalized conversion method for natural language commands to quantitative flight operation data based on machine learning. By designing standardized natural language command scripts, pilots are guided to perform a series of operations in flight simulation or real flight environments. Real-time data is collected on the pilot's control of the flight simulator or aircraft (such as the amount, speed, and amplitude of throttle, stick, and control wheel inputs) and the flight simulator's or aircraft's response data to the control data (such as speed and attitude), constructing a multi-dimensional feature vector. Using a supervised learning model, a personalized mapping relationship is established between natural language commands and the pilot's actual operating parameters. By defining a loss function and employing an iterative algorithm to continuously adjust the model parameters, the deviation between the output control commands and the pilot's actual operations gradually converges, ultimately forming a command conversion model unique to each pilot. Unlike traditional methods that rely on average standards or extensive manual annotation, this invention can dynamically identify and adapt to each pilot's individual operating characteristics, achieving efficient and accurate mapping between natural language and quantitative commands, significantly improving the fluency and efficiency of human-computer interaction.

[0025] like Figure 1 and Figure 2 As shown, the specific process is as follows: Step 1. Calibration script settings This step first designs a natural language instruction script, then determines one or more of the action level, operation quantity, operation speed, or response delay, defines the initial physical parameter range, and finally the pilot executes the flight mission. The designed natural language instruction script calibrates the pilot's flight habits. Taking the local five-way flight script as an example, the local five-way is a standard flight term referring to the five-way route that the aircraft completes, clockwise or counterclockwise, within a rectangle in the local space (one for ascent and one for descent in takeoff and landing spaces, respectively), starting from takeoff and using the runway as a reference. Each operation in the script is described using natural language and the action level is clearly defined (e.g., slow, constant speed, fast). Pilot operating habits are stable control patterns and preferences extracted from actual pilot operation data based on multiple missions; these are the parameterized features being estimated. A calibration example is as follows: (1) "Rapid throttle" to accelerate the aircraft's taxiing speed; (2) After reaching takeoff speed, "pull the control stick at a constant speed" to lift the aircraft off the ground; (3) "Slowly adjust the control stick and control wheel" to maintain the aircraft's attitude; (4) "Adjust the steering wheel at a constant speed" to make the aircraft complete three turns; (5) "Adjust the throttle at a constant speed" to keep the aircraft airspeed; (6) "Quickly adjust the control stick, quickly adjust the control wheel, and quickly adjust the throttle" to make the aircraft glide along the runway centerline and land.

[0026] Other descriptions are shown in Table 1, including 15 natural language command descriptions such as rapid left turn, right turn and / or steering wheel adjustment, constant speed left turn, right turn and / or steering wheel adjustment, slow left turn, right turn and / or steering wheel adjustment. Each natural language command has a preset corresponding physical parameter range as the initial mapping relationship of the model. The parameters include: operation duration, action level, operation amount, operation speed, and response latency. The specific initial parameter range and natural language commands can be flexibly adjusted according to different flight mission requirements and can be expanded according to mission requirements. This is only for illustrating the method of the present invention.

[0027] Based on these natural language instruction descriptions, the action level, operation amount, operation speed, and / or response delay are determined, as shown in Table 1. "Rapid throttle" accelerates the aircraft's taxiing, and the corresponding action level is rapid, with an operation amount of 20%~40%, an operation speed of 20%~40%, and a response delay of 0.2s~0.7s. Table 1 describes the action level, operation amount, operation speed, and response delay corresponding to other natural language instruction descriptions.

[0028] Table 1: Initial Physical Parameter Ranges for Natural Language Commands and Their Action Levels

[0029] By setting the script and parameter range in Table 1 above, this invention achieves an initial quantitative mapping between natural language action commands and pilots' actual operations, providing a benchmark for subsequent data collection, label generation, and model training.

[0030] Define the initial physical parameter range: For each natural language instruction, define the initial action range for each action, defining the value range of the instruction in dimensions such as operation amount, operation speed, and response latency. The initial physical parameters defined here are empirical values, which will be used as the basis for subsequent model iterations. The more accurate the initial empirical values ​​are, the fewer iterations will be required.

[0031] Finally, the pilot carried out the flight mission.

[0032] Step 1 is the pre-flight preparation work, which does not require pilot participation and is a prerequisite for the execution of the method of this invention. The core is to clearly define "natural language → action level → initial physical range → phase process and state criteria" to ensure that the pilot is always in a clear flight phase and / or configuration state during subsequent execution.

[0033] 2. Data Acquisition and Feature Definition: This invention guides pilots to complete specific flight missions based on calibrated script settings and collects their control data in real time. The collected data will be used to construct a mapping model between each pilot's unique natural language commands and quantitative operating parameters.

[0034] During flight calibration, natural language commands are issued sequentially according to the script preset in step 1. The pilot executes the corresponding operations based on the commands, and the aircraft generates response data accordingly. Each time a natural language command is issued, a data acquisition is triggered, continuously collecting flight operation data over a period of time. The issued natural language command is correlated with the pilot's subsequent operations and the aircraft's response data to form a complete data sample. During this process, the pilot's control data (such as the amount, speed, amplitude, and frequency of control parameters for throttle, control stick, and control wheel) and the aircraft's response data (such as aircraft speed, attitude, rate of change of attitude angle, and position) are recorded in real time. The collected pilot control data is a numerical record of the phenomena after the pilot's actual operation, i.e., "what the pilot did," while the aircraft response data is the objective state change resulting from the pilot's actions, i.e., "what the aircraft became."

[0035] Pilot control data and aircraft response data have a two-way influence: on the one hand, pilot control operations (such as pushing the throttle or pulling the stick) directly lead to changes in the aircraft's state (such as increased speed or changed attitude); on the other hand, the aircraft's current state also influences the pilot's subsequent operational decisions (such as reducing throttle when the speed is too high). This two-way influence means that the same natural language command may correspond to different quantitative operations in different aircraft states. For example, if a pilot says "accelerate quickly" during takeoff and cruise, although both mean "quickly," the actual amount and speed of the pilot's operation will inevitably differ due to the different aircraft states (speed, altitude, etc.). Pilot control data and aircraft response data include: (1) q: Operational quantity: such as throttle percentage, steering wheel deflection angle, etc.; (2) v: Operation speed: The rate of change of the controlled quantity per unit time; (3) A: Operation range: The absolute range of change of the control quantity; (4) T: Response delay: The time interval between receiving an instruction and executing an operation; (5) f: Operation frequency: The number of times the control device is operated per unit time; (6) S: Aircraft state vector, including aircraft speed, attitude, rate of change of attitude angle, position, etc.

[0036] Based on the above data, a feature vector is constructed for the pilot's response to each natural language command. ,Right now ,in: For the first The feature vector corresponding to the instruction; For the first The operands corresponding to this instruction; For the first The operation speed corresponding to this instruction; For the first The operation range corresponding to this instruction; For the first The response delay corresponding to this instruction; For the first Multidimensional feature vectors of the aircraft state corresponding to the execution of each natural language instruction; For the first During the execution of this command, the pilot adjusts the frequency of operation of the control devices. Take a positive integer.

[0037] Step 3. Data Standardization The purpose of standardization is to ensure that all input features are on the same scale, thereby avoiding unfair impacts on model training caused by some features having larger or smaller value ranges. Therefore, all features are standardized (normalized). The specific steps are as follows: First, all feature vectors Together, they constitute the training set. For the j-th feature (including control parameters) of the i-th sample data in the training set... and aircraft state subvectors (all components in), the standardized formula is: (1) in: For the j-th feature of the i-th sample data in the training set, such as throttle percentage, lever angle, etc. Let be the mean of the j-th feature in the training set, and be a known value. For the first j The standard deviation of each feature in the training set is a known value; represents the standardized or normalized feature value.

[0038] All features obtained after standardization have dimensionless properties with a mean of 0 and a variance of 1, making the model's dependence on different features more balanced, which is beneficial for convergence and generalization.

[0039] 4. Model Building Based on the collected pilot operation data and commands, a supervised learning method is used to construct a mapping model between natural language commands and actual pilot operations. The input is assumed to be a standardized feature vector. That is, the feature vector of each sample i contains j standardized feature values.

[0040] The model prediction formula is: (2) Where: Z and These are the parameters of function F, which expresses the mapping relationship from natural language to machine language. It's a non-linear expression, representing a machine learning model between natural language and pilot control data. This expression represents: Z and The predicted value obtained after mapping through a nonlinear model or function F Z is the input feature vector, consisting of standardized components. composition; These are the model parameters, which are the internal adjustable variables of the machine learning model F, and are used to make predictions based on the established model.

[0041] 5. Definition of loss function This step defines and constructs the loss function to calculate the difference between the predictions made in step 4. During model training, the loss function is defined to measure the difference between the predictions and the actual results. This applies to all parameters defined in step 2. If all parameters are within the standard range corresponding to the instruction, then the operation is determined to be a complete match with the natural language instruction stated in the script. In this case, the loss function... Minimum; if any parameter exceeds the range, a bias is determined, and the loss function is applied. The value is set to >0, and individual differences are automatically assessed based on the bias results. The loss function is expressed in the form of weighted mean square error as follows: (3) Where: N is the total number of valid data records corresponding to the number of natural language instructions in the script calibrated in step 1. Assuming there are 10 natural language instructions in the script (rapid acceleration, constant speed throttle adjustment, etc.), and each natural language instruction is collected once, then N is 10. A complete execution of one natural language instruction is recorded as one sample. Each sample contains: control data (which may have multiple parameters, such as operation amount, operation speed, operation amplitude, etc.) and response data (which may also have multiple parameters, such as speed, attitude angle, etc.). For the number of manipulation parameters that need to be regressed, for example, These correspond to operation quantity, operation speed, and response latency; For the i-th sample, the m-th control parameter is the true value, i.e., the actual pilot operation data recorded during the data acquisition phase, such as... According to formula (2) described above, we can obtain: ,here For function Output the m-th component of the vector. Let m be the predicted value of the model for the m-th control parameter in the i-th sample, where i and m are independent variables and take positive integer values. The weight of the m-th parameter is used to adjust the importance of different manipulation parameters in the total loss. For example, the amount of manipulation can be considered more important than the frequency of manipulation. The sum of all weights is 1. As a loss function, the smaller its value, the more closely the control parameters predicted by the model match the actual operation or control data of the pilot under the command, that is, the more accurately the model has learned each pilot's personalized control habits.

[0042] 6. Model parameter training and optimization To address situations where bias exists, model parameters are iteratively updated during model training. θ, Its update formula is: (4) in: and These are the model parameters at the (t+1)th and tth iterations, respectively, both including weights and biases, where t is a positive integer greater than or equal to 1; The learning rate is used to control the magnitude of each parameter update. Let be the gradient of the current loss function with respect to the parameters; then calculate the error between the (t+1)th and tth predicted values. for: (5), where k is the iteration number, taking values ​​of 0, 1, 2, ... To avoid the model failing to converge, a maximum number of iterations is also set. and convergence threshold If convergence does not meet the conditions or If t = t + 1, then return to the model building section in step 4, regenerate the predicted values, and iterate through the subsequent steps until the prediction error between the two steps is less than the set convergence threshold. At that time, that is Then the optimal parameters are obtained through optimization. Or reach the maximum number of iterations. At that time, that is Then the optimal parameters are obtained through optimization. , that is, θ: (6) Output it as the optimal parameter, and use this optimal parameter θ in the corresponding function. The calculation is updated to obtain or generate a personalized mapping model, that is, to generate a mapping model between natural language and the pilot's actual operation.

[0043] The training data obtained after the above steps is most suitable for the current pilot is updated in Table 1, forming a natural language command operation description unique to the current pilot. When the pilot speaks ambiguous natural language again, the model outputs a precise quantitative command that best matches the pilot's personal style based on its understanding of the individual pilot.

[0044] As an embodiment of the present invention, the present invention also discloses an intelligent flight operation generation device, the device being used to implement the method, comprising: The configuration module is used to set the calibration script before the flight begins; The data acquisition module is used to collect and define the characteristics of the aircraft's response data in real time after the start of flight. The standardization module is used to standardize the collected data. Modules for building models based on standardized data; The prediction module is used to apply a loss function to the model to predict discrepancies. The optimization module is used to train and optimize model parameters based on differences.

[0045] As an embodiment of the present invention, the present invention also discloses an electronic device, the electronic device comprising: Memory, which stores executable instructions; A processor that executes the executable instructions in the memory to implement the method of the present invention.

[0046] As an embodiment of the present invention, the present invention also discloses a computer storage medium storing a computer program, which is executed by a processor to implement the method described in the present invention.

[0047] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0048] The foregoing description illustrates and describes several preferred embodiments of the present invention. However, as previously stated, it should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the inventive concept described herein through the foregoing teachings or techniques or knowledge in related fields. Any modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A method for generating intelligent flight operations, characterized in that, The method includes the following steps: S1. Before the flight begins, calibrate the script settings; S2. Collect and define the characteristics of the aircraft's response data under the pilot's control in real time after the start of flight; S3. Standardize the collected data; S4. Building models based on standardized data; S5. Calculate the prediction difference using a loss function on the model; S6. Train and optimize model parameters based on differences.

2. The method of claim 1, wherein, S1 specifically includes: S11. Design natural language instruction scripts for pilot flight operations; S12. Determine the action level of the pilot's flight operations; S13. Define the initial physical parameter range for pilot flight operations; S14. The pilot performs a flight mission by flying.

3. The method of claim 2, wherein, The physical parameters include at least one or a combination of several of the following: action level, operation quantity, operation range, operation speed, and response delay.

4. The method of claim 1, wherein, The aircraft control response data includes the pilot's control data and the aircraft's response data, and the aircraft control response data is constructed into a feature vector of the pilot's response to each natural voice command.

5. The method of claim 3, wherein, The model in S4 is: where, is the predicted value, F is the machine learning model, is the normalized feature vector, is the model parameter.

6. The method according to claim 1, characterized in that, The expression for the loss function is: ,in, Let N be the loss function, and M be the total number of samples and the number of control parameters to be regressed, respectively. i and m are independent variables. This represents the true value of the m-th manipulation parameter in the i-th sample; Let m be the predicted value of the control parameter in the i-th sample. The weight of the m-th control parameter.

7. The method of claim 5, wherein, The training and optimization in S6 are achieved by iteratively updating the model parameters θ, and the update formula is as follows: , in: These are the model parameters at the (t+1)th iteration. These are the model parameters at the t-th iteration. η For learning rate, The gradient of the current loss function with respect to the parameters is t, where t is a positive integer.

8. An intelligent flight operation generation apparatus characterized by comprising: The apparatus is used to implement the method according to any one of claims 1-7, comprising: The configuration module is used to set the calibration script before the flight begins; The data acquisition module is used to collect and define the characteristics of the aircraft's response data in real time after the start of flight. The standardization module is used to standardize the collected data. Modules for building models based on standardized data; The prediction module is used to calculate the prediction discrepancies by applying a loss function to the model. The optimization module is used to train and optimize model parameters based on differences.

9. An electronic device, comprising: The electronic device includes: Memory, which stores executable instructions; A processor that executes the executable instructions in the memory to implement the method of any one of claims 1-7.

10. A computer storage medium, characterized in that, The medium stores a computer program, which is executed by a processor to implement the method according to any one of claims 1-7.

Images