Unmanned aerial vehicle flight intention recognition method and system

Abstract

The invention provides an unmanned aerial vehicle flight intention recognition method and system, and the method comprises the steps: receiving the flight data of an unmanned aerial vehicle, and constructing a multi-dimensional behavior expectation model in combination with a preset behavior pattern library and an airspace use rule; utilizing a deep belief network to map flight data into the model, calculating probability distribution of different flight intentions, and generating simulated behavior performance and probability distribution results; through a Markov decision process, comparing the deviation measurement of the simulated flight activity and the actual flight activity, and evaluating and predicting the optimal flight intention; and generating an operation suggestion according to the prediction result and the instant airspace safety state, sending the operation suggestion to the unmanned aerial vehicle, and generating a flight guidance suggestion based on execution feedback information. The flight intention recognition precision of the unmanned aerial vehicle and the reliability of real-time operation suggestions are improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of behavior prediction and intelligent management of unmanned aerial vehicles, and in particular, to a method and system for identifying the flight intention of an unmanned aerial vehicle. Background Art

[0002] With the widespread application of drones in logistics distribution, agricultural monitoring, environmental monitoring, and military reconnaissance, higher requirements are placed on the accurate identification and real-time response of drone flight intentions. In order to ensure the safe and efficient operation of drones, it is necessary to develop a technology that can accurately analyze the flight intentions of drones. This technology not only needs to process a large amount of flight parameters and environmental perception data from drones, but also needs to ensure the accuracy and consistency of data through spatiotemporal consistency calibration. In addition, the system needs to combine the preset behavior pattern library and airspace usage rules to build a multi-dimensional behavior expectation model, and use advanced deep learning methods such as deep belief networks to map flight data to the model, calculate the probability distribution of different flight intentions, and generate simulated behavior performance and probability distribution results. Based on these prediction results, the system also needs to evaluate and predict the optimal flight intention of the drone through the Markov decision process, generate operation suggestions based on the immediate airspace safety status, send operation instructions to the drone, and further optimize the flight guidance suggestions based on the feedback information after execution, so as to achieve closed-loop control and improve the success rate and stability of drone mission execution.

[0003] Existing technical solutions usually use spatiotemporal consistency calibration to process flight parameters and environmental perception data to ensure data consistency and accuracy. Then, a behavior pattern library is established based on historical flight data or theoretical models, and a multi-dimensional behavior expectation model is constructed in combination with airspace usage rules. Next, a deep belief network is used to map the flight data to the model, calculate the probability distribution of different flight intentions, and generate simulated behavior performance and probability distribution results. Through the Markov decision process, the deviation measures between the simulated behavior performance and the actual flight activities are compared, such as position deviation, speed change, and time difference in response to environmental interactions, to evaluate and predict the flight intention of the drone. Finally, based on the prediction results and the immediate airspace safety status, operation suggestions are generated, instructions are sent to the drone, and flight guidance suggestions are generated based on the feedback information after the drone executes to optimize subsequent operations.

[0004] Although the existing solutions have solved the problem of drone flight intention identification to a certain extent, there are still several key defects. First, there is a lack of detailed description of the specific methods, algorithms or standards for data calibration and preprocessing, which leads to insufficient operability in practical applications. Secondly, the establishment and update mechanism of the behavior pattern library is unclear, and it cannot dynamically adapt to new data and changing environmental conditions, which affects the adaptability and long-term accuracy of the system. In addition, the application details of the deep belief network are missing, including key information such as the specific architecture, loss function selection during training, and optimizer configuration, which makes it difficult to reproduce and optimize the method. At the same time, the selection of deviation metrics and the quantification of their relative importance are not detailed. The importance of these factors may vary in different mission scenarios, affecting the accuracy of the prediction. Finally, there is a lack of specific description of the core elements of the Markov decision process, such as the state space definition, transition probability calculation, and reward function design, which makes it difficult to fully understand and effectively implement the method. Summary of the invention

[0005] The embodiments of the present application provide a method and system for identifying UAV flight intentions, so as to solve the problems of low accuracy in identifying UAV flight intentions and poor reliability of real-time operation suggestions in the prior art.

[0006] In a first aspect, an embodiment of the present application provides a method for identifying a drone's flight intention, comprising: Receiving flight data from the UAV, wherein the flight data includes flight parameters and environmental perception data after being calibrated for spatiotemporal consistency; Using the flight data, a multi-dimensional behavior expectation model of the UAV is constructed according to a preset behavior pattern library and airspace use rules; Using a deep belief network, the flight data is mapped to the multi-dimensional behavior expectation model to calculate the probability distribution of the UAV under different flight intentions, and generate simulated behavior performance and corresponding probability distribution results; Based on the probability distribution result, the deviation metric between the simulated behavior performance and the actual flight activity is compared through a Markov decision process to evaluate and predict the flight intention of the UAV and obtain an optimal flight intention prediction, wherein the deviation metric includes position deviation, speed change, and time difference in response to environmental interaction; According to the flight intention and in combination with the immediate airspace safety status, an operation suggestion for the UAV is generated, and the operation suggestion is sent to the UAV to obtain feedback information after the UAV executes the operation, and based on the feedback information, a flight guidance suggestion is generated.

[0007] Optionally, the deep belief network is used to map the flight data to the multi-dimensional behavior expectation model to calculate the probability distribution of the UAV under different flight intentions, and generate simulated behavior performance and corresponding probability distribution results, including: Using a deep belief network, performing nonlinear transformation processing on the flight data, and using an attention mechanism to dynamically adjust the weight of the flight data to obtain a flight data representation; mapping the flight data representation to different portions of the multi-dimensional behavior expectation model to identify behavior patterns that match the flight data representation; Based on the behavior pattern, a flight intention model of the UAV in the multi-dimensional behavior expectation model is constructed, and a similarity metric learning method is applied to evaluate the similarity between the flight intention models to determine the optimal parameter configuration of the flight intention model and generate a behavior pattern configuration; According to the behavior pattern configuration combined with the Bayesian reasoning framework, the probability distribution of the drone executing each flight intention is calculated, and the variational inference technique is used to obtain the probability distribution result; Based on the probability distribution results and the behavior pattern configuration, a Monte Carlo method is used to generate a series of simulated behavior performances and corresponding probability distribution results.

[0008] Optionally, based on the behavior pattern, constructing a flight intention model of the UAV in the multi-dimensional behavior expectation model, and applying a similarity metric learning method to evaluate the similarity between the flight intention models to determine the optimal parameter configuration of the flight intention model, and generating a behavior pattern configuration, including: Mapping the flight intention of the behavior pattern to different parts of the multi-dimensional behavior expectation model to construct multiple flight intention models; Using the Bayesian optimization method, the best initial parameter combination is selected from the flight behavior patterns in the pre-training data set, and the Gaussian process is used in combination with the best initial parameter combination to initialize the parameter configuration of the flight intention model; Utilizing a variety of distance measurement methods and kernel methods combined with twin network learning, calculating the similarity between the flight data representation and the flight intention model, and quantifying the degree of matching between the flight data representation and the flight intention model, and generating a similarity measurement result; According to the similarity measurement result combined with the initialization parameter configuration, a gradient descent algorithm and a reinforcement learning mechanism are applied to determine the optimal parameter configuration from the parameters of the flight intention model; Based on the optimal parameter configuration, a behavior pattern configuration is generated using a situational awareness reasoning mechanism.

[0009] Optionally, the determining the optimal parameter configuration from the parameters of the flight intention model by applying a gradient descent algorithm and a reinforcement learning mechanism according to the similarity measurement result in combination with the initialization parameter configuration includes: According to the similarity measurement result and the initialization parameter configuration, an initial loss function is set, and a gradient descent algorithm is applied to obtain initial parameters of the flight intention model by calculating the gradient of the initial loss function with respect to the initialization parameter configuration and adopting an adaptive learning rate strategy; Using a reinforcement learning mechanism, a reward function is defined, and based on the reward function combined with actual flight data feedback, initial parameters of the flight intention model are dynamically adjusted to generate an intermediate parameter configuration; According to the flight data, the intermediate parameter configuration is evaluated to obtain a verification result, and the weights in the loss function and the reward function are adjusted according to the verification result to generate a closed-loop feedback mechanism, and based on the closed-loop feedback mechanism, the optimal parameter configuration is determined.

[0010] Optionally, based on the probability distribution result, by using a Markov decision process, comparing the deviation measure between the simulated behavior performance and the actual flight activity to evaluate and predict the flight intention of the UAV and obtain the optimal flight intention prediction, including: Based on the probability distribution result, a state transition model is constructed; according to the state transition model, in combination with a preset behavior pattern library and airspace usage rules, a Markov decision process model is initialized, a reward function is defined, and based on the reward function, an initial Markov decision process configuration is generated; By comparing the deviation measures between the simulated behavior performance and the actual flight activities, a series of deviation measure results are obtained; Based on the deviation measurement result and the initial Markov decision process configuration, applying a dynamic time warping algorithm to evaluate the match between the simulated behavior performance and the actual flight activity, and generating a match evaluation result; According to the matching evaluation result, the flight intention corresponding to the simulated behavior performance of the actual flight activity is selected as the prediction result to obtain the optimal flight intention prediction.

[0011] Optionally, the applying a dynamic time warping algorithm based on the deviation measurement result and the initial Markov decision process configuration to evaluate the matching degree between the simulated behavior performance and the actual flight activity and generate a matching degree evaluation result includes: Preprocessing the simulated behavior performance according to the deviation measurement result and in combination with the initial Markov decision process configuration to obtain the preprocessed simulated behavior performance; Determining the parameter configuration of the dynamic time warping algorithm based on the characteristics of the simulated behavior performance after the preprocessing and the deviation measurement result; Using the parameter configuration of the dynamic time warping algorithm, the time series matching degree between the simulated behavior performance and the actual flight activity is calculated to obtain a time series matching degree result; quantifying the similarity between the simulated behavior performance and the actual flight activity according to the time series matching result, and generating a matching score; Based on the matching score combined with the Markov decision process configuration, the matching of the simulated behavior performance is evaluated to generate a matching evaluation result.

[0012] Optionally, the use of the flight data to construct a multi-dimensional behavior expectation model of the drone according to a preset behavior pattern library and airspace usage rules includes: Performing feature extraction processing on the flight data to obtain a set of key feature sets; According to a preset behavior pattern library, the key feature set is compared with the flight behavior patterns in the preset behavior pattern library, the patterns related to the flight behavior of the UAV are identified, and the behavior pattern results are generated; Using the behavior pattern results, applying airspace use rules, evaluating whether the UAV flight behavior complies with the predicted aviation management regulations and safety requirements, and obtaining the flight behavior pattern; According to the flight behavior pattern, a multi-dimensional behavior expectation model is constructed.

[0013] In a second aspect, an embodiment of the present application provides a drone flight intention recognition system, comprising: A receiving module, used to receive flight data from the UAV, wherein the flight data includes flight parameters and environmental perception data after time-space consistency calibration processing; A construction module, for constructing a multi-dimensional behavior expectation model of the UAV based on the flight data and a preset behavior pattern library and airspace use rules; A calculation module, used to map the flight data into the multi-dimensional behavior expectation model using a deep belief network, so as to calculate the probability distribution of the UAV under different flight intentions, and generate simulated behavior performance and corresponding probability distribution results; A comparison module, for comparing the deviation metric between the simulated behavior performance and the actual flight activity through a Markov decision process based on the probability distribution result, so as to evaluate and predict the flight intention of the UAV and obtain an optimal flight intention prediction, wherein the deviation metric includes position deviation, speed change, and time difference of interactive response to the environment; The generation module is used to generate operation suggestions for the drone according to the flight intention and the immediate airspace safety status, send the operation suggestions to the drone, obtain feedback information after the drone executes the operation, and generate flight guidance suggestions based on the feedback information.

[0014] In a third aspect, an embodiment of the present application provides a computing device, comprising a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute a method for identifying a drone flight intention as described in any one of the first aspects.

[0015] In a fourth aspect, an embodiment of the present application provides a computer storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement a method for identifying the flight intention of a drone as described in any one of the first aspects.

[0016] In an embodiment of the present application, flight data from a drone is received, wherein the flight data includes flight parameters and environmental perception data that have been calibrated for spatiotemporal consistency; a multidimensional behavior expectation model of the drone is constructed using the flight data according to a preset behavior pattern library and airspace usage rules; a deep belief network is used to map the flight data to the multidimensional behavior expectation model to calculate the probability distribution of the drone under different flight intentions, and to generate simulated behavior performance and corresponding probability distribution results; based on the probability distribution results, a deviation metric between the simulated behavior performance and the actual flight activity is compared through a Markov decision process to evaluate and predict the flight intention of the drone, and obtain an optimal flight intention prediction, wherein the deviation metric includes position deviation, speed change, and time difference in response to environmental interaction; based on the flight intention and in combination with the immediate airspace safety status, an operation suggestion for the drone is generated, and the operation suggestion is sent to the drone to obtain feedback information after the drone executes the suggestion, and based on the feedback information, a flight guidance suggestion is generated.

[0017] The technical solution of this application has the following beneficial effects: This application integrates advanced technologies such as deep learning, behavior modeling, and Markov decision process. This invention not only improves the accuracy of drone flight intention recognition, but also provides timely and effective operational suggestions in a dynamically changing airspace environment, greatly enhancing the safety and operating efficiency of drones, and providing a solid technical guarantee for the large-scale application of drones. In addition, the flight guidance strategy is continuously optimized based on the feedback mechanism, further ensuring the success rate and stability of drone mission execution.

[0018] Furthermore, the flight data is processed nonlinearly by using a deep belief network, and the weight of the flight data is dynamically adjusted using an attention mechanism to obtain a flight data representation. Subsequently, the representation is mapped to different parts of the multidimensional behavior expectation model to identify the matching behavior patterns. Based on these behavior patterns, a flight intention model of the UAV is constructed, and a similarity metric learning method is applied to evaluate the similarity between different flight intention models, determine the optimal parameter configuration, and generate a behavior pattern configuration. Next, the probability distribution of each flight intention is calculated in combination with the Bayesian reasoning framework, and the variational inference technique is used to optimize the probability distribution results. Finally, based on the obtained probability distribution results and behavior pattern configuration, the Monte Carlo method is used to generate a series of simulated behavior performances and their corresponding probability distribution results.

[0019] Through the above methods, the deep belief network and attention mechanism enhance the nonlinear expression ability of flight data and ensure the effectiveness and flexibility of feature extraction. Through the multi-dimensional behavior expectation model and similarity metric learning method, the flight intention of the drone can be accurately identified and predicted, improving the accuracy and reliability of the prediction. Combining the Bayesian reasoning framework and variational inference technology, the calculation process of the probability distribution is further optimized, enabling the system to respond quickly to changes in complex environments. Finally, the simulated behavior performance generated by the Monte Carlo method not only improves the adaptability and robustness of the system, but also provides more accurate operation suggestions for the drone, significantly improving the execution efficiency and safety of the drone in actual tasks. This series of technical means work together to achieve a high degree of automation and intelligence in the recognition of drone flight intentions, effectively ensuring the safe operation of drones and the successful completion of missions.

[0020] These and other aspects of the present application will become more clearly understood in the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 A flowchart of a method for identifying the flight intention of a drone provided in an embodiment of the present application; Figure 2 A schematic diagram of the structure of a UAV flight intention recognition system provided in an embodiment of the present application; Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0024] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to different types.

[0025] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0026] Figure 1 A flowchart of a method for identifying the flight intention of a drone is provided for an embodiment of the present application. Figure 1 As shown, the method includes: Step 101: receiving flight data from a drone, wherein the flight data includes flight parameters and environment perception data after time-space consistency calibration; In this step, the flight data includes flight parameters (such as position, speed, heading, etc.) and environmental perception data (such as weather conditions, obstacle information) after time-space consistency calibration. Time-space consistency calibration ensures the synchronization of these data in time and space, thus providing a reliable basis for subsequent analysis. This step is the prerequisite for the operation of the entire system and ensures the quality and accuracy of the input data.

[0027] In actual operation, first, the drone collects flight parameters and environmental perception data through various sensors on board. Then, a specific algorithm is used to calibrate the temporal and spatial consistency of these raw data to eliminate data inconsistencies caused by time delays or spatial differences. The calibrated data is then sent to a ground station or cloud server for further analysis and processing.

[0028] For example, in an urban logistics delivery project, drones are responsible for delivering packages from warehouses to customers. After the drone takes off, it continuously sends a data stream containing its current location, flight speed, and surrounding environment information (such as wind speed and building distribution). These data are calibrated for temporal and spatial consistency to ensure that they remain highly accurate even in dynamically changing urban environments, providing a solid foundation for subsequent path planning.

[0029] Step 102: Using the flight data, a multi-dimensional behavior prediction model of the UAV is constructed according to a preset behavior pattern library and airspace usage rules; In this step, the multi-dimensional behavior prediction model is a behavior prediction framework that comprehensively considers multiple factors (such as drone performance, mission requirements, and airspace rules). It is based on a preset behavior pattern library and real-time flight data, and aims to simulate the expected behavior of drones in different scenarios to help identify and predict possible flight intentions.

[0030] In actual operation, the system first loads the preset behavior pattern library, which contains different types of task behavior templates (such as cruising, hovering, and obstacle avoidance). Then, combined with the currently received flight data and airspace usage rules, the model parameters are adjusted to generate a multi-dimensional behavior expectation model that reflects the current task requirements. This step provides a theoretical basis and reference framework for subsequent flight intention recognition.

[0031] For example, continuing with the logistics delivery example above, when the drone begins to perform a delivery mission, the system builds a multi-dimensional behavior expectation model specifically for this mission based on factors such as the distance between the drone's starting point and destination, the estimated flight time, and real-time traffic conditions and weather forecasts. This model not only takes into account the basic operating characteristics of the drone, but also incorporates external environmental constraints, such as no-fly zones and severe weather areas, to ensure that the mission is carried out safely and smoothly.

[0032] Step 103: using a deep belief network, mapping the flight data to the multi-dimensional behavior expectation model to calculate the probability distribution of the drone under different flight intentions, and generating simulated behavior performance and corresponding probability distribution results; In this step, the Deep Belief Network (DBN) is used to map the calibrated flight data into a multi-dimensional behavior expectation model, calculate the probability distribution of the drone under different flight intentions, and generate the corresponding simulated behavior performance and its probability distribution results. This process helps to evaluate the different action paths that the drone may take and their likelihood.

[0033] In actual operation, DBN first performs nonlinear transformation on flight data and dynamically adjusts weights using the attention mechanism to obtain a representation of the flight data. Next, this representation is mapped to different parts of the multi-dimensional behavior expectation model to identify matching behavior patterns. Based on these patterns, the system builds a flight intention model for the drone and applies similarity metric learning methods to determine the optimal parameter configuration and generate a behavior pattern configuration.

[0034] For example, in the logistics delivery process, when the drone approaches the target location, the system uses DBN to analyze its current flight status and environmental conditions and predict the behavior that the drone may take, such as slowing down to prepare for landing or bypassing temporary obstacles. By generating multiple simulated behavior performances and their probability distribution, the system can plan the best landing path in advance and optimize the efficiency of task execution.

[0035] Step 104: Based on the probability distribution result, the deviation metric between the simulated behavior performance and the actual flight activity is compared through a Markov decision process to evaluate and predict the flight intention of the UAV and obtain an optimal flight intention prediction, wherein the deviation metric includes position deviation, speed change, and time difference in response to environmental interaction; In this step, based on the probability distribution results generated in the previous step, the Markov Decision Process (MDP) is used to compare the deviation metrics (such as position deviation, speed change, and response time difference) between the simulated behavior performance and the actual flight activity, evaluate and predict the flight intention of the drone, and obtain the optimal flight intention prediction.

[0036] In actual operation, the system uses MDP to evaluate the deviation between each simulated behavior and the actual flight activity, and selects the behavior with the smallest deviation as the optimal flight intention prediction. This process not only takes into account the changes in physical distance and speed, but also incorporates the drone's response time to environmental changes to ensure the comprehensiveness and accuracy of the prediction results.

[0037] For example, in a logistics delivery mission, when a drone approaches a target location, the system uses MDP to quickly evaluate the effectiveness of various response strategies, such as adjusting the flight altitude to avoid sudden obstacles or speeding up to cope with upcoming bad weather. Ultimately, the system selects the most appropriate flight intention, allowing the drone to complete the delivery mission safely and efficiently.

[0038] Step 105: Based on the flight intention and the immediate airspace safety status, generate operation suggestions for the UAV, send the operation suggestions to the UAV, obtain feedback information after the UAV executes the operation, and generate flight guidance suggestions based on the feedback information.

[0039] In this step, based on the predicted flight intention and the immediate airspace safety status, operation suggestions for the drone are generated and sent to the drone. The system also receives feedback from the drone after execution, and based on this information, further generates flight guidance suggestions to optimize subsequent operations.

[0040] In actual operation, the system comprehensively considers the flight intention of the drone and the safety status of the current airspace, and formulates detailed operational suggestions, such as adjusting the flight route to avoid obstacles or enter a new monitoring area. After the suggestions are sent to the drone, the system will monitor the execution effect and continuously optimize the flight guidance suggestions based on the feedback information.

[0041] For example, when the logistics delivery mission was nearing its end, the system developed a detailed return plan based on the drone's flight intentions and the remaining mission volume. At the same time, considering the upcoming storm, the system recommended that the drone return to the base faster. After the drone executed the command, the system adjusted the subsequent task schedule based on its feedback, ensuring the smooth completion of the entire project and accumulating valuable experience for similar missions in the future.

[0042] Through the above five steps, this method significantly improves the accuracy and real-time response capability of UAV flight intention recognition. Specifically, spatiotemporal consistency calibration ensures the accuracy and consistency of input data; the multi-dimensional behavior expectation model enhances the adaptability and flexibility of the system; the application of deep belief networks and Markov decision processes improves the reliability of flight intention prediction; and the closed-loop feedback mechanism optimizes the effectiveness of operational recommendations. Overall, this process not only ensures the safe operation of UAVs, but also greatly improves the success rate and efficiency of their mission execution, and realizes efficient automated management in complex and changing environments.

[0043] In order to further improve the accuracy and adaptability of drone flight intention recognition, in some embodiments, the deep belief network described in step 103 is used to map the flight data to the multi-dimensional behavior expectation model to calculate the probability distribution of the drone under different flight intentions, and generate simulated behavior performance and corresponding probability distribution results, including: The flight data is nonlinearly transformed by using a deep belief network, and the weight of the flight data is dynamically adjusted by using an attention mechanism to obtain a flight data representation; the flight data representation is mapped to different parts of the multidimensional behavior expectation model to identify behavior patterns that match the flight data representation; based on the behavior pattern, a flight intention model of the UAV in the multidimensional behavior expectation model is constructed, and a similarity metric learning method is applied to evaluate the similarity between the flight intention models to determine the optimal parameter configuration of the flight intention model and generate a behavior pattern configuration; based on the behavior pattern configuration combined with a Bayesian reasoning framework, the probability distribution of the UAV executing each flight intention is calculated, and a variational inference technique is used to obtain a probability distribution result; based on the probability distribution result combined with the behavior pattern configuration, a Monte Carlo method is used to generate a series of simulated behavior performances and corresponding probability distribution results.

[0044] In this embodiment, DBN is used to perform nonlinear transformation processing on flight data, and the weight of each feature is dynamically adjusted by introducing an attention mechanism to obtain a more effective representation of flight data. The multi-dimensional behavior expectation model is a behavior prediction framework that comprehensively considers multiple factors (such as drone performance, mission requirements, and airspace rules), and can identify behavior patterns that match the current flight data representation. Based on these patterns, the system constructs a flight intention model of the drone, and determines the optimal parameter configuration through a similarity metric learning method, and finally generates a behavior pattern configuration. Combining the Bayesian reasoning framework and variational inference technology, the calculation process of the probability distribution is further optimized to ensure the accuracy and reliability of the prediction.

[0045] In an embodiment of the present application, first, the system uses DBN to perform nonlinear transformation processing on the flight data, and dynamically adjusts the weights through the attention mechanism to obtain a representation of the flight data. Then, the representation is mapped to different parts of the multi-dimensional behavior expectation model to identify the behavior patterns that match it. Based on these behavior patterns, the system builds a flight intention model for the drone, and applies a similarity metric learning method to evaluate the similarity between different flight intention models, determine the optimal parameter configuration, and generate a behavior pattern configuration. Next, the probability distribution of each flight intention is calculated in combination with the Bayesian reasoning framework, and the variational inference technique is used to optimize the probability distribution results. Finally, based on the obtained probability distribution results and behavior pattern configuration, the Monte Carlo method is used to generate a series of simulated behavior performances and their corresponding probability distribution results to provide more accurate operational recommendations.

[0046] Here is a specific example: In an urban emergency response project, drones are deployed to quickly assess the situation at the fire scene and guide rescue operations. In step 103, the drone first collects various data at the fire scene through its sensors, including temperature, smoke concentration, obstacle location, etc. The system uses DBN to perform nonlinear transformation processing on these raw data, and dynamically adjusts the weight of each feature through the attention mechanism to obtain a more accurate flight data representation. Then, the system maps the flight data representation to the pre-built multi-dimensional behavior expectation model to identify behavior patterns that match the current environmental conditions, such as bypassing high temperature areas or avoiding smoke areas. Based on these behavior patterns, the system constructs a flight intention model specifically for this emergency response mission, and evaluates the similarity between different flight intention models through similarity metric learning methods, determines the optimal parameter configuration, and generates a behavior pattern configuration.

[0047] Subsequently, the system combines the Bayesian reasoning framework to calculate the probability distribution of each flight intention, and uses variational inference technology to optimize the probability distribution results. Based on these probability distribution results, the system uses the Monte Carlo method to generate a series of simulated behavior performances and their corresponding probability distribution results, providing multiple possible flight paths for selection. In actual operation, the system selects the safest and most efficient flight path, enabling the drone to quickly reach the core area of ​​the fire, obtain key information, and provide real-time feedback to the ground command center. In this way, not only the efficiency of drone mission execution is improved, but also the overall effect of emergency response is significantly enhanced, ensuring the safety of personnel and property.

[0048] In order to further improve the accuracy of UAV flight intention recognition and the effectiveness of parameter optimization, in some embodiments, the step 103 constructs a UAV flight intention model in the multi-dimensional behavior expectation model based on the behavior pattern, and applies a similarity metric learning method to evaluate the similarity between the flight intention models to determine the optimal parameter configuration of the flight intention model and generate a behavior pattern configuration, further comprising: The flight intention of the behavior pattern is mapped to different parts of the multi-dimensional behavior expectation model to construct multiple flight intention models; the Bayesian optimization method is used to select the best initial parameter combination from the flight behavior pattern in the pre-training data set, and the Gaussian process is used in combination with the best initial parameter combination to initialize the parameter configuration of the flight intention model; a variety of distance measurement methods and kernel methods are used in combination with twin network learning to calculate the similarity between the flight data representation and the flight intention model, and the degree of matching between the flight data representation and the flight intention model is quantified to generate a similarity measurement result; according to the similarity measurement result combined with the initialization parameter configuration, the gradient descent algorithm and the reinforcement learning mechanism are applied to determine the optimal parameter configuration from the parameters of the flight intention model; based on the optimal parameter configuration, the behavior pattern configuration is generated using the situational awareness reasoning mechanism. Optionally, the method is to determine the optimal parameter configuration from the parameters of the flight intention model based on the similarity measurement result in combination with the initialization parameter configuration, applying the gradient descent algorithm and the reinforcement learning mechanism, including: setting an initial loss function based on the similarity measurement result and the initialization parameter configuration, and applying the gradient descent algorithm, by calculating the gradient of the initial loss function with respect to the initialization parameter configuration, and adopting an adaptive learning rate strategy to obtain the initial parameters of the flight intention model; using the reinforcement learning mechanism, defining a reward function, dynamically adjusting the initial parameters of the flight intention model based on the reward function combined with actual flight data feedback, and generating an intermediate parameter configuration; evaluating the intermediate parameter configuration based on the flight data to obtain a verification result, and adjusting the weights in the loss function and the reward function based on the verification result to generate a closed-loop feedback mechanism, and determining the optimal parameter configuration based on the closed-loop feedback mechanism.

[0049] In this embodiment, multiple flight intention models are constructed by mapping flight intention to different parts of a multi-dimensional behavior expectation model. These models are initialized with parameter configurations using the Bayesian optimization method and the Gaussian process to ensure that the starting point of the model has a high degree of accuracy. Subsequently, a variety of distance measurement methods and kernel methods are combined with twin network learning to calculate the similarity between the flight data representation and the flight intention model, and to quantify the degree of matching. Based on the similarity measurement results and the initialization parameter configuration, the gradient descent method and the reinforcement learning mechanism are used to determine the optimal parameter configuration from the parameters of the flight intention model, and finally the situational awareness reasoning mechanism is used to generate the behavior pattern configuration. This method not only improves the accuracy of flight intention recognition, but also enhances the effectiveness of parameter optimization.

[0050] In an embodiment of the present application, first, the system maps the flight intention of the behavior pattern to different parts of the multi-dimensional behavior expectation model to construct multiple flight intention models. Then, the Bayesian optimization method is used to select the best initial parameter combination from the pre-trained data set, and the Gaussian process is used to initialize the parameter configuration of the flight intention model. Next, a variety of distance measurement methods and kernel methods are combined with twin network learning to calculate the similarity between the flight data representation and the flight intention model, and generate a similarity measurement result. Based on the similarity measurement results and the initialization parameter configuration, the gradient descent algorithm and reinforcement learning mechanism are applied to determine the optimal parameter configuration. Finally, based on the optimal parameter configuration, the behavior pattern configuration is generated using the situational awareness reasoning mechanism.

[0051] Here is a specific example: In an urban fire rescue project, drones are used to monitor the fire scene in real time and guide rescue operations. In step 103, the drone first collects various data of the fire scene through its sensors, including temperature, smoke concentration, obstacle location, etc. The system inputs these raw data into the deep belief network, dynamically adjusts the weights through nonlinear transformation processing and attention mechanism, and obtains the flight data representation.

[0052] Next, the system maps the flight data representation to different parts of the multi-dimensional behavior expectation model and constructs multiple flight intention models, such as bypassing high temperature areas, avoiding smoke areas, or quickly reaching a designated location. The Bayesian optimization method is used to select the best initial parameter combination from the pre-trained data set, and the Gaussian process is used to initialize the parameter configuration of the flight intention model. Subsequently, the system uses a variety of distance measurement methods and kernel methods combined with twin network learning to calculate the similarity between the flight data representation and the flight intention model, quantify the degree of matching, and generate similarity measurement results. Based on the similarity measurement results and the initialization parameter configuration, the gradient descent algorithm and reinforcement learning mechanism are applied to determine the optimal parameter configuration. In this process, the system first sets the initial loss function and applies the gradient descent algorithm. By calculating the gradient of the initial loss function for the initialization parameter configuration, the adaptive learning rate strategy is used to obtain the initial parameters of the flight intention model. Then, the reinforcement learning mechanism is used to define the reward function, and the initial parameters of the flight intention model are dynamically adjusted based on the reward function combined with the actual flight data feedback to generate the intermediate parameter configuration.

[0053] The system evaluates the intermediate parameter configuration based on the flight data, obtains the verification results, and adjusts the weights in the loss function and reward function based on the verification results to generate a closed-loop feedback mechanism. Based on the closed-loop feedback mechanism, the system finally determines the optimal parameter configuration and generates the behavior pattern configuration using the situational awareness reasoning mechanism. In actual operation, the system selects the safest and most efficient flight path, enabling the drone to quickly reach the core area of ​​the fire, obtain key information, and provide real-time feedback to the ground command center. In this way, not only the mission execution efficiency of the drone is improved, but also the overall effect of the emergency response is significantly enhanced, ensuring the safety of personnel and property.

[0054] In order to further improve the accuracy of the UAV flight intention prediction and the effectiveness of the matching evaluation, in some embodiments, the step 104 is based on the probability distribution result, and the deviation measure between the simulated behavior performance and the actual flight activity is compared through a Markov decision process to evaluate and predict the UAV's flight intention, and obtain the optimal flight intention prediction, including: Based on the probability distribution result, a state transition model is constructed; according to the state transition model, in combination with a preset behavior pattern library and airspace usage rules, a Markov decision process model is initialized, a reward function is defined, and based on the reward function, an initial Markov decision process configuration is generated; by comparing the deviation measurement between the simulated behavior performance and the actual flight activity, a series of deviation measurement results are obtained; based on the deviation measurement results and the initial Markov decision process configuration, a dynamic time warping algorithm is applied to evaluate the match between the simulated behavior performance and the actual flight activity, and a match evaluation result is generated; according to the match evaluation result, the flight intention corresponding to the simulated behavior performance of the actual flight activity is selected as the prediction result, and the optimal flight intention prediction is obtained. Optionally, based on the deviation measurement result and the initial Markov decision process configuration, a dynamic time warping algorithm is applied to evaluate the match between the simulated behavior performance and the actual flight activity to generate a match evaluation result, including: preprocessing the simulated behavior performance according to the deviation measurement result in combination with the initial Markov decision process configuration to obtain a preprocessed simulated behavior performance; determining the parameter configuration of the dynamic time warping algorithm based on the characteristics of the preprocessed simulated behavior performance and the deviation measurement result; calculating the time series match between the simulated behavior performance and the actual flight activity using the parameter configuration of the dynamic time warping algorithm to obtain a time series match result; quantifying the similarity between the simulated behavior performance and the actual flight activity according to the time series match result to generate a match score; evaluating the match of the simulated behavior performance based on the match score in combination with the Markov decision process configuration to generate a match evaluation result.

[0055] In this embodiment, this embodiment constructs a state transition model based on the probability distribution results, and initializes the MDP in combination with the preset behavior pattern library and airspace usage rules. The MDP model guides the prediction of flight intention by defining a reward function. Subsequently, a series of deviation measurement results are generated by comparing the deviation measurement between the simulated behavior performance and the actual flight activity. Using the Dynamic Time Warping (DTW) algorithm, the system can accurately evaluate the match between the simulated behavior performance and the actual flight activity, and finally generate a match evaluation result, and select the flight intention that best matches the actual flight activity as the prediction result.

[0056] In an embodiment of the present application, first, a state transition model is constructed based on the probability distribution results, and the Markov decision process model is initialized in combination with the preset behavior pattern library and airspace usage rules. After defining the reward function, an initial MDP configuration is generated. Next, the system compares the deviation measurement between the simulated behavior performance and the actual flight activity to obtain a series of deviation measurement results. Then, based on these deviation measurement results and the initial MDP configuration, the dynamic time warping algorithm is applied to pre-process the simulated behavior performance, and the DTW parameter configuration is determined according to its characteristics. Next, the DTW algorithm is used to calculate the time series matching degree between the simulated behavior performance and the actual flight activity to obtain the time series matching degree result. Finally, the similarity between the two is quantified according to the time series matching degree result, a matching degree score is generated, and the matching degree of the simulated behavior performance is evaluated based on this score combined with the MDP configuration to generate the final matching degree evaluation result.

[0057] Here is a specific example: In an urban logistics distribution project, drones are used to deliver packages from warehouses to customers. In step 104, the drones collect a large amount of flight data, including location, speed, heading, and other information. The system first builds a state transition model based on the probability distribution results generated in the previous steps, and initializes the Markov decision process (MDP) model in combination with the preset behavior pattern library and airspace usage rules. Define the reward function, such as positive rewards for successfully completing the task, negative rewards for encountering obstacles or deviating from the path, and generate the initial MDP configuration.

[0058] Next, the system obtains a series of deviation measurement results by comparing the deviation measurement between the simulated behavior performance and the actual flight activities (such as position deviation, speed change, response time difference). Based on these deviation measurement results and the initial MDP configuration, the system applies the dynamic time warping algorithm (DTW) to preprocess the simulated behavior performance to ensure the consistency and accuracy of the data. According to the simulated behavior performance and its characteristics after preprocessing, the system determines the parameter configuration of the DTW algorithm.

[0059] Then, using these parameter configurations, the system calculated the time series matching between the simulated behavior performance and the actual flight activities, and obtained the time series matching results. Based on these matching results, the system quantified the similarity between the simulated behavior performance and the actual flight activities, and generated a matching score. Based on the matching score combined with the MDP configuration, the system evaluated the matching of the simulated behavior performance and generated the final matching evaluation result. In actual operation, the system selected the flight intention corresponding to the simulated behavior performance that best matches the actual flight activities as the prediction result, thereby optimizing the UAV's mission execution path. This not only improves the UAV's mission execution efficiency, but also significantly enhances its ability to cope with complex environments, ensuring the safe and efficient delivery of packages. In this way, the entire system achieves more accurate flight intention prediction and improves the overall performance and safety of UAV operations.

[0060] In order to further improve the accuracy of drone flight behavior recognition and the effectiveness of safety assessment, in some embodiments, the flight data described in step 102 is used to build a multi-dimensional behavior expectation model of the drone according to a preset behavior pattern library and airspace usage rules, including: The flight data is subjected to feature extraction processing to obtain a set of key feature sets; according to a preset behavior pattern library, the key feature set is compared with the flight behavior pattern in the preset behavior pattern library to identify the pattern related to the UAV flight behavior and generate a behavior pattern result; using the behavior pattern result, applying the airspace use rules, evaluating whether the UAV flight behavior complies with the predicted aviation management regulations and safety requirements, and obtaining the flight behavior pattern; according to the flight behavior pattern, constructing a multi-dimensional behavior expectation model.

[0061] In this embodiment, a set of key features is extracted from the original flight data through feature extraction processing. These features may include position, speed, heading, etc. Then, these key features are compared with the flight behavior patterns in the preset behavior pattern library to identify the pattern that best matches the current UAV flight behavior. Based on these behavior pattern results and combined with airspace usage rules (such as no-fly zones, route restrictions, etc.), the system can evaluate whether the UAV flight behavior complies with aviation management regulations and safety requirements. Finally, a multi-dimensional behavior expectation model is constructed based on the evaluation results for subsequent flight intention prediction and operation suggestion generation.

[0062] In the embodiment of the present application, first, the system performs feature extraction processing on the flight data to extract a set of key feature sets. Next, these key features are compared with the flight behavior patterns in the preset behavior pattern library to identify patterns related to the UAV flight behavior and generate behavior pattern results. Subsequently, using these behavior pattern results, the system applies airspace usage rules to evaluate whether the UAV flight behavior complies with the predicted aviation management regulations and safety requirements, and obtains the flight behavior pattern. Finally, based on these flight behavior patterns, the system constructs a multi-dimensional behavior expectation model to provide a basis for subsequent flight intention identification and operation recommendations.

[0063] Here is a specific example: In an urban logistics distribution project, drones are used to deliver packages from warehouses to customers. In step 102, the drone first collects a large amount of flight data, including information such as position, speed, and heading. The system performs feature extraction on the flight data and extracts a set of key feature sets, such as the distance of the current position relative to the target point, the current flight speed, and the position of obstacles in the surrounding environment.

[0064] Next, the system compares these key features with the flight behavior patterns in the preset behavior pattern library. The behavior pattern library contains a variety of typical flight behavior patterns, such as cruising, hovering, obstacle avoidance, etc. Through comparison, the system identifies the pattern that best matches the current drone flight behavior, such as the behavior pattern of the drone slowing down and preparing to land when approaching the target point, and generates a behavior pattern result. Then, the system uses these behavior pattern results, combined with the airspace use rules, to evaluate whether the drone's flight behavior complies with the foreseen aviation management regulations and safety requirements. For example, the system checks whether the drone has entered a no-fly zone or followed a prescribed route. If any non-compliant behavior is found, the system will issue a warning and recommend adjusting the flight path.

[0065] Finally, based on these flight behavior patterns, the system built a multi-dimensional behavior expectation model. This model not only takes into account the basic operating characteristics of the drone, but also incorporates external environmental constraints, such as no-fly zones and severe weather areas, to ensure the safe and smooth progress of the mission. In actual operation, the model helped the drone choose the best flight path, avoided potential risks, and ensured the safe and efficient delivery of the package. In addition, the model also provided a solid foundation for subsequent flight intention identification and operation recommendations, improving the reliability and flexibility of the entire system. In this way, drones can achieve smarter and safer operations in complex urban environments.

[0066] This application considers that in the field of UAV flight intention recognition and behavior prediction, accurate analysis of the UAV's flight intention is crucial to ensure its safe and efficient operation. However, traditional flight intention recognition methods often face problems such as high data complexity, inaccurate feature extraction, and poor model adaptability. Therefore, a new optional solution is proposed, which includes: The deep belief network is used to map the flight data to the multi-dimensional behavior expectation model to calculate the probability distribution of the drone under different flight intentions, generate simulated behavior performance and corresponding probability distribution results, including: The flight data is processed nonlinearly using a deep belief network, and the weight of the flight data is dynamically adjusted using an attention mechanism to obtain a flight data representation. The weight of each feature in the flight data is calculated using the following formula: : ; in, represents the weight of each feature in the flight data, is the temperature parameter, is the number of features, and is the regulating factor, Representation characteristics time delay or spatial distance, and are the weight matrix and bias term in the attention mechanism, respectively. It is feature vectors, and is an additional tuning parameter; The following is a detailed explanation of each parameter: Indicates the flight data The weight of a feature reflects the importance of the feature in identifying the flight intention of the drone.

[0067] and are the weight matrix and bias term in the attention mechanism. These parameters are obtained through training and are used to adjust the importance of each feature. They are designed to enable the model to adaptively focus on features that are more important to the current task.

[0068] Indicates This is one of the key features extracted from the raw flight data, such as position, speed, etc.

[0069] Represents the temperature parameter, which is used to control the smoothness of the softmax function. Values ​​of will make all weights equal, while lower Smaller values ​​increase the variance between the weights. The optimal value is usually determined through experimentation or cross-validation.

[0070] and represents an additional tuning parameter. is a global adjustment factor, and These are adjustment parameters for each feature. These parameters are set through training or experience to ensure that the model can flexibly adjust feature weights in different situations.

[0071] and Represents the adjustment factor, used to control time delay or spatial distance Impact on feature weights. These parameters are tuned experimentally or based on domain knowledge to optimize model performance.

[0072] Representation characteristics This reflects the temporal or spatial characteristics of the feature and helps the model understand the temporal dynamics or spatial distribution of the feature.

[0073] The following is an introduction to the design reasons of each sub-item: This is a standard linear transformation used to initially adjust the importance of features. The weight matrix learned through training and the bias term It can help the model better capture the relationship between features and targets.

[0074] This part of the design is to introduce nonlinear components so that the model can handle more complex feature dependencies. Using the hyperbolic sine function (sinh) can enhance the expressiveness of the model and enable it to capture subtle differences between features.

[0075] This part is used to consider the influence of the time delay or spatial distance of the feature on the weight. By introducing a combination of logarithmic function and exponential function, the influence of the time or spatial characteristics of the feature on the weight can be effectively balanced. This method can prevent some features from having too low weight due to excessive time delay or spatial distance.

[0076] The purpose of summing up the sub-items is to combine them together in an additive way, mainly to comprehensively consider the impact of multiple factors on the feature weight. Linear transformations, nonlinear components, and time or space effects are all considered independent but related factors, which together determine the importance of the feature.

[0077] mapping the flight data representation to different portions of the multi-dimensional behavior expectation model to identify behavior patterns that match the flight data representation; Based on the behavior pattern, the flight intention model of the UAV in the multi-dimensional behavior expectation model is constructed, and the similarity between the flight intention models is evaluated by applying the similarity metric learning method to determine the optimal parameter configuration of the flight intention model and generate the behavior pattern configuration. The similarity between the flight data representation and the behavior pattern is evaluated by the following similarity metric formula : ; in, represents the UAV flight data representation vector, Indicates The vector representation of the behavior pattern, represents the Euclidean norm, express The transpose of Representation vector The Euclidean norm of , Representation vector The Euclidean norm of , is the regularization parameter, express and The Kullback-Leibler divergence between the probability distributions of ; The following is a detailed explanation of each parameter: Represents the UAV flight data representation vector With Vector representation of behavior patterns The higher the value, the more similar the two are.

[0078] and Here is a representation vector of UAV flight data, which is usually extracted from the raw flight data through a deep learning model such as a deep belief network. It is a behavior pattern vector in the preset behavior pattern library. These patterns are usually built based on historical data or expert knowledge and represent the typical behavior of the drone in a specific situation.

[0079] This is a vector and The dot product (inner product) is used to measure the consistency of the directions of two vectors. The larger the dot product, the closer the directions of the two vectors are.

[0080] and These two symbols represent vectors. and The Euclidean norm of (i.e., the length of the vector). The norm is used to normalize the dot product result so that the similarity measure is not affected by the length of the vector.

[0081] Represents the regularization parameter, which is used to balance the weights of the dot product part and the Kullback-Leibler divergence (KL divergence) part. This parameter is usually set through cross-validation or experience.

[0082] This is and The Kullback-Leibler divergence between the probability distributions of is used to measure the difference between two distributions. KL divergence is asymmetric, indicating that the distribution To distribution Relative abstract.

[0083] The following is an introduction to the design reasons of each sub-item: This is the formula for calculating cosine similarity. Cosine similarity measures the angular difference between two vectors, rather than their absolute distance. This metric is well suited for vector comparisons in high-dimensional spaces because it ignores the magnitude of the vector and only focuses on the direction. The reason for the design is that using cosine similarity can effectively capture the directional consistency between flight data representation and behavior patterns, which is very important for identifying the flight intention of a drone.

[0084] The KL divergence is a measure of the difference between two probability distributions. Here, it is used to quantify the probability distribution represented by the flight data. Probability distribution of behavior patterns The reason for this design is that although cosine similarity can measure the directional consistency of vectors well, it cannot capture the subtle differences between probability distributions. Introducing KL divergence can help the system more comprehensively evaluate the similarity between two distributions, especially when dealing with scenarios with high uncertainty.

[0085] The purpose of adding the sub-items together is to comprehensively consider two different similarity measurement methods. Cosine similarity focuses on the directional consistency of vectors, while KL divergence focuses on the differences between probability distributions. Adding these two parts can provide a more comprehensive and robust similarity measurement method.

[0086] Here is a specific example: In a city logistics distribution project, suppose we have the following specific values: Flight data representation vector , behavior pattern vector , the regularization parameter .

[0087] Calculation steps: Calculate cosine similarity: Dot Product ; Euclidean norm ; Euclidean norm ; Cosine similarity ; Calculating KL divergence: Assumptions and The probability distributions are and , and calculate Since the specific probability distribution needs to be defined according to the actual application background, it is assumed here that .

[0088] Calculate the final similarity: Final Similarity .

[0089] From the calculation results, we can see that the value close to 1 indicates that the flight data represents the vector Vector with behavior pattern The closeness in direction means that the current flight data is highly consistent with the preset behavior pattern. The small KL divergence value indicates that the difference between the two probability distributions is not large, which further confirms the consistency of the flight data and the behavior pattern. The high final similarity of 0.9805 indicates that the current flight data matches the behavior pattern well, and the system can accurately identify the flight intention of the drone and make corresponding operational recommendations accordingly. In this way, the system can not only accurately identify the flight intention of the drone, but also dynamically adjust the flight strategy according to real-time data to ensure the safe and efficient completion of the mission.

[0090] According to the above behavior pattern configuration combined with the Bayesian reasoning framework, the probability distribution of the drone executing each flight intention is calculated, and the variational inference technology is used to obtain the probability distribution result. The probability distribution result is calculated by the following Bayesian formula: ; in, Indicates flight intention, is the mixing coefficient, and Flight Intent The mean vector and covariance matrix of the flight data are represented below, is the number of ingredients in the mixture, is the data dimension, represents the prior probability distribution, is the total number of flight intentions; The following is a detailed explanation of each parameter: Indicates that given flight data In the case of Flight Intent This is the core output of the Bayesian formula and is used to identify the most likely flight intention.

[0091] The representation vector representing the UAV flight data is usually extracted from the raw flight data through a deep learning model such as a deep belief network.

[0092] Indicates flight intentions, such as cruising, hovering or obstacle avoidance.

[0093] Indicates the number of mixture components, indicating that each flight intention can be a mixture of multiple Gaussian distributions. This allows the model to adapt to complex data distributions more flexibly.

[0094] Represents the mixing coefficient, which represents the weight of each mixture component in the overall model. These coefficients are usually estimated through the training process and satisfy .

[0095] and They are The flight intention The mean vector and covariance matrix of the mixture components. These parameters describe the central location of each mixture component and the shape of the data distribution. They are also estimated during the training process.

[0096] Represents the data dimension, that is, the feature vector Length.

[0097] represents the prior probability distribution, which represents the initial estimate of each flight intention before any data is observed. These probabilities can be obtained through historical data or expert knowledge.

[0098] represents the likelihood function, which means that under the assumption that the drone performs the In the case of flight intention, the flight data is observed probability.

[0099] The following is an introduction to the design reasons of each sub-item: This is the core part of the probability density function of the multidimensional Gaussian distribution, which measures the data points Relative to the mean vector The smaller the index part, the and The design reason is that using Gaussian distribution can effectively model the probability distribution of continuous data, especially in the presence of noise. This method is very suitable for dealing with uncertainties in flight data.

[0100] This is the standard normalization factor of the multidimensional Gaussian distribution, which ensures that the integral of the probability density function over the entire space is equal to 1. The design reason is that the normalization factor ensures the validity of the probability density function, so that it can correctly reflect the probability distribution of the data points.

[0101] is the mixing coefficient, which is used to balance the contributions of different mixture components. The importance of each component is determined by its corresponding mixing coefficient. The reason for the design is that the introduction of the mixing coefficient enables the model to better fit the complex actual data distribution, rather than being limited to a single Gaussian distribution.

[0102] It represents the prior probability, which reflects the initial estimation of each flight intention before any data is observed. The design reason is that the prior probability provides important background information to help the system make reasonable judgments when the data is insufficient or ambiguous.

[0103] This is the denominator of the Bayesian formula, which is used to normalize the posterior probabilities to ensure that the sum of all posterior probabilities is 1. The design reason is that the normalization factor ensures the validity and consistency of the posterior probabilities, allowing them to be directly used for comparison and decision-making.

[0104] Based on the probability distribution results In combination with the behavior pattern configuration, the Monte Carlo method is used to generate a series of simulated behavior performances and corresponding probability distribution results.

[0105] Here is a specific example: In a city logistics distribution project, suppose we have the following specific values: Flight data representation vector , flight intention , respectively cruise and hover , the number of mixed ingredients , data dimension .

[0106] Calculation steps: Compute the probability density function of each mixture component: Assume that the flight intention , the mean vector and covariance matrix of the mixture components are: ; The mixing coefficients are ; For flight intention , assuming similar parameter settings.

[0107] Calculate the likelihood function for each flight intention: Using the formula Compute the probability density function of each mixture component.

[0108] Compute the posterior probability: The likelihood function of each flight intention is combined with the prior probability Combined, substitute into the Bayesian formula to calculate the posterior probability .

[0109] Example calculation: Assume that after calculation, we get the following results: For flight intention , the weighted sum of the likelihood functions is 0.75.

[0110] For flight intention , the weighted sum of the likelihood functions is 0.25.

[0111] The final posterior probabilities are: ; ; Assume prior probability ,but: ; ; From the calculation results, it can be seen that the flight intention The posterior probability of (cruise) is significantly higher than the flight intention (Hovering). This means that under the current flight data, the system is more inclined to believe that the drone is performing cruising behavior. In this way, the system can not only accurately identify the flight intention of the drone, but also dynamically adjust the flight strategy based on real-time data to ensure the safe and efficient completion of the mission.

[0112] Figure 2 A schematic diagram of a UAV flight intention recognition system is provided for an embodiment of the present application. Figure 2 As shown, the system includes: A receiving module 21 is used to receive flight data from the UAV, wherein the flight data includes flight parameters and environment perception data after time-space consistency calibration processing; A construction module 22 is used to construct a multi-dimensional behavior expectation model of the UAV based on the flight data and a preset behavior pattern library and airspace use rules; The calculation module 23 is used to map the flight data into the multi-dimensional behavior expectation model using a deep belief network to calculate the probability distribution of the UAV under different flight intentions and generate simulated behavior performance and corresponding probability distribution results; A comparison module 24 is used to compare the deviation metric between the simulated behavior performance and the actual flight activity based on the probability distribution result through a Markov decision process to evaluate and predict the flight intention of the UAV and obtain an optimal flight intention prediction, wherein the deviation metric includes position deviation, speed change, and time difference of interactive response to the environment; The generation module 25 is used to generate operation suggestions for the drone according to the flight intention and the immediate airspace safety status, send the operation suggestions to the drone, obtain feedback information after the drone executes the operation, and generate flight guidance suggestions based on the feedback information.

[0113] Figure 2 The UAV flight intention recognition system can perform Figure 1 The implementation principle and technical effect of the method for identifying the flight intention of a drone described in the embodiment are not described in detail. The specific manner in which each module and unit performs operations in the drone flight intention identification system in the above embodiment has been described in detail in the embodiment of the method, and will not be described in detail here.

[0114] In one possible design, Figure 2 A drone flight intention recognition system of the illustrated embodiment can be implemented as a computing device, such as Figure 3As shown, the computing device may include a storage component 31 and a processing component 32; The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0115] The processing component 32 is used for the above Figure 1 The embodiment provides a method for identifying the flight intention of a UAV.

[0116] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0117] The storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0118] Of course, the computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0119] The input / output interface provides an interface between the processing component and the peripheral interface module, which may be an output device, an input device, etc.

[0120] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0121] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0122] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 A method for identifying the flight intention of a UAV in the illustrated embodiment.

[0123] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0124] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0125] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for identifying the flight intention of a UAV, characterized in that: include: Receiving flight data from the UAV, wherein the flight data includes flight parameters and environmental perception data after being calibrated for spatiotemporal consistency; Using the flight data, a multi-dimensional behavior expectation model of the UAV is constructed according to a preset behavior pattern library and airspace use rules; Using a deep belief network, the flight data is mapped to the multi-dimensional behavior expectation model to calculate the probability distribution of the UAV under different flight intentions, and generate simulated behavior performance and corresponding probability distribution results; Based on the probability distribution result, the deviation metric between the simulated behavior performance and the actual flight activity is compared through a Markov decision process to evaluate and predict the flight intention of the UAV and obtain an optimal flight intention prediction, wherein the deviation metric includes position deviation, speed change, and time difference in response to environmental interaction; According to the flight intention and in combination with the immediate airspace safety status, an operation suggestion for the UAV is generated, and the operation suggestion is sent to the UAV to obtain feedback information after the UAV executes the operation, and based on the feedback information, a flight guidance suggestion is generated.

2. The method according to claim 1, characterized in that The deep belief network is used to map the flight data to the multi-dimensional behavior expectation model to calculate the probability distribution of the drone under different flight intentions, generate simulated behavior performance and corresponding probability distribution results, including: Using a deep belief network, performing nonlinear transformation processing on the flight data, and using an attention mechanism to dynamically adjust the weight of the flight data to obtain a flight data representation; mapping the flight data representation to different portions of the multi-dimensional behavior expectation model to identify behavior patterns that match the flight data representation; Based on the behavior pattern, a flight intention model of the UAV in the multi-dimensional behavior expectation model is constructed, and a similarity metric learning method is applied to evaluate the similarity between the flight intention models to determine the optimal parameter configuration of the flight intention model and generate a behavior pattern configuration; According to the behavior pattern configuration combined with the Bayesian reasoning framework, the probability distribution of the drone executing each flight intention is calculated, and the variational inference technique is used to obtain the probability distribution result; Based on the probability distribution results and the behavior pattern configuration, a Monte Carlo method is used to generate a series of simulated behavior performances and corresponding probability distribution results.

3. The method according to claim 2, characterized in that Based on the behavior pattern, the flight intention model of the UAV in the multi-dimensional behavior expectation model is constructed, and a similarity metric learning method is applied to evaluate the similarity between the flight intention models to determine the optimal parameter configuration of the flight intention model and generate a behavior pattern configuration, including: Mapping the flight intention of the behavior pattern to different parts of the multi-dimensional behavior expectation model to construct multiple flight intention models; Using the Bayesian optimization method, the best initial parameter combination is selected from the flight behavior patterns in the pre-training data set, and the Gaussian process is used in combination with the best initial parameter combination to initialize the parameter configuration of the flight intention model; Utilizing a variety of distance measurement methods and kernel methods combined with twin network learning, calculating the similarity between the flight data representation and the flight intention model, and quantifying the degree of matching between the flight data representation and the flight intention model, and generating a similarity measurement result; According to the similarity measurement result combined with the initialization parameter configuration, a gradient descent algorithm and a reinforcement learning mechanism are applied to determine the optimal parameter configuration from the parameters of the flight intention model; Based on the optimal parameter configuration, a behavior pattern configuration is generated using a situational awareness reasoning mechanism.

4. The method according to claim 3, characterized in that The method of determining the optimal parameter configuration from the parameters of the flight intention model by applying a gradient descent algorithm and a reinforcement learning mechanism based on the similarity measurement result and the initialization parameter configuration includes: According to the similarity measurement result and the initialization parameter configuration, an initial loss function is set, and a gradient descent algorithm is applied to obtain initial parameters of the flight intention model by calculating the gradient of the initial loss function with respect to the initialization parameter configuration and adopting an adaptive learning rate strategy; Using a reinforcement learning mechanism, a reward function is defined, and based on the reward function combined with actual flight data feedback, initial parameters of the flight intention model are dynamically adjusted to generate an intermediate parameter configuration; According to the flight data, the intermediate parameter configuration is evaluated to obtain a verification result, and the weights in the loss function and the reward function are adjusted according to the verification result to generate a closed-loop feedback mechanism, and based on the closed-loop feedback mechanism, the optimal parameter configuration is determined.

5. The method according to claim 1, characterized in that Based on the probability distribution result, the deviation measure between the simulated behavior performance and the actual flight activity is compared through a Markov decision process to evaluate and predict the flight intention of the UAV and obtain the optimal flight intention prediction, including: Based on the probability distribution result, a state transition model is constructed; according to the state transition model, in combination with a preset behavior pattern library and airspace usage rules, a Markov decision process model is initialized, a reward function is defined, and based on the reward function, an initial Markov decision process configuration is generated; By comparing the deviation measures between the simulated behavior performance and the actual flight activities, a series of deviation measure results are obtained; Based on the deviation measurement result and the initial Markov decision process configuration, applying a dynamic time warping algorithm to evaluate the match between the simulated behavior performance and the actual flight activity, and generating a match evaluation result; According to the matching evaluation result, the flight intention corresponding to the simulated behavior performance of the actual flight activity is selected as the prediction result to obtain the optimal flight intention prediction.

6. The method according to claim 5, characterized in that The step of applying a dynamic time warping algorithm based on the deviation measurement result and the initial Markov decision process configuration to evaluate the matching degree between the simulated behavior performance and the actual flight activity and generating a matching degree evaluation result includes: Preprocessing the simulated behavior performance according to the deviation measurement result and in combination with the initial Markov decision process configuration to obtain the preprocessed simulated behavior performance; Determining the parameter configuration of the dynamic time warping algorithm based on the characteristics of the simulated behavior performance after the preprocessing and the deviation measurement result; Using the parameter configuration of the dynamic time warping algorithm, the time series matching degree between the simulated behavior performance and the actual flight activity is calculated to obtain a time series matching degree result; quantifying the similarity between the simulated behavior performance and the actual flight activity according to the time series matching result, and generating a matching score; Based on the matching score combined with the Markov decision process configuration, the matching of the simulated behavior performance is evaluated to generate a matching evaluation result.

7. The method according to claim 1, characterized in that The method utilizes the flight data to construct a multi-dimensional behavior prediction model of the UAV according to a preset behavior pattern library and airspace usage rules, including: Performing feature extraction processing on the flight data to obtain a set of key feature sets; According to a preset behavior pattern library, the key feature set is compared with the flight behavior patterns in the preset behavior pattern library, the patterns related to the flight behavior of the UAV are identified, and the behavior pattern results are generated; Using the behavior pattern results, applying airspace use rules, evaluating whether the UAV flight behavior complies with the predicted aviation management regulations and safety requirements, and obtaining the flight behavior pattern; According to the flight behavior pattern, a multi-dimensional behavior expectation model is constructed.

8. A UAV flight intention recognition system, characterized in that: include: A receiving module, used to receive flight data from the UAV, wherein the flight data includes flight parameters and environmental perception data after time-space consistency calibration processing; A construction module, for constructing a multi-dimensional behavior expectation model of the UAV based on the flight data and a preset behavior pattern library and airspace use rules; A calculation module, used to map the flight data into the multi-dimensional behavior expectation model using a deep belief network, so as to calculate the probability distribution of the UAV under different flight intentions, and generate simulated behavior performance and corresponding probability distribution results; A comparison module, for comparing the deviation metric between the simulated behavior performance and the actual flight activity through a Markov decision process based on the probability distribution result, so as to evaluate and predict the flight intention of the UAV and obtain an optimal flight intention prediction, wherein the deviation metric includes position deviation, speed change, and time difference of interactive response to the environment; The generation module is used to generate operation suggestions for the drone according to the flight intention and the immediate airspace safety status, send the operation suggestions to the drone, obtain feedback information after the drone executes the operation, and generate flight guidance suggestions based on the feedback information.

9. A computing device, characterized in that It comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a method for identifying the flight intention of a drone as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a computer, a method for identifying the flight intention of a drone as described in any one of claims 1 to 7 is implemented.

Images