Road accident prediction method, system and device under air-ground networking based on RBF-DNN approximation

Abstract

The invention discloses a road accident prediction method, system and device under an air-ground network based on RBF-DNN approximation. The method comprises the following steps: step 1, obtaining road data through a mobile sensor and a fixed sensor of the network and processing the road data; step 2, according to the air-ground networking system model and a possible unknown fault type model, in combination with a tracking error function and a robust adaptive fault-tolerant sliding mode function, obtaining a dynamic system model with an internal actuator unknown fault; step 3, according to the obtained dynamic system model, designing an adaptive law with projection and sigma-correction and a non-Zeno event triggering mechanism, and constructing a DNN closed-loop control air-ground networking prediction control model based on RBF; 4, after the obtained prediction control model is trained, road accident prediction is carried out on the obtained road real-time data; the problem that communication link defects still exist in an existing prediction technology in a highway sudden dangerous case scene is solved.

Description

Technical Field

[0001] This invention relates to the field of traffic safety technology, and in particular to a method, system, device, and storage medium for predicting road accidents using an air-to-ground network based on RBF-DNN approximation. Background Technology

[0002] In both emergency response and accident prevention, integrated space-air-ground (SAG) collaborative sensing and communication provides rapid coverage and resilient recovery capabilities for high-risk road sections, becoming a crucial technological tool for intelligent transportation and public safety. Simultaneously, 5G / NR-V2X sidechains support direct vehicle-to-vehicle / vehicle-to-UAV communication without relying on base station coverage, enabling "temporary self-organizing networks" to maintain availability and scalability under conditions of high speed, sudden incidents, and network outages. However, strong Doppler and beam shift caused by highly dynamic motion, sudden link interruptions and random packet loss, as well as partial sensor / execution failures of field equipment, present the air-ground two-layer network with coupled challenges of "time-varying nonlinearity + random network constraints + hardware degradation." A robust adaptive control framework with provable stability and performance upper bounds is urgently needed to support QoS and safety indicators. These pain points directly relate to the crucial "golden minutes" of road emergency communication and accident risk prediction and intervention capabilities, possessing significant engineering and social value.

[0003] Despite the significant progress made in the above research, there are still gaps in the following aspects for the highly dynamic, strongly coupled, and power-constrained scenario of "sudden highway emergencies": (1) When multiple air-to-ground links experience sudden interruptions / random packet loss and hardware failures, the transmitted data fluctuates, affecting subsequent operations. The distributed collaborative control of "beam-power-time-frequency-routing-topology" lacks provable upper bounds for H∞ and stable domain estimates; (2) Most NN-enhanced controls are still insufficient in suppressing parameter drift, avoiding Zeno, and meeting communication budgets. The collaborative design of event trigger thresholds and projection / σ-correction needs a unified theoretical and implementation framework; (3) Work that places "robust fault-tolerant control" and "accident risk prediction" together in real road scenarios for collaborative evaluation is still scarce. There is a lack of causal closed-loop evidence based on real road accident data and engineering deployment. Summary of the Invention

[0004] The main technical problem solved by this invention is to provide a road accident prediction method, system, device and storage medium based on RBF-DNN approximation in air-to-ground networking, which solves the problem that existing prediction technologies still have communication link defects in the high dynamic, strong coupling and power-constrained scenario of sudden highway accidents.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: a road accident prediction method based on RBF-DNN approximation in an air-to-ground network, characterized by comprising the following steps:

[0006] Road data is acquired and processed through a network of mobile and fixed sensors.

[0007] Based on the air-to-ground networking system model and the model of possible unknown fault types, and combined with the tracking error function and the robust adaptive fault-tolerant sliding mode function, a dynamic system model with unknown internal actuator faults is obtained.

[0008] Based on the obtained dynamic system model, and by designing an adaptive law with projection and σ-correction and a non-Zeno event triggering mechanism, a DNN closed-loop control air-to-ground network predictive control model based on RBF is constructed.

[0009] After training the obtained predictive control model, road accident prediction is performed using the obtained real-time road data.

[0010] Furthermore, the process of acquiring and processing road data is as follows: real-time road data is collected using mobile and fixed sensors, and the collected data is transmitted to the back-end terminal through a network. The back-end terminal classifies the road data according to time nodes and timestamps the classified road data.

[0011] Furthermore, the air-to-ground networking system model is shown in the following equation:

[0012] (1)

[0013] In the formula, Indicates the system status; Indicates control input; Represents a positive definite matrix; Represents the inertia matrix; Represents the inertia matrix; Represents small variables in the system; Indicates unknown interfering factors; This represents a class of bounded, continuous perturbation equations that always exist in the system; Indicates the system's adjustment output; This indicates the system's trace output; Represents a known constant matrix with the appropriate number of bits.

[0014] Furthermore, the unknown fault type model is shown in the following equation:

[0015] (2)

[0016] (3)

[0017] In the formula, Represents an unknown constant. Indicates the system's control output. This indicates that the internal actuators in the system have experienced random interruptions, partial failures, or time-delay faults.

[0018] Furthermore, the dynamic system equations with unknown internal actuator faults are shown below:

[0019] (4)

[0020] (5)

[0021] In the formula, Indicates the system status; Indicates control input; Represents a positive definite matrix; Represents the inertia matrix; Represents the inertia matrix; This indicates unknown interfering factors. This represents a class of bounded, continuous perturbation equations that always exist in the system; This represents the system's adjustment output, which is an information variable set manually. This indicates the system's trace output; Represents a known constant matrix with an appropriate number of bits; Indicates the degree of suppression against unknown interference / faults. The smaller the value, the better the system's anti-interference capability; This represents the tracking error function.

[0022] Furthermore, the predictive control model for air-to-ground network based on RBF DNN closed-loop control is shown in the following equation:

[0023] (6)

[0024] (7)

[0025] In the formula, This represents the fixed control gain matrix obtained through LMI; Represent the given control gain equation; Represents a given fixed gain matrix; This represents the output term of the RBF deep neural network; This indicates that in order to overcome the approximation error of RBF deep neural networks Robust fault tolerance term.

[0026] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is: to provide a road accident prediction system based on RBF-DNN approximation in an air-to-ground network, characterized in that it includes a data processing module, a dynamic system model construction module, and a prediction and control model, wherein:

[0027] Data processing module: Real-time road data is collected using unmanned mobile data collectors and fixed data collectors. The collected data is transmitted to the back-end terminal through a network. The back-end terminal classifies the road data according to time nodes and timestamps the classified road data.

[0028] Dynamic system model construction module: Based on the air-to-ground networking system model and the model of possible unknown fault types, and combined with the tracking error function and the robust adaptive fault-tolerant sliding mode function, a dynamic system model with unknown faults of internal actuators is obtained;

[0029] Predictive control model construction module: Based on the obtained dynamic system model, and by designing an adaptive law with projection and σ-correction and a non-Zeno event triggering mechanism, a DNN closed-loop control air-to-ground network predictive control model based on RBF is constructed.

[0030] Road accident prediction module: After training the obtained predictive control model, it performs road accident prediction on the obtained real-time road data.

[0031] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is: to provide a road accident prediction device based on RBF-DNN approximation in an air-to-ground network, characterized in that it includes:

[0032] Memory, used to store computer programs;

[0033] A processor is configured to read and execute the computer program stored in the memory, wherein when the computer program is executed, the processor executes the road accident prediction method based on RBF-DNN approximation in air-to-ground networking as described in any of the preceding claims.

[0034] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is to provide a computer-readable storage medium, characterized in that: the computer-readable storage medium stores instructions, which, when the computer instructions are executed on a computer, cause the computer to execute the road accident prediction method based on RBF-DNN approximation under air-to-ground networking as described in any of the above claims.

[0035] The beneficial effects of this invention are as follows: This invention proposes and verifies a robust adaptive H∞ prediction fault-tolerant control framework embedded with RBF-DNN approximation for air-ground integrated self-organizing networks in response to sudden highway emergencies. By constructing a two-layer "UAV-road" networked control model that integrates link quality, queue evolution, and resource execution constraints, it explicitly characterizes the coupling effect of sudden communication link interruptions / packet loss / delay and actuator / sensor degradation on the closed loop. Furthermore, it designs an adaptive law with projection + σ-correction and a non-Zeno event triggering mechanism, effectively solving the data anomaly caused by communication link failures when transmitting and collecting data from various sensors in a network. Based on real road accident data, it evaluates the accident risk prediction effectiveness, demonstrating the feasibility and benefits of "control-enhanced risk prediction" in highway emergency communication and police deployment. This application eliminates the need for supplementing or processing the collected road data with missing information, and simultaneously solves the problem of communication link failures causing abnormal road data, thereby affecting the accuracy of accident prediction.

[0036] To make the above and other objects, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only eleven of the drawings in this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0038] Figure 1 This is a flowchart of the method described in this application;

[0039] Figure 2 This is a line graph of the loss-accuracy of the training set in this application;

[0040] Figure 3 This is the PR-AUC curve of the training set in this application;

[0041] Figure 4 This is the ROC-AUC curve of the training set in this application;

[0042] Figure 5 This is the regression curve of the training set in this application;

[0043] Figure 6 This is a calibration curve of the predicted probability and the actual occurrence probability of the training set in this application;

[0044] Figure 7 This is a comparison chart of the actual occurrences and predicted occurrences of events in the verification set of this application;

[0045] Figure 8 This is a clustering heatmap of future traffic accident predictions for the validation set of this application;

[0046] Figure 9 This is a structural block diagram of the system in this application;

[0047] Figure 10 This is a structural block diagram of the device in this application;

[0048] Figure 11 This is a structural block diagram of the storage medium of this application. Detailed Implementation

[0049] To facilitate understanding of the present invention, a more detailed description is provided below with reference to the accompanying drawings and specific embodiments. Preferred embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a thorough and complete understanding of the disclosure of the invention.

[0050] It should be noted that, unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items.

[0051] Please see Figure 1 , Figure 1 This invention presents a road accident prediction method based on RBF-DNN approximation in an air-to-ground network, characterized by the following steps:

[0052] S1. Obtain road data through a network of mobile and fixed sensors and process the road data;

[0053] S2. Based on the air-to-ground networking system model and the model of possible unknown fault types, and combined with the tracking error function and the robust adaptive fault-tolerant sliding mode function, a dynamic system model with unknown internal actuator faults is obtained.

[0054] S3. Based on the obtained dynamic system model, and by designing an adaptive law with projection and σ-correction and a non-Zeno event triggering mechanism, construct a DNN closed-loop control air-to-ground network predictive control model based on RBF.

[0055] S4. After training the obtained predictive control model, predict road accidents using the obtained real-time road data.

[0056] In step S1, the process of acquiring and processing road data is as follows: real-time road data is collected using mobile and fixed sensors, and the collected data is transmitted to the back-end terminal via a network. The back-end terminal classifies the road data according to time nodes and timestamps the classified road data.

[0057] Mobile data collected by mobile sensors can be collected by drones, satellites, or vehicle-mounted sensors; fixed data collected by fixed sensors is collected by fixed sensors set up along the roadside; data collected by drones, satellites, vehicle-mounted sensors, and fixed sensors communicate wirelessly through a controller.

[0058] The data is categorized by time point: data collected by various sensors are grouped together according to the same time period to form a road information dataset for that time period. The time points can be automatically matched based on clock information or manually set by personnel. Manual setting methods include: 1 minute, 1 hour, 1 day, 1 month, 1 year, etc. The time can be flexibly adjusted according to needs. Short time periods are convenient for real-time prediction, while longer time periods are convenient for overall prediction.

[0059] In step S2, the air-to-ground networking system model is shown in the following equation:

[0060] (1)

[0061] In the formula, Indicates the system status; Indicates control input; Represents a positive definite matrix; Represents the inertia matrix; Represents the inertia matrix; Represents small variables in the system; Indicates unknown interfering factors; This represents a class of bounded, continuous perturbation equations that always exist in the system; Indicates the system's adjustment output; This indicates the system's trace output; Represents a known constant matrix with the appropriate number of bits.

[0062] The fault types considered in this application include three types: random interruption, partial failure, and deadlock delay. The unknown fault type model is shown in the following formula:

[0063] (2)

[0064] (3)

[0065] In the formula, Represents an unknown constant. Indicates the system's control output. This indicates that the internal actuators in the system have experienced random interruptions, partial failures, or time-delay faults.

[0066] Furthermore, the tracking error function defined in this application is shown in the following equation:

[0067] (4)

[0068] Robust adaptive fault tolerance as defined in this application The sliding mode function is shown in the following equation:

[0069] (5)

[0070] In equation (5), It is a positive definite matrix that satisfies .

[0071] Therefore, the combined equation (1-5) for the dynamic system with an unknown fault in the internal actuator is shown below:

[0072] (6)

[0073] (7)

[0074] In the formula, Indicates the system status; Indicates control input; Represents a positive definite matrix; Represents the inertia matrix; Represents the inertia matrix; This indicates unknown interfering factors. This represents a class of bounded, continuous perturbation equations that always exist in the system; This represents the system's adjustment output, which is an information variable set manually. This indicates the system's trace output; Represents a known constant matrix with an appropriate number of bits; Indicates the degree of suppression against unknown interference / faults. The smaller the value, the better the system's anti-interference capability; This represents the tracking error function.

[0075] Among them, due to the actual testing process, the model in Since this is an uncertain term, in order to better reflect practical engineering applications, it is necessary to... Theoretical approximation is performed. This application uses an RBF neural network to approximate the theoretical approximation. To approximate. Meanwhile, according to Given the model expression, the input selection for the chosen RBF approximation network is as follows:

[0076] (8)

[0077] For the RBF neural network used in this application, the corresponding feedback regulation control law is as follows:

[0078] (9)

[0079] In the formula, for The approximate estimate.

[0080] Substituting equation (9) into equation (6) yields the following equation:

[0081] (10)

[0082] In the formula, the relevant parameters are defined as follows:

[0083] (11)

[0084] At this point, it should be pointed out that The system's adjustment output refers to information variables set by the user.

[0085] Among them, in the system unknown fault model, satisfy And for equation (7). Matrix equations with appropriate dimensions This makes the following equation true:

[0086] (12)

[0087] Next, robustness Performance metrics are defined as follows: Considering a linear time-invariant continuous-time closed-loop system, as shown in the following equation:

[0088] (13)

[0089] In the formula, Indicates the system status; An unknown external disturbance signal with bounded energy satisfies the following equation:

[0090] (14)

[0091] in, This indicates that the system output is being called. , , Both represent known constant matrices with appropriate dimensions. Defined in equation (1). This represents the system state feedback control gain matrix. Furthermore, for a given constant... If the system equation (13) has the following properties: (1) the system as a whole is asymptotically stable; (2) external disturbances To the modulated output transfer function of The norm does not exceed a given constant. That is, under zero initial conditions The following equation holds true:

[0092] (15)

[0093] That is, equation (14) is equivalent to the following equation:

[0094] (16)

[0095] That is, the system (13) is said to have robust adaptive properties. performance Equation (16) expresses the ability of the air-ground integrated highway emergency self-organizing network system to suppress unknown external disturbances. Indicates the degree of suppression against unknown interference / faults. The smaller the value, the better the system's anti-interference capability.

[0096] To ensure the generality of the RBF-based DNN closed-loop control air-to-ground network predictive control model constructed in this application, the following constraints are given.

[0097] The constraint is: considering equation (13), let ,if real matrix And suppress positive scalar This makes the following equation true:

[0098] (17)

[0099] Then the system as a whole remains stable, and satisfy In equation (17), The definition is consistent with that in equation (13), and the terms in equation (17) are consistent with those in equation (13). and There exists a term This will lead to the solution obtained by the system. performance It exhibits a certain degree of boundary conservatism.

[0100] Meanwhile, considering equation (13), if Symmetric positive definite matrix and matrix and This makes the following equation true:

[0101] (18)

[0102] Then equation (17) holds true.

[0103] Among them, in order to reduce the system The conservative generality of performance optimization continues to satisfy the following constraint: for a given symmetric matrix ,in for A dimensional matrix. Then, in equation (19)... The three conditional expressions are equivalent:

[0104] (19)

[0105] Based on the above analysis, the controller designed in this application is shown in the following formula:

[0106] (20)

[0107] In equation (20), the relevant parameters are defined as follows:

[0108] (twenty one)

[0109] In equation (21), This represents the fixed control gain matrix obtained through LMI. This represents the given control gain equation. This represents a given fixed gain matrix. This represents the output term of the RBF deep neural network. This indicates that in order to overcome the approximation error of RBF deep neural networks The robust fault-tolerance term. Other relevant parameters are defined as shown in equation (22).

[0110] (twenty two)

[0111] exist The fault-tolerant system of DNN closed-loop control based on RBF is defined as follows:

[0112] (1) Under normal operating conditions, the closed-loop system operates stably and is optimized. Performance, while being modulated output Tracking the ideal signal And there is no steady-state error, that is, the following equation holds:

[0113] (twenty three)

[0114] (2) When an unknown actuator inside the system fails, the closed-loop system remains stable, and the modulated output continues to function. Tracking the ideal signal Furthermore, there is no steady-state error, and the system will always satisfy certain conditions after a fault occurs. Performance; Next, by combining equations (7), (21), and (23), the comprehensive RBF-based DNN closed-loop control fault-tolerant system equation is shown below:

[0115] (twenty four)

[0116] In equation (24), the relevant parameters are defined as follows:

[0117] (25)

[0118] In the formula, This represents the fixed control gain matrix obtained through LMI; Represent the given control gain equation; Represents a given fixed gain matrix; This represents the output term of the RBF deep neural network; This indicates that in order to overcome the approximation error of RBF deep neural networks Robust fault tolerance term.

[0119] The following is about Perform stability analysis: provide scalar values. , , , , If for , matrix ,as well as A matrix with appropriate dimensions This makes the following equation true:

[0120] (26)

[0121] In equation (26), the relevant parameters are defined as follows:

[0122] (27)

[0123] Combining equations (26) and (27), we have Feedback controller gain Make it a Fault-tolerant controller. Meanwhile, in equation (26), The definition is shown in the following equation. The controller considered in this application is in equation (20). In The equation is given by:

[0124] (28)

[0125] In equation (29), and The definition is as follows:

[0126] (29)

[0127] And equation (29) can be obtained from LMI equation (27), while It is Small positive numbers. For a given positive constant, the following equation holds:

[0128] (30)

[0129] In equation (30), the relevant parameters are defined as follows:

[0130] (31)

[0131] When time At that time, the system state will converge to a very small set. Online fault-tolerant adjustment is performed using the robust adaptive control law shown in the following equation:

[0132] (32)

[0133] In equation (32), the constant is for A weighted coefficient and the following formula holds:

[0134] (33)

[0135] in addition, It is based on the robust adaptive control gain designed in practice. According to equation (32), it can be seen that if... Then only The following conditions are given:

[0136] (34)

[0137] Additionally, due to If it is an unknown constant, then we have: The error system is changed to the following formula:

[0138] (35)

[0139] Therefore, for the closed-loop system equation (24) and the error system equation (35), the following conditions guarantee that all signals of the system are uniformly bounded and possess suboptimal performance. Fault tolerance.

[0140] The conditions are as follows: This invention considers that the system formula (24) and the error system formula (35) satisfy the following conditions: 1) In the unknown fault model of the system, satisfy ;2) and for equation (7), Matrix equations with appropriate dimensions This makes the following equation true:

[0141] (36)

[0142] if matrix as well as Appropriate dimension matrix If equation (26) holds, then we have the following expression in equation (20): At the same time, select the corresponding controller parameters. and given by equation (28) The adaptive control law (32) can guarantee the uniform boundedness of the closed-loop fault-tolerant system and track the error. It asymptotically converges to 0, and for Having suboptimal Performance, and under normal circumstances, when When the following equation holds:

[0143] (37)

[0144] Meanwhile, in the event of actuator failure, when When the following equation holds:

[0145] (38).

[0146] Below, this invention combines the traffic accident data of Provincial Highway S*** provided by the **City Traffic Police Team and uses the prediction method designed in this application to conduct experimental verification.

[0147] The data includes 372 sets of traffic accident data. Influencing factors include road segment, temperature, weather, road surface conditions, and lighting conditions. Prediction factors include month and specific time period. The data is divided into training and validation sets in a 2:8 ratio based on time. The prediction simulation is as follows: Figures 2 to 8 As shown.

[0148] from Figure 2It is quite obvious that the validation set loss in Train Loss / ValLoss reaches its lowest point and stabilizes after several epochs, indicating normal generalization, and the lowest point is reached near the early stopping point. The two curves in Train Acc / ValAcc are close and stable, indicating no overfitting, and the curve spacing is small.

[0149] from Figure 3 It is quite obvious that PR-AUC=0.683 is a significant improvement over the baseline of p=0.5135 in the prior test set.

[0150] from Figure 4 It can be clearly seen that ROC-AUC=0.7245 indicates that the designed controller can effectively distinguish between high-risk and low-risk samples, has good ranking ability, and can rank the samples that are "more likely to be injured or killed" at the top.

[0151] from Figure 5 It can be clearly seen that, under the feedback regulation control of the controller designed in this invention, the regression coefficient of the training set is 0.97713, the regression coefficient of the validation set is 0.86681, the regression coefficient of the test set is 0.69678, and the overall regression coefficient is 0.88121. The controller has good adaptive adjustment performance and self-iterative performance.

[0152] from Figure 6 It is quite obvious that the controller, by combining previous traffic accident data from Provincial Highway S**, generally provides a predicted probability that falls within the range of... Near the diagonal.

[0153] from Figure 7 It can be clearly seen that by collecting traffic accident data from previous occurrences on Provincial Highway S in **City, and using the controller designed in this invention for fault-tolerant prediction analysis, the predicted traffic accident occurrence time is basically consistent with the overall trend of actual traffic accidents, with a prediction accuracy rate between 70-75%, laying a certain foundation for subsequent traffic accident prediction and response measures.

[0154] from Figure 8 It is readily apparent that the peak times (predicted probability ≥70%) are concentrated at 10:00 AM in January, 11:00 AM in February, 6:00, 7:00, and 10:00 AM in March, 9:00 AM in April, 6:00 and 7:00 AM in May, 9:00 AM in June, 7:00 AM in July, 5:00 AM in November, and 3:00 AM in December. By analyzing the training set data and using the validation set data, we can predict the time points when traffic accidents are most likely to occur on Provincial Highway S***, thereby optimizing the deployment of police resources.

[0155] Please see Figure 9 , Figure 9 This application provides a direct robust, non-fragile, adaptive fault-tolerant control system for an air-to-ground ad hoc network system, characterized by comprising a data processing module, a dynamic system model construction module, and a predictive control model, wherein:

[0156] Data processing module: Real-time road data is collected using unmanned mobile data collectors and fixed data collectors. The collected data is transmitted to the back-end terminal through a network. The back-end terminal classifies the road data according to time nodes and timestamps the classified road data.

[0157] Dynamic system model construction module: Based on the air-to-ground networking system model and the model of possible unknown fault types, and combined with the tracking error function and the robust adaptive fault-tolerant sliding mode function, a dynamic system model with unknown faults of internal actuators is obtained;

[0158] Predictive control model construction module: Based on the obtained dynamic system model, and by designing an adaptive law with projection and σ-correction and a non-Zeno event triggering mechanism, a DNN closed-loop control air-to-ground network predictive control model based on RBF is constructed.

[0159] Road accident prediction module: After training the obtained predictive control model, it performs road accident prediction on the obtained real-time road data.

[0160] Please see Figure 10 , Figure 10 This application provides a distributed robust neural network adaptive fault-tolerant control device for self-organizing network UAVs, comprising:

[0161] Memory is used to store computer programs.

[0162] A processor is configured to read and execute the computer program stored in the memory. When the computer program is executed, the processor executes any of the aforementioned road accident prediction methods based on RBF-DNN approximation in air-to-ground networking.

[0163] The processor executes program instructions stored in the memory to implement the steps of any of the above embodiments of the air-to-ground network road accident prediction method based on RBF-DNN approximation. In a specific implementation scenario, the distributed robust neural network adaptive fault-tolerant control device for self-organizing UAVs may include, but is not limited to, microcomputers and servers. In addition, the distributed robust neural network adaptive fault-tolerant control device for self-organizing UAVs may also include mobile devices such as laptops and tablets, without limitation.

[0164] Specifically, the processor controls itself and the memory to implement the steps of any of the above embodiments of the road accident prediction method based on RBF-DNN approximation in a ground-to-air network. The processor can also be called a CPU (Central Processing Unit). The processor may be an integrated circuit chip with signal processing capabilities. The processor can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor can be a microprocessor or any conventional processor. Furthermore, the processor can be implemented using integrated circuit chips.

[0165] Please see Figure 11 , Figure 11 This is a schematic diagram of a framework of an embodiment of the computer-readable storage medium provided in this application. The computer-readable storage medium stores program instructions that can be executed by a processor. The program instructions are used to implement the steps of any of the above embodiments of the road accident prediction method based on RBF-DNN approximation in air-to-ground networking.

[0166] In some embodiments, the functions or modules of the apparatus provided in this disclosure can be used to perform the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0167] The description of the various embodiments above tends to emphasize the differences between the various embodiments. The similarities or similarities can be referred to each other. For the sake of brevity, this application will not repeat them.

[0168] In the several embodiments provided in this application, it should be understood that the disclosed methods and apparatus can be implemented in other ways. For example, the apparatus implementations described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of devices or units may be electrical, mechanical, or other forms.

[0169] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0170] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0171] The above are merely embodiments of the present invention and are not intended to limit the scope of the patent of the present invention. Any equivalent structural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

[0172] The above are merely embodiments of the present invention and are not intended to limit the scope of the patent of the present invention. Any equivalent structural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A road accident prediction method based on RBF-DNN approximation in air-ground networking, characterized in that: The method comprises the following steps: acquiring road data through networked mobile sensors and fixed sensors and processing the road data; obtaining a dynamic system model with unknown internal actuator faults according to an air-ground networking system model and a possible unknown fault type model, in combination with a tracking error function and a robust adaptive fault-tolerant sliding mode function; constructing an RBF-based DNN closed-loop control air-ground networking predictive control model according to the obtained dynamic system model and designing an adaptive law with projection and sigma-modification and a non-Zeno event triggering mechanism; after training the obtained predictive control model, predicting road accidents according to obtained real-time road data.

2. The road accident prediction method based on RBF-DNN approximation in air-ground networking according to claim 1, characterized in that: The process of acquiring road data and processing the road data is as follows: real-time acquisition of road data through mobile sensors and fixed sensors, transmission of the acquired data to a background terminal through a networking network, classification of the road data according to time nodes by the background terminal, and time marking of the classified road data.

3. The road accident prediction method based on RBF-DNN approximation in air-ground networking according to claim 1, characterized in that: The air-ground networking system model is as shown in the following formula: （1） In the formula, represents the system state; represents the control input; represents a positive definite matrix; represents an inertia matrix; represents an inertia matrix; represents a small variable of the system; represents an unknown disturbance factor; represents a class of bounded continuous perturbation equations that always exist in the system; represents a system regulation output; represents a system tracking output; represents a known constant matrix with appropriate bit number.

4. The road accident prediction method based on RBF-DNN approximation in air-ground networking according to claim 1, characterized in that: The unknown fault type model is as shown in the following formula: （2） （3） wherein represents an unknown constant, represents the control output of the system, represents a time delay fault due to random interruption or partial failure or jamming of the internal actuators in the system.

5. The road accident prediction method based on RBF-DNN approximation in air-ground networking according to claim 1, characterized in that: The dynamic system equation with unknown internal actuator faults is as shown in the following formula: （4） （5） In the formula, represents the system state; represents the control input; represents a positive definite matrix; represents an inertia matrix; represents an inertia matrix; represents an unknown disturbance factor, represents a class of bounded continuous perturbation equations that always exist in the system; represents the system regulation output, which is an artificial information variable; represents the system tracking output; represents a known constant matrix with appropriate bit number; represents the disturbance / fault suppression degree, The smaller, the better the anti-interference ability of the system; represents the tracking error function.

6. The road accident prediction method based on RBF-DNN approximation in air-ground networking according to claim 1, characterized in that: The RBF-based DNN closed-loop control air-ground networking predictive control model is as shown in the following formula: （6） （7） In the formula, represents a fixed control gain matrix solved by LMI; represents a given control gain equation; represents a given fixed gain matrix; represents an output term of the RBF deep neural network; represents a robust fault-tolerant term for overcoming the approximation error of the RBF deep neural network .

7. A road accident prediction system based on RBF-DNN approximation in air-ground networking, characterized in that: The method comprises a data processing module, a dynamic system model construction module, and a predictive control model, wherein: The data processing module: real-time acquisition of road data through unmanned mobile data collectors and fixed data collectors, transmission of the acquired data to a background terminal through a networking network, classification of the road data according to time nodes by the background terminal, and time marking of the classified road data; The dynamic system model construction module: obtaining a dynamic system model with unknown internal actuator faults according to an air-ground networking system model and a possible unknown fault type model, in combination with a tracking error function and a robust adaptive fault-tolerant sliding mode function; The predictive control model construction module: constructing an RBF-based DNN closed-loop control air-ground networking predictive control model according to the obtained dynamic system model and designing an adaptive law with projection and sigma-modification and a non-Zeno event triggering mechanism; The road accident prediction module: after training the obtained predictive control model, predicting road accidents according to obtained real-time road data.

8. A road accident prediction device based on RBF-DNN approximation in air-ground networking, characterized by, It comprises: a memory for storing a computer program; a processor for reading and executing the computer program stored in the memory, when the computer program is executed, the processor executes the road accident prediction method under air-ground networking based on RBF-DNN approximation according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that: The computer readable storage medium stores instructions, when the computer instructions run on the computer, the computer executes the road accident prediction method under air-ground networking based on RBF-DNN approximation according to any one of claims 1-6.

Images