Unmanned aerial vehicle intelligent decision support method and system

Abstract

The invention discloses an unmanned aerial vehicle intelligent decision support method and system, and relates to the technical field of unmanned aerial vehicle intelligent decision. The method comprises the following steps: constructing a modal environment logic topology, and defining a modal operator for representing a causal relationship among environmental parameters; based on the effective directed acyclic graph topology, performing modal analysis on multi-scale environmental parameters, and converting the multi-scale environmental parameters into modal logic propositions; based on the modal logic proposition and the effective directed acyclic graph topology, generating a main rule and an alternative rule set through a modal inference engine; and verifying the effective rule by adopting a double-engine adversarial architecture, optimizing the rule and directed acyclic graph topology through an arbitration mechanism, and generating a final decision rule. According to the method, the decision accuracy of the unmanned aerial vehicle in an unknown environment can be remarkably improved, manual intervention is greatly reduced, and the new environment deployment time is effectively shortened; according to the method, the extreme scene robustness is enhanced, the fault early warning time is prolonged, the misjudgment rate is reduced, and more reliable intelligent decision support is provided for cross-scene operation of the unmanned aerial vehicle.

Description

Technical Field

[0001] This invention relates to the field of intelligent decision-making technology for unmanned aerial vehicles (UAVs), specifically to an intelligent decision support method and system for UAVs. Background Technology

[0002] With the rapid development of drone technology, its application scenarios have expanded from conventional plains to complex environments such as plateaus, coastlines, and typhoon zones, placing extremely high demands on the environmental adaptability of intelligent decision-making systems. Drone operations require real-time processing of multi-dimensional environmental parameters such as air pressure, humidity, wind speed, and terrain. These parameters exhibit complex implicit causal relationships, and the decision-making system must accurately capture these relationships to dynamically adjust equipment operating thresholds to ensure operational safety and efficiency.

[0003] Existing UAV decision-making systems mostly rely on manually preset rules, that is, they make decisions by encoding fixed environmental parameter coupling relationships (such as "adjusting the temperature threshold when the air pressure is below a certain value"). However, such systems cannot autonomously discover new causal relationships in unknown environments. When encountering scenarios with unpreset parameter coupling (such as the combined effects of low air pressure at high altitudes and strong winds from typhoons), the lack of corresponding logical rules leads to decision failure, making it difficult to meet the operational needs in complex environments. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent decision support method and system for unmanned aerial vehicles (UAVs) to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent decision support method for unmanned aerial vehicles (UAVs), comprising the following steps:

[0006] Step S100: Construct a modal environment logical topology, define modal operators to characterize the causal relationship between environmental parameters, generate a dynamically updated directed acyclic graph topology through bootstrap causal chain mining, and perform validity judgment on the directed acyclic graph topology to output a valid directed acyclic graph topology.

[0007] Step S200: Based on the effective directed acyclic graph topology, perform modal analysis on the multi-scale environmental parameters and convert them into modal logic propositions;

[0008] Step S300: Based on the modal logic proposition and the effective directed acyclic graph topology, generate the main rule and alternative rule set through the modal reasoning engine, use the cross-scale fusion mechanism to process the spatiotemporal correlation, and perform confidence judgment on the rules to output the effective rules;

[0009] Step S400: The effective rules are verified using a dual-engine adversarial architecture, and the rules and directed acyclic graph topology are optimized through an arbitration mechanism to generate the final decision rules.

[0010] Preferably, step S100 includes:

[0011] Step S110: Define modal operators, which include necessary operators □, possible operators ◇, and time-series operators △, wherein:

[0012] The logical expression for the necessary operator □ is □(x→y), which means that the antecedent environmental parameter x necessarily causes the consequent environmental parameter y to change.

[0013] The logical expression for the possible operator ◇ is ◇(x→y), which means that the antecedent environmental parameter x may cause the consequent environmental parameter y to change;

[0014] The logical expression of the timing operator △ is △(x→y,t), which means that the antecedent environmental parameter x causes the consequent environmental parameter y to change within the time window t (unit: seconds);

[0015] Step S120: Bootstrap causal chain mining, including converting continuous sensor data into discrete event symbols, determining the causal direction through active intervention experiments, and updating the topology of the directed acyclic graph based on the co-occurrence frequency of event symbols. The topology update formula is:

[0016] G(t+1)=G(t)∪{<x,y,m> |P(y|x)>θ}

[0017] In the formula, G(t) is the topology of the directed acyclic graph at time t.<x,y,m> For the newly discovered causal chain, x is the cause parameter, y is the result parameter, m is the modal operator, P(y|x) is the conditional probability, and θ is the association threshold;

[0018] Step S130: Topology validity judgment. The updated directed acyclic graph topology is checked for contradictory causal chains using the modal resolution principle. If there are no contradictions, a valid topology is output. If there are contradictions, low-confidence conflict chains are removed and the process returns to step S120.

[0019] Preferably, in step S120, converting continuous sensor data into discrete event symbols includes: extracting features from environmental parameter data collected by sensors, generating event symbols based on parameter thresholds and changing trends, wherein the event symbols include parameter feature values ​​and time changes; determining the causal direction through active intervention experiments includes: actively perturbing the target environmental parameters, recording the response time sequence of the associated parameters, and determining the causal derivation direction based on the response time difference.

[0020] Preferably, step S200 includes:

[0021] Step S210: Divide the environmental parameters into spatiotemporal scales, into instantaneous parameters, slowly varying parameters, and spatial parameters. Instantaneous parameters are those with a sampling frequency ≥ 10 Hz, slowly varying parameters are those with a sampling frequency ≤ 0.1 Hz, and spatial parameters are those generated based on GPS and elevation data.

[0022] Step S220: Based on the causal chains and modal operators in the effective directed acyclic graph topology, the partitioned multi-scale environment parameters are converted into modal logic propositions containing the necessary operator □, the possible operator ◇, and the temporal operator △.

[0023] Preferably, step S300 includes:

[0024] Step S310: The modal reasoning engine generates rules, generating the main rule R based on modal logic propositions and the topology of a valid directed acyclic graph. a and an alternative rule set, wherein the alternative rule set includes rule R β and R e The main rule is the rule with the highest initial confidence, and the alternative rules are other candidate rules that meet the association threshold. Conflict resolution is performed using the modal resolution principle. The confidence of the main rule and the alternative rules are compared, and the rule with the highest confidence is selected as the candidate rule.

[0025] Step S320: The cross-scale fusion mechanism handles spatiotemporal correlations, including:

[0026] Time-scale coupling, through formula t 综合 =t 瞬时 ×w1+t avg ×w2 combines instantaneous and slowly varying parameters, where t 综合 t represents the timescale parameter after fusion. 瞬时 For instantaneous parameter values, t avg Let w1 be the mean of the slowly varying parameters, w2 be the weight of the instantaneous parameters, and w1 + w2 = 1;

[0027] Spatial-temporal correlation uses spatial parameters as constraints for rule premises to generate composite rules that include spatial features;

[0028] Step S330: Rule confidence judgment. If the rule confidence is ≥0.7, it is output as a valid rule. If the confidence is <0.7, return to step S310 to regenerate the rule.

[0029] Preferably, in step S310, the conflict resolution determination formula is: if Conf(R) a )>Conf(R β )∧Conf(R a )>Conf(R e If ), then activate rule R. aIn the formula Conf(R) a ), Conf(R) β ), Conf(R) e ) are respectively rules R a R β R e The confidence level.

[0030] Preferably, step S400 includes:

[0031] Step S410: Generate rules for the dual-engine adversarial architecture. The main engine generates decision rules R based on the current directed acyclic graph topology. 主 The adversarial engine generates challenge rules R by perturbing the causal chain. 对抗 ;

[0032] Step S420: The arbitration mechanism evaluates the rules and calculates the probability of equipment damage after the rule is triggered. The probability of risk is calculated as Risk(R) = P(damage|triggering R), where Risk(R) is the probability of risk of rule R and P(damage|triggering R) is the conditional probability of equipment damage after rule R is triggered.

[0033] Step S430: Topology and rule optimization, if Risk(R 对抗 ) <Risk(R 主 If the causal chain of the adversarial engine is valid for three consecutive times, then the causal chain of the adversarial engine will be incorporated into the main topology, as shown in the formula G. 主 =G 主 ∪G 对抗中有效链 In the formula G 主 For the optimized main topology, G 对抗中有效链 To counteract the effective causal chains generated by the engine, generate the most unfavorable scenario to verify the robustness of the rules. If the rules fail, return to step S100 to rebuild the topology.

[0034] The present invention also provides an intelligent decision support system for unmanned aerial vehicles, comprising:

[0035] The modal topology self-evolution module is used to construct a modal environment logical topology, define modal operators to characterize the causal relationship between environmental parameters, generate dynamically updated directed acyclic graph topologies through bootstrapping causal chain mining, and perform validity judgment on the directed acyclic graph topologies, outputting valid directed acyclic graph topologies.

[0036] The multi-scale modal analysis module is used to perform modal analysis on multi-scale environmental parameters based on the effective directed acyclic graph topology and convert them into modal logic propositions.

[0037] The modal rule reasoning module is used to generate master rules and candidate rule sets through the modal logic proposition and the effective directed acyclic graph topology, and to process spatiotemporal correlations by adopting a cross-scale fusion mechanism, and to judge the confidence of the rules and output the effective rules.

[0038] The adversarial verification optimization module is used to verify the effective rules using a dual-engine adversarial architecture, optimize the rules and the directed acyclic graph topology through an arbitration mechanism, and generate the final decision rules.

[0039] The present invention also provides an electronic device, which is a physical device, comprising:

[0040] The processor and the memory are communicatively connected.

[0041] The memory is used to store at least one executable instruction executed by the processor, which executes the executable instruction to implement an unmanned aerial vehicle (UAV) intelligent decision support method as described above.

[0042] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described intelligent decision support method for unmanned aerial vehicles.

[0043] Compared with the prior art, the beneficial effects of the present invention are:

[0044] This invention effectively solves the problem of environmental dimension collapse in rule trigger thresholds by characterizing the causal relationships of environmental parameters using modal logic, bootstrapping to mine causal chains, fusing spatiotemporal parameters across scales, and optimizing adversarial verification. It significantly improves the decision-making accuracy of UAVs in unknown environments, greatly reduces human intervention, and effectively shortens deployment time in new environments; it also enhances robustness in extreme scenarios, extends fault warning time, reduces false positive rates, and provides more reliable intelligent decision support for UAV operations across different scenarios. Attached Figure Description

[0045] Figure 1 This is a main flowchart of an intelligent decision support method for unmanned aerial vehicles (UAVs) provided in an embodiment of the present invention.

[0046] Figure 2 This is a schematic diagram of the structure of an intelligent decision support system for unmanned aerial vehicles (UAVs) provided in an embodiment of the present invention;

[0047] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] The method in this embodiment is executed by a terminal, which can be a mobile phone, tablet computer, PDA, laptop or desktop computer, etc. Of course, it can also be other devices with similar functions, and this embodiment does not limit them.

[0050] Please see Figure 1 This invention provides an intelligent decision support method and system for unmanned aerial vehicles (UAVs), the method being applied to, including:

[0051] Step S100: Construct a modal environment logical topology, define modal operators to characterize the causal relationships between environmental parameters, generate dynamically updated directed acyclic graph topologies through bootstrapping causal chain mining, and perform validity judgment on the directed acyclic graph topologies to output valid directed acyclic graph topologies.

[0052] It should be noted that in this embodiment, step S100 is the foundational building step of the entire decision support method. Its core purpose is to establish a logical framework that can accurately characterize the causal relationships between environmental parameters, providing underlying support for the generation of subsequent decision rules. Here, "modal environmental logical topology" refers to a directed graph structure that uses modal operators to label the causal relationship attributes (such as necessity, possibility, and temporality) between environmental parameters; "modal operators" refer to symbolic tools used to describe the characteristics of causal relationships, including the necessity operator □, the possibility operator ◇, and the temporality operator △; "bootstrapping causal chain mining" refers to the process by which the system autonomously discovers causal relationships between environmental parameters through data symbolization, proactive intervention experiments, and probability analysis, without relying on pre-set knowledge; "directed acyclic graph topology" refers to a graph structure where environmental parameters are nodes, causal relationships are directed edges, and there are no cyclic dependencies; "validity judgment" refers to the verification process that ensures there are no contradictory causal chains in the topology through logical consistency checks. This step, by dynamically constructing and verifying the topology, enables the autonomous accumulation of environmental knowledge, overcomes the limitations of traditional rule engines that rely on manually pre-set causal relationships, and provides a foundational logic for subsequent decision-making that is adaptable to different scenarios.

[0053] Additionally, it should be noted that in one possible implementation, the definition of modal operators can be extended according to the complexity of the scenario. For example, for extreme environments, a "strong correlation operator" □+ (representing a causal relationship confidence level ≥ 0.95) and a "weak correlation operator" ◇- (representing a causal relationship confidence level ≤ 0.3) can be added. The active intervention experiment in bootstrap causal chain mining can adopt a gradient perturbation strategy, that is, to fine-tune the target parameters in multiple gradients and record the continuous response curves of the correlation parameters to improve the accuracy of causal direction determination. In addition to detecting contradictory chains, the validity judgment can also be verified by backtracking historical data to verify the adaptability of the topology to the known scenario. If the adaptability rate is lower than 85%, topology reconstruction is triggered.

[0054] For example, in one specific implementation, when the UAV first enters the plateau environment, step S100 initiates bootstrap causal chain mining: sensor data such as air pressure, temperature, and air density are converted into event symbols such as "sudden drop in air pressure" and "temperature fluctuation"; by actively increasing the flight altitude to create air pressure disturbances, the necessary relationship of "air pressure decrease → air density decrease" (marked as □) and the possible relationship of "humidity increase → equipment heat dissipation efficiency decrease" (marked as ◇) are discovered; after generating a directed acyclic graph topology containing these causal chains, the contradictory chain of "air pressure increase → air density decrease" is detected through the modal resolution principle, and the effective topology is output after removing the low-confidence chain, laying the foundation for decision-making in plateau scenarios.

[0055] Specifically, step S100 includes:

[0056] Step S110: Define modal operators, which include necessary operators □, possible operators ◇, and time-series operators △, wherein:

[0057] The logical expression for the necessary operator □ is □(x→y), which means that the antecedent environmental parameter x necessarily causes the consequent environmental parameter y to change.

[0058] The logical expression for the possible operator ◇ is ◇(x→y), which means that the antecedent environmental parameter x may cause the consequent environmental parameter y to change;

[0059] The logical expression of the timing operator △ is △(x→y,t), which means that the antecedent environmental parameter x causes the consequent environmental parameter y to change within the time window t (unit: seconds);

[0060] Step S120: Bootstrap causal chain mining, including converting continuous sensor data into discrete event symbols, determining the causal direction through active intervention experiments, and updating the topology of the directed acyclic graph based on the co-occurrence frequency of event symbols. The topology update formula is:

[0061] G(t+1)=G(t)∪{<x,y,m> |P(y|x)>θ}

[0062] In the formula, G(t) is the topology of the directed acyclic graph at time t.<x,y,m> For the newly discovered causal chain, x is the cause parameter, y is the result parameter, m is the modal operator, P(y|x) is the conditional probability, and θ is the association threshold;

[0063] Step S130: Topology validity judgment. The updated directed acyclic graph topology is checked for contradictory causal chains using the modal resolution principle. If there are no contradictions, a valid topology is output. If there are contradictions, low-confidence conflict chains are removed and the process returns to step S120.

[0064] Optionally, the sub-steps of step S100 aim to construct a basic logical framework for the causal relationships of environmental parameters. This is achieved through modal operator definition, bootstrap causal chain mining, and topological validity judgment, enabling a precise characterization of environmental associations. Modal operators refer to symbols used to describe the deterministic and temporal characteristics of causal relationships between environmental parameters, including the necessary operator □, the possible operator ◇, and the time-series operator △. Bootstrap causal chain mining refers to the process by which the system autonomously discovers parameter associations without relying on pre-defined knowledge, through data symbolization, proactive intervention experiments, and topological updates. The directed acyclic graph topology refers to a graphical structure using nodes to represent environmental parameters and directed edges to represent causal relationships, used to store parameter association logic. The core purpose of this step is to overcome the limitations of traditional rule engines that rely on manually pre-defined coupling relationships. It enhances the system's adaptability to unknown environments through dynamic topology generation. The implementation approach involves first defining a logical symbol system, then mining associations through data-driven methods, and finally verifying topological consistency to ensure reliability.

[0065] In one possible implementation, the necessary operator □ can be used to characterize correlations with clearly defined physical laws (such as the relationship between air pressure and air density), the possible operator ◇ can be used to describe correlations affected by random factors (such as the relationship between humidity and equipment short circuits), and the time window t of the time series operator △ can be dynamically adjusted according to the rate of parameter change (such as setting t to 10 seconds for wind speed correlation and t to 300 seconds for temperature correlation).

[0066] Additionally, it should be noted that the topology update formula G(t+1)=G(t)∪{<x,y,m> In the formula |P(y|x)>θ}, P(y|x) refers to the conditional probability that the consequent parameter y occurs when the consequent parameter x occurs, and θ is the association threshold (default 0.85, which can be adjusted according to the complexity of the scenario). This formula ensures that newly discovered high-confidence causal chains can be incorporated into the topology in a timely manner. The topology validity judgment is implemented through the modal resolution principle, which refers to detecting whether there are contradictory causal chains (such as the simultaneous existence of □(x→y) and □(x→y)) through logical deduction. If a chain with low confidence exists, it is removed to ensure the self-consistency of the topological logic.

[0067] For example, in one feasible implementation, after the UAV enters the plateau environment, step S110 defines Δ(sudden drop in air pressure → decrease in battery heat dissipation efficiency, 60 seconds); step S120 converts the sensor data into the event symbol "air pressure < 70 kPa", and through an intervention experiment of actively increasing the altitude by 100 meters, it is found that the heat dissipation efficiency decreases by 80% within 60 seconds after the sudden drop in air pressure, generating a causal chain <air pressure, heat dissipation efficiency, Δ(60 seconds)> and updating the topology; step S130 detects that the new topology does not contradict the existing physical laws, and outputs the valid topology for subsequent reasoning.

[0068] In step S120, converting continuous sensor data into discrete event symbols includes: extracting features from environmental parameter data collected by sensors, generating event symbols based on parameter thresholds and trends, wherein the event symbols include parameter feature values ​​and time changes; determining the causal direction through active intervention experiments includes: actively perturbing the target environmental parameters, recording the response time sequence of related parameters, and determining the causal derivation direction based on the response time difference.

[0069] It should be noted that, in this embodiment, the detailed content of step S120 is the core implementation link of bootstrap causal chain mining. Through data symbolization and active intervention experiments, the accuracy and directionality of the causal chain are ensured. Continuous sensor data refers to analog data collected by sensors in real time (such as air pressure and humidity values ​​collected per second); discrete event symbols refer to converting continuous data into discrete identifiers containing feature values ​​and time-varying quantities (such as "air pressure < 70 kPa and a decrease of 8 kPa in 30 minutes"); active intervention experiments refer to the system actively changing a parameter (such as adjusting flight altitude to change air pressure) and observing the responses of other parameters to determine the causal direction. The purpose of this step is to solve the problem that traditional methods struggle to autonomously discover causal relationships. Data symbolization enables the logical processing of continuous data, and active intervention avoids misjudging correlation as causality, thereby improving the reliability of causal chain mining.

[0070] In one possible implementation, data symbolization can be achieved by combining threshold segmentation with trend recognition. For example, temperature data can be divided into “low temperature (<0℃)”, “normal temperature (0-35℃)”, and “high temperature (>35℃)”, and marked with trends such as “continuous rise”, “fluctuation”, and “continuous fall”. Active intervention experiments can set a minimum disturbance amplitude (e.g., the air pressure can be adjusted by no less than 5 kPa each time) to ensure that the response signal is identifiable.

[0071] For example, in one feasible implementation, when the UAV is operating in a coastal environment, step S120 first converts the continuous humidity data into an event symbol of "humidity > 90% and increases by 10% every 5 minutes"; then, by actively reducing the flight altitude (increasing humidity), it is observed that the probability of condensation on the equipment surface increases by 70%, and the causal direction is determined to be "increased humidity → increased risk of condensation"; finally, based on the co-occurrence frequency (the probability of this association is 0.92 > θ = 0.85), the causal chain <humidity, condensation risk, ◇> is incorporated into the topology of the directed acyclic graph.

[0072] Step S200: Based on the effective directed acyclic graph topology, perform modal analysis on the multi-scale environmental parameters and convert them into modal logic propositions.

[0073] It should be noted that in this embodiment, step S200 serves as a bridge between environmental perception and logical reasoning, aiming to convert the raw environmental data into symbolic propositions that can be processed by the reasoning engine. Here, "effective directed acyclic graph topology" refers to the consistent causal relationship graph structure verified in step S100; "multi-scale environmental parameters" refers to environmental data divided by time sampling frequency and spatial attributes, including instantaneous parameters (e.g., real-time wind speed), slowly varying parameters (e.g., daily average temperature), and spatial parameters (e.g., terrain slope); "modal parsing" refers to the process of converting parameter data into logical propositions based on the causal chains and modal operators in the topology; and "modal logical proposition" refers to symbolic statements containing modal operators, used to describe the current state of environmental parameters and the activation status of causal relationships. This step, by symbolizing multi-dimensional, multi-scale environmental data, achieves the adaptation of raw data to logical reasoning, providing standardized input for subsequent rule generation.

[0074] Additionally, it should be noted that in one possible implementation, the classification criteria for multi-scale environmental parameters can be dynamically adjusted. For example, in a typhoon scenario, the sampling frequency of instantaneous parameters can be increased to 20Hz to capture rapidly changing wind speed data. During modal analysis, if the parameter value is in a fuzzy range (such as humidity approaching a threshold), a fuzzy logic membership function can be introduced to convert "humidity 90%" into the propositional form of "◇(humidity > 85% → equipment leakage), membership degree 0.9", thereby enhancing the ability to represent uncertain data.

[0075] For example, in one specific implementation, when the UAV is patrolling in a mountainous area, step S200 analyzes the multi-scale parameters based on the effective topology output in step S100: the real-time wind speed (15m / s, lasting 10s) with a sampling frequency of 10Hz is converted into "△ (wind speed > 12m / s → turbulence risk, 10s)"; the average daily temperature (35℃) with a sampling frequency of 0.1Hz is converted into "□ (average daily temperature > 30℃ → increased battery load)"; the terrain slope (30°) generated based on GPS is converted into "spatial parameter s = 30°"; and finally, it is integrated into a modal logic proposition containing temporal, necessary, and spatial attributes, providing symbolic input for rule generation.

[0076] Specifically, step S200 includes:

[0077] Step S210: Divide the environmental parameters into spatiotemporal scales, into instantaneous parameters, slowly varying parameters, and spatial parameters. Instantaneous parameters are those with a sampling frequency ≥ 10 Hz, slowly varying parameters are those with a sampling frequency ≤ 0.1 Hz, and spatial parameters are those generated based on GPS and elevation data.

[0078] Step S220: Based on the causal chains and modal operators in the effective directed acyclic graph topology, the partitioned multi-scale environment parameters are converted into modal logic propositions containing the necessary operator □, the possible operator ◇, and the temporal operator △.

[0079] It should be noted that, in this embodiment, the subdivision of step S200 aims to convert the original environmental data into symbolic propositions that can be used for logical reasoning. This is achieved through spatiotemporal scale division and modal proposition transformation, realizing a unified representation of multi-dimensional parameters. Spatiotemporal scale division refers to classifying environmental parameters into instantaneous parameters (rapidly changing), slowly changing parameters (gradually changing), and spatial parameters (related to geographical location) based on their rate of change and spatial characteristics. Modal proposition transformation refers to converting the divided parameters into logical statements containing modal operators based on an effective directed acyclic graph topology. The core purpose of this step is to solve the reasoning gap problem caused by differences in the spatiotemporal characteristics of parameters. By unifying the symbol system, multi-scale parameters can participate in logical deduction. The implementation approach is to first classify and process the parameters to adapt to different change characteristics, and then combine topological associations to generate logical propositions.

[0080] In one possible implementation, instantaneous parameters may include wind speed, acceleration, etc. (sampling frequency 20Hz), slowly varying parameters may include daily average temperature, average humidity, etc. (sampling frequency 0.05Hz), and spatial parameters may include terrain slope, altitude, etc. (updated every 10 meters via GPS and elevation data); modal proposition transformation can employ rule matching, such as automatically generating the proposition "□(p<70kPa→ρ<0.9kg / m³)" when "air pressure < 70kPa" is detected and the topology contains □ (air pressure → air density).3 )".

[0081] For example, in one feasible implementation, when the drone is operating in a mountainous area, step S210 divides the real-time wind speed (15m / s, sampling frequency 10Hz) into instantaneous parameters, the daily average temperature (25℃, sampling frequency 0.1Hz) into slowly varying parameters, and the terrain slope (30°) into spatial parameters; step S220, based on the association between "△(wind speed > 12m / s → turbulence risk, 10s)" and "□(slope > 25° → airflow turbulence)" in the topology, transforms and generates the modal proposition "△(v > 12m / s → turbulence risk, 10s) ∧ □(s > 25° → airflow turbulence)", providing logical input for subsequent rule generation.

[0082] Step S300: Based on the modal logic proposition and the effective directed acyclic graph topology, the main rule and alternative rule set are generated through the modal reasoning engine, the spatiotemporal correlation is processed by the cross-scale fusion mechanism, the confidence of the rules is judged, and the effective rules are output.

[0083] It should be noted that in this embodiment, step S300 is the core generation step of the decision rule, and its purpose is to generate executable rules adapted to the current scenario based on symbolic environment propositions. Here, "modal logic proposition" refers to the symbolic statement containing modal operators output in step S200; "modal inference engine" refers to the program module capable of performing inference operations based on modal logic rules; "master rule and alternative rule set" refers to the rule set generated by the inference engine, with priority distinguished by confidence level. The master rule is the rule with the highest initial confidence level, and the alternative rules are other candidate rules that meet the basic threshold; "cross-scale fusion mechanism" refers to the logical processing method that fuses different time scales (instantaneous / gradually changing) and spatial dimension parameters; "confidence level judgment" refers to the verification process of filtering effective rules through a preset threshold (e.g., ≥0.7). This step, through multi-scale parameter fusion and confidence level filtering, can generate decision rules that balance timeliness and reliability, solving the problem of traditional rules' dependence on a single scale parameter.

[0084] Additionally, it should be noted that in one possible implementation, the time scale coupling weight in the cross-scale fusion mechanism can be dynamically adjusted. For example, when the instantaneous parameter remains abnormal for more than 5 minutes, the instantaneous parameter weight w1 is increased from 0.3 to 0.7 to prioritize response to sudden environmental changes. In addition to using a fixed threshold, the confidence judgment can also be dynamically adjusted in conjunction with the scenario risk level. For example, in an emergency rescue scenario, the confidence threshold can be reduced to 0.6 to prioritize ensuring response speed.

[0085] For example, in one specific implementation, when the drone performs a logistics delivery task, step S300 is based on modal logic propositions and valid topology generation rules: main rule R aFor “IF□(atmospheric pressure <70kPa)THEN battery temperature threshold = 60℃ × 0.8” (confidence level 0.8), alternative rule R β The given rule is "IF◇(humidity>90%)THEN battery temperature threshold=60℃×0.9" (confidence level 0.6), alternative rule R. e The formula is: "IF□(air pressure <70kPa)∧◇(humidity >90%)THEN battery temperature threshold = 60℃ × 0.75" (confidence level 0.85); instantaneous wind speed and daily average temperature are coupled through cross-scale fusion to adjust the rule triggering conditions; based on the confidence level, R... e Because the confidence level is ≥0.7, it is selected as a valid rule output, thus realizing multi-factor collaborative decision-making.

[0086] Specifically, step S300 includes:

[0087] Step S310: The modal reasoning engine generates rules, generating the main rule R based on modal logic propositions and the topology of a valid directed acyclic graph. a and an alternative rule set, wherein the alternative rule set includes rule R β and R e The main rule is the rule with the highest initial confidence, and the alternative rules are other candidate rules that meet the association threshold. Conflict resolution is performed using the modal resolution principle. The confidence of the main rule and the alternative rules are compared, and the rule with the highest confidence is selected as the candidate rule.

[0088] Step S320: The cross-scale fusion mechanism handles spatiotemporal correlations, including:

[0089] Time-scale coupling, through formula t 综合 =t 瞬时 ×w1+t avg ×w2 combines instantaneous and slowly varying parameters, where t 综合 t represents the timescale parameter after fusion. 瞬时 For instantaneous parameter values, t avg Let w1 be the mean of the slowly varying parameters, w2 be the weight of the instantaneous parameters, and w1 + w2 = 1;

[0090] Spatial-temporal correlation uses spatial parameters as constraints for rule premises to generate composite rules that include spatial features;

[0091] Step S330: Rule confidence judgment. If the rule confidence is ≥0.7, it is output as a valid rule. If the confidence is <0.7, return to step S310 to regenerate the rule.

[0092] It should be noted that, in this embodiment, the sub-steps of step S300 are the core of rule generation and optimization. Rules are generated through a modal inference engine, spatiotemporal correlations are processed through cross-scale fusion, and confidence levels are determined to output reliable decision rules. Here, the main rule refers to the candidate rule with the highest initial confidence, and the candidate rule set refers to the set of rules that potentially conflict with the main rule but meet the basic correlation conditions. Cross-scale fusion refers to the mechanism of incorporating parameter correlations at different spatiotemporal scales into the same rule, including temporal scale coupling and spatial-temporal correlation. Rule confidence refers to the probability (range 0-1) that the rule prediction result matches the actual environment. The purpose of this step is to address the difficulty of traditional rules in handling multi-scale parameter correlations and uncertainties. It improves the robustness of decision-making through multi-rule generation and filtering. The implementation approach is to first generate candidate rules based on logical propositions, then fuse multi-scale parameters to optimize the rules, and finally ensure rule reliability through confidence level filtering.

[0093] In one possible implementation, the time-scale coupling formula t 综合 =t 瞬时 ×w1+t avg The weight w1 in ×w2 can be dynamically adjusted according to the duration of the instantaneous parameter anomaly (e.g., w1 = 0.3 if the anomaly lasts for 5 minutes, and w1 = 0.7 if it lasts for more than 5 minutes); the spatial-temporal association can use terrain parameters as a prerequisite for the rule to take effect (e.g., "the wind speed rule is only activated when the slope is >30°").

[0094] For example, in one feasible implementation, when the UAV operates in a high-altitude typhoon environment, step S310 generates the master rule R based on modal propositions. a (IF□(p<70kPa)THENt=48℃, confidence level 0.7) and alternative rule R β (IF◇(h>90%)THENt=45℃, confidence level 0.8); Step S320 fuses the instantaneous temperature (52℃) and the daily average temperature (15℃) using a time-scale coupling formula to obtain t. 综合 =52×0.7+15×0.3=41.9℃, and generate a composite rule by combining the spatial parameter "slope 35°"; step S330 determines R β A confidence level of 0.8 or higher than 0.7 indicates a valid rule.

[0095] In step S310, the conflict resolution determination formula is: if Conf(R) a )>Conf(R β )∧Conf(R a )>Conf(R e If ), then activate rule R. a In the formula Conf(R) a ), Conf(R)β ), Conf(R) e ) are respectively rules R a R β R e The confidence level.

[0096] It should be noted that in this embodiment, the conflict resolution mechanism in step S310 is used to resolve decision-making contradictions when multiple rules are activated simultaneously, selecting the optimal rule through confidence comparison. Here, rule confidence refers to the matching probability between the rule and the actual environment, Conf(R) a ), Conf(R) β ), Conf(R) e ) represent the main rule R respectively a and alternative rules R β R e The confidence level; conflict resolution refers to the process of selecting a unique rule to execute when the triggering conditions of multiple rules are met simultaneously, through preset logic. The core purpose of this mechanism is to ensure the uniqueness and rationality of the decision. Its implementation idea is based on the confidence level ranking when the rules are generated, selecting the rule with the highest confidence level, avoiding the subjectivity of traditional "priority preset".

[0097] In one possible implementation, if a special case occurs where two rules have equal confidence levels (such as Conf(R)), a ) = Conf(R β =0.8), rule complexity can be introduced as a secondary criterion for decision-making (choosing rules with fewer conditions) to improve decision-making efficiency.

[0098] For example, in one feasible implementation, when the drone detects "air pressure < 70 kPa" and "humidity > 90%", the main rule R... a (Confidence level 0.7) and alternative rule R β (Confidence level 0.8) Simultaneous triggering, determine Conf(R) using the conflict resolution formula. β )>Conf(R a ), and Conf(R) β Other alternative rules (such as R) e (Confidence level 0.6), ultimately activating R β Perform temperature threshold adjustment.

[0099] Step S400: The effective rules are verified using a dual-engine adversarial architecture, and the rules and directed acyclic graph topology are optimized through an arbitration mechanism to generate the final decision rules.

[0100] It should be noted that in this embodiment, step S400 is the optimization and verification stage of the decision rules, the purpose of which is to improve the robustness of the rules and optimize the underlying topology through adversarial verification. Here, "dual-engine adversarial architecture" refers to a parallel verification structure in which the main engine generates the rule to be verified and the adversarial engine generates the challenge rule; "valid rule" refers to the decision rule that passes the confidence judgment in step S300; "arbitration mechanism" refers to a method of determining the security of the main rule and the challenge rule through risk probability assessment; and "final decision rule" refers to the final executable instruction after adversarial verification optimization. This step, by introducing adversarial verification and topology feedback mechanism, can discover the defects of the rules in extreme scenarios, realize the self-evolution of the decision system, and solve the problem of insufficient robustness of traditional rule engines.

[0101] Additionally, it should be noted that in one possible implementation, the adversarial engine in the dual-engine adversarial architecture can adopt a directional perturbation strategy, that is, to perturb the causal chain with the highest dependency in the main rule (such as reversing the causal direction of "air pressure → heat dissipation efficiency") to verify the stability of the core logic of the rule; the risk probability calculation in the arbitration mechanism can introduce the device health parameter, for example, when the battery health is below 80%, the risk probability weight is increased by 20% to prioritize the protection of device safety.

[0102] For example, in one specific implementation, when the UAV performs an inspection mission in a typhoon environment, step S400 initiates dual-engine countermeasures: the main engine generates R_main based on valid rules as "IF△(wind speed>15m / s,10s)THEN speed decreases by 30%"; the countermeasure engine disturbs the causal chain of "wind speed→turbulence risk" and generates R_countermeasure as "IF△(wind speed>15m / s,10s)THEN speed decreases by 40%"; the arbitration mechanism calculates Risk(R) 主 Risk(R) = 0.08 对抗 ) = 0.05, and R 对抗 The causal chain of the adversarial engine was validated three times in a row, so it was incorporated into the main topology; finally, the final decision rule of "reducing the speed by 40%" was generated, which improved the safety in extreme scenarios.

[0103] Specifically, step S400 includes:

[0104] Step S410: Generate rules for the dual-engine adversarial architecture. The main engine generates decision rules R based on the current directed acyclic graph topology. 主 The adversarial engine generates challenge rules R by perturbing the causal chain. 对抗 ;

[0105] Step S420: The arbitration mechanism evaluates the rules and calculates the probability of equipment damage after the rule is triggered. The probability of risk is calculated as Risk(R) = P(damage|triggering R), where Risk(R) is the probability of risk of rule R and P(damage|triggering R) is the conditional probability of equipment damage after rule R is triggered.

[0106] Step S430: Topology and rule optimization, if Risk(R 对抗 ) <Risk(R 主 If the causal chain of the adversarial engine is valid for three consecutive times, then the causal chain of the adversarial engine will be incorporated into the main topology, as shown in the formula G. 主 =G 主 ∪G 对抗中有效链 In the formula G 主 For the optimized main topology, G 对抗中有效链 To counteract the effective causal chains generated by the engine, generate the most unfavorable scenario to verify the robustness of the rules. If the rules fail, return to step S100 to rebuild the topology.

[0107] It should be noted that in this embodiment, the sub-steps of step S400 achieve robustness verification and topology optimization of the rules through a dual-engine adversarial and arbitration mechanism. The dual-engine adversarial architecture refers to a structure where the main engine and the adversarial engine work in parallel. The main engine is the module that generates the rules to be verified based on the current topology, while the adversarial engine is the module that generates challenge rules by perturbing the topology. The arbitration mechanism is a mechanism that compares the main rules and challenge rules through risk probability assessment. Topology optimization refers to the process of incorporating the valid associations discovered by the adversarial engine into the main topology. The purpose of this step is to address the insufficient robustness of rules in extreme scenarios and improve the reliability of system decision-making through a "generation-challenge-optimization" closed loop. The implementation idea is to first construct adversarial rules, then screen for better rules through risk assessment, and finally feed back into the topology to optimize subsequent rule generation.

[0108] In one possible implementation, the conditional probability in the risk probability formula Risk(R) = P(damage|trigger R) can be obtained by training with historical failure data, such as calculating the damage occurrence frequency as an estimate of Risk(R) based on 1000 samples of "equipment damage after triggering rule R"; the worst-case scenario refers to an environment with multiple extreme parameters superimposed (such as "low air pressure + high humidity + steep slope + strong wind"), which is used to test the rule's extreme adaptability.

[0109] For example, in one feasible implementation, in a high-altitude typhoon environment, the main engine of the UAV generates R in step S410. 主 (t=48℃), the counter-engine generates R by perturbing the correlation between "humidity and heat dissipation efficiency". 对抗 (t=42℃); Step S420 calculates Risk(R) 主Risk(R) = 0.05 对抗 =0.01; Step S430 incorporates its causal chain into the main topology because R has a lower risk of adversarial action and has been verified to be effective three times in a row. The robustness of the rule is verified in the worst-case scenario test, and finally the decision rule "t = 42℃ and the speed is reduced by 20%" is generated.

[0110] In this embodiment, the present invention effectively solves the problem of environmental dimension collapse in rule triggering thresholds by characterizing the causal relationship of environmental parameters through modal logic, bootstrapping to mine causal chains, fusing spatiotemporal parameters across scales, and optimizing adversarial verification. It can significantly improve the decision-making accuracy of UAVs in unknown environments, greatly reduce human intervention, and effectively shorten the deployment time in new environments; enhance robustness in extreme scenarios, extend fault warning time, reduce misjudgment rate, and provide more reliable intelligent decision support for UAV cross-scenario operations.

[0111] Based on the above embodiments, such as Figure 2 As shown, the present invention also provides an intelligent decision support system for unmanned aerial vehicles (UAVs) to support an intelligent decision support method for UAVs according to the above embodiments. The intelligent decision support system for UAVs includes:

[0112] The modal topology self-evolution module 11 is used to construct a modal environment logical topology, define modal operators to characterize the causal relationship between environmental parameters, generate dynamically updated directed acyclic graph topologies through bootstrapping causal chain mining, and perform validity judgment on the directed acyclic graph topologies to output valid directed acyclic graph topologies.

[0113] The multi-scale modal analysis module 12 is used to perform modal analysis on multi-scale environmental parameters based on the effective directed acyclic graph topology and convert them into modal logic propositions.

[0114] The modal rule reasoning module 13 is used to generate a set of master rules and alternative rules through a modal reasoning engine based on the modal logic propositions and the topology of the effective directed acyclic graph, to process the spatiotemporal correlation using a cross-scale fusion mechanism, to judge the confidence of the rules, and to output the effective rules.

[0115] The adversarial verification optimization module 14 is used to verify the effective rules using a dual-engine adversarial architecture, optimize the rules and the directed acyclic graph topology through an arbitration mechanism, and generate the final decision rules.

[0116] In an optional embodiment, the modal topology self-evolution module 11 is further used to define modal operators, which include a necessary operator □, a possible operator ◇, and a timing operator △, wherein: the logical expression of the necessary operator □ is □(x→y), indicating that the antecedent environmental parameter x necessarily causes the consequent environmental parameter y to change; the logical expression of the possible operator ◇ is ◇(x→y), indicating that the antecedent environmental parameter x may cause the consequent environmental parameter y to change; and the logical expression of the timing operator △ is △(x→y,t), indicating that the antecedent environmental parameter x causes the consequent environmental parameter y to change within a time window t (unit: seconds).

[0117] Bootstrap causal chain mining includes converting continuous sensor data into discrete event symbols, determining causal direction through active intervention experiments, and updating the topology of the directed acyclic graph based on the co-occurrence frequency of event symbols. The topology update formula is as follows:

[0118] G(t+1)=G(t)∪{<x,y,m> |P(y|x)>θ}

[0119] In the formula, G(t) is the topology of the directed acyclic graph at time t.<x,y,m> For the newly discovered causal chain, x is the cause parameter, y is the result parameter, m is the modal operator, P(y|x) is the conditional probability, and θ is the association threshold;

[0120] The topology validity judgment uses the modal resolution principle to detect whether there are contradictory causal chains in the updated directed acyclic graph topology. If there are no contradictions, the valid topology is output. If there are contradictions, the low-confidence conflict chains are removed and the bootstrap causal chain mining is returned.

[0121] The process of converting continuous sensor data into discrete event symbols includes: extracting features from environmental parameter data collected by sensors, generating event symbols based on parameter thresholds and trends, wherein the event symbols include parameter feature values ​​and time changes; and determining the causal direction through active intervention experiments includes: actively perturbing the target environmental parameters, recording the response time sequence of the associated parameters, and determining the causal derivation direction based on the response time difference.

[0122] In an optional embodiment, the multi-scale modal analysis module 12 is further used to divide the environmental parameters into spatiotemporal scales, namely instantaneous parameters, slowly varying parameters, and spatial parameters, wherein: instantaneous parameters are parameters with a sampling frequency ≥ 10 Hz, slowly varying parameters are parameters with a sampling frequency ≤ 0.1 Hz, and spatial parameters are parameters generated based on GPS and elevation data;

[0123] Based on causal chains and modal operators in the topology of effective directed acyclic graphs, the partitioned multi-scale environment parameters are transformed into modal logic propositions containing necessary operators □, possible operators ◇, and time-series operators △.

[0124] In an optional embodiment, the modal rule reasoning module 13 is further used to generate rules for the modal reasoning engine, generating a master rule R based on modal logic propositions and the topology of a valid directed acyclic graph. a and an alternative rule set, wherein the alternative rule set includes rule R β and R e The main rule is the rule with the highest initial confidence, and the alternative rules are other candidate rules that meet the association threshold. Conflict resolution is performed using the modal resolution principle. The confidence of the main rule and the alternative rules are compared, and the rule with the highest confidence is selected as the candidate rule.

[0125] Cross-scale fusion mechanisms handle spatiotemporal correlations, including: temporal scale coupling, through formula t. 综合 =t 瞬时 ×w1+t avg ×w2 combines instantaneous and slowly varying parameters, where t 综合 t represents the timescale parameter after fusion. 瞬时 For instantaneous parameter values, t avg The mean of the slowly varying parameters is w1, the weight of the instantaneous parameters is w2, and the weight of the slowly varying parameters is w1+w2=1; the spatial-temporal correlation uses spatial parameters as constraints on the premise of the rules to generate composite rules containing spatial features.

[0126] Rule confidence is determined. If the rule confidence is ≥ 0.7, it is output as a valid rule. If the confidence is < 0.7, the process returns to step S310 to regenerate the rule.

[0127] The formula for determining conflict resolution is: if Conf(R) a )>Conf(R β )∧Conf(R a )>Conf(R e If ), then activate rule R. a In the formula Conf(R) a ), Conf(R) β ), Conf(R) e ) are respectively rules R a R β R e The confidence level.

[0128] In an optional embodiment, the adversarial verification optimization module 14 is further used to generate rules for the dual-engine adversarial architecture, wherein the main engine generates decision rules R based on the current directed acyclic graph topology. 主 The adversarial engine generates challenge rules R by perturbing the causal chain. 对抗 ;

[0129] The arbitration mechanism assesses the rules and calculates the probability of equipment damage after the rule is triggered. The risk probability formula is Risk(R) = P(damage|triggering R), where Risk(R) is the risk probability of rule R and P(damage|triggering R) is the conditional probability of equipment damage after rule R is triggered.

[0130] Topology and rule optimization, if Risk(R) 对抗 ) <Risk(R 主 If the causal chain of the adversarial engine is valid for three consecutive times, then the causal chain of the adversarial engine will be incorporated into the main topology, as shown in the formula G. 主 =G 主 ∪G 对抗中有效链 In the formula G 主 For the optimized main topology, G 对抗中有效链 To counteract the effective causal chains generated by the engine, generate the most unfavorable scenario to verify the robustness of the rules. If the rules fail, return to step S100 to rebuild the topology.

[0131] In this embodiment, the present invention effectively solves the problem of environmental dimension collapse in rule triggering thresholds by characterizing the causal relationship of environmental parameters through modal logic, bootstrapping to mine causal chains, fusing spatiotemporal parameters across scales, and optimizing adversarial verification. It can significantly improve the decision-making accuracy of UAVs in unknown environments, greatly reduce human intervention, and effectively shorten the deployment time in new environments; enhance robustness in extreme scenarios, extend fault warning time, reduce misjudgment rate, and provide more reliable intelligent decision support for UAV cross-scenario operations.

[0132] Furthermore, the UAV intelligent decision support system can run the aforementioned UAV intelligent decision support method and system. For specific implementation details, please refer to the method embodiment, which will not be repeated here.

[0133] Based on the above embodiments, such as Figure 3 As shown, the present invention also provides an electronic device, the electronic device comprising:

[0134] The processor 22 includes at least one processor 22, at least one memory 21, a communication interface 23, and a communication bus 24, wherein the processor 22 is communicatively connected to the memory 21.

[0135] In this embodiment, the memory 21 can be implemented in any suitable manner, for example, the memory 21 can be a read-only memory, a hard disk drive, a solid-state drive, or a USB flash drive, etc.; the memory 21 is used to store at least one executable instruction executed by the processor;

[0136] In this embodiment, the processor 22 can be implemented in any suitable manner. For example, the processor 22 can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) that can be executed by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers, etc.; the processor is used to execute the executable instructions to implement an unmanned aerial vehicle (UAV) intelligent decision support method as described above.

[0137] Based on the above embodiments, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described intelligent decision support method for unmanned aerial vehicles.

[0138] Those skilled in the art will recognize that the modules and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0139] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, equipment, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0140] In the several embodiments provided in this application, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or units may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or equipment, and may be electrical, mechanical, or other forms.

[0141] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0142] In addition, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0143] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program instructions, such as USB flash drives, portable hard drives, read-only storage servers, random access storage servers, magnetic disks, or optical disks.

[0144] Furthermore, it should be noted that the combination of the various technical features in this case is not limited to the combination methods described in the claims of this case or the combination methods described in the specific embodiments. All technical features described in this case can be freely combined or combined in any way, unless they contradict each other.

[0145] It should be noted that the above examples are merely specific embodiments of the present invention, and the present invention is obviously not limited to the above embodiments, with many similar variations. All modifications that can be directly derived or conceived by those skilled in the art from the content disclosed in this invention should fall within the protection scope of this invention.

[0146] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent decision support for unmanned aerial vehicles, characterized in that, The method comprises the following steps: Step S100: constructing a modalized environment logical topology, defining a modal operator for representing a causal relationship between environment parameters, generating a dynamically updated directed acyclic graph topology through bootstrap causal chain mining, judging the validity of the directed acyclic graph topology, and outputting a valid directed acyclic graph topology; Step S200: based on the valid directed acyclic graph topology, performing modal analysis on multi-scale environment parameters and converting them into modal logic propositions; Step S300: based on the modal logic propositions and the valid directed acyclic graph topology, generating a main rule and a set of alternative rules through a modal reasoning engine, processing the spatiotemporal correlation by using a cross-scale fusion mechanism, judging the confidence of the rules, and outputting valid rules; Step S400: verifying the valid rules by using a double-engine confrontation architecture, optimizing the rules and the directed acyclic graph topology through an arbitration mechanism, and generating final decision rules.

2. The UAV intelligent decision support method of claim 1, wherein, The step S100 comprises: Step S110: defining a modal operator, wherein the modal operator comprises a necessity operator □, a possibility operator ◇, and a temporal operator △, and wherein: The logical expression of the necessity operator □ is □(x→y), which means that the antecedent environment parameter x necessarily leads to the change of the consequent environment parameter y; The logical expression of the possibility operator ◇ is ◇(x→y), which means that the antecedent environment parameter x possibly leads to the change of the consequent environment parameter y; The logical expression of the temporal operator △ is △(x→y, t), which means that the antecedent environment parameter x leads to the change of the consequent environment parameter y within a time window t (unit: second); Step S120: bootstrap causal chain mining, comprising converting continuous sensor data into discrete event symbols, determining a causal direction through active intervention experiments, and updating a directed acyclic graph topology based on the co-occurrence frequency of event symbols, wherein the topology update formula is: G(t+1)=G(t)∪{<x,y,m>|P(y|x)>θ} In the formula, G(t) is the directed acyclic graph topology at time t, <x, y, m> is a newly discovered causal chain, x is a cause parameter, y is a result parameter, m is a modal operator, P(y|x) is a conditional probability, and θ is a correlation threshold; Step S130: topology validity judgment, detecting whether there is a contradictory causal chain in the updated directed acyclic graph topology through modal resolution principles, outputting a valid topology if there is no contradiction, and eliminating low-confidence conflict chains and returning to step S120 if there is a contradiction.

3. The UAV intelligent decision support method of claim 2, wherein, In the step S120, converting continuous sensor data into discrete event symbols comprises: performing feature extraction on environment parameter data collected by a sensor, generating event symbols based on parameter thresholds and change trends, wherein the event symbols contain parameter feature values and time change amounts; and determining a causal direction through active intervention experiments comprises: actively disturbing a target environment parameter, recording the response time sequence of associated parameters, and determining the causal inference direction according to the response time difference.

4. The UAV intelligent decision support method of claim 1, wherein, The step S200 comprises: Step S210: Temporal and spatial scale division is performed on the environmental parameters, and the environmental parameters are divided into instantaneous parameters, slowly changing parameters and spatial parameters. The instantaneous parameters are parameters with a sampling frequency of greater than or equal to 10 Hz, the slowly changing parameters are parameters with a sampling frequency of less than or equal to 0.1 Hz, and the spatial parameters are parameters generated based on GPS and elevation data; Step S220: Based on the causal chain and the modal operator in the effective directed acyclic graph topology, the multi-scale environmental parameters after division are converted into modal logic propositions containing the necessity operator □, the possibility operator ◇ and the time sequence operator △.

5. The method of claim 2, wherein, The step S300 includes: Step S310: the modal reasoning engine generates rules, generates the main rule R based on the modal logic proposition and the valid directed acyclic graph topology a and the alternative rule set, the alternative rule set includes rules R β and R e , wherein the main rule is the rule with the highest initial confidence, and the alternative rule is other candidate rules that meet the correlation threshold; conflict resolution is performed through the modal resolution principle, the confidence of the main rule and the alternative rule is compared, and the rule with the highest confidence is selected as the candidate rule; Step S320: A cross-scale fusion mechanism processes the space-time correlation, including: Time scale coupling, by formula t 综合 = t 瞬时 × w1 + t avg × w2 Fuses instantaneous parameter and slowly varying parameter, wherein t 综合 is fused time scale parameter, t 瞬时 is instantaneous parameter value, t avg is slowly varying parameter mean value, w1 is instantaneous parameter weight, w2 is slowly varying parameter weight, and w1 + w2 = 1; Space-time correlation, taking the spatial parameters as constraint conditions of the rule premise, to generate a composite rule containing spatial features; Step S330: Rule confidence judgment, if the rule confidence is greater than or equal to 0.7, it is regarded as an effective rule output, if the confidence is less than 0.7, it returns to step S310 to generate a rule.

6. The UAV intelligent decision support method of claim 1, wherein, In step S310, the conflict resolution determination formula is: if Conf(R) a )>Conf(R β )∧Conf(R a )>Conf(R e If ), then activate rule R. a In the formula Conf(R) a ), Conf(R) β ), Conf(R) e ) are respectively rules R a R β R e The confidence level.

7. The UAV intelligent decision support method of claim 1, wherein, The step S400 includes: Step S410: The dual-engine adversarial architecture generates rules, and the main engine generates decision rules R based on the current directed acyclic graph topology 主 The adversarial engine generates challenge rules R by perturbing the causal chain 对抗 ; Step S420: Arbitration mechanism evaluates the rule, calculates the device damage risk probability after the rule is triggered, and the risk probability formula is Risk(R) = P(damage | trigger R), wherein Risk(R) is the risk probability of the rule R, and P(damage | trigger R) is the conditional probability of device damage after the rule R is triggered; Step S430: topology and rule optimization, if Risk(R 对抗 )<Risk(R 主 ) and three consecutive verifications are valid, then the causal chain of the countermeasure engine is incorporated into the main topology, and the formula is G 主 =G 主 ∪G 对抗中有效链 , wherein G 主 is the optimized main topology, and G 对抗中有效链 is the valid causal chain generated by the countermeasure engine; the most unfavorable scenario is generated to verify the robustness of the rule, and if it is invalid, the topology is reconstructed again in step S100.

8. An intelligent decision support system for unmanned aerial vehicles, characterized in that, The method includes the following steps: A modal topology self-evolution module is configured to construct a modalized environmental logic topology, define a modal operator for representing the causal relationship between environmental parameters, generate a dynamically updated directed acyclic graph topology through a bootstrap causal chain mining, and perform an effectiveness judgment on the directed acyclic graph topology, and output an effective directed acyclic graph topology; A multi-scale modal analysis module is configured to perform modal analysis on multi-scale environmental parameters based on the effective directed acyclic graph topology, and convert the multi-scale environmental parameters into modal logic propositions; A modal rule reasoning module is configured to generate a main rule and a set of alternative rules through a modal reasoning engine based on the modal logic propositions and the effective directed acyclic graph topology, process space-time correlation through a cross-scale fusion mechanism, perform a confidence judgment on the rules, and output an effective rule; An adversarial verification optimization module is configured to verify the effective rule through a double-engine adversarial architecture, optimize the rule and the directed acyclic graph topology through an arbitration mechanism, and generate a final decision rule.

9. An electronic device, comprising: The electronic device includes: A processor and a memory, the memory being in communication connection with the processor; The memory is used to store executable instructions executed by the processor, and the processor is used to execute the executable instructions to realize the unmanned aerial vehicle intelligent decision support method and system according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer program is stored in the computer readable storage medium, and the computer program is executed by the processor to realize the unmanned aerial vehicle intelligent decision support method and system according to any one of claims 1 to 7.

Images