A functional navigation method, vehicle diagnostic device, and storage medium

Abstract

The application relates to the technical field of vehicle diagnosis, and discloses a function navigation method, a vehicle diagnosis device and a storage medium. The function navigation method comprises the following steps: constructing a training data set of a function navigation model of a vehicle terminal; performing feature extraction based on the training data set, and determining feature information of a candidate function instruction; constructing a function navigation model based on the training data set and the feature information; and matching a preset query instruction input through the function navigation model, and generating a recommendation list of recommended function instructions. The application improves the function navigation experience on the vehicle diagnosis device, improves the accuracy of function navigation, and further improves the work efficiency of users.

Description

Technical Field

[0001] This application relates to the field of vehicle diagnostic technology, and in particular to a functional navigation method, vehicle diagnostic equipment, and storage medium. Background Technology

[0002] Vehicle diagnostic equipment covers a wide range of functions. However, once connected to related devices or systems, users often need to navigate through multiple pages and go through a cumbersome process to access specific diagnostic functions. This not only wastes time but also increases the risk of errors due to the complexity of the operation, potentially disrupting the entire diagnostic process or related business operations.

[0003] The existing vehicle diagnostic equipment's function navigation method is to identify the corresponding function through human description. However, human description may be inaccurate, resulting in the inability to accurately match the corresponding function. Moreover, different vehicle models may have different descriptions or names for the same diagnostic function, which may also lead to the inability to accurately identify the corresponding function. Summary of the Invention

[0004] In view of this, embodiments of this application provide a functional navigation method, a vehicle diagnostic device, and a storage medium, which can effectively solve the problem of inaccurate identification of corresponding functions.

[0005] In a first aspect, embodiments of this application provide a function navigation method, including:

[0006] Construct a training dataset for the functional navigation model of the vehicle terminal;

[0007] Based on the training dataset, feature extraction is performed to determine the feature information of candidate function instructions;

[0008] The functional navigation model is constructed based on the training dataset and the feature information;

[0009] The function navigation model matches the input preset query commands to generate a recommended list of recommended function commands.

[0010] In a first possible embodiment of the first aspect, the training dataset includes a vehicle model feature fingerprint database and a diagnostic instruction database, and the training dataset for constructing the functional navigation model of the vehicle terminal includes:

[0011] The extracted ECU feature parameters are used to generate vehicle model feature vectors to establish the vehicle model feature fingerprint database.

[0012] The system collects original diagnostic commands from multiple vehicle models, determines the call frequency of each candidate function command based on the call frequency of the original diagnostic commands, and determines the usage popularity index of each candidate function command to establish the diagnostic command library.

[0013] In a second possible embodiment of the first aspect, the training dataset further includes a semantic association graph, the candidate functional instructions include multiple identifier diagnostic instructions, each of the identifier diagnostic instructions includes functional parameters, and the training dataset for constructing the functional navigation model of the vehicle terminal further includes:

[0014] The name, function parameter list, and operation scenario of each candidate function instruction are encoded to obtain the semantic vector of each candidate function instruction;

[0015] A semantic association graph of the candidate functional instructions is constructed based on the semantic vectors of each candidate functional instruction.

[0016] In a third possible embodiment of the first aspect, constructing a semantic association graph of the candidate functional instructions based on the semantic vectors of each of the candidate functional instructions includes:

[0017] Each of the candidate function instructions is used as a node in the semantic association graph;

[0018] The semantic vector and call frequency of each candidate function instruction, as well as the vehicle model feature vector to which the candidate function instruction belongs, are used as node attributes of the semantic association graph.

[0019] The edge weights of the semantic association graph are determined based on the semantic similarity and co-occurrence frequency among the candidate functional instructions.

[0020] In a fourth possible embodiment of the first aspect, constructing the functional navigation model based on the training dataset and the feature information includes:

[0021] Construct an initial classifier and train it based on the training dataset and the feature information;

[0022] The training of the initial classifier is stopped under preset conditions by an early stopping mechanism to obtain the optimal functional navigation model parameters and thus the initial functional navigation model.

[0023] The initial functional navigation model is optimized online through an incremental learning pipeline to obtain the final functional navigation model.

[0024] In a fifth possible embodiment of the first aspect, the step of matching the input preset query instruction with the functional navigation model to generate a recommended list of recommended functional instructions includes:

[0025] The preset query command is semantically parsed to obtain the semantic feature vector of the preset query command;

[0026] The initial recommended function instruction is determined by searching for multiple nearest neighbors of the semantic feature vector of the preset query instruction in the feature space of the candidate function instruction through nearest neighbor retrieval.

[0027] After filtering the initial recommended function commands by vehicle type and sorting them by popularity, the final recommended function commands are determined.

[0028] In a sixth possible embodiment of the first aspect, the feature information includes the vehicle model matching degree of the candidate function instructions, and the step of filtering and ranking the initial recommended function instructions by vehicle model includes:

[0029] Based on the vehicle matching degree, a recommended function command that is compatible with the target vehicle model is selected;

[0030] Based on the popularity index of the filtered recommended function commands, the recommended function commands are sorted in descending order to obtain the recommended list.

[0031] In a seventh possible embodiment of the first aspect, the function navigation method further includes:

[0032] A reward signal is determined based on the user's selection, and the parameters of the functional navigation model are updated based on the reward signal.

[0033] Secondly, embodiments of this application provide a vehicle diagnostic device, which includes a processor and a memory. The memory stores a computer program, and the processor executes the computer program to implement the above-described functional navigation method.

[0034] Thirdly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed on a processor, implements the above-described function navigation method.

[0035] The embodiments of this application have the following beneficial effects:

[0036] This embodiment of a function navigation method includes: constructing a training dataset for a function navigation model of an in-vehicle terminal; extracting features based on the training dataset to determine feature information of candidate function commands; constructing the function navigation model based on the training dataset and the feature information; and matching input preset query commands through the function navigation model to generate a recommended list of recommended function commands. Based on the above solution, this application achieves a more intelligent and flexible user interaction experience. It eliminates the need for precise descriptions of the complete function names for the vehicle model; users only need to input commonly used preset query commands to match the corresponding specific function commands for the vehicle model, lowering the barrier to entry and enhancing the user experience. Attached Figure Description

[0037] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This paper illustrates a first flowchart of a functional navigation method according to an embodiment of the present application.

[0039] Figure 2 A second flowchart of the functional navigation method according to an embodiment of this application is shown;

[0040] Figure 3 A third flowchart of the functional navigation method according to an embodiment of this application is shown;

[0041] Figure 4 A schematic diagram of the fourth type of functional navigation method according to an embodiment of this application is shown. Detailed Implementation

[0042] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0043] The components of the embodiments of this application described and illustrated in the accompanying drawings can be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of this application provided in the drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0044] In the following text, the terms "comprising," "having," and their cognates, which may be used in various embodiments of this application, are intended only to indicate a particular feature, number, step, operation, element, component, or combination thereof, and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, elements, components, or combinations thereof, or adding the possibility of one or more combinations thereof. Furthermore, the terms "first," "second," "third," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance.

[0045] Unless otherwise specified, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which the various embodiments of this application pertain. Terms (such as those defined in commonly used dictionaries) shall be interpreted as having the same meaning as in their contextual meaning in the relevant technical field and shall not be construed as having an idealized or overly formal meaning, unless clearly defined in the various embodiments of this application.

[0046] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0047] Vehicle diagnostic equipment functions include: reading and clearing vehicle fault codes; monitoring operating status and data changes in real time; recording operating data to analyze fault causes or optimize performance; testing various vehicle sensors to determine their proper functioning; simulating various driving conditions to check the vehicle's actuators and control systems; and providing corresponding repair suggestions and guidance based on fault codes and test results. Vehicle diagnostic equipment also includes other functions such as system upgrades and viewing vehicle information.

[0048] In automotive repair and diagnostics, a significant efficiency bottleneck commonly encountered when using diagnostic equipment is the need to navigate through multiple menus from connecting to the device to the target function page. This can lead to problems such as novices needing to memorize complex paths and easily getting lost in the menus, while even experienced technicians may waste valuable time due to the excessive number of steps (e.g., during emergency troubleshooting), and needing to re-run the process after accidental clicks, thus reducing overall efficiency.

[0049] While large AI models can simplify the operation of automotive diagnostic equipment through natural language interaction, issues such as ambiguous user expressions and terminology differences between vehicle models still exist in practical applications. Therefore, to address these shortcomings of existing technologies, this application proposes a function navigation method, vehicle diagnostic equipment, and storage medium. This method constructs a function navigation model based on the feature information of a training dataset and candidate function commands. The function navigation model matches input preset query commands to obtain corresponding recommended function commands. Users can select one or more of these recommended function commands to detect the target vehicle. This application improves the function navigation experience on vehicle diagnostic equipment, increases the accuracy of responses, and further enhances user efficiency.

[0050] The following describes the functional navigation method using specific examples.

[0051] Figure 1 A flowchart of a function navigation method according to an embodiment of this application is shown. Exemplarily, the function navigation method includes the following steps:

[0052] S110, the training dataset for constructing the functional navigation model of the vehicle terminal.

[0053] In one embodiment, the training dataset includes a vehicle model feature fingerprint database. This application generates vehicle model feature vectors based on the extracted ECU (Electronic Control Unit) feature parameters to establish the vehicle model feature fingerprint database.

[0054] In this embodiment, the ECU characteristic parameters include powertrain type, ECU version number, bus architecture, and production batch. The vehicle model feature vector can be represented as: V_car = [Powertrain type, ECU version number, bus architecture, production batch]. The powertrain type determines the differences in the ECU's functional modules and diagnostic logic, including gasoline vehicles, hybrid vehicles, pure electric vehicles, and fuel cell vehicles. The ECU version number is a unique identifier for the electronic control unit's software, hardware, and calibration data, used to distinguish ECU designs at different iteration stages. The bus architecture defines the communication method between ECUs, affecting the protocol selection and connection logic of diagnostic instruments, including CAN bus (Controller Area Network), LIN bus (Local Interconnect Network), and Ethernet. The production batch is the production group identifier for the ECU during the manufacturing process, used to trace raw materials, processes, and quality inspection information.

[0055] In one embodiment, this application can access an OEM (Original Equipment Manufacturer) technical document library to obtain ECU characteristic parameters for various vehicle models. The OEM technical document library is a resource repository used to store relevant technical documents from the original equipment manufacturer.

[0056] In another embodiment, the training dataset also includes a diagnostic instruction library. This application collects original diagnostic instructions from multiple vehicle models, determines the call frequency of each candidate function instruction based on the call frequency of the diagnostic instructions, and determines the usage popularity index of each diagnostic function to establish a diagnostic instruction library.

[0057] In one embodiment, this application can acquire raw diagnostic commands from multiple vehicle models based on a dynamic acquisition protocol stack. Each raw diagnostic command includes the characteristic parameters of its respective ECU. The dynamic acquisition protocol stack is a software protocol stack used to realize the dynamic acquisition of vehicle diagnostic data. The dynamic acquisition protocol stack mainly includes [SAE J2534-1 standard instruction set, OEM-specific instruction decoder, and user operation behavior recorder]. The SAE J2534-1 standard instruction set enables standardized communication between the diagnostic equipment and the vehicle ECU, covering the basic diagnostic functions of mainstream vehicle models. The OEM-specific instruction decoder is used to decode the diagnostic instruction set, converting the instructions into operations that the equipment can execute. The user operation behavior recorder is used to record and analyze user operation behavior data.

[0058] Exemplary, the candidate function instructions in this application include multiple identification diagnostic instructions. These instructions are the original diagnostic instructions for standardized identifications used for additional functions. Different function parameters can be added when invoking these instructions. For example, the steering angle zero-learning function could have parameters such as steering wheel angle and steering angle zero point, which are not limited here.

[0059] In another embodiment, this application records the call frequency of each identifier diagnostic instruction through a feature data layer, and records the call frequency of candidate function instructions based on the call frequency of the identifier diagnostic instructions.

[0060] The feature data layer is a structured data storage and analysis layer, primarily used to record, statistically analyze, and interpret the function call characteristics of diagnostic instructions. Its core function is to transform diagnostic instructions into quantifiable feature data to facilitate subsequent analysis and model training. This includes function call frequency statistics, multi-dimensional feature labels, and dynamic behavior modeling.

[0061] Function call frequency statistics refer to the number of times each diagnostic instruction is called across different vehicle models and scenarios. Multidimensional feature labels refer to functions (standardized identifiers for diagnostic instructions), associated parameters (functional parameters attached when a diagnostic instruction is called), temporal characteristics (the temporal relationships such as call time, interval, and sequence of diagnostic instructions), and vehicle model association (associated with the vehicle model feature vector; the ECU feature parameters to which the diagnostic instruction belongs, including powertrain type, ECU version number, bus architecture, and production batch). This application can determine the ECU feature parameters and vehicle model feature vector to which candidate function instructions belong based on the ECU feature parameters applicable to the diagnostic instruction. Dynamic behavior modeling involves discovering typical diagnostic patterns and identifying abnormal calls through frequency statistics of diagnostic instructions.

[0062] For example, the formula for calculating the usage popularity index of candidate function instructions is: Popularity Index = ln(average number of calls per week) × 10 + Urgency Coefficient. The urgency coefficient refers to the urgency or impact of the function, and the range of the urgency coefficient can be [1, 10]. This application sets the urgency coefficient corresponding to each candidate function instruction according to the data items involved in each candidate function instruction, with special functions > action tests > data flow.

[0063] Special functions refer to advanced functions in vehicle diagnostics that require in-depth operation or manufacturer authorization. These functions are used to perform non-standard tasks, such as coding (modifying ECU parameters, such as light coding or vehicle configuration), matching / adaptation (calibrating sensors or actuators), programming (flashing ECU firmware, such as software upgrades or initialization after replacing control units), and secure access (removing manufacturer protections to perform sensitive operations). Motion testing refers to actively controlling vehicle components through functional commands to verify their proper functioning, such as activating actuators like fuel injectors, relays, and solenoid valves. This is primarily used to locate hardware faults and verify the restoration of functionality after repair. Data streams refer to the ECU's operating parameters, such as engine speed, coolant temperature, and vehicle speed, primarily used to assist in diagnostics.

[0064] In one embodiment, the training dataset also includes a semantic association graph. This application encodes the name of each candidate functional instruction, the list of functional parameters of the candidate functional instruction's identifier diagnostic instruction, and the operation scenario to obtain the semantic vector of each candidate functional instruction.

[0065] In one embodiment, this application uses a Bi-LSTM model to perform contextual analysis on the original DTC code (Diagnostic Trouble Code) and execution logs to generate a distributed representation of the function name: E_func = Encoder(function name + function parameter list + operation scenario), where E_func represents the semantic vector of the candidate function instruction, and Encoder specifically refers to the Bi-LSTM model.

[0066] This application uses a Bi-LSTM model to encode the function name, function parameter list, and operation scenario to obtain a semantic vector. For example, the diagnostic function is "ABS pump reset", its function name is "ABS pump reset", the function parameter list may be empty (because this function does not require additional parameters), and the operation scenario may be "to perform a reset operation after a fault occurs in the ABS system".

[0067] Function name encoding: Input the string "ABS pump reset" into the Bi-LSTM model. The Bi-LSTM model will process the string character by character and output a fixed-dimensional vector representation. This vector captures the semantic information of the function "ABS pump reset".

[0068] Function parameter list encoding: If the parameter list is empty, it can be represented directly by a zero vector or a specific placeholder vector. If the parameter list is not empty, each parameter can be encoded as a string, and these encoded vectors can be aggregated (e.g., summed, averaged, etc.) to obtain a unified parameter list vector.

[0069] Operation scenario encoding: Similarly, the string "perform a reset operation after the ABS (Antilock Brake System) system fails" is input into the Bi-LSTM model to obtain a vector representing the operation scenario.

[0070] Merge vectors: Merge the function name vector, function parameter list vector, and operation scenario vector (e.g., by concatenation, weighted summation, etc.) to obtain a final semantic vector E_func.

[0071] For example, after Bi-LSTM encoding, the function name vector for "ABS pump reset" might be [0.1,-0.2,0.3,...], and the operation scenario vector might be [0.2,0.1,-0.3,...]. If the parameter list is empty, it is replaced by a zero vector [0,0,0,...]. Concatenating these vectors yields the final semantic vector E_func = [0.1,-0.2,0.3,0.2,0.1,-0.3,0,0,0,...].

[0072] In another embodiment, this application constructs a semantic association graph of candidate functional instructions based on the semantic vectors of each candidate functional instruction. For example... Figure 2 As shown, constructing a semantic association graph includes the following steps:

[0073] S111, each candidate function instruction is used as a node in the semantic association graph.

[0074] For example, the semantic association graph is a relational network graph built on each candidate function instruction. It uses nodes and edges as basic elements to reveal the logical relationships between candidate function instructions, and maps the complex relationships between candidate function instructions in terms of semantics, vehicle type, and call frequency, thereby improving the accuracy of function navigation.

[0075] S112, the semantic vector and call frequency of each candidate function instruction, as well as the vehicle model feature vector to which the candidate function instruction belongs, are used as node attributes of the semantic association graph.

[0076] S113, determine the edge weights of the semantic association graph based on the semantic similarity and co-occurrence frequency among the candidate functional instructions.

[0077] In one embodiment, the directed edge eij of the semantic association graph represents the association relationship from diagnostic function i to diagnostic function j, and the edge weight wij combines semantic similarity and co-occurrence frequency. The edge weight calculation formula is: w_ij=sim(E_i,E_j)0.8+co-occurrence0.2.

[0078] Co-occurrence represents the frequency of co-occurrence, i.e., the frequency at which candidate functional instruction i and candidate functional instruction j appear together. This application can statistically analyze the co-occurrence frequency of candidate functional instruction i and candidate functional instruction j within the same time period based on the time and frequency of the diagnostic instruction. sim(E_i, E_j) represents the semantic similarity between candidate functional instruction i and candidate functional instruction j, where E_i represents the semantic vector of candidate functional instruction i and E_j represents the semantic vector of candidate functional instruction j. This application calculates the cosine similarity between candidate functional instruction i and candidate functional instruction j based on the semantic vectors of candidate functional instruction i and candidate functional instruction j.

[0079] It is understandable that during the training of a functional navigation model, information from the semantic association graph is incorporated as input into the model training, thereby enhancing the model's capabilities in knowledge representation and reasoning. Knowledge augmentation can help large models better utilize domain knowledge for learning, improving the model's generalization ability.

[0080] In one embodiment, the training dataset also includes historical diagnostic logs and a semantic generalization result set. The historical diagnostic logs are a collection of structured operational data recorded during vehicle diagnostics, used to track the diagnostic process, analyze fault modes, and optimize maintenance strategies. The semantic generalization result set represents the semantic generalization results of the function names of each candidate function instruction. It expands or enhances the function names of candidate function instructions using specific methods, enabling the model to understand broader semantic concepts. For example, synonym substitution can be used to replace the original words with semantically similar words to obtain semantic generalization results.

[0081] In one embodiment, the training dataset includes labeled data. This application labels each candidate function instruction to construct labeled data. Based on the correspondence between the preset query instruction and each candidate function instruction, this application obtains candidate function instructions labeled with the preset query instruction. Subsequently, supervised learning is performed by learning the correspondence between input (preset query instruction) and output (candidate function instruction) through a large amount of labeled data.

[0082] S120 performs feature extraction based on the training dataset to determine the feature information of candidate function instructions.

[0083] For example, the feature information of the candidate function instruction includes the vehicle matching degree between the candidate function instruction and the target vehicle model, the historical success rate of the candidate function instruction, and the semantic similarity between the candidate function instruction and the preset query instruction.

[0084] This application can match the target vehicle model with the ECU characteristic parameters to which the candidate function instruction belongs, and obtain the vehicle model matching degree. For example, if the power type, ECU version number and bus architecture in the ECU characteristic parameters to which the candidate function instruction belongs are the same as the target vehicle model, then the vehicle model matching rate is 75%.

[0085] Historical success rate is the frequency with which candidate function instructions successfully complete their expected operations in historical executions, which can be determined from the number of successes and calls in historical diagnostic logs. The semantic similarity between candidate function instructions and preset query instructions can be calculated by performing cosine similarity calculations on their semantic vectors.

[0086] S130, a functional navigation model is built based on the training dataset and feature information.

[0087] In one embodiment, such as Figure 3 As shown, the functional navigation model constructed in this application includes the following steps:

[0088] S131, Construct the initial classifier, and train the initial classifier based on the training dataset and feature information.

[0089] Exemplarily, this application combines multiple weak classifiers (decision trees) using the XGBoost algorithm (Extreme Gradient Boosting) to obtain an initial classifier. Each decision tree is trained based on the residuals of the previous tree, and the residuals are gradually reduced by iteratively optimizing the loss function. The XGBoost algorithm constructs a strong learner by iteratively training weak classifiers on the training dataset to obtain the initial classifier. In each iteration, the XGBoost algorithm adjusts the model parameters based on the current model's performance on the training dataset and feature information to improve the model's accuracy. Model parameters include, but are not limited to, the learning rate, regularization parameters, tree-related parameters, and data sampling parameters.

[0090] S132, the training of the initial classifier is stopped under preset conditions through an early stopping mechanism to obtain the optimal functional navigation model parameters and thus the initial functional navigation model.

[0091] In one embodiment, the early stopping mechanism of this application uses a dynamic monitoring strategy to randomly select 20% of the training data as a validator. During model training, the training process of going through all samples in the training dataset is called an epoch. Each epoch evaluates the validation set and calculates the accuracy. The basic condition for early stopping is that the accuracy of the validation set increases by <0.2% for 5 consecutive epochs. The enhanced condition is that if the AUC index drops by more than 0.5% at the same time, the model will be terminated immediately. A backtracking mechanism is added to intelligently revert to the historical best model state after the early stopping condition or the enhanced condition is triggered, that is, to save a snapshot of the model parameters when the validation set performs best.

[0092] AUC (Area Under the Curve) is a metric used to evaluate model performance, measuring the area under the ROC (Receiver Operating Characteristic) curve. The ROC curve is plotted with the false positive rate on the horizontal axis and the true positive rate on the vertical axis. A higher AUC value indicates better model classification performance. If, while meeting basic requirements, the AUC drops by more than 0.5%, it indicates that the model is not only improving slowly in accuracy but also experiencing a significant decline in overall classification performance. This further suggests that the model may be overfitting or trapped in a local optimum, and training should be stopped in this case.

[0093] In an optional embodiment, this application can verify the trained functional navigation model. For example, when a user inputs a preset query command for the target vehicle model, "perform steering wheel angle learning," the functional navigation model automatically matches: Steering Angle Sensor Reset, X208A calibration, and steering angle zeroing learning corresponding to the target vehicle model. This application verifies the accuracy of the functional navigation model based on the matching degree between the input preset query command and the matched functions. If the matching degree reaches the preset matching degree, the initial functional navigation model that achieves the target can be verified.

[0094] The matching score is calculated as follows: final_score = α×a + β×b + γ×c + δ×d, where α + β + γ + δ = 1. Here, final_score represents the matching score between the preset query command and the matched function; 'a' represents the semantic similarity between the input preset query command and the matched candidate function command; 'b' represents the co-occurrence frequency between the input preset query command and the matched candidate function command; 'c' represents the vehicle model matching score between the input preset query command and the matched candidate function command; 'd' represents the user's preference for the matched candidate function command; 'α' represents the semantic similarity weight; 'β' represents the co-occurrence frequency weight; 'γ' represents the vehicle model matching score weight; and 'δ' represents the user preference weight.

[0095] S133, the initial functional navigation model is optimized online through an incremental learning pipeline to obtain the final functional navigation model.

[0096] As an example, the incremental learning pipeline is an online learning implementation method. By fine-tuning the initial functional navigation model with new data based on the original initial functional navigation model, the model performance can be continuously optimized, and the forgetting of old knowledge due to the introduction of new data can be avoided.

[0097] In one embodiment, the incremental learning pipeline includes: new data input, feature extraction, similar sample retrieval, and local parameter update. New data arrives incrementally. For each new data sample, a feature extraction process is performed to obtain a semantic vector and a corresponding vehicle model feature vector. Similar sample retrieval is used to quickly find the set of samples most similar to the new data from the historical dataset; the new samples share similar vehicle model features with the historical database.

[0098] This application employs a high-efficiency nearest neighbor search algorithm, including index structure optimization, a vector database, and rule filtering. The index structure optimization primarily utilizes kd-trees (K-Dimension Trees, binary search trees for dimensional space) (suitable for low-dimensional feature spaces, quickly segmenting vehicle model feature spaces) and locality-sensitive hashing (suitable for high-dimensional semantic vectors, performing hash bucketing for approximate nearest neighbor search). The vector space library uses dedicated tools (FAISS, Milvus) to manage historical feature vectors, supporting millisecond-level retrieval. Rule filtering includes hard matching of vehicle models (if the data source is a pure electric vehicle, directly filtering samples from gasoline vehicles) and protocol version constraints (only retrieving historical data supporting the current bus architecture).

[0099] Finally, based on the identification of similar samples, this application employs a local parameter update operation. By fine-tuning or updating some parameters on the basis of existing model parameters, it adapts to the characteristics of the new data. This process can help improve the model's adaptability and generalization ability to new data.

[0100] In an optional embodiment, this application employs an Elastic Weight Consolidation (EWC) algorithm to limit adjustments to weights important in previous tasks, thereby protecting these weights from excessive alteration and maintaining the model's memory of old tasks, preventing catastrophic forgetting. The Elastic Weight Consolidation algorithm is based on the Fisher information matrix, which measures the importance of each weight in previous tasks. Specifically, EWC slows down the rate of change of each weight by calculating the importance of each weight and imposing regularization constraints on the weights when training new tasks.

[0101] S140: The function navigation model matches the input preset query commands and generates a recommended list of recommended function commands.

[0102] In one embodiment, the preset query command is derived from a predefined diagnostic command library. For example... Figure 4 As shown, the recommended function instructions obtained based on the function navigation model in this application include the following steps:

[0103] S141, perform semantic parsing on the preset query command to obtain the semantic feature vector of the preset query command.

[0104] Exemplary, the preset query instruction includes the target function name corresponding to the preset query instruction, a list of function parameters identifying the diagnostic instruction, and the target operation scenario. Similar to determining the semantic vector of candidate function instructions, this application inputs the string of the preset query instruction into the Bi-LSTM model. The Bi-LSTM model processes the string character by character and outputs a fixed-dimensional vector representation to obtain the semantic feature vector of the preset query instruction.

[0105] S142, by searching for multiple nearest neighbors of the semantic feature vector of the preset query instruction in the feature space of the candidate function instruction through nearest neighbor retrieval, the initial recommended function instruction is determined.

[0106] By way of example, this application can perform intelligent matching through a functional navigation model, calculate semantic similarity based on the semantic feature vector of the preset query instruction and the semantic vector of the candidate functional instructions, and find the top n candidate diagnostic instructions with the highest semantic similarity as the initial recommended functional instructions.

[0107] S143: After filtering the initial recommended function commands by vehicle type and sorting by popularity, the final recommended function commands are determined and a recommendation list is generated.

[0108] In one embodiment, the feature information includes the vehicle model matching degree of the candidate function instructions. This application filters recommended function instructions that are compatible with the target vehicle model based on the vehicle model matching degree. Based on the popularity index of the filtered recommended function instructions, the recommended function instructions are sorted in descending order to obtain a recommendation list.

[0109] Exemplary, this application determines the vehicle model matching degree between the initial recommended function command and the preset query command based on the target ECU characteristic parameters to which the preset query command belongs and the ECU characteristic parameters to which the initial recommended function command belongs, and filters out and deletes the initial recommended function commands that do not match the vehicle model of the preset query command. For example, a preset vehicle model matching degree can be set, retaining the initial recommended function commands that exceed the preset vehicle model matching degree, and filtering out and deleting the initial recommended function commands that do not exceed the preset vehicle model matching degree.

[0110] In one embodiment, the present application further determines a reward signal based on the user's selection result, so as to update the parameters of the functional navigation model according to the reward signal.

[0111] This application employs a feedback learning mechanism to collect user selection results from the recommendation list and calculate a reward signal: r = 1 - (selection position / list length). The parameters of the functional navigation model are updated based on the expression used to update model parameters in reinforcement learning.

[0112] The expression for updating model parameters is: Where θ , ε represents the updated model parameters, θ represents the original model parameters, and ε represents the learning rate. The learning rate controls the influence of new data on the old parameters during the update process; the larger the learning rate, the greater the magnitude of the model parameter update. r represents the reward signal, which indicates the reward obtained after performing a certain action. The gradient of the Q-value obtained by performing action a in state s is represented by , where Q-value represents the expected long-term reward for performing an action in a given state. The gradient represents the rate of change of the Q-value relative to the model parameters, thus indicating how to adjust the model parameters to maximize the Q-value. s is the state, representing the current environment or situation. In this application, the state can be information such as the semantic vector of a preset query instruction or a vehicle model feature vector. A is the action, representing the instruction selected by the model in the given state. In this application, the action can be a recommended function instruction recommended by the model to the user from candidate function instructions.

[0113] This application also provides a vehicle diagnostic device, which can be an intelligent vehicle diagnostic device, a remote vehicle diagnostic device, a remote video vehicle diagnostic device, or other vehicle diagnostic devices, etc., and is not limited here.

[0114] Exemplary, the vehicle diagnostic device includes a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program to enable the vehicle diagnostic device to perform the aforementioned functional navigation method.

[0115] The processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, including at least one of a Central Processing Unit (CPU), Graphics Processing Unit (GPU), Network Processor (NP), Digital Signal Processor (DSP), Application-Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application.

[0116] The memory can be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The memory is used to store computer programs, and the processor can execute the computer programs accordingly after receiving execution instructions.

[0117] This application also provides a computer-readable storage medium for storing the computer program used in the aforementioned vehicle diagnostic equipment. For example, the computer-readable storage medium may include, but is not limited to, various media capable of storing program code, such as a USB flash drive, a portable hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0118] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that, in alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0119] In addition, the functional modules or units in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0120] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a smartphone, personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0121] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method of functional navigation, characterized by, include: A training dataset for constructing a functional navigation model for an in-vehicle terminal is provided, the training dataset including a vehicle model feature fingerprint database and a diagnostic instruction database; A vehicle feature vector is generated based on the extracted ECU feature parameters to establish the vehicle feature fingerprint database; original diagnostic commands for multiple vehicle models are collected, and the call frequency of each candidate function command is determined based on the call frequency of the original diagnostic commands, as well as the usage popularity index of each candidate function command, to establish the diagnostic command database. Feature extraction is performed based on the training dataset to determine the feature information of the candidate function instructions, the feature information including the vehicle model matching degree of the candidate function instructions; The functional navigation model is constructed based on the training dataset and the feature information; The function navigation model matches the input preset query commands to generate a recommended list of recommended function commands.

2. The method of claim 1, wherein, The training dataset also includes a semantic association graph, the candidate functional instructions include multiple identifier diagnostic instructions, each of the identifier diagnostic instructions includes functional parameters, and the training dataset for constructing the functional navigation model of the vehicle terminal also includes: The name, function parameter list, and operation scenario of each candidate function instruction are encoded to obtain the semantic vector of each candidate function instruction; A semantic association graph of the candidate functional instructions is constructed based on the semantic vectors of each candidate functional instruction.

3. The method of claim 2, wherein, The construction of the semantic association graph of the candidate functional instructions based on the semantic vectors of each candidate functional instruction includes: Each of the candidate function instructions is used as a node in the semantic association graph; The semantic vector and call frequency of each candidate function instruction, as well as the vehicle model feature vector to which the candidate function instruction belongs, are used as node attributes of the semantic association graph. The edge weights of the semantic association graph are determined based on the semantic similarity and co-occurrence frequency among the candidate functional instructions.

4. The method of claim 1, wherein, The construction of the functional navigation model based on the training dataset and the feature information includes: Construct an initial classifier and train it based on the training dataset and the feature information; The training of the initial classifier is stopped under preset conditions by an early stopping mechanism to obtain the optimal functional navigation model parameters and thus the initial functional navigation model. The initial functional navigation model is optimized online through an incremental learning pipeline to obtain the final functional navigation model.

5. The method of claim 1, wherein, The step of matching the input preset query command with the functional navigation model to generate a recommended list of recommended functional commands includes: The preset query command is semantically parsed to obtain the semantic feature vector of the preset query command; The initial recommended function instruction is determined by searching for multiple nearest neighbors of the semantic feature vector of the preset query instruction in the feature space of the candidate function instruction through nearest neighbor retrieval. After filtering the initial recommended function commands by vehicle type and sorting them by popularity, the final recommended function commands are determined.

6. The method of claim 5, wherein, The step of filtering and ranking the initial recommended function commands by vehicle type includes: Based on the vehicle matching degree, a recommended function command that is compatible with the target vehicle model is selected; Based on the popularity index of the filtered recommended function commands, the recommended function commands are sorted in descending order to obtain the recommended list.

7. The method of claim 1, wherein, The method further includes: A reward signal is determined based on the user's selection, and the parameters of the functional navigation model are updated based on the reward signal.

8. A vehicle diagnosis device characterized by comprising: The vehicle diagnostic device includes a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the functional navigation method according to any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed on a processor, implements the functional navigation method according to any one of claims 1-7.

Images