Aircraft control method and system for collision avoidance at long distances

Abstract

The application relates to the technical field of aircraft route planning, in particular to an aircraft control method and system for preventing long-distance flight collision, which comprises the following steps: acquiring the offset distance of an aircraft from a set route at each moment during flight, and the measured distance between the aircraft and the nearest obstacle; obtaining the dynamic stability of the aircraft in each time interval; determining the long-term balance degree of the aircraft in each time interval based on the autocorrelation of the flight height of the aircraft at all moments in each time interval, the linear growth trend of the distance between the aircraft and the departure point in each time interval, and the fluctuation of the extreme value distribution of the flight speed of the aircraft at all moments in each time interval; correcting the Q value updating method in the deep Q network; and planning the route of the aircraft by using the corrected deep Q network. Thus, the long-distance flight collision risk of the aircraft is reduced.

Description

Technical Field

[0001] This application relates to the field of aircraft flight path planning technology, specifically to an aircraft control method and system for preventing long-distance flight collisions. Background Technology

[0002] Long-range collision avoidance control is a method that ensures the safe operation of aircraft in complex flight environments or high-density airspace through a real-time perception-decision-execution closed loop. Its core lies in dynamically constructing a three-dimensional safe zone model of the aircraft and combining it with trajectory prediction algorithms to anticipate potential conflicts. With the rapid development of the air traffic sector, long-range collision avoidance has gradually become an important research area in flight safety.

[0003] Unmanned Aerial Vehicle (UAV) route planning technology based on the DQN (Deep Q-Network) algorithm combines the advantages of deep learning and reinforcement learning, effectively improving the autonomous navigation capability of UAVs in complex environments and reducing the risk of obstacle collisions. In the DQN algorithm, the reward function determines the immediate reward obtained by the UAV after taking a certain action. However, in long-distance complex flight missions, the Q-value update in the traditional reward function ignores the sparse reward phenomenon caused by environmental dynamics, resulting in a lack of timely feedback for intermediate actions and slow Q-value updates. Furthermore, introducing environmental dynamics into the Q-value update process can easily disrupt the balance between immediate rewards and long-term objective tasks, potentially leading to short-sighted decision-making by the UAV. This can result in getting trapped in local optima during long-distance UAV missions, severely impacting the efficiency and success rate of UAV flight missions. Summary of the Invention

[0004] To address the aforementioned technical problems, the purpose of this application is to provide an aircraft control method and system for preventing long-distance flight collisions. The specific technical solution adopted is as follows:

[0005] In a first aspect, embodiments of this application provide an aircraft control method for preventing long-distance flight collisions, the method comprising the following steps:

[0006] Obtain the offset distance of the aircraft from the set flight path at each moment during flight, as well as the measured distance to the nearest obstacle;

[0007] Analyze the dispersion of wind speed distribution faced by the aircraft during flight in each time interval, as well as the correlation of wind speed faced between different time periods within the time interval, to determine the degree of flight maneuver adjustment of the aircraft in each time interval.

[0008] The coupling status between the offset distance and the measured distance at each time point is determined. By analyzing the dispersion of the offset distance during the flight of the aircraft in each time interval and the changing trend of the coupling status, the anomaly of the aircraft's state offset in each time interval is determined.

[0009] By combining the flight maneuver adjustment degree and the state deviation anomaly degree, the dynamic stability of the aircraft in each time interval is obtained;

[0010] Based on the autocorrelation of the flight altitude of the aircraft at all times within each time interval, and the linear growth trend of the distance between the aircraft and the starting point within each time interval, combined with the fluctuation of the extreme value distribution of the flight speed of the aircraft at all times within each time interval, the long-term balance of the aircraft in each time interval is determined.

[0011] Combining the dynamic stability and long-term equilibrium, the Q-value update method in the deep Q-network is modified, and the modified deep Q-network is used to plan the flight path of the aircraft.

[0012] In one embodiment, determining the flight maneuver adjustment degree includes:

[0013] The degree of dispersion of the wind speed faced by the aircraft at all times within each time interval is denoted as the first degree of dispersion. Each time interval is divided into time periods, and the cumulative sum of the correlations between any two time periods within each time interval and the corresponding wind speeds faced by the aircraft is calculated.

[0014] The flight maneuver adjustment degree is positively correlated with the first dispersion and negatively correlated with the summation.

[0015] In one embodiment, determining the coupling status between the offset distance and the measured distance at each time step includes: calculating the ratio of the offset distance to the measured distance at each time step as the coupling status between the offset distance and the measured distance at each time step.

[0016] In one embodiment, determining the anomaly of the aircraft's state deviation in each time interval includes:

[0017] Obtain the fitting curve of the coupling status at all times within each time interval, calculate the slope of the coupling status at each time on the fitting curve, and calculate the Hurst exponent of the slope at all times within each time interval.

[0018] The degree of dispersion of the offset distance at all times within each time interval is denoted as the second degree of dispersion;

[0019] The anomaly of the aircraft's state deviation in each time interval is the product of the Hurst exponent and the second discreteness.

[0020] In one embodiment, the determination of dynamic stability includes:

[0021] The anomaly of the aircraft's state deviation in each time interval is calculated and summed with a preset value greater than 0. The dynamic stability is the normalized value of the ratio of the aircraft's flight maneuver adjustment degree in each time interval to the summed result.

[0022] In one embodiment, determining the long-term balance of the aircraft across different time intervals includes:

[0023] The flight altitude of the aircraft at all times within each time interval is compiled into a flight altitude sequence, and the mean of all autocorrelation coefficients of the flight altitude sequence is calculated.

[0024] Linear fitting is performed on the distance between the aircraft and the starting point at all times within each time interval. Based on the fitting error of the distance between the aircraft and the starting point at all times within each time interval on the fitted line, the nonlinearity measure of the aircraft in each time interval is determined.

[0025] Calculate the degree of dispersion of the extreme values ​​of the flight speed of the aircraft at all times within each time interval, and denote it as the third degree of dispersion;

[0026] By combining the mean, the nonlinear metric, and the third dispersion, the long-term balance of the aircraft in each time interval is determined.

[0027] In one embodiment, the nonlinear metric is the sum of the fitting errors of the distance between the aircraft and the starting point at all times within each time interval on the fitted straight line.

[0028] In one embodiment, determining the long-term equilibrium includes:

[0029] Calculate the product of the nonlinear metric and the third dispersion, sum the product with a preset value greater than 0, and calculate the normalized value of the ratio of the mean to the summation result as the long-term balance.

[0030] In one embodiment, the Q-value update method in the modified deep Q-network includes:

[0031] The product of the dynamic stability and long-term equilibrium of the adjacent previous time interval at each time point is calculated as the reward adjustment factor at each time point.

[0032] The Q-value update method in deep Q-networks is as follows:

[0033] In the formula, Let the spacecraft be in state at time t. Take action Q-value update value, For learning rate, Let the spacecraft be in state at time t. Take action The reward received later; Let be the reward adjustment factor for the aircraft at time t. As a discount factor, Let t be the next state of the spacecraft. The maximum Q value of all possible actions, Let the spacecraft be in state at time t. Take action The Q value.

[0034] Secondly, embodiments of this application also provide an aircraft control system for preventing long-distance flight collisions, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of any of the methods described above.

[0035] This application has at least the following beneficial effects:

[0036] This application addresses the shortcomings of traditional DQN-based unmanned aerial vehicle (UAV) route planning, such as the neglect of sparse rewards caused by dynamic environmental changes in the Q-value update of the reward function, and the easy disruption of the balance between immediate rewards and long-term objectives. It provides a specific quantification method for dynamic stability and long-term balance, which can more accurately reflect the degree of immediate reward improvement during long-distance flight missions through the stability characteristics of UAV flight status data and the anomalies in flight status data imbalance. This ensures that the UAV can better adapt to complex weather conditions, effectively improves flight stability in unstable wind environments, reduces the interference of wind speed changes on the flight path, and effectively avoids the risk of collisions due to deviation from the flight path, thus contributing to improved flight path tracking accuracy. By obtaining reward adjustment factors through dynamic stability and long-term balance, and using these factors to update the Q-values ​​of the SA table in the DQN deep Q-network, the drawbacks of sparse rewards for UAVs in the DQN deep network can be effectively avoided. While introducing immediate rewards, it prevents the DQN deep Q-network model from making short-term decisions and getting stuck in local optima. It balances immediate rewards and long-term goal tasks, incentivizes positive reinforcement of UAVs, and enables real-time adjustment of flight paths according to environmental changes. This helps improve the autonomous navigation capability of UAVs, reduces the risk of collisions during long-distance flights, and ensures the efficiency and success rate of long-distance UAV flight missions. Attached Figure Description

[0037] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 A flowchart illustrating the steps of an aircraft control method for preventing long-distance flight collisions, provided as an embodiment of this application;

[0039] Figure 2 Flowchart for determining the reward adjustment factor. Detailed Implementation

[0040] To further illustrate the technical means and effects adopted by this application to achieve the intended purpose of the invention, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an aircraft control method and system for preventing long-distance flight collisions proposed in this application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0041] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0042] The following description, in conjunction with the accompanying drawings, details a specific scheme for an aircraft control method and system for preventing long-distance flight collisions provided in this application.

[0043] Please see Figure 1 The diagram illustrates a flowchart of an aircraft control method for preventing long-distance flight collisions according to an embodiment of this application. The method includes the following steps:

[0044] S1 collects various flight status data at different times during the flight of the unmanned aerial vehicle and performs preprocessing.

[0045] This embodiment deploys GPS sensors, an IMU (Inertial Measurement Unit), and millimeter-wave radar on the unmanned aerial vehicle (UAV) to acquire various flight status data of the UAV during long-distance flight missions. Specifically, it collects the flight distance between the UAV and the starting point at each moment during flight, i.e., the straight-line distance between the two points; the flight speed of the UAV at each moment during flight; the flight altitude of the UAV at each moment during flight; the wind speed encountered by the UAV at each moment during flight; the offset distance between the UAV and the set flight path at each moment during flight; and the measured distance between the UAV and the nearest obstacle at each moment during flight.

[0046] The specific method for obtaining the offset distance between the unmanned aerial vehicle (UAV) and the set route is as follows: compare the real-time GPS position of the UAV with the preset route, calculate the vertical distance between the real-time GPS position and the set route, and use it as the offset distance between the UAV and the set route.

[0047] The specific method for obtaining the distance between the UAV and the nearest obstacle is as follows: the UAV uses millimeter-wave radar to transmit and receive millimeter-wave signals to identify obstacles during the flight mission, and calculates the distance between the UAV and each obstacle based on the reception duration of the millimeter-wave signals. The distance between the UAV and the nearest obstacle at each moment is recorded as the measured distance between the UAV and the nearest obstacle at that moment. Since the calculation of obstacle distance using millimeter-wave signals is a well-known technology, the specific acquisition process will not be elaborated further.

[0048] In this embodiment, all flight status data are collected synchronously at a frequency of 1Hz. The implementer can set the frequency according to the actual situation, and this embodiment does not impose any restrictions on this.

[0049] To prevent data loss during transmission of unmanned aerial vehicle (UAV) flight status data due to network fluctuations or external interference, and to avoid the impact of different data units on subsequent analysis, all flight status data at each moment during long-distance UAV flight missions were imputed using cubic spline interpolation (CFI) and standardized using Z-Score to unify the units of measurement. CFI is a well-known existing technique, and its specific process will not be elaborated upon.

[0050] S2, analyze the dispersion of wind speed distribution faced by the aircraft during flight in each time interval, as well as the correlation of wind speed faced between different time periods within the time interval, and determine the degree of adjustment of the aircraft's flight maneuvers in each time interval.

[0051] In the process of UAV route planning based on the DQN (Deep Q-Network) algorithm, the Q-value update using the traditional reward function only yields a reward when the UAV reaches the target point or under specific circumstances, resulting in sparse rewards. In long-distance, complex flight missions, sparse rewards mean that intermediate actions of the UAV mission cannot receive timely reward feedback. This makes it difficult for the deep Q-network to evaluate the UAV route planning and optimize intermediate actions through limited feedback during the learning process, thus reducing the UAV mission execution efficiency.

[0052] Specifically, in the unmanned aerial vehicle (UAV) route planning process based on the DQN deep Q-network algorithm, when the UAV is affected by dynamic environmental changes during long-distance flight missions, but should receive action rewards for good performance, the UAV will try to minimize the impact of wind speed on flight when choosing a flight path, thereby reducing the energy consumption of the UAV. The UAV shows high stability in wind speed data and significant short-term correlation with the timeliness of attitude adjustment.

[0053] Based on the above analysis, this embodiment sets each 5-minute interval as a flight interval, and divides each flight interval into N time periods, which are called short time intervals. In this embodiment, N=10. The value of N and the length of the flight interval can be set by the implementer according to the actual situation. This embodiment does not impose any restrictions on this.

[0054] Calculate the dispersion of the wind speed faced by the aircraft at all times within each flight interval, and denote it as the first dispersion. Then, arrange all the wind speeds faced by the aircraft in each short time interval within each flight interval into a short-time wind speed sequence according to the time sequence, and calculate the correlation between any two short-time wind speed sequences within each flight interval.

[0055] It should be noted that the degree of dispersion can be calculated using methods such as variance, standard deviation, and coefficient of variation, while the correlation can be calculated using methods such as Pearson correlation coefficient and cosine similarity.

[0056] In this embodiment, the coefficient of variation of the wind speed faced by the aircraft at all times within each flight interval is calculated and denoted as the first dispersion. The Pearson correlation coefficient between any two short-time wind speed sequences within each flight interval is calculated. The sum of all the Pearson correlation coefficients within each flight interval is calculated. The flight maneuver adjustment degree of the aircraft in each time interval is positively correlated with the first dispersion and negatively correlated with the sum.

[0057] Specifically, the normalized value of the accumulated sum is obtained using the Sigmoid normalization function, and the ratio of the first dispersion to the normalized value is used as the flight maneuver adjustment degree of the aircraft in each time interval.

[0058] S3, determine the coupling status between the offset distance and the measured distance at each time point, and determine the state offset anomaly of the aircraft in each time interval by the degree of dispersion of the offset distance during the flight of the aircraft in each time interval and the changing trend of the coupling status.

[0059] In addition, during flight, unmanned aerial vehicles should generally choose good paths to avoid frequently performing obstacle avoidance operations, so that the deviation distance of the unmanned aerial vehicle from the predetermined route is less volatile and the coupling change between the unmanned aerial vehicle and the nearest obstacle is less drastic.

[0060] The offset distance between the UAV and the set flight path is obtained at each moment within the flight interval, and the measured distance between the UAV and the nearest obstacle is also obtained at each moment. The ratio of the offset distance to the measured distance at each moment is calculated as the coupling status between the offset distance and the measured distance at each moment. The more severe the offset-obstacle coupling status, the higher the offset distance between the UAV and the set flight path or the smaller the distance to the nearest obstacle, indicating a poorer flight path selection and a higher risk of flight collision. The coupling status at all moments within the flight interval is arranged in chronological order, and the least squares method is used to obtain the fitting curve of the coupling status at all moments within the flight interval. The slope of the coupling status at each moment at the corresponding position on the fitting curve is calculated to reflect the changing trend of the coupling status.

[0061] Furthermore, the Hurst exponent of the slope at all times within each time interval is calculated, and the variance of the offset distance at all times within each time interval is calculated and denoted as the second dispersion. The product of the Hurst exponent and the second dispersion is taken as the state offset anomaly degree of the aircraft in each time interval.

[0062] S4. Combining the flight maneuver adjustment degree and the state deviation anomaly degree, the dynamic stability of the aircraft in each time interval is obtained.

[0063] This embodiment constructs the dynamic stability of the aircraft in various time intervals to characterize the timeliness of the unmanned aerial vehicle's action adjustments and path adjustment status under the dynamic changes of the long-distance flight mission environment. The specific expression is as follows:

[0064] In the formula, Let represent the dynamic stability of the unmanned aerial vehicle during the i-th time interval in a long-distance flight mission. Let be the flight maneuver adjustment degree of the unmanned aerial vehicle during the i-th time interval in a long-distance flight mission. Let be the state deviation anomaly degree of the unmanned aerial vehicle in the i-th time interval during a long-distance flight mission. 'a' is a preset value greater than 0, used to avoid a denominator of 0. In this embodiment, a = 0.01. Implementers can set this value according to actual conditions; this embodiment does not impose any restrictions. 'norm()' is a normalization function that makes ... The value range is within the range of [0,1].

[0065] It should be understood that dynamic stability reflects the timeliness of flight maneuver adjustments and path selection deviations of an unmanned aerial vehicle (UAV) under the influence of dynamic environmental changes within each flight interval during a long-distance flight mission. Flight maneuver adjustment reflects the stability of the UAV in the face of wind speed within the flight interval, and is significantly correlated with the short-term wind speed caused by timely flight path adjustments. State deviation anomaly characterizes the frequency of UAV flight operations that avoid obstacles and deviate from the set flight path, and the severity of the coupling changes between the deviation distance and the obstacle distance. In long-distance UAV flight missions, the more effective the UAV's maneuver adjustments in the face of dynamic environmental changes, the more timely the UAV's path adjustments to reduce the impact of wind speed, the higher the stability of the UAV's wind speed data, and the more obvious the short-term correlation of the wind speed data. These are the key performance indicators. The larger the value, the less frequent the path selection operations of the unmanned aerial vehicle to avoid obstacles in long-distance dynamic environmental changes, the more stable the deviation from the set route distance, and the less the change in the coupling state of obstacle deviation, i.e., the more accurate the calculated indicators. The reward for updating the Q-value in the DQN deep Q-network algorithm should be increased at this point to avoid the drawback of sparse rewards for the UAV during dynamic environmental changes.

[0066] S5. Based on the autocorrelation of the flight altitude of the aircraft at all times within each time interval, and the linear growth trend of the distance between the aircraft and the starting point within each time interval, combined with the fluctuation of the extreme value distribution of the flight speed of the aircraft at all times within each time interval, the long-term balance of the aircraft in each time interval is determined.

[0067] In long-distance flight missions of unmanned aerial vehicles (UAVs), relying solely on dynamic stability as the basis for Q-value updates in the DQN (Deep Q-Network) can have certain drawbacks. Specifically, improving the reward situation through the UAV's excellent performance in dynamic environments ignores the balance between short-term rewards and long-term objective tasks. Over-reliance on the limited performance of the UAV's intermediate actions can easily lead to short-term decision-making in Q-value updates, causing the UAV's route planning to fall into local optima and affecting the efficiency and success rate of the UAV in the complete flight mission.

[0068] Specifically, in the unmanned aerial vehicle (UAV) route planning process based on the DQN (Deep Q-Network) algorithm, the stronger the balance between the obstacle avoidance or path optimization actions performed by the UAV under the influence of environmental dynamics during long-distance flight missions and the long-term target mission, the milder the short-term detours and excessive obstacle avoidance of the UAV. That is, the flight distance data shows a strong linear enhancement phenomenon, and the frequency of aircraft speed fluctuations is more concentrated rather than the continuous change of aircraft speed caused by frequent adjustment of obstacle avoidance actions. At the same time, the stronger the balance between the short-term obstacle avoidance behavior of the UAV and the long-term target mission, the higher the long-term correlation of the UAV's flight altitude.

[0069] Based on the above analysis, this application constructs a long-term equilibrium degree for the aircraft in each time interval to characterize the flight state imbalance and long-term correlation caused by the unmanned aerial vehicle (UAV) performing obstacle avoidance or path optimization maneuvers during long-distance flight missions. Specifically, the flight distances of the UAV at all times within each flight interval are arranged in chronological order to form a flight distance data sequence. The least squares method is used to obtain the fitted straight line of the flight distance data sequence, and the fitting error of each element in the flight distance data sequence relative to the fitted straight line is determined. That is, the absolute value of the difference between each element in the flight distance data sequence and its corresponding fitted value on the fitted straight line is calculated. The sum of the absolute values ​​of the differences of all elements in the flight distance data sequence is calculated as a nonlinearity measure of the aircraft in each time interval. When the value of the nonlinearity measure is higher, it indicates that the linear growth trend of the UAV's flight distance is more ambiguous, and the flight stability of the UAV is lower.

[0070] The flight velocities of the unmanned aerial vehicle (UAV) within each flight interval are arranged in chronological order to form a flight velocity data sequence. The AMPD (Automatic Multiscale-based Peak Detection) algorithm is used to obtain the extreme values ​​(maximum and minimum) in the flight velocity data sequence. The flight altitudes of the UAV at all times within each time interval are arranged in chronological order to form a flight altitude sequence, and the mean of all autocorrelation coefficients of the flight altitude sequence is calculated. The long-term equilibrium of the UAV in each time interval is determined by the following expression:

[0071] In the formula, Let represent the long-term equilibrium of the unmanned aerial vehicle during the i-th time interval in a long-distance flight mission. Let be the mean of all autocorrelation coefficients of the flight altitude sequence of the unmanned aerial vehicle in the i-th time interval during a long-distance flight mission; calculate the standard deviation of the extreme values ​​of the flight velocity of the aircraft at all times within each time interval, denoted as the third dispersion. The product of the third discreteness and the nonlinear metric for the i-th time interval during a long-distance flight mission of the unmanned aerial vehicle. To ensure that the value is greater than 0 and to avoid a denominator of 0, this embodiment... The implementer can set it themselves; norm() is the normalization function, making... The value range is within the range of [0,1].

[0072] It should be understood that the long-term balance reflects the flight state imbalance caused by the unmanned aerial vehicle (UAV) performing obstacle avoidance or path optimization maneuvers during long-distance flight missions, as well as the long-term correlation of flight altitude; the mean reflects the long-term autocorrelation strength of flight altitude caused by the balance between short-term obstacle avoidance behavior and long-term target mission. This characterizes the degree of ambiguity in the linear growth trend of flight distance and the temporal discreteness of flight speed fluctuations when an unmanned aerial vehicle (UAV) performs obstacle avoidance or path optimization maneuvers. In long-distance UAV missions, the stronger the balance between the UAV's obstacle avoidance or path optimization behavior influenced by environmental dynamics and its long-term flight objective, the less pronounced the short-term detours and obstacle avoidance, the more obvious the long-term linear increase trend in flight distance, and the more concentrated the frequency of flight speed fluctuations. This is reflected in the calculated indicators. The smaller the value; at the same time, the higher the long-term correlation strength of the unmanned aerial vehicle's flight altitude, i.e., the higher the calculated index. It gets bigger.

[0073] S6. Combining the dynamic stability and long-term balance, the Q-value update method in the deep Q-network is modified, and the modified deep Q-network is used to plan the flight path of the aircraft.

[0074] In the process of using the DQN-based deep Q-network algorithm for unmanned aerial vehicle (UAV) route planning, the more timely and effective the UAV's action adjustments and path adjustments are under the dynamic changes of the long-distance flight mission environment, and the stronger the balance between the obstacle avoidance or path optimization actions performed by the UAV and the long-term flight mission objectives, the higher the reward coefficient of the UAV should be to incentivize and strengthen the positive incentive of the UAV.

[0075] Therefore, this embodiment constructs a reward adjustment factor to characterize the degree of reward coefficient improvement of the UAV within any flight interval in a long-distance flight mission. This factor can be obtained through the dynamic stability and long-term balance of the flight interval. Specifically, the product of the dynamic stability and long-term balance of the flight interval is used as the reward adjustment factor. A larger reward adjustment factor indicates that the UAV's actions within the flight interval are timely, its path adjustment is good, and the balance between obstacle avoidance or path optimization actions and the long-term flight mission objective is stronger. Therefore, the Q-value in the DQN deep Q-network algorithm should be updated based on this factor to incentivize positive reinforcement of the UAV and avoid collision risks and flight instability drawbacks during long-distance flights.

[0076] This embodiment calculates the product of dynamic stability and long-term balance in the preceding 5 minutes of each time point, i.e., the time interval adjacent to the previous time point, as the reward adjustment factor for each time point. The flowchart for determining the reward adjustment factor is as follows: Figure 2 As shown.

[0077] Initialize the DQN deep Q-network and the target network. Set the learning rate of the DQN deep Q-network to 0.001 and the discount factor to 0.9. Use various flight state data from the long-distance flight mission of the unmanned aerial vehicle as input to train the DQN deep Q-network. Through an experience playback mechanism, sample data from the playback buffer and update the Q-network parameters using gradient descent. The specific method for updating the Q-values ​​in the SA table is as follows:

[0078] In the formula, Let the spacecraft be in state at time t. Take action Q-value update value, For learning rate, Let the spacecraft be in state at time t. Take action The reward received later; Let be the reward adjustment factor for the aircraft at time t. As a discount factor, Let t be the next state of the spacecraft. The maximum Q value of all possible actions, Let the spacecraft be in state at time t. Take action The Q value.

[0079] The modified Q-value update method is used to update the SA table, and the trained DQN deep Q network is used to realize the flight path planning of the UAV. The flight path planning of UAV based on DQN deep Q network is a well-known existing technology, and the specific acquisition process will not be described in detail.

[0080] Based on the same inventive concept as the above method, this application embodiment also provides an aircraft control system for preventing long-distance flight collisions, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-described aircraft control methods for preventing long-distance flight collisions.

[0081] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments of this specification have been described above. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0082] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0083] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.

Claims

1. A method for controlling an aircraft to prevent collisions during long-distance flight, characterized in that, The method includes the following steps: Obtain the offset distance of the aircraft from the set flight path at each moment during flight, as well as the measured distance to the nearest obstacle; Analyze the dispersion of wind speed distribution faced by the aircraft during flight in each time interval, as well as the correlation of wind speed faced between different time periods within the time interval, to determine the degree of flight maneuver adjustment of the aircraft in each time interval. The coupling status between the offset distance and the measured distance at each time point is determined. By analyzing the dispersion of the offset distance during the flight of the aircraft in each time interval and the changing trend of the coupling status, the anomaly of the aircraft's state offset in each time interval is determined. By combining the flight maneuver adjustment degree and the state deviation anomaly degree, the dynamic stability of the aircraft in each time interval is obtained; Based on the autocorrelation of the flight altitude of the aircraft at all times within each time interval, and the linear growth trend of the distance between the aircraft and the starting point within each time interval, combined with the fluctuation of the extreme value distribution of the flight speed of the aircraft at all times within each time interval, the long-term balance of the aircraft in each time interval is determined. Combining the dynamic stability and long-term equilibrium, the Q-value update method in the deep Q-network is modified, and the modified deep Q-network is used to plan the flight path of the aircraft.

2. The aircraft control method for preventing long-distance flight collisions as described in claim 1, characterized in that, The determination of the flight maneuver adjustment degree includes: The degree of dispersion of the wind speed faced by the aircraft at all times within each time interval is denoted as the first degree of dispersion. Each time interval is divided into time periods, and the cumulative sum of the correlations between any two time periods within each time interval and the corresponding wind speeds faced by the aircraft is calculated. The flight maneuver adjustment degree is positively correlated with the first dispersion and negatively correlated with the summation.

3. The aircraft control method for preventing long-distance flight collisions as described in claim 1, characterized in that, Determining the coupling status between the offset distance and the measurement distance at each time step includes: calculating the ratio of the offset distance to the measurement distance at each time step as the coupling status between the offset distance and the measurement distance at each time step.

4. The aircraft control method for preventing long-distance flight collisions as described in claim 1, characterized in that, The determination of the anomaly degree of the aircraft's state deviation in each time interval includes: Obtain the fitting curve of the coupling status at all times within each time interval, calculate the slope of the coupling status at each time on the fitting curve, and calculate the Hurst exponent of the slope at all times within each time interval. The degree of dispersion of the offset distance at all times within each time interval is denoted as the second degree of dispersion; The anomaly of the aircraft's state deviation in each time interval is the product of the Hurst exponent and the second discreteness.

5. The aircraft control method for preventing long-distance flight collisions as described in claim 1, characterized in that, The determination of dynamic stability includes: The anomaly of the aircraft's state deviation in each time interval is calculated and summed with a preset value greater than 0. The dynamic stability is the normalized value of the ratio of the aircraft's flight maneuver adjustment degree in each time interval to the summed result.

6. The aircraft control method for preventing long-distance flight collisions as described in claim 1, characterized in that, Determining the long-term equilibrium of the aircraft in each time interval includes: The flight altitude of the aircraft at all times within each time interval is compiled into a flight altitude sequence, and the mean of all autocorrelation coefficients of the flight altitude sequence is calculated. Linear fitting is performed on the distance between the aircraft and the starting point at all times within each time interval. Based on the fitting error of the distance between the aircraft and the starting point at all times within each time interval on the fitted line, the nonlinearity measure of the aircraft in each time interval is determined. Calculate the degree of dispersion of the extreme values ​​of the flight speed of the aircraft at all times within each time interval, and denote it as the third degree of dispersion; By combining the mean, the nonlinear metric, and the third dispersion, the long-term balance of the aircraft in each time interval is determined.

7. The aircraft control method for preventing long-distance flight collisions as described in claim 6, characterized in that, The nonlinear metric is the sum of the fitting errors of the distance between the aircraft and the starting point at all times within each time interval on the fitted straight line.

8. The aircraft control method for preventing long-distance flight collisions as described in claim 6, characterized in that, The determination of the long-term equilibrium includes: Calculate the product of the nonlinear metric and the third dispersion, sum the product with a preset value greater than 0, and calculate the normalized value of the ratio of the mean to the summation result as the long-term balance.

9. The aircraft control method for preventing long-distance flight collisions as described in claim 1, characterized in that, The Q-value update method in the modified deep Q-network includes: The product of the dynamic stability and long-term equilibrium of the adjacent previous time interval at each time point is calculated as the reward adjustment factor at each time point. The Q-value update method in deep Q-networks is as follows: In the formula, Let the spacecraft be in state at time t. Take action Q-value update value, For learning rate, Let the spacecraft be in state at time t. Take action The reward received later; Let be the reward adjustment factor for the aircraft at time t. As a discount factor, Let t be the next state of the spacecraft. The maximum Q value of all possible actions, Let the spacecraft be in state at time t. Take action The Q value.

10. An aircraft control system for preventing long-distance flight collisions, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-9.

Images