Unmanned aerial vehicle flight trajectory optimization system and method based on artificial intelligence and internet of things

Abstract

The application discloses a UAV flight trajectory optimization system and method based on artificial intelligence and the Internet of Things, and relates to the technical field of UAV control.The method comprises the following steps: collecting communication delay, environmental data and UAV state in real time; analyzing historical communication delay data by using a Markov chain to predict a basic delay value in a short time in the future; correcting the basic delay value by using a machine learning model; outputting a final delay prediction result; calculating a delay risk index to judge a delay risk level; based on the delay risk level, when the delay risk level is low risk, sending a control instruction in advance according to the final delay prediction result; when the delay risk level is medium or high risk, calling a pre-trained reinforcement learning strategy library to generate an adaptive control instruction.The application constructs a multi-dimensional risk index, and provides various strategy adjustment flight trajectories according to real-time data and the risk index, thereby improving the flight stability of the UAV.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle control, in particular to an unmanned aerial vehicle flight trajectory optimization system and method based on artificial intelligence and the Internet of Things. BACKGROUND

[0002] With the rapid development of unmanned aerial vehicle technology, unmanned aerial vehicles are increasingly widely used in various industries, especially in the fields of logistics distribution, agricultural monitoring, environmental monitoring, etc. Unmanned aerial vehicle flight trajectory optimization is one of the important technologies to ensure efficient and safe flight of unmanned aerial vehicles. Currently, most unmanned aerial vehicle flight control systems rely on traditional preset flight paths and fixed control instructions, lacking sufficient intelligence and adaptability, especially in complex environments, unmanned aerial vehicles often face communication delays, environmental interference and unstable flight states, affecting the stability and efficiency of flight.

[0003] Traditional communication technology has limitations, Wi-Fi transmission range is limited, 4G / 5G cellular network has a long transmission range but has communication delay, especially control link delay which can cause safety hazards. Environmental data and unmanned aerial vehicle state data are complex and variable, and existing methods cannot fully consider the impact of these factors on communication delay, resulting in inaccurate delay prediction. In view of the risks caused by communication delay, there is a lack of effective evaluation and response mechanism, and it is difficult to provide accurate control strategies according to different risk levels to ensure the safe flight and task execution of unmanned aerial vehicles in complex situations. SUMMARY

[0004] The purpose of the present application is to provide an unmanned aerial vehicle flight trajectory optimization system and method based on artificial intelligence and the Internet of Things to solve the problems in the prior art.

[0005] To achieve the above purpose, the present application provides the following technical solution: an unmanned aerial vehicle flight trajectory optimization method based on artificial intelligence and the Internet of Things, the method comprising:

[0006] Real-time collection of communication delay, environmental data and unmanned aerial vehicle state;

[0007] Using Markov chain to analyze historical communication delay data to predict the basic delay value in the near future; using a machine learning model to analyze environmental data and unmanned aerial vehicle state to correct the predicted basic delay value; combining the basic delay value and the correction result to output the final delay prediction result;

[0008] Calculate the delay risk index to determine the delay risk level;

[0009] Based on the delay risk level, when the delay risk level is low, send the control instruction in advance according to the final delay prediction result; when the delay risk level is medium or high, call the pre-trained reinforcement learning strategy library to generate adaptive control instructions.

[0010] According to the above scheme, the historical communication delay data is analyzed by using the Markov chain to predict the basic delay value in a short future time, comprising:

[0011] Collecting historical communication delay data, arranging in time sequence, and constructing continuous historical communication delay time series data set; according to the distribution characteristics of the historical communication delay data, the continuous delay value is divided into several discrete state intervals with clear boundaries, and each interval represents a specific delay range level;

[0012] Statistical transfer frequency between different delay states in the historical data, calculate the transition probability between each pair of states, and construct the Markov state transition probability matrix;

[0013] The real-time collected communication delay is mapped to the corresponding discrete state in the Markov state transition probability matrix as the initial state of the Markov chain; based on the initial state and the state transition probability matrix, the delay state distribution in a short future time is iteratively calculated, and the delay state probability distribution in a short future time is deduced through multiple state transition simulations;

[0014] The predicted discrete state probability distribution is converted into a specific delay value range, and a time series smoothing algorithm is used to optimize the prediction result, effectively reducing the prediction fluctuation introduced by state discretization, and outputting the basic delay value.

[0015] According to the above scheme, the environment data and the state of the unmanned aerial vehicle are analyzed by using the machine learning model to correct the predicted basic delay value, comprising:

[0016] The environment data includes real-time wind speed, temperature gradient and topography; the state of the unmanned aerial vehicle includes flight height, horizontal speed and battery remaining capacity;

[0017] The environment data and the state of the unmanned aerial vehicle are converted into environment features and unmanned aerial vehicle state features;

[0018] Based on the environment features and unmanned aerial vehicle state features, a multi-dimensional input vector is constructed; the multi-dimensional input vector is input into a pre-trained gradient boosting decision tree model, and a delay correction amount is output through feature importance weighted analysis, the delay correction amount includes positive correction amount and negative correction amount;

[0019] The basic delay prediction value and the delay correction amount are dynamically weighted and fused, wherein the weight coefficient is dynamically adjusted according to the gradient boosting decision tree model confidence and feature stability; the gradient boosting decision tree model confidence is obtained by the prediction error of the gradient boosting decision tree model; the feature stability is evaluated by the feature importance variance in the sliding window;

[0020] The deviation between the actual communication delay and the prediction result is recorded, and when the deviation continuously exceeds a set threshold, an incremental learning update mechanism of the gradient boosting decision tree model is triggered to adaptively optimize and correct the parameters.

[0021] According to the above scheme, the conversion of the environment data and the unmanned aerial vehicle state into environment features and unmanned aerial vehicle state features comprises:

[0022] The mean and standard deviation of the wind speed data in a sliding time window are calculated to generate a wind speed stability feature; a linear regression of the temperature data and real-time height data is calculated to extract a vertical temperature gradient feature; and a complexity feature of the topography is quantified based on the topography.

[0023] The height change acceleration feature of the flight height is calculated; the horizontal speed is decomposed into longitudinal and lateral speed component features through coordinate transformation; and a discharge acceleration feature is extracted by fitting a quadratic curve to the battery power time series data.

[0024] According to the above scheme, the combination of the basic delay value and the correction result to output the final delay prediction result comprises:

[0025] A correction amount constraint is applied to the basic delay value to generate a preliminary fusion result, and positive and negative compensation intervals are distinguished according to the directionality feature of the correction amount;

[0026] A dynamic confidence evaluation mechanism is used to classify the fusion result according to its reliability, which comprises calculating the mean square error of the prediction value and the actual delay in the last three times, and combining the feature stability index to generate a confidence score;

[0027] A sliding time window smoothing algorithm is applied to eliminate short-term fluctuations in the prediction result, and an abnormal prediction value is logically checked in combination with the current network topology state to generate a delay prediction range with a confidence interval;

[0028] When the prediction result confidence is higher than a set threshold, the final prediction value is directly output; and when the confidence is lower than the threshold, a redundant communication link verification mechanism is triggered.

[0029] The redundant communication link verification mechanism comprises automatically activating at least one backup communication link when the delay prediction confidence of the main communication link is lower than a set threshold; synchronously sending probe data packets and collecting real-time communication delay data through the main and backup links; and performing weighted fusion calculation on the multi-link measurement results, wherein the weights are dynamically allocated according to the link signal strength, historical stability and environmental adaptability; and comparing the fusion result with the original prediction value, when the difference exceeds a fault tolerance threshold, the verification value is used and the main link is marked as abnormal, and the communication strategy is updated.

[0030] According to the above scheme, the delay risk index is calculated, and the delay risk level is judged, comprising:

[0031] Based on the final delay prediction result, the deviation degree of the predicted delay from the safety threshold is calculated, the fluctuation amplitude of the delay prediction result in the recent time window is counted, the environmental features and the unmanned aerial vehicle state features are extracted, and a multi-dimensional risk element set is constructed, which contains the deviation degree, the fluctuation amplitude, the environmental features and the unmanned aerial vehicle features; and the weight distribution ratio is dynamically adjusted according to the current task type of the unmanned aerial vehicle, and the standardized risk index value is output;

[0032] A multi-level delay risk threshold interval is set, and the calculated risk index value is mapped to the corresponding delay risk level; the delay risk level includes low risk, medium risk and high risk;

[0033] The low risk is that the predicted delay does not exceed the safety threshold and the fluctuation is gentle, the medium risk is that the predicted delay exceeds the safety threshold but the standby link is available, and the high risk is that the predicted delay exceeds the safety threshold and there is no reliable standby link.

[0034] According to the above scheme, when the delay risk level is low, the control instruction is sent in advance according to the final delay prediction result, comprising:

[0035] The basic compensation time is determined according to the final delay prediction value, and the compensation amount is adjusted in combination with the current motion state of the unmanned aerial vehicle; the higher the flight speed, the longer the compensation time, so as to ensure the timeliness of the instruction when flying at high speed; the control instruction is added with a time stamp and a serial number mark; after the unmanned aerial vehicle receives the control instruction, the instruction is executed according to the time stamp and the serial number mark, so as to avoid control errors caused by out-of-order or delay; at the same time, the instruction execution state is monitored, the matching degree of the instruction sending time and the predicted communication delay is recorded, the actual instruction execution time of the unmanned aerial vehicle is monitored, and the deviation between the actual instruction execution time and the expected time is compared, and the deviation data is updated to the prediction error accumulation record in synchronization; the prediction accuracy is continuously optimized through incremental learning by accumulating the prediction error data.

[0036] According to the above scheme, when the delay risk level is medium or high, a pre-trained reinforcement learning strategy library is called to generate adaptive control instructions, comprising:

[0037] According to the current risk level, in combination with the real-time collected environmental data and the state of the unmanned aerial vehicle, a candidate strategy of the corresponding risk level is matched from the pre-trained reinforcement learning strategy library; the expected income and the execution cost of the candidate strategy are evaluated, the candidate strategy with the highest comprehensive score is selected as the preliminary strategy; the expected income includes communication reliability, flight safety and task efficiency, and the execution cost includes energy consumption and operation risk;

[0038] Integrate and encode environmental data and UAV state into a standard input vector for a reinforcement learning model; based on the standard input vector, use the reinforcement learning model to generate a preliminary control instruction sequence, and combine with the dynamics constraints to check the feasibility of the preliminary control instruction;

[0039] Apply dynamics constraints and obstacle avoidance rules to the instructions that pass the feasibility check; the dynamics constraints include maximum acceleration and angular velocity limits; the obstacle avoidance rules include maintaining a minimum safety distance from obstacles;

[0040] Output adaptive control instructions with safety margins; the safety margins take into account various uncertainty factors and potential risks, providing additional safety for UAV flight, ensuring that the UAV can still safely and stably perform tasks in complex and variable environments.

[0041] According to the above scheme, the pre-trained reinforcement learning strategy library includes:

[0042] Establish a UAV dynamics model and a communication environment model to simulate various risk scenarios; use deep reinforcement learning algorithms to generate strategies through virtual trial and error training;

[0043] According to the risk level classification, the medium risk strategy focuses on communication recovery, and the high risk strategy focuses on emergency risk avoidance; the medium risk strategy includes communication frequency band switching strategy, flight speed adjustment strategy and data transmission compression strategy; the high risk strategy includes emergency obstacle avoidance maneuver strategy, forced return strategy and emergency landing strategy;

[0044] During the use of the strategy, record the execution effect of the strategy and dynamically update the priority of the strategy.

[0045] An unmanned aerial vehicle flight trajectory optimization system based on artificial intelligence and the Internet of Things, the system comprises: a data acquisition module, a delay prediction module, a risk assessment module and a control decision module;

[0046] The data acquisition module includes a communication data unit, an environmental data unit and a UAV state unit; the communication data unit is used to collect communication delay, packet loss rate, jitter, signal strength indicators, signal-to-noise ratio, frequency band information and retransmission information; the environmental data unit is used to collect real-time wind speed, temperature gradient and topography; the UAV state unit is used to collect the flight height, horizontal speed and remaining battery capacity of the UAV;

[0047] The delay prediction module comprises a prediction unit, a correction unit and a result fusion unit; the prediction unit generates a basic prediction value based on historical communication delay data; the correction unit dynamically corrects the basic prediction value through environmental data and unmanned aerial vehicle state characteristics; and the result fusion unit integrates the basic prediction value and the correction amount by weighting and outputs a final delay prediction result;

[0048] The risk assessment module is configured to quantify the communication delay risk level and trigger a hierarchical response strategy.

[0049] The control decision module comprises a low-risk control unit, a reinforcement learning strategy library and an instruction verification unit; the communication delay compensation unit sends control instructions in advance according to the final delay prediction result when the risk assessment unit assesses the communication delay risk as low risk; the reinforcement learning strategy library generates adaptive control instructions by calling the pre-trained reinforcement learning strategy library when the delay risk level is medium or high risk; and the instruction verification unit verifies the feasibility of the instructions by dynamic constraint and obstacle avoidance rules.

[0050] Compared with the prior art, the present application has the following advantages:

[0051] 1. The present application analyzes historical communication delay data to predict a basic delay value, and combines machine learning models to correct according to environmental and unmanned aerial vehicle state data, thereby accurately predicting communication delay.

[0052] 2. The present application constructs a multi-dimensional risk index, and provides various strategy adjustment flight trajectories according to real-time data and the risk index, thereby improving the flight stability of the unmanned aerial vehicle.

[0053] 3. The present application performs dynamic constraint and obstacle avoidance rule checking on control instructions, and outputs instructions with safety margin, thereby reducing flight accidents caused by unreasonable instructions and enhancing the reliability of the unmanned aerial vehicle flight. BRIEF DESCRIPTION OF DRAWINGS

[0054] Fig. 1 The present application is a step flow chart of the unmanned aerial vehicle flight trajectory optimization method based on artificial intelligence and the Internet of Things.

[0055] Fig. 2 The present application is a structural schematic diagram of the unmanned aerial vehicle flight trajectory optimization system based on artificial intelligence and the Internet of Things. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0057] Embodiment: As shown in the figure, the present application provides a technical solution, a UAV flight trajectory optimization method based on artificial intelligence and Internet of Things, which comprises the following steps: Figs. 1-2 S1, collecting communication delay, environmental data and UAV state in real time;

[0058]

[0059] Specifically, the environmental data includes real-time wind speed, temperature gradient and topography; the UAV state includes flight height, horizontal speed and battery remaining capacity; for example, the collected data: the current link is a 4G main link, and the latest three communication delays of the 4G main link are 120ms, 150ms and 110ms respectively; the real-time wind speed is 8.2m / s, the temperature gradient is-6.5°C / 100m, and the topography quantization value is 0.78; the flight height is 152m, the horizontal speed is 12m / s (decomposed into longitudinal speed 10.2m / s and lateral speed 6.5m / s), and the battery capacity is 65%; this is only for illustration and is not limited.

[0060] S2, using Markov chain to analyze historical communication delay data to predict the basic delay value in a short time in the future;

[0061] Specifically, collect historical communication delay data, arrange them in time sequence, and construct a continuous historical communication delay time series data set; according to the distribution characteristics of the historical communication delay data, divide the continuous delay values into a plurality of discrete state intervals with clear boundaries, each interval representing a specific delay range level, for example: discretize the historical delay data into 3 state intervals, among which, state 1: [100ms, 130ms), state 2: [130ms, 160ms), and state 3: ≥160ms; count the transition frequency between different delay states in the historical data, calculate the transition probability between each pair of states, and construct a Markov state transition probability matrix; map the real-time collected communication delay to the corresponding discrete state in the Markov state transition probability matrix as the initial state of the Markov chain; based on the initial state and the state transition probability matrix, iteratively calculate the delay state distribution in a short time in the future, and derive the delay state probability distribution in a short time in the future through multiple state transition simulations; convert the predicted discrete state probability distribution into a specific delay value range, and use a time series smoothing algorithm to optimize the prediction result, effectively reducing the prediction fluctuation introduced by state discretization, and output the basic delay value; for example: the current communication delay is 125ms, which is mapped to state 1, the predicted next state probability is 60% for state 1, and the predicted communication delay range is [100ms, 130ms); after smoothing, the output basic delay value is 118ms.

[0062] ​S3, correcting the predicted basic delay value by analyzing the environmental data and the state of the unmanned aerial vehicle through a machine learning model;

[0063] Specifically, a multi-dimensional input vector is constructed based on the environmental features and the unmanned aerial vehicle state features; for example: [wind speed stability = 0.15, temperature gradient = -6.5, terrain complexity = 0.78, height acceleration = 0.12, longitudinal velocity = 10.2, discharge acceleration = -1.2], which is only for illustration and does not limit; the multi-dimensional input vector is input into a pre-trained gradient boosting decision tree model, and a delay correction amount is output through feature importance weighted analysis, the delay correction amount including a positive correction amount and a negative correction amount; the basic delay prediction value and the delay correction amount are dynamically weighted and fused, wherein the weight coefficients are dynamically adjusted according to the gradient boosting decision tree model confidence and the feature stability; the gradient boosting decision tree model confidence is obtained by the prediction error of the gradient boosting decision tree model; the feature stability is evaluated by the feature importance variance in the sliding window; the deviation between the actual communication delay and the prediction result is recorded, and when it continuously exceeds the set threshold, the incremental learning update mechanism of the gradient boosting decision tree model is triggered to adaptively optimize the correction parameters;

[0064] Further, the mean and standard deviation of the wind speed data in the sliding time window are calculated to generate the wind speed stability feature; the linear regression of the temperature data and the real-time height data is calculated to extract the vertical temperature gradient feature; based on the terrain and topography, the complexity feature of the terrain and topography is quantified; the height change acceleration feature of the flight height is calculated; the horizontal speed is decomposed into longitudinal and lateral speed component features through coordinate transformation; the battery power time series data is fitted with a quadratic curve to extract the discharge acceleration feature.

[0065] S4, combining the basic delay value and the correction result, outputting a final delay prediction result;

[0066] Further, a correction amount constraint is applied to the base delay value, a preliminary fusion result is generated, and positive compensation and negative compensation intervals are distinguished according to the directionality characteristics of the correction amount; a dynamic confidence evaluation mechanism is used to perform confidence classification on the fusion result, the confidence classification includes calculating the mean square error of the predicted value and the actual delay of the last three times, combining the feature stability index to generate a confidence score; a sliding time window smoothing algorithm is applied to eliminate short-term fluctuations in the prediction result, and logical verification is performed on abnormal prediction values in combination with the current network topology state to generate a delay prediction range with a confidence interval; when the prediction result confidence is higher than a set threshold, the final prediction value is directly output; when the confidence is lower than the threshold, a redundant communication link verification mechanism is triggered; the redundant communication link verification mechanism includes automatically activating at least one backup communication link when the delay prediction confidence of the main communication link is lower than the set threshold; probe data packets are synchronously sent through the main and backup links, and real-time communication delay data is collected; weighted fusion calculation is performed on the multi-link measurement results, and the weights in the weighted fusion calculation are dynamically allocated according to the link signal strength, historical stability and environmental adaptability; the fusion result is compared with the original prediction value, and when the difference exceeds the fault tolerance threshold, the verification value is used and the main link is marked as abnormal, and the communication strategy is updated.

[0067] S5, calculate a delay risk index and determine a delay risk level;

[0068] Specifically, based on the final delay prediction result, the deviation degree of the predicted delay from the safety threshold is calculated, the fluctuation amplitude of the delay prediction result in the recent time window is counted, the environmental features and the unmanned aerial vehicle state features are extracted, a multi-dimensional risk factor set is constructed, the multi-dimensional risk factor set includes the deviation degree, the fluctuation amplitude, the environmental features and the unmanned aerial vehicle features; and the weight distribution ratio is dynamically adjusted according to the current task type of the unmanned aerial vehicle, and a standardized risk index value is output; a plurality of delay risk threshold intervals are set, and the calculated risk index value is mapped to the corresponding delay risk level; for example: through delay risk index calculation, the deviation from the safety threshold is-10ms, the fluctuation amplitude is the standard deviation of the delay in the sliding window, which is 18ms, the environmental risk is the terrain complexity and the wind speed, the output delay risk index is 0.35, and the delay risk level is low risk; the delay risk level includes low risk, medium risk and high risk; the low risk means that the predicted delay does not exceed the safety threshold and the fluctuation is gentle, the medium risk means that the predicted delay exceeds the safety threshold but the backup link is available, and the high risk means that the predicted delay exceeds the safety threshold and there is no reliable backup link.

[0069] S6, when the delay risk level is low risk, a control instruction is sent in advance according to the final delay prediction result;

[0070] Specifically, when the delay risk level is low risk, the control instruction is sent in advance according to the final delay prediction result, including: determining the basic compensation time according to the final delay prediction value, adjusting the compensation amount combined with the current motion state of the unmanned aerial vehicle, the higher the flight speed, the longer the compensation time, ensuring the timeliness of the instruction when flying at high speed; adding a time stamp and a serial number mark to the control instruction, and the unmanned aerial vehicle executes the instruction according to the time stamp and the serial number mark after receiving the control instruction, avoiding control errors caused by out-of-order or delay; for example: send the pitch angle adjustment instruction 140ms in advance, and the time stamp is +140ms; at the same time, the instruction execution state is monitored, the matching degree of the instruction sending time and the predicted communication delay is recorded, the actual instruction execution time of the unmanned aerial vehicle is monitored, and the deviation between the actual instruction execution time and the expected time is compared, and the deviation data is updated to the prediction error accumulation record at the same time; for example: the actual execution deviation is +5ms, and the prediction error accumulation record is updated; by continuously accumulating and analyzing the prediction error data, the communication delay prediction accuracy is continuously optimized.

[0071] S7, when the delay risk level is medium or high, a pre-trained reinforcement learning strategy library is called to generate adaptive control instructions;

[0072] Specifically, when the delay risk level is medium or high risk, a pre-trained reinforcement learning strategy library is called to generate adaptive control instructions, for example: when encountering strong wind of 15 m / s, causing the communication delay to change to 210 ms, the delay risk level is high risk; according to the current risk level, and combining the real-time collected environmental data and the state of the unmanned aerial vehicle, a candidate strategy corresponding to the risk level is matched from the pre-trained reinforcement learning strategy library; the expected return and execution cost of the candidate strategy are evaluated, and the candidate strategy with the highest comprehensive score is selected as the preliminary strategy; for example: the matched candidate strategies are strategy A: forced return, and strategy B: emergency landing; the candidate strategies are evaluated, and the evaluation score of strategy B is higher, so strategy B is selected; the expected return includes communication reliability, flight safety and task performance, and the execution cost includes energy consumption and operation risk; the environmental data and the state of the unmanned aerial vehicle are integrated and coded to convert into a standard input vector of the reinforcement learning model; based on the standard input vector, the reinforcement learning model is used to generate a preliminary control instruction sequence, and the preliminary control instruction is checked for feasibility in combination with the dynamics constraint; the instruction that passes the feasibility check is subjected to dynamics constraint and obstacle avoidance rule; for example: after strategy B is selected, a landing instruction is generated, and the dynamics constraint is a maximum descent acceleration of 2 m / s2, and the obstacle avoidance rule is to ensure that the distance from the mountain is greater than 50 m; the dynamics constraint includes maximum acceleration and angular velocity limit; the obstacle avoidance rule includes maintaining a minimum safety distance from the obstacle; the adaptive control instruction with a safety margin is output; the safety margin takes into account various uncertainty factors and potential risks, providing additional safety for the flight of the unmanned aerial vehicle, ensuring that the unmanned aerial vehicle can still safely and stably perform tasks in complex and variable environments.

[0073] Further, an unmanned aerial vehicle dynamics model and a communication environment model are established to simulate various risk scenarios; a deep reinforcement learning algorithm is used to generate strategies through virtual trial and error training; the strategies are stored according to risk levels, and the medium risk strategies focus on communication recovery, and the high risk strategies focus on emergency avoidance; the medium risk strategies include communication frequency band switching strategy, flight speed adjustment strategy and data transmission compression strategy, and the high risk strategies include emergency obstacle avoidance maneuver strategy, forced return strategy and emergency landing strategy; during the use of the strategies, the strategy execution effect is recorded, and the strategy priority is dynamically updated; for example: after strategy B is executed, the strategy execution effect is recorded as the delay is reduced to 170 ms, the task progress is delayed by 8%, and the delay risk level is still high risk, which does not achieve good results, and the strategy priority is updated, and the weight of strategy A in the same scenario is increased by 15%.

[0074] The present application provides another technical solution, an unmanned aerial vehicle flight trajectory optimization system based on artificial intelligence and Internet of Things, which comprises a data acquisition module, a delay prediction module, a risk assessment module and a control decision module.

[0075] The data collection module comprises a communication data unit, an environment data unit and a UAV state unit; the communication data unit is used to collect communication delay, packet loss rate, jitter condition, signal strength index, signal-to-noise ratio, frequency band information and retransmission information; the environment data unit is used to collect real-time wind speed, temperature gradient and topography; the UAV state unit is used to collect UAV flight height, horizontal speed and battery remaining capacity;

[0076] The delay prediction module comprises a prediction unit, a correction unit and a result fusion unit; the prediction unit generates a basic prediction value based on historical communication delay data; the correction unit dynamically corrects the basic prediction value through environment data and UAV state characteristics; the result fusion unit integrates the basic prediction value and the correction amount by weighting, and outputs a final delay prediction result;

[0077] The risk assessment module is used to quantify the communication delay risk level and trigger a hierarchical response strategy;

[0078] The control decision module comprises a low-risk control unit, a reinforcement learning strategy library and an instruction verification unit; the communication delay compensation unit sends control instructions in advance according to the final delay prediction result when the risk assessment unit assesses the communication delay risk as low risk; the reinforcement learning strategy library generates adaptive control instructions by calling the pre-trained reinforcement learning strategy library when the delay risk level is medium or high risk; the instruction verification unit verifies the feasibility of the instructions through dynamic constraints and obstacle avoidance rules.

[0079] It is apparent to those skilled in the art that the present application is not limited to the details of the foregoing exemplary embodiments, but can be implemented in other particular forms without departing from the spirit or essential characteristics of the present application. Therefore, the embodiments should be considered in all respects as illustrative and not restrictive, the scope of the present application being defined by the appended claims rather than the foregoing description, and it is intended that all changes which come within the meaning and range of equivalency of the claims are embraced therein. Any reference signs in the claims should not be construed as limiting the claims to the figures in which the reference signs are used.

Claims

1. An unmanned aerial vehicle flight trajectory optimization method based on artificial intelligence and Internet of Things, characterized in that: The method comprises: Real-time collection of communication delay, environmental data and UAV state; Markov chain analysis of historical communication delay data to predict the basic delay value in the near future; historical communication delay data is arranged in chronological order to construct a continuous historical communication delay time series dataset; according to the distribution characteristics of the historical communication delay data, the continuous delay value is divided into several discrete state intervals with clear boundaries, and each interval represents a specific delay range level; the transition frequency between different delay states in the historical data is counted, the transition probability between each pair of states is calculated, and a Markov state transition probability matrix is constructed; the real-time collected communication delay is mapped to the corresponding discrete state in the Markov state transition probability matrix as the initial state of the Markov chain; based on the initial state and the state transition probability matrix, the delay state distribution in the near future is iteratively calculated, and the delay state probability distribution in the near future is deduced through multiple state transition simulations; the discrete state probability distribution obtained by prediction is converted into a specific delay value range, and a time series smoothing algorithm is used to optimize the prediction results, effectively reducing the prediction fluctuations introduced by state discretization, and outputting the basic delay value; environmental data and UAV state are converted into environmental features and UAV state features; based on the environmental features and UAV state features, a multi-dimensional input vector is constructed; the multi-dimensional input vector is input into a pre-trained gradient boosting decision tree model, and a delay correction amount is output through feature importance weighted analysis, the delay correction amount including positive and negative correction amounts; the basic delay prediction value and the delay correction amount are dynamically weighted and fused, wherein the weight coefficients are dynamically adjusted according to the gradient boosting decision tree model confidence and feature stability; the gradient boosting decision tree model confidence is obtained through the prediction error of the gradient boosting decision tree model; the feature stability is evaluated by the variance of the feature importance in the sliding window; the deviation between the actual communication delay and the prediction result is recorded, and when it continuously exceeds the set threshold, the incremental learning update mechanism of the gradient boosting decision tree model is triggered to adaptively optimize the correction parameters; The final delay prediction result is output by combining the basic delay value and the correction result; Calculate the delay risk index to determine the delay risk level; Based on the delay risk level, when the delay risk level is low, send the control command in advance according to the final delay prediction result; when the delay risk level is medium or high, call the pre-trained reinforcement learning strategy library to generate adaptive control instructions. 2.The method of claim 1, wherein: The environmental data and UAV state are converted into environmental features and UAV state features, which comprise: Calculate the mean and standard deviation of the wind speed data within a sliding time window to generate a wind speed stability feature; calculate the linear regression of the temperature data and real-time altitude data to extract a vertical temperature gradient feature; based on the topography, quantify the complexity of the topography feature; Calculate the height change acceleration feature of the flight height; through coordinate transformation, decompose the horizontal speed into longitudinal and lateral speed component features; fit a quadratic curve to the battery power time series data to extract a discharge acceleration feature. 3.The AI and IoT-based UAV flight trajectory optimization method of claim 1, wherein: The combined base delay value and correction result is output as the final delay prediction result, including: Apply a correction amount constraint to the base delay value to generate a preliminary fusion result, and distinguish between positive compensation and negative compensation intervals according to the directionality feature of the correction amount; Through a dynamic confidence evaluation mechanism, the fusion result is classified by credibility; Apply a sliding time window smoothing algorithm to eliminate short-term fluctuations in the prediction result, and perform logical verification on abnormal prediction values in combination with the current network topology state to generate a delay prediction range with a confidence interval; When the prediction result confidence is higher than the set threshold, the final prediction value is directly output; when the confidence is lower than the threshold, a redundant communication link verification mechanism is triggered. 4.The method of claim 1, wherein the method further comprises: receiving a flight plan of the UAV; and determining a flight trajectory of the UAV based on the flight plan and the at least one flight parameter. The delay risk index is calculated to determine the delay risk level, including: Based on the final delay prediction result, calculate the deviation of the predicted delay from the safety threshold, and calculate the fluctuation amplitude of the delay prediction result in the recent time window to extract environmental features and unmanned aerial vehicle state features, construct a multi-dimensional risk factor set, and the multi-dimensional risk factor set contains deviation, fluctuation amplitude, environmental features and unmanned aerial vehicle features; and dynamically adjust the weight distribution ratio according to the current task type of the unmanned aerial vehicle, and output the standardized risk index value; Set multiple delay risk threshold intervals to map the calculated risk index value to the corresponding delay risk level; the delay risk level includes low risk, medium risk and high risk. 5.The unmanned aerial vehicle flight trajectory optimization method based on artificial intelligence and internet of things according to claim 1, characterized in that: When the delay risk level is low, the control instruction is sent in advance according to the final delay prediction result, including: Determine the base compensation time according to the final delay prediction value, adjust the compensation amount according to the current motion state of the unmanned aerial vehicle, the higher the flight speed, the longer the compensation time, to ensure the timeliness of the instruction when flying at high speed; add a timestamp and a serial number mark to the control instruction, and the unmanned aerial vehicle receives the control instruction according to the timestamp and the serial number mark to execute the instruction, avoiding control errors caused by out-of-order or delay; At the same time, monitor the instruction execution state, record the matching degree of the instruction sending time and the predicted communication delay, monitor the actual instruction execution time of the unmanned aerial vehicle, and compare the deviation between the actual instruction execution time and the expected time, and update the deviation data to the prediction error accumulation record. 6.The method of claim 1, wherein the method further comprises: receiving a flight plan of the UAV; and determining a flight trajectory of the UAV based on the flight plan and the at least one flight parameter. When the delay risk level is medium or high, a pre-trained reinforcement learning strategy library is called to generate adaptive control instructions, including: According to the current risk level, combine environmental data and unmanned aerial vehicle state to match candidate strategies of corresponding risk level from the pre-trained reinforcement learning strategy library; evaluate the expected return and execution cost of the candidate strategies, and select the candidate strategy with the highest comprehensive score as the preliminary strategy; Integrate and encode the environmental data and the unmanned aerial vehicle state into a standard input vector of a reinforcement learning model; based on the standard input vector, generate a preliminary control instruction sequence by using the reinforcement learning model, and perform feasibility checking on the preliminary control instruction in combination with a dynamic constraint; Apply a dynamic constraint and an obstacle avoidance rule to the instruction that passes the feasibility checking; the dynamic constraint includes a maximum acceleration and angular velocity limit; the obstacle avoidance rule includes maintaining a minimum safety distance from an obstacle; Output an adaptive control instruction with a safety margin. 7.The method of claim 6, wherein the method further comprises: receiving a flight plan of the UAV; and determining the flight trajectory of the UAV based on the flight plan. The pre-trained reinforcement learning strategy library includes: Establish an unmanned aerial vehicle dynamics model and a communication environment model to simulate various risk scenarios; use a deep reinforcement learning algorithm to generate a strategy through virtual trial and error training; According to the risk level, store the strategy, and focus on communication recovery for medium-risk strategies and emergency avoidance for high-risk strategies; the medium-risk strategy includes a communication frequency band switching strategy, a flight speed adjustment strategy, and a data transmission compression strategy; the high-risk strategy includes an emergency obstacle avoidance maneuvering strategy, a forced return strategy, and an emergency landing strategy; During the use of the strategy, record the execution effect of the strategy and dynamically update the priority of the strategy.

8. The UAV flight trajectory optimization system based on artificial intelligence and Internet of Things, applied to the UAV flight trajectory optimization method based on artificial intelligence and Internet of Things in any one of claims 1-7, characterized in that: The system includes a data acquisition module, a delay prediction module, a risk assessment module, and a control decision module; The data acquisition module includes a communication data unit, an environmental data unit, and an unmanned aerial vehicle state unit; the communication data unit is used to acquire communication delay, packet loss rate, jitter condition, signal strength indicator, signal-to-noise ratio, frequency band information, and retransmission information; the environmental data unit is used to acquire real-time wind speed, temperature gradient, and topography; the unmanned aerial vehicle state unit is used to acquire unmanned aerial vehicle flight height, horizontal speed, and battery remaining capacity; The delay prediction module includes a prediction unit, a correction unit, and a result fusion unit; the prediction unit generates a basic prediction value based on historical communication delay data; the correction unit dynamically corrects the basic prediction value through environmental data and unmanned aerial vehicle state characteristics; the result fusion unit integrates the basic prediction value and the correction amount by weighting to output a final delay prediction result; The risk assessment module is used to quantify the communication delay risk level and trigger a graded response strategy; The control decision module includes a low-risk control unit, a reinforcement learning strategy library, and an instruction checking unit; the communication delay compensation unit sends control instructions in advance according to the final delay prediction result when the risk assessment unit assesses the communication delay risk as low risk; the reinforcement learning strategy library generates adaptive control instructions when the delay risk level is medium or high risk by calling the pre-trained reinforcement learning strategy library; the instruction checking unit checks the feasibility of the instruction by dynamic constraint and obstacle avoidance rule.

Images