An unmanned aerial vehicle task termination decision method, system, device and medium

Abstract

The application provides a UAV task termination decision method, system, device and medium, and belongs to the field of UAV operation and management. The task completion time of a UAV is described as a random variable subject to an exponential distribution, the number of attacks on the UAV is described as a random variable subject to a Poisson distribution, and a prior distribution is modeled through a gamma distribution. According to the sustained flight time and the number of attacks after the UAV attempts to perform a task last time, the prior distribution is updated through a Bayesian updating mechanism to determine the current belief state of the UAV. A multi-stage recursive optimization equation is established, a mapping table is obtained by solving, and when the UAV attempts to perform a task, the current optimal termination threshold is determined based on the mapping table and the current belief state of the UAV. When the sustained flight time of the UAV reaches the current optimal termination threshold, the task attempt action of the UAV is terminated and the UAV is recalled in advance. The method improves the accuracy of UAV task termination decision.

Description

Technical Field

[0001] This invention belongs to the field of unmanned aerial vehicle (UAV) operation and maintenance management, and specifically relates to a UAV mission termination decision method, system, equipment, and medium. Background Technology

[0002] Unmanned aerial vehicles (UAVs) are a typical type of autonomous unmanned platform, characterized by high mobility, flexible deployment, and strong environmental adaptability. They have been widely used, primarily in high-risk mission scenarios such as military reconnaissance, disaster relief, dynamic monitoring, and inspection and maintenance. However, UAVs face significant complexity and uncertainty challenges during mission execution. This uncertainty manifests in two main aspects: First, the uncertainty of mission performance. Complex weather conditions and strong electromagnetic interference (EMI) can affect the attitude control of UAVs, increasing control difficulty and potentially leading to navigation signal loss, resulting in significant random fluctuations in mission completion time. Second, the unpredictability of external survival threats. In adversarial environments, enemy defense systems typically detect and intercept attacks randomly, exhibiting strong heterogeneity in attack intensity, frequency, and mode of action, making accurate modeling difficult before the mission begins.

[0003] To address these needs, existing termination decisions primarily rely on pre-mission offline planning. For example, they assume key parameters such as mission completion efficiency and external threat intensity are known constants, and formulate fixed termination rules based on pre-defined risk maps or offline-calibrated threat parameters. However, in real-world adversarial environments, key information such as the opponent's firepower intensity and the actual difficulty of the mission changes constantly. Existing termination decisions struggle to reflect the characteristics of unknown parameters that are constantly updated with observed information in real-world environments, resulting in low decision accuracy. Summary of the Invention

[0004] To address the issue of low accuracy in existing UAV mission termination decisions, this invention provides a UAV mission termination decision method, system, computer equipment, and medium.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for making decisions on the termination of a drone mission includes: The mission completion time of the drone is described as a random variable following an exponential distribution, and the number of attacks on the drone is described as a random variable following a Poisson distribution. The prior distribution of the parameters of the Poisson and exponential distributions is modeled using a gamma distribution. Based on the continuous flight time of the drone after its last mission attempt and the number of attacks it has suffered, the prior distribution is updated using a Bayesian update mechanism to determine the current belief state of the drone. When a drone attempts to perform a task, the current optimal termination threshold is determined based on the drone's current belief state using a mapping table. When the drone's continuous flight time reaches the current optimal termination threshold, the drone's mission attempt is terminated and the drone is recalled ahead of schedule.

[0006] Optionally, the UAV mission termination decision method provided by the present invention further includes: Set the mission completion reward, downing penalty, early recall cost, and mission failure penalty for the drone attempting to perform the mission. Among these, the downing penalty is greater than the early recall cost, and the mission failure penalty is greater than the early recall cost. The expected total cost of a drone completing a mission is determined based on the mission completion reward, the penalty for being shot down, the cost of early recall, the penalty for mission failure, and the probability distribution of random variables.

[0007] Optionally, in the UAV mission termination decision method provided by the present invention, there are multiple UAVs, and the state variables of the multi-stage recursive optimization equation also include the number of remaining available UAVs.

[0008] Optionally, in the UAV mission termination decision method provided by the present invention, the boundary conditions of the multi-stage recursive optimization equation include: when the number of remaining available UAVs is zero, the mission fails and a mission failure penalty is imposed; when the number of attacks suffered by the UAV reaches a preset value, the number of remaining available UAVs is reduced by one; when the UAV reaches the optimal termination threshold and is recalled in advance, the number of remaining available UAVs remains unchanged.

[0009] Optionally, the UAV mission termination decision method provided by the present invention further includes: When the drone's continuous flight time does not reach the current optimal termination threshold and the mission is completed, the mission attempt is terminated and the mission is determined to be completed. If the drone's continuous flight time does not reach the current optimal termination threshold and the number of attacks it has suffered reaches a preset value, the mission attempt is terminated and the drone is determined to have been shot down.

[0010] Optionally, the belief state includes four hyperparameters of the gamma distribution, wherein the first hyperparameter is a shape parameter describing the posterior distribution of the exponential distribution, the second hyperparameter is a scale parameter describing the posterior distribution of the exponential distribution, the third hyperparameter is a shape parameter describing the posterior distribution of the Poisson distribution, and the fourth hyperparameter is a scale parameter describing the posterior distribution of the Poisson distribution. The UAV mission termination decision method provided by this invention also includes: When the drone is recalled early from its last attempt to perform a mission, the continuous flight time is added to the second and fourth hyperparameters, and the number of attacks suffered is added to the third hyperparameter. When the last time the drone attempted to perform a mission was determined to be shot down, the continuous flight time will be added to the second and fourth hyperparameters, and the number of attacks suffered will be added to the third hyperparameter. When the drone's last attempt to perform a mission is deemed successful, the continuous flight time is added to the second and fourth hyperparameters, the number of attacks suffered is added to the third hyperparameter, and the first hyperparameter is incremented by one.

[0011] Optionally, the UAV mission termination decision method provided by the present invention further includes: Using four hyperparameters as indexes, perform matching queries in the mapping table; When the current belief state of the drone falls into the discretized grid point of the belief state in the mapping table, the optimal termination threshold corresponding to the discretized grid is used as the current optimal termination threshold. When the current belief state of the drone does not fall into the discretized grid point of the belief state in the mapping table, the current optimal termination threshold is calculated by interpolation.

[0012] The present invention also provides a UAV mission termination decision system, comprising: The belief state construction module is used to describe the mission completion time of the UAV as a random variable following an exponential distribution, and the number of attacks the UAV has suffered as a random variable following a Poisson distribution. The prior distribution of the parameters of the Poisson and exponential distributions is modeled by a gamma distribution. Based on the continuous flight time of the UAV after its last attempt to perform a mission and the number of attacks it has suffered, the prior distribution is updated through a Bayesian update mechanism to determine the current belief state of the UAV. Mapping building blocks are used for; The threshold determination module is used to determine the current optimal termination threshold based on the current belief state of the drone when the drone attempts to perform a task, according to the mapping table. The mission termination module is used to terminate the drone's mission attempt and recall the drone in advance when the drone's continuous flight time reaches the current optimal termination threshold.

[0013] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement any of the steps in a method for making a decision on the termination of a drone mission.

[0014] The present invention also provides a computer-readable storage medium storing a computer program that, when loaded by a processor, can execute any step of a method for making a decision on the termination of a drone mission.

[0015] The UAV mission termination decision method provided by this invention has the following beneficial effects: This invention abandons the static preset mode of treating task completion efficiency and external threat intensity as known fixed parameters in traditional methods. Instead, it uses gamma distribution to perform prior modeling of unknown exponential distribution rate parameters and Poisson distribution intensity parameters. Combined with real-time observation data such as the actual flight time of the UAV after its last attempt to perform a mission and the number of attacks it has suffered, it uses a Bayesian update mechanism to dynamically determine the current belief state of the UAV. This allows the UAV to continuously correct its understanding of the real mission environment as observation information accumulates, thereby effectively overcoming the decision-making inaccuracy problem caused by unknown and dynamically changing parameters in real adversarial environments. Attached Figure Description

[0016] To more clearly illustrate the embodiments and design schemes of the present invention, the accompanying drawings required for this embodiment will be briefly described below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of a UAV mission termination decision method provided in an embodiment of the present invention; Figure 2 The optimal process index function provided for embodiments of the present invention Examples of the monotonicity of parameters; Figure 3 Different cost parameter scenarios provided in the embodiments of the present invention With termination threshold The changing pattern. Detailed Implementation

[0018] To enable those skilled in the art to better understand and implement the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0019] In the field of UAV operation and maintenance management research, for mission execution and risk control in uncertain environments, most existing research on risk perception and probabilistic constraint planning relies mainly on static risk maps or threat parameters obtained through offline calibration. However, in real adversarial environments, threat intensity and threat patterns change rapidly with time and space, sometimes exhibiting characteristics of randomly arriving event flows. Fixed parameters or static risk descriptions are insufficient to consistently reflect the true environmental state. Furthermore, while existing research incorporates environmental uncertainty, it largely focuses on uncertainties such as position errors, model perturbations, or motion deviations. For situations where external shocks arrive randomly and parameters are unknown, there is still a lack of a method to update threat intensity using mission process observation information and feed the updated results back to decision rules in real time. Moreover, during the mission execution phase, existing research mainly focuses on path selection, trajectory optimization, or adjustments to control strategies, but there is limited discussion on the crucial task termination issue of whether to continue mission execution.

[0020] To address the aforementioned shortcomings, and considering the randomness of task completion time and survival risk, and the difficulty for decision-makers to accurately grasp the true information about the task environment and system state before the task begins, this invention treats the problem as a sequential decision-making problem under conditions of incomplete information. As the task progresses, the system continuously acquires observational information such as duration and number of attacks, thus changing its perception of task difficulty and external threats. Based on this characteristic, this invention no longer treats the task environment as fixed, unchanging known conditions, but rather unifies parameter learning with dynamic decision-making, enabling the task termination strategy to be adjusted accordingly as information accumulates.

[0021] Specifically, this invention first models two key uncertainties in the task execution process. On the one hand, the randomness of the time required to complete the task is considered as a manifestation of the uncertainty of task execution efficiency; on the other hand, the randomness of encountering attacks during execution is considered as a manifestation of the uncertainty of external threats. To address these unknown factors, this invention employs Bayesian statistical learning, treating relevant parameters as random variables and updating them posteriorly using real-time observation data gradually acquired during task execution. Through this process, the system's understanding of the task environment can gradually transform from initial prior judgments into a more realistic dynamic cognition, providing a basis for subsequent optimal termination decisions. Based on the dynamic updating of parameters, this invention further studies the optimal termination strategy for UAVs under the constraint of limited trial opportunities. First, taking a single UAV scenario as the research object, three possible outcomes—task completion, UAV being shot down, and recall after reaching a threshold—are incorporated into a unified analytical framework, upon which a corresponding cost structure and state transition relationship are constructed. Subsequently, using Markov decision process theory and dynamic programming methods, a multi-stage recursive optimization model is established to analyze the optimal decision the system should make under different cognitive states. This approach integrates mission rewards, downtime losses, recall costs, and mission failure costs into a unified consideration, resulting in a termination criterion that can adaptively adjust to changes in environmental perception. Building upon single-UAV research, this invention further extends the model to multi-UAV sequential trial scenarios. Considering the limited carrying capacity and risk resistance of a single platform in complex adversarial environments, missions often require multiple UAVs to execute sequentially. Therefore, in multi-UAV scenarios, the system needs to consider not only whether the current mission should continue but also the preservation of remaining UAV resources and subsequent mission capabilities. Based on this, this invention introduces the number of remaining available UAVs into the original state description, expanding the system state space and establishing a corresponding recursive decision-making mechanism to analyze the formation pattern of the optimal termination strategy under resource constraints. Finally, numerical experiments are used to verify the established model and related conclusions, and the impact of different parameter and cost settings on the optimal termination threshold and system decision results is further analyzed.

[0022] Example 1 This invention provides a method for making decisions on the termination of unmanned aerial vehicle (UAV) missions, specifically as follows: Figure 1 As shown, it includes the following steps: Step 1: Describe the mission completion time of the UAV as a random variable following an exponential distribution, and the number of attacks the UAV has suffered as a random variable following a Poisson distribution. Model the prior distribution of the parameters of the Poisson and exponential distributions using a gamma distribution. Based on the UAV's continuous flight time after its last mission attempt and the number of attacks it has suffered, update the prior distribution using a Bayesian update mechanism to determine the UAV's current belief state.

[0023] Step 2: Define the mission completion reward, downing penalty, early recall cost, and mission failure penalty for the UAV attempting to execute the mission, wherein the downing penalty is greater than the early recall cost, and the mission failure penalty is greater than the early recall cost. Determine the expected total cost for the UAV to complete the mission based on the mission completion reward, downing penalty, early recall cost, mission failure penalty, and the probability distribution of random variables. With the objective of minimizing the expected total cost for the UAV to complete the mission, using belief state as the state variable and the optimal termination threshold as the decision variable, establish a multi-stage recursive optimization equation based on a Markov decision process. Solve the multi-stage recursive optimization equation point-by-point on the discretized grid of belief states using a backward dynamic programming algorithm, and establish a mapping table between the optimal termination threshold and belief states.

[0024] Step 3: When the UAV attempts to perform a task, the current optimal termination threshold is determined based on the mapping table and the UAV's current belief state. For example, a matching query is performed in the mapping table using the four hyperparameters as indices; when the UAV's current belief state falls into a discretized grid point of belief state in the mapping table, the optimal termination threshold corresponding to that discretized grid point is used as the current optimal termination threshold; when the UAV's current belief state does not fall into a discretized grid point of belief state in the mapping table, the current optimal termination threshold is calculated through interpolation.

[0025] Step 4: When the continuous flight time of the drone reaches the current optimal termination threshold, terminate the drone's mission attempt and recall the drone ahead of schedule. When the continuous flight time of the drone does not reach the current optimal termination threshold and the mission is completed, end the mission attempt and determine that the mission is completed; when the continuous flight time of the drone does not reach the current optimal termination threshold and the number of attacks suffered reaches a preset value, end the mission attempt and determine that the drone has been shot down.

[0026] Step 5: When the UAV was recalled early during its last mission attempt, the sustained flight time is added to the second and fourth hyperparameters, and the number of attacks suffered is added to the third hyperparameter; when the UAV was determined to have been shot down during its last mission attempt, the sustained flight time is added to the second and fourth hyperparameters, and the number of attacks suffered is added to the third hyperparameter; when the UAV was determined to have completed its last mission attempt, the sustained flight time is added to the second and fourth hyperparameters, the number of attacks suffered is added to the third hyperparameter, and the first hyperparameter is incremented by one. The belief state includes four hyperparameters of a gamma distribution: the first hyperparameter is a shape parameter describing the posterior distribution of the exponential distribution parameters; the second hyperparameter is a scale parameter describing the posterior distribution of the exponential distribution parameters; the third hyperparameter is a shape parameter describing the posterior distribution of the Poisson distribution parameters; and the fourth hyperparameter is a scale parameter describing the posterior distribution of the Poisson distribution parameters.

[0027] Specifically, in complex adversarial environments, UAVs face profound dual uncertainties when performing high-risk missions. First, there is the uncertainty of inherent execution efficiency: influenced by mission complexity, weather interference, and fluctuations in their own performance, the time required to complete a mission often exhibits significant randomness. Second, there is the uncertainty of external survival threats: the detection and interception behaviors of enemy defense systems are highly discrete and random, making the UAV's survivability during mission execution subject to real-time threat parameters.

[0028] During mission execution, premature termination decisions can lead to frequent mission interruptions, reducing overall mission reliability; conversely, delayed termination exposes the drone to prolonged exposure to unknown shock environments, increasing the risk of being shot down. To address this decision-making challenge, this chapter constructs a decision model with online state awareness and global optimization capabilities. The core idea of ​​this model is to treat mission performance and external shocks as dynamically evolving stochastic processes, incorporating Bayesian theory to achieve real-time learning and correction of unknown environmental parameters. Based on this, this invention utilizes dynamic programming to model the multi-stage trial process, minimizing the global expected cost encompassing mission gains, system failures, and timeout recall costs. The aim is to provide a scientific and robust mission termination criterion for drones in safety-critical environments.

[0029] First, in the model assumptions stage, we consider the drone performing the task and the time required to complete the task. Obtain the parameter as The exponential distribution, in time Number of times the interior was attacked Obtain the parameter as The Poisson distribution, The time a drone is exposed to an unknown environment, when it is attacked a certain number of times. At that time, the drone was shot down, and the moment of being shot down was Following the principles of Bayesian statistics, the parameters are... and Treat them as random variables, respectively using and This means that the two random variables are independent and obey parameters respectively. and The gamma distribution. That is... .in, express The probability density function is shown in equation (1):

[0030] (1) in, For gamma function, the expression is: Based on the above assumptions, The conditional probability density function is The conditional distribution function is It can be deduced The probability density function and distribution function are shown in Equations (2) and (3):

[0031] (2) (3) and The conditional probability density function and conditional distribution function are shown in Equation (4) and Equation (5), respectively: (4) (5) Launching at this time The distribution function is shown in equation (6): (6) Right now It follows a negative binomial distribution, that is... ,at this time The distribution function is shown in formula (7): (7) Furthermore, the present invention defines As assumed in the above implementation method, the moment when the drone is shot down is the moment when the attack threshold is reached. The minimum moment, i.e. ,Depend on Can be launched The conditional probability density function is shown in equation (8): (8) Therefore, it can be deduced that The probability density function is shown in equation (9): (9) Based on the above basic variable symbols and their distribution, the following section presents the termination decision for the drone's mission, with the drone being recalled at a specific time. This refers to the mission termination threshold, which is the threshold set when the duration of a drone's mission reaches a certain threshold. If the mission is still not completed, continuing to perform the mission at this point may result in being shot down. Therefore, in Termination decisions are made at all times. Based on the above model assumptions, there are three scenarios during the execution of a mission by a drone: First, the mission is completed, i.e., when the actual duration of the drone's mission is... The task was completed successfully without any system malfunction, and the execution time did not reach the termination threshold. The decision to terminate the mission was not implemented; secondly, the drone was shot down, i.e., when the actual duration of the drone's mission was... At this point, the drone is shot down; the mission is not yet complete, and no termination decision has been made. Finally, there is the mission termination scenario, where the actual duration of the drone's mission is... At that point, the decision to terminate the mission is executed; the drone has neither crashed nor completed its mission.

[0032] Based on this, parameters are established for different scenarios in the three cases mentioned above. and The Bayesian update formula is derived, and a cost function is established based on it. The optimal termination threshold is then determined using dynamic programming. The value of .

[0033] Specifically, for a given first On the second attempt Prior distribution parameters ,as well as Prior distribution parameters With termination threshold For the first After the second attempt and The posterior distribution and parameter update formulas are shown in formulas (10) and (11):

[0034] (10) (11) There are three possible scenarios during a drone's mission: mission completion, drone being shot down, and mission termination. In the mission completion scenario, the first... The second attempt The mission was successfully completed; the drone was neither shot down nor executed a termination decision. Therefore, [the following is possible]. Define this time and The joint posterior distribution is ,in Indicates the first The relationship between the time taken to complete the task and the number of attacks suffered by the drone during each attempt, based on Bayes' theorem, is shown in formula (12):

[0035] (12) Normalizing formula (12) yields the result shown in formula (13): (13) From formulas (10), (11), and (12), it can be seen that in the first case, i.e. when the task is completed... and Each parameter and The update formulas are shown in formulas (14) and (15); (14) (15) In the scenario where the drone is shot down, the first The second attempt The drone was shot down before the mission was completed and no termination decision was made. Define this time and The joint posterior distribution is The relational expression obtained based on Bayes' theorem is shown in formula (16): (16) Normalizing equation (16) yields equation (17): (17) From formulas (10), (11), and (17), it can be seen that in the second scenario, when the drone is shot down... and Each parameter and The update formulas are shown in formulas (18) and (19): (18) (19) In the case of mission termination, the first The secondary task is The termination decision is executed at any time, at which point the task duration is... Arrived However, the mission was not yet complete and the drone had not been shot down, therefore Define at this time and The joint posterior distribution is The relational expression obtained based on Bayes' theorem is shown in formula (20): (20) Normalizing formula (20) yields formula (21): (twenty one) From formulas (10), (11), and (21), it can be seen that in the third case, i.e. when the task terminates... and Each parameter and The update formulas are shown in formulas (22) and (23): (twenty two) (twenty three) In summary, due to the parameters for characterizing task completion rate... Its shape parameters The update depends only on the indicator variable indicating whether the task is completed, while the scale parameter... Regardless of the final outcome of the mission, the actual exposure time of that attempt will be strictly accumulated. This applies to parameters characterizing the strength of external threats. Its shape parameters It accurately recorded the total number of attacks encountered by the drone during mission execution, and the scale parameters. This also adds up to the total exposure time in uncertain environments.

[0036] The aforementioned Bayesian update framework based on conjugate priors plays a significant role in dimensionality reduction. It cleverly integrates and transforms the discrete and heterogeneous real-time observation information acquired by the UAV in successive sequential attempts into iterative belief states in the system's state space. This approach not only effectively overcomes the curse of dimensionality caused by massive historical observation data in long-term sequential decision-making but also lays a parameterized foundation for subsequent construction of recursive equations and dynamic programming models based on Markov Decision Processes (MDPs). This enables the system to adaptively adjust the optimal termination decision based on the latest posterior knowledge.

[0037] Subsequently, during the termination decision-making process, the drone had... Given several attempts, assume that the time to complete the task in each attempt follows a parameter of... The exponential distribution, the number of times attacked in the unknown environment follows the parameter . The Poisson distribution is assumed, and the optimization objective is to find the optimal task termination threshold such that the cost of completing the task is minimized. Let be the value of the threshold. The punitive costs of having a drone shot down To cover the penalties for being recalled early, The benefits resulting from completing the task. Regarding the penalty costs of mission failure, based on actual circumstances, the cost of being shot down and failing to complete a mission is greater than the cost of being recovered early. Therefore, this invention assumes... .

[0038] drones The termination threshold for attempting to complete this task is If in Previously, if the mission was successfully completed and the drone was not shot down, the mission would terminate; if... Previously, the drone was shot down, and since there was only one drone, the mission was terminated; if the drone were in... If the drone is recalled, then the drone still has The second attempt, the first Before the next attempt, after updating the parameters, all historical information was used. It means, that is , At this time, the first The expected cost of the next attempt is shown in formula (24):

[0039] (twenty four) Next, define the optimal process index function. Indicates the first The state before the next attempt ,forward All attempts failed to complete the mission, and the drone was not shot down. The minimum expected cost from the start of the trial to the end of the task is shown in formulas (25) and (26): (25) (26) in, Indicates the first The first attempt failed to complete the task, and the drone was successfully recalled. The expected cost of subsequently dispatching a drone to retrieve the optimal termination threshold is denoted as... ,but In particular, , indicating the first The task was still not completed on the second attempt, at which point the maximum number of attempts had been reached, and the task failed.

[0040] Subsequently, this invention establishes the monotonicity of two variables—the duration of the UAV's mission completion and the moment it is shot down—with respect to parameters under the likelihood ratio order. Specifically, this is based on the proposition that, given... The random variable of the time required to complete the task. , and given The random variable of the time of subsequent shooting down They each possess the following properties: Under the likelihood ratio order, as the shape parameter... The increase, Randomly decreasing; under the likelihood ratio order, with the scale parameter The increase, Randomly increasing; under the likelihood ratio order, as the shape parameter... The increase, Randomly decreasing; under the likelihood ratio order, with the scale parameter The increase, Randomly increasing.

[0041] Specifically, assuming At this point, it is as shown in formula (27): (27) right Taking the logarithm and differentiating, we obtain the relationship shown in formula (28): (28) Right now about Monotonically non-decreasing, meaning that this distribution family has no change with respect to the parameters. It has a monotonically increasing likelihood ratio, while Therefore, under the likelihood ratio order, as the shape parameter... The increase, Randomly decreasing; under the likelihood ratio order, with the scale parameter The increase, The random increment pattern is consistent. Then, assume... ,remember , The likelihood ratio function is shown in formula (29):

[0042] (29) right Taking the logarithm and differentiating, we obtain the relationship shown in formula (30): (30) Right now about Strictly monotonically decreasing. That is, this distribution family has respect to the parameter... It has a monotonically decreasing likelihood ratio, and Therefore, under the likelihood ratio order, as the shape parameter... The increase, Randomly decreasing; under the likelihood ratio order, with the scale parameter The increase, Random increments all meet the criteria.

[0043] In summary, the above propositions demonstrate that changes in prior beliefs directly affect the task execution process: when the expected operational efficiency reflected by the task completion rate parameter increases, the time required to complete the task tends to shorten stochastically; when the airspace danger level reflected by the external threat parameter increases, the survival time of the UAV before suffering a fatal attack also decreases accordingly. Therefore, changes in decision-makers' perceptions of task efficiency and threat intensity affect the final task outcome through both the task completion process and the survival process. This conclusion provides a theoretical basis for further analysis of the monotonicity of the expected cost function and the optimal process index function with respect to state parameters.

[0044] Furthermore, this invention is based on the theorem that for any _th Number of attempts and termination threshold Regarding its expected cost Yes, about Monotonically non-increasing; Regarding Monotonous and undiminished; about Monotonous and undiminished; about Monotonically non-increasing. Specifically, suppose the first... On the second attempt, For parameters The optimal threshold at that time, For parameters The optimal threshold at time , assuming for any , about Monotonic and non-increasing, that is, when hour, ,when When the relationship is as shown in formula (31):

[0045] (31) because right The threshold is a feasible threshold, not an optimal threshold. The first inequality in formula (31) holds. Combining this with the above theorem, we can determine that the second inequality holds. From the above proposition, we can deduce that the last inequality holds, thus proving that the expected cost is... Yes, about Monotonicity is non-increasing, and other monotonicity characteristics are similar, which will not be discussed in this invention.

[0046] In summary, with other parameters remaining constant, as the parameters... The increase, The increase shortens the average time required to complete the task. This not only increases the time required to reach the termination threshold. This increases the probability of completing the mission before it is completed and also reduces the time the drone is exposed to unknown and challenging environments. Therefore, the expected cost of a single attempt... and optimal process index function about It is monotonically non-increasing. Similarly, when other parameters remain constant, it increases with the parameter... The increase, The smaller the size of the drone, the slower the mission execution becomes, requiring the drone to spend more time completing the mission. This forces it to stay in dangerous environments for longer periods, increasing the cumulative probability of being shot down and also increasing the risk of mission timeouts.

[0047] because ,when An increase in the value indicates a more severe unknown environment and a higher frequency of attacks per unit of time, which directly leads to a greater cumulative number of attacks suffered by the drone during the mission. The probability of being shot down increases significantly with each subsequent attack, as being shot down incurs high penalties. Therefore, the expected cost of a single attempt and optimal process index function Increase. When Increasing the value implies that we perceive the environment to be relatively safer, and the intervals between attacks become longer. Within the same mission duration, the likelihood of the drone suffering a fatal attack decreases, thus increasing the probability of mission success. Therefore, the expected cost per attempt... and optimal process index function reduce.

[0048] Based on the above implementation methods, this invention establishes a stochastic process model for single-UAV mission execution, treating mission completion rate parameters and external threat parameters as unknown random variables. Combining real-time observation information such as mission duration and number of attacks, the posterior update formulas for relevant parameters are derived using Bayesian statistical learning methods, achieving dynamic learning of unknown environmental parameters. Furthermore, considering mission rewards, downtime penalties, recall costs, and mission failure penalties, an optimal termination decision model under finite attempt constraints is constructed, and corresponding recursive equations are established using Markov decision processes and dynamic programming methods. The variation patterns of the expected cost function and the optimal process index function with state parameters are analyzed, providing a theoretical basis for determining single-UAV mission termination strategies in unknown environments and laying the foundation for subsequent research on multi-UAV sequential attempt scenarios.

[0049] Furthermore, when there are multiple drones, the state variables of the multi-stage recursive optimization equation also include the number of remaining available drones, and the boundary conditions of the stage recursive optimization equation include: when the number of remaining available drones is zero, the task fails and the task failure penalty is applied; when the number of attacks suffered by the drone reaches a preset value, the number of remaining available drones is reduced by one; when the drone reaches the optimal termination threshold and is recalled in advance, the number of remaining available drones remains unchanged.

[0050] Specifically, considering that in actual high-intensity combat scenarios or large-scale operational tasks, the carrying capacity and risk resistance of a single platform are limited, the successful completion of a mission usually relies on multiple drones to conduct sequential attempts and coordinate with each other.

[0051] Compared to single-drone models, multi-drone collaborative scenarios introduce more complex decision variables. In this context, decision-making logic must not only weigh immediate gains against risks but also consider the dynamic depletion of limited resources and its long-term impact on subsequent mission potential. Since drones are a non-renewable strategic resource, balancing the investment in current attempts with the preservation of future capabilities is crucial to improving overall mission reliability.

[0052] Based on this, this embodiment extends the logic of the above implementation by introducing a swarm size parameter to expand the system state space and constructing an optimal termination decision model under multi-UAV sequential attempts. Through an improved dynamic programming recursive mechanism, this embodiment aims to analyze the optimal decision criteria under multi-platform constraints, providing a theoretical basis for the scientific allocation of limited resources in complex adversarial environments.

[0053] Specifically, assuming the system initially has one available drone, when the drone performs a mission in an unknown environment, its mission completion time characteristics, impact arrival process, and shot-down threshold, among other random mechanisms, remain consistent with the settings for a single drone. Assuming only one drone is dispatched for each attempt, the outcome of a single mission execution can fall into three categories: mission completion, i.e., when the actual duration of the drone's mission execution is... The task was completed successfully without any system malfunction, and the execution time did not reach the termination threshold. The decision to terminate the operation was not implemented; the drone was shot down, meaning the actual duration of the drone's mission was... If a drone is shot down, the mission is not yet complete, and no termination decision is made, the number of available drones in the system decreases by 1. Mission termination occurs when the actual duration of the drone's mission is [not specified]. At this point, a termination decision is executed. At this time, the drone has neither crashed nor completed its mission, and the number of drones available to the system remains unchanged.

[0054] In order to distinguish it from the single-UAV mission termination optimization model symbol, this embodiment will use the first symbol in the multi-UAV scenario. The expected cost of the first attempt is denoted as: As shown in formula (32): (32) in, Indicates the first The state of mind before starting the next attempt For the first The number of available drones remaining at the time of the next attempt. and This formula comprehensively considers three types of potential outcomes and their costs for a single task. The three terms correspond to the expected benefits of completing the task, the expected penalties of the drone being shot down, and the expected costs of recalling the drone if the task is not completed.

[0055] Next, we define the optimal process index function. , indicating the first Before the next attempt, the system is in a state. ,forward The task was not completed on the first attempt, from the... The minimum expected cost from the first attempt to the end of the task is shown in formulas (33) and (34): (33) (34) in, Indicates the first On the second attempt, the task was not completed, and The minimum expected cost of continuing to dispatch drones until the mission ends. Specifically, Indicates the first Before the mission could be attempted again, all drones had been shot down, leaving zero available drones. No further attempts were possible, and the mission was terminated.

[0056] Similar to the single-drone scenario, for any given first... Number of attempts and termination threshold Regarding its expected cost Regarding Monotonically non-increasing; Regarding Monotonous and undiminished; about Monotonous and undiminished; about Monotonically non-increasing, thus obtaining the parameter state Expected cost per order attempt Regarding the monotonicity of each parameter, further research will be conducted on the optimal process index function. Regarding the changing patterns of each state parameter.

[0057] Subsequently, this embodiment is based on the following theorem: for any _th _ Each attempt yields an optimal process index function. about Monotonically non-increasing; Regarding Monotonic and non-decreasing. Specifically, the above theorem can be proved by reverse mathematical induction, assuming the first... On the second attempt, For parameters The optimal threshold at that time, For parameters The optimal threshold at time, for any , about Monotonic and non-increasing, that is, when hour, The following proves the above formula. This also holds true, as shown in formula (35):

[0058] (35) because right Since the threshold is a feasible threshold rather than an optimal threshold, the first inequality holds. Furthermore, combining this with the aforementioned theorem, we can deduce that the second inequality holds. Therefore Then, based on the assumption This leads to the conclusion that the third inequality sign holds true.

[0059] Then, assuming the first On the second attempt, For parameters The optimal threshold at that time, For parameters The optimal threshold at time, for any , about Monotonous and undiminished, that is, when hour, The following proves the above formula. This also holds true, as shown in formula (36): (36) because right Since the threshold is a feasible threshold rather than an optimal threshold, the first inequality holds. Furthermore, combining this with the aforementioned theorem, we can deduce that the second inequality holds. Therefore Then, based on the assumption This leads to the conclusion that the third inequality sign holds true.

[0060] In summary, the expected cost function of a single attempt under a multi-drone sequential attempt. and optimal process index function Regarding the monotonicity of the parameters, and the conclusions above, we will now explain them from another perspective: as... The increase (or (This reduces the average time required for drones to complete their missions). In the multi-drone sequential attempt model, the improved efficiency of a single drone not only directly reduces the risk of being shot down in the current attempt, but more importantly, it indirectly protects backup drone resources by increasing the success rate of a single attempt, thereby optimizing the success probability of subsequent sequences on a global scale. This positive feedback mechanism ensures that the parameters... and Consistency of monotonicity.

[0061] and, Regarding parameters It may exhibit non-monotonic fluctuations, as... The increase (or (The reduction in the number of drones), and the harsher the unknown environment. In the single-drone model, environmental degradation inevitably leads to... Increased. However, in multi-drone models, environmental threats... The increase (or The reduction) will lead to resources from Reduce to The risk of being shot down has increased. To avoid irreversible penalties to subsequent missions due to a reduction in the number of available drones, the system may significantly shorten the termination threshold. Stabilize the number of available drones at This artificial risk-cutting mechanism, arising from resource conservation, allows the system to offset external losses by abandoning current attempts in extremely adverse environments. This breaks the logic that total cost increases linearly with worsening environmental parameters, leading to a globally optimal performance function. Environmental threat parameters no longer exhibit simple monotonicity.

[0062] In summary, this embodiment establishes an optimal termination decision model under multi-drone conditions by introducing the resource variable of the remaining available drones, and constructs the corresponding dynamic programming recursive relationship. Based on this, the properties of the expected cost function of a single attempt and the optimal process index function in a multi-drone scenario are analyzed. Research shows that the expected cost function of a single attempt is related to... The monotonicity of the optimal process index function with respect to The monotonicity remains consistent with the above embodiments, while the optimal process index function is related to It may exhibit non-monotonic variations. This indicates that multi-drone mission termination strategies are not only affected by immediate gains and risks, but are also closely related to the preservation of limited resources and the ability to conduct subsequent missions.

[0063] Example 2 Based on Example 1, this example establishes... For the optimal termination decision model under the constraint of multiple attempts, this embodiment designs an algorithm to solve for the optimal termination threshold and the corresponding task cost based on the established optimal process index function. The pseudocode is shown in Table 1: Table 1. Pseudocode table for calculating the optimal termination threshold. This algorithm primarily uses backward dynamic programming to calculate the optimal threshold. The algorithm first initializes the prior parameters. and cost parameters Attack frequency threshold The maximum termination threshold , far from a walk and maximum number of attempts In the first During the trial, state parameters With the decision-making process being dynamically updated Iterate through all possible parameter values. Specifically, considering the first... The maximum number of attacks a drone can withstand during each attempt is [number]. Therefore The range of values ​​is defined within Within.

[0064] Then, the experimental baseline was set as follows: The monotonicity of the optimal process index function with respect to parameters was investigated, as well as the existence of the optimal threshold and its evolution trend with changes in cost parameters.

[0065] Among them, the optimal process index function in the first attempt Regarding the changing trends of different parameters, as follows: Figure 2 As shown, overall, all four sets of curves exhibit a clear and stable monotonic trend, and the curve shapes are basically consistent under different cost scenarios, with differences only in numerical levels. Furthermore, when all other parameters remain constant, Follow The increase in cost shows a significant decreasing trend. This result indicates that as task completion efficiency improves, the system tends to complete tasks in a shorter time, thereby shortening the exposure time in hazardous environments and reducing the probability of triggering failure events, thus lowering expected costs. The similar downward slopes of the curves under different cost combinations indicate that this improvement effect is not dominated by any single cost, but rather has a consistent impact on the overall risk and benefit trade-off.

[0066] In fixed Under the conditions, Follow It increases and rises steadily. (And) relatively, The increase in can be understood as a decrease in task completion efficiency, leading to an increase in the expected execution time per unit task. Since a longer execution time simultaneously increases the probability of triggering termination decisions and encountering risk events, rise.

[0067] when When fixed, Follow The value increases with the magnitude of the threat and then gradually flattens out. This phenomenon indicates that an increase in the prior level related to the intensity of the environmental threat will increase the expected loss of the failure risk, thereby raising the overall expected cost; at the same time, as the threat level increases to a higher range, the value continues to increase. The incremental contribution to risk shows diminishing marginal returns, meaning that the system’s sensitivity to further deterioration of priors decreases in the high-risk range, exhibiting a certain “risk saturation” characteristic.

[0068] In fixed Under the conditions, Follow As the value increases, the marginal utility also decreases. An intuitive explanation is: Increasing the prior transfer of the corresponding environment from "more dangerous" to "safer" reduces the probability of failure events occurring within a given task duration, thus decreasing the expected contribution of the failure penalty term and consequently reducing... ,when When the risk is large enough, it is already at a relatively low level, and there is limited room for further improvement to "become safer". Therefore, the rate of decline of the curve gradually converges.

[0069] Under the five cost settings, the relative positions of the curves in each subplot are consistent: higher failure penalties or higher termination-related costs correspond to higher... Higher-yield scenarios correspond to lower costs. More importantly, regardless of how costs change, right The monotonic directions remain unchanged, indicating that the cost parameters mainly affect the absolute level and the size of the gap in the value function, without changing the direction of the effect of the prior parameters on the optimal trade-off.

[0070] Under different cost parameter scenarios With termination threshold The pattern of change is as follows Figure 3 As shown, each curve exhibits a relatively obvious unimodal characteristic, that is, with the termination threshold... The increase, The value first decreases and then increases, reaching a minimum at a certain threshold. This indicates the optimal termination threshold for the first attempt under the current model settings. Yes, it exists. This result indicates that the decision to terminate a drone mission is essentially a trade-off between "continuing to execute the mission" and "timely termination and recall": if the termination threshold is set too low, the drone will be recalled before it has had sufficient mission execution time, reducing the probability of mission completion and increasing the number of repeated attempts, thus raising the total expected cost; if the termination threshold is set too high, the drone's exposure time in the unknown adversarial environment will be significantly prolonged, increasing the risk of being shot down and the risk of mission failure, which will also lead to a higher total expected cost. Therefore, the lowest point of the curve corresponds to... It is the optimal decision result that balances returns and risks.

[0071] Further comparison of the curve positions and minimum point changes under different cost scenarios reveals that changes in cost parameters not only affect the overall level of the process indicator function but also significantly alter the position of the optimal termination threshold. When the task is completed, the benefits... As the threshold increases, the positive rewards for successfully completing a mission are amplified, and the system tends to extend the drone's mission duration to increase the chances of mission completion. Therefore, the optimal termination threshold typically shifts to the right; the penalty cost for being shot down... As the threshold increases, the cost of continued exposure of the drone to hazardous environments rises significantly, leading the system to adopt a more conservative strategy: terminating the current task earlier, thus shifting the optimal termination threshold to the left; a penalty is applied when the task fails. When the threshold is increased, to avoid premature recall leading to the rapid depletion of attempt opportunities, the system usually tends to appropriately extend the current attempt time. Therefore, the optimal termination threshold shows a certain rightward shift trend; while when the cost of early recall is increased... As the threshold increases, the cost of recall itself rises, and the system will correspondingly reduce premature termination, thus increasing the optimal termination threshold. This phenomenon indicates that the optimal termination threshold is not a fixed constant, but a dynamic result determined by the task's benefits, the cost of being shot down, the penalty for failure, and the cost of recall.

[0072] In summary, existing studies often assume that key parameters such as task completion efficiency and external threat intensity are known, or analyze optimal strategies only under fixed parameter conditions. This makes it difficult to reflect the characteristics of real-world environments where parameters are unknown and constantly updated with observational information. To address this issue, this invention extends static parameter setting into a dynamic learning framework, enhancing the model's adaptability to unknown environments. Specifically, relevant parameters are modeled as unknown random variables, and combined with real-time observational data such as task execution duration and attack counts, Bayesian statistical learning methods are used for posterior updates. The parameter learning process is also incorporated into the task termination decision, thereby improving the model's ability to characterize and adapt to uncertain environments.

[0073] Furthermore, existing research on UAV mission decision-making largely focuses on path planning, task allocation, and collaborative control, while specialized research on mission termination is relatively limited. In fact, in high-risk mission scenarios, the timing of termination directly impacts mission rewards, platform survival probability, and subsequent resource allocation. Therefore, this invention extends learning from single-parameter learning to multi-parameter joint learning, enhancing the model's ability to express complex uncertainties. Specifically, it treats the UAV mission termination problem as an independent research object, constructing a dynamic optimization model within a unified framework of mission completion, UAV downing, and mission termination and recall. The model analyzes the formation mechanism of the optimal termination strategy, thereby enriching the research content of UAV mission decision-making.

[0074] Furthermore, existing research on learning-based dynamic decision-making often focuses on updating and optimizing single unknown parameters, with less emphasis on the joint learning of multiple key parameters. Addressing the coexistence of randomness in mission duration and survival risk during UAV mission execution, this invention models the UAV mission termination problem, expanding the theoretical research on UAV mission decision-making. Specifically, it incorporates mission completion rate parameters and attack arrival process parameters into a unified learning framework. Through synchronous posterior updates of multi-dimensional parameters, it characterizes the dual uncertainties of mission difficulty and environmental threats, and based on this, studies the optimal termination strategy under the combined influence of multiple uncertainties. Compared to single-parameter learning, this embodiment can more comprehensively reflect the characteristics of complex mission environments.

[0075] Example 3 The present invention also provides a UAV mission termination decision system, comprising: The belief state construction module is used to describe the mission completion time of the UAV as a random variable following an exponential distribution, and the number of attacks suffered by the UAV as a random variable following a Poisson distribution. The prior distribution of the parameters of the Poisson distribution and the exponential distribution is modeled by a gamma distribution. Based on the continuous flight time of the UAV after its last attempt to perform a mission and the number of attacks suffered, the prior distribution is updated through a Bayesian update mechanism to determine the current belief state of the UAV. Mapping building blocks are used for; The threshold determination module is used to determine the current optimal termination threshold based on the current belief state of the drone when the drone attempts to perform a task, according to a mapping table. The mission termination module is used to terminate the mission attempt of the drone and recall the drone in advance when the continuous flight time of the drone reaches the current optimal termination threshold.

[0076] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the steps in an embodiment of a UAV mission termination decision method. Specific implementation methods can be found in the method embodiments, and will not be repeated here.

[0077] Furthermore, the present invention also provides a non-transitory computer-readable storage medium containing instructions on which a computer program is stored. For example, a memory containing instructions that can be executed by a processor of a computer device to perform the above-described method. For example, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc. When the computer program is executed by the processor, it can implement the steps in an embodiment of a UAV mission termination decision method. Specific implementation methods can be found in the method embodiments, which will not be repeated here.

[0078] Those skilled in the art will understand that embodiments of the present invention can provide methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0079] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0080] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0081] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0082] It should be noted that the specific embodiments described above enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. Therefore, although the present invention has been described in detail in this specification and embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention; and all technical solutions and improvements that do not depart from the spirit and scope of the present invention are covered within the protection scope of the present invention patent. No reference numerals in the claims should be construed as limiting the scope of the claims. Any simple variations or equivalent substitutions of technical solutions that can be readily obtained by those skilled in the art within the scope of the technology disclosed in the present invention are within the protection scope of the present invention.

Claims

1. A method for making decisions on terminating a drone mission, characterized in that, include: The mission completion time of the UAV is described as a random variable following an exponential distribution, and the number of attacks on the UAV is described as a random variable following a Poisson distribution. The prior distribution of the parameters of the Poisson and exponential distributions is modeled using a gamma distribution. Based on the continuous flight time of the UAV after its last mission attempt and the number of attacks it has suffered, the prior distribution is updated using a Bayesian update mechanism to determine the current belief state of the UAV. With the goal of minimizing the expected total cost of the drone completing the mission, and with belief state as the state variable and the optimal termination threshold as the decision variable, a multi-stage recursive optimization equation is established. On the discretized grid of belief states, the multi-stage recursive optimization equation is solved to obtain the optimal termination threshold and mapping table; When a drone attempts to perform a task, the current optimal termination threshold is determined based on the drone's current belief state using a mapping table. When the continuous flight time of the drone reaches the current optimal termination threshold, the drone's mission attempt is terminated and the drone is recalled in advance.

2. The UAV mission termination decision method according to claim 1, characterized in that, Before aiming to minimize the expected total cost of the drone completing the mission, the following also applies: The system sets the mission completion reward, downing penalty, early recall cost, and mission failure penalty for the drone attempting to perform the mission, wherein the downing penalty is greater than the early recall cost, and the mission failure penalty is greater than the early recall cost. The expected total cost for the UAV to complete the mission is determined based on the mission completion reward, the penalty for being shot down, the cost of early recall, the mission failure penalty, and the probability distribution of random variables.

3. The UAV mission termination decision method according to claim 1, characterized in that, The drones are multiple, and the state variables of the multi-stage recursive optimization equation also include the number of remaining available drones.

4. The UAV mission termination decision method according to claim 3, characterized in that, The boundary conditions of the multi-stage recursive optimization equation include: when the number of remaining available drones is zero, the mission fails and a mission failure penalty is applied; when the number of attacks suffered by a drone reaches a preset value, the number of remaining available drones is reduced by one; when a drone reaches the optimal termination threshold and is recalled early, the number of remaining available drones remains unchanged.

5. The UAV mission termination decision method according to claim 1, characterized in that, After determining the current optimal termination threshold based on the mapping table and the current belief state of the UAV, the process further includes: When the drone's continuous flight time does not reach the current optimal termination threshold and the task is completed, the task attempt is terminated and the task is determined to be completed. If the drone's continuous flight time does not reach the current optimal termination threshold and the number of attacks it has suffered reaches a preset value, the mission attempt is terminated and the drone is determined to have been shot down.

6. The UAV mission termination decision method according to claim 5, characterized in that, The belief state includes four hyperparameters of the gamma distribution, where the first hyperparameter is the shape parameter describing the posterior distribution of the exponential distribution, the second hyperparameter is the scale parameter describing the posterior distribution of the exponential distribution, the third hyperparameter is the shape parameter describing the posterior distribution of the Poisson distribution, and the fourth hyperparameter is the scale parameter describing the posterior distribution of the Poisson distribution. Based on the UAV's sustained flight time since its last mission attempt and the number of attacks it has suffered, the prior distribution is updated using a Bayesian update mechanism to determine the UAV's current belief state, including: When the drone was recalled early from its last attempt to perform a mission, the continuous flight time was added to the second and fourth hyperparameters, and the number of attacks suffered was added to the third hyperparameter. When the last time the drone attempted to perform a mission was determined to be shot down, the continuous flight time is added to the second hyperparameter and the fourth hyperparameter, and the number of attacks suffered is added to the third hyperparameter; When the last time the drone attempted to perform a task, the task was determined to be completed, the continuous flight time was added to the second and fourth hyperparameters, the number of attacks suffered was added to the third hyperparameter, and the first hyperparameter was incremented by one.

7. The UAV mission termination decision method according to claim 1, characterized in that, Determining the current optimal termination threshold based on the mapping table and the current belief state of the drone includes: Using the four hyperparameters as indexes, a matching query is performed in the mapping table; When the current belief state of the UAV falls into the discretized grid point of the belief state in the mapping table, the optimal termination threshold corresponding to the discretized grid is used as the current optimal termination threshold. When the current belief state of the UAV does not fall into the discretized grid point of the belief state in the mapping table, the current optimal termination threshold is calculated by interpolation.

8. A UAV mission termination decision system, characterized in that, include: The belief state construction module is used to describe the mission completion time of the UAV as a random variable following an exponential distribution, and the number of attacks suffered by the UAV as a random variable following a Poisson distribution. The prior distribution of the parameters of the Poisson distribution and the exponential distribution is modeled by a gamma distribution. Based on the continuous flight time of the UAV after its last attempt to perform a mission and the number of attacks suffered, the prior distribution is updated through a Bayesian update mechanism to determine the current belief state of the UAV. Mapping building blocks are used for; The threshold determination module is used to determine the current optimal termination threshold based on the current belief state of the drone when the drone attempts to perform a task, according to a mapping table. The mission termination module is used to terminate the mission attempt of the drone and recall the drone in advance when the continuous flight time of the drone reaches the current optimal termination threshold.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the unmanned aerial vehicle (UAV) mission termination decision method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is loaded by the processor, it is able to execute the steps of the unmanned aerial vehicle mission termination decision method according to any one of claims 1 to 7.

Images