Navigation methods, devices, equipment, media and products

Abstract

This application discloses a navigation method, apparatus, device, medium, and product, relating to the field of satellite navigation. The method includes acquiring first and second navigation spatiotemporal data; using a pre-constructed spatiotemporal encoder to extract features from the first and second navigation spatiotemporal data, obtaining a first and second spatiotemporal encoded feature matrix; determining a first and a second confidence level corresponding to the first and second spatiotemporal encoded feature matrices, respectively; fusing the first and second spatiotemporal encoded feature matrices based on the first and second confidence levels to obtain a fused feature matrix; and determining navigation data based on the first, second, and fused feature matrices. This application solves the technical problem of low fusion accuracy in the prior art when fusing two satellite navigation data using fixed fusion weights.

Description

Technical Field

[0001] This application relates to the field of satellite navigation technology, and in particular to a navigation method, apparatus, device, medium and product. Background Technology

[0002] The BeiDou Navigation Satellite System (BDS), as a global satellite navigation system, boasts advantages such as high precision, high reliability, and global coverage, and has been widely applied in various military and civilian navigation scenarios. However, in complex environments (such as urban canyons, dense forests, and areas with electromagnetic interference), a single BeiDou navigation module is prone to signal blockage, interference, or anomalies, leading to decreased navigation accuracy and navigation interruptions, thus failing to meet the requirements for high-reliability navigation.

[0003] To address these issues, existing technologies often employ a dual-BeiDou redundancy design. This involves using two independent BeiDou navigation modules to collect data in parallel, achieving redundancy backup. If one module malfunctions, the system switches to the other, which is functioning normally. However, the data collected in parallel by the two BeiDou modules needs to be fused. Traditional fusion methods often use fixed weighted averages, where the fusion weights are fixed and cannot be dynamically adjusted based on the real-time operating status of the BeiDou modules. When the navigation environment undergoes sudden changes (such as sudden electromagnetic interference), the fusion accuracy drops significantly.

[0004] In summary, existing technologies employ fixed fusion weights when fusing two satellite navigation data sets, resulting in low fusion accuracy. Summary of the Invention

[0005] The purpose of this application is to provide a navigation method, apparatus, device, medium, and product that can solve the technical problem of low fusion accuracy when fusing two satellite navigation data using fixed fusion weights in the prior art.

[0006] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a navigation method, including: First navigation spatiotemporal data and second navigation spatiotemporal data are acquired, wherein the first navigation spatiotemporal data and the second navigation spatiotemporal data are respectively collected by two independent satellite navigation modules; Using a pre-constructed spatiotemporal encoder, features are extracted from the first and second navigation spatiotemporal data to obtain the first and second spatiotemporal coding feature matrices. Determine the first confidence level and the second confidence level corresponding to the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix, respectively; Based on the first confidence level and the second confidence level, feature fusion is performed on the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix to obtain the fused feature matrix; Navigation data is determined based on the first spatiotemporal coding feature matrix, the second spatiotemporal coding feature matrix, and the fused feature matrix.

[0007] Optionally, the construction of the spatiotemporal encoder includes: determining an initial encoder based on a selective state-space model; initializing the parameters in the initial encoder to obtain the spatiotemporal encoder, wherein the parameters in the initial encoder include at least state-space parameters, convolution parameters, time-step parameters, dimension expansion parameters, activation function, and input dimension.

[0008] Optionally, determining the first confidence level and the second confidence level corresponding to the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix respectively includes: determining the variance of the features corresponding to each time step in the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix to obtain a first variance sequence and a second variance sequence; normalizing the first variance sequence and the second variance sequence to obtain a first confidence level sequence and a second confidence level sequence; and determining the first confidence level and the second confidence level respectively based on the first confidence level sequence and the second confidence level sequence.

[0009] Optionally, based on the first confidence level and the second confidence level, feature fusion is performed on the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix to obtain a fused feature matrix. Specifically, this includes: determining the first weight corresponding to the first spatiotemporal coding feature matrix and the second weight corresponding to the second spatiotemporal coding feature matrix based on the first confidence level and the second confidence level; and performing weighted fusion on the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix based on the first weight and the second weight to obtain a fused feature matrix.

[0010] Optionally, navigation data is determined based on the first spatiotemporal coding feature matrix, the second spatiotemporal coding feature matrix, and the fused feature matrix. Specifically, this includes: performing anomaly detection on the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix to obtain anomaly detection results; determining target features from the first spatiotemporal coding feature matrix, the second spatiotemporal coding feature matrix, and the fused feature matrix based on the anomaly detection results; and determining navigation data based on the target features.

[0011] Optionally, based on the anomaly detection result, the target feature is determined from the first spatiotemporal coding feature matrix, the second spatiotemporal coding feature matrix, and the fused feature matrix. Specifically, this includes: in response to the anomaly detection result indicating that the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix are normal, the fused feature matrix is ​​used as the target feature; or, in response to the anomaly detection result indicating that one of the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix is ​​abnormal, the normal matrix in the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix is ​​used as the target feature; or, in response to the anomaly detection result indicating that both the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix are abnormal, historical fused features are obtained, and a linear prediction algorithm is used to predict the target feature based on the historical fused features.

[0012] Secondly, this application provides a navigation device, comprising: The system comprises the following modules: an acquisition module for acquiring first and second navigation spatiotemporal data, wherein the first and second navigation spatiotemporal data are collected by two independent navigation modules; an extraction module for extracting features from the first and second navigation spatiotemporal data using a pre-built spatiotemporal encoder to obtain a first and second spatiotemporal coding feature matrix; a first determination module for determining the first and second confidence levels corresponding to the first and second spatiotemporal coding feature matrices, respectively; a fusion module for fusing features from the first and second spatiotemporal coding feature matrices based on the first and second confidence levels to obtain a fused feature matrix; and a second determination module for determining navigation data based on the first, second, and fused feature matrices.

[0013] Optionally, the extraction module is also used to: determine an initial encoder based on a selective state-space model; initialize the parameters in the initial encoder to obtain a spatiotemporal encoder, wherein the parameters in the initial encoder include at least state-space parameters, convolution parameters, time-step parameters, dimension expansion parameters, activation function, and input dimension.

[0014] Optionally, the first determining module is further configured to: determine the variance of the features corresponding to each time step in the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix to obtain a first variance sequence and a second variance sequence; normalize the first variance sequence and the second variance sequence to obtain a first confidence sequence and a second confidence sequence; and determine the first confidence level and the second confidence level according to the first confidence sequence and the second confidence sequence, respectively.

[0015] Optionally, the fusion module is further configured to: determine, based on a first confidence level and a second confidence level, a first weight corresponding to the first spatiotemporal coding feature matrix and a second weight corresponding to the second spatiotemporal coding feature matrix; and perform weighted fusion of the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix based on the first weight and the second weight to obtain a fused feature matrix.

[0016] Optionally, the second determining module is further configured to: perform anomaly detection on the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix to obtain anomaly detection results; determine target features from the first spatiotemporal coding feature matrix, the second spatiotemporal coding feature matrix and the fused feature matrix based on the anomaly detection results; and determine navigation data based on the target features.

[0017] Optionally, the second determining module is further configured to: in response to an anomaly detection result indicating that the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix are normal, use the fused feature matrix as the target feature; or, in response to an anomaly detection result indicating that one of the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix is ​​abnormal, use the normal matrix of the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix as the target feature; or, in response to an anomaly detection result indicating that both the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix are abnormal, obtain historical fused features and use a linear prediction algorithm to predict the target feature based on the historical fused features.

[0018] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the navigation method described in any one of the above.

[0019] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the navigation method described in any one of the above.

[0020] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the navigation method described in any one of the above descriptions.

[0021] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a navigation method, apparatus, device, medium, and product. The method independently collects navigation spatiotemporal data through two modules to form a redundant data foundation. Then, a spatiotemporal encoder is used to extract highly stable and expressive spatiotemporal coding feature matrices. Based on the coding features, the corresponding confidence level is determined to achieve adaptive quantitative evaluation of the quality of the two navigation data streams. Subsequently, weights are dynamically allocated according to the confidence level for feature fusion, replacing the traditional fixed-weight fusion and effectively improving the feature fusion accuracy. Finally, the final navigation data is determined based on the two coding features and the fused feature matrix. Even in complex environments, it can still output stable and accurate results, thus solving the technical problem of low fusion accuracy when using fixed fusion weights in the prior art when fusing two satellite navigation data. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is an application environment diagram of a navigation method according to an embodiment of this application; Figure 2 A flowchart illustrating a navigation method provided in one embodiment of this application; Figure 3 This is a schematic diagram of the architecture of a navigation system provided in one embodiment of this application; Figure 4 A schematic diagram of the architecture of a spatiotemporal encoder provided in an embodiment of this application; Figure 5 A flowchart illustrating a navigation method provided in one embodiment of this application; Figure 6 This is a schematic diagram of a fault-tolerant detection and switching logic provided in one embodiment of this application; Figure 7 This is a schematic diagram of the structure of a navigation device provided in one embodiment of this application; Figure 8 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] The navigation method provided in this application embodiment can be applied to, for example, Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on another server. Terminal 102 can send first navigation spatiotemporal data and second navigation spatiotemporal data to server 104. After receiving the first and second navigation spatiotemporal data, server 104 uses a pre-built spatiotemporal encoder to extract features from the first and second navigation spatiotemporal data, obtaining a first spatiotemporal coding feature matrix and a second spatiotemporal coding feature matrix; it determines the first confidence level and the second confidence level corresponding to the first and second spatiotemporal coding feature matrices, respectively; based on the first and second confidence levels, it performs feature fusion on the first and second spatiotemporal coding feature matrices to obtain a fused feature matrix; and based on the first spatiotemporal coding feature matrix, the second spatiotemporal coding feature matrix, and the fused feature matrix, it determines the navigation data. Server 104 can then feed back the obtained navigation data to terminal 102. In addition, in some embodiments, the navigation method can also be implemented by the server 104 or the terminal 102 separately. For example, the terminal 102 can directly process the first navigation spatiotemporal data and the second navigation spatiotemporal data, or the server 104 can obtain the first navigation spatiotemporal data and the second navigation spatiotemporal data from the data storage system and process them.

[0027] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, and IoT devices. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers, or it can be a cloud server.

[0028] In one exemplary embodiment, such as Figure 2As shown, a navigation method is provided. This method is executed by a computer device, specifically, it can be executed by a computer device such as a terminal or a server alone, or it can be executed by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 201 to 205. Wherein: Step S201: Obtain first navigation spatiotemporal data and second navigation spatiotemporal data, wherein the first navigation spatiotemporal data and the second navigation spatiotemporal data are respectively collected by two independent satellite navigation modules.

[0029] In this embodiment, step S201 is the data acquisition step of the navigation method. In step S201, two independently configured satellite navigation modules (exemplarily using dual Beidou navigation modules, with their hardware devices and data transmission paths completely physically isolated to effectively avoid electromagnetic interference and data crosstalk between modules) are used to synchronously collect satellite navigation observation data from the same navigation carrier (such as various vehicles, ships, drones, outdoor mobile devices, and other common navigation carriers), and output the first navigation spatiotemporal data and the second navigation spatiotemporal data respectively.

[0030] Each set of navigation spatiotemporal data includes time and spatial dimensions. The time dimension corresponds to continuous fixed-interval time steps (e.g., one time step every 1 second, with the acquisition frequency flexibly adjusted according to actual navigation accuracy requirements). The spatial dimension corresponds to navigation parameters, including key indicators such as longitude, latitude, speed, heading, satellite signal-to-noise ratio, and positioning accuracy factor. Furthermore, the acquisition timestamps of the two sets of data are aligned to avoid fusion errors caused by time sequence deviations.

[0031] Step S201 employs a redundant design with independent acquisition of dual modules. On the one hand, it effectively avoids navigation data interruption or distortion caused by factors such as obstruction (e.g., tall buildings in cities, dense forests), electromagnetic interference (e.g., industrial equipment, communication signals), and hardware failures in a single navigation module. On the other hand, it provides two independent and synchronous data sources to support subsequent dynamic weight fusion, fault tolerance detection, and switching.

[0032] For example, two independent BeiDou-3 navigation modules are used to collect navigation spatiotemporal data in parallel. The two modules are physically isolated and work independently to avoid the failure of the entire navigation system due to the failure of a single module. The collected data includes, but is not limited to: carrier latitude and longitude, speed, heading angle, timestamp, satellite signal-to-noise ratio (SNR), number of visible satellites, satellite azimuth angle and elevation angle. The sampling frequency is set to 10Hz~50Hz, which can be dynamically adjusted according to the needs of the navigation scenario.

[0033] Optionally, in some embodiments of the present invention, it is necessary to preprocess the raw data collected by the two BeiDou modules to eliminate noise interference, achieve spatiotemporal synchronization, ensure data consistency, and provide high-quality data input for subsequent spatiotemporal coding. The specific processing steps include: Noise Reduction: A Kalman filter algorithm is used to denoise continuous data such as latitude, longitude, and velocity, removing errors caused by random noise and electromagnetic interference. The filter gain is dynamically adjusted according to the data noise characteristics. Spatiotemporal Synchronization: Using the timestamps of the two BeiDou modules as a reference, linear interpolation is used to achieve time synchronization of the data between the two modules (synchronization error ≤ 1ms). At the same time, the spatial position deviation of the two modules is calibrated according to the satellite azimuth and elevation angles to ensure spatial data consistency. Outlier Removal: The 3σ criterion is used to remove outliers in the collected data that significantly deviate from the normal range (such as abrupt changes in latitude and longitude caused by signal blockage, a sudden drop in signal-to-noise ratio, etc.) to avoid outliers affecting the subsequent coding and fusion accuracy.

[0034] Step S202: Using a pre-built spatiotemporal encoder, feature extraction is performed on the first navigation spatiotemporal data and the second navigation spatiotemporal data to obtain the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix.

[0035] In this embodiment, step S202 is a feature preprocessing step for navigation data, which converts the original navigation data into a spatiotemporal feature vector with high stability and high expressiveness, providing high-quality data support for subsequent confidence assessment and dynamic fusion.

[0036] Specifically, the first and second navigation spatiotemporal data obtained in step S201 are independently input into a pre-trained and debugged spatiotemporal encoder. The encoder's built-in spatiotemporal modeling unit performs deep feature extraction and feature reconstruction on both the first and second navigation spatiotemporal data. The spatiotemporal encoder automatically mines the continuous variation patterns in the time dimension of the original navigation data (such as position change trends and velocity fluctuation characteristics at different time steps) and the correlation information in the spatial dimension (such as the coordinated changes in longitude and latitude, and the matching relationship between speed and heading). Simultaneously, through a built-in noise suppression module, it effectively filters out irrelevant interference such as random measurement noise and environmental noise from the original observation data. Finally, it outputs a structurally sound first and second spatiotemporal coding feature matrix, respectively.

[0037] Understandably, the two sets of encoded feature matrices maintain completely consistent dimensions, forming a two-dimensional matrix of [number of time steps × feature dimension] (e.g., 100 time steps, 8-dimensional features, with the dimension flexibly adjusted according to the number of navigation parameters). The purpose of this step is to transform the original, low-dimensional navigation observation data into high-dimensional, high-quality features suitable for subsequent confidence assessment and dynamic weighted fusion, effectively solving the technical problems of high noise, weak feature correlation, and insufficient extraction of effective information in the original navigation data, which leads to low fusion accuracy.

[0038] Optionally, the spatiotemporal encoder is the Mamba spatiotemporal encoder, which has the ability to model long temporal data with linear complexity, can efficiently adapt to the continuous temporal characteristics of navigation data, and avoids the computational redundancy problem of traditional Transformer architecture.

[0039] Step S203: Determine the first confidence level and the second confidence level corresponding to the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix, respectively.

[0040] In this embodiment, step S203 is a prerequisite for dynamic weight allocation, and its purpose is to accurately quantify and evaluate the real-time reliability of the two navigation data.

[0041] Specifically, based on the first and second spatiotemporal coding feature matrices output in step S202, a unified confidence calculation algorithm is used to calculate the confidence indices corresponding to the two sets of features, namely the first confidence (corresponding to the first spatiotemporal coding feature matrix) and the second confidence (corresponding to the second spatiotemporal coding feature matrix). The confidence level is a quantitative indicator specifically characterizing the stability and reliability of the current output data of the corresponding navigation module.

[0042] Step S203 calculates the confidence level of the features to achieve adaptive and quantitative evaluation of the quality of the two navigation data streams. This replaces the unreasonable method of setting fixed weights based on experience in existing technologies, providing a precise basis for weight allocation in subsequent dynamic fusion steps. The higher the confidence level, the more reliable the navigation data of the corresponding module, and the higher the weight should be assigned during subsequent fusion; conversely, the lower the confidence level, the lower the weight should be assigned.

[0043] Step S204: Based on the first confidence level and the second confidence level, perform feature fusion on the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix to obtain the fused feature matrix.

[0044] In this embodiment of the application, step S204 realizes adaptive dynamic weighted fusion of the two navigation features.

[0045] For example, based on the first confidence level (denoted as conf1) and the second confidence level (denoted as conf2) calculated in step S203, the fusion weights of the two spatiotemporal coding features are dynamically calculated according to the adaptive allocation principle that the higher the confidence level, the greater the weight. The specific weight calculation formulas are: w1=conf1 / (conf1+conf2), w2=conf2 / (conf1+conf2). Here, w1 is the fusion weight of the first spatiotemporal coding feature matrix, and w2 is the fusion weight of the second spatiotemporal coding feature matrix, satisfying the constraint w1+w2=1 to ensure the stability of the magnitude of the fused feature matrix and avoid feature distortion caused by weight imbalance. After the weight calculation is completed, a weighted summation method is used step-by-step and feature-dimension-by-feature to perform fusion calculation on the first and second spatiotemporal coding feature matrices. That is, the fusion value at each time step and each feature dimension is the sum of the products of the two feature values ​​and their respective weights in the corresponding dimension, ultimately obtaining the complete fused feature matrix.

[0046] It should be noted that, compared with existing fixed-weight fusion technology, the fusion weights in this step can be dynamically adjusted according to the real-time confidence levels of the two navigation data. When the confidence level of a certain navigation module is reduced due to external interference (such as obstruction or electromagnetic interference), its corresponding fusion weight will automatically decrease, while the weight of the other reliable module (which maintains a high level of confidence) will automatically increase. This effectively offsets the fusion errors caused by noise and abnormal data jumps, significantly improves the accuracy and stability of the fusion feature matrix, and solves the problems of fixed weights being unable to adapt to changes in data quality and low fusion accuracy.

[0047] Step S205: Determine navigation data based on the first spatiotemporal coding feature matrix, the second spatiotemporal coding feature matrix, and the fused feature matrix.

[0048] In this embodiment, step S205 is the result output stage, which outputs high-precision, high-reliability, and high-robust final navigation data to meet the usage requirements of actual navigation scenarios.

[0049] Specifically, the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix obtained in step S202, as well as the fusion feature matrix obtained in step S204, are used together as decoding inputs and connected to the inverse coding mapping (decoding) module corresponding to the spatiotemporal encoder. The high-dimensional spatiotemporal coding features are reversed and restored into usable navigation parameters through the decoding algorithm, including core navigation data such as latitude and longitude coordinates, driving speed, navigation heading, and timing information.

[0050] Optionally, during the decoding process, anomaly detection and fault tolerance judgment are simultaneously completed by combining the confidence differences and feature change trends of the two encoded features. For example, it can determine in real time whether there are data anomalies or faults in a certain navigation module. If an anomaly is detected, it automatically switches to the data or fusion result of another reliable module to ensure the continuity of navigation output. The final output navigation data not only fully retains the high reliability advantage of the redundant data of the two Beidou modules, but also highlights the dominant role of high-quality data through dynamic weighted fusion. This effectively solves the technical problems of low accuracy, poor reliability, and weak anti-interference ability of existing fixed-weight fusion in complex environments (such as urban obstruction, electromagnetic interference, and complex outdoor terrain), ensuring that the navigation results can continuously, stably, and accurately meet the navigation needs of various carriers.

[0051] In summary, through the description of the above embodiments, the method provided by the present invention independently collects navigation spatiotemporal data through two modules to form a redundant data foundation; then, it uses a spatiotemporal encoder to extract spatiotemporal coding feature matrices with high stability and high expressive power; and determines the corresponding confidence level based on the coding features to achieve adaptive quantitative evaluation of the quality of the two navigation data streams; furthermore, it dynamically allocates weights based on the confidence level for feature fusion, replacing the traditional fixed-weight fusion, effectively improving the feature fusion accuracy; finally, it determines the final navigation data based on the two coding features and the fused feature matrix, and can still output stable and accurate results in complex environments, thus solving the technical problem of low fusion accuracy when using fixed fusion weights in the prior art when fusing two satellite navigation data.

[0052] Optionally, the construction of the spatiotemporal encoder includes: Step a: Determine the initial encoder based on the selective state-space model.

[0053] In this embodiment of the application, the spatiotemporal encoder needs to be pre-built. Step a is the basic framework building step for the spatiotemporal encoder. Based on the selective state space model (SSM), an initial encoder structure adapted to BeiDou navigation long time series data is built.

[0054] For example, firstly, the core architecture of the initial encoder is determined to be a linear layout consisting of an input layer, a local convolutional module, a selective scanning module, and an output layer. This layout can efficiently extract spatiotemporal features from navigation data, balancing computational efficiency and feature representation capabilities. The core module is the selective scanning module, which is composed of a selective state-space model (SSM). This module differs from traditional encoders in that it can achieve long-term data modeling with linear complexity (O(n)), breaking through the O(n) time complexity of the traditional Transformer attention mechanism. 2To overcome the limitations of complexity, the initial encoder is adapted to the processing requirements of continuous time-step data in BeiDou navigation. Simultaneously, the overall structure of the initial encoder needs to match the characteristics of BeiDou navigation data, and is designed as a bidirectional scanning architecture, performing forward (time steps from early to late, spatial parameters from primary to secondary) and reverse scanning of the navigation data respectively, ensuring comprehensive capture of the spatiotemporal dependencies of the navigation data.

[0055] In addition, the initial encoder reserves a parameter configuration interface, which clarifies the core parameter categories that need to be configured, including state space parameters, convolution parameters, time step parameters, dimension expansion parameters, activation functions and input dimensions. This provides a basis for parameter initialization in step b, enabling the initial encoder to have debuggable and optimizable characteristics, and adapt to the accuracy requirements of different navigation scenarios.

[0056] Understandably, the spatiotemporal encoder framework built in this step leverages the advantages of the selective state-space model to address the problems of computational redundancy and insufficient feature extraction in traditional encoders when processing long-term navigation data.

[0057] Step b involves initializing the parameters in the initial encoder to obtain the spatiotemporal encoder. The parameters in the initial encoder include at least the state space parameters, convolution parameters, time step parameters, dimension expansion parameters, activation function, and input dimension.

[0058] In this embodiment, step b is the parameter configuration step for constructing the spatiotemporal encoder, which initializes the parameters in the initial encoder determined in step a to ensure that the encoder can operate normally and efficiently extract the spatiotemporal features of BeiDou navigation.

[0059] For example, firstly, the parameters that the initial encoder needs to initialize are clearly defined. The initialization principles, specific values, and functions of various parameters are as follows. All parameters are adapted to BeiDou navigation data (e.g., the input dimension matches the number of navigation parameters, and the time step parameter matches the data acquisition frequency) to ensure the relevance and rationality of the parameter settings: State space parameters, including state matrix A, input projection matrix B, output projection matrix C, and direct mapping term D, are the core parameters of the selective state space model. State matrix A is initialized as a diagonal matrix with dimensions set to "hidden layer dimension / / 2, hidden layer dimension / / 2", and matrix elements are negative to ensure decay characteristics during state evolution and avoid feature redundancy. Input projection matrix B and output projection matrix C are both initialized as random normal distribution matrices, with dimensions matching the input feature dimension and hidden layer dimension, ensuring efficient mapping of input features to the state space and accurate output of state features. Direct mapping term D is initialized as a zero vector and is gradually optimized through training to supplement the direct contribution of input features and improve the completeness of feature extraction.

[0060] Convolution parameter: d_conv, the size of the local convolution kernel, is initialized to 4 (which can be flexibly adjusted according to the feature dimension of the navigation data). It corresponds to the local convolution module of the spatiotemporal encoder in step S202. It is used to capture the local spatiotemporal correlation features of the navigation data, filter local noise, improve the local stability of the features, and provide a high-quality local feature foundation for the long-term modeling of the subsequent selective scanning module.

[0061] Time step parameter: This is the initial value of the state update step size Δt. It is initialized as a learnable parameter with an initial value range of 0.1 to 0.5. It is subsequently adaptively adjusted through training to control the speed of state updates in the selective scanning module. This enables selective modeling of extending the memory of critical time steps and shortening the memory of redundant time steps, adapting to the temporal fluctuation characteristics of BeiDou navigation data. For example, dt_rank=auto, dt_min=0.001, dt_max=0.1, dt_init=random, dt_scale=1.0, dt_init_floor=1e-4.

[0062] It should be noted that `dt_rank=auto` sets the rank of the time step parameter Δt. When set to `auto`, the model automatically determines the rank based on the input feature dimension and hidden layer dimension, eliminating the need for manual setting. This adapts to BeiDou navigation data of different dimensions, simplifies parameter configuration, and ensures that the rank setting matches the data characteristics, avoiding computational redundancy due to an excessively large rank or insufficient time step control precision due to an excessively small rank. `dt_min=0.001` limits the minimum value of the time step Δt, ensuring that Δt cannot be less than 0.001. This prevents excessively frequent state updates due to an excessively small time step, which wastes computational resources, and also prevents over-refinement of feature memory due to an excessively small Δt, which can amplify noise, thus ensuring the rationality of time step control. `dt_max=0.1`: This setting limits the maximum value of time step Δt, ensuring that Δt cannot exceed 0.1. This prevents lag in state updates caused by excessively large time steps, which could hinder timely capture of temporal fluctuations in BeiDou navigation data (such as sudden velocity changes or position shifts). It ensures that the time step accurately tracks the dynamic changes in navigation data, improving the accuracy of temporal feature extraction. `dt_init=random`: This setting determines the initial value generation method for time step Δt. When set to `random`, the initial value of Δt is randomly generated between `dt_min` (0.001) and `dt_max` (0.1). Compared to fixed initial values, random initialization increases model diversity, avoids local optima in model training caused by single initial values, and improves the model's adaptability to the temporal characteristics of BeiDou navigation data. `dt_scale=1.0`: This parameter scales the initial value of time step Δt. A value of 1.0 keeps the randomly generated initial value of Δt unchanged. If adjustments are needed based on navigation scenario requirements, this parameter can be modified to scale Δt overall, flexibly adapting to BeiDou navigation data with different acquisition frequencies (e.g., the scaling factor can be reduced for high-frequency data and increased for low-frequency data). `dt_init_floor=1e-4`: This parameter sets the minimum threshold for the initial value of time step Δt. If the randomly generated initial value of Δt is lower than 1e-4, it will be automatically adjusted to 1e-4, further limiting the minimum value of Δt. This avoids computational anomalies caused by extremely small values, ensuring the stability of time step parameter initialization and providing a reliable foundation for subsequent model training.

[0063] Dimension expansion parameter: The core is the feature expansion factor expand, which is initialized to 2 (to adapt to the expansion requirements of Beidou navigation 8-dimensional features). It is used to expand the input feature dimension to the specified hidden layer dimension, improve the expressive power of the features, ensure that the encoder can mine the deep spatiotemporal correlation information hidden in the navigation data, and avoid the problem of insufficient fusion accuracy caused by low-dimensional features.

[0064] Activation function: The GELU activation function is selected for initialization.

[0065] Input dimension: Initialize based on the number of spatial dimension parameters of BeiDou navigation data. If the navigation data contains 8 core parameters such as longitude, latitude, speed, and heading, the input dimension is initialized to 8 to ensure that the encoder input layer can accurately match the dimension of the original navigation data and avoid feature distortion caused by dimension mismatch.

[0066] After all parameters are initialized, the initial encoder is debugged to ensure that the parameters of each module work together and there are no dimensional conflicts, and finally a spatiotemporal encoder that can be directly used for feature extraction in step S202 is obtained.

[0067] Optionally, in step S203, determining the first confidence level and the second confidence level corresponding to the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix respectively includes the following steps: Step S2031: Determine the variance of the features corresponding to each time step in the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix to obtain the first variance sequence and the second variance sequence.

[0068] In this embodiment of the application, step S2031 is the first step in the confidence calculation, which uses variance quantification to determine the stability of the feature at each time step.

[0069] For example, for the first spatiotemporal coding feature matrix output in step S202 (with dimensions of "number of time steps × feature dimension"), the variance of the feature vector corresponding to each time step is calculated step by step. For the feature vector of the i-th time step (containing d feature dimensions, where d is the number of feature dimensions, such as 8 dimensions), the variance of the features at all time steps is calculated sequentially using the variance calculation formula, resulting in a one-dimensional sequence consistent with the number of time steps, which is the first variance sequence. Using the same calculation method, the variance of the feature vectors of the second spatiotemporal coding feature matrix is ​​calculated step by step to obtain the second variance sequence.

[0070] Understandably, the magnitude of variance directly reflects the stability of the features at the corresponding time step. The smaller the variance, the smaller the fluctuation of the values ​​of each dimension of the feature vector at that time step, the more stable the features, and the higher the data reliability of the navigation module at that time step. The larger the variance, the more drastic the feature fluctuations, the more severe the interference to the module, and the lower the data reliability, thus providing basic data support for subsequent confidence quantification.

[0071] Step S2032: Normalize the first variance sequence and the second variance sequence to obtain the first confidence sequence and the second confidence sequence.

[0072] In this embodiment of the application, step S2032 is the second step in the confidence calculation, which converts the variance sequence into a confidence sequence with a uniform interval, so as to facilitate subsequent calculation and weight allocation.

[0073] For example, a linear normalization algorithm is used to normalize the first variance sequence and the second variance sequence obtained in step S2031. The core logic of normalization is that the smaller the variance, the higher the confidence level. Therefore, each variance sequence is first reverse-mapped (by subtracting the variance value of each time step from the maximum value of the variance sequence), and then the values ​​after reverse mapping are scaled to the range of 0 to 1 to ensure that the confidence level sequences have uniform and comparable values. After normalization calculation, the first variance sequence is normalized to obtain the first confidence level sequence, and the second variance sequence is normalized to obtain the second confidence level sequence. Both sequences have the same number of time steps as the original variance sequence. The confidence level of each time step is between 0 and 1. The closer the value is to 1, the more stable the characteristics of that time step and the more reliable the data.

[0074] Understandably, the purpose of step S203 is to eliminate the magnitude difference between different variance sequences, realize the quantitative comparison of the reliability of the two navigation data at each time step, and provide a standardized basis for the subsequent calculation of the overall confidence level.

[0075] Step S2033: Determine the first confidence level and the second confidence level according to the first confidence level sequence and the second confidence level sequence, respectively.

[0076] In this embodiment of the application, step S2033 is the final step in the confidence calculation, which obtains the overall confidence of the two navigation data based on the instantaneous confidence of each time step.

[0077] For example, for the first confidence sequence obtained in step S2032, the arithmetic mean of the instantaneous confidence of all time steps in the sequence is calculated. This average value is the first confidence corresponding to the first spatiotemporal coding feature matrix, used to characterize the overall reliability of the first BeiDou navigation module throughout the entire data acquisition cycle. Using the same calculation method, the arithmetic mean of the instantaneous confidence of all time steps in the second confidence sequence is calculated to obtain the second confidence corresponding to the second spatiotemporal coding feature matrix, which characterizes the overall reliability of the second BeiDou navigation module.

[0078] Understandably, by taking the average value, the reliability of each time step in the entire data acquisition cycle is comprehensively considered, avoiding the impact of abnormal data in a single time step on the overall confidence level. This yields a quantitative indicator that can objectively reflect the overall data quality of the two navigation modules, providing an accurate and reliable basis for the dynamic weight allocation in step S204.

[0079] Optionally, in step S204, feature fusion is performed on the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix based on the first confidence level and the second confidence level to obtain a fused feature matrix, specifically including the following steps: Step S2041: Determine the first weight corresponding to the first spatiotemporal coding feature matrix and the second weight corresponding to the second spatiotemporal coding feature matrix based on the first confidence level and the second confidence level.

[0080] In this embodiment, step S2041 is a dynamic fusion weight allocation step, which accurately calculates the fusion weights of the two navigation data based on the overall confidence of the two navigation data.

[0081] For example, using the first confidence level (conf1) and the second confidence level (conf2) obtained in step S203 as the core calculation basis, following the principle that confidence level is positively correlated with weight, a normalized weight calculation method is adopted to determine the first weight (denoted as w1) corresponding to the first spatiotemporal coding feature matrix and the second weight (denoted as w2) corresponding to the second spatiotemporal coding feature matrix. The specific calculation formula is: w1=conf1 / (conf1+conf2), w2=conf2 / (conf1+conf2). This formula ensures that the sum of the two weights is always 1 (w1+w2=1), avoiding weight imbalance that leads to distortion of the magnitude of the fused feature matrix, while realizing adaptive weight allocation. When conf1 is greater than conf2, w1 is greater than w2, that is, the data of the first navigation module is more reliable and obtains a higher fusion weight; when conf2 is greater than conf1, w2 is greater than w1, that is, the data of the second navigation module is more reliable and obtains a higher fusion weight; when the two confidence levels are equal, the weights are evenly distributed to achieve balanced fusion.

[0082] Understandably, step S2041 abandons the fixed weight setting of existing technologies and achieves an adaptive allocation where higher data quality results in a greater fusion weight through the direct correlation between confidence and weight. This provides scientific weight support for subsequent accurate fusion and solves the problem that fixed weights cannot adapt to changes in data quality.

[0083] Step S2042: Based on the first weight and the second weight, the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix are weighted and fused to obtain the fused feature matrix.

[0084] In this embodiment of the application, step S2042 is a feature fusion step, in which the two spatiotemporal coding feature matrices are precisely weighted and fused based on the first weight (w1) and the second weight (w2) determined in step S2041.

[0085] For example, the dimensions of both the first spatiotemporal coding feature matrix (denoted as M1) and the second spatiotemporal coding feature matrix (denoted as M2) are explicitly defined as "T×d" (T is the total number of time steps, and d is the number of feature dimensions, such as 100×8), ensuring that the dimensions of the two matrices are completely matched and satisfying the condition for element-wise weighted fusion. Subsequently, a weighted summation method is used to perform fusion calculation on M1 and M2 step-by-step and feature dimension-by-feature. After completing the weighted calculation for all time steps and all feature dimensions in sequence, a fused feature matrix with the same dimensions as M1 and M2 is obtained.

[0086] Understandably, step S2042, through weighted fusion, fully preserves the effective features of the two navigation data streams, while allowing the features of the high-confidence (high-weight) module to dominate, effectively offsetting the noise and abnormal interference of the low-confidence module, improving the accuracy and stability of the fused features, and providing core support for the output of high-precision navigation data in step S205.

[0087] Optionally, in step S205, navigation data is determined based on the first spatiotemporal coding feature matrix, the second spatiotemporal coding feature matrix, and the fused feature matrix, specifically including the following steps: Step S2051: Perform anomaly detection on the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix to obtain anomaly detection results.

[0088] In this embodiment, step S2051 is the anomaly detection step of the feature matrix. For example, a normal feature library is constructed (obtained through training with coded features from a large number of normal navigation scenarios). The Euclidean distance between the real-time coded features of the two modules and the normal feature library is calculated. When the distance is greater than a set threshold (the threshold is calibrated experimentally, with a default of 0.5), the corresponding BeiDou module is determined to be abnormal (such as signal interruption, interference, etc.). Simultaneously, feature confidence is used to assist in the judgment. When the confidence is lower than 0.3, the module is further confirmed to be abnormal, reducing the false positive rate.

[0089] Step S2052: Based on the anomaly detection results, determine the target features from the first spatiotemporal coding feature matrix, the second spatiotemporal coding feature matrix, and the fused feature matrix.

[0090] In this embodiment, step S2052 is a redundancy and fault-tolerance step for navigation data output. Based on the anomaly detection results, the optimal target features are adaptively selected to ensure the accuracy and reliability of the decoded navigation data. According to the anomaly detection results obtained in step S2051, following the selection principle of prioritizing non-anomaly features and secondarily selecting fused features, the corresponding target features are determined respectively. Then, the target features are input into the decoding module to reversely restore the final navigation data.

[0091] Step S2053: Determine navigation data based on target characteristics.

[0092] In this embodiment, after the target features are determined, they are connected to the decoding module corresponding to the spatiotemporal encoder. Through a reverse encoding mapping algorithm, the high-dimensional target features are reverse-engineered into usable core navigation parameters, specifically including latitude and longitude coordinates, driving speed, heading, and timing information. During the decoding process, the temporal variation trend of the target features is simultaneously considered to smooth the navigation parameters, further reducing data fluctuations and improving the stability of the output navigation data.

[0093] Understandably, the above steps achieve redundancy and fault tolerance for dual BeiDou navigation through anomaly detection and adaptive target feature filtering, ensuring that reliable navigation data can still be output when single or dual-path features are abnormal. This solves the problems of weak fault tolerance and sharp drop in navigation accuracy under abnormal scenarios in existing technologies, and meets the navigation needs in complex environments.

[0094] Optionally, in step S2052, the target features are determined from the first spatiotemporal coding feature matrix, the second spatiotemporal coding feature matrix, and the fused feature matrix based on the anomaly detection results. This specifically includes the following steps: If the anomaly detection result indicates that the first and second spatiotemporal coding feature matrices are normal, the fused feature matrix is ​​used as the target feature; or, if the anomaly detection result indicates that one of the first and second spatiotemporal coding feature matrices is abnormal, the normal matrix in the first and second spatiotemporal coding feature matrices is used as the target feature; or, if the anomaly detection result indicates that both the first and second spatiotemporal coding feature matrices are abnormal, historical fused features are obtained and a linear prediction algorithm is used to predict the target feature based on the historical fused features.

[0095] In this embodiment, in response to the anomaly detection result in step S2051, both the first spatiotemporal coding feature matrix (M1) and the second spatiotemporal coding feature matrix (M2) show no anomalies (dual normal scenario). At this time, both original coding features possess high stability and high reliability, with no abnormal interference. Based on this scenario, the fused feature matrix obtained in step S204 is preferentially used as the target feature. The fused feature matrix has integrated the core advantages of the two highly reliable features through dynamic weighted fusion. Compared with a single original coding feature, it can more comprehensively capture the spatiotemporal correlation information of navigation data, with better accuracy and stability. Using the fused feature matrix as the target feature input to the decoding module can fully leverage the core value of dynamic fusion, outputting higher-precision navigation data after decoding, meeting the usage requirements of high-precision navigation scenarios.

[0096] In response to the anomaly detection result in step S2051, an anomaly exists in either the first spatiotemporal coding feature matrix or the second spatiotemporal coding feature matrix (single anomaly scenario). This scenario is divided into two sub-cases, and the specific target feature selection logic is as follows: If only the first spatiotemporal coding feature matrix (M1) is abnormal, while the second spatiotemporal coding feature matrix (M2) is normal: In this case, M2 is not affected by the abnormality, possesses complete spatiotemporal features and high reliability, and is prioritized as the target feature, directly inputting it into the decoding module for decoding. At the same time, the abnormal M1 is completely discarded to avoid abnormal features participating in the decoding process, preventing interference with the accuracy of navigation results, fully demonstrating the fault-tolerance value of the dual Beidou redundancy design, and ensuring the reliability of navigation data.

[0097] If only the second spatiotemporal coding feature matrix (M2) is abnormal and the first spatiotemporal coding feature matrix (M1) is normal: consistent with the above sub-case logic, the M1 without abnormality is given priority as the target feature and input into the decoding module for decoding, while the abnormal M2 is discarded, effectively avoiding interference from abnormal data and ensuring the stability and accuracy of navigation data.

[0098] In response to the anomaly detection results in step S2051, both the first spatiotemporal coding feature matrix (M1) and the second spatiotemporal coding feature matrix (M2) exhibit anomalies (double anomaly scenario). In this case, neither of the two original features can be directly used as target features (which can easily lead to a sharp drop in navigation accuracy). Furthermore, the current fused feature matrix is ​​unreliable because it is based on the fusion of the original features from the double anomalies. Therefore, the target features are obtained by using historical fused features + linear prediction, as follows: Historical fusion features are defined as the fusion feature matrices corresponding to N consecutive normal data acquisition cycles prior to the current data acquisition cycle (N is preferably 5~10, and can be flexibly adjusted according to the motion characteristics of the navigation vehicle). All historical fusion features must satisfy the condition that M1 and M2 are not abnormal within their corresponding cycles, ensuring the reliability of historical data. Each historical fusion feature matrix has the same dimension as the current fusion feature matrix (both are "T×d"), and a historical fusion feature sequence is constructed in chronological order.

[0099] A linear prediction algorithm is used to predict target features: a univariate linear prediction algorithm (adapted to the linear temporal variation characteristics of navigation data, computationally efficient, and with strong real-time performance) is selected. Using historical fused feature sequences as input, the algorithm predicts target features step-by-step and feature-dimension-by-feature. Specifically, for each feature dimension d, feature values ​​for all time steps of that dimension are extracted from the historical fused feature sequence to construct a one-dimensional time-series data sequence. A linear prediction model is then fitted using the least squares method to predict the target feature values ​​for the corresponding time step and feature dimension in the current period. After predicting all time steps and all feature dimensions sequentially, a complete target feature matrix is ​​obtained. This matrix has the same dimensions as the current fused feature matrix, accurately matching the input requirements of the decoding module.

[0100] Reference Figure 3 This invention also provides a navigation system, comprising the following six core modules, the connection relationships and functions of which are as follows: Dual BeiDou signal acquisition module: It consists of two independent BeiDou-3 navigation modules (first BeiDou navigation module and second BeiDou navigation module). The model can be BDS-3 receiver module. The two modules work in parallel to acquire carrier navigation spatiotemporal data and output raw navigation data to the data preprocessing module. Data preprocessing module: Electrically connected to the dual Beidou signal acquisition module, it uses algorithms such as Kalman filtering, linear interpolation, and 3σ criterion to denoise, synchronize time and space, and remove outliers from the raw data, and outputs the preprocessed data to the Mamba time-space encoder module; Mamba Spatiotemporal Encoder Module: Electrically connected to the data preprocessing module, it uses the Mamba model configured above to encode the preprocessed spatiotemporal sequence data, extract deep spatiotemporal correlation features, and output the encoded features to the redundancy fusion module and the fault tolerance detection and switching module. Redundant fusion module: Electrically connected to the Mamba spatiotemporal encoder module, it adopts a dynamic weighted fusion algorithm, allocates weights based on the confidence of coding features, realizes dual Beidou data fusion, and outputs initial fused navigation data to the fault-tolerant detection and switching module; Fault-tolerant detection and switching module: It is electrically connected to the Mamba spatiotemporal encoder module and the redundant fusion module respectively. It uses Euclidean distance anomaly detection and feature confidence to assist in judgment, detects module anomalies and performs fault-tolerant switching, and outputs the final navigation data to the navigation output module. Navigation output module: Electrically connected to the fault-tolerant detection and switching module, it is used to receive the final navigation data and output it through the corresponding communication interface for use by the carrier's navigation display, control and other equipment.

[0101] Taking vehicle navigation as an example, the implementation of each module is as follows: Dual Beidou signal acquisition module: Two single-frequency receiver modules are selected, with the sampling frequency set to 10Hz, to collect vehicle navigation data in parallel, including latitude and longitude (accuracy ±1m), speed (accuracy ±0.1m / s), heading angle (accuracy ±0.5°), timestamp (accuracy ±1μs), satellite signal-to-noise ratio, number of visible satellites, satellite azimuth angle and elevation angle; the two modules are installed at the front and rear of the vehicle respectively, physically isolated to avoid mutual interference.

[0102] Data preprocessing module: Kalman filter parameters are set as follows: process noise variance Q=diag([1e-4,1e-4,1e-6,1e-6,1e-6]), observation noise variance R=diag([1e-3,1e-3,1e-5]); spatiotemporal synchronization adopts linear interpolation method, and the synchronization error is controlled within 1ms; outlier removal adopts the 3σ criterion, and the threshold is calibrated according to the normal range of vehicle navigation data (latitude and longitude deviation threshold ±5m, speed deviation threshold ±2m / s).

[0103] The Mamba spatiotemporal encoder module has the following input features: D=8, time series length T=10 (corresponding to 1 second of continuous data); core parameter configuration: d_state=32, d_conv=4, expand=2, dt_config is configured to the default optimal setting, activation function is GELU, and output feature dimension is 8; the encoder is implemented using the PyTorch framework and runs on an embedded processor, with a single frame encoding time ≤5ms.

[0104] Redundant fusion module: Feature confidence is calculated by normalizing the variance of the encoded feature vector, with a normalization range of [0,1]; Dynamic weight allocation adopts the above formula, and the accuracy of the fused navigation data is improved to ±0.8m (latitude and longitude) and ±0.08m / s (speed).

[0105] Fault-tolerant detection and switching module: The anomaly detection threshold is set to 0.5 (Euclidean distance), and the feature confidence threshold is set to 0.3; the fault-tolerant switching response time is ≤0.1 seconds. When the signal of one Beidou module is interrupted, it immediately switches to another module. The navigation interruption time is ≤0.1 seconds, which meets the requirements of vehicle navigation.

[0106] Navigation output module: adopts CAN bus interface, output frequency 10Hz, output data includes latitude and longitude, speed, heading angle, timestamp and navigation status (normal / abnormal / switching), for display by the vehicle navigation system and use by the vehicle control system.

[0107] Reference Figure 4This diagram illustrates the internal structure and data flow of the Mamba spatiotemporal encoder, a feature extraction module in a dual-BeiDou redundant fault-tolerant navigation system. It employs a linear layout with left-right flow, consisting of: input, input layer, local convolution module, selective scan module, output layer, and output. Data flows between these modules, with arrows clearly indicating the order of encoding processing.

[0108] The input layer receives raw navigation data with an input dimension of (T×D, D=8), where T represents the number of time steps and D=8 represents 8-dimensional navigation features (such as longitude, latitude, speed, heading, etc.). The raw data is converted into a feature format that can be processed internally by the encoder, preparing it for subsequent convolution and scanning.

[0109] The local convolution module (d_conv=4) indicates that the local convolution kernel size is 4. It performs local spatiotemporal convolution on the input features to capture local correlation information of the navigation data (such as the short-term trend of position change) while filtering local noise and improving the local stability of the features.

[0110] The selective scan module (d_state=32, expand=2), with parameters d_state=32 (state space dimension) and expand=2 (feature dimension expansion factor), is the core of the Mamba encoder. It uses a selective state-space model (SSM) to model long-term temporal data. It dynamically adjusts the memory strength for different time steps, efficiently capturing the long-term spatiotemporal dependencies of navigation data while maintaining linear complexity, thus avoiding the computational bottleneck of traditional Transformers.

[0111] The output layer also has an output dimension of (T×D, D=8), consistent with the input dimension. The high-dimensional features processed by convolution and selective scanning are restored to a spatiotemporally encoded feature matrix that can be used for subsequent fusion and fault-tolerant detection. The output spatiotemporally encoded feature matrix provides a high-quality and highly stable feature foundation for subsequent dynamic confidence assessment and adaptive feature fusion, and is the core support for solving the problem of low accuracy in fixed-weight fusion.

[0112] Reference Figure 5 and Figure 6 Taking in-vehicle navigation as an example, the method is implemented as follows: Step 1: Dual Beidou signal acquisition. Two Beidou modules work in parallel to acquire vehicle navigation data in real time at a sampling frequency of 10Hz. The acquired raw data (latitude, longitude, speed, etc.) is transmitted to the data preprocessing module via serial port.

[0113] Step 2: Data preprocessing. After receiving the raw data, the data preprocessing module first uses Kalman filtering to remove random noise from the velocity and latitude / longitude data. Then, using the timestamps of the two modules as a reference, time synchronization is achieved through linear interpolation to calibrate the spatial position deviation between the two modules. Finally, the 3σ criterion is used to remove outliers such as abrupt changes in latitude / longitude and a sudden drop in signal-to-noise ratio, and outputs high-quality preprocessed spatiotemporal sequence data.

[0114] Step 3: Mamba spatiotemporal coding. The preprocessed spatiotemporal sequence data of the two modules (each module is a 10×8 matrix) are input into the Mamba spatiotemporal encoder. The encoder captures the spatial correlation between satellite signal-to-noise ratio and satellite azimuth angle through the local convolution module (d_conv=4), and captures the velocity change trend of 10 consecutive time steps through the selective scanning module. It outputs the 10×8 encoded feature matrix corresponding to the two modules. This feature matrix contains deep spatiotemporal correlation information.

[0115] Step 4: Dynamic redundancy fusion. Calculate the average confidence (conf1, conf2) of the two module coding feature matrices. Assuming conf1=0.8 and conf2=0.7, the dynamic weights w1=0.8 / (0.8+0.7)=0.53 and w2=0.7 / (0.8+0.7)=0.47. Use a weighted summation method to fuse the coding features of the two modules to obtain a fused feature matrix. Then, through inverse coding mapping, convert it into vehicle navigation data (latitude, longitude, speed, etc.) as the initial fusion result.

[0116] Step 5: Fault tolerance detection and switching. Calculate the Euclidean distance between the encoded features of the two modules and the normal feature library. If the distance of the first Beidou module is 0.6 (greater than the threshold of 0.5) and the confidence level is 0.25 (lower than the threshold of 0.3), then the first module is determined to be abnormal. At this time, the fault tolerance detection and switching module immediately switches to the navigation data of the second module (obtained by inverse mapping based on the encoded features of the second module) to ensure that navigation is not interrupted. When the first module recovers to normal (distance ≤ 0.5, confidence level ≥ 0.3), switch back to the fused navigation result.

[0117] Step 6: Navigation result output. The navigation output module outputs the final navigation data (latitude and longitude, speed, heading angle, etc.) to the vehicle navigation system and vehicle control system via the CAN bus at a frequency of 10Hz, thus completing the vehicle navigation.

[0118] Optionally, in some embodiments of the present invention, in an in-vehicle navigation scenario, the present invention is compared with an existing dual BeiDou redundant fault-tolerant navigation scheme (using a Transformer encoder + fixed weight fusion). The test environment includes three scenarios: urban roads (with obstructions), highways (without obstructions), and areas with electromagnetic interference. The test results are as follows: Navigation accuracy: The average latitude and longitude accuracy of this invention is ±0.75m and the speed accuracy is ±0.07m / s in three scenarios; the average latitude and longitude accuracy of existing solutions is ±1.0m and the speed accuracy is ±0.1m / s, and the accuracy of this invention is improved by more than 25%. Fault tolerance performance: The fault error rate of the present invention is 0.5%, and the fault tolerance response time is 0.08 seconds; the fault error rate of the existing solution is 1.2%, and the fault tolerance response time is 0.2 seconds. The present invention reduces the fault error rate by more than 40% and improves the response speed by more than 60%. Real-time performance: The processing time for a single frame of data in this invention is 8ms, while the existing solution is 20ms, thus improving the real-time performance by more than 60%. Reliability: In electromagnetic interference areas, the navigation interruption time of the present invention is ≤0.1 seconds, while the navigation interruption time of the existing solution is ≤0.3 seconds, and the reliability of the present invention is significantly improved.

[0119] Test results show that the present invention can effectively improve the accuracy, reliability and real-time performance of dual Beidou redundant fault-tolerant navigation, fully meet the needs of vehicle navigation scenarios, and can also be adapted to other high-requirement navigation scenarios.

[0120] Based on the description of the above embodiments, the present invention has the following technical effects: By employing the Mamba spatiotemporal encoder, which combines its local convolution and selective scanning mechanisms, the long-term spatiotemporal correlation features of BeiDou navigation data can be efficiently captured. Compared with existing Transformer and CNN encoders, the feature extraction error is reduced by more than 30%, providing high-quality feature support for subsequent redundant fusion. At the same time, the parameter configuration of the Mamba model has been optimized to balance accuracy and computational efficiency, resolving the contradiction between accuracy and real-time performance in existing encoders.

[0121] The dynamic weighted fusion algorithm based on the confidence level of coded features can adjust the fusion weights according to the real-time working status of the BeiDou module, which improves the fusion navigation accuracy compared with fixed weight fusion. The fault-tolerant detection combines the anomaly degree and confidence level of coded features, which reduces the false judgment rate and shortens the fault-tolerant response time. It can quickly respond to abnormal situations such as signal blockage and electromagnetic interference, ensuring uninterrupted navigation.

[0122] The computational complexity of the Mamba model is much lower than that of the Transformer (the number of parameters is reduced by more than 50%). When processing high-frequency BeiDou data of 10Hz~50Hz, the processing time of a single frame is ≤10ms, which meets the real-time navigation requirements of high-speed mobile carriers such as vehicles and drones.

[0123] Based on the same inventive concept, this application also provides a navigation device for implementing the navigation method described above. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations in one or more navigation device embodiments provided below can be found in the limitations of the navigation method described above, and will not be repeated here.

[0124] In one exemplary embodiment, such as Figure 7 As shown, a navigation device 70 includes: an acquisition module 701 for acquiring first navigation spatiotemporal data and second navigation spatiotemporal data, wherein the first navigation spatiotemporal data and the second navigation spatiotemporal data are respectively acquired by two independent navigation modules; an extraction module 702 for extracting features from the first navigation spatiotemporal data and the second navigation spatiotemporal data using a pre-constructed spatiotemporal encoder to obtain a first spatiotemporal coding feature matrix and a second spatiotemporal coding feature matrix; a first determination module 703 for determining a first confidence level and a second confidence level corresponding to the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix, respectively; a fusion module 704 for fusing features from the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix according to the first confidence level and the second confidence level to obtain a fused feature matrix; and a second determination module 705 for determining navigation data based on the first spatiotemporal coding feature matrix, the second spatiotemporal coding feature matrix, and the fused feature matrix.

[0125] Optionally, the extraction module 702 is further configured to: determine an initial encoder based on a selective state-space model; initialize the parameters in the initial encoder to obtain a spatiotemporal encoder, wherein the parameters in the initial encoder include at least state-space parameters, convolution parameters, time-step parameters, dimension expansion parameters, activation function, and input dimension.

[0126] Optionally, the first determining module 703 is further configured to: determine the variance of the features corresponding to each time step in the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix, to obtain a first variance sequence and a second variance sequence; normalize the first variance sequence and the second variance sequence to obtain a first confidence sequence and a second confidence sequence; and determine the first confidence level and the second confidence level according to the first confidence sequence and the second confidence sequence, respectively.

[0127] Optionally, the fusion module 704 is further configured to: determine, based on the first confidence level and the second confidence level, a first weight corresponding to the first spatiotemporal coding feature matrix and a second weight corresponding to the second spatiotemporal coding feature matrix; and perform weighted fusion of the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix based on the first weight and the second weight to obtain a fused feature matrix.

[0128] Optionally, the second determining module 705 is further configured to: perform anomaly detection on the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix to obtain anomaly detection results; determine target features from the first spatiotemporal coding feature matrix, the second spatiotemporal coding feature matrix and the fused feature matrix based on the anomaly detection results; and determine navigation data based on the target features.

[0129] Optionally, the second determining module 705 is further configured to: in response to an anomaly detection result indicating that the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix are normal, use the fused feature matrix as the target feature; or, in response to an anomaly detection result indicating that one of the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix is ​​abnormal, use the normal matrix of the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix as the target feature; or, in response to an anomaly detection result indicating that both the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix are abnormal, obtain historical fused features and use a linear prediction algorithm to predict the target feature based on the historical fused features.

[0130] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 8 As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores navigation spatiotemporal data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a navigation method.

[0131] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0132] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0133] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0134] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0135] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0136] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0137] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0138] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0139] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A navigation method, characterized in that, The navigation method includes: First navigation spatiotemporal data and second navigation spatiotemporal data are acquired, wherein the first navigation spatiotemporal data and the second navigation spatiotemporal data are respectively acquired through two independent satellite navigation modules; Using a pre-constructed spatiotemporal encoder, features are extracted from the first navigation spatiotemporal data and the second navigation spatiotemporal data to obtain a first spatiotemporal coding feature matrix and a second spatiotemporal coding feature matrix; Determine the first confidence level and the second confidence level corresponding to the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix, respectively; Based on the first confidence level and the second confidence level, feature fusion is performed on the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix to obtain a fused feature matrix; Navigation data is determined based on the first spatiotemporal coding feature matrix, the second spatiotemporal coding feature matrix, and the fusion feature matrix.

2. The navigation method according to claim 1, characterized in that, The construction of the spatiotemporal encoder includes: The initial encoder is determined based on the selective state-space model; The parameters in the initial encoder are initialized to obtain the spatiotemporal encoder, wherein the parameters in the initial encoder include at least state space parameters, convolution parameters, time step parameters, dimension expansion parameters, activation function, and input dimension.

3. The navigation method according to claim 1, characterized in that, The determination of the first confidence level and the second confidence level corresponding to the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix, respectively, specifically includes: Determine the variance of the features corresponding to each time step in the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix to obtain the first variance sequence and the second variance sequence; Normalize the first variance sequence and the second variance sequence to obtain the first confidence sequence and the second confidence sequence; The first confidence level and the second confidence level are determined based on the first confidence level sequence and the second confidence level sequence, respectively.

4. The navigation method according to claim 1, characterized in that, The step of fusing features between the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix based on the first confidence level and the second confidence level to obtain a fused feature matrix specifically includes: Based on the first confidence level and the second confidence level, determine the first weight corresponding to the first spatiotemporal coding feature matrix and the second weight corresponding to the second spatiotemporal coding feature matrix; Based on the first weight and the second weight, the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix are weighted and fused to obtain the fused feature matrix.

5. The navigation method according to claim 1, characterized in that, The step of determining navigation data based on the first spatiotemporal coding feature matrix, the second spatiotemporal coding feature matrix, and the fused feature matrix specifically includes: Anomaly detection is performed on the first and second spatiotemporal coding feature matrices to obtain anomaly detection results. Based on the anomaly detection results, target features are determined from the first spatiotemporal coding feature matrix, the second spatiotemporal coding feature matrix, and the fused feature matrix; The navigation data is determined based on the target characteristics.

6. The navigation method according to claim 5, characterized in that, The step of determining the target feature from the first spatiotemporal coding feature matrix, the second spatiotemporal coding feature matrix, and the fused feature matrix based on the anomaly detection result specifically includes: In response to the anomaly detection result indicating that the first and second spatiotemporal coding feature matrices are normal, the fused feature matrix is ​​used as the target feature; or, In response to the anomaly detection result indicating that one of the first and second spatiotemporal coding feature matrices is anomaly, the normal matrix in the first and second spatiotemporal coding feature matrices is taken as the target feature; or, In response to the anomaly detection result indicating that both the first and second spatiotemporal coding feature matrices are abnormal, historical fusion features are obtained and the target feature is predicted based on the historical fusion features using a linear prediction algorithm.

7. A navigation device, characterized in that, The navigation device includes: The acquisition module is used to acquire first navigation spatiotemporal data and second navigation spatiotemporal data, wherein the first navigation spatiotemporal data and the second navigation spatiotemporal data are respectively acquired through two independent navigation modules; The extraction module is used to extract features from the first navigation spatiotemporal data and the second navigation spatiotemporal data using a pre-built spatiotemporal encoder, to obtain a first spatiotemporal coding feature matrix and a second spatiotemporal coding feature matrix; The first determining module is used to determine the first confidence level and the second confidence level corresponding to the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix, respectively; The fusion module is used to perform feature fusion on the first spatiotemporal coding feature matrix and the second spatiotemporal coding feature matrix based on the first confidence level and the second confidence level to obtain a fused feature matrix; The second determining module is used to determine navigation data based on the first spatiotemporal coding feature matrix, the second spatiotemporal coding feature matrix, and the fusion feature matrix.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of the navigation method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the navigation method according to any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the navigation method according to any one of claims 1-6.

Images