An in-vehicle AI information processing system based on Ethernet TSN and a processing method thereof
By using the neural network adaptive parameter configuration model in the on-board AI information processing system and adjusting the TSN configuration parameters according to the driving environment, the problem of difficult TSN network configuration in the existing technology is solved, and better driving experience and network flexibility are achieved sex.
Patent Information
- Application Number
- CN202411216712.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-02
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-09-02
AI Technical Summary
The prior art is difficult to adaptively regulate the configuration of the on-board TSN network according to the real-time driving environment, resulting in limited applicability of the TSN network and affecting the driving experience.
Design an on-board AI information processing system based on Ethernet TSN, including TSN network, on-board computer ECU and TSN configuration server, use neural network to establish an adaptive parameter configuration model, adjust TSN configuration parameters in real time according to driving environment, and improve the flexibility of TSN network sex.
It realizes adaptive regulation of the TSN network, enhances the driving experience, improves the flexibility and applicability of the on-board network, and ensures low-cost and high-performance data transmission in different driving environments.
Smart Images

Figure CN118740632B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of vehicle-mounted TSN technology, and in particular to an Ethernet TSN-based vehicle-mounted AI information processing system and a processing method thereof. Background Art
[0002] Traditional in-vehicle network systems and TSN systems are mainly burned into the in-vehicle ECU in a statically planned and pre-configured manner. However, this static method is extremely complex and difficult to operate for the configuration and update of TSN scenarios.
[0003] Nowadays, the TSN protocol is used in more and more scenarios in vehicle networks, and more and more parameters need to be configured, and the interdependent configurations are becoming more and more complex. It is almost impossible to statically plan all TSN configuration parameters simply by relying on the design stage of vehicle electronic and electrical architecture; in addition, it is extremely difficult to dynamically customize TSN parameters in multiple scenarios based on the flexible and changeable TSN application scenarios.
[0004] Therefore, it is difficult in the existing technology to adaptively control the configuration of the TSN network according to the real-time driving environment, which limits the applicability of the TSN network and ultimately affects the driving experience. Summary of the invention
[0005] The purpose of the present invention is to provide an in-vehicle AI information processing system based on Ethernet TSN and a processing method thereof, so as to solve the technical problem in the prior art that the configuration of the TSN network is adaptively controlled according to the real-time driving environment, resulting in limited applicability of the TSN network and ultimately affecting the driving experience.
[0006] In order to solve the above technical problems, the present invention specifically provides the following technical solutions:
[0007] An in-vehicle AI information processing system based on Ethernet TSN includes a TSN network, an in-vehicle computer ECU, at least one AI functional domain unit, and a TSN configuration server, wherein:
[0008] The TSN network is arranged between the vehicle computer ECU and each AI functional domain unit to realize information interconnection between the vehicle computer ECU and the AI functional domain units;
[0009] The TSN configuration server is used to adaptively control the TSN configuration parameters between the on-board computer ECU and each AI functional domain unit according to the driving environment, thereby improving the flexibility of the TSN network and enhancing the driving experience.
[0010] As a preferred solution of the present invention, the TSN configuration parameters include: a transmission path and a transmission protocol.
[0011] As a preferred solution of the present invention, the AI functional domain unit includes at least one intelligent terminal, and each intelligent terminal is connected by a CAN bus to achieve information interconnection between the intelligent terminals.
[0012] As a preferred solution of the present invention, the TSN configuration server and each AI functional domain unit are connected using a CAN bus to achieve information interconnection between the TSN configuration server and each AI functional domain unit.
[0013] As a preferred solution of the present invention, the TSN configuration server is loaded with an adaptive parameter configuration model for adaptively configuring the TSN network according to the driving environment, wherein:
[0014] When the TSN network is transiently configured according to the driving environment,
[0015] The adaptive parameter configuration model is:
[0016] ;
[0017] In the formula, The output of the model is The TSN configuration parameters of the AI functional domain unit at the tth time, For the The AI information of the AI functional domain unit at the tth moment, , For the AI functional domain unit The AI information of a smart terminal at the tth moment, m is the number of smart terminals in the AI functional domain unit, and BP is the neural network;
[0018] The loss function of the adaptive parameter configuration model is:
[0019] ;
[0020] Where Ls(t) is the loss function at the tth moment, For the The degree of change in the driving environment monitored by the AI functional domain unit at the tth moment, The output of the model is The TSN configuration parameters of the AI functional domain unit at the tth time, For the The true value of the TSN configuration parameter of an AI functional domain unit at the tth moment, is the total number of AI functional domain units;
[0021] When the TSN network is configured in a steady state according to the driving environment,
[0022] The adaptive parameter configuration model is:
[0023] ;
[0024] In the formula, The output of the model is The TSN configuration parameters of an AI functional domain unit in the Tth time period, For the The AI information of the AI functional domain unit in the Tth time period, , For the AI functional domain unit The first intelligent terminal in the Tth time period AI information at a certain moment, is the first moments, m is the number of intelligent terminals in the AI functional domain unit, n is the total number of moments in the time period, and BP is the neural network;
[0025] The loss function of the adaptive parameter configuration model is:
[0026] ;
[0027] Where Ls(T) is the loss function in the Tth time period, For the The degree of change in the driving environment monitored by an AI functional domain unit in the Tth time period, The output of the model is The TSN configuration parameters of an AI functional domain unit in the Tth time period, For the The true value of the TSN configuration parameter of an AI functional domain unit in the Tth time period, is the total number of AI functional domain units.
[0028] As a preferred solution of the present invention, the present invention provides an in-vehicle AI information processing method based on Ethernet TSN, which is applied to the in-vehicle AI information processing system based on Ethernet TSN. The in-vehicle AI information processing method includes the following steps:
[0029] The TSN configuration server uses a neural network to establish a configuration measurement network based on the AI information generated on the AI functional domain unit to determine the TSN configuration parameters on the AI functional domain unit;
[0030] The TSN configuration server uses a similarity function to analyze the changes in the driving environment for the AI information generated on the AI functional domain unit, and obtains the analysis results of the changes in the driving environment;
[0031] The change analysis result is used to modify the configuration measurement network, thereby obtaining an adaptive parameter configuration model for adaptively configuring the TSN network according to the driving environment.
[0032] As a preferred solution of the present invention, the method for constructing the configuration and calculation network includes:
[0033] Record the AI information generated on each AI functional domain unit at multiple times;
[0034] According to the AI information generated on each AI functional domain unit, the TSN configuration parameters on the AI functional domain unit are manually matched;
[0035] A neural network is used to map and train the AI information generated on the AI functional domain unit with the TSN configuration parameters on the AI functional domain unit to obtain a configuration measurement network;
[0036] The configuration and calculation network is:
[0037] ;
[0038] In the formula, Configure TSN parameters for model output, is AI information, BP is neural network.
[0039] As a preferred solution of the present invention, the method of analyzing the change of the driving environment includes transient change analysis and steady-state change analysis, wherein:
[0040] Transient change analysis methods include:
[0041] ;
[0042] In the formula, For the The degree of change in the driving environment monitored by the AI functional domain unit at the tth moment, For the AI functional domain unit AI information of a smart terminal at the tth moment, For the AI functional domain unit AI information of a smart terminal at the t-1th moment, for and The similarity function between them, m is the number of intelligent terminals in the AI functional domain unit;
[0043] Steady-state change analysis methods include:
[0044] ;
[0045] In the formula, For the The degree of change in the driving environment monitored by the AI functional domain unit at the Tth time period, For the AI functional domain unit The first intelligent terminal in the Tth time period AI information at a certain moment, For the AI functional domain unit The first intelligent terminal in the T-1th time period AI information at all times, is the first a moment, is the first time in the T-1th time period a moment, for and The similarity function between them is shown in Figure 2, where m is the number of intelligent terminals in the AI functional domain unit and n is the total number of moments in the time period.
[0046] As a preferred solution of the present invention, the method for constructing the adaptive parameter configuration model includes:
[0047] When the change analysis result is :
[0048] According to the change analysis result, the loss function of the configuration measurement network is weightedly modified to obtain the loss function of the adaptive parameter configuration model;
[0049] Using the loss function of the adaptive parameter configuration model, training the configuration measurement network to obtain the adaptive parameter configuration model;
[0050] The adaptive parameter configuration model is:
[0051] ;
[0052] In the formula, The output of the model is The TSN configuration parameters of the AI functional domain unit at the tth time, For the The AI information of the AI functional domain unit at the tth moment, , For the AI functional domain unit The AI information of a smart terminal at the tth moment, m is the number of smart terminals in the AI functional domain unit, and BP is the neural network;
[0053] The loss function of the adaptive parameter configuration model is:
[0054] ;
[0055] Where Ls(t) is the loss function at the tth moment, For the The degree of change in the driving environment monitored by the AI functional domain unit at the tth moment, The output of the model is The TSN configuration parameters of the AI functional domain unit at the tth time, For the The true value of the TSN configuration parameter of an AI functional domain unit at the tth moment, is the total number of AI functional domain units;
[0056] When the change analysis result is :
[0057] According to the change analysis result, the loss function of the configuration measurement network is weightedly modified to obtain the loss function of the adaptive parameter configuration model;
[0058] Using the loss function of the adaptive parameter configuration model, training the configuration measurement network to obtain the adaptive parameter configuration model;
[0059] The adaptive parameter configuration model is:
[0060] ;
[0061] In the formula, The output of the model is The TSN configuration parameters of an AI functional domain unit in the Tth time period, For the The AI information of the AI functional domain unit in the Tth time period, , For the AI functional domain unit The first intelligent terminal in the Tth time period AI information at a certain moment, is the first moments, m is the number of intelligent terminals in the AI functional domain unit, n is the total number of moments in the time period, and BP is the neural network;
[0062] The loss function of the adaptive parameter configuration model is:
[0063] ;
[0064] Where Ls(T) is the loss function in the Tth time period, For the The degree of change in the driving environment monitored by an AI functional domain unit in the Tth time period, The output of the model is The TSN configuration parameters of an AI functional domain unit in the Tth time period, For the The true value of the TSN configuration parameter of an AI functional domain unit in the Tth time period, is the total number of AI functional domain units.
[0065] As a preferred solution of the present invention, the similarity function includes Euclidean distance, Jaccard similarity coefficient, and cosine similarity.
[0066] Compared with the prior art, the present invention has the following beneficial effects:
[0067] The present invention establishes an adaptive parameter configuration model in the TSN configuration server, which is used to configure the TSN parameters between the on-board computer ECU and each AI functional domain unit, so as to realize adaptive regulation according to the driving environment, improve the flexibility of the TSN network and enhance the driving experience. BRIEF DESCRIPTION OF THE DRAWINGS
[0068] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the implementation methods or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.
[0069] Figure 1 A block diagram of an in-vehicle AI information processing system provided by an embodiment of the present invention;
[0070] Figure 2 Flowchart of the vehicle-mounted AI information processing method provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0071] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0072] like Figure 1 As shown, the present invention provides an in-vehicle AI information processing system based on Ethernet TSN, including a TSN network, an in-vehicle computer ECU, at least one AI functional domain unit, and a TSN configuration server, wherein:
[0073] The TSN network is deployed between the vehicle computer ECU and each AI functional domain unit to realize the information interconnection between the vehicle computer ECU and the AI functional domain units;
[0074] The TSN configuration server is used to adaptively adjust the TSN configuration parameters between the on-board computer ECU and each AI functional domain unit according to the driving environment, thereby improving the flexibility of the TSN network and enhancing the driving experience.
[0075] The present invention arranges the TSN network in the vehicle, transforming the vehicle network into a domain-based structured architecture, interconnecting the main in-vehicle subsystems with the on-board computer, thereby realizing dynamic sharing of the entire vehicle resource pool and achieving a high-performance, low-cost and wire-harness-reduced architecture model.
[0076] The AI functional domain unit in the present invention is the in-vehicle subsystem, such as intelligent automatic driver assistance / braking (ADAS), in-vehicle entertainment, body control, power transmission and other functional domains. Through the TSN network layout, data information can be transmitted efficiently and at low cost.
[0077] In addition, functional domains such as automatic driver assistance / braking (ADAS), in-vehicle entertainment, body control, power transmission, etc. will also include their own intelligent terminals, such as GPS positioning terminals, Beidou positioning terminals, trajectory recording terminals, vehicle diagnostic terminals, TCU terminals, DMS driver behavior monitoring terminals, ADAS advanced driving assistance terminals, BSD blind spot monitoring terminals, AVM 360° surround view terminals, video acquisition terminals, audio acquisition terminals, face recognition terminals, and mobile network data transceiver terminals.
[0078] TSN configuration parameters include: transmission path and transmission protocol; among them, transmission path, such as AI functional domain unit 1 (driver assistance / braking functional domain) or AI functional domain unit 2 (body control functional domain)-on-board computer ECU-AI functional domain unit 3 (powertrain functional domain); transmission protocol, such as TSN802.1Qbv, TSN802.1Qcc and 802.1Qdj protocols.
[0079] The AI functional domain unit includes at least one intelligent terminal, and each intelligent terminal is connected by a CAN bus to achieve information interconnection between the intelligent terminals.
[0080] The TSN configuration server and each AI functional domain unit are connected using the CAN bus to achieve information interconnection between the TSN configuration server and each AI functional domain unit.
[0081] In order to achieve adaptive control according to the driving environment and improve the flexibility of the TSN network to enhance the driving experience, the present invention sets up a TSN configuration server and builds an adaptive parameter configuration model in the TSN configuration server. The model can adjust the configuration parameters of the TSN network at each AI functional domain unit in real time according to changes in the driving environment to adapt to the current driving environment, that is, to adjust the data transmission mode of each AI functional domain according to the current driving environment, so as to achieve low-cost and high-performance transmission in various driving environments.
[0082] The present invention can achieve low-cost and high-performance transmission in various driving environments, can improve data analysis and control operations that rely on transmitted data, and enhance the driving experience. For example, when the road environment changes from dry to wet, the TSN configuration is also adjusted from a dry environment to adapt to a wet environment. The data transmission performance in a wet environment is improved, which can enable the ADAS functional domain to realize data analysis more quickly, take braking adjustments to adapt to the wet environment, enhance environmental adaptability, and improve the driving experience.
[0083] The adaptive parameter configuration model in the present invention is composed of a configuration parameter calculation network combination of each AI functional domain unit, which can realize real-time calculation of the configuration parameters of each AI functional domain unit, ensuring all-round configuration calculation on the vehicle-mounted TSN network, and making the calculation more comprehensive.
[0084] Furthermore, the present invention modifies the configuration parameter calculation network of each AI functional domain unit according to the changes in the driving environment to ensure that it is still applicable under the changes in the driving environment and enhance the robustness of the model. Therefore, the environmental changes are associated with the loss function of the model to achieve the update of the loss function when the environment changes, so that the configuration parameter calculation network of each AI functional domain unit needs to be updated along with the update of the loss function, so as to achieve the configuration parameter calculation network that adapts to the changed environment, and then regain the TSN configuration parameters that adapt to the changed environment, achieving the effect of reconfiguring the TSN network as the driving environment changes.
[0085] Moreover, the present invention characterizes the environmental changes in the loss function by their degree of change, and refines the degree of environmental change to each AI functional domain unit, that is, the degree of environmental change of each AI functional domain unit is perceived. The degree of environmental change of each AI functional domain unit essentially refers to the degree to which each AI functional domain unit is affected by the environmental changes, thereby realizing the priority of adaptive planning configuration of each AI functional domain unit.
[0086] The higher the degree of impact on the AI functional domain unit, the more the AI functional domain unit configuration parameter calculation network needs to be retrained and calculated to adapt to the current environment, and its configuration priority is higher. Similarly, the smaller the degree of impact on the AI functional domain unit, the less the AI functional domain unit configuration parameter calculation network needs to be retrained and calculated to adapt to the current environment, and its configuration priority is lower. Therefore, the configuration priority can be planned according to the degree of impact when the environment changes, effectively ensuring the balance of configuration resources.
[0087] The present invention is divided into two modes when configuring parameters in the TSN network, the first is transient configuration, and the second is steady-state configuration. Among them, the first transient configuration refers to real-time adjustment according to the real-time changes of the driving environment, and the configuration adjustment at a single moment has strong timeliness and high sensitivity in environmental adaptation, but has high requirements on hardware and computing power. The second steady-state configuration refers to configuring TSN parameters at regular intervals, and the configuration adjustment within the time period. During the adjustment process, the data transmission volume is small, the system stability is high, and the requirements on hardware and computing power are not high, but the timeliness and sensitivity are insufficient, and it can be selected in actual use.
[0088] The TSN configuration server is loaded with an adaptive parameter configuration model for adaptively configuring the TSN network according to the driving environment, where:
[0089] When the TSN network is transiently configured according to the driving environment,
[0090] The adaptive parameter configuration model is:
[0091] ;
[0092] In the formula, The output of the model is The TSN configuration parameters of the AI functional domain unit at the tth time, For the The AI information of the AI functional domain unit at the tth moment, , For the AI functional domain unit The AI information of a smart terminal at the tth moment, m is the number of smart terminals in the AI functional domain unit, and BP is the neural network;
[0093] The loss function of the adaptive parameter configuration model is:
[0094] ;
[0095] Where Ls(t) is the loss function at the tth moment, For the The degree of change in the driving environment monitored by the AI functional domain unit at the tth moment, The output of the model is The TSN configuration parameters of the AI functional domain unit at the tth time, For the The true value of the TSN configuration parameter of an AI functional domain unit at the tth moment, is the total number of AI functional domain units;
[0096] When the TSN network is configured in a steady state according to the driving environment,
[0097] The adaptive parameter configuration model is:
[0098] ;
[0099] In the formula, The output of the model is The TSN configuration parameters of an AI functional domain unit in the Tth time period, For the The AI information of the AI functional domain unit in the Tth time period, , For the AI functional domain unit The first intelligent terminal in the Tth time period AI information at a certain moment, is the first moments, m is the number of intelligent terminals in the AI functional domain unit, n is the total number of moments in the time period, and BP is the neural network;
[0100] The loss function of the adaptive parameter configuration model is:
[0101] ;
[0102] Where Ls(T) is the loss function in the Tth time period, For the The degree of change in the driving environment monitored by an AI functional domain unit in the Tth time period, The output of the model is The TSN configuration parameters of an AI functional domain unit in the Tth time period, For the The true value of the TSN configuration parameter of an AI functional domain unit in the Tth time period, is the total number of AI functional domain units.
[0103] like Figure 2 As shown, the present invention provides an in-vehicle AI information processing method based on Ethernet TSN, which is applied to an in-vehicle AI information processing system based on Ethernet TSN. The in-vehicle AI information processing method includes the following steps:
[0104] The TSN configuration server uses a neural network to establish a configuration measurement network based on the AI information generated on the AI functional domain unit to determine the TSN configuration parameters on the AI functional domain unit;
[0105] The TSN configuration server uses a similarity function to analyze the changes in the driving environment for the AI information generated on the AI functional domain unit, and obtains the analysis results of the changes in the driving environment;
[0106] By using the change analysis results, the configuration measurement network is modified to obtain an adaptive parameter configuration model for adaptively configuring the TSN network according to the driving environment.
[0107] The construction method of configuring the measurement network includes:
[0108] Record the AI information generated on each AI functional domain unit at multiple times;
[0109] According to the AI information generated on each AI functional domain unit, the TSN configuration parameters on the AI functional domain unit are manually matched;
[0110] A neural network is used to map and train the AI information generated on the AI functional domain unit with the TSN configuration parameters on the AI functional domain unit to obtain a configuration measurement network;
[0111] Configure the measurement network as follows:
[0112] ;
[0113] In the formula, Configure TSN parameters for model output, is AI information, BP is neural network.
[0114] The present invention is divided into two modes when configuring parameters in the TSN network, the first is transient configuration, and the second is steady-state configuration. The first transient configuration refers to real-time adjustment according to the real-time changes in the driving environment, and the configuration adjustment at a single moment has strong timeliness and high sensitivity in the adjustment environment adaptation, but has high requirements on hardware and computing power. The second steady-state configuration refers to configuring TSN parameters at regular intervals, and the configuration adjustment within the time period. During the adjustment process, the data transmission volume is small, the system stability is high, and the requirements on hardware and computing power are not high, but the timeliness and sensitivity are insufficient. It can be selected in actual use, as follows:
[0115] The methods of driving environment change analysis include transient change analysis and steady-state change analysis.
[0116] Transient change analysis methods include:
[0117] ;
[0118] In the formula, For the The degree of change in the driving environment monitored by the AI functional domain unit at the tth moment, For the AI functional domain unit AI information of a smart terminal at the tth moment, For the AI functional domain unit AI information of a smart terminal at the t-1th moment, for and The similarity function between them, m is the number of intelligent terminals in the AI functional domain unit;
[0119] and The greater the gap, the The bigger, the The greater the impact of environmental changes on an AI functional domain unit at the tth moment, the more inappropriate the current TSN parameter configuration is, the more it needs to be adjusted, and the higher the configuration priority.
[0120] Steady-state change analysis methods include:
[0121] ;
[0122] In the formula, For the The degree of change in the driving environment monitored by the AI functional domain unit at the Tth time period, For the AI functional domain unit The first intelligent terminal in the Tth time period AI information at a certain moment, For the AI functional domain unit The first intelligent terminal in the T-1th time period AI information at all times, is the first a moment, is the first time in the T-1th time period a moment, for and The similarity function between them is shown in Figure 2, where m is the number of intelligent terminals in the AI functional domain unit and n is the total number of moments in the time period.
[0123] and The greater the gap, the The bigger, the The greater the impact of environmental changes on an AI functional domain unit in the Tth time period, the more inappropriate the current TSN parameter configuration is, the more it needs to be adjusted, and the higher the configuration priority.
[0124] The adaptive parameter configuration model in the present invention is composed of the configuration parameter calculation network of each AI functional domain unit, which can realize the real-time calculation of the configuration parameters of each AI functional domain unit, ensure the comprehensive configuration calculation on the vehicle-mounted TSN network, and make the calculation more comprehensive, as follows:
[0125] The method for constructing the adaptive parameter configuration model includes:
[0126] When the change analysis result is :
[0127] Through the change analysis results, the loss function of the configuration measurement network is weighted and modified to obtain the loss function of the adaptive parameter configuration model;
[0128] Using the loss function of the adaptive parameter configuration model, the configuration measurement network is trained to obtain the adaptive parameter configuration model;
[0129] The adaptive parameter configuration model is:
[0130] ;
[0131] In the formula, The output of the model is The TSN configuration parameters of the AI functional domain unit at the tth time, For the The AI information of the AI functional domain unit at the tth moment, , For the AI functional domain unit The AI information of a smart terminal at the tth moment, m is the number of smart terminals in the AI functional domain unit, and BP is the neural network;
[0132] The loss function of the adaptive parameter configuration model is:
[0133] ;
[0134] Where Ls(t) is the loss function at the tth moment, For the The degree of change in the driving environment monitored by the AI functional domain unit at the tth moment, The output of the model is The TSN configuration parameters of the AI functional domain unit at the tth time, For the The true value of the TSN configuration parameter of an AI functional domain unit at the tth moment, is the total number of AI functional domain units;
[0135] When the change analysis result is :
[0136] Through the change analysis results, the loss function of the configuration measurement network is weighted and modified to obtain the loss function of the adaptive parameter configuration model;
[0137] Using the loss function of the adaptive parameter configuration model, the configuration measurement network is trained to obtain the adaptive parameter configuration model;
[0138] The adaptive parameter configuration model is:
[0139] ;
[0140] In the formula, The output of the model is The TSN configuration parameters of an AI functional domain unit in the Tth time period, For the The AI information of the AI functional domain unit in the Tth time period, , For the AI functional domain unit The first intelligent terminal in the Tth time period AI information at a certain moment, is the first moments, m is the number of intelligent terminals in the AI functional domain unit, n is the total number of moments in the time period, and BP is the neural network;
[0141] The loss function of the adaptive parameter configuration model is:
[0142] ;
[0143] Where Ls(T) is the loss function in the Tth time period, For the The degree of change in the driving environment monitored by an AI functional domain unit in the Tth time period, The output of the model is The TSN configuration parameters of an AI functional domain unit in the Tth time period, For the The true value of the TSN configuration parameter of an AI functional domain unit in the Tth time period, is the total number of AI functional domain units.
[0144] The present invention modifies the configuration parameter calculation network of each AI functional domain unit according to the changes in the driving environment to ensure that it is still applicable under the changes in the driving environment and enhance the robustness of the model. Therefore, the environmental changes are associated with the loss function of the model to achieve the update of the loss function when the environment changes, so that the configuration parameter calculation network of each AI functional domain unit needs to be updated along with the update of the loss function, so as to achieve the configuration parameter calculation network that adapts to the changed environment, and then regain the TSN configuration parameters that adapt to the changed environment, achieving the effect of reconfiguring the TSN network as the driving environment changes.
[0145] Moreover, the present invention characterizes the environmental changes in the loss function by their degree of change, and refines the degree of environmental change to each AI functional domain unit, that is, the degree of environmental change of each AI functional domain unit is perceived. The degree of environmental change of each AI functional domain unit essentially refers to the degree to which each AI functional domain unit is affected by the environmental changes, thereby realizing the priority of adaptive planning configuration of each AI functional domain unit.
[0146] The higher the degree of impact on the AI functional domain unit, the more the AI functional domain unit configuration parameter calculation network needs to be retrained and calculated to adapt to the current environment, and its configuration priority is higher. Similarly, the smaller the degree of impact on the AI functional domain unit, the less the AI functional domain unit configuration parameter calculation network needs to be retrained and calculated to adapt to the current environment, and its configuration priority is lower. Therefore, the configuration priority can be planned according to the degree of impact when the environment changes, effectively ensuring the balance of configuration resources.
[0147] Similarity functions include Euclidean distance, Jaccard similarity coefficient, and cosine similarity.
[0148] The present invention establishes an adaptive parameter configuration model in the TSN configuration server, which is used to configure the TSN parameters between the on-board computer ECU and each AI functional domain unit, so as to realize adaptive regulation according to the driving environment, improve the flexibility of the TSN network and enhance the driving experience.
[0149] The above embodiments are only exemplary embodiments of the present application and are not intended to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present application within the essence and protection scope of the present application, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present application.
Claims
1. An in-vehicle AI information processing system based on Ethernet TSN, It is characterized in that It includes a TSN network, an onboard computer ECU, at least one AI functional domain unit, and a TSN configuration server, wherein: The TSN network is arranged between the vehicle computer ECU and each AI functional domain unit to realize information interconnection between the vehicle computer ECU and the AI functional domain units; The TSN configuration server is used to adaptively control the TSN configuration parameters between the vehicle computer ECU and each AI functional domain unit according to the driving environment; The TSN configuration server is loaded with an adaptive parameter configuration model for adaptively configuring the TSN network according to the driving environment, where: When the TSN network is transiently configured according to the driving environment, The adaptive parameter configuration model is: ; In the formula, The output of the model is The TSN configuration parameters of the AI functional domain unit at the tth time, For the The AI information of the AI functional domain unit at the tth moment, , For the AI functional domain unit The AI information of a smart terminal at the tth moment, m is the number of smart terminals in the AI functional domain unit, and BP is the neural network; The loss function of the adaptive parameter configuration model is: ; Where Ls(t) is the loss function at the tth moment, For the The degree of change in the driving environment monitored by the AI functional domain unit at the tth moment, The output of the model is The TSN configuration parameters of the AI functional domain unit at the tth time, For the The true value of the TSN configuration parameter of an AI functional domain unit at the tth moment, is the total number of AI functional domain units; When the TSN network is configured in a steady state according to the driving environment, The adaptive parameter configuration model is: ; In the formula, The output of the model is The TSN configuration parameters of an AI functional domain unit in the Tth time period, For the The AI information of the AI functional domain unit in the Tth time period, , For the AI functional domain unit The first intelligent terminal in the Tth time period AI information at a certain moment, is the first moments, m is the number of intelligent terminals in the AI functional domain unit, n is the total number of moments in the time period, and BP is the neural network; The loss function of the adaptive parameter configuration model is: ; Where Ls(T) is the loss function in the Tth time period, For the The degree of change in the driving environment monitored by an AI functional domain unit in the Tth time period, The output of the model is The TSN configuration parameters of an AI functional domain unit in the Tth time period, For the The true value of the TSN configuration parameter of an AI functional domain unit in the Tth time period, is the total number of AI functional domain units.
2. According to claim 1, an in-vehicle AI information processing system based on Ethernet TSN, Features: The TSN configuration parameters include: transmission path and transmission protocol.
3. According to the vehicle-mounted AI information processing system based on Ethernet TSN as claimed in claim 1, Features: The AI functional domain unit includes at least one intelligent terminal, and each intelligent terminal is connected by a CAN bus to achieve information interconnection between the intelligent terminals.
4. According to claim 1, an in-vehicle AI information processing system based on Ethernet TSN, Features: The TSN configuration server and each AI functional domain unit are connected using a CAN bus to achieve information interconnection between the TSN configuration server and each AI functional domain unit.
5. A vehicle-mounted AI information processing method based on Ethernet TSN, It is characterized in that The vehicle-mounted AI information processing system based on Ethernet TSN as described in any one of claims 1 to 4, the vehicle-mounted AI information processing method comprises the following steps: The TSN configuration server uses a neural network to establish a configuration measurement network based on the AI information generated on the AI functional domain unit to determine the TSN configuration parameters on the AI functional domain unit; The TSN configuration server uses a similarity function to analyze the changes in the driving environment for the AI information generated on the AI functional domain unit, and obtains the analysis results of the changes in the driving environment; The change analysis result is used to modify the configuration measurement network, thereby obtaining an adaptive parameter configuration model for adaptively configuring the TSN network according to the driving environment.
6. According to claim 5, a vehicle-mounted AI information processing method based on Ethernet TSN, Features: The method for constructing the configuration and calculation network includes: Record the AI information generated on each AI functional domain unit at multiple times; According to the AI information generated on each AI functional domain unit, the TSN configuration parameters on the AI functional domain unit are manually matched; A neural network is used to map and train the AI information generated on the AI functional domain unit with the TSN configuration parameters on the AI functional domain unit to obtain a configuration measurement network; The configuration and calculation network is: ; In the formula, Configure TSN parameters for model output, is AI information, BP is neural network.
7. According to claim 6, a vehicle-mounted AI information processing method based on Ethernet TSN, Features: The methods of analyzing the changes in the driving environment include transient change analysis and steady-state change analysis. Transient change analysis methods include: ; In the formula, For the The degree of change in the driving environment monitored by the AI functional domain unit at the tth moment, For the AI functional domain unit AI information of a smart terminal at the tth moment, For the AI functional domain unit AI information of a smart terminal at the t-1th moment, for and The similarity function between them, m is the number of intelligent terminals in the AI functional domain unit; Steady-state change analysis methods include: ; In the formula, For the The degree of change in the driving environment monitored by the AI functional domain unit at the Tth time period, For the AI functional domain unit The first intelligent terminal in the Tth time period AI information at a certain moment, For the AI functional domain unit The first intelligent terminal in the T-1th time period AI information at all times, is the first a moment, is the first time in the T-1th time period a moment, for and The similarity function between them is shown in Figure 2, where m is the number of intelligent terminals in the AI functional domain unit and n is the total number of moments in the time period.
8. According to the vehicle-mounted AI information processing method based on Ethernet TSN as claimed in claim 7, Features: The method for constructing the adaptive parameter configuration model includes: When the change analysis result is : According to the change analysis result, the loss function of the configuration measurement network is weightedly modified to obtain the loss function of the adaptive parameter configuration model; Using the loss function of the adaptive parameter configuration model, training the configuration measurement network to obtain the adaptive parameter configuration model; The adaptive parameter configuration model is: ; In the formula, The output of the model is The TSN configuration parameters of the AI functional domain unit at the tth time, For the The AI information of the AI functional domain unit at the tth moment, , For the AI functional domain unit The AI information of a smart terminal at the tth moment, m is the number of smart terminals in the AI functional domain unit, and BP is the neural network; The loss function of the adaptive parameter configuration model is: ; In the formula, is the loss function at the tth moment, For the The degree of change in the driving environment monitored by the AI functional domain unit at the tth moment, The output of the model is The TSN configuration parameters of the AI functional domain unit at the tth time, For the The true value of the TSN configuration parameter of an AI functional domain unit at the tth moment, is the total number of AI functional domain units; When the change analysis result is : According to the change analysis result, the loss function of the configuration measurement network is weightedly modified to obtain the loss function of the adaptive parameter configuration model; Using the loss function of the adaptive parameter configuration model, training the configuration measurement network to obtain the adaptive parameter configuration model; The adaptive parameter configuration model is: ; In the formula, The output of the model is The TSN configuration parameters of an AI functional domain unit in the Tth time period, For the The AI information of the AI functional domain unit in the Tth time period, , For the AI functional domain unit The first intelligent terminal in the Tth time period AI information at a certain moment, is the first moments, m is the number of intelligent terminals in the AI functional domain unit, n is the total number of moments in the time period, and BP is the neural network; The loss function of the adaptive parameter configuration model is: ; Where Ls(T) is the loss function in the Tth time period, For the The degree of change in the driving environment monitored by an AI functional domain unit in the Tth time period, The output of the model is The TSN configuration parameters of an AI functional domain unit in the Tth time period, For the The true value of the TSN configuration parameter of an AI functional domain unit in the Tth time period, is the total number of AI functional domain units.
9. According to claim 8, a vehicle-mounted AI information processing method based on Ethernet TSN, It is characterized in that The similarity functions include Euclidean distance, Jaccard similarity coefficient, and cosine similarity.
