Vehicle machine application preloading method and device, medium and equipment
By collecting multi-source vehicle data and using decision trees and recurrent neural networks to generate pre-loading instructions, the system solves the problems of vehicle application startup delay and resource conflicts, achieves accurate pre-loading and safety-first resource allocation, and improves system stability and response speed.
Patent Information
- Application Number
- CN202511733959.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-24
- Publication Date
- 2026-02-24
AI Technical Summary
Existing in-vehicle application preloading technology suffers from a lack of vehicle scene perception, resource conflict risks, and insufficient cold start optimization, resulting in startup delays and resource consumption that affect safety-critical functions.
Collect multi-source heterogeneous vehicle data, generate vehicle scene labels and preloaded action instructions through a pre-set decision tree model and recurrent neural network to achieve accurate preloading, and optimize resource allocation by combining decision process and time series network.
Reduce vehicle infotainment application startup latency, improve preloading accuracy, ensure critical functional resources, enhance system stability, and achieve driving safety and rapid response.
Smart Images
Figure CN121560411A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of vehicle technology, and in particular to a preloading method, apparatus, medium and equipment for vehicle infotainment applications. Background Technology
[0002] The preloading technologies currently used in in-vehicle infotainment systems can be mainly divided into three categories:
[0003] 1. Frequency preloading: Based on a statistical model of the number of times an app has been launched in the past / runtime, high-frequency applications are preloaded in the background.
[0004] 2. Related preloading: Trigger related loading based on application call relationships (such as navigation → charging payment).
[0005] 3. Layout preloading: Multi-threaded asynchronous loading of UI components reduces resource parsing time during cold starts.
[0006] In the context of in-vehicle infotainment systems, the above-mentioned existing technologies have at least the following drawbacks:
[0007] 1. Lack of scene awareness:
[0008] (1) It relies on historical frequency or simple location fences, without integrating vehicle status and driving behavior temporal features.
[0009] (2) Example: The video app is preloaded when the vehicle is traveling at high speed, or the charging station search tool is not preloaded when the battery is low.
[0010] 2. Resource conflict risk:
[0011] (1) Preloading occupies CPU / memory resources, which may affect key safety functions such as ADAS and emergency braking.
[0012] (2) Lack of application priority classification mechanism.
[0013] 3. Insufficient cold start optimization:
[0014] Traditional layout preloading requires the main thread to render UI components (such as DecorView), and complex applications still require 1-3 seconds to start. Operation delays while driving can easily lead to distraction accidents. Summary of the Invention
[0015] The purpose of this invention is to provide a method, apparatus, medium, and device for preloading in-vehicle applications, which can at least reduce the startup delay of in-vehicle applications, improve the preloading accuracy of in-vehicle applications, ensure key in-vehicle functional resources, and enhance the stability of the in-vehicle system.
[0016] To address the aforementioned technical problems, in a first aspect, the present invention provides a method for preloading in-vehicle applications, comprising at least:
[0017] Collect multi-source heterogeneous data of vehicles and perform data encoding operations on the multi-source heterogeneous data of vehicles to form multi-dimensional feature vectors;
[0018] Based on the vehicle multi-source heterogeneous data, the multi-dimensional feature vector, the preset decision tree model, and the preset recurrent neural network (RNN) model, at least vehicle scene labels and preloaded action instructions are generated.
[0019] The vehicle scene tag and the preloading action instruction are parsed to obtain at least one preloading type of vehicle-mounted application, and then the preloading action of the corresponding vehicle-mounted application is executed according to each preloading type.
[0020] Optionally, the preset decision tree model uses at least one vehicle state data from the multi-source heterogeneous vehicle data as at least one root node of the model;
[0021] The preset decision tree model uses at least one type of context information corresponding to each vehicle state data as at least one branch node of the model's root node.
[0022] The preset decision tree model takes at least one vehicle system application requirement scenario and pre-loaded action corresponding to each of the context information as at least one leaf node of the corresponding branch node of the model.
[0023] Optionally, the preset decision tree model monitors the prediction success rate of at least each model decision branch; when the frequency of user rejection or misjudgment rate of any model decision branch in vehicle application reaches a preset threshold, the preset decision tree model performs a pruning operation on the corresponding model decision branch.
[0024] Optionally, the generation of vehicle scene labels and preloaded action instructions based on the vehicle multi-source heterogeneous data, the multi-dimensional feature vector, the preset decision tree model, and the preset recurrent neural network (RNN) model at least includes:
[0025] A first decision result is generated based on the vehicle's multi-source heterogeneous data and the preset decision tree model;
[0026] A second decision result is generated based on the multidimensional feature vector and the preset RNN model;
[0027] The first decision result and the second decision result are integrated to generate the vehicle scene label and the preloaded action instructions;
[0028] Specifically, when the deviation between the first decision result and the second decision result exceeds a preset deviation threshold, the preset decision tree model triggers a decision branch addition process, generating at least one model decision branch by analyzing context deviation features.
[0029] Optionally, the step of parsing the vehicle scene tag and the preloading action instruction to obtain at least one preloading type of the vehicle-mounted application, and then executing the preloading action of the corresponding vehicle-mounted application according to each preloading type, specifically includes at least the following:
[0030] The vehicle scene tags and the preloaded action instructions are parsed to obtain at least one type of preloaded vehicle-mounted application.
[0031] When the preloading type of any vehicle infotainment application is the first preloading level, at least all core services and user interface (UI) components of the corresponding vehicle infotainment application will be fully loaded into memory.
[0032] When the preloading type of any vehicle infotainment application is the second preloading level, at least the corresponding vehicle infotainment application process is pre-created, and the core UI rendering and layout calculation are completed.
[0033] When the preloading type of any vehicle infotainment application is the third preloading level, at least the corresponding vehicle infotainment application process is pre-created and a preset memory size is allocated to it.
[0034] Based on the same concept, in a second aspect, the present invention also provides a vehicle application preloading device for performing the vehicle application preloading method described in any one of the first aspects;
[0035] The in-vehicle application preloading device includes at least:
[0036] The data acquisition module is used to collect multi-source heterogeneous data of the vehicle and perform data encoding operations on the multi-source heterogeneous data of the vehicle to form a multi-dimensional feature vector.
[0037] The instruction generation module is used to generate at least vehicle scene labels and preloaded action instructions based on the vehicle multi-source heterogeneous data, the multi-dimensional feature vector, the preset decision tree model and the preset recurrent neural network (RNN) model.
[0038] An action execution module is used to parse the vehicle scene label and the preload action instruction to obtain at least one preload type of vehicle-mounted application, and then execute the preload action of the corresponding vehicle-mounted application according to each preload type.
[0039] Optionally, the instruction generation module is at least used for:
[0040] A first decision result is generated based on the vehicle's multi-source heterogeneous data and the preset decision tree model;
[0041] A second decision result is generated based on the multidimensional feature vector and the preset RNN model;
[0042] The first decision result and the second decision result are integrated to generate the vehicle scene label and the preloaded action instructions;
[0043] Specifically, when the deviation between the first decision result and the second decision result exceeds a preset deviation threshold, the preset decision tree model triggers a decision branch addition process, generating at least one model decision branch by analyzing context deviation features.
[0044] Optionally, the action execution module is specifically used for at least:
[0045] The vehicle scene tags and the preloaded action instructions are parsed to obtain at least one type of preloaded vehicle-mounted application.
[0046] When the preloading type of any vehicle infotainment application is the first preloading level, at least all core services and user interface (UI) components of the corresponding vehicle infotainment application will be fully loaded into memory.
[0047] When the preloading type of any vehicle infotainment application is the second preloading level, at least the corresponding vehicle infotainment application process is pre-created, and the core UI rendering and layout calculation are completed.
[0048] When the preloading type of any vehicle infotainment application is the third preloading level, at least the corresponding vehicle infotainment application process is pre-created and a preset memory size is allocated to it.
[0049] Based on the same concept, in a third aspect, the present invention also provides an electronic device, including a memory and a processor, the memory storing a computer program executable on the processor, wherein the processor, when executing the program, implements the steps in the vehicle application preloading method of any one of the first aspects.
[0050] Based on the same concept, in a fourth aspect, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps in the vehicle application preloading method of any one of the first aspects.
[0051] The technical solution provided by this invention first collects multi-source heterogeneous vehicle data and performs data encoding operations on the multi-source heterogeneous vehicle data to form a multi-dimensional feature vector. Further, based on the multi-source heterogeneous vehicle data, the multi-dimensional feature vector, a preset decision tree model, and a preset recurrent neural network (RNN) model, at least vehicle scene labels and pre-loaded action instructions are generated. Finally, the vehicle scene labels and pre-loaded action instructions are parsed to obtain at least one pre-loading type for a vehicle infotainment system application, and then the corresponding pre-loading action for each pre-loading type is executed. This invention can at least reduce the startup latency of vehicle infotainment systems, improve the pre-loading accuracy of vehicle infotainment systems, ensure key vehicle infotainment system functional resources, and enhance the stability of the vehicle infotainment system. Attached Figure Description
[0052] Figure 1 This is a flowchart of a vehicle infotainment system application preloading method provided in an embodiment of the present invention;
[0053] Figure 2 This is a schematic diagram of the structure of a vehicle infotainment preloading device provided in an embodiment of the present invention;
[0054] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation
[0055] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0056] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the application. The singular forms “a,” “said,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.
[0057] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.
[0058] It should be understood that although the terms first, second, third, etc., may be used in the embodiments of this application, these descriptions should not be limited to these terms. These terms are only used to distinguish the descriptions. For example, first may also be referred to as second without departing from the scope of the embodiments of this application, and similarly, second may also be referred to as first.
[0059] Depending on the context, the words “if” or “suppose” as used here can be interpreted as “when” or “in response to determination” or “in response to detection.” Similarly, depending on the context, the phrases “if determination” or “if detection (of the stated condition or event)” can be interpreted as “when determination” or “in response to determination” or “when detection (of the stated condition or event)” or “in response to detection (of the stated condition or event).”
[0060] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or device that includes said element.
[0061] It should be noted that any symbols and / or numbers present in the specification that are not marked in the accompanying drawings are not reference numerals.
[0062] Figure 1 This is a flowchart of a vehicle infotainment application preloading method provided by an embodiment of the present invention. This embodiment is applicable to at least any vehicle infotainment application control and optimization scenario. The vehicle infotainment application preloading method can be, but is not limited to, executed by the vehicle infotainment application preloading device in this embodiment as the execution subject, which can be implemented in software and / or hardware. Figure 1 As shown, the in-vehicle application preloading method includes at least the following steps:
[0063] S1. Collect multi-source heterogeneous data of vehicles and perform data encoding operations on the multi-source heterogeneous data of vehicles to form multi-dimensional feature vectors.
[0064] The multi-source heterogeneous vehicle data can be collected from the vehicle perception layer. Specifically, this data can include vehicle speed, gear position, and battery level data collected via CAN bus or Ethernet interface; GPS location, current time, and driving route obtained from positioning devices; and user behavior data, such as voice, gestures, and blinking, uploaded to the cloud in real time via the vehicle's infotainment app's data tracking system. Furthermore, the fixed-length feature vector formed by encoding the above data is the aforementioned multi-dimensional feature vector; the multi-dimensional feature vector can have 72 dimensions.
[0065] S2. Based on multi-source heterogeneous vehicle data, multi-dimensional feature vectors, a preset decision tree model, and a preset recurrent neural network (RNN) model, generate at least vehicle scene labels and preloaded action instructions.
[0066] The preset RNN model can be specifically a Long Short-Term Memory (LSTM) network.
[0067] In one specific implementation, optionally, the preset decision tree model uses at least one vehicle state data from the multi-source heterogeneous vehicle data as at least one root node of the model.
[0068] The default decision tree model uses at least one type of contextual information corresponding to each vehicle state data as at least one branch node of the root node of the model.
[0069] The pre-defined decision tree model takes at least one vehicle system application requirement scenario and pre-loaded action corresponding to each context information as at least one leaf node of the corresponding branch node of the model.
[0070] In another specific implementation, optionally, the preset decision tree model monitors the prediction success rate of at least each model decision branch; when the user rejection frequency or misjudgment rate of any model decision branch in vehicle application reaches a preset threshold, the preset decision tree model performs a pruning operation on the corresponding model decision branch.
[0071] In another specific implementation, optionally, the vehicle scene labels and preloaded action instructions generated based on multi-source heterogeneous vehicle data, multi-dimensional feature vectors, a preset decision tree model, and a preset recurrent neural network (RNN) model include at least:
[0072] The first decision result is generated based on multi-source heterogeneous vehicle data and a pre-set decision tree model.
[0073] A second decision result is generated based on multidimensional feature vectors and a pre-defined RNN model;
[0074] Integrate the results of the first and second decisions to generate vehicle scene labels and preloaded action instructions;
[0075] Specifically, when the deviation between the first decision result and the second decision result exceeds the preset deviation threshold, the preset decision tree model triggers the decision branch addition process, generating at least one model decision branch by analyzing the context deviation characteristics.
[0076] For example, a pre-defined decision tree model can be configured with a hierarchical conditional judgment structure:
[0077] 1. Root Node: As the starting point for decision-making, it is responsible for detecting the most critical and widely influential single vehicle state. For example, vehicle speed is a key detection indicator for the root node. The model can continuously monitor whether the vehicle speed exceeds a certain threshold (such as 60km / h), and this judgment can serve as the basis for subsequent speed decision branches.
[0078] 2. Branch Node: This node refines and deepens the conditions of the root node by introducing contextual information from multiple dimensions (such as time and location) to create a more specific scenario. For example, given a vehicle speed > 60 km / h, a branch node can further check if the time information falls within the "morning rush hour" (e.g., 7:00-9:00 on weekdays). When both the vehicle speed > 60 km / h and the morning rush hour conditions are met, the branch node successfully matches the initial conditions for "commuting mode".
[0079] 3. Leaf Node: Represents the endpoint of a complete scene judgment. Upon reaching a leaf node, the model can output a clear scene label (such as the aforementioned commuting mode) and a pre-loaded action instruction bound to that label (such as loading a navigation app). The two together constitute the aforementioned first decision result.
[0080] Furthermore, the pre-defined decision tree model can also be configured with a priority mechanism:
[0081] To handle emergency or special situations, the model's decision-making process can incorporate a strict priority ranking. Judgment rules for emergency scenarios are assigned the highest priority and can interrupt or override judgments for regular scenarios.
[0082] For example, a priority mechanism can be triggered when the battery level is detected to be below 20%. At this point, the model can immediately enforce the leaf node action corresponding to the "emergency low battery" scenario—loading the charging station search tool and pausing or downgrading background entertainment apps to ensure the normal operation of core functions. Understandably, this mechanism ensures driving safety and meets core needs.
[0083] Furthermore, the pre-defined decision tree model can also be configured with a dynamic optimization mechanism:
[0084] Pruning: The model continuously tracks the prediction success rate of each decision branch. If a branch (such as a "long-distance mode" built on historical data) is frequently "rejected" by users in real-world applications (i.e., the user does not take the corresponding action after the prediction), its misjudgment rate will continue to rise. When the misjudgment rate (or the frequency of user "rejection") exceeds a preset threshold, the model can automatically "prune," that is, temporarily disable or directly delete the decision branch to prevent it from continuously outputting incorrect predictions and ensure the accuracy of the rule base.
[0085] Furthermore, the pre-defined decision tree model can also be combined with LSTM / GRU to add new rules:
[0086] When the prediction result of the model's decision-making process (i.e., the aforementioned first decision result) deviates significantly from the prediction result of the LSTM / GRU time series prediction network (i.e., the aforementioned second decision result) (e.g., the deviation is greater than 15%, i.e., the aforementioned deviation exceeds the preset deviation threshold), and the user's actual behavior is more consistent with the LSTM / GRU prediction, the model can determine that there is a "knowledge blind spot" in the decision-making process. At this time, the model can trigger the process of adding new rules, automatically generating or prompting the administrator to create a new rule (e.g., high temperature reduction mode) by analyzing the key contextual features that caused the deviation (e.g., the temperature was high at the time), thereby achieving self-learning and self-improvement of the rule base.
[0087] More specifically, the pre-defined RNN model can be used to handle the non-linear, personalized, and long-term dependent habitual patterns that are difficult for the pre-defined decision tree model to capture and are hidden in the massive amount of user behavior.
[0088] Preset RNN models may include:
[0089] 1. Input Layer: Precisely integrates a 72-dimensional feature vector. This vector comprehensively depicts the complete state of "human-vehicle-environment" at each time step, including vehicle dynamic data (such as vehicle speed, battery level, gear position, etc.), spatiotemporal information (such as GPS, timestamps, etc.), and user historical behavior statistics (such as application usage frequency, order, etc.), providing the model with rich and detailed perceptual input.
[0090] 2. Hidden Layers: A three-layer stacked gated recurrent unit (GRU) structure can be used. Choosing GRU instead of standard LSTM is an optimization of the limited computing resources on the vehicle while ensuring model performance. It has fewer parameters and faster training and inference speeds.
[0091] (1) Layered learning: The first layer GRU can learn local and short-term behavioral associations (such as "checking navigation after listening to music"); the deeper second and third layers GRU can abstract longer-term and more complex macro patterns based on shallow features (such as identifying users' unique application combination habits when "going out for leisure on weekend evenings").
[0092] (2) Regularization: The strategy of "Dropout=0.2" is applied after each GRU layer. This will randomly ignore a part of the neurons during training, effectively preventing the model from overfitting to specific training data, and greatly enhancing the model's ability to generalize and predict new behaviors and new environments.
[0093] 3. Output Layer: The hidden state vector, which is highly condensed with the essence of historical information and output by the third-layer GRU at the last time step, is transformed into a clear application launch probability distribution through a Softmax activation function. For example, the output might be {Navigation App: 0.7, Music App: 0.2, Utility App: 0.1}, quantitatively expressing the model's prediction of the user's next intention.
[0094] S3. Parse vehicle scene tags and preload action instructions to obtain at least one preload type of vehicle infotainment application, and then execute the preload action of the corresponding vehicle infotainment application according to each preload type.
[0095] There can be multiple types of preloading, and each preloading type can correspond to a preloading level.
[0096] In yet another specific implementation, step S3 may optionally include at least the following:
[0097] (3-1) Parse vehicle scene tags and preloaded action instructions to obtain at least one type of preloaded vehicle application;
[0098] (3-2) When the preloading type of any vehicle infotainment application is the first preloading level, at least all core services and UI components of the corresponding vehicle infotainment application shall be fully loaded into memory;
[0099] (3-3) When the preloading type of any vehicle infotainment application is the second preloading level, at least the corresponding vehicle infotainment application process shall be created in advance, and the core UI rendering and layout calculation shall be completed.
[0100] (3-4) When the preloading type of any vehicle infotainment application is the third preloading level, at least the corresponding vehicle infotainment application process shall be pre-created and a preset memory size shall be allocated to it.
[0101] More specifically, the technical implementation of the aforementioned step S3 can be:
[0102] 1. Command Reception and Parsing: Receive scene tags and preloaded commands, such as "Commuter Mode → Load Navigation App".
[0103] 2. Categorized Execution Strategy:
[0104] (1) Navigation applications (corresponding to the first preloading level mentioned above; preloading level: full component loading):
[0105] Reason: Navigation applications are complex, requiring the loading of multiple core components such as map data, positioning modules, and route planning algorithms upon startup, resulting in the longest cold start time. Furthermore, user needs for navigation are typically highly deterministic and immediate.
[0106] Action: Upon receiving a load command, the application can be allocated significant system resources to fully load all its core services and UI components into memory. This is almost equivalent to fully launching the application and placing it in the background, thus achieving a "zero-latency" response when the user clicks.
[0107] (2) Entertainment applications (corresponding to the second preloading level mentioned above; preloading level: core UI pre-rendering):
[0108] Reason: The core of entertainment apps (such as music and podcasts) lies in content presentation and user interface; their backend data logic is relatively lightweight. Users have requirements for their startup speed, but their tolerance is slightly higher than that for navigation.
[0109] Action: Only the application process is pre-created, and its core UI rendering and layout calculations are completed, but no audio or video content is actually loaded. When the user clicks, the application framework is already ready, and content can be quickly populated, significantly reducing "white screen" time and achieving a good balance between speed and resource consumption.
[0110] (3) Utility applications (corresponding to the third preloading level mentioned above; preloading level: process pre-start):
[0111] Reason: Utility applications (such as charging station search and vehicle settings) have simple functions and start up quickly. Excessive preloading is not very meaningful and only wastes resources.
[0112] Action: Only the basic process of the application is pre-created, and a small amount of memory is allocated to it. This changes the application from a "not running" state to a "ready" state. When the user clicks, the overhead of creating a process is eliminated, and the startup speed is still slightly but effectively improved.
[0113] The technical solution provided in this embodiment firstly collects multi-source heterogeneous vehicle data and performs data encoding operations on the multi-source heterogeneous vehicle data to form a multi-dimensional feature vector; further, based on the multi-source heterogeneous vehicle data, the multi-dimensional feature vector, a preset decision tree model, and a preset recurrent neural network (RNN) model, at least vehicle scene labels and preloaded action instructions are generated; finally, the vehicle scene labels and preloaded action instructions are parsed to obtain at least one preloaded type of vehicle-mounted application, and then the preloaded action of the corresponding vehicle-mounted application is executed according to each preloaded type.
[0114] In summary, this embodiment achieves two key advantages: firstly, it enables precise pre-loading for adaptive driving scenarios, dynamically predicting needs by integrating multi-dimensional data such as time, location, and vehicle status; secondly, it establishes a safety-first resource allocation model to ensure the supply of resources for critical vehicle functions; and thirdly, it achieves "first-screen opening in seconds" (≤0.5 seconds) through component-level hierarchical pre-loading.
[0115] More specifically, this embodiment can achieve at least the following beneficial effects:
[0116] 1. Reduced startup latency: Achieve a step-by-step, refined improvement in application startup speed by preloading execution processes in a categorized manner.
[0117] Differentiated preloading strategies are adopted for different application characteristics, thereby significantly reducing the user-perceived startup latency. Specifically:
[0118] For high-priority, heavy-load applications such as navigation, a "full component loading" strategy is implemented to fully load their core services, UI, and data into memory, achieving a "second-open" effect with almost zero latency when the user clicks.
[0119] For medium-priority applications such as entertainment, the "core UI pre-rendering" strategy is implemented to pre-compile the interface layout calculation, avoiding the most time-consuming UI rendering process during startup. After the user clicks, they only need to fill in the content, which greatly reduces the white screen waiting time.
[0120] For low-frequency, lightweight applications such as utility applications, "process pre-start" is performed to pre-create basic processes, saving the system overhead of creating processes from scratch and achieving lightweight startup acceleration.
[0121] This tiered strategy replaces the traditional "one-size-fits-all" cold start process, avoiding unnecessary main thread blocking and precisely allocating resources to optimize the startup process that users are most sensitive to.
[0122] 2. Improved accuracy through preloading: Based on a dual-core engine of "decision process + temporal network", the prediction accuracy and generalization ability far exceed those of traditional solutions.
[0123] The "scene prediction engine" in this solution achieves high-precision prediction through a hybrid intelligent architecture.
[0124] Multi-dimensional data integration: Unlike traditional solutions that rely solely on geofencing, this solution integrates 72-dimensional feature vectors composed of data from CAN bus / Ethernet, including vehicle speed, gear position, battery level, GPS, time, and user historical behavior. This rich data input fundamentally improves the granularity and accuracy of scene perception.
[0125] Dual-core collaborative prediction:
[0126] (1) Decision-making process: It handles logically clear deterministic scenarios (such as morning and evening rush hour commuting and long-distance driving), ensuring the stability and reliability of basic scenario prediction.
[0127] (2) LSTM / GRU time series prediction network: As a data-driven core, it utilizes three-layer stacked GRU and Dropout regularization to deeply mine users' personalized and non-linear implicit habits (such as special application combinations for weekend travel), effectively making up for the blind spots of the rule system.
[0128] (3) The two work in parallel, enabling the system to deal with both deterministic logic and capture personalized patterns, ultimately stabilizing the overall confidence level of scene classification at over 92%.
[0129] 3. Critical Function Resource Guarantee: Relying on the priority mechanism in the decision-making process, dynamic resource scheduling with security as the priority is achieved.
[0130] The "decision-making process" of this solution incorporates a strict prioritization mechanism to ensure that resources for critical functions are guaranteed under any circumstances.
[0131] High-priority scenario enforcement: When a high-priority event is detected (such as battery level below 20% or ADAS system activation), the priority mechanism is triggered.
[0132] Dynamic resource reallocation: The system will immediately interrupt or override regular entertainment scenario predictions and force the execution of safety-related leaf node instructions (such as loading a charging station search tool). At the same time, the system will automatically reduce the CPU usage of non-critical applications (such as entertainment apps) (e.g., from 15% to 5%) and can complete resource preemption for safety processes with a response speed of less than 10 milliseconds, ensuring that driving safety and core needs are given the highest priority.
[0133] 4. Enhanced system stability: Through the dynamic optimization mechanism of the decision-making process, the system model achieves self-learning and self-adaptation.
[0134] This solution has the ability to self-iterate and optimize, ensuring long-term stability and high efficiency.
[0135] Automatic pruning: The system continuously monitors the prediction success rate of each branch in the decision-making process. When the misjudgment rate of a certain rule branch exceeds a preset threshold (e.g., 20%), the system will automatically "prune" (i.e., disable or delete) it to avoid interference from inefficient rules continuously outputting incorrect predictions, thus ensuring the overall accuracy of the rule base.
[0136] Rule self-growth (Collaborative LSTM / GRU): When the decision-making process deviates significantly from the LSTM network's prediction, and the user's actual behavior verifies the accuracy of the LSTM, the system determines that there is a "knowledge blind spot" in the decision-making process. At this point, the system triggers a new rule-addition process, analyzes key contextual features, and automatically generates new rules to improve the decision tree. This self-feedback learning mechanism enables the system to continuously learn from operation, constantly improving its adaptability to new scenarios and behaviors, thereby enhancing the stability and vitality of the entire system.
[0137] Figure 2 This is a schematic diagram of a vehicle-mounted application preloading device provided in an embodiment of the present invention. This embodiment is applicable to at least any vehicle-mounted application control and optimization scenario. The vehicle-mounted application preloading device can be implemented in software and / or hardware. Figure 2 As shown, the vehicle application preloading device is used to execute the vehicle application preloading method of any of the foregoing embodiments or implementations.
[0138] The in-vehicle infotainment system preloading device includes at least:
[0139] The data acquisition module 110 is used to acquire multi-source heterogeneous data of the vehicle and perform data encoding operations on the multi-source heterogeneous data of the vehicle to form a multi-dimensional feature vector.
[0140] The instruction generation module 120 is used to generate at least vehicle scene labels and preloaded action instructions based on multi-source heterogeneous vehicle data, multi-dimensional feature vectors, preset decision tree models and preset recurrent neural network (RNN) models.
[0141] The action execution module 130 is used to parse vehicle scene tags and preload action instructions to obtain at least one preload type of vehicle infotainment application, and then execute the preload action of the corresponding vehicle infotainment application according to each preload type.
[0142] Optionally, the instruction generation module 120 is used at least for:
[0143] The first decision result is generated based on multi-source heterogeneous vehicle data and a pre-set decision tree model.
[0144] A second decision result is generated based on multidimensional feature vectors and a pre-defined RNN model;
[0145] Integrate the results of the first and second decisions to generate vehicle scene labels and preloaded action instructions;
[0146] Specifically, when the deviation between the first decision result and the second decision result exceeds the preset deviation threshold, the preset decision tree model triggers the decision branch addition process, generating at least one model decision branch by analyzing the context deviation characteristics.
[0147] Optionally, the action execution module 130 is specifically used for at least:
[0148] Parse vehicle scene tags and preloaded action instructions to obtain at least one type of preloaded vehicle infotainment application;
[0149] When the preloading type of any vehicle infotainment application is the first preloading level, at least all core services and user interface (UI) components of the corresponding vehicle infotainment application will be fully loaded into memory.
[0150] When the preloading type of any vehicle infotainment application is the second preloading level, at least the corresponding vehicle infotainment application process is created in advance, and the core UI rendering and layout calculation are completed.
[0151] When the preloading type of any vehicle infotainment application is the third preloading level, at least the corresponding vehicle infotainment application process is pre-created and a preset memory size is allocated to it.
[0152] Optionally, the preset decision tree model uses at least one vehicle state data from the multi-source heterogeneous vehicle data as at least one root node of the model;
[0153] The default decision tree model uses at least one type of contextual information corresponding to each vehicle state data as at least one branch node of the root node of the model.
[0154] The pre-defined decision tree model takes at least one vehicle system application requirement scenario and pre-loaded action corresponding to each context information as at least one leaf node of the corresponding branch node of the model.
[0155] Optionally, the preset decision tree model monitors the prediction success rate of at least each model decision branch; when the user rejection frequency or misjudgment rate of any model decision branch in vehicle application reaches a preset threshold, the preset decision tree model performs a pruning operation on the corresponding model decision branch.
[0156] The technical solution provided in this embodiment firstly collects multi-source heterogeneous vehicle data through a data acquisition module and performs data encoding operations on the multi-source heterogeneous vehicle data to form a multi-dimensional feature vector. Further, an instruction generation module generates at least vehicle scene labels and pre-loaded action instructions based on the multi-source heterogeneous vehicle data, the multi-dimensional feature vector, a preset decision tree model, and a preset recurrent neural network (RNN) model. Finally, an action execution module parses the vehicle scene labels and pre-loaded action instructions to obtain at least one pre-load type for a vehicle infotainment application, and then executes the corresponding pre-load action for each pre-load type. This embodiment can at least reduce the startup latency of vehicle infotainment applications, improve the pre-loading accuracy of vehicle infotainment applications, ensure key vehicle infotainment functional resources, and enhance the stability of the vehicle infotainment system.
[0157] This embodiment provides an electronic device. Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. See also: Figure 3 The electronic device 1000 includes a processor 1001 and a memory 1002. The memory 1002 stores computer-readable instructions. When the computer-readable instructions are executed by the processor 1001, the steps in any of the above vehicle application preloading methods are performed. Through the above technical solution, the processor 1001 and the memory 1002 are interconnected and communicate with each other through a communication bus and / or other forms of connection mechanism (not shown). The memory 1002 stores a computer program that can be executed by the processor. When the electronic device 1000 is running, the processor 1001 executes the computer program to execute the vehicle application preloading method in any optional implementation of the above embodiments, so as to achieve at least the following functions: collecting multi-source heterogeneous data of the vehicle and performing data encoding operations on the multi-source heterogeneous data of the vehicle to form a multi-dimensional feature vector; generating at least vehicle scene labels and preloading action instructions based on the multi-source heterogeneous data of the vehicle, the multi-dimensional feature vector, the preset decision tree model and the preset recurrent neural network (RNN) model through the instruction generation module; parsing the vehicle scene labels and preloading action instructions through the action execution module to obtain at least one preloading type of vehicle application, and then executing the preloading action of the corresponding vehicle application according to each preloading type.
[0158] This embodiment provides a computer-readable storage medium storing a computer program. When executed by a processor, the program implements the vehicle application preloading method provided in all embodiments of this application: collecting multi-source heterogeneous vehicle data and performing data encoding operations on the multi-source heterogeneous vehicle data to form a multi-dimensional feature vector; generating at least vehicle scene labels and preloading action instructions based on the multi-source heterogeneous vehicle data, the multi-dimensional feature vector, a preset decision tree model, and a preset recurrent neural network (RNN) model through an instruction generation module; parsing the vehicle scene labels and preloading action instructions through an action execution module to obtain at least one preloading type of vehicle application, and then executing the corresponding preloading action of the vehicle application according to each preloading type.
[0159] Any combination of one or more computer-readable media may be used. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device.
[0160] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including—but not limited to—electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of transmitting, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.
[0161] The program code contained on a computer-readable medium may be transmitted using any suitable medium, including—but not limited to—wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.
[0162] Computer program code for performing the operations of this invention can be written in one or more programming languages or a combination thereof. Programming languages include object-oriented programming languages—such as Java, Smalltalk, and C++—as well as conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0163] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for preloading vehicle infotainment applications, characterized in that, At least including: Collect multi-source heterogeneous data of vehicles and perform data encoding operations on the multi-source heterogeneous data of vehicles to form multi-dimensional feature vectors; Based on the vehicle multi-source heterogeneous data, the multi-dimensional feature vector, the preset decision tree model, and the preset recurrent neural network (RNN) model, at least vehicle scene labels and preloaded action instructions are generated. The vehicle scene tag and the preloading action instruction are parsed to obtain at least one preloading type of vehicle-mounted application, and then the preloading action of the corresponding vehicle-mounted application is executed according to each preloading type.
2. The in-vehicle application preloading method according to claim 1, characterized in that, The preset decision tree model uses at least one vehicle state data from the multi-source heterogeneous vehicle data as at least one root node of the model; The preset decision tree model uses at least one type of context information corresponding to each vehicle state data as at least one branch node of the model's root node. The preset decision tree model takes at least one vehicle system application requirement scenario and pre-loaded action corresponding to each of the context information as at least one leaf node of the corresponding branch node of the model.
3. The in-vehicle application preloading method according to claim 1, characterized in that, The preset decision tree model monitors the prediction success rate of at least each model decision branch; when the user rejection frequency or misjudgment rate of any model decision branch in vehicle application reaches a preset threshold, the preset decision tree model performs a pruning operation on the corresponding model decision branch.
4. The in-vehicle application preloading method according to claim 1, characterized in that, The generation of vehicle scene labels and preloaded action instructions based on the vehicle multi-source heterogeneous data, the multi-dimensional feature vector, the preset decision tree model, and the preset recurrent neural network (RNN) model at least includes: A first decision result is generated based on the vehicle's multi-source heterogeneous data and the preset decision tree model; A second decision result is generated based on the multidimensional feature vector and the preset RNN model; The first decision result and the second decision result are integrated to generate the vehicle scene label and the preloaded action instructions; Specifically, when the deviation between the first decision result and the second decision result exceeds a preset deviation threshold, the preset decision tree model triggers a decision branch addition process, generating at least one model decision branch by analyzing context deviation features.
5. The in-vehicle application preloading method according to claim 1, characterized in that, The step of parsing the vehicle scene tags and the preloading action instructions to obtain at least one preloading type of in-vehicle application, and then executing the corresponding preloading action of the in-vehicle application according to each preloading type, specifically includes at least the following: The vehicle scene tags and the preloaded action instructions are parsed to obtain at least one type of preloaded vehicle-mounted application. When the preloading type of any vehicle infotainment application is the first preloading level, at least all core services and user interface (UI) components of the corresponding vehicle infotainment application will be fully loaded into memory. When the preloading type of any vehicle infotainment application is the second preloading level, at least the corresponding vehicle infotainment application process is pre-created, and the core UI rendering and layout calculation are completed. When the preloading type of any vehicle infotainment application is the third preloading level, at least the corresponding vehicle infotainment application process is pre-created and a preset memory size is allocated to it.
6. A preloading device for vehicle infotainment systems, characterized in that, Used to perform the vehicle application preloading method according to any one of claims 1-5; The in-vehicle application preloading device includes at least: The data acquisition module is used to collect multi-source heterogeneous data of the vehicle and perform data encoding operations on the multi-source heterogeneous data of the vehicle to form a multi-dimensional feature vector. The instruction generation module is used to generate at least vehicle scene labels and preloaded action instructions based on the vehicle multi-source heterogeneous data, the multi-dimensional feature vector, the preset decision tree model and the preset recurrent neural network (RNN) model. An action execution module is used to parse the vehicle scene label and the preload action instruction to obtain at least one preload type of vehicle-mounted application, and then execute the preload action of the corresponding vehicle-mounted application according to each preload type.
7. The vehicle infotainment application preloading device according to claim 6, characterized in that, The instruction generation module is used at least for: A first decision result is generated based on the vehicle's multi-source heterogeneous data and the preset decision tree model; A second decision result is generated based on the multidimensional feature vector and the preset RNN model; The first decision result and the second decision result are integrated to generate the vehicle scene label and the preloaded action instructions; Specifically, when the deviation between the first decision result and the second decision result exceeds a preset deviation threshold, the preset decision tree model triggers a decision branch addition process, generating at least one model decision branch by analyzing context deviation features.
8. The vehicle infotainment system preloading device according to claim 6, characterized in that, The action execution module is specifically used for at least: The vehicle scene tags and the preloaded action instructions are parsed to obtain at least one type of preloaded vehicle-mounted application. When the preloading type of any vehicle infotainment application is the first preloading level, at least all core services and user interface (UI) components of the corresponding vehicle infotainment application will be fully loaded into memory. When the preloading type of any vehicle infotainment application is the second preloading level, at least the corresponding vehicle infotainment application process is pre-created, and the core UI rendering and layout calculation are completed. When the preloading type of any vehicle infotainment application is the third preloading level, at least the corresponding vehicle infotainment application process is pre-created and a preset memory size is allocated to it.
9. An electronic device comprising a memory and a processor, the memory storing a computer program executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the vehicle application preloading method according to any one of claims 1 to 5.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps in the vehicle application preloading method according to any one of claims 1 to 5.