A method for unmanned inspection of expressways by unmanned aerial vehicles

Abstract

The application belongs to the technical field of intelligent traffic management, and provides a method for unmanned inspection of expressways by unmanned aerial vehicles, which comprises the following steps: acquiring traffic characteristic data of each road section; generating a road section accident occurrence prediction probability according to the traffic characteristic data; screening out a mobilizable unmanned aerial vehicle of each road section according to the road section accident occurrence prediction probability, and generating a mobilizable unmanned aerial vehicle set; screening out an additional road section requiring additional unmanned aerial vehicle inspection according to the road section accident occurrence prediction probability, and generating an additional road section set; taking the additional road section as an independent variable and the mobilization consumption score of the unmanned aerial vehicle in the mobilizable unmanned aerial vehicle set as a dependent variable, establishing an additional unmanned aerial vehicle optimization model, and generating an optimal additional unmanned aerial vehicle, so as to perform unmanned aerial vehicle addition for the additional road section until the quantity demand of the unmanned aerial vehicle of the additional road section is met; the application can accurately identify high-risk road sections and mobilizable unmanned aerial vehicles, optimize resource allocation, and improve inspection efficiency and response speed.

Description

Technical Field

[0001] This invention belongs to the field of intelligent traffic management technology, and in particular relates to a method for unmanned inspection of highways using drones. Background Technology

[0002] With the continuous increase in highway traffic volume and the increasing complexity of road networks, the timeliness requirements for emergency response to traffic accidents and road inspections are becoming increasingly stringent. Traditional manual inspections and fixed monitoring systems have limitations such as coverage blind spots and response delays. Drones, with their advantages of mobility, flexibility, and comprehensive field of view, have become an important means of dynamic highway inspection and accident early warning. However, current drone inspections mostly adopt a fixed-section stationing or periodic rotation mode, lacking an intelligent scheduling mechanism to dynamically adjust drone deployment based on real-time traffic risks. This results in insufficient inspection resources during high-accident periods or on high-risk sections, while resources remain idle in low-risk areas.

[0003] Currently, research on highway drone inspection and scheduling mainly focuses on route planning, communication relay, or scheduling based on simple event triggers. Existing technologies largely rely on historical accident data or fixed schedules for drone task allocation, failing to integrate multi-dimensional traffic characteristic data (such as traffic flow, vehicle speed, and weather) in real time to dynamically predict the probability of road segment accidents. At the resource scheduling level, existing methods often treat drones as independent, static resource points, failing to establish a cross-segment drone collaborative dispatch mechanism, and also failing to comprehensively consider the optimal dispatch scheme under practical constraints such as dispatch distance and endurance.

[0004] However, existing technologies have significant shortcomings: First, they lack dynamic prediction models for road accident probabilities based on real-time traffic characteristics, making it impossible to accurately identify road sections that require key inspections; second, they lack a mechanism for identifying and screening drones, making it impossible to release inspection resources from low-risk road sections; and third, the decision-making process for increasing drone deployment is simplistic, failing to consider multiple factors such as deployment costs and endurance constraints, resulting in low scheduling efficiency and uneven resource utilization. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method for unmanned inspection of highways using drones, thus solving the aforementioned problems.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for unmanned inspection of highways using drones, the method specifically comprising:

[0007] Assign several drones to each section of the highway to acquire traffic characteristic data for each section;

[0008] Based on traffic characteristic data, generate a predicted probability of road segment accidents.

[0009] Based on the predicted probability of road accidents, select the available drones for each road segment and generate a set of available drones. The selection method is as follows: preset a threshold for the probability of road accidents. If the predicted probability of road accidents is less than the threshold, then several of the drones allocated to that road segment are available drones.

[0010] Based on the predicted probability of road accidents, additional road sections that need to be inspected by drones are selected, and a set of additional road sections is generated.

[0011] Based on the set of available drones and the set of additional routes, with the additional routes as the independent variable and the dispatch consumption score of drones in the set of available drones as the dependent variable, an optimization model for additional drones is established to generate the optimal additional drones.

[0012] Based on the optimal deployment of additional drones, additional drones are deployed to the additional road sections until the number of drones required for the additional road sections is met; the number of drones required for the additional road sections is determined based on the predicted probability of accidents occurring on the additional road sections.

[0013] Based on the above technical solutions, the present invention also provides the following optional technical solutions:

[0014] Further technical solutions: The specific methods for generating the predicted probability of accidents occurring on the aforementioned road section include:

[0015] Through the formula: ;

[0016] Generate the predicted probability of accidents occurring on the generated road section ;

[0017] In the formula, This represents the normalized value of the i-th traffic feature data for road segment j. This represents the weight coefficient of the i-th traffic feature of road segment j. This represents the accident frequency score for road segment j in historical data. This represents the prediction accuracy score for the probability of road accidents occurring on road segment j in historical data.

[0018] Further technical solution: The specific method for obtaining the accident frequency score of road segment j includes:

[0019] Through the formula: ;

[0020] Generate an accident frequency score for road segment j. ;

[0021] In the formula, This represents the frequency of accidents occurring on road segment j in historical data. This represents the minimum frequency of accidents across all road sections in historical data. This represents the maximum frequency of accidents across all road sections in historical data.

[0022] Further technical solution: The method for obtaining the prediction accuracy score of the probability of road accidents occurring in road segment j specifically includes:

[0023] Through the formula: ;

[0024] The prediction accuracy score for the probability of road accidents occurring in road segment j. ;

[0025] In the formula, This represents the prediction accuracy of the probability of an accident occurring on road segment j based on historical data. This represents the prediction accuracy threshold.

[0026] Further technical solutions: The specific method for generating the optimal additional drones includes:

[0027] Based on the set of available drones and the set of additional dispatch routes, the dispatch consumption characteristic data of available drones being dispatched to additional dispatch routes is obtained, and a dispatch consumption score is generated based on the dispatch consumption characteristic data of available drones being dispatched to additional dispatch routes.

[0028] Using the additional dispatch route as the independent variable and the dispatch consumption score of the drones in the available drone set as the dependent variable, an optimization model for additional drones is established to generate the optimal additional drone.

[0029] Further technical solution: The method for generating the mobilization consumption score specifically includes:

[0030] Through the formula: ;

[0031] Generate mobilization consumption score ;

[0032] In the formula, This represents the normalized value of the k-th dispatch consumption characteristic data of the deployable drone y being dispatched to the additional dispatch segment j. This represents the weighting coefficient of the k-th dispatch consumption feature when a deployable drone y is dispatched to the additional dispatch segment j.

[0033] Further technical solution: The specific expression of the additional UAV optimization model is as follows: ;

[0034] In the expression, This indicates the optimal number of additional drones to be deployed to the additional deployment segment j. This represents the dispatch cost score for drone y that was deployed to the additional dispatch route j. This represents the normalized value of the flight time of the drone y deployed to the additional route segment j. , All are weighting coefficients, and , This refers to a mobilizable collection of drones.

[0035] This invention provides a method for unmanned inspection of highways using drones, which has the following advantages compared with the prior art:

[0036] This invention achieves efficient drone deployment by dynamically predicting the probability of road accidents, screening available drones and adding more road sections, and establishing an optimization model. It can dynamically predict the probability of road accidents based on real-time traffic characteristic data, accurately identify high-risk road sections and available drones, optimize resource allocation, and improve inspection efficiency and response speed. Attached Figure Description

[0037] Figure 1 This is a flowchart illustrating a method for unmanned inspection of highways using drones, as provided by the present invention.

[0038] Figure 2 This is a flowchart illustrating step S50 of the present invention.

[0039] Figure 3 This invention provides a structural schematic diagram of an unmanned inspection system for highways using drones.

[0040] Figure 4 This is a schematic diagram of the structure of the optimal drone deployment analysis unit provided by the present invention. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0042] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.

[0043] Please see Figure 1 The present invention provides a method for unmanned inspection of highways using drones, comprising the following steps:

[0044] Step S10: Assign several drones to each section of the highway to acquire traffic characteristic data for each section;

[0045] Step S20: Generate the predicted probability of road segment accidents based on traffic characteristic data;

[0046] Step S30: Based on the predicted probability of road accidents, select the available drones for each road segment and generate a set of available drones; the selection method is as follows: preset a threshold for the probability of road accidents. If the predicted probability of road accidents is less than the threshold, then several of the drones allocated to that road segment are available drones.

[0047] Step S40: Based on the predicted probability of road accidents, select the road sections that need to be inspected by additional drones and generate a set of additional road sections.

[0048] Step S50: Based on the set of available drones and the set of additional routes, with the additional routes as the independent variable and the dispatch consumption score of drones in the set of available drones as the dependent variable, establish an optimization model for additional drones and generate the optimal additional drones.

[0049] Step S60: Based on the optimal number of additional drones, additional drones are dispatched to the additional road sections until the number of drones required for the additional road sections is met; wherein, the number of drones required for the additional road sections can be determined based on the predicted probability of accidents occurring on the additional road sections.

[0050] Among them, a road segment refers to the smallest management unit in a highway that has a specific geographical scope and traffic attributes. Each road segment can independently collect traffic characteristic data and conduct accident risk assessment.

[0051] Traffic characteristic data refers to various types of data that reflect the real-time or historical traffic conditions of highway sections, such as traffic flow, average vehicle speed, vehicle density, weather conditions, road construction information, historical accident records, and vehicle violation information. These data are the basis for assessing the accident risk of road sections.

[0052] Dispatchable drones refer to drones that are currently in low-risk road sections or in an idle state, but can be reassigned to other high-risk road sections to perform additional tasks;

[0053] Additional road sections refer to road sections that are assessed based on the predicted probability of accidents and are deemed to have a high risk of accidents, requiring additional drones for inspection or stationing.

[0054] Mobilization cost score is a quantitative indicator that measures the cost or expense required for a deployable drone to move from its current location to a dispatch route, including factors such as flight distance, estimated flight time, energy consumption, and path complexity.

[0055] Specifically, in step S10, it is first necessary to allocate several drones to each segment of the highway and acquire traffic characteristic data for each segment. The allocation of drones can be done in various ways; for example, they can be evenly distributed according to the segment length to ensure that each segment has basic inspection capabilities in the initial stage. Traffic characteristic data can be acquired through manual periodic entry, real-time collection by roadside sensor networks, or uploading by vehicle-mounted devices. For example, patrol personnel can manually record information such as traffic flow and speed on the segment, or data can be periodically collected by fixed sensors such as radar and cameras deployed on the segment.

[0056] In step S20, based on the acquired traffic characteristic data, a predicted probability of accident occurrence is generated for each road segment. This probability can be generated based on simple statistical analysis, such as calculating the average historical accident frequency and combining it with a simple linear regression model of the current traffic flow; or it can be based on the current traffic conditions and historical data to assess the accident risk of each road segment.

[0057] In step S30, based on the generated predicted probability of road segment accidents, available drones for each road segment are selected, and a set of available drones is generated. The selection method includes a preset threshold for the probability of road segment accidents. If the predicted probability of a road segment accident is less than this threshold, several drones allocated to that road segment are identified as available drones. For example, a fixed probability threshold can be manually set; when the predicted probability of a road segment is lower than this threshold, all drones in that road segment are considered available. The specific number of available drones can be set manually; for example, half of all drones allocated to that road segment can be simply marked as available, or a fixed number, such as two, can be set.

[0058] In step S40, based on the predicted probability of road accidents, additional road sections requiring drone patrols are selected, and a set of additional road sections is generated. This selection process can be based on a probability threshold similar to that in step S30. For example, when the predicted probability of a road accident is higher than a certain preset higher threshold, the road section is identified as a section requiring additional drone patrols. Alternatively, management personnel can manually select road sections requiring additional drone patrols based on the predicted probability and the actual situation through manual review.

[0059] In step S50, based on the set of available drones and the set of additional dispatch routes, with the additional dispatch routes as the independent variable and the dispatch cost score of drones in the set of available drones as the dependent variable, an optimization model for additional drones is established, and the optimal additional drone is generated. This optimization model can be established using a simple greedy algorithm; for example, for each additional dispatch route, the drone closest to the available drones is selected for matching. The dispatch cost score can be simply calculated based on the straight-line distance between the drone and the additional dispatch route. The optimal additional drone can be generated by sorting according to this single distance index and selecting the drone with the shortest distance.

[0060] In step S60, based on the generated optimal deployment drones, additional drones are deployed to the designated road segment until the required number of drones for that segment is met. Drone deployment can be initiated manually, for example, by an operator manually controlling the optimal deployment drone to fly to the target road segment. The required number of drones for each road segment can be determined based on the predicted probability of accidents occurring on that segment. For example, a simple linear relationship can be established: the higher the probability, the greater the required number of drones. Alternatively, the required number of drones can be manually determined based on experience and different probability ranges.

[0061] This application enables the optimal deployment of additional drones based on the demand for drones in the additional road sections. This overall technical solution, based on dynamic prediction, intelligent screening, and optimized scheduling, effectively solves the technical problems of low drone scheduling efficiency, uneven resource utilization, and inability to accurately identify and respond to high-risk road sections in existing technologies, significantly improving the intelligence level and emergency response capability of drone inspections on highways.

[0062] Preferably, the present invention further proposes a method for generating the predicted probability of accidents occurring on the aforementioned road section, specifically including:

[0063] Through the formula: ;

[0064] Generate the predicted probability of accidents occurring on the generated road section ;

[0065] In the formula, This represents the normalized value of the i-th traffic feature data for road segment j. This represents the weight coefficient of the i-th traffic feature of road segment j. This represents the accident frequency score for road segment j in historical data. This represents the prediction accuracy score for the probability of road accidents occurring on road segment j in historical data.

[0066] Traffic feature data typically needs to be normalized before being input into the calculation formula to eliminate differences in dimensions and ensure fairness in the calculation of different features. Normalization methods can include min-max normalization, which linearly transforms the data to the [0,1] interval, or Z-score normalization, which converts the data into a distribution with a mean of 0 and a standard deviation of 1.

[0067] The weight coefficient of the i-th traffic feature of road segment j reflects the degree of influence of different traffic features on the predicted probability of accidents occurring on the road segment. For example, in rainy or snowy weather, the weight of road slipperiness may be higher than the weight of traffic flow in normal weather. These weight coefficients can be set by expert experience or learned from historical accident data through machine learning algorithms (such as regression analysis, decision trees, etc.) to optimize the performance of the prediction model.

[0068] The accident frequency score for road segment j in historical data quantifies the frequency of accidents occurring on road segment j over a past period. Historical accident frequency is an important indicator for predicting future accident risk; road segments with high frequency typically indicate higher potential risk. This score can be obtained based on historical accident records, calculated by statistically analyzing the number of accidents per unit time or unit mileage, and then undergoing appropriate normalization or score transformation.

[0069] The prediction accuracy score for the probability of road accidents occurring on road segment j in historical data is used to evaluate the historical prediction performance of the current prediction model on road segment j. If the historical prediction accuracy for a certain road segment is low, it indicates that the prediction results for that road segment may have significant uncertainty, and the prediction probability needs to be corrected. This score can be obtained by comparing historical prediction results with actual accident occurrences, calculating prediction error or accuracy indicators (such as F1 score, precision, recall, etc.), and performing corresponding score conversions.

[0070] This application's solution introduces a comprehensive prediction formula that organically combines current traffic characteristic data, historical accident frequency scores, and historical prediction accuracy scores for road segment j, all working together to generate the predicted probability of road segment accidents. Specifically, the summation term in the formula... First, a baseline accident risk value based on real-time or recent traffic characteristic data and its corresponding weighting coefficients is calculated. Then, this value is multiplied by a factor term. By incorporating the historical accident frequency score of road segment j into the calculation, road segments with a high historical accident rate receive a higher prediction probability under the same real-time traffic conditions, thus reflecting the cumulative effect of historical risk. Finally, it is multiplied by a factor term. The entire prediction result is then corrected to reflect the model's historical prediction accuracy on road segment j. If the historical prediction accuracy is high, then the factor terms... When the value is close to 1, the prediction result is less affected; if the historical prediction accuracy is low, the factor term... The value is low, so the prediction results are adjusted appropriately to reduce the risk caused by model uncertainty. This multi-dimensional, dynamically weighted prediction mechanism ensures that the generated road segment accident prediction probability not only reflects real-time traffic conditions but also integrates the inherent risk characteristics of the road segment and the predictive reliability of the model, thus significantly improving the accuracy and robustness of the prediction. In this way, the generated road segment accident prediction probability can more accurately reflect the true risk level of the road segment, providing a more reliable basis for the subsequent selection of deployable drones and the identification of additional road segments, thereby optimizing the allocation efficiency and emergency response capabilities of highway drone inspection resources.

[0071] The above technical solution, when generating the predicted probability of road segment accidents, not only considers real-time traffic characteristic data but also integrates the historical accident frequency of the road segment and the historical accuracy of the prediction model. This multi-factor fusion prediction method significantly improves the accuracy and reliability of the predicted probability of road segment accidents. Therefore, in subsequent drone inspection methods, high-risk road segments can be identified more accurately, avoiding resource waste or risk omissions caused by inaccurate predictions. For example, for road segments with a high historical accident rate but seemingly normal current traffic conditions, their predicted probability will be increased due to the weighting of historical frequency scores, prompting the system to maintain higher attention to them; while for road segments with poor prediction model performance, their predicted probability will be adjusted due to the correction of prediction accuracy scores, reducing the risk of misjudgment. This makes the allocation and deployment decisions of drone resources more scientific and reasonable, improving the overall efficiency and safety of drone inspections on highways.

[0072] Preferably, the present invention further proposes a method for obtaining the accident frequency score of road segment j, specifically including:

[0073] Through the formula: ;

[0074] Accident frequency score for road segment j ;

[0075] In the formula, This represents the frequency of accidents occurring on road segment j in historical data. This represents the minimum frequency of accidents across all road sections in historical data. This represents the maximum frequency of accidents across all road sections in historical data;

[0076] Among these, the accident frequency of road segment j in historical data is a raw indicator measuring the density of accidents on a specific road segment over a past period. It can be obtained by analyzing a historical traffic event database, counting the total number of accidents that occurred on a specific road segment over a past period (e.g., the past year, the past six months), and dividing that number by the length of that period or traffic flow to obtain the accident frequency per unit time or unit flow. Alternatively, it can be combined with accident records from traffic management departments, insurance company claims data, and sensor data (such as sudden braking, abnormal parking, etc.) to comprehensively calculate a more complete accident frequency.

[0077] The minimum accident frequency across all road segments in historical data is used as the lower bound of the normalization formula to determine the relative starting point for scoring. This is applied when calculating the accident frequency score. Previously, the system could perform a one-time statistical analysis of the historical accident frequency for all highway sections and select the minimum value. Alternatively, it could set a global, empirical minimum value, or perform stratified statistics based on factors such as region and road type to obtain the minimum values ​​for each category, thus more precisely reflecting the lowest risk level for different types of road sections.

[0078] The maximum accident frequency across all road segments in historical data is used as the upper limit of the normalization formula to determine the relative endpoint of the score. Similar to the minimum accident frequency across all road segments in historical data, the system can perform a one-time statistical analysis of the historical accident frequencies of all highway segments and select the maximum value. Alternatively, a global, empirical maximum value can be set, or stratified statistics can be performed based on factors such as region and road type to obtain their respective maximum values, thus more accurately reflecting the highest risk level of different types of road segments.

[0079] This application's solution standardizes the accident frequency score for road segment j, ensuring accuracy and comparability in generating the predicted probability of road segment accidents. Specifically, after acquiring traffic characteristic data for each road segment, the system needs to generate the predicted probability of road segment accidents based on this data and historical information. To ensure that the predicted probability of road segment accidents accurately reflects the actual risk of each road segment and avoids assessment bias caused by differences in the original data, this solution introduces a normalization process for the accident frequency of road segment j in historical data. The system first statistically analyzes the accident frequency of all highway segments within a specific historical period and identifies the minimum and maximum values ​​of the accident frequency for all segments. Subsequently, the original accident frequency of road segment j is substituted into the Min-Max normalization formula for calculation, resulting in a standardized score between 0 and 1. This standardization process effectively eliminates differences in the numerical range and dimensions of the original accident frequency data, allowing for comparison of accident frequencies of different road segments on a uniform scale, providing a more stable and representative historical accident risk assessment factor for calculating the predicted probability of road segment accidents. The accident frequency score obtained in this way, together with the normalized value and weighting coefficient of the i-th traffic feature data of road segment j, and the prediction accuracy score of the road segment accident prediction probability, accurately generates the road segment accident prediction probability. Based on these accurate prediction probabilities, the system can more rationally select deployable drones and additional road segments, and deploy additional drones, thereby optimizing the efficiency of drone inspections and accident response capabilities on highways.

[0080] The above technical solution normalizes the accident frequency of road segment j, generating an accident frequency score. This normalization effectively eliminates the differences in units and numerical ranges between the original accident frequency data of different road segments, making the accident frequency score more objective and comparable in reflecting the historical accident risk level of each road segment. This significantly improves the accuracy and stability of the predicted probability calculation for road segment accidents, avoiding prediction bias caused by fluctuations or extreme values ​​in the original data. This provides a more reliable basis for subsequent drone selection and deployment decisions, thereby optimizing the overall efficiency and accuracy of drone inspections on highways.

[0081] Preferably, the present invention further proposes a method for obtaining the prediction accuracy score of the probability of road accidents occurring in road segment j, specifically including:

[0082] Through the formula: ;

[0083] The prediction accuracy score for the probability of road accidents occurring in road segment j. ;

[0084] In the formula, This represents the prediction accuracy of the probability of an accident occurring on road segment j based on historical data. This represents the prediction accuracy threshold;

[0085] The prediction accuracy score for the probability of road accidents occurring on road segment j is a quantitative indicator used to evaluate the accuracy of the accident prediction model or algorithm for a specific road segment j based on historical data. Its purpose is to correct the final predicted probability of road segment accidents, making it more accurately reflect the reliability of the prediction model itself. This score can be obtained in various ways. For example, it can be calculated by comparing historical prediction results with actual accident occurrences and converting it into a score between 0 and 1. Alternatively, it can be obtained through expert experience or historical data analysis, subjectively or objectively evaluating the prediction accuracy of different road segments and assigning corresponding scores.

[0086] The prediction accuracy of the probability of road accidents occurring on road segment j in historical data refers to the actual accuracy achieved by the prediction model or algorithm for road segment j over a past period. This accuracy can be used as the raw input for calculating the prediction accuracy score. It can be obtained through backtesting, applying the historical prediction model to past data and comparing the prediction results with actual accident data to calculate the proportion of correct predictions or the error rate, thus obtaining the prediction accuracy of the probability of road accidents occurring on road segment j in historical data; or by using techniques such as cross-validation to train and validate the prediction model on historical datasets to evaluate its generalization ability and prediction accuracy on road segment j.

[0087] The prediction accuracy threshold is a preset benchmark value used to measure whether the prediction accuracy of road segment j has reached an acceptable level. Its purpose is to serve as a reference point for normalization and scoring calculations. This threshold can be set based on industry standards, historical experience, or expert consensus, or it can be dynamically set based on the average or median of the historical prediction accuracy of all road segments to adapt to changes in the overall prediction level.

[0088] This application's solution introduces a scoring mechanism based on prediction accuracy to quantitatively adjust the reliability of road segment accident prediction probabilities. Specifically, it uses the relationship between the prediction accuracy of road segment j in historical data and a preset prediction accuracy threshold to calculate the prediction accuracy score for road segment j using a formula. When the prediction accuracy of road segment j is higher than the threshold, A negative value leads to The value is 0, thus predicting the accuracy score. A value of 1 reflects that the higher the prediction accuracy, the smaller its correction effect on the final prediction probability. In other words, when the prediction accuracy is far above the threshold, it indicates that the prediction result is very reliable and requires little correction. Conversely, when the prediction accuracy of road segment j is below the threshold, A positive value results in a positive prediction accuracy score. A decrease in the value indicates low prediction accuracy, requiring a reduction in the prediction accuracy score. To correct its prediction probability of road accidents. The calculation process incorporates factors that affect the final predicted probability, allowing for a more significant adjustment to reflect its lower reliability. This mechanism ensures that the final predicted probability of road segment accidents not only considers traffic characteristics and historical accident frequencies but also incorporates the reliability assessment of the prediction model itself. This results in a more comprehensive and accurate reflection of the true risk level of the road segment. In this way, the scheme addresses the potential bias caused by the inadequacy of the prediction model's accuracy when generating the predicted probability of road segment accidents, making subsequent drone deployment and reinforcement decisions more scientific and rational.

[0089] The above technical solution allows for full consideration of the accuracy of the prediction model itself when generating the predicted probability of road segment accidents. By introducing a prediction accuracy score and dynamically adjusting it based on the relative relationship between the prediction accuracy of road segment j and the prediction accuracy threshold in historical data, the final predicted probability of road segment accidents can more accurately reflect the true risk level of the road segment. This avoids risk assessment bias caused by insufficient accuracy of the prediction model, improves the reliability and guidance of the prediction results, and provides a more solid data foundation for subsequent drone deployment and expansion decisions, thus helping to optimize the allocation efficiency of drone inspection resources and emergency response capabilities.

[0090] For preferred options, please refer to [link / reference]. Figure 2 The present invention further proposes a method for generating the optimal additional drones, specifically including:

[0091] Step S51: Based on the set of available drones and the set of additional dispatch routes, obtain the dispatch consumption characteristic data of available drones dispatched to additional dispatch routes, and generate a dispatch consumption score based on the dispatch consumption characteristic data of available drones dispatched to additional dispatch routes.

[0092] Step S52: Using the additional dispatch route as the independent variable and the dispatch consumption score of the drones in the set of available drones as the dependent variable, establish an optimization model for additional drones and generate the optimal additional drone.

[0093] The dispatch consumption characteristic data refers to data describing the resources or costs required to dispatch a deployable drone to a specific deployment route. This data may include the distance between the drone's current location and the deployment route, the estimated flight time, potential obstacles along the way, the power consumption during dispatch, and any additional costs that may be incurred during the dispatch process. This data can be obtained through comprehensive analysis of the drone's own sensor data, Geographic Information System (GIS) data, meteorological data, and historical operational data from the drone management system. Another method is to calculate various consumption indicators for different drones to different deployment routes using a pre-defined dispatch path planning algorithm.

[0094] The mobilization consumption score is a quantifiable indicator of the overall consumption incurred in mobilizing a specific deployable drone to a specific deployment route. This score aims to integrate various mobilization consumption characteristics into a single, comparable value to facilitate subsequent optimization decisions. It can be generated through methods such as weighted summation, fuzzy comprehensive evaluation, or machine learning models, processing and aggregating the various mobilization consumption characteristics. Another method is to assign different weights to each characteristic data point based on preset scoring rules and expert experience, and then calculate the final score.

[0095] The drone deployment optimization model is a mathematical model or algorithm used to select the optimal drone for each deployment segment from the set of available drones. The model's objective is typically to minimize total deployment costs. It can be built using optimization algorithms such as linear programming, integer programming, genetic algorithms, ant colony optimization, or deep reinforcement learning. Another approach is to employ a multi-objective optimization method, considering multiple objective functions including deployment cost, response time, and drone endurance. The optimal deployment drone is the one most suitable for deployment to a specific deployment segment, calculated by the drone deployment optimization model. This drone achieves the lowest deployment cost score or the highest overall benefit while meeting inspection requirements. It is generated directly from the model's output, assigning one or more specific drone identifiers to each deployment segment. Alternatively, the model can output a priority list, from which the system selects based on actual conditions.

[0096] This application's solution first identifies the additional dispatch routes requiring extra inspection and the available drone set for deployment. To make informed dispatch decisions, the system acquires various dispatch cost characteristics, such as distance, flight time, battery consumption, and drone type, required to deploy each available drone to each additional dispatch route. This raw data is then processed and integrated to generate a single, comparable dispatch cost score, quantifying the overall "cost" or "suboptimal level" of a particular dispatch task. Based on these scores, an optimization model for additional drone deployment is established. This model uses the additional dispatch route as the independent variable and the dispatch cost score of the drones in the available drone set as the dependent variable, aiming to find the optimal additional drone for each additional dispatch route, typically by minimizing total dispatch cost or other relevant factors. By systematically evaluating and scoring potential dispatch schemes and applying the optimization model, this application's solution ensures that the allocation of drone resources is efficient and economical, effectively addressing high-risk situations on highways. This approach goes beyond simple "first-come, first-served" or "nearest principle" scheduling strategies. Instead, it takes into account a range of factors to make truly optimal decisions, thereby improving the overall efficiency and resource utilization of the highway drone inspection system.

[0097] Through the above technical solution, this application provides a systematic, data-driven method for drone allocation. By first acquiring mobilization consumption characteristic data and generating mobilization consumption scores, and then establishing an optimization model, it achieves more accurate and efficient selection of drones to address high-risk sections of highways. Unlike scheduling methods that rely solely on simple criteria (such as distance or availability), this application's solution comprehensively considers various factors affecting mobilization costs, including distance, flight time, and the specific attributes of the drones themselves. Therefore, it ensures that the most suitable drones are scheduled, thereby minimizing resource waste and improving the effectiveness of inspections. This makes the drone inspection system more responsive and economical, directly addressing the challenge of efficiently deploying resources to dynamic high-risk areas.

[0098] Preferably, the present invention further proposes a method for generating the mobilization consumption score, specifically including:

[0099] Through the formula: ;

[0100] Generate mobilization consumption score ;

[0101] In the formula, This represents the normalized value of the k-th dispatch consumption characteristic data of the deployable drone y being dispatched to the additional dispatch segment j. This represents the weighting coefficient of the k-th dispatch consumption feature of the deployable drone y being dispatched to the additional dispatch segment j;

[0102] The mobilization cost score aims to quantify the overall cost or resource consumption required to move a specific mobilizable drone y to a specific deployment route j. This score is a key indicator for evaluating the merits of different drone mobilization schemes, and its value directly reflects the mobilization efficiency or economy. For example, this score can represent a combination of factors such as energy consumption, time cost, or potential risks required for a drone to fly from its current location to the target route.

[0103] The k-th dispatch consumption feature data for a deployable drone y to the additional dispatch segment j represents various specific factors affecting the drone's dispatch consumption. These factors could include the drone y's flight distance from its current location to the additional dispatch segment j, the estimated flight time, battery consumption, the degree of terrain complexity encountered, and the impact of specific weather conditions (such as wind speed and precipitation). Normalizing this raw data eliminates differences in units and numerical ranges between different feature data, allowing for comparison and weighted calculations on a unified scale. For example, the flight distance or the estimated flight time can be normalized to a value between 0 and 1.

[0104] The weighting coefficient of the k-th dispatch consumption feature of a deployable drone y to the additional dispatch segment j is used to characterize the relative importance of each dispatch consumption feature in the overall dispatch consumption score. These weights can be set according to the needs of the actual application scenario, historical data analysis results, or expert experience. For example, in some emergency situations, flight time may be given higher weight; while in routine inspections, battery consumption or flight distance may be more critical. These weighting coefficients can be preset or dynamically adjusted according to real-time conditions or through machine learning algorithms to adapt to different inspection tasks and environmental changes.

[0105] This application's solution generates a mobilization cost score by introducing a multi-factor weighted summation calculation model. Specifically, for each deployable UAV y and its corresponding dispatch segment j, the system collects and normalizes a series of feature data reflecting mobilization costs or efficiency. These feature data cover multiple dimensions such as flight distance, estimated flight time, battery consumption, and environmental impact, ensuring a comprehensive consideration of mobilization costs. Subsequently, the system assigns a corresponding weight coefficient to each normalized feature data, which reflects the importance of that feature in the overall mobilization cost. By multiplying these normalized feature data with their corresponding weight coefficients and summing them, a comprehensive mobilization cost score is finally obtained. This weighted summation mechanism allows mobilization cost factors of different natures and dimensions to be uniformly quantified and compared, thereby overcoming the limitations of single or a few indicators failing to accurately assess complex mobilization costs. In this way, the generated dispatch consumption score can more accurately reflect the actual consumption of dispatching drone y to the additional dispatch segment j, providing a more reliable input for the subsequent drone dispatch optimization model, thereby enabling the optimization model to make better decisions and ensuring that drone resources are allocated efficiently and reasonably.

[0106] Through the above technical solution, this application provides a systematic and multi-dimensional method for generating mobilization cost scores. This method normalizes various feature data affecting UAV mobilization efficiency and cost, assigns different weight coefficients according to their importance, and finally generates a mobilization cost score through weighted summation. This allows the assessment of mobilization cost to no longer be limited to a single or few indicators, but to comprehensively and accurately reflect the actual mobilization cost. Therefore, in subsequent UAV deployment optimization models, decisions can be made based on more accurate mobilization cost scores, thereby significantly improving the accuracy and rationality of selecting the optimal UAV deployment, effectively avoiding resource waste or inefficiency caused by inaccurate assessments, and ultimately optimizing the overall efficiency and response speed of highway UAV inspections.

[0107] Preferably, the present invention further proposes the following expression for the optimization model of the additional UAVs: ;

[0108] In the expression, This indicates the optimal number of additional drones to be deployed to the additional deployment segment j. This represents the dispatch cost score for drone y that was deployed to the additional dispatch route j. This represents the normalized value of the remaining flight time of the drone y that has been deployed to the additional route segment j. , All are weighting coefficients, and , This refers to a mobilizable collection of drones;

[0109] The drone deployment optimization model aims to select the most suitable drone from the available drone set for deployment to a specific route. Its function is to achieve intelligent and efficient drone scheduling by quantifying and weighing multiple key factors.

[0110] Optimal deployment of additional drones The objective of the model solution is to determine the most suitable drone y for each road segment j that requires additional drones. This drone y is selected from the set of available drones, and the selection criterion is to minimize the objective function value.

[0111] The normalized value of the remaining flight time of drone y, dispatched to additional route segment j, reflects drone y's remaining flight capability during mission execution. Flight time is a key indicator for drones performing inspection missions, directly affecting mission completion efficiency and whether a mid-mission return for charging is necessary. The purpose of normalization is to unify the flight time of different drones to a comparable scale; for example, it can be mapped to a range of 0 to 1, where 1 represents the longest flight time and 0 represents the shortest. Another normalization method is to calculate drone y's effective flight capability based on its remaining battery power, battery health, and expected inspection duration on additional route segment j, and then normalize the result.

[0112] , These two coefficients are used to balance the relative importance of mobilization consumption scores and normalized values ​​of endurance time in optimization decisions. By adjusting... , The value of can be adjusted based on actual needs (e.g., a greater emphasis on reducing mobilization costs or a greater emphasis on ensuring task continuity). For example, when When the size is large, the model will tend to select drones with lower mobilization costs; when When the weights are large, the model will be more inclined to select drones with longer battery life. These weighting coefficients can be set through expert experience or trained and optimized using machine learning methods, combining historical scheduling data and task completion performance.

[0113] The proposed solution involves a weighted combination of the dispatch cost score for moving drone y to the additional dispatch segment j and the normalized value of drone y's remaining flight time. The optimal additional drone is determined by minimizing this combined value. Specifically, the model uses the argmin function to find a drone y for each additional dispatch segment j from the set of available drones, achieving an optimal balance between dispatch cost and flight time. The weighting coefficients are as follows: , The system can flexibly adjust the importance of dispatch costs and battery life in decision-making based on actual scheduling strategies and task priorities. For example, in emergency situations, more emphasis may be placed on rapid response. It can be set higher; however, in regular inspections, the focus may be more on the continuity and efficiency of the task. It can be set even higher. This optimization mechanism, which comprehensively considers both deployment costs and endurance, ensures that the selected drones can not only reach the target road segment at a lower cost, but also have sufficient endurance to efficiently complete the inspection task. This avoids suboptimal decisions caused by optimizing a single indicator, and significantly improves the intelligence and practicality of drone inspection and scheduling.

[0114] Through the aforementioned technical solution, the drone deployment optimization model, when selecting the optimal drone for deployment, not only considers the cost of dispatching the drone to the deployment segment but also innovatively incorporates the drone's endurance. This comprehensive consideration avoids the limitations that may arise from making decisions based on a single indicator. For example, considering only the dispatch cost, the system might select the closest drone but with insufficient endurance, causing the drone to need to return or recharge quickly after arriving at the deployment segment, thus increasing the complexity of subsequent scheduling and the risk of mission interruption. The solution in this application, by introducing a normalized value for endurance and flexibly adjusting it in conjunction with weighting coefficients, allows the system to achieve the optimal balance between dispatch efficiency and mission continuity based on actual needs. This significantly improves the execution efficiency and reliability of drone inspection tasks, reduces secondary scheduling or mission delays caused by insufficient drone endurance, and thus ensures the timeliness and effectiveness of highway inspections.

[0115] For preferred options, please refer to [link / reference]. Figure 3 In another embodiment, the present invention also proposes a system for unmanned inspection of highways using drones. This system is used to execute the above-described method for unmanned inspection of highways using drones, specifically including:

[0116] The data acquisition unit 10 is used to allocate several drones to each section of the highway to acquire traffic characteristic data of each section.

[0117] Accident probability prediction unit 20 is used to generate the predicted probability of road segment accidents based on traffic characteristic data;

[0118] The adjustable drone screening unit 30 is used to screen the adjustable drones for each road segment based on the predicted probability of road segment accidents, and generate a set of adjustable drones. The screening method is as follows: a preset road segment accident probability threshold is set. If the predicted probability of road segment accidents is less than the road segment accident probability threshold, then several of the drones allocated to that road segment are adjustable drones.

[0119] The additional road segment screening unit 40 is used to screen out additional road segments that need to be inspected by additional drones based on the predicted probability of road segment accidents, and generate a set of additional road segments.

[0120] The optimal drone deployment analysis unit 50 is used to establish an optimization model for drone deployment based on the set of available drones and the set of deployment routes, with the deployment routes as independent variables and the deployment consumption score of drones in the set of available drones as dependent variables, and to generate the optimal drone deployment.

[0121] The inspection and dispatch unit 60 is used to dispatch additional drones to the additional road sections according to the optimal dispatch drones, until the number of drones required for the additional road sections is met; wherein, the number of drones required for the additional road sections can be determined based on the predicted probability of road accidents occurring in the additional road sections.

[0122] For preferred options, please refer to [link / reference]. Figure 4 The present invention further proposes that the optimal deployment of additional UAVs analysis unit 50 specifically includes:

[0123] The mobilization consumption analysis module 51, based on the set of mobilizable drones and the set of additional dispatch routes, is used to obtain mobilization consumption characteristic data of mobilizable drones to additional dispatch routes, and generate a mobilization consumption score based on the mobilization consumption characteristic data of mobilizable drones to additional dispatch routes.

[0124] The optimal additional drone generation module 52 is used to establish an optimization model for additional drones, with the additional route as the independent variable and the dispatch consumption score of drones in the set of available drones as the dependent variable, and generate the optimal additional drone.

[0125] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for unmanned inspection of highways using drones, characterized in that, The method specifically includes: Assign several drones to each section of the highway to acquire traffic characteristic data for each section; Based on traffic characteristic data, generate a predicted probability of road segment accidents. Based on the predicted probability of road accidents, select the available drones for each road segment and generate a set of available drones. The selection method is as follows: preset a threshold for the probability of road accidents. If the predicted probability of road accidents is less than the threshold, then several of the drones allocated to that road segment are available drones. Based on the predicted probability of road accidents, additional road sections that need to be inspected by drones are selected, and a set of additional road sections is generated. Based on the set of available drones and the set of additional routes, with the additional routes as the independent variable and the dispatch consumption score of drones in the set of available drones as the dependent variable, an optimization model for additional drones is established to generate the optimal additional drones. Based on the optimal deployment of additional drones, additional drones are deployed to the additional road sections until the number of drones required for the additional road sections is met; the number of drones required for the additional road sections is determined based on the predicted probability of accidents occurring on the additional road sections.

2. The method for unmanned inspection of highways using unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The specific methods for generating the predicted probability of accidents occurring on the aforementioned road section include: Through the formula: ; Generate the predicted probability of accidents occurring on the generated road section ; In the formula, This represents the normalized value of the i-th traffic feature data for road segment j. This represents the weight coefficient of the i-th traffic feature of road segment j. This represents the accident frequency score for road segment j in historical data. This represents the prediction accuracy score for the probability of road accidents occurring on road segment j in historical data.

3. The method for unmanned inspection of highways using unmanned aerial vehicles (UAVs) according to claim 2, characterized in that, The specific methods for obtaining the accident frequency score for road segment j include: Through the formula: ; Generate an accident frequency score for road segment j. ; In the formula, This represents the frequency of accidents occurring on road segment j in historical data. This represents the minimum frequency of accidents across all road sections in historical data. This represents the maximum frequency of accidents across all road sections in historical data.

4. The method for unmanned inspection of highways using unmanned aerial vehicles according to claim 2, characterized in that, The methods for obtaining the prediction accuracy score of the probability of road accidents occurring in road segment j specifically include: Through the formula: ; The prediction accuracy score for the probability of road accidents occurring in road segment j. ; In the formula, This represents the prediction accuracy of the probability of an accident occurring on road segment j based on historical data. This represents the prediction accuracy threshold.

5. The method for unmanned inspection of highways using unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The specific methods for generating the optimal additional drones include: Based on the set of available drones and the set of additional dispatch routes, the dispatch consumption characteristic data of available drones being dispatched to additional dispatch routes is obtained, and a dispatch consumption score is generated based on the dispatch consumption characteristic data of available drones being dispatched to additional dispatch routes. Using the additional dispatch route as the independent variable and the dispatch consumption score of the drones in the available drone set as the dependent variable, an optimization model for additional drones is established to generate the optimal additional drone.

6. The method for unmanned inspection of highways using unmanned aerial vehicles according to claim 5, characterized in that, The specific methods for generating the mobilization consumption score include: Through the formula: ; Generate mobilization consumption score ; In the formula, This represents the normalized value of the k-th dispatch consumption characteristic data of the deployable drone y being dispatched to the additional dispatch segment j. This represents the weighting coefficient of the k-th dispatch consumption feature when a deployable drone y is dispatched to the additional dispatch segment j.

7. The method for unmanned inspection of highways using unmanned aerial vehicles according to claim 5, characterized in that, The specific expression of the optimization model for increasing the number of drones is as follows: ; In the expression, This indicates the optimal number of additional drones to be deployed to the additional deployment segment j. This represents the dispatch cost score for drone y that was deployed to the additional dispatch route j. This represents the normalized value of the flight time of the drone y deployed to the additional route segment j. , All are weighting coefficients, and , This refers to a mobilizable collection of drones.

Images