Vehicle driving anomaly detection method and system based on AI

Abstract

The invention relates to the technical field of vehicle management, and discloses an AI-based vehicle driving anomaly detection method and system. Comprising the following modules: a fault detection module which is used for collecting environment information, gas information and driving information, respectively generating first one-dimensional structure data, second one-dimensional structure data and third one-dimensional structure data, inputting the first one-dimensional structure data, the second one-dimensional structure data and the third one-dimensional structure data into a fault detection model, and outputting a fault type; the data acquisition module is used for acquiring comprehensive rescue information and generating second two-dimensional structure data; the rescue scheme generation module is used for inputting the second two-dimensional structure data into a rescue model and outputting a rescue scheme; and the scheme execution module is used for decoding the rescue scheme and then sending the rescue scheme to a receiving terminal of a rescue worker and a fault vehicle so as to specifically execute rescue work. According to the invention, the fault is judged by combining the gas component deviation influenced by the environment, and the rescue department plans and executes a high-efficiency rescue scheme according to the help information and the multi-aspect information.

Description

Technical Field

[0001] The present invention relates to the field of vehicle management technology, and more specifically, to an AI-based vehicle driving abnormality detection method and system. Background Art

[0002] Amphibious vehicles often face harsh environmental conditions and high-intensity workloads when performing missions, which poses severe challenges to the vehicle's power system.

[0003] Rescue amphibious vehicles are usually managed by government emergency management departments, fire departments or professional rescue organizations. These vehicles are mainly used for rescue work in emergency situations such as floods, water accidents, etc. However, when a vehicle breaks down, it may be difficult to continue the rescue work, and the staff may also be in danger. Due to the harsh environment, traditional anomaly detection methods mainly focus on the physical parameters of the engine, such as speed, temperature, etc., but it is difficult to detect some potential chemical anomalies, such as incomplete combustion, catalyst failure, etc.

[0004] Since amphibious vehicles need to work in two completely different environments, on land and on water, the engine load, combustion conditions, cooling methods, etc. when driving on the water are different from those when driving on land, which causes the composition and concentration of exhaust gas to change. Therefore, the abnormality detection system needs to be able to adapt to the multi-environment working characteristics of amphibious vehicles. Exhaust gas is a direct product of the combustion process and can reflect the working status of the engine in real time. It also has low detection cost, fast response speed, and is easy to perform real-time detection, making it suitable for large-scale applications.

[0005] However, the exhaust gas composition may change due to changes in environmental conditions such as ambient temperature, air pressure, humidity, etc., but this does not necessarily mean that the engine has a fault. If the fault detection model is too sensitive, it may misjudge normal environmental changes as faults, resulting in too many false alarms.

[0006] Due to the particularity of rescue amphibious vehicles, the surrounding environment may be land or water. When a rescue amphibious vehicle breaks down, timely rescue is required. However, when the rescue occurs on the water or in some extremely harsh environments, the rescue is difficult, the rescue time is long, and the rescue cost is high, resulting in low efficiency of subsequent rescue. Summary of the invention

[0007] The present invention provides an AI-based vehicle driving abnormality detection method and system to solve the technical problems in the related technology that the rescue environment is harsh, the fault detection model is too sensitive, the potential faults are difficult to find, the rescue environment is harsh, and the rescue efficiency is low.

[0008] The present invention provides an AI-based vehicle driving abnormality detection system, comprising the following modules:

[0009] Fault detection module: it is used to collect environmental information, gas information and driving information at n time points before the current time point, sort the environmental information to obtain No. 1 one-dimensional structure data, generate No. 2 one-dimensional structure data from the collected gas information, sort the collected driving information to obtain No. 3 one-dimensional structure data, input the No. 1 one-dimensional structure data, No. 2 one-dimensional structure data and No. 3 one-dimensional structure data into the fault detection model, and output the type of fault that occurred;

[0010] In a preferred embodiment, the fault detection module comprises the following steps:

[0011] Step 101, collecting environmental information, gas information and driving information at n time points before the current time point;

[0012] Step 102, sorting the environmental information to obtain No. 1 one-dimensional structure data, the No. 1 one-dimensional structure data includes n data items sorted by time, and the t-th data item represents the environmental information collected at the t-th moment; generating No. 2 one-dimensional structure data from the collected gas information, the No. 2 one-dimensional structure data includes n data items sorted by time, and the t-th data item represents the No. 1 two-dimensional structure data generated by the gas information collected at the t-th moment; sorting the collected driving information to obtain No. 3 one-dimensional structure data, the No. 3 one-dimensional structure data includes n data items sorted by time, and the t-th data item represents the driving information collected at the t-th moment;

[0013] The No. 1 two-dimensional structure data includes a No. 1 data matrix and a No. 1 relationship matrix. One unit of the No. 1 data matrix represents gas information of an independent object. The independent object includes various types of gases. One unit of the No. 1 data matrix only includes gas information of the independent object represented by it.

[0014] The element in the i-th row and j-th column of the No. 1 relationship matrix represents the association between the i-th unit of the No. 1 data matrix and the independent object represented by the j-th unit. If there is an association, the value of this element of the No. 1 relationship matrix is ​​1, otherwise it is 0;

[0015] Step 103, input the one-dimensional structure data No. 1, the one-dimensional structure data No. 2 and the one-dimensional structure data No. 3 into the fault detection model, the fault detection model includes a first intermediate layer, a second intermediate layer, a third intermediate layer, a fourth intermediate layer, a first feature fusion layer, a first output layer, a second output layer and a third output layer, the first intermediate layer inputs the one-dimensional structure data No. 1, outputs the first intermediate representation data to the feature fusion layer, the second intermediate representation data inputs the one-dimensional structure data No. 2, outputs the second intermediate representation data to the third intermediate layer, the third intermediate layer outputs the third intermediate representation data to the first feature fusion layer, the fourth intermediate layer inputs the one-dimensional structure data No. 3, outputs the fourth intermediate representation data to the first feature fusion layer and the third output layer, the third output layer outputs an indication of whether a fault occurs, the feature fusion layer outputs the first feature fusion representation data to the first output layer and the second output layer, the first output layer outputs information indicating gas composition deviation, and the second output layer outputs an indication of the type of fault that occurs.

[0016] Data collection module: It is used to collect comprehensive rescue information at the current time point and generate No. 2 two-dimensional structure data from the collected comprehensive rescue information;

[0017] Environmental information includes: temperature, humidity, air pressure, wind speed, rainfall, and solar radiation intensity;

[0018] Gas information includes exhaust gas information and gas composition deviation information;

[0019] The emission gas information includes: the type and concentration of the vehicle's emission gas;

[0020] Driving information includes: driving environment, engine speed, engine torque, coolant temperature, fuel injection amount, injection timing, and fuel pressure;

[0021] Comprehensive rescue information includes: faulty vehicle information, environmental information, map information, and rescue information;

[0022] Faulty vehicle information: fault type, fault location, fault environment, fault time, fuel tank level, fuel pressure, battery voltage, remaining power percentage, coolant temperature, brake fluid level, and tire pressure;

[0023] Map information: road networks, topography, hydrology, location and attributes of buildings;

[0024] Rescue information: professional classification of rescue personnel, corresponding number and location, type of rescue vehicles, corresponding number and location.

[0025] Rescue plan generation module: it is used to input the second two-dimensional structure data into the rescue model, the rescue model includes a fifth intermediate layer and a fourth output layer, the fifth intermediate layer inputs comprehensive rescue information and outputs fifth intermediate representation data to the fourth output layer, and the fourth output layer outputs the rescue plan;

[0026] Plan execution module: It is used to decode the rescue plan output by the rescue model and send it to the receiving terminals of the rescue personnel and the faulty vehicle. After receiving the information, the faulty vehicle and the rescue personnel will make specific arrangements to go to the target location for rescue work.

[0027] In a preferred embodiment, the types of faults that occur include: incomplete combustion, excessive combustion, intake duct blockage, insufficient intake air, catalyst failure, oxygen sensor failure, insufficient cooling, excessive cooling, piston ring wear, and valve wear.

[0028] In a preferred embodiment, when the probability of a fault occurring is greater than 0.5, it indicates that the fault has occurred; otherwise, the fault has not occurred; and when all faults have not occurred, it indicates that everything is normal.

[0029] In a preferred embodiment, when the probability of whether a fault occurs is greater than 0.5, it indicates that a fault occurs; otherwise, no fault occurs.

[0030] In a preferred embodiment, a method for detecting abnormal vehicle driving based on AI comprises the following steps:

[0031] Step 100, collecting environmental information, gas information and driving information at n time points before the current time point;

[0032] Step 200, sorting the environmental information to obtain a first one-dimensional structure data; generating a second one-dimensional structure data from the collected gas information; sorting the collected driving information to obtain a third one-dimensional structure data;

[0033] Step 300, inputting the first one-dimensional structure data, the second one-dimensional structure data and the third one-dimensional structure data into a fault detection model, and the fault detection model outputs an indication of the type of fault that has occurred;

[0034] Step 400, collecting comprehensive rescue information at the current time point;

[0035] Step 500, generating second two-dimensional structure data from the collected comprehensive rescue information;

[0036] Step 600, inputting the second two-dimensional structure data into the rescue model, and the rescue model output represents the rescue plan;

[0037] Step 700, decode the rescue plan output by the rescue model and send it to the receiving terminals of the rescue personnel and the faulty vehicle. After receiving the information, the faulty vehicle and the rescue personnel make specific arrangements to go to the target location for rescue work.

[0038] The beneficial effects of the present invention are as follows: the present invention obtains the gas composition deviation by analyzing the influence of the environment, and then determines the fault that has occurred in combination with the influence of the gas composition deviation. The rescue department then plans and executes a high-efficiency rescue plan based on the distress information combined with various information. The implemented high-efficiency rescue plan includes not only the rescue route of the rescue vehicle, but also the rescue route of the faulty vehicle, thereby ensuring high efficiency of the rescue. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a module schematic diagram of a vehicle driving abnormality detection system based on AI of the present invention;

[0040] Figure 2 It is a flow chart of a vehicle driving abnormality detection method based on AI of the present invention. DETAILED DESCRIPTION

[0041] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the present specification. Various examples may omit, replace, or add various processes or components as needed. In addition, the features described in some examples may also be combined in other examples.

[0042] At least one embodiment of the present invention discloses an AI-based vehicle driving abnormality detection system, such as Figure 1 As shown, it includes the following modules:

[0043] Fault detection module 100: used to collect environmental information, gas information and driving information at n time points before the current time point, sort the environmental information to obtain No. 1 one-dimensional structure data, generate No. 2 one-dimensional structure data from the collected gas information, sort the collected driving information to obtain No. 3 one-dimensional structure data, input the No. 1 one-dimensional structure data, No. 2 one-dimensional structure data and No. 3 one-dimensional structure data into the fault detection model, and output the type of fault that occurred;

[0044] In one embodiment of the present invention, the execution steps of the fault detection module include the following steps:

[0045] Step 101, collecting environmental information, gas information and driving information at n time points before the current time point;

[0046] Step 102, sorting the environmental information to obtain No. 1 one-dimensional structure data, the No. 1 one-dimensional structure data includes n data items sorted by time, and the t-th data item represents the environmental information collected at the t-th moment; generating No. 2 one-dimensional structure data from the collected gas information, the No. 2 one-dimensional structure data includes n data items sorted by time, and the t-th data item represents the No. 1 two-dimensional structure data generated by the gas information collected at the t-th moment; sorting the collected driving information to obtain No. 3 one-dimensional structure data, the No. 3 one-dimensional structure data includes n data items sorted by time, and the t-th data item represents the driving information collected at the t-th moment;

[0047] Environmental information includes: temperature, humidity, air pressure, wind speed, rainfall, and solar radiation intensity;

[0048] Gas information includes exhaust gas information and gas composition deviation information;

[0049] The emission gas information includes: the type and concentration of the vehicle's emission gas;

[0050] Driving information includes: driving environment (land or water), engine speed, engine torque, coolant temperature, fuel injection amount, injection timing, and fuel pressure;

[0051] The No. 1 two-dimensional structure data includes a No. 1 data matrix and a No. 1 relationship matrix. One unit of the No. 1 data matrix represents gas information of an independent object. The independent object includes various types of gases. One unit of the No. 1 data matrix only includes gas information of the independent object represented by it.

[0052] The element in the i-th row and j-th column of the No. 1 relationship matrix represents the association between the i-th unit of the No. 1 data matrix and the independent object represented by the j-th unit. If there is an association, the value of this element of the No. 1 relationship matrix is ​​1, otherwise it is 0;

[0053] The gases are related in the sense that: they are all gases emitted by vehicles;

[0054] In one embodiment of the present invention, the types of gas include CO, HC, O2, NOx, and CO2.

[0055] In one embodiment of the present invention, the driving environment includes land and water.

[0056] Step 103, input the one-dimensional structure data No. 1, the one-dimensional structure data No. 2 and the one-dimensional structure data No. 3 into a fault detection model, the fault detection model includes a first intermediate layer, a second intermediate layer, a third intermediate layer, a fourth intermediate layer, a first feature fusion layer, a first output layer, a second output layer and a third output layer, the first intermediate layer inputs the one-dimensional structure data No. 1, outputs the first intermediate representation data to the feature fusion layer, the second intermediate representation data inputs the one-dimensional structure data No. 2, outputs the second intermediate representation data to the third intermediate layer, the third intermediate layer outputs the third intermediate representation data to the first feature fusion layer, the fourth intermediate layer inputs the one-dimensional structure data No. 3, outputs the fourth intermediate representation data to the first feature fusion layer and the third output layer, the third output layer outputs whether a fault occurs, the feature fusion layer outputs the first feature fusion representation data to the first output layer and the second output layer, the first output layer outputs information indicating gas composition deviation, and the second output layer outputs information indicating the type of fault that occurs;

[0057] The gas composition deviation information includes: the type of gas and the corresponding gas composition deviation value;

[0058] In one embodiment of the present invention, the calculation formula of the first intermediate layer is as follows:

[0059] Long short-term memory network, the long short-term memory network includes forget gate, input gate, cell state and output gate, the tth output of the forget gate for:

[0060] ;

[0061] in, is the weight matrix of the forget gate, is the bias term of the forget gate, is the t-1th intermediate representation data and The splicing, is the sigmoid function; Represents the tth data item of the one-dimensional structure data;

[0062] The tth output of the input gate and the tth candidate cell state The calculation method is:

[0063] ;

[0064] ;

[0065] in, is the weight matrix of the input gate, is the bias term of the input gate, is the hyperbolic tangent activation function, is the weight matrix for calculating candidate cell states, is the bias term of the candidate cell state;

[0066] The t-th cell state is the state of the t-1th cell and Weighted synthesis, the calculation method is:

[0067] ;

[0068] in, Represents element-by-element multiplication operation;

[0069] The tth activation value of the output gate and the tth intermediate representation data The calculation method is:

[0070] ;

[0071] ;

[0072] in, is the weight matrix of the output gate, is the bias term of the output gate.

[0073] In one embodiment of the present invention, the calculation formula of the second intermediate layer is as follows:

[0074] ;

[0075] in The second intermediate representation data of the vth unit of the data matrix No. 1 representing the tth data item in the one-dimensional structure data No. 2, is the set of cells of the No. 1 data matrix associated with the v-th cell of the No. 1 data matrix of the t-th data item in the No. 2 one-dimensional structure data, and Respectively represent the gas information of the independent object represented by the vth and uth cells of the data matrix No. 1 of the tth data item in the one-dimensional structure data, is a learnable scalar parameter, represents the two-dimensional recognition weight parameter, Represents a multilayer perceptron.

[0076] In one embodiment of the present invention, the calculation formula of the third intermediate layer is as follows:

[0077] ;

[0078] in represents the tth third intermediate representation data, represents the t-1th third intermediate representation data, = , The second intermediate representation data of the vth unit of the data matrix No. 1 representing the tth data item in the one-dimensional structure data No. 2, is the set of cells of the No. 1 data matrix associated with the v-th cell of the No. 1 data matrix of the t-th data item in the No. 2 one-dimensional structure data, and are the first and second weight parameters, is the bias parameter and tanh is the hyperbolic tangent function.

[0079] In one embodiment of the present invention, the calculation formula of the fourth intermediate layer is as follows:

[0080] ;

[0081] ;

[0082] ;

[0083] ;

[0084] in, , , represents the first, second, and third weight parameters, , , represents the first, second and third bias parameters, represents the dot product, , and Represent the first, second, and third intermediate states respectively, where Represents the tth data item of the three-dimensional structure data, and They represent the t-th and t-1-th fourth intermediate representation data respectively, n≥t≥1, n represents the total number of data items of the three-dimensional structure data, and when t=1 , tanh is the hyperbolic tangent function, Represents the sigmoid function.

[0085] In one embodiment of the present invention, the calculation formula of the first feature fusion layer is as follows:

[0086] ;

[0087] in, represents the tth first fusion representation data, represents an element-wise multiplication operation, represents the fusion function (concatenation function or summation function), represents the tth third intermediate representation data, represents the tth fourth intermediate representation data, represents the tth intermediate representation data, represents the sum weight matrix, Represents the summed bias parameter.

[0088] In one embodiment of the present invention, the calculation formula of the first output layer is as follows:

[0089] ;

[0090] in Represents a matrix, the 1st row and the ith column of the matrix represent the type of the ith gas, the 2nd row and the ith column represent the composition deviation of the ith gas, represents the tth first fusion representation data, n represents the total number of data items of the one-dimensional structure data, is the weight parameter, is the bias parameter, Represents the sigmoid function.

[0091] In one embodiment of the present invention, the calculation formula of the second output layer is as follows:

[0092] ;

[0093] in It means outputting a matrix, the first row and the jth column of the matrix represent the type of the jth fault, the second row and the jth column represent the probability of the jth fault. represents the tth first fusion representation data, n represents the total number of data items of the one-dimensional structure data, is the weight parameter, is the bias parameter, Represents the sigmoid function.

[0094] In one embodiment of the present invention, when the probability of a fault occurring is greater than 0.5, it indicates that the fault has occurred; otherwise, the fault has not occurred; and when all faults have not occurred, it indicates that the system is normal.

[0095] In one embodiment of the present invention, the types of faults that occur include: incomplete combustion, excessive combustion, intake duct blockage, insufficient intake volume, catalyst failure, oxygen sensor failure, insufficient cooling, excessive cooling, piston ring wear, and valve wear.

[0096] In one embodiment of the present invention, the calculation formula of the third output layer is as follows:

[0097] ;

[0098] in Indicates the probability of failure occurring. represents the tth fourth intermediate representation data, n represents the total number of data items of the third one-dimensional structure data, is the weight parameter, is the bias parameter, Represents the sigmoid function.

[0099] In one embodiment of the present invention, when the probability of whether a fault occurs is greater than 0.5, it indicates that a fault occurs, otherwise, no fault occurs.

[0100] Data collection module 200: used to collect comprehensive rescue information at the current time point, and generate No. 2 two-dimensional structure data from the collected comprehensive rescue information, the No. 2 two-dimensional structure data includes No. 2 data matrix and No. 2 relationship matrix, one unit of the No. 2 data matrix represents comprehensive rescue information of an independent object, the independent object includes a faulty vehicle, environmental information, map information, rescue personnel and rescue vehicle, and one unit of the No. 2 data matrix only contains the comprehensive rescue information of the independent object it represents;

[0101] The element in the i-th row and j-th column of the second relationship matrix represents the association between the i-th unit and the independent object represented by the j-th unit of the second data matrix. If there is an association, the value of this element of the second relationship matrix is ​​1, otherwise it is 0;

[0102] Association between fault information nodes and environmental information nodes: Environmental information belongs to the fault location;

[0103] Association between faulty vehicle nodes and map information nodes: the fault occurs at a certain location on the map;

[0104] The relationship between the rescuer node and the rescue vehicle node: rescuers of different professions need to be equipped with corresponding rescue vehicles;

[0105] The association between the rescuer node and the fault information node: rescuers of different professions need to go to the fault site for rescue;

[0106] The relationship between rescue vehicle nodes and fault nodes: Different types of rescue vehicles need to go to the fault location to provide rescue;

[0107] The comprehensive rescue information includes: faulty vehicle information, environmental information, map information, and rescue information;

[0108] Faulty vehicle information: fault type, fault location, fault environment (land or water), fault time, fuel tank level, fuel pressure, battery voltage, remaining power percentage, coolant temperature, brake fluid level, and tire pressure;

[0109] Map information: road networks, topography, hydrology, location and attributes of buildings;

[0110] Rescue information: professional classification of rescuers, corresponding number and location, type of rescue vehicles, corresponding number and location;

[0111] In one embodiment of the present invention, the gas information, driving information and comprehensive rescue information are input into the model after feature engineering;

[0112] Gas information, driving information and comprehensive rescue information include text modalities, and the Word Embedding algorithm is used as a feature engineering method.

[0113] Rescue plan generation module 300: used to input the second two-dimensional structure data into the rescue model, the rescue model includes a fifth intermediate layer and a fourth output layer, the fifth intermediate layer inputs the comprehensive rescue information and outputs the fifth intermediate representation data to the fourth output layer, and the fourth output layer outputs the rescue plan;

[0114] In one embodiment of the present invention, the calculation formula of the fifth intermediate layer is as follows:

[0115] ;

[0116] ;

[0117] ;

[0118] in The fifth intermediate representation data of the vth unit of the second data matrix, and Respectively represent the comprehensive rescue information of the independent objects represented by the vth and uth units of the second data matrix, represents the attention coefficient of the vth and uth units, represents the attention coefficient of the vth and sth units, represents the normalized attention coefficients of the vth and uth units, represents the set of cells of the second data matrix that are associated with the vth cell of the second data matrix, represents an exponential function with a natural constant as base, represents the trainable attention vector parameter, CONCAT represents the concatenation operation, represents transpose, represents the two-dimensional recognition weight parameter, and LeakyReLU is the LeakyReLU activation function.

[0119] In one embodiment of the present invention, the calculation formula of the fourth output layer is as follows:

[0120] ;

[0121] in Represents a vector. The first component of the vector represents a matrix. The 1st row and the mth column of the matrix represent the professional classification of the mth rescuer, and the 2nd row and the mth column represent the number of rescuers of the mth professional classification. The second component of the vector represents a matrix. The 1st row and the nth column of the matrix represent the type of the nth rescue vehicle, and the 2nd row and the nth column represent the number of rescue vehicles of the nth type. The third component of the vector represents a matrix. If the value of the mth row and the nth column of the matrix is ​​1, it means that the faulty vehicle passes through the coordinate (m, n), and if it is 0, it does not pass. The fourth component of the vector represents a matrix. If the value of the mth row and the nth column of the matrix is ​​1, it means that the rescue vehicle passes through the coordinate (m, n), and if it is 0, it does not pass. , represents the fifth intermediate representation data of the vth unit of the second data matrix, M represents the set of all units of the second data matrix, is the weight parameter, is the bias parameter, Represents the sigmoid function.

[0122] Solution execution module 400: It is used to decode the rescue solution output by the rescue model and send it to the receiving terminals of the rescue personnel and the faulty vehicle. After receiving the information, the faulty vehicle and the rescue personnel make specific arrangements to go to the target location for rescue work.

[0123] In one embodiment of the present invention, the loss function of the fault detection model is a cross entropy loss function, and the calculation formula is as follows:

[0124] ;

[0125] in represents the loss value, Indicates the total number of gas types, represents the total number of training samples, It is The symbol value of the training sample, if the The true classification of the training samples is The value of a classification category is 1, otherwise it is 0. The gas composition deviation model predicts The training samples belong to The probability of the classification category, Represents the logarithmic function with the natural constant e as the base.

[0126] ;

[0127] in represents the loss value, Indicates the total number of fault types, represents the total number of training samples, It is The symbol value of the training sample, if the The true classification of the training samples is The value of a classification category is 1, otherwise it is 0. is the first The training samples belong to The probability of the classification category, Represents the logarithmic function with the natural constant e as the base.

[0128] In one embodiment of the present invention, the fourth intermediate representation data is connected to the third output layer for pre-training, and the loss function is a cross entropy loss function, and the calculation formula is as follows:

[0129] ;

[0130] in represents the loss value, represents the total number of training samples, It is The true classification label of the training sample is 0 or 1, 1 means that the vehicle of the training sample has a fault, and 0 means that the vehicle of the training sample has no fault. It is the predicted The classification labels of training samples, Represents the logarithmic function with the natural constant e as the base.

[0131] In one embodiment of the present invention, the steps of training the rescue model include:

[0132] Step 201, initializing the parameters of the rescue model;

[0133] Step 202: Observe the comprehensive rescue information at time e , the strategy executed at time e , Comprehensive rescue information at e+1 time , execution strategy Rewards ;

[0134] Step 203, then calculate the policy error:

[0135] ;

[0136] in represents the policy error at time e, represents the discount factor, , Represents the rescue model input The maximum probability value in the first output vector output when Represents the rescue model input The first output vector corresponding to the strategy The probability value of

[0137] ;

[0138] in Indicates the cost of repairing the broken-down vehicle. represents the maximum rescue time under ideal conditions, Indicates the actual rescue time. Indicates the actual rescue cost. represents the maximum acceptable rescue cost, , , Represent the first, second, and third weight coefficients respectively.

[0139] Step 204, update the rescue model, the updated formula is as follows:

[0140] ;

[0141] , represents the step size of deep learning, Indicates delivery update;

[0142] Step 205, iterate steps 202-204 until the rescue model converges or the number of iterations reaches a set value. The default value of this value is 100.

[0143] At least one embodiment of the present invention discloses a vehicle driving abnormality detection method based on AI, such as Figure 2 As shown, the following steps are included:

[0144] Step 301, collecting environmental information, gas information and driving information at n time points before the current time point;

[0145] Step 302, sorting the environmental information to obtain a first one-dimensional structure data; generating a second one-dimensional structure data from the collected gas information; sorting the collected driving information to obtain a third one-dimensional structure data;

[0146] Step 303, inputting the one-dimensional structure data No. 1, the one-dimensional structure data No. 2 and the one-dimensional structure data No. 3 into a fault detection model, and the fault detection model outputs an indication of the type of fault that has occurred;

[0147] Step 304, collecting comprehensive rescue information at the current time point;

[0148] Step 305, generating second two-dimensional structure data from the collected comprehensive rescue information;

[0149] Step 306, inputting the second two-dimensional structure data into the rescue model, and the rescue model output represents the rescue plan;

[0150] Step 307, decode the rescue plan output by the rescue model and send it to the receiving terminals of the rescue personnel and the faulty vehicle. After receiving the information, the faulty vehicle and the rescue personnel make specific arrangements to go to the target location for rescue work.

[0151] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation mode. The above-mentioned specific implementation mode is merely illustrative and not restrictive. Under the guidance of this embodiment, ordinary technicians in this field can also make more forms of equivalent embodiments, all of which are within the protection of this embodiment.

Claims

1. A vehicle driving abnormality detection system based on AI, characterized in that: Includes the following modules: Fault detection module: it is used to collect environmental information, gas information and driving information at n time points before the current time point, sort the environmental information to obtain No. 1 one-dimensional structure data, generate No. 2 one-dimensional structure data from the collected gas information, sort the collected driving information to obtain No. 3 one-dimensional structure data, input the No. 1 one-dimensional structure data, No. 2 one-dimensional structure data and No. 3 one-dimensional structure data into the fault detection model, and output the type of fault that occurred; Data collection module: It is used to collect comprehensive rescue information at the current time point and generate No. 2 two-dimensional structure data from the collected comprehensive rescue information; Rescue plan generation module: it is used to input the second two-dimensional structure data into the rescue model, the rescue model includes a fifth intermediate layer and a fourth output layer, the fifth intermediate layer inputs comprehensive rescue information and outputs fifth intermediate representation data to the fourth output layer, and the fourth output layer outputs the rescue plan; Plan execution module: It is used to decode the rescue plan output by the rescue model and send it to the receiving terminals of the rescue personnel and the faulty vehicle. After receiving the information, the faulty vehicle and the rescue personnel will make specific arrangements to go to the target location for rescue work.

2. The AI-based vehicle driving abnormality detection system according to claim 1, characterized in that: The fault detection module includes the following steps: Step 101, collecting environmental information, gas information and driving information at n time points before the current time point; Step 102, sorting the environmental information to obtain No. 1 one-dimensional structure data, the No. 1 one-dimensional structure data includes n data items sorted by time, and the t-th data item represents the environmental information collected at the t-th moment; generating No. 2 one-dimensional structure data from the collected gas information, the No. 2 one-dimensional structure data includes n data items sorted by time, and the t-th data item represents the No. 1 two-dimensional structure data generated by the gas information collected at the t-th moment; sorting the collected driving information to obtain No. 3 one-dimensional structure data, the No. 3 one-dimensional structure data includes n data items sorted by time, and the t-th data item represents the driving information collected at the t-th moment; Step 103, inputting the first one-dimensional structure data, the second one-dimensional structure data and the third one-dimensional structure data into a fault detection model, and outputting an indication of the type of fault that has occurred.

3. The AI-based vehicle driving abnormality detection system according to claim 1, characterized in that: Environmental information includes: temperature, humidity, air pressure, wind speed, rainfall, and solar radiation intensity; Gas information includes exhaust gas information and gas composition deviation information; The emission gas information includes: the type and concentration of the vehicle's emission gas; Driving information includes: driving environment, engine speed, engine torque, coolant temperature, fuel injection amount, injection timing, and fuel pressure; Comprehensive rescue information includes: faulty vehicle information, environmental information, map information, and rescue information; Faulty vehicle information: fault type, fault location, fault environment, fault time, fuel tank level, fuel pressure, battery voltage, remaining power percentage, coolant temperature, brake fluid level, and tire pressure; Map information: road networks, topography, hydrology, location and attributes of buildings; Rescue information: professional classification of rescue personnel, corresponding number and location, type of rescue vehicles, corresponding number and location.

4. The AI-based vehicle driving abnormality detection system according to claim 1, characterized in that: The second two-dimensional structure data includes a second data matrix and a second relationship matrix, wherein one unit of the second data matrix represents comprehensive rescue information of an independent object, and the independent object includes a faulty vehicle, environmental information, map information, rescue personnel, and rescue vehicles, and one unit of the second data matrix only includes comprehensive rescue information of the independent object represented by it; The element in the i-th row and j-th column of the second relationship matrix represents the association between the i-th unit and the independent object represented by the j-th unit of the second data matrix. If there is an association, the value of this element of the second relationship matrix is ​​1, otherwise it is 0; Association between fault information nodes and environmental information nodes: Environmental information belongs to the fault location; Association between faulty vehicle nodes and map information nodes: the fault occurs at a certain location on the map; The relationship between the rescuer node and the rescue vehicle node: rescuers of different professions need to be equipped with corresponding rescue vehicles; The association between the rescuer node and the fault information node: rescuers of different professions need to go to the fault site for rescue; The association between rescue vehicle nodes and fault nodes: Different types of rescue vehicles need to go to the fault site to provide rescue.

5. The AI-based vehicle driving abnormality detection system according to claim 1, characterized in that: The types of faults that occur include: incomplete combustion, excessive combustion, intake duct blockage, insufficient intake air, catalyst failure, oxygen sensor failure, insufficient cooling, excessive cooling, piston ring wear, and valve wear.

6. The AI-based vehicle driving abnormality detection system according to claim 2, characterized in that: The No. 1 two-dimensional structure data includes a No. 1 data matrix and a No. 1 relationship matrix. One unit of the No. 1 data matrix represents gas information of an independent object. The independent object includes various types of gases. One unit of the No. 1 data matrix only includes gas information of the independent object represented by it. The element in the i-th row and j-th column of the No. 1 relationship matrix represents the association between the i-th unit of the No. 1 data matrix and the independent object represented by the j-th unit. If there is an association, the value of this element of the No. 1 relationship matrix is ​​1, otherwise it is 0; The correlation between gases means that they are all gases emitted by vehicles.

7. The AI-based vehicle driving abnormality detection system according to claim 2, characterized in that: When the probability of a fault occurring is greater than 0.5, it indicates that the fault has occurred; otherwise, the fault has not occurred. When no faults occur, it indicates that the system is normal.

8. The AI-based vehicle driving abnormality detection system according to claim 2 is characterized in that: The fault detection model includes a first intermediate layer, a second intermediate layer, a third intermediate layer, a fourth intermediate layer, a first feature fusion layer, a first output layer, a second output layer and a third output layer. The first intermediate layer inputs one-dimensional structure data No. 1 and outputs first intermediate representation data to the feature fusion layer. The second intermediate representation data inputs one-dimensional structure data No. 2 and outputs second intermediate representation data to the third intermediate layer. The third intermediate layer outputs third intermediate representation data to the first feature fusion layer. The fourth intermediate layer inputs one-dimensional structure data No. 3 and outputs fourth intermediate representation data to the first feature fusion layer and the third output layer. The output of the third output layer indicates whether a fault occurs. The feature fusion layer outputs the first feature fusion representation data to the first output layer and the second output layer. The output of the first output layer indicates gas composition deviation information. The output of the second output layer indicates the type of fault that has occurred.

9. The AI-based vehicle driving abnormality detection system according to claim 8, characterized in that: The fourth intermediate representation data is connected to the third output layer for pre-training. The loss function is the cross entropy loss function, and the calculation formula is as follows: ； in represents the loss value, represents the total number of training samples, It is The true classification label of the training sample is 0 or 1, 1 means that the vehicle of the training sample has a fault, and 0 means that the vehicle of the training sample has no fault. It is the predicted The classification labels of training samples, Represents the logarithmic function with the natural constant e as the base.

10. A vehicle driving abnormality detection method based on AI, characterized in that: The following steps are involved: Step 100, collecting environmental information, gas information and driving information at n time points before the current time point; Step 200, sorting the environmental information to obtain first-dimensional structure data; generating second-dimensional structure data from the collected gas information; Sorting the collected driving information to obtain three-dimensional structure data; Step 300, inputting the first one-dimensional structure data, the second one-dimensional structure data and the third one-dimensional structure data into a fault detection model, and the fault detection model outputs an indication of the type of fault that has occurred; Step 400, collecting comprehensive rescue information at the current time point; Step 500, generating second two-dimensional structure data from the collected comprehensive rescue information; Step 600, inputting the second two-dimensional structure data into the rescue model, and the rescue model output represents the rescue plan; Step 700, decode the rescue plan output by the rescue model and send it to the receiving terminals of the rescue personnel and the faulty vehicle. After receiving the information, the faulty vehicle and the rescue personnel make specific arrangements to go to the target location for rescue work.

Images