A timing covariance variable prediction perception adaptive scheduling vehicle-mounted GUI dynamic refreshing method

Abstract

The application discloses a timing covariant variable prediction perception adaptive scheduling vehicle-mounted GUI dynamic refreshing method, aiming at the problems of visual response delay, unbalanced computing power distribution, multi-source data redundant interference and the absence of safety update mechanism of the vehicle-mounted GUI system, the binding relationship of signals and GUI controls is managed through hierarchical dynamic mapping table, and the safety thereof is ensured by using an ordered summary cascade seal chain; multi-order time domain characteristics of the signals are extracted to input a lightweight prediction model, the signal values at future time points are deduced in advance, and picture pre-rendering is carried out accordingly; meanwhile, an emergency factor based on signal change rate is introduced to realize adaptive scheduling, multi-source heterogeneous signals are fused through confidence propagation, and the prediction model weight is updated online based on the residual error. The whole can effectively eliminate end-to-end display delay, reduce system load, and improve the fluency, safety and reliability of human-computer interaction of the intelligent cockpit.

Description

Technical Field

[0001] This invention belongs to the technical field of intelligent cockpit embedded systems and computer graphics, specifically involving a method for dynamic refresh of vehicle GUI based on temporal covariate prediction, perception, and adaptive scheduling. Background Technology

[0002] With the evolution of modern automotive electrical and electronic architecture (EEA) towards domain control and central computing, the instrument panel and central control screen of the smart cockpit have become key interactive interfaces for drivers to obtain core vehicle status information. However, due to the increasing complexity of signal links, existing in-vehicle GUI systems generally suffer from the following technical shortcomings:

[0003] End-to-end display latency: The physical latency of the entire signal chain, from sensor acquisition, CAN bus transmission and reception, gateway forwarding, to reception, parsing, and rendering by the graphics engine, is typically between 30ms and 70ms. Under severe dynamic conditions such as emergency braking or rapid acceleration, this latency can cause instrument pointers to appear "sticky" or their values ​​to "jump," deviating from the driver's immediate perception and impacting driving safety and the human-machine interface experience.

[0004] Static resource allocation leads to wasted computing power: Traditional solutions employ a uniform static refresh rate strategy for all signal mappings. In actual driving, water temperature, fuel level, etc., are low-frequency, slowly changing signals, while vehicle speed, engine speed, etc., are high-frequency, rapidly changing signals. Oversampling and rendering low-frequency signals unnecessarily consumes valuable MCU computing power and bus bandwidth, reducing the system's real-time response to emergency events.

[0005] Multi-source heterogeneous data conflicts: In advanced automated driving assistance systems, key state variables (such as vehicle speed) can simultaneously originate from the CAN bus, GPS, wheel speed sensors, and inertial navigation systems. Due to differences in sampling frequency, noise characteristics, and data update cycles, data inconsistencies and even contradictions often occur among these information sources, posing challenges to decision fusion.

[0006] Vulnerabilities of secure binding mechanisms: Traditional signal-to-GUI control mapping logic is usually embedded in firmware code and does not support dynamic security updates at runtime. Furthermore, when facing illegal data injection attacks targeting in-vehicle networks (such as CAN bus), existing systems lack effective signal authenticity verification and anti-tampering mechanisms, posing serious security risks. Summary of the Invention

[0007] To address the aforementioned issues, the present invention aims to provide a method for dynamic refresh of in-vehicle GUI based on temporal covariate prediction perception adaptive scheduling, thereby resolving the technical deficiencies in existing in-vehicle GUI systems, such as visual response delay, unbalanced computing power allocation, multi-source data redundancy interference, and the lack of a secure update mechanism.

[0008] To achieve the above objectives, the technical solution adopted by the present invention includes: A method for dynamically refreshing an onboard GUI based on temporal covariate prediction perception adaptive scheduling includes the following steps: S1. Configure a hierarchical dynamic mapping table in the non-volatile memory of the vehicle control unit MCU. The hierarchical dynamic mapping table is used to manage the binding relationship between at least one physical quantity to be monitored and its corresponding GUI control. The hierarchical dynamic mapping table contains multiple blocks, and each block stores at least a unique signal identifier, a memory mapping address of the GUI control, a signal parsing operator, an initial weight of the prediction model, and a block integrity certificate. S2 acquires the original signal sequence of the physical quantity to be monitored in real time through the vehicle communication interface, establishes an adaptive time-series perception window, extracts the first-order rate of change, second-order acceleration and historical fluctuation deviation value of the original signal sequence, constructs a time-domain semantic embedding feature tensor and outputs it. S3, input the temporal semantic embedding feature tensor to the lightweight timing prediction model running on the MCU side of the vehicle control unit. The lightweight timing prediction model calculates and outputs the expected values ​​of the physical quantity predictions for the first and second future moments based on the temporal semantic embedding feature tensor at the current moment. S4. In the preprocessing period before the original signal sequence is parsed and processed according to the signal parsing operator and the true value is obtained, the graphical user interface (GUI) rendering task reads the expected values ​​of the physical quantities at the first and second future moments according to the memory mapping addresses of the GUI controls recorded in the hierarchical dynamic mapping table, calls the graphics acceleration unit or graphics display controller to pre-draw the screen, generates pre-rendered screen data, and stores the pre-rendered screen data in the preset buffer video memory. S5 calculates the signal change rate of the original signal sequence in real time and calculates the emergency factor based on the signal change rate; when the emergency factor is greater than or equal to the preset dynamic threshold, the kernel preemption logic is triggered, which raises the priority of the task that parses and processes the original signal sequence according to the signal parsing operator recorded in the hierarchical dynamic mapping table to the real-time queue and shortens the response interval of the display frame synchronization pulse. S6. For the physical quantity to be monitored, obtain the current measurement value provided by at least two data sources and the error statistics of each data source in the historical period; calculate the real-time weight coefficient of each data source according to the historical error distribution, and perform weighted summation on each current measurement value according to the real-time weight coefficient to obtain the global optimal fusion estimate. S7. After the real physical signal is parsed into the real value according to the signal parsing operator, the expected value of the physical quantity prediction corresponding to the current time is obtained from the expected value of the physical quantity prediction. The residual between the real value of the real physical signal and the obtained expected value of the physical quantity prediction corresponding to the current time is calculated. When the absolute value of the residual is greater than or equal to the preset perception deviation threshold, the residual-driven forced alignment correction is performed. At the same time, according to the initial weight of the prediction model, the output layer weight of the lightweight time series prediction model is updated online through the gradient optimization algorithm.

[0009] Preferably, in S1, the block integrity certificate is a digital signature generated based on the asymmetric elliptic curve self-verification signature operator, and the hierarchical dynamic mapping table achieves tamper-proofing through an ordered digest cascaded seal chain structure; the ordered digest cascaded seal chain structure is as follows: the content hash value of the current block is concatenated with the hash value of the previous block and then hashed again to generate the link hash value of the current block, and stored in the next block.

[0010] Preferably, in S3, the model weights of the lightweight time series prediction model are processed by structured topology sparse compression and fixed-point quantization encoding so that the single inference time of the lightweight time series prediction model is less than 0.5ms and the static random access memory resources occupied do not exceed 4KB.

[0011] Preferably, in S4, the preset buffer video memory includes a three-level predictive frame asynchronous co-caching system (TPAFB). The three-level predictive frame asynchronous co-caching system (TPAFB) includes a physical frame buffer and two logical prediction buffers. The physical frame buffer is used to store the current display screen, and the two logical prediction buffers are used to alternately store the screen data pre-rendered based on the expected values ​​of the physical quantities predicted at the first and second future moments, and achieve a zero-latency smooth transition of the display output through buffer switching.

[0012] Preferably, in S5, the emergency factor The specific formula is as follows:

[0013] Where t represents the current time, This represents the rate of change of the original signal sequence at the current moment. The signal increment represents the current moment. The signal rate of change and the signal increment are both dimensionless parameters after normalization. α and β are preset positive weighting coefficients. The values ​​of α and β are both 0 to 1, and the sum of α and β is 1. Shortening the response interval of the display frame synchronization pulse specifically involves shortening the response interval. From the basic interval The compression interval is shortened as follows:

[0014] Wherein, γ is a preset sensitivity coefficient, and the lower limit of the compression interval is 5ms.

[0015] Preferably, in S5, when the emergency factor When the power consumption is below the preset low power threshold, the corresponding rendering task is attached to the low power background queue, and the rendering frame rate is switched from 60FPS to 10FPS.

[0016] Preferably, in S6, the error statistics of each data source during the historical period include the historical residual sequence and real-time error variance of each data source; Real-time weighting coefficients are calculated using the following confidence propagation method: First, based on the residual of the i-th data source at the previous time step... The channel confidence level is updated exponentially to reflect the current time. The specific formula is as follows:

[0017] Where μ is a preset forgetting factor, and the value of μ ranges from 0.1 to 1 after the residuals have been normalized. Then, based on the channel confidence at the current moment Real-time error variance with the i-th data source Calculate the normalized weight coefficient of the i-th data source at the current time. The specific formula is as follows:

[0018] Where j represents the index of all data sources participating in the fusion; Finally, the measured values ​​at the current time are weighted and summed using the normalized weight coefficients of each data source to obtain the globally optimal fusion estimate. Global optimal fusion estimate Calculated using the following formula:

[0019] in, Let be the measurement value of the i-th data source at time t.

[0020] Preferably, S6 further includes an anomaly source removal step: calculating the chi-square test statistic for the i-th data source. The specific formula is as follows:

[0021] When the chi-square test statistic When the channel confidence of the i-th data source at the current time is greater than or equal to the preset chi-square threshold, Force a zeroing process to eliminate abnormal data sources.

[0022] Preferably, in S7, the gradient optimization algorithm is a constrained sensing domain adaptive gradient collaborative optimization operator, which only fine-tunes the output layer weights of the lightweight time series prediction model; residual The calculation method is as follows: ,in, For the true value, This refers to the expected value of the physical quantity prediction corresponding to the current moment in the expected value of the physical quantity prediction. The online update method for output layer weights is as follows: ,in, This is the output layer weight matrix at the current time. This refers to the activation vectors of the output layer of a lightweight time series prediction model. Adaptive learning rate; Adaptive learning rate Attenuation is performed according to the following formula: in, ρ is the initial learning rate, and ρ is the preset decay coefficient.

[0023] A time-series covariate prediction and perception adaptive scheduling vehicle GUI dynamic refresh system is disclosed in this application.

[0024] Compared with the prior art, the advantages of the present invention are: (1) The time-series covariate prediction perception adaptive scheduling vehicle GUI dynamic refresh method of the present invention deploys a lightweight time-series prediction model on the MCU side, completes the deduction of the signal value at future time before the physical signal actually arrives, and calls the graphics acceleration unit in advance to pre-draw the screen based on the prediction result, so that the visual feedback and the driver's operation are almost synchronized, which in principle compensates for the inherent delay of the physical link and effectively eliminates the end-to-end display delay.

[0025] (2) The time-series covariate prediction perception adaptive scheduling vehicle GUI dynamic refresh method of the present invention introduces an emergency factor based on the signal change rate and an adaptive scheduling mechanism to monitor the dynamic characteristics of the signal in real time: when the signal changes drastically, the parsing task is promoted to the real-time queue and the VSYNC response interval is shortened; when the signal tends to be stable, the rendering task is automatically mounted to the low-power background queue and the rendering frame rate is switched from 60FPS to 10FPS. Simulation experiments show that under typical working conditions, the system power consumption of non-critical working conditions can be reduced by about 20%, which effectively extends the service life of electronic components.

[0026] (3) The time-series covariate prediction perception adaptive scheduling vehicle GUI dynamic refresh method of the present invention uses a multi-source heterogeneous confidence propagation fusion estimator to combine the historical error distribution of each data source with the real-time channel quality to dynamically calculate the fusion weight. Even in scenarios where GPS signals are attenuated, such as tunnels, it can still output a stable and reliable vehicle speed display. When a certain data source is abnormal, its confidence can be forcibly reduced to zero and removed by chi-square test to ensure that the instrument always provides accurate reference data.

[0027] (4) The time-series covariate prediction perception adaptive scheduling vehicle GUI dynamic refresh method of the present invention encapsulates the signal mapping table through an ordered digest cascaded sealing chain structure, and concatenates and re-hashes the content hash value of the current block with the hash value of the previous block to form a time-series associated anti-tampering chain. This structure allows OEMs to dynamically change the UI layout and signal binding relationship through a secure channel during the vehicle's operating life cycle without having to fully flash the firmware, while effectively resisting malicious tampering of the signal mapping relationship. Attached Figure Description

[0028] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the following detailed description to explain the invention, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the overall architecture of a time-series covariate prediction, perception, adaptive scheduling, vehicle GUI dynamic refresh system according to this application; Figure 2 A schematic diagram of the entire signal flow and the intervention nodes of each functional module in the vehicle-mounted GUI signal dynamic binding system; Figure 3 A flowchart illustrating the internal deduction and online weight evolution logic of a lightweight time series prediction model; Figure 4 A schematic diagram of the task queue structure and computing power allocation of the perception-driven adaptive priority scheduler under different driving conditions; Figure 5 This is a schematic diagram of the hierarchical structure and security verification link of the in-vehicle GUI signal mapping block based on the ordered digest cascaded seal chain. Detailed Implementation

[0029] The invention is not limited to the specific embodiments described below. All equivalent modifications made based on the technical solutions of this application fall within the protection scope of this invention. Unless otherwise specified, all components and devices in this invention utilize components and devices known in the prior art.

[0030] Example like Figure 1As shown, the overall architecture of the vehicle-mounted GUI dynamic refresh system provided by this invention includes the following core functional modules: a hierarchical dynamic mapping table (responsible for the secure binding management of signals and GUI controls), an edge-side perception coprocessing unit (responsible for signal feature extraction and timing prediction), a GUI rendering engine (responsible for screen pre-drawing and buffer management based on prediction results), an adaptive scheduler (responsible for dynamic adjustment of task priorities based on urgency factors), and a multi-source heterogeneous belief propagation fusion estimator (responsible for intelligent fusion of multiple redundant signals). The data flow and interaction relationships between the modules are as follows: Figure 1 As shown, the signal flow throughout the entire link and the intervention nodes of each functional module are as follows: Figure 2 As shown.

[0031] This embodiment discloses a method for dynamically refreshing an onboard GUI based on temporal covariate prediction and adaptive scheduling, including the following steps: Step S1: Configuration of hierarchical dynamic mapping table In this embodiment, a hierarchical dynamic mapping table is first configured in the non-volatile memory (such as embedded Flash) of the vehicle control unit (MCU). This mapping table is organized in a block-based hierarchical structure, with each block corresponding to the binding relationship between a physical quantity to be monitored and a GUI control.

[0032] Specifically, each block stores at least the following information: Signal ID: Used to uniquely identify a physical signal in the system; GUI control memory mapping address: points to the storage location of the graphical control corresponding to this signal in the display cache or memory; Signal parsing operators: analytical functions or algorithms used to convert raw CAN bus messages or raw data received from vehicle communication interfaces into the true values ​​of physical quantities; Initial weights for the prediction model: Initial model parameters provided for the lightweight time series prediction model; Block integrity certificate: A digital signature generated based on an asymmetric elliptic curve self-verifying signature operator (such as ECDSA) to verify the integrity and authenticity of the block data.

[0033] To enhance the mapping table's resistance to malicious tampering during runtime, this embodiment employs an ordered digest cascaded seal chain structure to protect each block in series. Specifically, the entire content of the current block is hashed to obtain a content hash value. This hash value is then concatenated with the link hash value of the previous block and hashed again to generate the link hash value of the current block, which is then stored in the next block. In this way, any tampering with the data in any block will invalidate the link hash values ​​of all subsequent blocks, thus forming a time-series, tamper-proof verification chain. The hierarchical diagram of the above structure is shown below. Figure 5As shown.

[0034] Step S2: Signal Feature Extraction and Serialization Processing After the mapping table is configured, the system can obtain the raw signal sequence of the physical quantities to be monitored (including but not limited to vehicle speed, engine speed, battery power, etc.) in real time through the vehicle communication interface (such as CAN transceiver).

[0035] To transform the raw signal into structured features usable by the prediction model, this embodiment establishes an adaptive convergent topological time-series sensing window. The length N of this window can be dynamically adjusted according to the historical fluctuations of the signal; the window is shortened when fluctuations are severe to improve response sensitivity, and extended when fluctuations are gentle to suppress noise interference.

[0036] Within this window, the system extracts three types of time-domain features: First-order rate of change: reflects the instantaneous trend of signal change, calculated by the difference ratio between adjacent sampling points; Second-order acceleration: reflects the degree of drastic change in the signal, that is, the second difference of the first-order rate of change; Historical fluctuation deviation value: reflects the degree of dispersion of the signal within the historical window, and is calculated by the standard deviation of each sampling point within the window relative to the mean.

[0037] The three features mentioned above constitute a three-dimensional column vector, namely the temporal semantic embedding feature tensor, which serves as the input to the subsequent prediction model. The adaptive adjustment of the window length is driven by historical fluctuation bias and is achieved through an exponential convergence method.

[0038] Step S3, Signal Forward-Looking Inference The temporal semantic embedding feature tensor constructed in step S2 is input into the lightweight temporal prediction model running in the MCU edge perception coprocessing unit.

[0039] In this embodiment, the prediction model employs a recursive covariance-driven phase tracking estimator, achieving lightweight deployment requirements of less than 0.5ms single inference time and no more than 4KB SRAM usage, thus adapting to the stringent computing power constraints of automotive-grade MCUs. The estimator achieves real-time estimation of the future state of the signal through the following recursive process: first, a one-step prediction is performed based on the posterior state of the previous moment; then, the prediction error covariance is propagated; next, the adaptive gain matrix is ​​calculated; then, the prediction is corrected based on the current observations to obtain the posterior state estimate; finally, the posterior covariance is updated. Through these five recursive steps, the model can output the expected values ​​of the physical quantities predicted at the first future moment (t+1) and the second future moment (t+2).

[0040] Step S4: Proactive Pre-rendering and Buffer Management During the preprocessing period before the original signal sequence is parsed and processed according to the signal parsing operator to obtain the true value, the GUI rendering task performs screen rendering in advance according to the predicted expected value output in step S3.

[0041] Specifically, the GUI rendering task reads the predicted expected values ​​for the first and second future moments based on the memory mapping addresses of the GUI controls recorded in the hierarchical dynamic mapping table, and calls the graphics acceleration unit (GPU) or graphics display controller (GDC) to pre-draw the screen. The pre-drawing content includes, but is not limited to: trigonometric function calculation of the instrument pointer deflection angle, interpolation generation of numerical gradient textures, and pixel-level calculation of the progress bar fill ratio.

[0042] To ensure that the pre-rendered image can be instantly switched to the screen in the next display cycle, this embodiment employs a three-tier predictive frame asynchronous collaborative buffer (TPAFB) system to manage the rendering buffer. This system comprises three buffer areas: a physical frame buffer to store the image currently displayed on the screen; and two logical prediction buffers to alternately store pre-rendered image data based on the predicted expectations of the first and second future timeframes. Through the rapid switching of the logical prediction buffers, the system can instantly switch the displayed content when the next VSYNC signal arrives, achieving a visually smooth transition with zero latency.

[0043] Step S5: Adaptive Scheduling Optimization The system kernel maintains a task weight table in real time to monitor the load and signal activity of each bus. The core of this step is to introduce an urgency factor to quantify the urgency of a signal. The urgency factor is determined by both the signal change rate and the signal increment, and is calculated using the following formula:

[0044] Where t represents the current time, This represents the rate of change of the original signal sequence at the current moment. The value represents the signal increment at the current moment, and α and β are preset positive weighting coefficients. The values ​​of α and β both range from 0 to 1, and their sum is 1 to ensure that the output of the emergency factor is within a uniform dimension. In this embodiment, α is preferably 0.6, and β is preferably 0.4.

[0045] In a preferred embodiment, and Both were normalized to the [0,1] interval.

[0046] When the urgency factor is greater than or equal to a preset dynamic threshold, it indicates that the current signal is in a state of drastic change and requires priority processing. The dynamic threshold can be selected in the range of 0.4 to 0.8, and is preferably 0.6 in this embodiment. At this time, the system triggers the kernel preemption logic, which raises the priority of the task of parsing the original signal sequence according to the signal parsing operator recorded in the hierarchical dynamic mapping table to the real-time queue, and at the same time dynamically shortens the response interval of the display frame synchronization pulse (VSYNC), specifically from the base interval T_base (usually 16.6ms, corresponding to a 60FPS frame rate) to the compressed interval:

[0047] Wherein, γ is a preset sensitivity coefficient, and the lower limit of the compression interval is 5ms (corresponding to the highest effective refresh rate of 200FPS).

[0048] When the emergency factor drops below the preset low-power threshold, indicating that the signal change is becoming smoother, the system automatically assigns the corresponding rendering task to the low-power background queue and switches the rendering frame rate from 60FPS to 10FPS to reduce MCU power consumption and bus usage. The task queue structure and computing power allocation of the adaptive scheduler under different driving conditions are illustrated below. Figure 4 As shown.

[0049] Step S6: Belief propagation fusion of multi-source heterogeneous signals For certain critical physical quantities (such as vehicle speed), the system may receive information from multiple data sources simultaneously, such as the CAN bus, GPS, and wheel speed sensors. Due to differences in sampling frequency, update cycle, and environmental noise characteristics, the measured values ​​from each data source may vary.

[0050] This embodiment employs a belief propagation fusion estimator to intelligently fuse multiple redundant inputs. The workflow of this fusion estimator is as follows: Based on the residual of the i-th data source at the previous time step The channel confidence level is updated exponentially to reflect the current time. The specific formula is as follows:

[0051] Wherein, μ is a preset forgetting factor, and when the residual is normalized, the value of μ ranges from 0.1 to 1; in this embodiment, μ is preferably 0.5; Then, based on the channel confidence at the current time... Real-time error variance with the i-th data source Calculate the normalized weight coefficient of the i-th data source at the current time. The specific formula is as follows:

[0052] Where j represents the index of all data sources participating in the fusion; the higher the channel confidence and the smaller the real-time error variance, the higher the weight of the data source.

[0053] Finally, the measured values ​​at the current time are weighted and summed using the normalized weight coefficients of each data source to obtain the globally optimal fusion estimate. The global optimal fusion estimate Calculated using the following formula:

[0054] in, Let be the measurement value of the i-th data source at time t.

[0055] Furthermore, this embodiment introduces a chi-square test mechanism for real-time removal of outliers. Specifically, the chi-square test statistic for the i-th data source is calculated as follows:

[0056] When the chi-square test statistic When the chi-square threshold is greater than or equal to a preset threshold, the channel confidence of the data source at the current moment is forcibly reduced to zero, causing its weight in subsequent fusion to be reduced to zero, thereby achieving automatic isolation of abnormal data sources. For example, in scenarios where GPS signals attenuate, such as in tunnels, the chi-square test statistic of the GPS source will increase significantly, its confidence will be automatically reduced, and the weight of the wheel speed sensor will be increased accordingly to ensure the continuous stability of vehicle speed display.

[0057] Step S7: Residual calibration and online evolution of the prediction model After the actual physical signal is resolved into its true value using the signal analysis operator, the system obtains the predicted expected value of the physical quantity corresponding to the current moment (i.e., the value previously predicted for the current moment) and calculates the residual between the true value and the obtained predicted expected value. The residual is... The calculation method is as follows: ,in, For the true value, The predicted expected value of the physical quantity is the predicted expected value of the physical quantity corresponding to the current time.

[0058] If the absolute value of the residual is less than the preset perception deviation threshold, it indicates that the current prediction accuracy meets the requirements, and the system directly uses the pre-rendered image corresponding to the prediction value for display output without additional processing.

[0059] The perception deviation threshold in this embodiment is 3% to 10% of the full scale of the physical quantity to be monitored, preferably 5%.

[0060] If the absolute value of the residual is greater than or equal to the preset perception bias threshold, it indicates that the prediction bias exceeds the acceptable range, and the system performs two operations: First, residual-driven forced alignment correction is performed. Within a preset time window (e.g., 300ms), the system gradually transitions the displayed value from the pre-rendered value to the real value using a smooth curve, avoiding visual interference to the driver caused by sudden changes in pointers or values.

[0061] Second, the lightweight time-series prediction model is updated online. Specifically, a gradient optimization algorithm is used to fine-tune the weights of the output layer of the prediction model. In this embodiment, the gradient optimization algorithm is a constraint-aware domain adaptive gradient collaborative optimization operator, which only performs gradient backpropagation and weight updates on the output layer without changing the hidden layers of the model, thereby meeting the strict computing power constraints of automotive-grade chips.

[0062] The online update method for output layer weights is as follows: ;in, This is the output layer weight matrix at the current time. The activation vector of the output layer of the lightweight time series prediction model. It is an adaptive learning rate.

[0063] The adaptive learning rate decays according to the following formula: in, ρ is the initial learning rate, ρ is the preset decay coefficient, and t is the current iteration step.

[0064] By gradually decaying the learning rate, the model can quickly adapt to changes in operating conditions in the early stages and tend to stabilize in the later stages, avoiding oscillations caused by over-adjustment.

[0065] Through the aforementioned online evolution mechanism, the prediction model can continuously adapt to the current driving conditions (such as highway cruising, urban congestion, etc.) and keep the prediction accuracy within an acceptable range.

[0066] This invention can be widely applied to various motor vehicles equipped with graphical user interfaces, especially in scenarios such as instrument panel displays, central control screen information displays, and head-up display systems in intelligent cockpit systems. The method proposed in this invention can be deployed on existing automotive-grade MCU hardware platforms through software upgrades, without requiring modifications to the vehicle's existing hardware architecture, thus exhibiting good compatibility and widespread application value.

[0067] To verify the technical effectiveness of the predictive perception dynamic refresh method, this embodiment constructs a vehicle speed signal computer simulation environment for simulation verification. The simulation time is 80s, the signal sampling period is 10ms, and the vehicle speed signal includes constant speed, normal acceleration, normal braking, and emergency braking conditions. The equivalent signal delay caused by vehicle communication, signal parsing, and task scheduling is assumed to be 50ms, and random noise with a standard deviation of 0.18km / h is superimposed on the measured signal. The basic refresh method directly uses the delayed measurement value to update the GUI control, while the predictive perception refresh method calculates the vehicle speed change trend based on the most recent 15 sampling points and predicts the vehicle speed for the next 50ms to generate a pre-rendered image.

[0068] Table 1. Simulation results of predictive sensing refresh

[0069] As shown in Table 1, compared with the basic delayed refresh method, the root mean square error (RMSE) of the predictive perception refresh method under all operating conditions decreased from 0.253 km / h to 0.149 km / h, a reduction of approximately 41.3%; under acceleration and braking dynamic conditions, the RMSE decreased from 0.357 km / h to 0.164 km / h, a reduction of approximately 54.1%. The mean absolute error under dynamic conditions decreased by approximately 56.6%, indicating that using short-time prediction results for pre-rendering can reduce the lag in GUI display values ​​caused by communication and parsing delays.

[0070] Further simulations were conducted on the adaptive refresh strategy. The in-vehicle GUI contained 12 controls to be refreshed, of which 3 safety-critical dynamic controls were always refreshed at 60 FPS, and the remaining 9 low-dynamic controls were refreshed at 10 FPS when the emergency factor was below the low-power threshold, reverting to 60 FPS under dynamic conditions such as vehicle acceleration and braking. In the simulated conditions, the low-dynamic state accounted for approximately 63.8% of the total runtime, and the dynamic state accounted for approximately 36.2%.

[0071] Table 2 Simulation results of adaptive refresh load

[0072] As shown in Table 2, without reducing the refresh rate of safety-critical controls, the adaptive refresh method reduces the average number of control updates from 720.0 times per second to 433.1 times per second, resulting in an equivalent reduction in rendering call load of approximately 39.8%. Therefore, adjusting the refresh frequency of low-dynamic controls according to the degree of change of the monitored physical quantity can reduce repeated parsing and redrawing under stable operating conditions while maintaining display responsiveness under dynamic operating conditions.

[0073] The above data is computer simulation data generated based on preset vehicle motion conditions and signal delay conditions. It is used to illustrate the working process and technical effects of the technical solution described in this invention. The specific performance may vary depending on the target controller, display system, communication cycle and model parameters.

[0074] The preferred embodiments of this disclosure have been described in detail above with reference to the accompanying drawings. However, this disclosure is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this disclosure, various simple modifications can be made to the technical solutions of this disclosure, and these simple modifications all fall within the protection scope of this disclosure.

[0075] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, this disclosure will not describe the various possible combinations separately.

[0076] Furthermore, the various implementation methods disclosed in this solution can be combined in any way, as long as they do not violate the spirit of this disclosure, they should also be regarded as the content invented by this disclosure.

Claims

1. A method for dynamically refreshing an onboard GUI based on temporal covariate prediction, perception, and adaptive scheduling, characterized in that: Includes the following steps: S1, Configure a hierarchical dynamic mapping table in the non-volatile memory of the vehicle control unit MCU. The hierarchical dynamic mapping table is used to manage the binding relationship between at least one physical quantity to be monitored and its corresponding GUI control. The hierarchical dynamic mapping table contains multiple blocks, and each block stores at least a unique signal identifier, a GUI control memory mapping address, a signal parsing operator, the initial weights of the prediction model, and a block integrity certificate. S2, real-time acquisition of the original signal sequence of the physical quantity to be monitored through the vehicle communication interface, establishment of an adaptive time-series perception window, extraction of the first-order rate of change, second-order acceleration and historical fluctuation deviation value of the original signal sequence, construction of time-domain semantic embedding feature tensor and output; S3, the time-domain semantic embedding feature tensor is input to the lightweight timing prediction model running on the MCU side of the vehicle control unit. The lightweight timing prediction model calculates and outputs the expected values ​​of the physical quantity predictions for the first and second future moments based on the time-domain semantic embedding feature tensor at the current moment. S4, during the preprocessing period before parsing the original signal sequence according to the signal parsing operator and obtaining the true value, the graphical user interface (GUI) rendering task reads the expected values ​​of the physical quantities at the first and second future moments according to the memory mapping addresses of the GUI controls recorded in the hierarchical dynamic mapping table, calls the graphics acceleration unit or graphics display controller to pre-draw the screen, generates pre-rendered screen data, and stores the pre-rendered screen data in a preset buffer video memory. S5, calculate the signal change rate of the original signal sequence in real time, and calculate the emergency factor based on the signal change rate; when the emergency factor is greater than or equal to a preset dynamic threshold, trigger the kernel preemption logic, upgrade the priority of the task of parsing the original signal sequence according to the signal parsing operator recorded in the hierarchical dynamic mapping table to the real-time queue, and shorten the response interval of the display frame synchronization pulse. S6, for the physical quantity to be monitored, obtain the current measurement value provided by at least two data sources and the error statistics of each data source in the historical period; The real-time weighting coefficients of each data source are calculated based on the historical error distribution, and the current measurement values ​​are weighted and summed based on the real-time weighting coefficients to obtain the global optimal fusion estimate. S7. After the real physical signal is parsed into a real value according to the signal parsing operator, the expected value of the physical quantity prediction corresponding to the current time is obtained from the expected value of the physical quantity prediction, and the residual between the real value of the real physical signal and the obtained expected value of the physical quantity prediction corresponding to the current time is calculated. When the absolute value of the residual is greater than or equal to the preset perception bias threshold, residual-driven forced alignment correction is performed. At the same time, the output layer weights of the lightweight time series prediction model are updated online using a gradient optimization algorithm based on the initial weights of the prediction model.

2. The method for dynamically refreshing the onboard GUI by temporal covariate prediction perception adaptive scheduling as described in claim 1, characterized in that, In S1, the block integrity certificate is a digital signature generated based on an asymmetric elliptic curve self-verification signature operator, and the hierarchical dynamic mapping table achieves tamper-proofing through an ordered digest cascaded seal chain structure. The ordered digest cascaded seal chain structure is as follows: the content hash value of the current block is concatenated with the hash value of the previous block and then hashed again to generate the link hash value of the current block, and then stored in the next block.

3. The method for dynamically refreshing the onboard GUI based on temporal covariate prediction, perception, and adaptive scheduling according to claim 1, characterized in that, In step S3, the model weights of the lightweight time series prediction model are processed by structured topology sparse compression and fixed-point quantization encoding to ensure that the single inference time of the lightweight time series prediction model is less than 0.5ms and the static random access memory resources occupied do not exceed 4KB.

4. The method for dynamically refreshing the onboard GUI based on temporal covariate prediction, perception, and adaptive scheduling according to claim 1, characterized in that, In S4, the preset buffer memory includes a three-level predictive frame asynchronous co-caching system TPAFB, which includes a physical frame buffer and two logical predictive buffers. The physical frame buffer is used to store the current display screen, and the two logical prediction buffers are used to alternately store the screen data pre-rendered based on the expected values ​​of the physical quantities predicted at the first and second future moments, and to achieve a smooth transition with zero latency in the display output through buffer switching.

5. The method for dynamically refreshing the onboard GUI based on temporal covariate prediction, perception, and adaptive scheduling according to any one of claims 1 to 4, characterized in that, In S5, the emergency factor The specific formula is as follows: Where t represents the current time, This represents the rate of change of the original signal sequence at the current moment. The signal increment at the current moment is represented by the signal change rate and the signal increment, which are dimensionless parameters after normalization. α and β are preset positive weighting coefficients, and the values ​​of α and β are both in the range of 0 to 1, and the sum of α and β is 1. The shortening of the response interval of the display frame synchronization pulse specifically refers to the response interval From the basic interval The compression interval is shortened as follows: Wherein, γ is a preset sensitivity coefficient, and the lower limit of the compression interval is 5ms.

6. The method for dynamically refreshing the onboard GUI based on temporal covariate prediction, perception, and adaptive scheduling according to claim 5, characterized in that, In S5, when the emergency factor When the power consumption is below the preset low power threshold, the corresponding rendering task is attached to the low power background queue, and the rendering frame rate is switched from 60FPS to 10FPS.

7. The method for dynamically refreshing the onboard GUI based on temporal covariate prediction, perception, and adaptive scheduling according to claim 6, characterized in that, In S6, the error statistics of each data source during the historical period include the historical residual sequence and real-time error variance of each data source. The real-time weighting coefficients are calculated using the following confidence propagation method: First, based on the residual of the i-th data source at the previous time step... The channel confidence level is updated exponentially to reflect the current time. The specific formula is as follows: Wherein, μ is a preset forgetting factor, and when the residual is normalized, the value of μ ranges from 0.1 to 1; Then, based on the channel confidence at the current moment... Real-time error variance with the i-th data source Calculate the normalized weight coefficient of the i-th data source at the current time. The specific formula is as follows: Where j represents the index of all data sources participating in the fusion; Finally, the measured values ​​at the current time are weighted and summed using the normalized weight coefficients of each data source to obtain the globally optimal fusion estimate. The global optimal fusion estimate Calculated using the following formula: in, Let be the measurement value of the i-th data source at time t.

8. The method for dynamically refreshing the onboard GUI based on temporal covariate prediction, perception, and adaptive scheduling according to claim 7, characterized in that, S6 further includes an anomaly source removal step: Calculate the chi-square test statistic for the i-th data source. The specific formula is as follows: When the chi-square test statistic When the channel confidence of the i-th data source is greater than or equal to the preset chi-square threshold, the channel confidence of the i-th data source at the current time is... Force a zeroing process to eliminate abnormal data sources.

9. The method for dynamically refreshing the onboard GUI based on temporal covariate prediction, perception, and adaptive scheduling according to claim 8, characterized in that, In S7, the gradient optimization algorithm is a constrained sensing domain adaptive gradient collaborative optimization operator, which only fine-tunes the output layer weights of the lightweight time series prediction model; The residual The calculation method is as follows: ,in, For the true value, The predicted expected value of the physical quantity is the predicted expected value of the physical quantity that corresponds to the current time. The online update method for the output layer weights is as follows: ,in, This is the output layer weight matrix at the current time. The activation vector of the output layer of the lightweight time series prediction model. Adaptive learning rate; The adaptive learning rate Attenuation is performed according to the following formula: in, ρ is the initial learning rate, and ρ is the preset decay coefficient.

10. A time-series covariate prediction-sensing adaptive scheduling vehicle GUI dynamic refresh system, characterized in that, The system is used to execute the time-series covariate prediction perception adaptive scheduling vehicle GUI dynamic refresh method as described in any one of claims 1 to 9.

Images