A method and system for intelligent network system time synchronization of adaptive kalman filtering

Abstract

This invention provides a method and system for time synchronization in intelligent connected systems using adaptive Kalman filtering. The method first establishes a three-dimensional state-space model containing time deviation, frequency deviation, and drift rate; then executes the prediction and update steps of Kalman filtering; next, it adaptively adjusts the filtering parameters and detects and recovers filter divergence; finally, it fuses multiple time sources—GNSS, NTP, and local hardware RTC—through federated Kalman filtering; and it also optimizes the computational complexity of the filtering algorithm. The system adopts a four-layer architecture—application layer, time service layer, driver layer, and hardware layer—integrating multiple modules to achieve integrated time source acquisition, evaluation, fusion, and calibration. This invention improves time synchronization accuracy and robustness, adapts to complex in-vehicle environments, ensures the continuity of time synchronization, meets the real-time requirements of in-vehicle terminals, and provides reliable time assurance for intelligent connected applications such as V2X communication.

Description

Technical Field

[0001] This invention relates to the field of intelligent connected vehicle technology, and specifically to a method and system for time synchronization of intelligent connected vehicle systems using adaptive Kalman filtering. Background Technology

[0002] Intelligent connected terminals are a core component of the Internet of Vehicles (IoV). They not only need to adapt to the complex road and network environment in China, but also need to cope with different usage scenarios internationally. The accuracy and timeliness of data are the key to the reliable operation of intelligent connected systems. Precise terminal time is the foundation for ensuring accurate data interaction between the vehicle and the cloud, and can provide accurate and effective timestamp data for application scenarios such as V2X communication, remote diagnostics, data recording, and OTA upgrades.

[0003] Currently, intelligent connected terminals mainly rely on three time sources for time synchronization: GNSS PNT time, network time NTP, and local hardware time RTC. Among them, GNSS PNT time has the highest accuracy and is the first choice for terminal time synchronization. However, this time source is greatly affected by the signal environment. In obstructed scenarios such as urban canyons, under overpasses, and tunnels, GNSS signals attenuate severely, and time accuracy drops rapidly. NTP network time is the next most accurate, but it suffers from transmission delay issues in mobile vehicle terminal scenarios. At the same time, its time accuracy is affected by factors such as network coverage quality and network bandwidth, making it unusable in weak or no network environments. Local hardware RTC time is the terminal's local clock and does not depend on external signals and networks. However, it is affected by factors such as ambient temperature, hardware component precision, and circuit design, resulting in significant time drift. Improving the accuracy of hardware time requires the use of high-precision crystal oscillators and other components, which significantly increases the terminal's hardware cost.

[0004] In summary, existing time synchronization solutions for intelligent connected terminals suffer from poor robustness of single time sources, ineffective fusion of multiple time sources, and time asynchrony or errors in complex environments. These issues fail to meet the continuous time accuracy requirements of intelligent connected systems. Therefore, there is an urgent need for a method that can adapt to complex in-vehicle environments, integrate the advantages of multiple time sources, and improve time synchronization accuracy under the same hardware conditions. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method and system for time synchronization of intelligent connected systems using adaptive Kalman filtering.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for time synchronization of an intelligent connected system using adaptive Kalman filtering, comprising the following steps: S1: Establish a state-space model for time synchronization of intelligent connected terminals: Define the state vector: x(k) = [Δt(k), Δf(k), Δḟ(k)]ᵀ; Where Δt(k) is the time deviation at time k, Δf(k) is the frequency deviation at time k, and Δḟ(k) is the frequency drift rate at time k; The state transition equation is derived as follows: x(k+1)=F(k)・x(k)+w(k), where F(k) is a 3×3 state transition matrix, x(k) is the state vector at time k, w(k) is the process noise vector and satisfies w(k)~N(0,Q), and Q is the process noise covariance matrix. The state transition matrix F is Δt is the sampling interval time; And the observation equation: z(k)=H(k)・x(k)+v(k), where z(k) is the observation value at time k; H(k) is a 1×3 observation matrix; v(k) is the observation noise and satisfies v(k)~N(0,R); The observation matrix H =

[100] .

[0007] S2: Perform the prediction step based on the Kalman filter algorithm; Based on the state transition equation and the state estimate x^(k-1|k-1) from the previous time step, we obtain the state prediction x^(k|k-1) = F(k-1)·x^(k-1|k-1) at time k. Based on the covariance matrix P(k-1|k-1) from the previous time step, combined with the state transition matrix F and the process noise covariance matrix, the predicted covariance value at time step k is obtained: P(k|k-1)=F(k-1)・P(k-1|k-1)・F(k-1)ᵀ+Q(k-1), where Q is the process noise covariance matrix.

[0008] S3: Perform update steps based on the Kalman filter algorithm; Calculate the Kalman gain K(k), update the state estimate x(k|k) based on the observed value z(k), and update the covariance matrix P(k|k); The update step is performed based on the Kalman filter algorithm. First, the Kalman gain K(k) is calculated, with the formula: K(k) = P(k|k-1)・H(k)ᵀ・[H(k)・P(k|k-1)・H(k)ᵀ+R(k)] -1 ; Based on the observed value z(k) and the predicted state value x^(k|k-1) at time k, the state estimate at time k is updated as follows: x^(k|k)=x^(k|k-1)+K(k)・[z(k)-H(k)・x^(k|k-1)]; To ensure the positive definiteness and numerical stability of the covariance matrix, the covariance matrix is ​​updated using the Joseph form: P(k|k)=[IK(k)・H]・P(k|k-1)・[IK(k)・H]ᵀ+K(k)・R・K(k)ᵀ, where I is the identity matrix.

[0009] S4: Adaptively adjust the parameters of the Kalman filter; This includes adaptive estimation of the measurement noise matrix R based on the innovation adjustment process, noise matrix Q based on the innovation adjustment process, and filtering divergence detection and recovery.

[0010] S5: Integrates multiple time sources such as GNSS, NTP, and local hardware RTC to construct multi-sensor observation equations and corresponding measurement noise matrices; Federated Kalman filtering is used for information distribution and fusion to obtain the optimal time estimate and achieve time synchronization of the intelligent connected system.

[0011] Preferably, in step S2, when the sampling interval Δt = 1 second, the state transition matrix F is: The process noise covariance matrix Q is: ,in The variance of the time-biased white noise. The variance of the white noise with frequency deviation. The frequency drift rate is the white noise variance; when the sampling interval Δt = 1 second, the process noise covariance matrix Q is... .

[0012] Preferably, in step S4, the specific process of adaptive innovation estimation is as follows: Calculate the innovation sequence ε(k)=z(k)-H・x^(k|k-1); Calculate the theoretical value of the new information covariance S theory (k) = H・P(k|k-1)・Hᵀ+R and the actual estimated value S actual (k)=(1 / N)・∑[i=k-N+1tok]ε(i)・ε(i)ᵀ, where N is the statistical window; When |S actual -S theory When the value exceeds the set threshold, update R to R. new =R old •(S) actual / S theory ).

[0013] Preferably, in step S4, the adaptive adjustment formula for the process noise matrix Q is: Q adaptive =Q nominal •α(k), Where the fitness factor α(k) = tr(S) actual) / tr(S theory ), where tr is the trace operation of a matrix; The condition for detecting filter divergence is tr(P(k|k))>M・tr(P(0|0)), where M is a set threshold. The divergence recovery method is P(k|k)=λ・P(k|k), where λ>1.

[0014] Preferably, in step S5, the multi-sensor observation equation is: = + ; where z GNSS、 z NTP、 z LOCAL These are observations from three different time sources; v GNSS、 v NTP、 v LOCAL These represent the corresponding observation noise; the corresponding measurement noise matrix R is... Typical values ​​are set according to the accuracy of each time source: where R GNSS =1×10 -16 (10ns precision), R NTP =1×10 -12 (1μs precision), R LOCAL =1×10 -10 (10μs accuracy).

[0015] Preferably, in step S5, the information allocation strategy of the federated Kalman filter is β. GNSS +β NTP +β LOCAL =1, β is the information allocation coefficient; Local filter information matrix Y i =β i •P -1 (k|k-1), local measurement noise matrix R i =R i / β i ; During information fusion, the global covariance matrix is ​​calculated first. The information fusion formula is P. -1 (k|k)=∑Y i Then calculate the global state estimate, x^(k|k)=P(k|k)・∑[Y i ・x i ^(k|k)], where i is the GNSS, NTP, or LOCAL time source; finally, the optimal time estimate is obtained, and the time synchronization of the intelligent connected system is realized.

[0016] Ideally, the computational complexity of the Kalman filter is optimized. For scalar observation scenarios, the Kalman gain is simplified to K(k) = P(k|k-1)・H / (H・P(k|k-1)・H+R), avoiding matrix inversion operations and improving computational efficiency. At the same time, Cholesky decomposition is used to implement square root filtering, decomposing the covariance matrix into P(k|k-1) = S(k|k-1)・S(k|k-1)ᵀ, ensuring that the P matrix is ​​always a positive definite matrix and guaranteeing the positive definiteness of the covariance matrix P.

[0017] An adaptive Kalman filter intelligent connected system time synchronization system, the system is a TBOX time synchronization system, which includes, from top to bottom, an application layer, a time service layer, a driver layer and a hardware layer; The application layer includes V2X communication, data recording, remote diagnostics, and OTA upgrade modules. The application layer is a time synchronization application scenario. The application layer is the final application scenario for time synchronization, receiving precise timestamps output by the time service layer to ensure the data accuracy and timeliness of each application. The time service layer includes a time manager, which integrates a time fusion algorithm module, a quality assessment module, a source switching control module, and a calibration algorithm module to realize multi-time source fusion, quality assessment, source switching, and time calibration. The time service layer is the core control layer of the system. The time fusion algorithm module incorporates a Kalman filter algorithm, an adaptive parameter adjustment module, and a multi-source fusion algorithm to achieve filtering and fusion of multiple time sources. The quality assessment module performs real-time assessments of the accuracy, signal quality, and network status of the three time sources (GNSS, NTP, and local hardware) and outputs the reliability level of each time source. The source switching control module automatically switches and adjusts the priority of time sources based on the quality assessment results and filtering status. The calibration algorithm module performs real-time calibration of the system clock based on the optimal time estimate output by the time fusion algorithm module. The driver layer includes GNSS driver, NTP client, crystal oscillator driver, and CAN driver modules, enabling communication between the hardware layer and the time service layer. As a communication bridge between the hardware layer and the time service layer, the driver layer drives and acquires data from each hardware module, converting the time source data from the hardware layer into digital signals that can be processed by the time service layer.

[0018] The hardware layer includes a GNSS module, a communication module, a crystal oscillator, and a CAN controller module. The hardware layer provides the system with three time sources: GNSS, NTP, and local hardware. The hardware layer provides the system with a physical layer time source and communication interface. The GNSS module collects PNT time, the communication module establishes network connection and obtains NTP time, the crystal oscillator provides local hardware RTC time, and the CAN controller enables communication on the vehicle's CAN bus.

[0019] Preferably, the time fusion algorithm module incorporates a Kalman filter algorithm, an adaptive parameter adjustment module, and a multi-source fusion algorithm. The quality assessment module evaluates the accuracy and reliability of GNSS, NTP, and local hardware time sources. The source switching control module automatically switches time sources based on the quality assessment results and filtering status. The calibration algorithm module calibrates the system clock based on the optimal time estimate.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) By establishing a three-dimensional state space model that includes time deviation, frequency deviation and frequency drift rate, the invention can accurately characterize the dynamic characteristics of the clock of the intelligent connected terminal. Combined with the Kalman filter algorithm, the optimal estimation of time deviation is achieved, which greatly improves the accuracy of time synchronization under the same hardware conditions and effectively solves the problem of the decrease in accuracy of a single time source in complex scenarios.

[0021] (2) The present invention introduces an adaptive parameter adjustment mechanism, which dynamically adjusts the process noise matrix Q and the measurement noise matrix R through the statistical characteristics of the information, and performs filter divergence detection and recovery, thereby improving the robustness and adaptability of Kalman filtering and being able to adapt to the dynamic changes of the vehicle environment.

[0022] (3) The present invention uses multi-sensor Kalman filtering combined with federated Kalman filtering to realize intelligent fusion of multiple time sources such as GNSS, NTP and local hardware RTC, giving full play to the advantages of each time source, and automatically switching to a reliable time source in scenarios such as poor GNSS signal and network interruption, so as to ensure the continuity and stability of time synchronization.

[0023] (4) The present invention optimizes the computational complexity of Kalman filtering. By simplifying Kalman gain under scalar observation and using square root filtering of Cholesky decomposition, the computational efficiency is improved while ensuring the filtering accuracy, thus meeting the real-time requirements of the vehicle terminal.

[0024] (5) The TBOX time synchronization system of the present invention adopts a layered architecture design. Each layer of modules has independent functions and works together to realize the integration of time source acquisition, quality assessment, filtering fusion and clock calibration, ensuring data consistency between the vehicle and the cloud, and providing reliable time guarantee for intelligent connected applications such as V2X communication, remote diagnosis and data recording. Attached Figure Description

[0025] Figure 1 This is a schematic diagram illustrating the principle of the method of the present invention. Detailed Implementation

[0026] To provide a further understanding of the purpose, structure, features, and functions of the present invention, detailed descriptions are provided below with reference to specific embodiments.

[0027] Please refer to the reference. Figure 1 This embodiment uses the vehicle-mounted TBOX intelligent connected terminal as the application carrier, and sets the sampling interval Δt to 1 second. It elaborates on the time synchronization method and system implementation process of the intelligent connected system based on Kalman filtering in complex urban road scenarios (including GNSS signal obstruction, network signal fluctuation, no network, weak network, etc.), and verifies the time synchronization accuracy and robustness of the present invention in complex vehicle environments.

[0028] The TBOX time synchronization system used in this embodiment has a four-layer architecture, from top to bottom: application layer (V2X communication, data recording, remote diagnostics, OTA upgrade), time service layer (time manager integrating time fusion algorithm, quality assessment, source switching control, and calibration algorithm), driver layer (GNSS driver, NTP client, crystal oscillator driver, CAN driver), and hardware layer (GNSS module, communication module, crystal oscillator, CAN controller). The system provides three time sources for time synchronization: GNSS, NTP, and local hardware RTC. Each layer works together to complete the entire process of time data acquisition, fusion, and calibration.

[0029] I. Parameter initialization before implementation: According to the Kalman filter algorithm requirements of this invention, the state vector, covariance matrix, noise matrix, adaptive adjustment parameters, and federated Kalman filter information allocation coefficients are initialized. All parameters are uniformly set to international standard units. The specific initialization values ​​are as follows: The initial state estimate is x^(0|0)=[0,0,0]ᵀ, meaning that the initial time deviation, frequency deviation, and frequency drift rate are all 0. Initial covariance matrix: P(0|0) = These correspond to the initial uncertainties of time, frequency, and frequency drift rate, respectively. Process noise covariance matrix: Q= GNSS measurement noise matrix R GNSS =1×10 -16 NTP measurement noise matrix R NTP =1×10 -12 Local hardware measurement noise matrix R LOCAL =1×10 -10 ; Adaptive parameter thresholds: the innovation covariance deviation threshold is 0.1, the filter divergence judgment threshold is M=100, and the divergence recovery coefficient is λ=10; Federal Kalman filter information allocation coefficients: β GNSS =0.6, β NTP =0.3, β LOCAL =0.1 (can be dynamically adjusted based on quality assessment results).

[0030] Simultaneously, the core matrix parameters are determined: when the sampling interval Δt = 1 second, the state transition matrix F = ; Observation matrix H=[1,0,0].

[0031] II. Implementation of Kalman Filter Core Process: In this embodiment, the Kalman filter algorithm is executed according to the process of state-space model establishment → prediction step → update step → adaptive parameter adjustment, completing one iteration calculation every 1 second. The specific implementation process is as follows: Define the state vector: x(k) = [Δt(k), Δf(k), Δḟ(k)]ᵀ; Where Δt(k) is the time deviation at time k, Δf(k) is the frequency deviation at time k, and Δḟ(k) is the frequency drift rate at time k; Establish the state transition equation: x(k+1)=F(k)・x(k)+w(k), where F(k) is a 3×3 state transition matrix, x(k) is the state vector at time k; w(k) is the process noise vector and satisfies w(k)~N(0,Q), where Q is the process noise covariance matrix; Establish the observation equation: z(k)=H(k)・x(k)+v(k), where z(k) is the observation value at time k; H(k) is a 1×3 observation matrix; v(k) is the observation noise and satisfies v(k)~N(0,R).

[0032] Based on the state estimate x^(k-1|k-1) and covariance matrix P(k-1|k-1) from the previous time step (k-1), calculate the state prediction x^(k|k-1) and covariance prediction P(k|k-1) at time k: State prediction: x^(k-1|k-1)=F(k-1)·x^(k-1|k-1); Expand the calculation: Δt^(k|k-1)=Δt^(k-1|k-1)+Δf^(k-1|k-1)+0.5·Δḟ^(k-1|k-1); Δf^(k|k-1)=Δf^(k-1|k-1)+Δḟ^(k-1|k-1); Δḟ^(k|k-1)=Δḟ^(k-1|k-1); Covariance prediction: P(k|k-1)=F(k-1)・P(k-1|k-1)・F(k-1)ᵀ+Q(k-1); Detailed calculation of matrix multiplication: P(k|k-1) = The calculation is performed according to the matrix multiplication rules to obtain the covariance prediction matrix at time k; the first iteration (k=1) is used. Due to the initial state x^(0∣0)=[0,0,0] T Then x^(1∣0)=F⋅x^(0∣0)=[0,0,0] T ; Covariance prediction calculation yields: P(1|0) = .

[0033] Based on the observations z(k) from each time source at time k, the Kalman gain K(k) and the innovation sequence y(k) are calculated sequentially to update the state estimate and covariance matrix. The covariance update uses the Joseph form to ensure numerical stability and positive definiteness. Innovation sequence: y(k)=z(k)-H·x^(k|k-1)=z(k)-Δt^(k|k-1); Innovation covariance: S(k) = H·P(k|k-1)·Hᵀ + R = P 11 (k|k-1)+R; Kalman gain: K(k) = P(k|k-1)·Hᵀ·[H·P(k|k-1)·Hᵀ+R] -1 Since this embodiment is a scalar observation, it is simplified to: K(k)=P(k|k-1)·H / (H·P(k|k-1)·H+R), avoiding matrix inversion; State update: x^(k|k) = x^(k|k-1) + K(k)·y(k); Covariance update: P(k|k)=[IK(k)·H]·P(k|k-1)·[IK(k)·H]ᵀ+K(k)·R·K(k)ᵀ, where I is a 3rd order identity matrix; In the first iteration (k=1), assume the observation value z(1) acquired by the GNSS module is 50×10 -9 If s (50ns time deviation), then: Innovation sequence: y(1) = 50 × 10 −9 −0 = 50 × 10 −9 s; Innovation covariance: S(1) = P 11 (1∣0)+R GNSS ≈1×10 −12 ; Kalman gain: K(1) = ; State update: x^(1|1)= That is, at k=1, the time deviation is 50ns, the frequency deviation is 0.005ppb, and the frequency drift rate is 0. The covariance matrix P(1∣1) is updated in Joseph form to obtain the covariance estimate at time k=1.

[0034] Based on the statistical characteristics of the innovation sequence, the measurement noise matrix R and the process noise matrix Q are dynamically adjusted, and the filter divergence state is detected in real time. If divergence occurs, a recovery operation is performed. Specific implementation details are as follows: Adaptive estimation adjustment of innovation R: Calculation of innovation sequence: ε(k)=z(k)-H・x^(k|k-1); Theoretical value of new covariance: S theory (k)=H・P(k|k-1)・Hᵀ+R; Actual estimated value: S actual (k)==(1 / N)・∑[i=k-N+1tok]ε(i)・ε(i)ᵀ; When |S actual -S theory When the value is greater than the set threshold, Update R to R new =R old •(S) actual / S theory ); Calculate the fitness factor α(k) = tr(S) actual ) / tr(S theory ); tr is the trace operation of a matrix; Then the noise matrix Q of the adaptive process adaptive =Q nominal •α(k); Calculate the trace of the covariance matrix tr(P(k∣k)). If tr(P(k∣k))>100・tr(P(0∣0)) (set threshold), If the filter is determined to be diverging, the recovery operation P(k∣k)=10・P(k∣k) (λ=10) is executed.

[0035] In this embodiment, when the vehicle travels under the overpass, the GNSS signal attenuates, the observation error increases, and the deviation between the actual and theoretical values ​​of the innovation covariance exceeds 0.1. The system automatically increases the RGNSS and reduces the weight of the GNSS time source. When the covariance matrix trace increases sharply during the filtering process, divergence recovery is triggered to ensure the stability of the filtering.

[0036] Multi-time source fusion implementation: This embodiment employs a multi-sensor Kalman filter + federated Kalman filter approach to achieve intelligent fusion of three time sources: GNSS, NTP, and local hardware RTC, fully leveraging the advantages of each time source. The specific implementation process is as follows: Constructing multi-sensor observation equations: = + ; The corresponding multi-sensor measurement noise matrix: The quality assessment module performs real-time evaluation of the signal quality, accuracy, and network status of the three time sources, and outputs the reliability level. In this embodiment, the initial information allocation coefficient β... GNSS =0.6、β NTP =0.3、β LOCAL =0.1; When GNSS signals are blocked, the quality assessment module will adjust β. GNSS Adjust to 0.1, β NTP Adjust to 0.6, β LOCAL Adjusting it to 0.3 enables dynamic weight adjustment; Calculate the local filter information matrix Y based on the allocation coefficients. i =β i •P -1 (k|k-1); Local measurement noise matrix R i =R i / β i ; Finally, multi-source information fusion: First, calculate the global covariance matrix: P -1 (k|k-1)=∑Y i Then calculate the global state estimate: x^(k|k)=P(k|k)・∑[Y i ・x i ^(k|k)]; Finally, the optimal time deviation is extracted from the global state estimate to obtain the precise time of the current moment, thus completing the fusion of multiple time sources.

[0037] The present invention has been described in the above-described embodiments; however, these embodiments are merely examples for implementing the present invention. It must be noted that the disclosed embodiments do not limit the scope of the present invention. Conversely, any modifications and refinements made without departing from the spirit and scope of the present invention are within the scope of patent protection of the present invention.

Claims

1. A method for time synchronization in an intelligent connected system using adaptive Kalman filtering, characterized in that: Includes the following steps: S1: Establish a state-space model for time synchronization of intelligent connected terminals: Define the state vector: x(k) = [Δt(k), Δf(k), Δḟ(k)]ᵀ; Where Δt(k) is the time deviation at time k, Δf(k) is the frequency deviation at time k, and Δḟ(k) is the frequency drift rate at time k; The state transition equation is derived as follows: x(k+1)=F(k)・x(k)+w(k), where F(k) is a 3×3 state transition matrix, x(k) is the state vector at time k, w(k) is the process noise vector and satisfies w(k)~N(0,Q), and Q is the process noise covariance matrix. And the observation equation: z(k)=H(k)・x(k)+v(k), where z(k) is the observation value at time k; H(k) is a 1×3 observation matrix; v(k) is the observation noise and satisfies v(k)~N(0,R); The state transition matrix F is ; Observation matrix H = [100]; S2: Perform the prediction step based on the Kalman filter algorithm; The predicted state value x^(k|k-1) at time k is obtained based on the state transition equation; x^(k|k-1)=F(k-1)·x^(k-1|k-1); The covariance prediction value at time k is obtained according to the covariance prediction formula P(k|k-1)=F(k-1)・P(k-1|k-1)・F(k-1)ᵀ+Q(k-1), where Q is the process noise covariance matrix. S3: Perform update steps based on the Kalman filter algorithm; Calculate the Kalman gain K(k), update the state estimate x(k|k) based on the observed value z(k), and update the covariance matrix P(k|k); S4: Adaptively adjust the parameters of the Kalman filter; This includes adaptive estimation of the measurement noise matrix R based on the innovation adjustment process, noise matrix Q based on the innovation adjustment process, and filtering divergence detection and recovery. S5: Integrates multiple time sources such as GNSS, NTP, and local hardware RTC to construct multi-sensor observation equations and corresponding measurement noise matrices; Federated Kalman filtering is used for information distribution and fusion to obtain the optimal time estimate and achieve time synchronization of the intelligent connected system.

2. The method for time synchronization of an intelligent connected system using adaptive Kalman filtering as described in claim 1, characterized in that: In step S2, when the sampling interval Δt = 1 second, the state transition matrix F is: ; The process noise covariance matrix Q is ,in The variance of the time-biased white noise. The variance of the white noise with frequency deviation. The frequency drift rate is the white noise variance; when the sampling interval Δt = 1 second, the process noise covariance matrix Q is... .

3. The method for time synchronization of an intelligent connected system using adaptive Kalman filtering as described in claim 1, characterized in that: In step S3, the formula for calculating the Kalman gain is: K(k)=P(k|k-1)・H(k)ᵀ・[H(k)・P(k|k-1)・H(k)ᵀ+R(k)] -1 ; The state update formula is: x(k|k)=x(k-1|k-1)+K(k)・[z(k)-H(k)・x(k|k-1)]; The covariance update adopts the Joseph form: P(k|k)=[IK(k)・H]・P(k|k-1), where I is the identity matrix.

4. The method for time synchronization of an intelligent connected system using adaptive Kalman filtering as described in claim 1, characterized in that: In step S4, the specific process of adaptive innovation estimation is as follows: Calculate the innovation sequence ε(k)=z(k)-H・x(k|k-1); Calculate the theoretical value of the new information covariance S theory (k) = H・P(k|k-1)・Hᵀ+R and the actual estimated value S actual (k)=(1 / N)・∑[i=k-N+1tok]ε(i)・ε(i)ᵀ, where N is the statistical window; When |S actual -S theory When the value exceeds the set threshold, update R to R. new =R old •(S) actual / S theory ).

5. The method for time synchronization of an intelligent connected system using adaptive Kalman filtering as described in claim 1, characterized in that: In step S4, the adaptive adjustment formula for the process noise matrix Q is: Q adaptive =Q nominal •α(k), Where the fitness factor α(k) = tr(S) actual ) / tr(S theory ), where tr is the trace operation of a matrix; The condition for detecting filter divergence is tr(P(k|k))>M・tr(P(0|0)), where M is a set threshold. The divergence recovery method is P(k|k)=λ・P(k|k), where λ>1.

6. The method for time synchronization of an intelligent connected system using adaptive Kalman filtering as described in claim 1, characterized in that: In step S5, the multi-sensor observation equation is: = + ; The corresponding measurement noise matrix R is , Where R GNSS =1×10 -16 R NTP =1×10 -12 R LOCAL =1×10 -10 .

7. The method for time synchronization of an intelligent connected system using adaptive Kalman filtering as described in claim 1, characterized in that: In step S5, the information allocation strategy of the federated Kalman filter is β. GNSS +β NTP +β LOCAL =1, β is the information allocation coefficient; Local filter information matrix Y i =β i •P -1 (k|k-1), local measurement noise matrix R i =R i / β i The information fusion formula is P. -1 (k|k)=∑Y i ; x^(k|k)=P(k|k)・∑[Y i ・x i [(k|k)], where i is the GNSS, NTP, or LOCAL time source.

8. A method for time synchronization of an intelligent connected system using adaptive Kalman filtering as described in any one of claims 1-7, characterized in that: The computational complexity of Kalman filtering is optimized. For scalar observation scenarios, the Kalman gain is simplified to K(k)=P(k|k-1)・H / (H・P(k|k-1)・H+R), avoiding matrix inversion operations. Cholesky decomposition is used to implement square root filtering, ensuring the positive definiteness of the covariance matrix P.

9. A time synchronization system for an intelligent connected system using adaptive Kalman filtering, characterized in that, The system is a TBOX time synchronization system, which includes, from top to bottom, an application layer, a time service layer, a driver layer, and a hardware layer. The application layer includes V2X communication, data recording, remote diagnostics, and OTA upgrade modules. The application layer is for time synchronization application scenarios. The time service layer includes a time manager, which integrates a time fusion algorithm module, a quality assessment module, a source switching control module, and a calibration algorithm module to realize multi-time source fusion, quality assessment, source switching, and time calibration. The driving layer includes GNSS driver, NTP client, crystal oscillator driver, and CAN driver modules, enabling communication between the hardware layer and the time service layer; The hardware layer includes a GNSS module, a communication module, a crystal oscillator, and a CAN controller module. The hardware layer provides the system with three time sources: GNSS, NTP, and local hardware.

10. The time synchronization system for an intelligent connected system with adaptive Kalman filtering as described in claim 9, characterized in that: The time fusion algorithm module incorporates a Kalman filter algorithm, an adaptive parameter adjustment module, and a multi-source fusion algorithm. The quality assessment module evaluates the accuracy and reliability of GNSS, NTP, and local hardware time sources. The source switching control module automatically switches time sources based on the quality assessment results and filtering status. The calibration algorithm module calibrates the system clock based on the optimal time estimate.

Images