Explanatable unmanned aerial vehicle task decision method and device

Abstract

The invention discloses an interpretable unmanned aerial vehicle task decision method and device. The unmanned aerial vehicle task decision-making method comprises the steps of obtaining a first input data set for an unmanned aerial vehicle decision-making task; the first input data set is input into a pre-constructed first preset decision model, a first strategy data set used for outputting a decision result is obtained, and the first strategy data set comprises an adjustment scheme used for indicating a next action of the unmanned aerial vehicle for the target task; the first strategy data set and the first input data set are input into a feature analysis model based on an SHAP algorithm, a decision feature data set for the unmanned aerial vehicle decision task is obtained, and the decision feature data set comprises unmanned aerial vehicle decision feature data corresponding to the adjustment scheme; and constructing a second preset decision model based on a decision tree algorithm by using the decision feature data set to obtain decision tree data for presenting a decision path and a decision basis for the adjustment scheme, and taking the decision tree data as an explanation data set for explaining the unmanned aerial vehicle decision task.

Description

Technical Field

[0001] The present disclosure generally relates to the fields of artificial intelligence algorithm technology and natural language processing technology, and more specifically, to an explainable drone mission decision-making method and device. Background Art

[0002] In related fields, with the rapid development of artificial intelligence related technologies, the processing of various intelligent tasks has gradually become a hot issue. Some intelligent decision-making methods and systems have been proposed for the processing of drone tasks in the field of drones. Traditional intelligent decision-making systems mainly rely on rule-based systems and expert systems to make decisions through predefined rule sets and logic sets. Although these decision-making systems have strong interpretability and show strong adaptability in some specific scenarios, due to their lack of adaptability and flexibility, traditional decision-making systems are difficult to cope with complex environments such as drone mission decisions.

[0003] For decision-making on UAV missions involving complex environments, there is an urgent need to study the explainability of such UAV mission-related decisions, and an improved UAV mission-related decision-making method is needed. Summary of the invention

[0004] The embodiments of the present disclosure provide an explainable UAV mission decision method and device, which provide a comprehensive explanation of the decision-making process of complex UAV adjustment decision schemes by utilizing the SHAP algorithm to perform feature analysis on dynamic decision results for UAV mission data including time series and dynamic environments.

[0005] In one general aspect, an explainable UAV task decision method is provided, the UAV task decision method comprising: obtaining a first input data set for a UAV decision task, the first input data set being UAV task data based on a time series and a dynamic environment stored in a table form, the UAV task data comprising position data for a target task, position data of at least one obstacle related to the target task, current route data related to the target task, and time data related to the target task; inputting the first input data set into a pre-built first preset decision model to obtain a first strategy data set for outputting a decision result, the first strategy data set comprising an adjustment scheme for indicating the next action of the UAV for the target task; inputting the first strategy data set and the first input data set into a feature analysis model based on a SHAP algorithm to obtain a decision feature data set for the UAV decision task, the decision feature data set comprising UAV decision feature data corresponding to the adjustment scheme; using the decision feature data set to construct a second preset decision model based on a decision tree algorithm, to obtain decision tree data for presenting a decision path and decision basis for the adjustment scheme as an explanation data set for explaining the UAV decision task.

[0006] Optionally, the step of inputting the first strategy data set and the first input data set into a feature analysis model based on the SHAP algorithm to obtain a decision feature data set for the UAV decision-making task may include: for each strategy in the first strategy data set, extracting current input data corresponding to each strategy from the first input data set; performing feature analysis based on the SHAP value on the current input data of each strategy, and performing feature extraction based on the marginal contribution of the analyzed different features in the contexts corresponding to different strategies, to obtain a feature set with the greatest degree of influence on the current strategy of each strategy, as the decision feature data set for the current input data of each strategy, and summarizing the decision feature data sets for the current input data of each strategy to obtain a decision feature data set for the UAV decision task.

[0007] Optionally, the marginal contribution corresponding to each feature included in the feature set that has the greatest impact on the current strategy of each strategy may be calculated by the following formula:

[0008] in, Represents the marginal contribution function corresponding to each feature, U represents the utility function obtained after the preset feature subsets cooperate, N represents the full set of features, n represents the number of feature sets, S represents the number of feature subsets,i Indicates the sequence number of the current feature, k is greater than or equal to 1 and less than or equal to n An integer.

[0009] Optionally, the construction process of the second preset decision model may include: selecting multiple features from the decision feature data set; constructing each node of the decision tree according to preset evaluation parameters for the multiple features, until the construction process is terminated when a preset termination condition is met.

[0010] Optionally, the first preset decision model may include at least one of a decision model based on a reinforcement learning algorithm, a decision model based on a deep learning algorithm, and a decision model based on a Bayesian network algorithm.

[0011] Optionally, the first preset decision model may include a decision model based on a reinforcement learning algorithm, and the step of inputting the first input data set into the pre-built first preset decision model to obtain a first strategy data set for outputting a decision result may include: based on the first input data set, performing reinforcement learning operations including action, feedback, adjustment, and re-action to obtain a decision result that satisfies the conditions for maximizing a preset cumulative reward target as the first strategy data set.

[0012] Optionally, the first preset decision model may include a decision model based on a deep learning algorithm, and the first preset decision model may be trained in the following manner: obtaining a first sample data set for the UAV decision task, the first sample data set is sample UAV task data based on time series and dynamic environment stored in a tabular form, the sample UAV task data includes position data for the preset task, position data of at least one obstacle related to the preset task, current route data related to the preset task, and time data related to the preset task; based on the first sample data set, determining a corresponding sample strategy data set, the corresponding sample strategy data set includes an adjustment plan for indicating the next action of the UAV for the preset task; based on the first sample data set and the corresponding sample strategy data set, calculating a preset loss function; according to the preset loss function, adjusting the model parameters of the first preset decision model to obtain the trained first preset decision model.

[0013] In another general aspect, an explainable UAV task decision device is provided, the UAV task decision device comprising: a data acquisition module, configured to: acquire a first input data set for the UAV decision task, the first input data set is UAV task data based on time series and dynamic environment stored in a table form, the UAV task data includes position data for a target task, position data of at least one obstacle related to the target task, current route data related to the target task, and time data related to the target task; a decision generation module, configured to: input the first input data set into a pre-built first preset decision model to obtain a first strategy data set for outputting a decision result, the first strategy data set includes an adjustment scheme for indicating the next action of the UAV for the target task; a feature analysis module, configured to: input the first strategy data set and the first input data set into a feature analysis model based on a SHAP algorithm to obtain a decision feature data set for the UAV decision task, the decision feature data set includes UAV decision feature data corresponding to the adjustment scheme; a decision interpretation module, configured to: construct a second preset decision model based on a decision tree algorithm using the decision feature data set, and obtain decision tree data for presenting a decision path and decision reasons for the adjustment scheme as an explanation data set for interpreting the UAV decision task.

[0014] Optionally, the operation of inputting the first strategy data set and the first input data set into a feature analysis model based on the SHAP algorithm to obtain a decision feature data set for the UAV decision-making task may include: for each strategy in the first strategy data set, extracting current input data corresponding to each strategy from the first input data set; performing feature analysis based on the SHAP value on the current input data of each strategy, and performing feature extraction based on the marginal contribution of the analyzed different features in the contexts corresponding to different strategies, to obtain a feature set with the greatest degree of influence on the current strategy of each strategy, as the decision feature data set for the current input data of each strategy, and summarizing the decision feature data sets for the current input data of each strategy to obtain a decision feature data set for the UAV decision task.

[0015] Optionally, the marginal contribution corresponding to each feature included in the feature set that has the greatest impact on the current strategy of each strategy may be calculated by the following formula:

[0016] in, Represents the marginal contribution function corresponding to each feature, U represents the utility function obtained after the preset feature subsets cooperate, N represents the full set of features,n represents the number of feature sets, S represents the number of feature subsets, i Indicates the sequence number of the current feature, k is greater than or equal to 1 and less than or equal to n An integer.

[0017] Optionally, the construction process of the second preset decision model may include: selecting multiple features from the decision feature data set; constructing each node of the decision tree according to preset evaluation parameters for the multiple features, until the construction process is terminated when a preset termination condition is met.

[0018] Optionally, the first preset decision model may include at least one of a decision model based on a reinforcement learning algorithm, a decision model based on a deep learning algorithm, and a decision model based on a Bayesian network algorithm.

[0019] Optionally, the first preset decision model may include a decision model based on a reinforcement learning algorithm, and the decision generation module inputs the first input data set into a pre-built first preset decision model, and the operation of obtaining a first strategy data set for outputting a decision result may include: based on the first input data set, performing reinforcement learning operations including action, feedback, adjustment, and re-action, to obtain a decision result that satisfies the conditions for maximizing the preset cumulative reward target as the first strategy data set.

[0020] Optionally, the first preset decision model may include a decision model based on a deep learning algorithm, and the first preset decision model may be trained in the following manner: obtaining a first sample data set for a drone decision task, the first sample data set being sample drone task data based on a time series and a dynamic environment stored in a table form, the sample drone task data including location data for the preset task, location data of at least one obstacle related to the preset task, current route data related to the preset task, and time data related to the preset task; based on the first sample data set, determining a corresponding sample strategy data set, the corresponding sample strategy data set including an adjustment scheme for indicating the next action of the drone for the preset task; based on the first sample data set and the corresponding sample strategy data set, calculating a preset loss function; according to the preset loss function, adjusting the model parameters of the first preset decision model to obtain the trained first preset decision model. In another general aspect, a computer program product is provided, the computer program product including a computer program / instruction, which can implement the above-described interpretable drone task decision method when executed by a processor.

[0021] In another general aspect, a computer-readable storage medium is provided, which, when instructions in the computer-readable storage medium are executed by at least one processor, enables the at least one processor to perform the explainable drone mission decision method as described above.

[0022] In another general aspect, a computing device is provided, comprising: at least one processor; and at least one memory storing computer executable instructions, wherein the computer executable instructions, when executed by the at least one processor, cause the at least one processor to execute the explainable UAV mission decision method as described above.

[0023] According to the explainable drone mission decision method and device of the embodiment of the present disclosure, by using the SHAP algorithm to perform feature analysis on the dynamic decision results for drone mission data including time series and dynamic environment, a comprehensive explanation of the decision process of complex drone decision schemes is provided. In addition, by using feature analysis based on the SHAP algorithm, the user can clearly know the impact of each drone decision feature corresponding to the drone adjustment scheme on the final drone decision, so as to understand the drone adjustment decision logic of the model, which helps to improve the trust in the model. In addition, by using a decision tree model to display the feature segmentation for drone decisions and the decision path for drone adjustment schemes, a detailed explanation of each drone adjustment decision point is provided, which is convenient for users to understand and trace the basis of each drone adjustment decision point. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The above and other objects and features of the embodiments of the present disclosure will become more apparent through the following description in conjunction with the accompanying drawings showing the embodiments, in which: Figure 1 is a flow chart illustrating an explainable drone mission decision method according to an embodiment of the present disclosure; Figure 2 is a block diagram showing an explainable drone mission decision device according to an embodiment of the present disclosure; Figure 3 is a block diagram illustrating a computing device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0025] The following specific embodiments are provided to help the reader obtain a comprehensive understanding of the methods, devices and / or systems described herein. However, after understanding the disclosure of the present application, various changes, modifications and equivalents of the methods, devices and / or systems described herein will be clear. For example, the order of operations described herein is only an example and is not limited to those orders set forth herein, but can be changed as will be clear after understanding the disclosure of the present application, except for operations that must occur in a specific order. In addition, for greater clarity and simplicity, the description of features known in the art may be omitted.

[0026] Reference will now be made in detail to the embodiments of the present disclosure, examples of which are illustrated in the accompanying drawings, wherein like reference numerals refer to like parts throughout. The embodiments will be described below with reference to the drawings in order to explain the present disclosure.

[0027] In the related art, as mentioned above, the traditional decision-making system is difficult to cope with the complex dynamic environment in the field of drones. At present, for the decision-making of drone tasks in this field including complex dynamic environments, it is urgent to realize the interpretability improvement method of the decision-making related to such drone tasks.

[0028] In order to solve the problems in the related technology, the present disclosure proposes an explainable UAV mission decision method, which can provide a comprehensive explanation of the decision-making process of complex UAV decision-making schemes by utilizing the SHAP algorithm to perform feature analysis on dynamic decision results for UAV mission data including time series and dynamic environments.

[0029] That is to say, the explainable UAV task decision method disclosed in the present invention extracts and analyzes features of a black box model (i.e., a model without interpretability) such as reinforcement learning decision-making through the SHAP algorithm, and uses these extracted features to regain the entire processing process of a white box model (a highly interpretable model) such as a decision tree-based model, thereby providing a comprehensive explanation of the decision-making process of complex UAV decision-making schemes related to time series and dynamic environments.

[0030] Refer to the following Figures 1 to 3 The present invention describes in detail an explainable UAV mission decision method and device according to an embodiment of the present invention.

[0031] Figure 1 is a flow chart illustrating an explainable drone mission decision method 100 according to an embodiment of the present disclosure.

[0032] Reference Figure 1According to an embodiment of the present disclosure, in step S101, a first input data set for a drone decision task is obtained. Here, the first input data set is drone task data based on a time series and a dynamic environment stored in a table form. For example, the drone task data includes position data for a target task, position data of at least one obstacle related to the target task, current route data related to the target task, and time data related to the target task.

[0033] As an example, the position data for a preset object may include coordinate data, for example, the position data for a target task may include target coordinate data, and the position data for at least one obstacle may include coordinate data for at least one obstacle. In addition, for example, the time data associated with the target task may include time data in different forms, for example, time representation data processed according to different rules. Here, it should be noted that the form of the data used as the model input is limited to a tabular form.

[0034] By executing the explainable decision-making method according to the present disclosure, which will be described in detail later, for UAV decision-making tasks that need to consider multiple factors (such as target position, dynamic position of obstacles, time series, etc.), the UAV decision-making system provides a decision basis for adjusting the UAV's execution action by analyzing input data including, for example, dynamic position coordinates and dynamic time series, which helps to formulate the best adjustment plan for the UAV for subsequent execution actions (for example, including performed operations and changes in routes).

[0035] According to an embodiment of the present disclosure, in step S102, a first input data set is input into a pre-built first preset decision model to obtain a first strategy data set for outputting a decision result. Here, the first strategy data set includes an adjustment plan for instructing the next action of the drone for the target task.

[0036] As an example, the first preset decision model may include at least one of a decision model based on a reinforcement learning algorithm, a decision model based on a deep learning algorithm, and a decision model based on a Bayesian network algorithm.

[0037] That is, the first preset decision model may include one of a decision model based on a reinforcement learning algorithm, a decision model based on a deep learning algorithm, and a decision model based on a Bayesian network algorithm, and preferably may include a decision model based on a reinforcement learning algorithm. In addition, an exemplary Bayesian network algorithm may be a naive Bayesian algorithm, but is not limited thereto.

[0038] According to an embodiment of the present disclosure, by utilizing the SHAP (SHapley Additive ex Planations) algorithm to be described later to perform feature analysis on drone input data corresponding to drone decisions generated by at least one of a decision model based on a reinforcement learning algorithm, a decision model based on a deep learning algorithm, and a decision model based on a Bayesian network algorithm, the computational complexity of interpreting data in these three cases can be reduced, and the computational efficiency and real-time application capability of the entire decision-making method can be improved.

[0039] In addition, by using the SHAP algorithm to be described later to perform feature analysis on the input data corresponding to the drone decision generated by the decision model based on the deep learning algorithm, the feature extraction capability of deep learning can be used to improve the decision quality and interpretability of the entire model.

[0040] In addition, by using the SHAP algorithm to be described later to perform feature analysis on the input data corresponding to the drone decision generated by the decision model based on the Bayesian network algorithm, it is possible to use the Bayesian network to describe the probabilistic relationship between variables and make drone decisions through an inference algorithm.

[0041] As an example, the first preset decision model may include a decision model based on a reinforcement learning algorithm, and in this case, step S102 may further include: based on the first input data set, performing reinforcement learning operations including action, feedback, adjustment, and re-action, to obtain a decision result that satisfies the conditions for maximizing the preset cumulative reward target as the first strategy data set.

[0042] That is, by adopting a reinforcement learning algorithm, the optimal drone decision strategy is generated based on the environmental state including the dynamic environmental changes associated with the drone mission (for example, the changes in the position and route of the drone over time and the changes in the position of obstacles related to the drone over time, etc.). The system gradually optimizes its decision-making process by constantly interacting with the environment to achieve the desired goal.

[0043] According to the embodiments of the present disclosure, by combining the decision model based on the reinforcement learning algorithm and the feature analysis model based on the SHAP algorithm to interpret the complex UAV mission strategy including time series and dynamic environments, it is possible to overcome the limitations of the SHAP algorithm itself that it is only applicable to perception tasks, the limitation that it is not suitable for processing scenarios including data of time series and dynamic environments, and the limitations that it has large computational complexity, high computational complexity and low real-time efficiency.

[0044] As another example, the first preset decision model may include a decision model based on a deep learning algorithm, and in this case, step S102 may further include steps S1021 to S1024: In step S1021, a first sample data set for a drone decision task is obtained. Here, the first sample data set is sample drone task data based on a time series and a dynamic environment stored in a table form, for example, the sample drone task data includes position data for a preset task, position data of at least one obstacle related to the preset task, current route data related to the preset task, and time data related to the preset task.

[0045] In step S1022, based on the first sample data set, a corresponding sample strategy data set is determined. Here, the corresponding sample strategy data set includes an adjustment plan for instructing the next action of the drone for the preset task.

[0046] In step S1023, a preset loss function is calculated based on the first sample data set and the corresponding sample strategy data set.

[0047] In step S1024, the model parameters of the first preset decision model are adjusted according to the preset loss function to obtain the trained first preset decision model.

[0048] According to an embodiment of the present disclosure, in step S103, the first strategy data set and the first input data set are input into a feature analysis model based on the SHAP algorithm to obtain a decision feature data set for the drone decision task. Here, the decision feature data set includes drone decision feature data corresponding to the adjustment plan.

[0049] As an example, step S103 may further include step S1031 and step S1032: In step S1031 , for each strategy in the first strategy data set, current input data corresponding to each strategy is extracted from the first input data set.

[0050] In step S1032, feature analysis based on SHAP values ​​is performed on the current input data of each strategy, and feature extraction is performed based on the marginal contributions of the analyzed different features in the scenarios corresponding to different strategies, so as to obtain the feature set with the greatest impact on the current strategy of each strategy as the decision feature data set for the current input data of each strategy, and the decision feature data sets for the current input data of each strategy are summarized to obtain the decision feature data set for the UAV decision-making task.

[0051] Here, the SHAP algorithm is based on the Shapley value theory and provides a unified feature importance measure by assigning the contribution of features to the model output. Therefore, by using the SHAP algorithm to perform feature analysis on the input data of a single decision strategy, after generating the decision result, the SHAP algorithm is used to perform feature importance analysis on the decision.

[0052] As an example, the marginal contribution of each feature included in the feature set that has the greatest impact on the current strategy for each strategy can be calculated using the following formula: (1) in, Represents the marginal contribution function corresponding to each feature, U represents the utility function obtained after the preset feature subsets cooperate, N represents the full set of features, n represents the number of feature sets, S represents the number of feature subsets, i Indicates the sequence number of the current feature, k is greater than or equal to 1 and less than or equal to n An integer.

[0053] also, U(S) express S The utility obtained by cooperating with a subset of features (for example, the flight path predicted by the neural network, the heading adjustment angle predicted by the neural network, etc.).

[0054] In addition, the meaning of the above formula (1) is for n characteristics, according to the S The result of executing the decision (i.e., the strategy set) when the features are U(S) ) to calculate the Shapley value of each feature. Here, the larger the value, the more important the corresponding feature is for the judgment of this flight path adjustment, and the more it should be retained.

[0055] For example, in a scenario where there are multiple obstacles related to the target task, features corresponding to parameters of a portion of the multiple obstacles may be more important than features corresponding to parameters of another portion of the multiple obstacles, and therefore features corresponding to these parameters of the portion of the multiple obstacles should be retained.

[0056] According to the embodiments of the present disclosure, by using feature analysis based on the SHAP algorithm, users can clearly know the impact of each drone decision feature on the final drone decision, so as to understand the decision logic of the model, which helps to improve the trust in the model. For example, for the strategy set obtained by reinforcement learning (the adjustment plan for the next action of the target task) and the data set before the reinforcement learning is performed (such as various drone information and obstacle information related to the target task), the feature analysis based on the SHAP algorithm in the present disclosure can be used to analyze which information is more important.

[0057] According to an embodiment of the present disclosure, in step S104, a second preset decision model based on a decision tree algorithm is constructed using a decision feature data set to obtain decision tree data for presenting a decision path and a decision basis for an adjustment plan as an explanatory data set for explaining the drone decision task.

[0058] According to an embodiment of the present disclosure, by utilizing the SHAP algorithm to perform feature analysis on dynamic decision results for data including time series and dynamic environments, a comprehensive explanation of the decision-making process of complex UAV decision-making tasks is provided.

[0059] In addition, by analyzing the feature results of the SHAP algorithm, we can extract the feature subset that has the greatest impact on the decision. These features in the feature subset not only help understand the decision logic of the model, but can also be used for further model optimization and debugging.

[0060] In addition, by using the decision tree model to display feature segmentation and decision paths, a detailed explanation of each decision point is provided, making it easier for users to understand and trace the basis of each decision point.

[0061] For example, the construction process of the second preset decision model may include: selecting multiple features from the decision feature data set; constructing each node of the decision tree according to preset evaluation parameters for the multiple features, until the construction process is terminated when a preset termination condition is met.

[0062] According to the embodiments of the present disclosure, by using the above-mentioned features to construct a decision tree model, the decision path and decision basis for drone decisions are displayed through a tree structure, so that each drone decision process can be clearly explained and understood, and the local explanation ability for drone decisions can be further enhanced. In addition, the decision tree model has a small amount of calculation in the prediction stage, can quickly provide explanations, is suitable for real-time applications, and improves the actual use efficiency and effect of the entire decision model (or system).

[0063] The practical application of the explainable UAV task decision method according to the embodiment of the present disclosure is illustrated below in conjunction with the following Table 1. In this example, it is assumed that the number of obstacles is three.

[0064] Table 1

[0065] Referring to Table 1, the first input data set for the UAV decision-making task includes UAV mission data, specifically including position data for the target task (for example, "target_X", "target_Y" in Table 1), position data of at least one obstacle related to the target task (for example, "obstacle 1_X", "obstacle 1_Y", "obstacle 2_X", "obstacle 2_Y", "obstacle 3_X", "obstacle 3_Y" in Table 1), current route data related to the target task (for example, "current heading" in Table 1), and time data related to the target task (for example, "time 1", "time 2", "time 3" in Table 1).

[0066] Specifically, obstacle1_X and obstacle1_Y represent the X and Y coordinates of the first obstacle, obstacle2_X and obstacle2_Y represent the X and Y coordinates of the second obstacle, obstacle3_X and obstacle3_Y represent the X and Y coordinates of the third obstacle, target_X and target_Y represent the X and Y coordinates of the target position of the drone, current heading represents the current heading angle of the drone, time1, time2 and time3 represent different normalized time representations, etc. These parameters together form the basis for the navigation and obstacle avoidance decision-making of the drone when performing tasks.

[0067] In this example, the first preset decision model can be a neural network based on reinforcement learning, and the first policy data set for outputting the decision result can be, for example, "Generate Decision" in Table 1. Here, "Generate Decision" means the final decision output, indicating the next action or adjustment of the drone. By inputting the above parameters into the above neural network based on reinforcement learning to obtain the decision result, the drone can adjust its flight path in real time to avoid obstacles and reach the target location smoothly.

[0068] Subsequently, by inputting the above parameters and generated decisions into the feature analysis model based on the SHAP algorithm, a decision feature data set for the UAV decision-making task is obtained. By calculating the Shapley value of each feature, for example, it can be concluded that the absolute values ​​of the Shapley values ​​of obstacle 3_X, obstacle 3_Y, and time parameters are small, so these data are secondary compared to other data and can be ignored, while other data will be retained.

[0069] For example, the third obstacle is farther from the target position than the other two obstacles, and has less impact on the decision result, so its corresponding data can be ignored; similarly, the three different normalized times also have less impact on the decision result, so their corresponding data can be ignored.

[0070] Finally, when using the decision feature data set to construct a second preset decision model based on the decision tree algorithm, only data related to the parameters "obstacle 1_X", "obstacle 1_Y", "obstacle 2_X", "obstacle 2_Y", "current heading", "target_X", and "target_Y" can be used to construct a second preset decision model based on the decision tree algorithm, and decision tree data for presenting the decision path and decision basis for the adjustment plan for the drone can be obtained as an explanatory data set for explaining the drone decision-making task.

[0071] Figure 2 is a block diagram showing an explainable drone mission decision device 200 according to an embodiment of the present disclosure.

[0072] Reference Figure 2 According to an embodiment of the present disclosure, an explainable UAV mission decision-making device 200 may include a data acquisition module 210, a decision generation module 220, a feature analysis module 230 and a decision interpretation module 240.

[0073] According to an embodiment of the present disclosure, the data acquisition module 210 performs the following operations: acquiring a first input data set for a drone decision task. Here, the first input data set is drone task data based on a time series and a dynamic environment stored in a table form. For example, the drone task data includes position data for a target task, position data of at least one obstacle related to the target task, current route data related to the target task, and time data related to the target task.

[0074] According to an embodiment of the present disclosure, the decision generation module 220 performs the following operations: inputting a first input data set into a pre-trained first preset decision model to obtain a first strategy data set for outputting a decision result. Here, the first strategy data set includes an adjustment scheme for instructing the next action of the drone for the target task.

[0075] As an example, the first preset decision model may include at least one of a decision model based on a reinforcement learning algorithm, a decision model based on a deep learning algorithm, and a decision model based on a Bayesian network algorithm.

[0076] In addition, further, in the case where the first preset decision model includes a decision model based on a reinforcement learning algorithm, the decision generation module 220 inputs the first input data set into the pre-built first preset decision model, and the operation of obtaining the first strategy data set for outputting a decision result may include: based on the first input data set, performing reinforcement learning operations including action, feedback, adjustment, and re-action, to obtain a decision result that satisfies the conditions for maximizing the preset cumulative reward target as the first strategy data set.

[0077] In addition, in the case where the first preset decision model includes a decision model based on a deep learning algorithm, the decision generation module 220 may train the pre-built first preset decision model through the following operations 1) to 4): In operation 1), a first sample data set for a drone decision task is obtained. Here, the first sample data set is sample drone task data based on a time series and a dynamic environment stored in a table form, for example, the sample drone task data includes position data for a preset task, position data of at least one obstacle related to the preset task, current route data related to the preset task, and time data related to the preset task.

[0078] In operation 2), based on the first sample data set, a corresponding sample strategy data set is determined. Here, the corresponding sample strategy data set includes an adjustment scheme for instructing the next action of the drone for the preset task.

[0079] In operation 3), a preset loss function is calculated based on the first sample data set and the corresponding sample strategy data set.

[0080] In operation 4), according to a preset loss function, the model parameters of the first preset decision model are adjusted to obtain a trained first preset decision model.

[0081] According to an embodiment of the present disclosure, the feature analysis module 230 performs the following operations: inputting the first strategy data set and the first input data set into a feature analysis model based on the SHAP algorithm to obtain a decision feature data set for the drone decision task. Here, the decision feature data set includes drone decision feature data corresponding to the adjustment scheme.

[0082] Here, as an example, the feature analysis module 230 inputs the first strategy data set and the first input data set into a feature analysis model based on the SHAP algorithm, and the operation of obtaining a decision feature data set for the UAV decision-making task may include: for each strategy in the first strategy data set, extracting the current input data corresponding to each strategy from the first input data set; performing feature analysis based on the SHAP value on the current input data of each strategy, and performing feature extraction based on the marginal contribution of the analyzed different features in the scenarios corresponding to different strategies, to obtain the feature set with the greatest impact on the current strategy of each strategy, as the decision feature data set for the current input data of each strategy, and summarizing the decision feature data sets for the current input data of each strategy to obtain a decision feature data set for the UAV decision task.

[0083] As an example, the marginal contribution value corresponding to each feature included in the feature set that has the greatest impact on the current strategy of each strategy can be calculated using the above formula (1).

[0084] According to an embodiment of the present disclosure, the decision interpretation module 240 performs the following operations: constructing a second preset decision model based on a decision tree algorithm using a decision feature data set, and obtaining decision tree data for presenting a decision path and decision reasons for an adjustment plan as an explanation data set for explaining the drone decision task.

[0085] As an example, the construction process of the second preset decision model may include: selecting multiple features from a decision feature data set; constructing each node of the decision tree according to preset evaluation parameters for the multiple features, until the construction process is terminated when a preset termination condition is met.

[0086] It should be noted that the operations performed on the above-mentioned structural frames can be referred to in Figure 1 The related contents described are similar and will not be repeated here.

[0087] Figure 3 is a block diagram illustrating a computing device 300 according to an embodiment of the present disclosure.

[0088] Reference Figure 3 , the computing device 300 according to an embodiment of the present disclosure may include a processor 310 and a memory 320. The processor 310 may include (but not limited to) a central processing unit (CPU), a digital signal processor (DSP), a microcomputer, a field programmable gate array (FPGA), a system on a chip (SoC), a microprocessor, an application-specific integrated circuit (ASIC), etc. The memory 320 may store computer executable instructions to be executed by the processor 310. The memory 320 includes a high-speed random access memory and / or a non-volatile computer-readable storage medium. When the processor 310 executes the computer executable instructions stored in the memory 320, the explainable drone mission decision method as described above may be implemented.

[0089] According to the embodiment of the present disclosure, the interpretable drone mission decision method can be written as a computer program / instruction to form a computer program product and stored on a computer-readable storage medium. When the computer program / instruction is executed by a processor, the interpretable drone mission decision method as described above can be implemented. When the instructions in the computer-readable storage medium are executed by the processor of the electronic device / server, the electronic device / server is enabled to execute the interpretable drone mission decision method as described above. Examples of computer-readable storage media include: read-only memory (ROM), random access programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), flash memory, non-volatile memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, Blu-ray or optical disk storage, hard disk drive (HDD), solid state drive (SSD), card storage (such as, multimedia card, secure digital (SD) card or extreme digital (XD) card), magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid state disk and any other device, any other device is configured to store computer program and any associated data, data files and data structures in a non-transitory manner and provide the computer program and any associated data, data files and data structures to a processor or computer so that the processor or computer can execute the computer program. In one example, the computer program and any associated data, data files and data structures are distributed on a networked computer system so that the computer program and any associated data, data files and data structures are stored, accessed and executed in a distributed manner by one or more processors or computers.

[0090] According to the explainable UAV mission decision method and device of the embodiment of the present disclosure, a comprehensive explanation of the decision-making process of complex UAV decision-making schemes is provided by utilizing the SHAP algorithm to perform feature analysis on dynamic decision results for UAV mission data including time series and dynamic environments.

[0091] On the other hand, by using feature analysis based on the SHAP algorithm, users can clearly understand the impact of each drone decision feature corresponding to the drone adjustment plan on the final drone decision, thereby understanding the drone decision logic of the model, which in turn helps to improve trust in the model.

[0092] On the other hand, by utilizing the decision tree model to demonstrate the feature segmentation for UAV decision-making and the decision path for the UAV adjustment plan, a detailed explanation of each UAV decision point is provided, making it easier for users to understand and trace the basis of each UAV decision point.

[0093] On the other hand, by optimizing the calculation of SHAP values ​​and the construction of the decision tree model, the computational complexity is reduced and the real-time application efficiency of the entire interpretable UAV decision model is improved, so that the entire interpretable UAV decision model can quickly provide explanations related to the respective UAV adjustment plans in actual use.

[0094] On the other hand, regarding the application aspects of the explainable drone mission decision method and device according to the embodiments of the present disclosure, they can be used in various fields involving data of time series and dynamic environments, for example, other data processing fields besides the drone field, etc.

[0095] Although some embodiments of the present disclosure have been disclosed and described, it will be appreciated by those skilled in the art that modifications and variations may be made to the embodiments without departing from the concept and spirit of the present disclosure, the scope of which is defined by the claims and their equivalents.

Claims

1. An explainable UAV mission decision method, characterized in that: The UAV mission decision method comprises: Acquire a first input data set for a drone decision task, wherein the first input data set is drone task data based on a time series and a dynamic environment stored in a table form, wherein the drone task data includes position data for a target task, position data of at least one obstacle related to the target task, current route data related to the target task, and time data related to the target task; Inputting the first input data set into a pre-built first preset decision model to obtain a first strategy data set for outputting a decision result, wherein the first strategy data set includes an adjustment plan for instructing the next action of the drone for the target task; Inputting the first strategy data set and the first input data set into a feature analysis model based on the SHAP algorithm to obtain a decision feature data set for the UAV decision task, wherein the decision feature data set includes UAV decision feature data corresponding to the adjustment scheme; The decision feature data set is used to construct a second preset decision model based on a decision tree algorithm, and decision tree data for presenting a decision path and decision basis for the adjustment plan is obtained as an explanation data set for explaining the drone decision task.

2. The UAV mission decision method according to claim 1, characterized in that: The step of inputting the first strategy data set and the first input data set into a feature analysis model based on the SHAP algorithm to obtain a decision feature data set for the UAV decision task includes: For each strategy in the first strategy data set, extracting current input data corresponding to each strategy from the first input data set; A feature analysis based on the SHAP value is performed on the current input data of each strategy, and feature extraction is performed based on the marginal contribution of different analyzed features in the situations corresponding to different strategies, so as to obtain the feature set with the greatest impact on the current strategy of each strategy, which is used as the decision feature data set for the current input data of each strategy, and the decision feature data sets for the current input data of each strategy are summarized to obtain the decision feature data set for the UAV decision task.

3. The UAV mission decision method according to claim 2, characterized in that: The marginal contribution of each feature included in the feature set that has the greatest impact on the current strategy of each strategy is calculated using the following formula: in, Represents the marginal contribution function corresponding to each feature, U represents the utility function obtained after the preset feature subsets cooperate, N represents the full set of features, n represents the number of feature sets, S represents the number of feature subsets, i Indicates the sequence number of the current feature, k is greater than or equal to 1 and less than or equal to n An integer.

4. The UAV mission decision method according to claim 1, characterized in that: The construction process of the second preset decision model includes: Selecting a plurality of features from the decision feature dataset; According to the preset evaluation parameters for the multiple features, each node of the decision tree is constructed until the construction process ends when a preset termination condition is met.

5. The UAV mission decision method according to claim 1, characterized in that: The first preset decision model includes at least one of a decision model based on a reinforcement learning algorithm, a decision model based on a deep learning algorithm, and a decision model based on a Bayesian network algorithm.

6. The UAV mission decision method according to claim 5, characterized in that: The first preset decision model includes a decision model based on a reinforcement learning algorithm, and the step of inputting the first input data set into the pre-built first preset decision model to obtain a first strategy data set for outputting a decision result includes: Based on the first input data set, reinforcement learning operations including action, feedback, adjustment, and re-action are performed to obtain a decision result that satisfies the conditions for maximizing the preset cumulative reward target as the first strategy data set.

7. The UAV mission decision method according to claim 6, characterized in that: The first preset decision model includes a decision model based on a deep learning algorithm, and the first preset decision model is trained in the following manner: Acquire a first sample data set for a drone decision task, wherein the first sample data set is sample drone task data based on a time series and a dynamic environment stored in a table form, wherein the sample drone task data includes position data for a preset task, position data of at least one obstacle related to the preset task, current route data related to the preset task, and time data related to the preset task; Based on the first sample data set, determining a corresponding sample strategy data set, the corresponding sample strategy data set including an adjustment scheme for instructing the next action of the drone for a preset task; Calculating a preset loss function based on the first sample data set and the corresponding sample strategy data set; According to the preset loss function, the model parameters of the first preset decision model are adjusted to obtain the trained first preset decision model.

8. An explainable UAV mission decision device, characterized in that: The UAV mission decision device comprises: The data acquisition module is configured to: acquire a first input data set for a drone decision task, wherein the first input data set is drone task data based on a time series and a dynamic environment stored in a table form, wherein the drone task data includes position data for a target task, position data of at least one obstacle related to the target task, current route data related to the target task, and time data related to the target task; A decision generation module is configured to: input the first input data set into a pre-built first preset decision model to obtain a first strategy data set for outputting a decision result, wherein the first strategy data set includes an adjustment plan for instructing the next action of the drone for the target task; A feature analysis module is configured to: input the first strategy data set and the first input data set into a feature analysis model based on the SHAP algorithm to obtain a decision feature data set for the UAV decision task, wherein the decision feature data set includes UAV decision feature data corresponding to the adjustment scheme; The decision interpretation module is configured to: use the decision feature data set to construct a second preset decision model based on a decision tree algorithm, and obtain decision tree data for presenting the decision path and decision reasons for the adjustment plan as an explanation data set for explaining the drone decision task.

9. A computer program product, characterized in that The computer program product comprises a computer program / instruction, which, when executed by a processor, implements the explainable UAV mission decision method according to any one of claims 1 to 7.

10. A computing device, characterized in that The computing device includes: at least one processor; and at least one memory storing computer-executable instructions, wherein the computer-executable instructions, when executed by the at least one processor, cause the at least one processor to execute the explainable drone mission decision method as described in any one of claims 1 to 7.

Images