CN112052959B - Automatic driving training method, equipment and medium based on federal learning - Google Patents
Automatic driving training method, equipment and medium based on federal learning Download PDFInfo
- Publication number
- CN112052959B CN112052959B CN202010931767.1A CN202010931767A CN112052959B CN 112052959 B CN112052959 B CN 112052959B CN 202010931767 A CN202010931767 A CN 202010931767A CN 112052959 B CN112052959 B CN 112052959B
- Authority
- CN
- China
- Prior art keywords
- preset
- model
- target
- trained
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The application discloses an automatic driving training method, equipment and medium based on federal learning, which are applied to a cloud server, wherein the method comprises the following steps: when a training instruction is detected, a preset model to be trained is obtained, and the preset model to be trained is sent to each target vehicle in a preset driving range; acquiring first target model parameters obtained after the target vehicles train the preset model to be trained based on respective real-time local data sets; executing a preset federation process based on the first target model parameters, and performing iterative training on the preset model to be trained to obtain second target model parameters; and sending the second target model parameters to the target vehicles so that each target vehicle can obtain a respective target model based on the second target model parameters to realize respective automatic driving functions. The method solves the technical problem that the effective automatic driving model is difficult to obtain quickly and accurately in the prior art.
Description
Technical Field
The application relates to the technical field of artificial intelligence of Internet science and technology, in particular to an automatic driving training method, equipment and medium based on federal learning.
Background
With the continuous development of financial technologies, especially internet technologies, more and more technologies (such as distributed, blockchain, artificial intelligence, etc.) are applied in the internet technology field, but the internet technology field industry also puts forward higher requirements on technologies, such as the internet technology field has higher requirements on automatic driving training based on federal learning.
Recently, the development of the automatic driving technology is very rapid, and the application of the automatic driving has the advantages of reducing the driving working intensity, relieving the driving fatigue, improving the driving safety, reducing the accident rate and the like, wherein the environment perception is one of the core technologies of the automatic driving, and the automatic driving is analyzed and decided through the result of the environment perception and is used as the basis of path planning, so that the automatic driving is realized, and the environment perception comprises lane detection, pedestrian detection, traffic sign recognition, obstacle detection and the like.
At present, environmental perception is realized through a model of an identification target obtained based on deep learning training, but in the prior art, the model of the identification target is often obtained through deep learning, namely through an offline mode, and then the actual application of each environmental scene is carried out, the model obtained through offline training often has locality, namely the problem that the model obtained through offline training is not matched with the actual environment, the offline training data is often huge, so that the data transmission load in the training process is excessive, the model training efficiency is low, namely the technical problem that an effective automatic driving model is difficult to quickly and accurately obtain exists in the prior art.
Disclosure of Invention
The application mainly aims to provide an automatic driving training method, equipment and medium based on federal learning, and aims to solve the technical problem that an effective automatic driving model is difficult to acquire quickly and accurately in the prior art.
In order to achieve the above object, the present application provides an automatic driving training method based on federal learning, which is applied to a cloud server, and the automatic driving training method based on federal learning includes:
when a training instruction is detected, a preset model to be trained is obtained, and the preset model to be trained is sent to each target vehicle in a preset driving range;
acquiring first target model parameters obtained after the target vehicles train the preset model to be trained based on respective real-time local data sets;
executing a preset federation process based on the first target model parameters, and performing iterative training on the preset model to be trained to obtain second target model parameters;
and sending the second target model parameters to the target vehicles so that each target vehicle can obtain a respective target model based on the second target model parameters to realize respective automatic driving functions.
Optionally, when the training instruction is detected, a preset model to be trained is obtained, and the step of sending the preset model to be trained to each target vehicle in a preset driving range includes:
when a training instruction is detected, selecting a preset model to be trained from a preset model set according to the region attribute carried in the training instruction;
and sending the preset model to be trained to each target vehicle in a preset driving range.
Optionally, when the training instruction is detected, the step of selecting the preset model to be trained from a preset model set according to the region attribute carried in the training instruction includes:
when a training instruction is detected, selecting a preset model subset from a preset model set according to the region attribute carried in the training instruction;
and selecting the preset model to be trained from a preset model subset according to the time information and the vehicle type information carried in the training instruction.
Optionally, the step of sending the second target model parameter to the target vehicle, so that each target vehicle obtains a respective target model based on the second target model parameter to implement a respective autopilot function, includes:
Encrypting the second target model parameters to obtain encrypted model parameters;
and sending the encryption model parameters to the target vehicles so that each target vehicle can obtain a respective target model based on the second target model parameters to realize respective automatic driving functions.
Optionally, when the training instruction is detected, a preset model to be trained is obtained, and the preset model to be trained is sent to each target vehicle in a preset driving range, which further includes:
when a training instruction is detected, a preset model to be trained is obtained;
and communicating with each base station in a preset running range, and distributing the preset model to be trained to each base station so that each base station can send the preset model to be trained to each target vehicle in the preset running range based on the distance information between each base station and each target vehicle.
Optionally, when the training instruction is detected, a preset model to be trained is obtained, and before the step of sending the preset model to be trained to each target vehicle in a preset driving range, the method includes:
acquiring a preset basic model;
performing iterative training on the preset basic model based on offline training data with a preset identification tag so as to train and update model parameters in the preset basic model until the preset basic model reaches a preset training completion condition;
And setting the preset basic model after reaching the preset training completion condition as a preset model to be trained.
Optionally, the offline training data with the preset identification tag includes offline laser radar scanning data with the preset identification tag, offline camera shooting data and offline millimeter wave radar data.
The application also provides an automatic driving training method based on federal learning, which is applied to vehicles and comprises the following steps:
receiving a preset model to be trained sent by a cloud server;
training the preset model to be trained based on a real-time local data set to obtain first target model parameters;
encrypting and transmitting the first target model parameters to the cloud server so that the cloud server executes a preset federal process, and performing iterative training on the preset model to be trained to obtain second target model parameters;
and receiving the second target model parameters sent by the cloud server in an encrypted mode, and obtaining a target model based on the second target model parameters so as to realize an automatic driving function.
The application also provides an automatic driving training device based on federal learning, which is applied to a cloud server, and comprises:
The first acquisition module is used for acquiring a preset model to be trained when a training instruction is detected, and transmitting the preset model to be trained to each target vehicle in a preset driving range;
the second acquisition module is used for acquiring first target model parameters obtained after the target vehicles train the preset model to be trained based on respective real-time local data sets;
the first training module is used for executing a preset federal process based on the first target model parameters, and carrying out iterative training on the preset model to be trained to obtain second target model parameters;
and the sending module is used for sending the second target model parameters to the target vehicles so that each target vehicle can obtain a respective target model based on the second target model parameters to realize respective automatic driving functions.
Optionally, the first acquisition module includes:
the selecting unit is used for selecting the preset model to be trained from a preset model set according to the region attribute carried in the training instruction when the training instruction is detected;
and the transmitting unit is used for transmitting the preset model to be trained to each target vehicle in a preset driving range.
Optionally, the selecting unit includes:
the first selecting subunit is used for selecting a preset model subset from a preset model set according to the region attribute carried in the training instruction when the training instruction is detected;
the second selecting subunit is configured to select the preset model to be trained from a preset model subset according to the time information and the vehicle type information carried in the training instruction.
Optionally, the sending module includes:
the encryption unit is used for carrying out encryption processing on the second target model parameters to obtain encryption model parameters;
and the sending unit is used for sending the encryption model parameters to the target vehicles so that each target vehicle can obtain a respective target model based on the second target model parameters to realize respective automatic driving functions.
Optionally, the first acquisition module further includes:
the detection unit is used for acquiring a preset model to be trained when the training instruction is detected;
the distribution unit is used for communicating with each base station in a preset running range, distributing the preset model to be trained to each base station, and enabling each base station to send the preset model to be trained to each target vehicle in the preset running range based on distance information between each base station and each target vehicle.
Optionally, the automatic driving training device based on federal learning further comprises:
the third acquisition module is used for acquiring a preset basic model;
the second training module is used for carrying out iterative training on the preset basic model based on offline training data with a preset identification tag so as to train and update model parameters in the preset basic model until the preset basic model reaches a preset training completion condition;
the setting module is used for setting the preset basic model after reaching the preset training completion condition as a preset model to be trained.
Optionally, the offline training data with the preset identification tag includes offline laser radar scanning data with the preset identification tag, offline camera shooting data and offline millimeter wave radar data.
The application also provides an automatic driving training device based on federal learning, which is applied to a vehicle, and the automatic driving training device based on federal learning comprises:
the first receiving module is used for receiving a preset model to be trained sent by the cloud server;
the fourth acquisition module is used for training the preset model to be trained based on a real-time local data set to obtain first target model parameters;
A fifth obtaining module, configured to encrypt and send the first target model parameter to the cloud server, so that the cloud server executes a preset federal procedure, and performs iterative training on the preset model to be trained to obtain a second target model parameter;
the second receiving module is used for receiving the second target model parameters sent by the cloud server in an encrypted mode, and obtaining a target model based on the second target model parameters so as to realize an automatic driving function.
The application also provides automatic driving training equipment based on federal learning, which is entity equipment, and comprises: the automatic driving training method based on the federal learning comprises a memory, a processor and a program of the automatic driving training method based on the federal learning, wherein the program of the automatic driving training method based on the federal learning is stored in the memory and can run on the processor, and the steps of the automatic driving training method based on the federal learning can be realized when the program of the automatic driving training method based on the federal learning is executed by the processor.
The application also provides a medium, wherein the medium is stored with a program for realizing the automatic driving training method based on federal learning, and the program for realizing the automatic driving training method based on federal learning realizes the steps of the automatic driving training method based on federal learning when being executed by a processor.
Compared with the prior art that after a model of an identification target is obtained through local offline data training, the model is put into different actual environments for use, when a training instruction is detected, a preset model to be trained is obtained, the preset model to be trained is sent to each target vehicle in a preset running range, and first target model parameters obtained after each target vehicle trains the preset model to be trained based on respective real-time local data sets are obtained; executing a preset federation process based on the first target model parameters, and performing iterative training on the preset model to be trained to obtain second target model parameters; and sending the second target model parameters to the target vehicles so that each target vehicle can obtain a respective target model based on the second target model parameters to realize respective automatic driving functions. According to the method, after the preset model to be trained is obtained, the federal model is trained on line and in real time based on each target vehicle in an actual environment scene, so that the problem that the federal model is not matched with the actual environment or has local limitation is avoided.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the application and together with the description, serve to explain the principles of the application.
In order to more clearly illustrate the embodiments of the application or the technical solutions of the prior art, the drawings which are used in the description of the embodiments or the prior art will be briefly described, and it will be obvious to a person skilled in the art that other drawings can be obtained from these drawings without inventive effort.
FIG. 1 is a schematic flow chart of a first embodiment of an automatic driving training method based on federal learning;
FIG. 2 is a detailed flowchart of steps for acquiring a preset model to be trained and transmitting the preset model to be trained to each target vehicle within a preset driving range when a training instruction is detected in a first embodiment of an automatic driving training method based on federal learning;
fig. 3 is a schematic device structure diagram of a hardware running environment according to an embodiment of the present application.
The achievement of the objects, functional features and advantages of the present application will be further described with reference to the accompanying drawings, in conjunction with the embodiments.
Detailed Description
It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the application.
In a first embodiment of the federal learning-based automatic driving training method, referring to fig. 1, the federal learning-based automatic driving training method is applied to a cloud server, and includes:
step S10, when a training instruction is detected, a preset model to be trained is obtained, and the preset model to be trained is sent to each target vehicle in a preset running range;
step S20, obtaining first target model parameters obtained after the target vehicles train the preset model to be trained based on respective real-time local data sets;
step S30, executing a preset federation process based on the first target model parameters, and performing iterative training on the preset model to be trained to obtain second target model parameters;
step S40, sending the second target model parameters to the target vehicles, so that each target vehicle obtains a respective target model based on the second target model parameters to implement a respective automatic driving function.
The method comprises the following specific steps:
step S10, when a training instruction is detected, a preset model to be trained is obtained, and the preset model to be trained is sent to each target vehicle in a preset running range;
in this embodiment, the automatic driving training method based on federal learning is applied to a cloud server, where the cloud server is in communication connection with each participant through a base station, and it should be noted that, in this embodiment, each participant is a target vehicle, where the target vehicle and the cloud server together form an automatic driving training system based on federal learning, and the automatic driving training system based on federal learning belongs to an automatic driving training device based on federal learning, and it should be noted that, in this embodiment, the automatic driving training system based on federal learning may be divided into automatic driving training subsystems of federal learning in different regions, specifically, for example, an automatic driving training subsystem of federal learning in region a and an automatic driving training subsystem of federal learning in region B, where, the training instruction can be triggered by a user through an automatic driving training system (or a subsystem) based on federal learning, the cloud server can detect the training instruction after the automatic driving training system based on federal learning triggers the training instruction, the cloud server acquires a preset model to be trained after detecting the training instruction, the preset model to be trained is sent to each target vehicle in a preset running range, it is required to say that a plurality of models exist in the cloud server, therefore, the preset model to be trained needs to be selected from the plurality of models, the preset model to be trained is sent to each target vehicle in the preset running range, wherein, the mode of selecting the preset model to be trained from the plurality of models can be extracting target model attribute information from the training instruction so as to select the preset model to be trained from the plurality of models, or model information related to the training instruction, selecting a preset model to be trained from a plurality of models.
When a training instruction is detected, a preset model to be trained is obtained, the preset model to be trained is sent to each target vehicle in a preset running range in real time on line, wherein the preset running range can be determined by the preset model to be trained, namely, the preset model to be trained is determined, the preset running range is determined, and the step of sending the preset model to be trained to each target vehicle in the preset running range can be that: and sending the preset model to be trained to all vehicles in a preset running range, or sending the preset model to be trained to a certain type of vehicle in the preset running range, for example, sending the preset model to be trained to a cargo truck type vehicle in the preset running range, sending the preset model to be trained to a car type vehicle in the preset running range, or sending the preset model to be trained to a passenger car type vehicle in the preset running range. In this embodiment, it should be noted that, the purpose of sending the preset model to be trained to a certain type of vehicle within a preset driving range is to: the training accuracy is improved, because the vehicles are different, the vehicle cameras or the radar have different environmental perceptions, and therefore, the training is independently carried out according to different vehicle types, the data pertinence is strong, and therefore, the training accuracy can be improved.
It should be noted that, in this embodiment, when the preset model to be trained is sent to a certain type of vehicle within the preset driving range, the vehicle type needs to be identified when the vehicle enters the preset driving range, specifically, the type of the vehicle may be identified according to the roadside units of the road system, and after the identification, the cloud server receives the identified vehicle type to finally determine the target vehicle.
Referring to fig. 2, the step of acquiring a preset model to be trained when a training instruction is detected, and transmitting the preset model to be trained to each target vehicle within a preset driving range includes:
step S11, when a training instruction is detected, selecting a preset model to be trained from a preset model set according to the region attribute carried in the training instruction;
in this embodiment, when a training instruction is detected, the preset model to be trained is selected from a preset model set according to an area attribute carried in the training instruction, for example, 100 models in the preset model set are shared, and the 100 models correspond to different cities, and then the preset model to be trained is selected from the preset model set according to city information carried in the training instruction.
And step S12, the preset model to be trained is sent to each target vehicle in a preset driving range.
After the preset model to be trained is obtained, the preset model to be trained is sent to each target vehicle in a preset running range in real time on line, namely, as long as the vehicle enters the preset running range, the preset model to be trained is sent to each target vehicle in the preset running range, so that the target vehicles can train locally.
When the training instruction is detected, selecting the preset model to be trained from a preset model set according to the region attribute carried in the training instruction, wherein the step comprises the following steps:
step S111, when a training instruction is detected, selecting a preset model subset from a preset model set according to the region attribute carried in the training instruction;
step S112, selecting the preset model to be trained from a preset model subset according to the time information and the vehicle type information carried in the training instruction.
In this embodiment, after selecting a subset of the preset models from the preset model set, the preset model to be trained is selected from the preset model subset according to the time information and the vehicle type information carried in the training instruction, that is, in this embodiment, the selection of the preset model to be trained is simultaneously associated with the vehicle type and the time, because, in different time periods, the sunlight is different, the light change rule is different, the ingestion of the vehicle-mounted camera on the vehicle is changed, thus, the preset model to be trained is simultaneously selected according to the vehicle type and the time, specifically, assuming that 100 models in the preset model subset are used in total, and 10 models are used in summer and passenger cars, the preset model to be trained is selected from the 10 models, in this embodiment, the preset model to be trained is selected from the preset model subset according to the time information and the vehicle type information carried in the training instruction, so that the rate and the accuracy of model training are improved, and the model to be trained is required to be updated in the spring environment when the preset model is not used in autumn.
When the training instruction is detected, a preset model to be trained is obtained, and before the step of sending the preset model to be trained to each target vehicle in a preset driving range, the method comprises the following steps:
step S01, a preset basic model is obtained;
in this embodiment, it should be noted that, although the model generated by offline training may have a certain bias, that is, the correlation between the sample of the offline data and the location, environment, etc. of the acquisition is relatively large, and there is often a case of local sample, for example, the data acquisition of the offline training is limited to a certain area and limited to a certain specific environment (illumination, weather, etc.), so as to affect the training result, in this embodiment, in order to ensure the security of the initial training stage, the offline data is subjected to offline training, and the obtained target model is used as the preset model to be trained in the initial training stage, that is, the preset model to be trained is obtained by offline training through the preset basic model.
Step S02, performing iterative training on the preset basic model based on offline training data with a preset identification tag so as to train and update model parameters in the preset basic model until the preset basic model reaches a preset training completion condition;
And carrying out iterative training on the preset basic model based on offline training data with preset identification tags so as to train and update model parameters in the preset basic model until the preset basic model reaches preset training completion conditions, wherein the offline training data with the preset identification tags comprises offline laser radar scanning data with the preset identification tags, offline camera shooting data, offline millimeter wave radar data and the like. Based on the offline training data with the preset identification tag, specifically, based on the offline training data with the preset identification tag of different areas (or different times and different vehicle types), the preset basic model is subjected to iterative training to obtain different types of preset models to be trained, specifically, based on the offline training data with the preset identification tag, the preset basic model is subjected to iterative federal training to perform training update on model parameters in the preset basic model until the preset basic model reaches preset training completion conditions, wherein the preset training completion conditions comprise that the training times reach a second preset times or the first preset loss function converges.
Step S03, setting the preset basic model after reaching the preset training completion condition as a preset model to be trained.
Setting the preset basic model which reaches the preset training completion condition as a preset model to be trained, wherein the preset model to be trained is a plurality of models.
Step S20, obtaining first target model parameters obtained after the target vehicles train the preset model to be trained based on respective real-time local data sets;
in this embodiment, after determining the target vehicles, acquiring each first target model parameter obtained after the target vehicles train the preset model to be trained based on the respective real-time local data sets, and specifically, acquiring each first target model parameter obtained after the target vehicles train the preset model to be trained based on the respective real-time local data sets with the respective labels, where each target vehicle may be trained for a second preset number of times, and then obtaining the corresponding first target model parameter.
Step S30, executing a preset federation process based on the first target model parameters, and performing iterative training on the preset model to be trained to obtain second target model parameters;
And executing a preset federal process based on the first target model parameters, performing iterative training on the preset model to be trained to obtain second target model parameters, specifically, transmitting the first target model parameters to a cloud server side by each target vehicle, performing aggregation means such as average value taking and the like on each first target model parameter after each first target model parameter is received by the cloud server side to obtain first aggregation parameters, transmitting the first aggregation parameters to each target vehicle after the first aggregation parameters are obtained, performing local iterative training on each target vehicle to obtain target model parameters corresponding to the next round, transmitting the target model parameters of the next round to the cloud server side after the target model parameters corresponding to the next round are obtained, performing iterative training on the basis of the first target model parameters by each target vehicle until a second preset loss function of the system converges, and setting the converged model parameters as the second target model parameters.
Step S40, sending the second target model parameters to the target vehicles, so that each target vehicle obtains a respective target model based on the second target model parameters to implement a respective automatic driving function.
In this embodiment, after the second target model parameters are obtained, the second target model parameters are sent to the target vehicles, and the target vehicles perfect the local target model based on the second target model parameters, that is, each target vehicle obtains a respective target model based on the second target model parameters, and after the target model is obtained, the automatic driving function of each target vehicle is realized.
Compared with the prior art that different sample data are converted into an embedded vector by a participant and interactive federal modeling is carried out on the embedded vector by a server, the method and the device acquire the sample data, wherein the sample data comprise user data and article data; determining, locally at the first party, a user-embedded vector of the user data and receiving an item-embedded vector sent by a server, the item-embedded vector being generated locally at the server based on the item data; and obtaining a preset prediction model of the first participant through federal learning training based on the user embedded vector and the article embedded vector. In the method, after sample data are obtained, a user embedding vector of the user data is determined at the local side of the first participant, the article embedding vector determined at the server side based on the article data is received, and then a preset prediction model of the first participant is obtained through federal learning training based on the user embedding vector and the article embedding vector, namely, different sample data are converted into different embedding vectors, and a large amount of calculated article data are particularly caused to be placed at the server side to generate the article embedding vector, and only the interactive federal modeling of the article embedding vector is carried out with the server side without the interaction of the user embedding vector, so that a large amount of calculation overhead and communication overhead are prevented from being generated at the local side of the participant, and the obtaining efficiency of the preset prediction model is improved.
In another embodiment of the federal learning-based automatic driving training method, the step of sending the second target model parameters to the target vehicles to obtain respective target models by the target vehicles based on the second target model parameters so as to realize respective automatic driving functions includes:
a1, carrying out encryption processing on the second target model parameters to obtain encrypted model parameters;
in this embodiment, in order to ensure security, when the target vehicle interacts with the cloud server, the target vehicle interacts with the cloud server through encryption, for example, the first target model parameter is encrypted and sent to the cloud server, and the cloud server encrypts the second target model parameter to obtain an encrypted model parameter.
And step A2, the encryption model parameters are sent to the target vehicles so that each target vehicle can obtain a respective target model based on the second target model parameters to realize respective automatic driving functions.
And sending the encryption model parameters to the target vehicles so that each target vehicle obtains a respective target model based on the second target model parameters to realize respective automatic driving functions.
In this embodiment, the encryption model parameter is obtained by performing encryption processing on the second target model parameter; and sending the encryption model parameters to the target vehicles so that each target vehicle can obtain a respective target model based on the second target model parameters to realize respective automatic driving functions. In this embodiment, the secure acquisition of the target model is realized.
In another embodiment of the federal learning-based automatic driving training method, when a training instruction is detected, a preset model to be trained is obtained, and the preset model to be trained is sent to each target vehicle in a preset driving range, and the method further comprises the steps of:
step B1, when a training instruction is detected, a preset model to be trained is obtained;
and step B2, communicating with each base station in a preset running range, and distributing the preset model to be trained to each base station so that each base station can send the preset model to be trained to each target vehicle in the preset running range based on the distance information between each base station and each target vehicle.
In this embodiment, when a training instruction is detected, the cloud server acquires a preset model to be trained, the cloud server communicates with each base station within a preset driving range, distributes the preset model to each base station, so that each base station sends the preset model to each target vehicle within the preset driving range based on distance information between the preset model and each target vehicle, that is, in this embodiment, in order to reduce communication duration, each target vehicle quickly acquires the preset model to be trained, so as to finally achieve quick acquisition of the target model, each base station sends the preset model to each target vehicle within the preset driving range based on distance information between the preset model to each target vehicle, and specifically, each target vehicle is in the shortest communication distance, so as to acquire the preset model to be trained.
In the embodiment, a preset model to be trained is obtained when a training instruction is detected; and communicating with each base station in a preset running range, and distributing the preset model to be trained to each base station so that each base station can send the preset model to be trained to each target vehicle in the preset running range based on the distance information between each base station and each target vehicle. In this embodiment, each target vehicle obtains a preset model to be trained based on the nearest distance, and model training efficiency is improved.
The embodiment of the application provides an automatic driving training method based on federal learning, which is applied to a vehicle in another embodiment of the automatic driving training method based on federal learning, and comprises the following steps:
receiving a preset model to be trained sent by a cloud server;
training the preset model to be trained based on a real-time local data set to obtain first target model parameters;
encrypting and transmitting the first target model parameters to the cloud server so that the cloud server executes a preset federal process, and performing iterative training on the preset model to be trained to obtain second target model parameters;
and receiving the second target model parameters sent by the cloud server in an encrypted mode, and obtaining a target model based on the second target model parameters so as to realize an automatic driving function.
In this embodiment, for a vehicle, a preset model to be trained sent by a cloud server is received; training the preset model to be trained based on a real-time local data set to obtain first target model parameters; encrypting and transmitting the first target model parameters to the cloud server so that the cloud server executes a preset federal process, and performing iterative training on the preset model to be trained to obtain second target model parameters; and receiving the second target model parameters sent by the cloud server in an encrypted mode, and obtaining a target model based on the second target model parameters so as to realize an automatic driving function. In this embodiment, the target vehicle trains the target model on line in real time through federation to achieve automatic driving.
Referring to fig. 3, fig. 3 is a schematic device structure diagram of a hardware running environment according to an embodiment of the present application.
As shown in fig. 3, the federal learning-based automatic driving training apparatus may include: a processor 1001, such as a CPU, memory 1005, and a communication bus 1002. Wherein a communication bus 1002 is used to enable connected communication between the processor 1001 and a memory 1005. The memory 1005 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also optionally be a storage device separate from the processor 1001 described above.
Optionally, the federal learning-based autopilot training apparatus may further include a rectangular user interface, a network interface, a camera, an RF (Radio Frequency) circuit, a sensor, an audio circuit, a WiFi module, and the like. The rectangular user interface may include a Display screen (Display), an input sub-module such as a Keyboard (Keyboard), and the optional rectangular user interface may also include a standard wired interface, a wireless interface. The network interface may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface).
It will be appreciated by those skilled in the art that the federal learning-based automatic driving training apparatus structure illustrated in fig. 3 is not limiting of the federal learning-based automatic driving training apparatus, and may include more or fewer components than illustrated, or certain components may be combined, or a different arrangement of components.
As shown in fig. 3, an operating system, a network communication module, and a federal learning-based autopilot training program may be included in a memory 1005 as a computer medium. The operating system is a program that manages and controls federally-learned autopilot training device hardware and software resources, supporting federally-learned autopilot training programs and the execution of other software and/or programs. The network communication module is used to enable communication between components within the memory 1005 and other hardware and software in the federal learning-based autopilot training system.
In the federal learning-based automatic driving training apparatus shown in fig. 3, the processor 1001 is configured to execute the federal learning-based automatic driving training program stored in the memory 1005, to implement the steps of the federal learning-based automatic driving training method described in any one of the above.
The specific implementation mode of the automatic driving training equipment based on federal learning is basically the same as the above-mentioned automatic driving training method based on federal learning, and is not repeated here.
The application also provides an automatic driving training device based on federal learning, which is applied to a cloud server and comprises:
The first acquisition module is used for acquiring a preset model to be trained when a training instruction is detected, and transmitting the preset model to be trained to each target vehicle in a preset driving range;
the second acquisition module is used for acquiring first target model parameters obtained after the target vehicles train the preset model to be trained based on respective real-time local data sets;
the first training module is used for executing a preset federal process based on the first target model parameters, and carrying out iterative training on the preset model to be trained to obtain second target model parameters;
and the sending module is used for sending the second target model parameters to the target vehicles so that each target vehicle can obtain a respective target model based on the second target model parameters to realize respective automatic driving functions.
Optionally, the first acquisition module includes:
the selecting unit is used for selecting the preset model to be trained from a preset model set according to the region attribute carried in the training instruction when the training instruction is detected;
and the transmitting unit is used for transmitting the preset model to be trained to each target vehicle in a preset driving range.
Optionally, the selecting unit includes:
the first selecting subunit is used for selecting a preset model subset from a preset model set according to the region attribute carried in the training instruction when the training instruction is detected;
the second selecting subunit is configured to select the preset model to be trained from a preset model subset according to the time information and the vehicle type information carried in the training instruction.
Optionally, the sending module includes:
the encryption unit is used for carrying out encryption processing on the second target model parameters to obtain encryption model parameters;
and the sending unit is used for sending the encryption model parameters to the target vehicles so that each target vehicle can obtain a respective target model based on the second target model parameters to realize respective automatic driving functions.
Optionally, the first acquisition module further includes:
the detection unit is used for acquiring a preset model to be trained when the training instruction is detected;
the distribution unit is used for communicating with each base station in a preset running range, distributing the preset model to be trained to each base station, and enabling each base station to send the preset model to be trained to each target vehicle in the preset running range based on distance information between each base station and each target vehicle.
Optionally, the automatic driving training device based on federal learning further comprises:
the third acquisition module is used for acquiring a preset basic model;
the second training module is used for carrying out iterative training on the preset basic model based on offline training data with a preset identification tag so as to train and update model parameters in the preset basic model until the preset basic model reaches a preset training completion condition;
the setting module is used for setting the preset basic model after reaching the preset training completion condition as a preset model to be trained.
Optionally, the offline training data with the preset identification tag includes offline laser radar scanning data with the preset identification tag, offline camera shooting data and offline millimeter wave radar data.
The application also provides an automatic driving training device based on federal learning, which is applied to a vehicle, and the automatic driving training device based on federal learning comprises:
the first receiving module is used for receiving a preset model to be trained sent by the cloud server;
the fourth acquisition module is used for training the preset model to be trained based on a real-time local data set to obtain first target model parameters;
A fifth obtaining module, configured to encrypt and send the first target model parameter to the cloud server, so that the cloud server executes a preset federal procedure, and performs iterative training on the preset model to be trained to obtain a second target model parameter;
the second receiving module is used for receiving the second target model parameters sent by the cloud server in an encrypted mode, and obtaining a target model based on the second target model parameters so as to realize an automatic driving function.
The specific implementation manner of the automatic driving training device based on federal learning is basically the same as the above-mentioned examples of the automatic driving training method based on federal learning, and is not repeated here.
Embodiments of the present application provide a medium, and the medium stores one or more programs that are further executable by one or more processors for implementing the steps of the federal learning-based autopilot training method described in any one of the above.
The specific implementation manner of the medium of the application is basically the same as the above-mentioned embodiments of the automatic driving training method based on federal learning, and will not be repeated here.
The foregoing description is only of the preferred embodiments of the present application, and is not intended to limit the scope of the application, but rather is intended to cover any equivalents of the structures or equivalent processes disclosed herein, or any application, directly or indirectly, within the scope of the application.
Claims (9)
1. The utility model provides an automatic driving training method based on federal study, which is characterized in that is applied to the cloud server, the cloud server constitutes the automatic driving training system based on federal study with each target vehicle jointly, and automatic driving training method based on federal study includes:
when a training instruction is detected, a preset model to be trained, which is obtained through offline training through a preset basic model and offline data, is obtained, and the preset model to be trained is sent to each target vehicle in a preset driving range in real time on line;
acquiring first target model parameters obtained after the target vehicles train the preset model to be trained based on respective real-time local data sets;
based on the first target model parameters, performing on-line preset federation flow, and performing iterative training on the preset model to be trained to obtain second target model parameters, wherein the preset federation flow comprises: each target vehicle sends first target model parameters to a cloud server side, the cloud server side receives each first target model parameter and then aggregates each first target model parameter to obtain a first aggregate parameter, after the first aggregate parameter is obtained, the first aggregate parameter is sent to each target vehicle so that each target vehicle carries out local iterative training of a next round to obtain a target model parameter corresponding to the next round, and after the target model parameter corresponding to the next round is obtained, each target vehicle sends the target model parameter of the next round to the cloud server side so as to be aggregated by the cloud server side;
The second target model parameters are sent to the target vehicles so that each target vehicle can obtain a local target model on line based on the second target model parameters, and therefore the automatic driving function of each target vehicle can be achieved;
when the training instruction is detected, a preset model to be trained is obtained, and the preset model to be trained is sent to each target vehicle in a preset driving range, which comprises the following steps:
when a training instruction is detected, selecting a preset model to be trained from a preset model set according to the region attribute carried in the training instruction;
and sending the preset model to be trained to each target vehicle in a preset driving range, wherein the preset driving range is determined by the preset model to be trained, namely the preset model to be trained, and the preset driving range is determined.
2. The automatic driving training method based on federal learning according to claim 1, wherein the step of selecting the preset model to be trained from a preset model set according to the region attribute carried in the training instruction when the training instruction is detected comprises:
when a training instruction is detected, selecting a preset model subset from a preset model set according to the region attribute carried in the training instruction;
And selecting the preset model to be trained from a preset model subset according to the time information and the vehicle type information carried in the training instruction.
3. The federal learning-based autopilot training method of claim 1 wherein the step of transmitting the second target model parameters to the target vehicles for each of the target vehicles to obtain a respective target model based on the second target model parameters to implement a respective autopilot function comprises:
encrypting the second target model parameters to obtain encrypted model parameters;
and sending the encryption model parameters to the target vehicles so that each target vehicle can obtain a respective target model based on the second target model parameters to realize respective automatic driving functions.
4. The federal learning-based automatic driving training method according to claim 1, wherein the step of acquiring a preset model to be trained and transmitting the preset model to be trained to each target vehicle within a preset driving range when a training instruction is detected, further comprises:
when a training instruction is detected, a preset model to be trained is obtained;
And communicating with each base station in a preset running range, and distributing the preset model to be trained to each base station so that each base station can send the preset model to be trained to each target vehicle in the preset running range based on the distance information between each base station and each target vehicle.
5. The federal learning-based automatic driving training method according to claim 1, wherein, before the step of transmitting the preset model to be trained to each target vehicle within a preset driving range in real time on line, the method comprises:
acquiring a preset basic model;
performing iterative training on the preset basic model based on offline training data with a preset identification tag so as to train and update model parameters in the preset basic model until the preset basic model reaches a preset training completion condition; and setting the preset basic model after the preset training completion condition is reached as a preset model to be trained.
6. The federal learning-based automatic driving training method according to claim 5, wherein the offline training data having a preset identification tag includes offline lidar scan data having a preset identification tag, offline camera data, and offline millimeter wave radar data.
7. An automatic driving training method based on federal learning, which is characterized by being applied to a vehicle, comprising the following steps:
receiving a preset model to be trained sent by a cloud server, wherein the cloud server acquires the preset model to be trained obtained through offline training through a preset basic model and offline data when detecting a training instruction, and sends the preset model to be trained to each target vehicle in a preset driving range in real time on line;
training the preset model to be trained based on a real-time local data set to obtain first target model parameters;
encrypting and sending the first target model parameters to the cloud server so that the cloud server executes a preset federal process, and performing iterative training on the preset model to be trained to obtain second target model parameters, wherein the preset federal process comprises: each target vehicle sends first target model parameters to a cloud server side, the cloud server side receives each first target model parameter and then aggregates each first target model parameter to obtain a first aggregate parameter, after the first aggregate parameter is obtained, the first aggregate parameter is sent to each target vehicle so that each target vehicle carries out local iterative training of a next round to obtain a target model parameter corresponding to the next round, and after the target model parameter corresponding to the next round is obtained, each target vehicle sends the target model parameter of the next round to the cloud server side so as to be aggregated by the cloud server side;
Receiving the second target model parameters sent by the cloud server in an encrypted mode, and obtaining a target model based on the second target model parameters so as to realize an automatic driving function;
the step of receiving the preset model to be trained sent by the cloud server comprises the following steps:
receiving a preset model to be trained sent by a cloud server, wherein when the cloud server detects a training instruction, the cloud server selects the preset model to be trained from a preset model set according to the region attribute carried in the training instruction; and sending the preset model to be trained to each target vehicle in a preset driving range, wherein the preset driving range is determined by the preset model to be trained, namely, the preset model to be trained is determined, and the preset driving range is determined.
8. Automatic driving training equipment based on federal study, characterized by, automatic driving training equipment based on federal study includes: a memory, a processor and a program stored on the memory for implementing the federal learning-based autopilot training method,
the memory is used for storing a program for realizing the automatic driving training method based on federal learning;
the processor is configured to execute a program implementing the federal learning-based autopilot training method to implement the steps of the federal learning-based autopilot training method of any one of claims 1 to 7.
9. A computer-readable storage medium, characterized in that the computer-readable storage medium has stored thereon a program for implementing a federal learning-based automatic driving training method, the program for implementing the federal learning-based automatic driving training method being executed by a processor to implement the steps of the federal learning-based automatic driving training method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010931767.1A CN112052959B (en) | 2020-09-04 | 2020-09-04 | Automatic driving training method, equipment and medium based on federal learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010931767.1A CN112052959B (en) | 2020-09-04 | 2020-09-04 | Automatic driving training method, equipment and medium based on federal learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112052959A CN112052959A (en) | 2020-12-08 |
| CN112052959B true CN112052959B (en) | 2023-08-25 |
Family
ID=73609868
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010931767.1A Active CN112052959B (en) | 2020-09-04 | 2020-09-04 | Automatic driving training method, equipment and medium based on federal learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112052959B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112486180A (en) * | 2020-12-10 | 2021-03-12 | 深圳前海微众银行股份有限公司 | Vehicle control method, device, equipment, storage medium and program product |
| CN112766138A (en) * | 2021-01-14 | 2021-05-07 | 深圳前海微众银行股份有限公司 | Positioning method, device and equipment based on image recognition and storage medium |
| CN113313264B (en) * | 2021-06-02 | 2022-08-12 | 河南大学 | Efficient federal learning method in Internet of vehicles scene |
| CN113920780B (en) * | 2021-09-01 | 2022-12-16 | 同济大学 | Cloud and mist collaborative personalized forward collision risk early warning method based on federal learning |
| CN114740970A (en) * | 2022-02-23 | 2022-07-12 | 广东工业大学 | Millimeter wave gesture recognition method and system based on federal learning |
| CN114548608B (en) * | 2022-04-26 | 2022-08-05 | 腾讯科技（深圳）有限公司 | Model processing method and device, target traffic equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106446754A (en) * | 2015-08-11 | 2017-02-22 | 阿里巴巴集团控股有限公司 | Image identification method, metric learning method, image source identification method and devices |
| CN109747659A (en) * | 2018-11-26 | 2019-05-14 | 北京汽车集团有限公司 | The control method and device of vehicle drive |
| CN110196593A (en) * | 2019-05-16 | 2019-09-03 | 济南浪潮高新科技投资发展有限公司 | A kind of more scene environments detections of automatic Pilot and decision system and method |
| CN110867096A (en) * | 2019-10-08 | 2020-03-06 | 江苏大学 | Mountain road safety control system and method suitable for vehicles with multiple automatic driving grades |
| CN111428881A (en) * | 2020-03-20 | 2020-07-17 | 深圳前海微众银行股份有限公司 | Recognition model training method, device, equipment and readable storage medium |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111413957B (en) * | 2018-12-18 | 2021-11-02 | 北京航迹科技有限公司 | System and method for determining driving actions in autonomous driving |
-
2020
- 2020-09-04 CN CN202010931767.1A patent/CN112052959B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106446754A (en) * | 2015-08-11 | 2017-02-22 | 阿里巴巴集团控股有限公司 | Image identification method, metric learning method, image source identification method and devices |
| CN109747659A (en) * | 2018-11-26 | 2019-05-14 | 北京汽车集团有限公司 | The control method and device of vehicle drive |
| CN110196593A (en) * | 2019-05-16 | 2019-09-03 | 济南浪潮高新科技投资发展有限公司 | A kind of more scene environments detections of automatic Pilot and decision system and method |
| CN110867096A (en) * | 2019-10-08 | 2020-03-06 | 江苏大学 | Mountain road safety control system and method suitable for vehicles with multiple automatic driving grades |
| CN111428881A (en) * | 2020-03-20 | 2020-07-17 | 深圳前海微众银行股份有限公司 | Recognition model training method, device, equipment and readable storage medium |
Non-Patent Citations (1)
| Title |
|---|
| 联邦学习可视化:挑战与框架;潘如晟等;《计算机辅助设计与图形学学报》;20200430(第04期);第513-519页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112052959A (en) | 2020-12-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112052959B (en) | Automatic driving training method, equipment and medium based on federal learning | |
| US10665105B2 (en) | Dynamic-map constructing method, dynamic-map constructing system, and moving terminal | |
| CN112712733B (en) | Vehicle-road cooperation-based collision early warning method and system and road side unit | |
| US11210770B2 (en) | AI-based inspection in transportation | |
| US20170127246A1 (en) | Method for Sharing Travelling Information, Vehicle Platform and Intelligent Traffic System | |
| EP4036886A2 (en) | Method and apparatus for monitoring vehicle, cloud control platform and system for vehicle-road collaboration | |
| SE541962C2 (en) | Method and apparatus for detecting vehicle contour based on point cloud data | |
| WO2021135948A1 (en) | Data transmission method and device | |
| CN112203216B (en) | Positioning information acquisition method, driving assistance method and vehicle end sensor detection method | |
| US20230078241A1 (en) | Driving assistance processing method and apparatus, computer-readable medium, and electronic device | |
| CN114228743B (en) | Unmanned logistics vehicle control method, device and system and readable storage medium | |
| KR20210090624A (en) | Data distribution systems, sensor devices and servers | |
| CN113246991B (en) | Data transmission method and device for remote driving end of unmanned vehicle | |
| CN113189989B (en) | Vehicle intention prediction method, device, equipment and storage medium | |
| CN113612844A (en) | Data sharing method and device for vehicle end, medium and electronic equipment | |
| Volk et al. | Environment-aware development of robust vision-based cooperative perception systems | |
| US20220207997A1 (en) | Traffic Flow Estimation Apparatus, Traffic Flow Estimation Method, Traffic Flow Estimation Program, And Storage Medium Storing Traffic Flow Estimation Program | |
| CN110789515A (en) | System and method for hardware validation in a motor vehicle | |
| CN108764722A (en) | Hazardous materials transportation vehicles management method, system and computer readable storage medium | |
| Yang et al. | Stochastic Parameter Identification Method for Driving Trajectory Simulation Processes Based on Mobile Edge Computing and Self-Organizing Feature Mapping | |
| CN112766138A (en) | Positioning method, device and equipment based on image recognition and storage medium | |
| US20220004777A1 (en) | Information processing apparatus, information processing system, information processing method, and program | |
| CN112150811A (en) | Urban road intersection traffic information acquisition system based on image recognition technology | |
| CN111800465A (en) | Vehicle message processing method, device, medium and electronic equipment | |
| AlKishri et al. | Object recognition for organizing the movement of self-driving car |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |