CN113345229B

CN113345229B - Road early warning method based on federal learning and related equipment thereof

Info

Publication number: CN113345229B
Application number: CN202110609350.8A
Authority: CN
Inventors: 李泽远; 王健宗
Original assignee: Ping An Technology Shenzhen Co Ltd
Current assignee: Ping An Technology Shenzhen Co Ltd
Priority date: 2021-06-01
Filing date: 2021-06-01
Publication date: 2022-04-19
Anticipated expiration: 2041-06-01
Also published as: CN113345229A

Abstract

The embodiment of the application belongs to the technical field of artificial intelligence, is applied to the field of intelligent city management, and relates to a road early warning method based on federal learning and related equipment thereof. The dispatching client judges the geographical position distance between the vehicle early warning signal and the first road risk early warning signal, and when the distance is smaller than the range threshold, the dispatching client sends a road regulation and control notice to designated personnel, so that the dispatching client helps the traffic department to reasonably distribute traffic resources, reduces the occurrence of road accidents and improves the traffic capacity. The target inverse neural network model and the target recurrent neural network model may be stored in a block chain. The data barrier is broken through, and the occurrence of road accidents is reduced.

Description

Road early warning method based on federal learning and related equipment thereof

Technical Field

The application relates to the technical field of artificial intelligence, in particular to a road early warning method based on federal learning and related equipment thereof.

Background

Many problems of the existing urban road traffic are not properly solved, and as seen from the endless emergence of traffic accidents, the identification and detection of traffic risks in cities are urgently needed to be improved. At present, in the aspect of road risk detection, a product mostly implements some basic state detection through a sensor or a camera and cannot carry out deeper risk early warning judgment.

Meanwhile, data among traffic control departments of provinces, cities, counties and counties are not intercommunicated, information management systems adopted by the departments are very different, and data information held by enterprises and individuals corresponding to individual automobiles is not intercommunicated at all, so that a data barrier among governments, enterprises and individual households is generated. It is difficult to accurately and quickly perform early warning of vehicles and roads, to quickly predict dangers in advance, and to frequently generate road accidents.

Disclosure of Invention

The embodiment of the application aims to provide a road early warning method based on federal learning and relevant equipment thereof, so that data barriers are broken, road accidents are reduced, and traffic capacity is improved.

In order to solve the above technical problem, an embodiment of the present application provides a road early warning method based on federal learning, which adopts the following technical scheme:

a road early warning method based on federal learning comprises the following steps:

each first local server receives the initial reverse neural network model transmitted by the central server respectively, acquires vehicle index data of a corresponding vehicle, trains the initial reverse neural network model based on the vehicle index data, acquires parameters of the trained initial reverse neural network model, serves as first parameters, and transmits the first parameters to the central server, wherein the first local servers and the vehicles are in one-to-one correspondence association relationship;

each second local server receives the initial recurrent neural network model transmitted by the central server respectively, acquires corresponding monitored road monitoring image data, trains the initial recurrent neural network model based on the road monitoring image data, acquires parameters of the trained initial recurrent neural network model, serves as second parameters, and transmits the second parameters to the central server, wherein the second local servers and the monitoring are in one-to-one correspondence association relationship;

the central server receives the first parameter and the second parameter respectively, performs aggregation processing on the first parameter and the second parameter through an enhanced algorithm respectively to obtain a first target parameter and a second target parameter respectively, and transmits the first target parameter and the second target parameter to the first local server and the second local server respectively;

the first local server receives the first target parameter, and iteratively updates the initial reverse neural network model based on the first target parameter until a preset stop condition is reached to obtain a target reverse neural network model;

the method comprises the steps that a first local server obtains vehicle index data to be detected of a current vehicle, the vehicle index data to be detected are input into a target reverse neural network model, the vehicle state and the vehicle abnormal probability output by the target reverse neural network model are obtained, whether the vehicle abnormal probability is larger than an abnormal threshold value or not is determined, and when the vehicle abnormal probability is larger than the abnormal threshold value, vehicle early warning signals are sent to vehicles and dispatching clients within a preset range;

the second local server receives the second target parameter, and iteratively updates the initial recurrent neural network model based on the second target parameter until a preset stop condition is reached to obtain a target recurrent neural network model;

the second local server acquires a road image in real time according to a time sequence, inputs the road image into the target recurrent neural network model, acquires a road risk probability output by the target recurrent neural network model, determines whether the road risk probability is greater than a first road risk threshold value, and sends a first road risk early warning signal to the scheduling client when the road risk probability is greater than the first road risk threshold value;

when the scheduling client receives a vehicle early warning signal and a first road risk early warning signal at the same time, the scheduling client judges whether the distance between a first geographical position carried by the vehicle early warning signal and a second geographical position carried by the first road risk early warning signal is smaller than a range threshold, and when the distance between the first geographical position and the second geographical position is smaller than the range threshold, a road regulation and control notice is sent to appointed personnel.

Further, after the step of determining whether the vehicle abnormal probability is greater than an abnormal threshold and sending a vehicle early warning signal to the vehicle and the scheduling client within a preset range when the vehicle abnormal probability is greater than the abnormal threshold, the method further includes:

the first server determines whether the vehicle abnormal probability is larger than an abnormal threshold value or not, and sends a safety early warning signal to the scheduling client when the vehicle abnormal probability is larger than the abnormal threshold value;

the scheduling client receives a safety early warning signal sent by the first local server, determines the first local server sending the safety early warning signal as an early warning local server, and determines a vehicle associated with the early warning local server as a target vehicle according to the early warning local server;

the scheduling client determines the geographic position of the target vehicle, determines target monitoring from the monitoring based on the geographic position of the target vehicle, and determines a second local server associated with the target monitoring as a target second local server based on the target monitoring;

the dispatching client sends road monitoring early warning signals to all the target second local servers;

after the step of obtaining a road image in real time according to a time sequence by the second local server, inputting the road image into the target recurrent neural network model, and obtaining a road risk probability output by the target recurrent neural network model, the method further includes:

when the second local server receives a road monitoring early warning signal sent by a scheduling client, the second local server determines whether the road risk probability is greater than a second road risk threshold, and sends the second road risk early warning signal to the scheduling client when the road risk probability is greater than the second road risk threshold, wherein the second road risk threshold is smaller than the first road risk threshold.

Further, the step of receiving, by the central server, the first parameter and the second parameter, respectively, and performing aggregation processing on the first parameter and the second parameter through an enhanced algorithm to obtain a first target parameter and a second target parameter, respectively, includes:

the central server receives the first parameter and the second parameter respectively, performs averaging operation on the first parameter and the second parameter respectively, and obtains a first target parameter and a second target parameter respectively, where the first target parameter is characterized in that:

wherein W' is the first target parameter, n is the number of vehicles, n_kFor the k-th vehicle, n_kThe value is always 1, and the number of the particles is constant,

is the first parameter;

the second target parameter is characterized by:

wherein V' is the second target parameter, m is the number of monitoring, m_pFor the p-th monitoring, m_pThe value is always 1, and the number of the particles is constant,

is the second parameter.

Further, before the step of receiving the initial inverse neural network model transmitted by the central server and acquiring the vehicle index data of the corresponding vehicle, the method further includes:

the central server determines the number n of vehicles according to a preset distribution ratio;

receiving the abnormal indexes transmitted by each candidate local server, and sorting the candidate local servers in a descending order according to the abnormal indexes to obtain an index list;

and selecting the first n candidate local servers in the index list as the first local server, and taking the vehicle corresponding to the first local server as the vehicle.

Further, the step of determining the number n of vehicles by the central server according to the preset distribution ratio includes:

the number of vehicles is characterized in that:

n is the maximum value between the product of the total number of vehicles and the preset distribution ratio and 1.

Further, the first local server iteratively updates the initial inverse neural network model based on the first target parameter until a preset stop condition is reached, and the step of obtaining a target inverse neural network model includes:

the first local server updates the initial reverse neural network model based on the first target parameter to obtain an intermediate reverse neural network model, trains the intermediate reverse neural network model based on the vehicle index data, and obtains parameters of the trained intermediate reverse neural network model;

transmitting the parameters of the trained intermediate reverse neural network model to a central server so that the central server generates third target parameters according to the parameters of the intermediate reverse neural network model;

and receiving a third target parameter transmitted by the central server, and iteratively updating the intermediate reverse neural network model through the third target parameter until a preset iteration number is reached to obtain the target reverse neural network model.

Further, the step of transmitting the first parameter to a central server comprises:

encrypting the first parameter through an AES encryption function to obtain encrypted data;

transmitting the encrypted data to the central server.

In order to solve the above technical problem, an embodiment of the present application further provides a road early warning system based on federal learning, which adopts the following technical scheme:

a federal learning-based road warning system including a central server, a first local server and a second local server, wherein,

the first local server is used for receiving the initial reverse neural network model transmitted by the central server, acquiring vehicle index data of a corresponding vehicle, training the initial reverse neural network model based on the vehicle index data, acquiring parameters of the trained initial reverse neural network model, taking the parameters as first parameters, and transmitting the first parameters to the central server, wherein the first local server and the vehicle are in one-to-one correspondence association relationship;

the second local server is used for receiving the initial recurrent neural network model transmitted by the central server, acquiring corresponding monitored road monitoring image data, training the initial recurrent neural network model based on the road monitoring image data, acquiring parameters of the trained initial recurrent neural network model, taking the parameters as second parameters, and transmitting the second parameters to the central server, wherein the second local server and the monitoring are in a one-to-one corresponding association relationship;

the central server is configured to receive the first parameter and the second parameter, aggregate the first parameter and the second parameter through an enhanced algorithm, obtain a first target parameter and a second target parameter, and transmit the first target parameter and the second target parameter to the first local server and the second local server, respectively;

the first local server is used for receiving the first target parameter, iteratively updating the initial reverse neural network model based on the first target parameter until a preset stop condition is reached, and obtaining a target reverse neural network model;

the first local server is used for acquiring vehicle index data to be detected of a current vehicle in real time, inputting the vehicle index data to be detected into the target reverse neural network model, acquiring a vehicle state and a vehicle abnormal probability output by the target reverse neural network model, determining whether the vehicle abnormal probability is greater than an abnormal threshold value, and sending a vehicle early warning signal to vehicles within a preset range when the vehicle abnormal probability is greater than the abnormal threshold value;

the second local server is configured to receive the second target parameter, and iteratively update the initial recurrent neural network model based on the second target parameter until a preset stop condition is reached to obtain a target recurrent neural network model;

the second local server is used for acquiring a road image according to a time sequence in real time, inputting the road image into the target recurrent neural network model, acquiring a road risk probability output by the target recurrent neural network model, determining whether the road risk probability is greater than a first road risk threshold value or not, and sending a first road risk early warning signal to the scheduling client when the road risk probability is greater than the first road risk threshold value.

In order to solve the above technical problem, an embodiment of the present application further provides a computer device, which adopts the following technical solutions:

a computer device comprising a memory having computer readable instructions stored therein and a processor that when executed implement the steps of the federal learning based road warning method described above.

In order to solve the above technical problem, an embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solutions:

a computer readable storage medium having computer readable instructions stored thereon which, when executed by a processor, implement the steps of the federal learning based road warning method described above.

Compared with the prior art, the embodiment of the application mainly has the following beneficial effects:

according to the method and the device, all data do not need to be concentrated together, the central server sends different neural network models to the first local server and the second local server, namely the initial reverse neural network model and the initial recurrent neural network model, the first local server and the second local server train the received neural network models respectively according to the local data, and then the trained model parameters are transmitted to the central server, so that the data can be comprehensively trained with the data of other servers without leaving the local, under the condition that the data privacy is guaranteed, the data barrier is effectively broken, and further the model expression effects of the trained target reverse neural network model and the trained target recurrent neural network model are effectively improved. And sending vehicle early warning signals to vehicles within a preset range, so that surrounding vehicles can avoid vehicles with possible faults in time, and traffic accidents are reduced. The dispatching client side sends road regulation and control notification to appointed personnel by judging the distance between the vehicle early warning signal and the first road risk early warning signal when the distance is smaller than the range threshold value, so that the dispatching client side can help the traffic department to dynamically adjust the police force distribution, reasonably distribute traffic resources, reduce the occurrence of road accidents and improve the traffic capacity.

Drawings

In order to more clearly illustrate the solution of the present application, the drawings needed for describing the embodiments of the present application will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present application, and that other drawings can be obtained by those skilled in the art without inventive effort.

FIG. 1 is an exemplary system architecture diagram in which the present application may be applied;

FIG. 2 is a flow diagram of one embodiment of a federal learning based road warning method in accordance with the present application;

FIG. 3 is a schematic block diagram of one embodiment of a computer device according to the present application.

Reference numerals: 100. a central server; 200. a first local server; 300. a second local server; 400. scheduling the client; 500. a computer device; 501. a memory; 502. a processor; 503. a network interface.

Detailed Description

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used in the description of the application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "including" and "having," and any variations thereof, in the description and claims of this application and the description of the above figures are intended to cover non-exclusive inclusions. The terms "first," "second," and the like in the description and claims of this application or in the above-described drawings are used for distinguishing between different objects and not for describing a particular order.

Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. It is explicitly and implicitly understood by one skilled in the art that the embodiments described herein can be combined with other embodiments.

In order to make the technical solutions better understood by those skilled in the art, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

The road risk detection method based on the federal learning in the embodiment of the application is applied to a road risk detection system based on the federal learning. Referring to fig. 1, the system architecture of the present application includes a central server 100, a first local server 200, a second local server 300, and a scheduling client 400. The connection between the central server 100 and the first local server 200, between the central server 100 and the second local server 300, between the scheduling client 400 and the first local server 200, and between the scheduling client 400 and the second local server 300 is performed through a network, which may include various connection types, such as a wired, wireless communication link, or an optical fiber cable, etc.

A user may use scheduling client 400 to interact with a target server or federated server over a network to receive or send messages, etc. The scheduling client can be installed with various communication client applications, such as a web browser application, a shopping application, a search application, an instant messaging tool, a mailbox client, social platform software, and the like.

The scheduling client 400 may be various electronic devices having a display screen and supporting web browsing, including but not limited to a smart phone, a tablet computer, an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III, mpeg compression standard Audio Layer 3), an MP4 player (Moving Picture Experts Group Audio Layer IV, mpeg compression standard Audio Layer 4), a laptop portable computer, a desktop computer, and the like.

It should be understood that the numbers of the central server 100, the first local server 200, the second local server 300, and the scheduling client 400 in fig. 1 are merely illustrative. There may be any number of central servers 100, first local servers 200, second local servers 300, and scheduling clients 400, as desired for an implementation.

With continued reference to fig. 2, a flow diagram of one embodiment of a federal learning based road warning method in accordance with the present application is shown. The road early warning method based on the federal learning comprises the following steps:

s1: each first local server receives the initial reverse neural network model transmitted by the central server respectively, acquires vehicle index data of a corresponding vehicle, trains the initial reverse neural network model based on the vehicle index data, acquires parameters of the trained initial reverse neural network model, and transmits the first parameters to the central server as first parameters, wherein the first local servers and the vehicles are in one-to-one correspondence association relationship.

In this embodiment, the central server transmits the initial inverse neural network model to each of the first local servers, respectively. A Back Propagation Neural Network (BPNN). The prediction task model is composed of an input layer, a hidden layer and an output layer, wherein the hidden layer transmits important information between the input layer and the output layer, and the reverse neural network model has better performance in a prediction task. The central server transmits the initial recurrent neural network model to each second local server respectively. A Recurrent Neural Network (RNN) model can update information over time, with the last input information persisting in the Network and affecting the output content of the next input information, the recurrence occurring in each step of the sequential process. The vehicle in S1 in the present application refers to a vehicle selected from all candidate vehicles that participates in the training process of the present application. The method comprises the steps that a first local server obtains vehicle index data of a corresponding vehicle, the vehicle index data are historical data, and the vehicle index data comprise: the performance indexes of the motor vehicle (such as vehicle load, engine heating rate, turbine gas compression ratio and the like), the internal state indexes of the vehicle under the driving state (such as steering, braking, driving and the like) and the vehicle state (including oil consumption, pollutant discharge, engine speed and the like). And carrying out structuralization processing on the vehicle index data to obtain structuralized vehicle index data. The historical data is acquired in real time through a sensor installed in the vehicle, structured processing is performed to acquire structured vehicle index data, and the structured vehicle index data is stored in a hard disk inside the vehicle in real time through an Operating System (OS) of the vehicle, so as to serve as historical data applied to a training process. The structured data is exemplified as follows:

vehicle number	Type (B)	Numerical value	Whether or not it is abnormal
				1	Load class	100	Is that

TABLE 1

The initial reverse neural network model is trained through structured vehicle index data, and parameters of the trained initial reverse neural network model are obtained. The training process of the first local server is performed by the following expression:

wherein eta represents the hyper-parameter, w represents the parameter of the initial inverse neural network model in the training process,

represents the gradient, l (w, β) represents the curve, and β represents the training data, i.e., the vehicle index data.

Specifically, in step S1, the step of transmitting the first parameter to the central server includes:

transmitting the encrypted data to the central server.

In this embodiment, the AES encryption belongs to symmetric encryption, the first parameter is encrypted by an AES encryption function, and after obtaining the encrypted data, the central server decrypts the encrypted data by an AES decryption function to obtain the first parameter, where a key used for AES decryption is the same as a key used for encryption. In the process of practical application, data transmission can be carried out in an asymmetric encryption mode according to actual needs, and the method is applicable.

S2: each second local server receives the initial recurrent neural network model transmitted by the central server respectively, acquires corresponding monitored road monitoring image data, trains the initial recurrent neural network model based on the road monitoring image data, acquires parameters of the trained initial recurrent neural network model, serves as second parameters, and transmits the second parameters to the central server, wherein the second local servers and the monitoring are in one-to-one correspondence relationship.

In this embodiment, the road monitoring image data is an image of captured historical traffic monitoring video data. Before training the recurrent neural network model, the road monitoring image data needs to be pre-labeled, and the image is labeled with labels, for example: collisions, congestion, etc. And obtaining the marked image, training the initial recurrent neural network model through the marked image, and obtaining the parameters of the trained initial recurrent neural network model. The monitoring at S2 in the present application refers to the monitoring selected from all the candidate monitoring to participate in the training process of the present application.

It should be noted that the expression of the training process of the initial recurrent neural network in the second local server is the same as the expression of the training process of the initial inverse neural network in the first local server, and details are not repeated here.

In this embodiment, the electronic device (for example, the first local server or the second local server shown in fig. 1) on which the federally learned road warning method operates may receive the initial inverse neural network model or the initial recurrent neural network model transmitted by the central server through a wired connection or a wireless connection. It should be noted that the wireless connection means may include, but is not limited to, a 3G/4G connection, a WiFi connection, a bluetooth connection, a WiMAX connection, a Zigbee connection, a uwb (ultra wideband) connection, and other wireless connection means now known or developed in the future.

S3: and the central server respectively receives the first parameter and the second parameter, respectively carries out aggregation processing on the first parameter and the second parameter through an enhanced algorithm, respectively obtains a first target parameter and a second target parameter, and respectively transmits the first target parameter and the second target parameter to the first local server and the second local server.

In this embodiment, the enhancement algorithm of the present application is a FedAvg enhancement algorithm. The FedAvg algorithm is an important method in federal learning, and the main content of FedAvg is to integrate and average the parameters trained by each user. According to the method and the device, the first parameter and the second parameter are subjected to aggregation processing to obtain the aggregated first target parameter and the aggregated second target parameter, and the aggregated first target parameter and the aggregated second target parameter are used for a subsequent federal training process.

Specifically, in step S3, the step of respectively receiving the first parameter and the second parameter by the central server, and respectively performing aggregation processing on the first parameter and the second parameter by using an enhanced algorithm to respectively obtain a first target parameter and a second target parameter includes:

is the first parameter;

the second target parameter is characterized by:

is the second parameter.

In this embodiment, the central server performs an averaging operation on the received first parameters of each first local server according to a formula to obtain first target parameters, where the first target parameters aggregate characteristics of the vehicle corresponding to each first local server. The central server performs average operation on the received second parameters of the second local servers according to a formula to obtain second target parameters, and the second target parameters also have monitoring characteristics corresponding to the second local servers, so that a model with better performance can be finally obtained.

S4: and the first local server receives the first target parameter, and iteratively updates the initial reverse neural network model based on the first target parameter until a preset stop condition is reached to obtain a target reverse neural network model.

In this embodiment, the initial inverse neural network model is updated iteratively through the first target parameter, and the target inverse neural network model is finally obtained, so that the data barrier is effectively broken, the training data cannot be local, and all the first local servers can still jointly train the model.

Specifically, in step S4, that is, the first local server iteratively updates the initial inverse neural network model based on the first target parameter until a preset stop condition is reached, the step of obtaining the target inverse neural network model includes:

In this embodiment, the trained parameters of the intermediate inverse neural network model are transmitted to the central server, so that the central server generates a third target parameter according to the parameters of the intermediate inverse neural network model, the central server performs the aggregation processing according to the received parameters to obtain aggregated parameters as the third target parameter, the central server transmits the third target parameter to the first local server, the first local server updates the intermediate inverse neural network model according to the third target parameter to obtain an updated intermediate inverse neural network model, and trains the updated intermediate inverse neural network model through locally stored training data, i.e., vehicle index data, and the process is repeated until the number of iterations E times is reached, so as to train the butterfly target inverse neural network model.

S5: the method comprises the steps that a first local server obtains vehicle index data to be detected of a current vehicle in real time, the vehicle index data to be detected are input into a target reverse neural network model, the vehicle state and the vehicle abnormal probability output by the target reverse neural network model are obtained, whether the vehicle abnormal probability is larger than an abnormal threshold value or not is determined, and vehicle early warning signals are sent to vehicles and dispatching clients within a preset range when the vehicle abnormal probability is larger than the abnormal threshold value.

In the embodiment, the target inverse neural network model obtained after training can perform defining and early warning operation on the type of the vehicle risk. When the Vehicle abnormal probability is larger than the abnormal threshold value, a Vehicle early warning signal is generated, and the Vehicle early warning signal is sent to surrounding Vehicle clusters through V2X (Vehicle to X, Vehicle wireless communication technology) to remind surrounding vehicles to avoid the abnormal Vehicle so as to improve the linkage meeting decision of driving. V2X (Vehicle to X) enables communication between cars, cars and base stations, and base stations.

S6: and the second local server receives the second target parameter, and iteratively updates the initial recurrent neural network model based on the second target parameter until a preset stop condition is reached to obtain a target recurrent neural network model.

In this embodiment, the initial recurrent neural network model is updated iteratively through the second target parameter, and the target recurrent neural network model is finally obtained, so that the data barrier is effectively broken under the condition of ensuring data privacy, and the target recurrent neural network model with better performance is obtained.

S7: the second local server acquires a road image in real time according to the time sequence, inputs the road image into the target recurrent neural network model, acquires a road risk probability output by the target recurrent neural network model, determines whether the road risk probability is greater than a first road risk threshold value, and sends a first road risk early warning signal to the scheduling client when the road risk probability is greater than the first road risk threshold value.

In this embodiment, a road is detected through a trained target recurrent neural network model, and when the road risk probability is greater than the first road risk threshold, a first road risk early warning signal is sent to a scheduling client, and meanwhile, a first road risk early warning signal is also sent to vehicles within a preset range, so that traffic accidents are detected and subjected to central control early warning, and vehicle-road cooperation and networking communication are achieved. The target recurrent neural network model trained by the federal learning mechanism can give more accurate and rapid early warning and identification inference for vehicle interior early warning and road detection in real scenes such as main traffic major roads, intersections with frequent violation, traffic jam roads and the like.

S8: when the scheduling client receives a vehicle early warning signal and a first road risk early warning signal at the same time, the scheduling client judges whether the distance between a first geographical position carried by the vehicle early warning signal and a second geographical position carried by the first road risk early warning signal is smaller than a range threshold, and when the distance between the first geographical position and the second geographical position is smaller than the range threshold, a road regulation and control notice is sent to appointed personnel.

In the embodiment, the reverse neural network model and the recurrent neural network model are utilized to automatically detect and early warn the situation of the possible traffic accident, when the distance between the geographic positions of the simultaneously received vehicle early warning signal and the first road risk early warning signal is smaller than the range threshold, the traffic accident, traffic jam and the like can be determined in the area between the two geographic positions, and then a road regulation and control notice is sent to designated personnel, so that the occurrence probability of the traffic accident is reduced, and the road safety is improved. The scheduling client receives the vehicle early warning signal and the first road risk early warning signal at the same time, namely, the time interval between the vehicle early warning signal and the first road risk early warning signal is smaller than a preset time interval threshold value, namely, the scheduling client receives the two signals at the same time, and the traffic police force can be reasonably allocated to achieve traffic intelligent management and operation through the scheme of the application.

In some optional implementation manners of this embodiment, in step S5, that is, after the step of determining whether the vehicle abnormality probability is greater than an abnormality threshold, and when the vehicle abnormality probability is greater than the abnormality threshold, sending a vehicle warning signal to vehicles within a preset range and a scheduling client, the electronic device may further perform the following steps:

in step S7, after the step of acquiring, by the second local server, the road image according to the time sequence in real time, inputting the road image into the target recurrent neural network model, and acquiring the road risk probability output by the target recurrent neural network model, the electronic device may further perform the following steps:

In this embodiment, after receiving the safety warning signal sent by the first local server, the scheduling client determines the target vehicle and the geographic position thereof based on the safety warning signal, and then takes monitoring within a preset range of the target vehicle as target monitoring according to the geographic position of the target vehicle, and further determines a target second local server, and sends a road monitoring warning signal to the target second local server, so as to perform key tracking on a road where the vehicle possibly having a problem is located. The method comprises the specific steps of reducing a road risk threshold, namely comparing a second road risk threshold with a road risk probability to determine whether a road accident is possible, determining that the accident is possible on the current road when the road risk probability is larger than the second road risk threshold, sending a second road risk early warning signal to a scheduling client so as to find the abnormal traffic condition or the accident more quickly, and performing traffic evacuation and police force scheduling in time after the scheduling client receives the second road risk early warning signal.

In some optional implementations of this embodiment, before step S1, that is, before the step of the first local servers respectively receiving the initial inverse neural network model transmitted by the central server and acquiring the vehicle index data of the corresponding vehicle, the electronic device may further perform the following steps:

In this embodiment, for the selection of vehicles, the candidate local servers obtain vehicle index data of each vehicle, determine an abnormality index, that is, an abnormality rate, of the vehicle index data of each vehicle, transmit the abnormality index to the central server, the central server sorts the candidate local servers according to the abnormality index to obtain an abnormality rate sorting table, and take the first n candidate local servers in the abnormality rate sorting table as first local servers and the vehicle corresponding to the first local server as a vehicle. Because the abnormality index of the vehicle is high, the vehicle is indicated to have more abnormal samples, and the model training is more convenient.

Of course, the application can also set the types in the vehicle index data, such as the load type and the fuel consumption type, to be corresponding weights with different sizes; respectively calculating the category abnormal rate of each category in the vehicle index data of each vehicle, and carrying out weighted summation on the category abnormal rate based on the weight corresponding to the category to obtain the comprehensive abnormal rate; the central server receives the comprehensive abnormal rate uploaded by the candidate local servers, sequences the candidate local servers according to the comprehensive abnormal rate to obtain a comprehensive abnormal rate sequencing table, and takes the first n candidate local servers of the comprehensive abnormal rate sequencing table as first local servers and the vehicles corresponding to the first local servers as the vehicles. Therefore, the typical required categories can be selected, and the weights of the categories are set to be higher, so that required training samples can be conveniently screened out for subsequent federal learning training, and a more targeted reverse neural network model can be obtained.

Specifically, the step of determining the number n of vehicles by the central server according to the preset distribution ratio includes:

the number of vehicles is characterized in that: n is the maximum value between the product of the total number of vehicles and the preset distribution ratio and 1.

In the present embodiment, n is max (C × K, 1), where C is the total number of vehicles and K is the distribution ratio. K is a distribution proportion, the range is 0-1, and the value of K in the application is 50%. And selecting n vehicles from the total vehicles as vehicles, participating in the process of the federal learning training, and controlling the number of the vehicles participating in the federal training in the total vehicles through distribution proportion. The step that n is the maximum value between the product of the total number of vehicles and the preset distribution proportion and 1 is set, so that the condition that the product of C and K is smaller than 1 due to the fact that the distribution proportion is set to be too small is avoided, and the number of the vehicles is smaller than 1.

It should be noted that: the determination method of the monitored number m is the same as the determination method of the vehicle number n, and is not described herein again.

It is emphasized that, in order to further ensure the privacy and security of the target inverse neural network model and the target recurrent neural network model, the target inverse neural network model and the target recurrent neural network model may also be stored in nodes of a block chain. The block chain referred by the application is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, a consensus mechanism, an encryption algorithm and the like. A block chain (Blockchain), which is essentially a decentralized database, is a series of data blocks associated by using a cryptographic method, and each data block contains information of a batch of network transactions, so as to verify the validity (anti-counterfeiting) of the information and generate a next block. The blockchain may include a blockchain underlying platform, a platform product service layer, an application service layer, and the like.

The method and the device can be applied to the field of smart city management, and therefore the construction of the smart city is promoted.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware associated with computer readable instructions, which can be stored in a computer readable storage medium, and when executed, can include processes of the embodiments of the methods described above. The storage medium may be a non-volatile storage medium such as a magnetic disk, an optical disk, a Read-Only Memory (ROM), or a Random Access Memory (RAM).

It should be understood that, although the steps in the flowcharts of the figures are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and may be performed in other orders unless explicitly stated herein. Moreover, at least a portion of the steps in the flow chart of the figure may include multiple sub-steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, which are not necessarily performed in sequence, but may be performed alternately or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

With continuing reference to fig. 1, as an implementation of the method shown in fig. 2, the present application provides an embodiment of a road early warning system based on federal learning, where the embodiment of the apparatus corresponds to the embodiment of the method shown in fig. 2, and the apparatus may be applied to various electronic devices.

As shown in fig. 1, the road warning system based on federal learning according to this embodiment includes: the road early warning system based on federal learning comprises a central server 100, a first local server 200, a second local server 300 and a scheduling client 400, wherein the first local server 200 is used for receiving an initial reverse neural network model transmitted by the central server 100, acquiring vehicle index data of a corresponding vehicle, training the initial reverse neural network model based on the vehicle index data, acquiring parameters of the trained initial reverse neural network model as first parameters, and transmitting the first parameters to the central server 100, wherein the first local server 200 and the vehicle are in a one-to-one corresponding incidence relation; the second local server 300 is configured to receive the initial recurrent neural network model transmitted by the central server 100, obtain corresponding monitored road monitoring image data, train the initial recurrent neural network model based on the road monitoring image data, obtain a parameter of the trained initial recurrent neural network model, serve as a second parameter, and transmit the second parameter to the central server 100, where the second local server 300 and the monitoring are in a one-to-one association relationship; the central server 100 is configured to receive the first parameter and the second parameter, aggregate the first parameter and the second parameter through an enhanced algorithm, obtain a first target parameter and a second target parameter, and transmit the first target parameter and the second target parameter to the first local server 200 and the second local server 300; the first local server 200 is configured to receive the first target parameter, iteratively update the initial inverse neural network model based on the first target parameter until a preset stop condition is reached, and obtain a target inverse neural network model; the first local server 200 is configured to obtain vehicle index data to be detected of a current vehicle in real time, input the vehicle index data to be detected into the target reverse neural network model, obtain a vehicle state and a vehicle abnormal probability output by the target reverse neural network model, determine whether the vehicle abnormal probability is greater than an abnormal threshold, and send a vehicle early warning signal to the vehicle and the scheduling client 400 within a preset range when the vehicle abnormal probability is greater than the abnormal threshold; the second local server 300 is configured to receive the second target parameter, and iteratively update the initial recurrent neural network model based on the second target parameter until a preset stop condition is reached, so as to obtain a target recurrent neural network model; the second local server 300 is configured to obtain a road image according to a time sequence in real time, input the road image into the target recurrent neural network model, obtain a road risk probability output by the target recurrent neural network model, determine whether the road risk probability is greater than a first road risk threshold, and send a first road risk early warning signal to the scheduling client 400 when the road risk probability is greater than the first road risk threshold; the scheduling client 400 is configured to, when the scheduling client 400 receives a vehicle early warning signal and a first road risk early warning signal at the same time, determine whether a distance between a first geographical position carried by the vehicle early warning signal and a second geographical position carried by the first road risk early warning signal is smaller than a range threshold, and send a road regulation and control notification to an appointed person when the distance between the first geographical position and the second geographical position is smaller than the range threshold.

In the embodiment, all data do not need to be collected together, the central server sends different neural network models, namely an initial reverse neural network model and an initial recurrent neural network model, to the first local server and the second local server, the first local server and the second local server train the received neural network models respectively according to the local data, and then the trained model parameters are transmitted to the central server, so that the data can be comprehensively trained with the data of other servers without leaving the local, under the condition of ensuring the privacy of the data, the data barrier is effectively broken, and the model expression effects of the trained target reverse neural network model and the trained target recurrent neural network model are effectively improved. And sending vehicle early warning signals to vehicles within a preset range, so that surrounding vehicles can avoid vehicles with possible faults in time, and traffic accidents are reduced. The dispatching client side sends road regulation and control notification to appointed personnel by judging the distance between the vehicle early warning signal and the first road risk early warning signal when the distance is smaller than the range threshold value, so that the dispatching client side can help the traffic department to dynamically adjust the police force distribution, reasonably distribute traffic resources, reduce the occurrence of road accidents and improve the traffic capacity.

The central server 100 includes an encryption module and a transmission module, wherein the encryption module is configured to perform encryption processing on the first parameter through an AES encryption function to obtain encrypted data. The transmission module is configured to transmit the encrypted data to the central server 100.

In some optional implementations of the present embodiment, the central server 100 is further configured to: respectively receiving the first parameter and the second parameter, respectively performing an averaging operation on the first parameter and the second parameter, and respectively obtaining the first target parameter and the second target parameter, where the first target parameter is characterized in that:

is the first parameter;

the second target parameter is characterized by:

is the second parameter.

The first local server 200 includes an update module, a parameter transmission module, and an iteration module. The updating module is used for updating the initial reverse neural network model based on the first target parameter to obtain an intermediate reverse neural network model, training the intermediate reverse neural network model based on the vehicle index data, and obtaining parameters of the trained intermediate reverse neural network model; the parameter transmission module is used for transmitting the parameters of the trained intermediate inverse neural network model to the central server so that the central server generates a third target parameter 100 according to the parameters of the intermediate inverse neural network model; the iteration module is configured to receive a third target parameter transmitted by the central server 100, and update the intermediate reverse neural network model through iteration of the third target parameter until a preset number of iterations is reached, so as to obtain the target reverse neural network model.

In some optional implementations of this embodiment, the scheduling client 400 may further be configured to: receiving a safety early warning signal sent by the first local server 200, determining the first local server 200 sending the safety early warning signal as an early warning local server, and determining a vehicle associated with the early warning local server according to the early warning local server as a target vehicle; the scheduling client 400 determines the geographic location of the target vehicle, determines a target monitoring from the monitoring based on the geographic location of the target vehicle, and determines a second local server 300 associated with the target monitoring as a target second local server 300 based on the target monitoring; the scheduling client 400 sends a road monitoring and early warning signal to all the target second local servers 300.

The second local server 300 may be further configured to determine whether the road risk probability is greater than a second road risk threshold when receiving a road monitoring early warning signal sent by the scheduling client 400, and send a second road risk early warning signal to the scheduling client 400 when the road risk probability is greater than the second road risk threshold, where the second road risk threshold is smaller than the first road risk threshold.

In some optional implementations of this embodiment, the central server 100 may further be configured to: determining the number n of vehicles according to a preset distribution ratio; receiving the abnormal indexes transmitted by each candidate local server, and sorting the candidate local servers in a descending order according to the abnormal indexes to obtain an index list; and selecting the first n candidate local servers in the index list as the first local server, and taking the vehicle corresponding to the first local server as the vehicle 200.

In some optional implementations of the present embodiment, the central server 100 is further configured to: the number of vehicles is characterized in that: n is the maximum value between the product of the total number of vehicles and the preset distribution ratio and 1.

In order to solve the technical problem, an embodiment of the present application further provides a computer device. Referring to fig. 3, fig. 3 is a block diagram of a basic structure of a computer device according to the present embodiment.

The computer device 200 comprises a memory 201, a processor 202, a network interface 203 communicatively connected to each other via a system bus. It is noted that only computer device 200 having components 201 and 203 is shown, but it is understood that not all of the illustrated components are required and that more or fewer components may alternatively be implemented. As will be understood by those skilled in the art, the computer device is a device capable of automatically performing numerical calculation and/or information processing according to a preset or stored instruction, and the hardware includes, but is not limited to, a microprocessor, an Application Specific Integrated Circuit (ASIC), a Programmable Gate Array (FPGA), a Digital Signal Processor (DSP), an embedded device, and the like.

The computer device can be a desktop computer, a notebook, a palm computer, a cloud server and other computing devices. The computer equipment can carry out man-machine interaction with a user through a keyboard, a mouse, a remote controller, a touch panel or voice control equipment and the like.

The memory 201 includes at least one type of readable storage medium including a flash memory, a hard disk, a multimedia card, a card type memory (e.g., SD or DX memory, etc.), a Random Access Memory (RAM), a Static Random Access Memory (SRAM), a Read Only Memory (ROM), an Electrically Erasable Programmable Read Only Memory (EEPROM), a Programmable Read Only Memory (PROM), a magnetic memory, a magnetic disk, an optical disk, etc. In some embodiments, the storage 201 may be an internal storage unit of the computer device 200, such as a hard disk or a memory of the computer device 200. In other embodiments, the memory 201 may also be an external storage device of the computer device 200, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), or the like, provided on the computer device 200. Of course, the memory 201 may also include both internal and external storage devices of the computer device 200. In this embodiment, the memory 201 is generally used for storing an operating system installed in the computer device 200 and various types of application software, such as computer readable instructions of the road warning method based on federal learning, and the like. Further, the memory 201 may also be used to temporarily store various types of data that have been output or are to be output.

The processor 202 may be a Central Processing Unit (CPU), controller, microcontroller, microprocessor, or other data Processing chip in some embodiments. The processor 202 is generally operative to control overall operation of the computer device 200. In this embodiment, the processor 202 is configured to execute computer readable instructions stored in the memory 201 or process data, such as computer readable instructions for executing the federal learning based road warning method.

The network interface 203 may comprise a wireless network interface or a wired network interface, and the network interface 203 is generally used for establishing communication connection between the computer device 200 and other electronic devices.

In the embodiment, the data barrier is broken, and the model expression effects of the trained target reverse neural network model and the trained target recurrent neural network model are effectively improved. The scheduling client 400 sends a road regulation and control notification to designated personnel, and can help the traffic department to dynamically adjust police force distribution, reasonably distribute traffic resources, reduce road accidents and improve traffic capacity.

The present application provides yet another embodiment, which is to provide a computer-readable storage medium having stored thereon computer-readable instructions executable by at least one processor to cause the at least one processor to perform the steps of the federal learning based road warning method as described above.

Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solutions of the present application may be embodied in the form of a software product, which is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk) and includes instructions for enabling a terminal device (such as a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the method according to the embodiments of the present application.

It is to be understood that the above-described embodiments are merely illustrative of some, but not restrictive, of the broad invention, and that the appended drawings illustrate preferred embodiments of the invention and do not limit the scope of the invention. This application is capable of embodiments in many different forms and is provided for the purpose of enabling a thorough understanding of the disclosure of the application. Although the present application has been described in detail with reference to the foregoing embodiments, it will be apparent to one skilled in the art that the present application may be practiced without modification or with equivalents of some of the features described in the foregoing embodiments. All equivalent structures made by using the contents of the specification and the drawings of the present application are directly or indirectly applied to other related technical fields and are within the protection scope of the present application.

Claims

1. A road early warning method based on federal learning is characterized by comprising the following steps:

the method comprises the steps that a first local server obtains vehicle index data to be detected of a current vehicle in real time, the vehicle index data to be detected are input into a target reverse neural network model, the vehicle state and the vehicle abnormal probability output by the target reverse neural network model are obtained, whether the vehicle abnormal probability is larger than an abnormal threshold value or not is determined, and when the vehicle abnormal probability is larger than the abnormal threshold value, vehicle early warning signals are sent to vehicles and dispatching clients within a preset range;

2. The federal learning-based road warning method as claimed in claim 1, wherein the step of determining whether the vehicle abnormal probability is greater than an abnormal threshold value, and after the step of sending a vehicle warning signal to vehicles and a scheduling client within a preset range when the vehicle abnormal probability is greater than the abnormal threshold value, further comprises:

3. The federal learning-based road warning method as claimed in claim 1, wherein the step of respectively receiving the first parameter and the second parameter by the central server, respectively performing aggregation processing on the first parameter and the second parameter by using an enhanced algorithm, and respectively obtaining a first target parameter and a second target parameter comprises:

is the first parameter;

the second target parameter is characterized by:

wherein V' is the second target parameter, m is the number of monitoring, m_pFor the p th personControl m_pThe value is always 1, and the number of the particles is constant,

is the second parameter.

4. The federal learning-based road warning method as claimed in claim 1, wherein before the step of receiving the initial inverse neural network model transmitted from the central server and obtaining vehicle index data of the corresponding vehicle, the method further comprises:

5. The federal learning-based road warning method as claimed in claim 4, wherein the step of determining the number n of vehicles by the central server according to a preset distribution ratio comprises:

the number of vehicles is characterized in that:

6. The federal learning-based road warning method as claimed in claim 1, wherein the first local server iteratively updates the initial inverse neural network model based on the first target parameter until a preset stop condition is reached, and the step of obtaining the target inverse neural network model comprises:

7. The federal learning based road warning method as claimed in claim 1, wherein the step of transmitting the first parameter to a central server comprises:

transmitting the encrypted data to the central server.

8. The road early warning system based on the federal learning is characterized by comprising a central server, a first local server, a second local server and a scheduling client, wherein,

the first local server is used for acquiring vehicle index data to be detected of a current vehicle in real time, inputting the vehicle index data to be detected into the target reverse neural network model, acquiring a vehicle state and a vehicle abnormal probability output by the target reverse neural network model, determining whether the vehicle abnormal probability is greater than an abnormal threshold value, and sending a vehicle early warning signal to the vehicles and the dispatching client-side within a preset range when the vehicle abnormal probability is greater than the abnormal threshold value;

the second local server is used for acquiring a road image according to a time sequence in real time, inputting the road image into the target recurrent neural network model, acquiring a road risk probability output by the target recurrent neural network model, determining whether the road risk probability is greater than a first road risk threshold value or not, and sending a first road risk early warning signal to the scheduling client when the road risk probability is greater than the first road risk threshold value;

the scheduling client is used for judging whether the distance between a first geographical position carried by the vehicle early warning signal and a second geographical position carried by the first road risk early warning signal is smaller than a range threshold value or not when the scheduling client receives the vehicle early warning signal and the first road risk early warning signal at the same time, and sending a road regulation and control notice to appointed personnel when the distance between the first geographical position and the second geographical position is smaller than the range threshold value.

9. A computer device comprising a memory having computer readable instructions stored therein and a processor that when executed implement the steps of the federal learning based road warning method as claimed in any of claims 1 to 7.

10. A computer readable storage medium having computer readable instructions stored thereon which, when executed by a processor, implement the steps of the federal learning based road warning method as claimed in any of claims 1 to 7.