GNSS / INS Atmospheric Integration Positioning Method and System Based on Inertial Navigation System

Abstract

The present application provides a satellite navigation and atmospheric fusion positioning method and system based on an inertial navigation system, which relates to the field of navigation. The method includes: first, obtaining multi-source navigation raw data, including inertial navigation data, satellite navigation data, and atmospheric information data; extracting features through deep learning to obtain multi-modal feature data; evaluating the quality of the satellite navigation data to determine whether the satellite navigation signal is valid. When the satellite navigation signal is valid, a federated filtering algorithm is used for integrated navigation solution, and an atmospheric information error model is established; when the satellite navigation signal is invalid, the atmospheric information is compensated using this error model and fused with the inertial navigation data to achieve high-precision navigation and ensure the flight safety of the unmanned aerial vehicle in complex environments.

Description

Technical Field

[0001] This application relates to the field of navigation, and particularly to a satellite-aided atmospheric fusion positioning method and system based on an inertial navigation system. Background Art

[0002] With the rapid development of unmanned aerial vehicle (UAV) technology, the performance and reliability of navigation systems are crucial for flight safety. Currently, UAV navigation systems generally adopt a combined navigation scheme of a global navigation satellite system (GNSS) and an inertial navigation system (INS). The satellite signals provide absolute position information, while the inertial navigation system provides short-term high-precision position, velocity, and attitude information. This combined navigation method can provide stable and reliable navigation results in open environments. However, in practical applications, UAVs often need to perform tasks in urban canyons, forest-covered areas, high-rise dense areas, or military confrontation environments. In these complex environments, satellite navigation signals are easily blocked, affected by multipath effects, or artificially interfered with, resulting in a significant decrease in navigation accuracy or even complete failure. When the satellite navigation signals fail, traditional combined navigation systems can only rely on the inertial navigation system for navigation, and the inertial navigation system itself has the problem of error accumulation, leading to a rapid decrease in position accuracy over time. Prolonged interruption of satellite navigation signals will expose UAVs to serious flight safety risks and may even cause UAVs to lose control or crash. Summary of the Invention

[0003] This application provides a satellite-aided atmospheric fusion positioning method and system based on an inertial navigation system, which solves the problem of how to make full use of atmospheric information and the inertial navigation system for fusion to provide continuous and stable high-precision navigation information for UAVs in the case of satellite navigation signals being interfered with or failing.

[0004] In a first aspect, this application provides a satellite-aided atmospheric fusion positioning method based on an inertial navigation system, the method comprising:

[0005] Obtaining inertial navigation data, satellite navigation data, and atmospheric information data to obtain multi-source navigation raw data;

[0006] Performing deep learning feature extraction on the multi-source navigation raw data to obtain multi-modal feature data;

[0007] Performing quality assessment on the satellite navigation data to obtain a satellite-aided navigation signal availability result;

[0008] When the satellite-aided navigation signal availability result indicates that the satellite-aided navigation information is valid: inputting the multi-modal feature data into a federated filtering algorithm to perform combined navigation solution on the inertial navigation data and the satellite-aided navigation information to obtain an initial navigation result; estimating and modeling the error amount in the atmospheric information based on the initial navigation result to obtain an atmospheric information error model;

[0009] When the availability result of the satellite navigation signal indicates that the satellite navigation information is invalid: compensate the real-time atmospheric information by using the atmospheric information error model to obtain the compensated atmospheric information; fuse and position the compensated atmospheric information with the inertial navigation data to obtain a high-precision navigation result.

[0010] By adopting the above technical solution, through multi-source data fusion and dynamic mode switching, the high reliability and high precision of the navigation system in complex environments are realized. First, this method simultaneously acquires inertial navigation data, satellite navigation data, and atmospheric information data, expands the dimension of available information sources, and processes the multi-source navigation raw data through deep learning feature extraction technology to effectively extract the spatio-temporal features contained in the data and obtain multi-modal feature data. This deep learning-based feature extraction method can automatically mine the deep associations between data and improve the feature expression ability compared with traditional data processing methods. Second, this method designs a dual-mode working mechanism to real-time judge the availability of the satellite navigation signal by evaluating the quality of the satellite navigation data. When the satellite navigation signal is available, the system inputs the multi-modal feature data into the federated filtering algorithm to achieve the optimal fusion of inertial navigation data and satellite navigation information, and simultaneously dynamically estimates and models the error amount in the atmospheric information to obtain an atmospheric information error model. This process not only provides a high-precision navigation result but also establishes a mapping relationship between the atmospheric information and the position, laying a foundation for backup navigation when the satellite navigation signal fails. When the satellite navigation signal is unavailable, the system can seamlessly switch to the auxiliary navigation mode based on the atmospheric information, compensate the real-time atmospheric information by using the previously established atmospheric information error model, and fuse and position the compensated atmospheric information with the inertial navigation data. This fusion method effectively suppresses the error accumulation problem of the inertial navigation system and breaks through the performance bottleneck of traditional integrated navigation in the case of satellite navigation signal failure. Through the organic combination of the above technical means, this method realizes the continuous high-precision positioning of the navigation system in an environment where the satellite navigation is interfered, greatly improves the environmental adaptability and anti-interference ability of the navigation system, and provides a reliable guarantee for the safe flight of the unmanned aerial vehicle in complex environments.

[0011] In the second aspect of the present application, a satellite navigation and atmospheric fusion positioning system based on an inertial navigation system is provided, including:

[0012] A data acquisition module, configured to acquire inertial navigation data, satellite navigation data, and atmospheric information data to obtain multi-source navigation raw data;

[0013] A feature extraction module, configured to perform deep learning feature extraction on the multi-source navigation raw data to obtain multi-modal feature data;

[0014] A quality assessment module, configured to evaluate the quality of the satellite navigation data to obtain the availability result of the satellite navigation signal;

[0015] An atmospheric error model construction module, when the availability result of the satellite navigation signal indicates that the satellite navigation information is valid: input the multi-modal feature data into the federated filtering algorithm, perform integrated navigation solution on the inertial navigation data and the satellite navigation information, and obtain an initial navigation result; estimate and model the error amount in the atmospheric information based on the initial navigation result, and obtain an atmospheric information error model.

[0016] A fusion positioning module, when the availability result of the satellite navigation signal indicates that the satellite navigation information is invalid: use the atmospheric information error model to compensate the real-time atmospheric information, and obtain the compensated atmospheric information; fuse and position the compensated atmospheric information and the inertial navigation data to obtain a high-precision navigation result.

[0017] In the third aspect of the present application, a computer storage medium is provided. The computer storage medium stores multiple instructions, and the instructions are adapted to be loaded and executed by a processor to perform the above method steps.

[0018] In the fourth aspect of the present application, an electronic device is provided, including a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to enable the electronic device to perform the above method.

[0019] In summary, one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0020] 1. Through multi-source data fusion and dynamic mode switching, the present application realizes high reliability and high precision of the navigation system in complex environments. First, the method simultaneously acquires inertial navigation data, satellite navigation data, and atmospheric information data, expands the dimension of available information sources, and processes the multi-source navigation raw data through deep learning feature extraction technology to effectively extract the spatio-temporal features contained in the data and obtain multi-modal feature data. This deep learning-based feature extraction method can automatically mine the deep associations between data and improve the feature expression ability compared with traditional data processing methods.

[0021] 2. The present application designs a dual-mode working mechanism. By evaluating the quality of satellite navigation data, it can judge the availability of satellite navigation signals in real time. When the satellite navigation signals are available, the system inputs multi-modal feature data into the federated filtering algorithm to achieve the optimal fusion of inertial navigation data and satellite navigation information. At the same time, it dynamically estimates and models the error amount in the atmospheric information to obtain an atmospheric information error model. This process not only provides high-precision navigation results but also establishes a mapping relationship between atmospheric information and position, laying a foundation for backup navigation when satellite navigation signals fail. When the satellite navigation signals are unavailable, the system can seamlessly switch to the auxiliary navigation mode based on atmospheric information, compensate the real-time atmospheric information using the previously established atmospheric information error model, and fuse the compensated atmospheric information with inertial navigation data for positioning. This fusion method effectively suppresses the error accumulation problem of the inertial navigation system and breaks through the performance bottleneck of traditional integrated navigation in the case of satellite navigation signal failure. Description of the Drawings

[0022] Figure 1 It is a schematic flowchart of a satellite navigation and atmosphere fusion positioning method based on an inertial navigation system provided by an embodiment of the present application. Detailed Embodiments

[0023] In order to enable those skilled in the art to better understand the technical solutions in this specification, the following will clearly and completely describe the technical solutions in the embodiments of this specification in conjunction with the accompanying drawings in the embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments.

[0024] In the description of the embodiments of the present application, words such as "for example" or "for illustration" are used to represent examples, illustrations, or explanations. Any embodiment or design solution described as "for example" or "for illustration" in the embodiments of the present application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Exactly speaking, using words such as "for example" or "for illustration" aims to present relevant concepts in a specific manner.

[0025] In the description of the embodiments of the present application, the meaning of the term "a plurality" refers to two or more. For example, a plurality of systems refers to two or more systems, and a plurality of screen terminals refers to two or more screen terminals. In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the technical features indicated. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. The terms "include", "comprise", "have" and their variants all mean "include but not limited to", unless otherwise specifically emphasized in other ways.

[0026] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments.

[0027] On the basis of the above background technology, further, please refer to Figure 1 , Figure 1 which is a schematic flowchart of a satellite-aided atmospheric fusion positioning method based on an inertial navigation system provided by an embodiment of the present application. This system can be implemented depending on a computer program or run as an independent tool application. In a preferred embodiment of the present invention, in the embodiments of the present application, this method can be applied to a server, but can also be applied to electronic devices such as a server. A satellite-aided atmospheric fusion positioning method based on an inertial navigation system includes the following steps:

[0028] S101, obtaining inertial navigation data, satellite navigation data, and atmospheric information data to obtain multi-source navigation raw data;

[0029] Specifically, when performing step S101, inertial navigation data is first obtained through an inertial measurement unit carried on the unmanned aerial vehicle. Inertial navigation data is navigation parameters obtained by measuring the acceleration and angular velocity of the carrier and integrating them in combination with initial position, velocity, and attitude information. Specifically, a three-axis accelerometer measures the linear acceleration of the unmanned aerial vehicle on three orthogonal axes, and a three-axis gyroscope measures the angular rate of the unmanned aerial vehicle. These raw inertial measurement data are output as the attitude, velocity, and position information of the unmanned aerial vehicle after coordinate transformation, compensation correction, and integration calculation. The purpose of collecting inertial navigation data is to provide continuous navigation information to ensure that the unmanned aerial vehicle can know its own motion state in real time. However, due to factors such as zero bias and scale factor error in inertial measurement, as well as error accumulation problems in the integration process, relying solely on inertial navigation data cannot maintain a high-precision navigation effect for a long time.

[0030] At the same time, the system obtains satellite navigation data through a satellite signal receiver. Satellite navigation data is the three-dimensional position and velocity information of the receiver calculated by receiving and processing signals sent by the global satellite navigation system (such as GPS, Beidou, GLONASS, etc.). The satellite navigation receiver receives signals with timestamps sent by multiple navigation satellites, calculates the distance from the receiver to each satellite by measuring the signal propagation time, and then calculates the precise position of the receiver through a three-dimensional space positioning algorithm. Satellite navigation data has the characteristics of global coverage and absolute positioning, can provide high-precision position and velocity information, and there is no error accumulation problem. The purpose of collecting satellite navigation data is to obtain a high-precision absolute position reference for correcting the cumulative error of the inertial navigation system.

[0031] Based on the above embodiments, as an alternative embodiment, obtaining the atmospheric information data includes:

[0032] Collecting atmospheric pressure, temperature, and humidity data at different spatial positions to obtain raw atmospheric data;

[0033] Based on the raw atmospheric data, establishing an airflow field model and performing feature analysis to obtain airflow field feature data;

[0034] Using the airflow field feature data to identify turbulence characteristics and obtain turbulence mode feature data;

[0035] Inputting the airflow field feature data and the turbulence mode feature data into a dynamic modeling algorithm to establish a time-varying model of atmospheric parameters and obtain atmospheric dynamic feature data;

[0036] Taking the atmospheric dynamic feature data as part of the multi-source navigation raw data.

[0037] Specifically, the process of obtaining the atmospheric information data first requires collecting atmospheric pressure, temperature, and humidity data at different spatial positions to obtain raw atmospheric data. This collection process is achieved through a multi-point distributed sensor array carried on the unmanned aerial vehicle (UAV), including high-precision barometric pressure sensors, temperature sensors, and humidity sensors. The barometric pressure sensor uses piezoresistive or capacitive sensing elements and can measure atmospheric pressure changes with an accuracy of 0.1 hPa; the temperature sensor uses a thermistor or a semiconductor thermometer with a measurement accuracy of 0.1 °C; the humidity sensor uses a capacitive humidity-sensitive element with a relative humidity measurement accuracy of 2%. These sensors are distributed at different positions on the UAV to form a spatial sampling network and continuously collect atmospheric parameters within the flight airspace. Collecting atmospheric data at different spatial positions is to capture the spatial distribution characteristics of atmospheric parameters, which have a strong correlation with geographical locations and provide environmental reference information for subsequent navigation.

[0038] Based on the obtained raw atmospheric data, the system establishes an airflow field model and performs feature analysis to obtain airflow field feature data. The airflow field model is a mathematical model that describes the airflow motion state within a given spatial region and is established by calculating the barometric pressure gradient, temperature gradient, and humidity gradient in combination with the principles of fluid mechanics. In specific implementation, the system uses a three-dimensional grid division method to discretize the sampling space, and then uses the Laplace equation and the fluid continuity equation to establish the spatial distribution functions of barometric pressure, temperature, and humidity. In the feature analysis stage, the main feature vectors of the airflow field are extracted through principal component analysis (PCA), the spectral characteristics of the airflow field are analyzed through Fourier transform, and the multi-scale characteristics of the airflow field are captured through wavelet transform. The obtained airflow field feature data contains key information such as the flow direction, flow velocity, and vorticity of the airflow, and these features have a deterministic association with geographical locations and time, providing important environmental context information for the navigation system.

[0039] Turbulence characteristics are identified using airflow field characteristic data to obtain turbulence mode characteristic data. Turbulence is a complex non-linear phenomenon in airflow motion, manifested as an irregular and random fluid motion state. Turbulence characteristics identification is achieved by analyzing parameters such as vorticity, velocity fluctuations, and energy spectra in the airflow field characteristic data. The system uses the Reynolds decomposition method to decompose the airflow velocity into the mean velocity and the pulsating velocity, calculates the variance, covariance, and autocorrelation function of the pulsating velocity, and extracts characteristics such as turbulence intensity, turbulence scale, and turbulence energy spectrum. At the same time, the system uses machine learning algorithms, such as support vector machine (SVM) or random forest, to classify the extracted characteristics and identify different types of turbulence modes, such as laminar flow, turbulence, and transition states. The turbulence mode characteristic data contains the intensity, distribution, and evolution characteristics of atmospheric disturbances, which are closely related to terrain, ground objects, and weather conditions, have obvious regional characteristics, and can provide additional position information for the navigation system.

[0040] The airflow field characteristic data and the turbulence mode characteristic data are input into a dynamic modeling algorithm to establish a time-varying model of atmospheric parameters and obtain atmospheric dynamic characteristic data. The dynamic modeling algorithm is a mathematical model construction method that can describe the time evolution characteristics of a system, and in this system, it is implemented by combining Kalman filtering and neural networks. First, the system uses a temporal convolutional neural network to process the airflow field characteristic data and the turbulence mode characteristic data to extract time-correlation characteristics; then, the extracted characteristics are input into a long short-term memory network (LSTM) to establish a prediction model with parameters varying over time; finally, the Kalman filter is used to correct and optimize the model output to form a time-varying model of atmospheric parameters. This model can accurately describe the variation law of atmospheric parameters over time and space and predict the atmospheric state at future moments. The obtained atmospheric dynamic characteristic data contains the spatio-temporal evolution law of atmospheric parameters, and these laws have a deterministic correspondence with geographical locations and navigation trajectories, providing rich environmental information for the navigation system.

[0041] Taking the atmospheric dynamic characteristic data as part of the multi-source navigation raw data is the basis for data fusion in the entire navigation system. These atmospheric dynamic characteristic data are synchronously processed with inertial navigation data and satellite navigation data through a unified data format and time stamp to jointly form a multi-source navigation raw data set. The purpose of this processing is to integrate atmospheric environment information with traditional navigation information and lay a data foundation for subsequent deep learning feature extraction and navigation fusion. By using the airflow field and turbulence characteristics as auxiliary navigation information, the system can rely on atmospheric environment characteristics to assist positioning when satellite navigation signals are interfered with or fail, significantly improving the environmental adaptability and anti-interference ability of the navigation system and ensuring the safe flight of unmanned aerial vehicles in complex environments.

[0042] S102. Perform deep learning feature extraction on the multi-source navigation raw data to obtain multi-modal feature data;

[0043] Specifically, in step S102, the system performs deep learning feature extraction on the multi-source navigation raw data to obtain multi-modal feature data. The multi-source navigation raw data includes different types of sensor information, such as inertial navigation data, satellite navigation data, and atmospheric information data. These data have different sampling frequencies, data dimensions, and physical meanings. To effectively utilize the information of these heterogeneous data, deep learning feature extraction technology needs to be used for processing. Deep learning feature extraction is a method that uses a multi-layer neural network to automatically learn data features, which can extract discriminative and expressive features from the raw data and is suitable for dealing with the fusion problem of multi-source heterogeneous data.

[0044] First, the system performs standardization processing on the multi-source navigation raw data to obtain a standardized data sequence. The standardization processing includes steps such as data cleaning, outlier detection and processing, time alignment, and normalization. Data cleaning removes noise through median filtering; outlier detection uses the 3σ criterion to identify and process outliers; time alignment uses an interpolation algorithm to unify data with different sampling frequencies to the same time base; normalization processing maps data with different dimensions to the same numerical range, usually using the Z-score normalization method, that is, converting the data into a distribution with a mean of 0 and a standard deviation of 1. The purpose of the standardization processing is to eliminate the scale difference and time asynchrony problems between different types of data and create conditions for subsequent feature extraction. The standardized data sequence retains the temporal characteristics of the raw data and has better numerical stability.

[0045] Next, the system uses a convolutional neural network to perform multi-scale feature extraction on the standardized data sequence to obtain initial feature data. A convolutional neural network is a deep learning model specifically used to process data with a grid structure. Through the characteristics of local connection and weight sharing, it can effectively capture the spatial features of the data. In this system, a one-dimensional convolutional neural network is used to process the temporal data, and the network structure includes multiple convolutional layers, pooling layers, and batch normalization layers. The convolutional layers use convolutional kernels of different sizes (such as 3×1, 5×1, 7×1) to extract features of different scales respectively; the pooling layers reduce the feature dimension through max pooling operations and improve the robustness of the model; the batch normalization layers accelerate the training convergence and improve the generalization ability of the model. The multi-scale feature extraction ability of the convolutional neural network enables the system to simultaneously capture the short-term fluctuations and long-term trends in the data and improve the expression ability of the features.

[0046] Based on the initial feature data, time series modeling is performed to obtain a time series feature sequence. Time series modeling is a key step in processing data with time dependence and is implemented using a Long Short-Term Memory network (LSTM) in this system. The Long Short-Term Memory network is a special type of recurrent neural network that has the ability to remember long-term dependencies and controls the flow of information through input gates, forget gates, and output gates. The system inputs the initial feature data extracted by the convolutional neural network into a bidirectional LSTM network in chronological order. The bidirectional LSTM can consider both past and future information simultaneously and model the time series relationship more comprehensively. The hidden layer dimension of the LSTM network is set to 128, and the network depth is 2 layers. The Dropout technique (dropout rate 0.3) is used to prevent overfitting. The purpose of time series modeling is to capture the time correlation of navigation data, which is of great significance for predicting the motion state and identifying navigation patterns.

[0047] Anomaly pattern recognition is performed on the time series feature sequence to obtain anomaly detection results. Anomaly pattern recognition is the process of discovering patterns in data that do not conform to the expected patterns and is of great significance for identifying sensor failures or environmental disturbances in the navigation system. The system uses an autoencoder-based anomaly detection method. The autoencoder is an unsupervised learning model that can discover anomaly patterns in data by learning the process of compressing and then reconstructing the input data. In the specific implementation, the system uses an autoencoder network composed of an encoder and a decoder. The encoder compresses the time series feature sequence into a low-dimensional latent space, and the decoder attempts to reconstruct the original features from the latent space. When the reconstruction error exceeds a preset threshold, the system considers this data point as an anomaly point. The anomaly detection results include the location, time, and degree of anomaly of the anomaly points, providing a data reliability assessment for subsequent navigation fusion.

[0048] Finally, the time series feature sequence and the anomaly detection results are fused to obtain multi-modal feature data. Feature fusion is achieved using an attention mechanism. The attention mechanism is a technique that can adaptively allocate the importance of different features, highlighting key information and suppressing irrelevant information. The system designs a multi-head self-attention module to calculate the correlation between different positions in the time series feature sequence and adjusts the attention weights through the anomaly detection results to reduce the impact of anomaly points. The fused multi-modal feature data combines the advantages of various types of sensor data, including the short-term high-precision characteristics of inertial navigation data, the absolute position information of satellite navigation data, and the environmental characteristics of atmospheric information data, forming a feature representation with rich information content. This multi-modal feature data provides high-quality input for subsequent navigation fusion, significantly improving the navigation accuracy and reliability of the system in complex environments.

[0049] Based on the above embodiments, as an alternative embodiment, the deep learning feature extraction of the multi-source navigation raw data to obtain multi-modal feature data includes:

[0050] S201. Standardize the multi-source navigation raw data to obtain a standardized data sequence;

[0051] Specifically, in step S201, the system standardizes the multi-source navigation raw data to obtain a standardized data sequence. The multi-source navigation raw data includes different types of data such as inertial navigation data, satellite navigation data, and atmospheric information data. These data have different physical dimensions, sampling frequencies, and value ranges. For example, the unit of acceleration in inertial navigation data is m / s², and the unit of angular velocity is rad / s; the position information in satellite navigation data is represented by longitude and latitude; the unit of air pressure in atmospheric information data is hPa, the unit of temperature is °C, and the humidity is in percentage. If these heterogeneous data are directly used in the deep learning model without standardization, it will lead to difficult model training and poor feature extraction effect. Therefore, standardization processing is required to make different types of data comparable and facilitate the subsequent model to learn the internal laws in the data.

[0052] The standardization process first performs data pre-cleaning to remove obvious outliers and noise. The system uses a sliding window median filtering algorithm to process various types of data. The window size is dynamically adjusted according to the data type. Usually, the window size for inertial navigation data is 5 sampling points, for satellite navigation data is 3 sampling points, and for atmospheric information data is 7 sampling points. Median filtering can effectively remove impulse noise and occasional outliers while keeping the edge characteristics of the data from being blurred. For example, for acceleration data, median filtering can remove the spike noise caused by mechanical vibration; for satellite navigation data, it can filter out the position jumps caused by multipath effects.

[0053] Next, the system performs time alignment processing to solve the problem of inconsistent sampling frequencies of different sensor data. The sampling frequency of inertial navigation data is usually 100Hz, that of satellite navigation data is 1 - 10Hz, and that of atmospheric information data is 1Hz. The system selects 10Hz as the unified sampling frequency, downsamples the high-frequency data, and performs interpolation processing on the low-frequency data. Downsampling uses the mean filtering method, that is, taking the average value of all data within a sampling period as the representative value of that period; interpolation processing uses the cubic spline interpolation algorithm, which can ensure the continuity of the first and second derivatives of the interpolation curve and generate a smoother data sequence. Time alignment ensures the consistency of different types of data in the time dimension and provides a basis for subsequent feature extraction.

[0054] Then, the system normalizes the aligned data, mapping data with different dimensions to a unified numerical range. The system adopts the Z-score normalization method, that is, for each data type, it calculates its mean μ and standard deviation σ, and then transforms the data x: x' = (x - μ) / σ. This transformation adjusts the data to a distribution with a mean of 0 and a standard deviation of 1. For inertial navigation data, the three-axis acceleration and three-axis angular velocity are normalized respectively; for satellite navigation data, the longitude, latitude, and altitude are normalized respectively; for atmospheric information data, the air pressure, temperature, and humidity are normalized respectively. The normalization process eliminates the dimensional differences between data, making different types of data have similar influences in the deep learning model and avoiding the situation where a certain type of data dominates the model training due to large numerical values.

[0055] Finally, the system organizes the processed data into a standardized data sequence in chronological order. This sequence is stored in matrix form, with each row representing the data at a time point and each column representing a specific data type. The standardized data sequence maintains the original data's temporal relationship, has a unified numerical range and sampling frequency, and creates a good data foundation for subsequent feature extraction. This processing method greatly improves the training efficiency and feature extraction effect of the deep learning model, enabling the model to better learn the internal laws and interrelationships in multi-source data and laying a data foundation for achieving high-precision navigation fusion.

[0056] S202. Use a convolutional neural network to perform multi-scale feature extraction on the standardized data sequence to obtain initial feature data, and perform temporal modeling based on the initial feature data to obtain a temporal feature sequence;

[0057] Specifically, in step S202, the system uses a convolutional neural network to perform multi-scale feature extraction on the standardized data sequence to obtain initial feature data, and performs temporal modeling based on the initial feature data to obtain a temporal feature sequence. A convolutional neural network is a deep learning model specifically designed to process data with a grid structure. Through mechanisms such as local receptive fields, weight sharing, and multi-layer feature extraction, it can effectively capture the spatial features and multi-scale patterns in the data. In navigation data processing, multi-scale feature extraction is particularly important because navigation data contains information at different time scales, such as the high-frequency vibration characteristics of inertial navigation data, the low-frequency position changes of satellite navigation data, and the medium-frequency environmental changes of atmospheric information data.

[0058] First, the system constructs a multi-branch convolutional neural network structure for multi-scale feature extraction. The network contains three parallel convolutional branches, each branch using convolutional kernels of different sizes, namely 3×1, 5×1, and 7×1, corresponding to capturing short-term, medium-term, and long-term temporal features respectively. Each convolutional branch consists of three convolutional layers, and after each convolution, a batch normalization layer and a ReLU activation function are followed. The batch normalization layer accelerates network convergence and improves generalization ability by normalizing the input of each layer; the ReLU activation function introduces non-linearity to enhance the expressive power of the network. The number of filters in the first convolutional layer is 32, the second layer is 64, and the third layer is 128. As the network depth increases, the extracted features gradually evolve from low-level to high-level. After each convolutional layer, the system also applies a max pooling operation with a pooling window size of 2 and a stride of 2 to reduce the feature dimension and extract the most significant features. Max pooling selects the maximum value within the window as the output, which enhances the model's invariance to small displacements and improves the robustness of the features.

[0059] After the processing of the three parallel branches is completed, the system concatenates the output feature maps of the three branches along the channel dimension to form multi-scale fusion features. To further improve the expressive power of the features, the system applies a 1×1 convolution to the fusion features for inter-channel information interaction, with the number of filters being 256. This operation is equivalent to weighted fusion of features at different scales, enhancing the expressive power of the features. Finally, the system converts the feature map into a fixed-length feature vector through global average pooling to obtain the initial feature data. Global average pooling takes the average value of each feature channel, greatly reducing the number of parameters and effectively preventing overfitting. The initial feature data contains multi-scale spatio-temporal features in the standardized data sequence, but the temporal dependence relationship of the data has not been fully utilized.

[0060] Next, the system performs temporal modeling based on the initial feature data to obtain a temporal feature sequence. The purpose of temporal modeling is to capture the temporal dependence relationship between data, which is crucial for accurately predicting the navigation state. The system uses a long short-term memory network (LSTM) for temporal modeling. LSTM is a special recurrent neural network that solves the problems of gradient vanishing and gradient explosion in traditional recurrent neural networks through a gating mechanism and can effectively learn long-term dependence relationships in long sequence data. The LSTM unit contains three gates: an input gate, a forget gate, and an output gate, as well as a memory unit. The input gate controls the degree to which new information enters the memory unit; the forget gate determines how much old information to discard; the output gate controls how much information in the memory unit is output. This structural design enables LSTM to selectively remember and forget information and effectively handle long-term dependence problems.

[0061] In specific implementation, the system reorganizes the initial feature data in chronological order to form a sequence input and feeds it into the LSTM network. The LSTM network consists of two layers, with 128 hidden units in each layer, and adopts a bidirectional LSTM structure. The bidirectional LSTM contains two LSTM layers, a forward LSTM and a backward LSTM. The forward LSTM processes the information flow from the past to the future, and the backward LSTM processes the information flow from the future to the past. The outputs of the two are concatenated at each time step to form a feature representation containing complete context information. This bidirectional processing method enables the model to utilize both past and future information and comprehensively understand the time-series data. To prevent overfitting, the system adds a Dropout layer between the LSTM layers, with a dropout rate set to 0.3. Dropout prevents the model from overfitting the training data by randomly discarding a portion of neurons during the training process.

[0062] After the LSTM network finishes processing, the system obtains the hidden state sequence at each time step, i.e., the time-series feature sequence. The time-series feature sequence not only contains the multi-scale spatial features of the original data but also encodes the temporal dependencies between the data, providing a high-quality feature representation for subsequent anomaly detection and feature fusion. This method that combines the multi-scale feature extraction of the convolutional neural network and the temporal modeling of the LSTM fully utilizes the advantages of deep learning in feature extraction, can automatically learn the complex patterns in the data without manual feature design, and greatly improves the adaptability and robustness of the navigation system in complex environments.

[0063] S203. Perform anomaly pattern recognition on the time-series feature sequence to obtain an anomaly detection result, and fuse the time-series feature sequence and the anomaly detection result to obtain multi-modal feature data.

[0064] Specifically, in step S203, the system performs anomaly pattern recognition on the time-series feature sequence to obtain an anomaly detection result, and fuses the time-series feature sequence and the anomaly detection result to obtain multi-modal feature data. Anomaly pattern recognition is of great significance in the navigation system because during actual flight, sensor data may be affected by external interference, equipment failures, or environmental changes, resulting in outliers or abnormal patterns. If these anomalies are not recognized and processed in a timely manner, they will have a negative impact on navigation accuracy and may even cause the navigation system to fail. Therefore, it is necessary to perform anomaly pattern recognition on the time-series feature sequence to identify potential anomaly points and appropriately process them in subsequent processing.

[0065] First, the system uses an autoencoder-based method for anomaly pattern recognition. An autoencoder is an unsupervised learning model consisting of an encoder and a decoder. The encoder compresses the input data into a low-dimensional latent space, and the decoder attempts to reconstruct the original input from the latent space. The training objective of the autoencoder is to minimize the reconstruction error, i.e., to make the reconstructed data as similar as possible to the original input. An autoencoder trained on normal data can reconstruct normal patterns well, but for abnormal data, since the model has not learned these patterns, the reconstruction error is usually large. Utilizing this property, the system identifies anomalies by calculating the reconstruction error.

[0066] S103. Perform quality assessment on the satellite navigation data to obtain the availability result of the satellite navigation signal;

[0067] Specifically, in step S103, the system performs quality assessment on the satellite navigation data to obtain the availability result of the satellite navigation signal. The quality assessment of satellite navigation data is a key link in the entire navigation fusion system because satellite navigation signals are vulnerable to factors such as occlusion, multipath effects, and ionospheric interference in complex environments, resulting in a decrease in positioning accuracy or even complete failure. Accurately assessing the availability of satellite navigation signals is an important basis for the system to decide whether to enable the backup navigation mode. If the quality assessment of satellite navigation signals is not performed, the system may still use them for navigation when the satellite navigation signals are already unreliable, resulting in a significant decrease in navigation accuracy and even endangering the flight safety of the UAV.

[0068] First, the system obtains the signal-to-noise ratio (SNR) and carrier-to-noise ratio (CNR) data of the satellite navigation data. The signal-to-noise ratio refers to the ratio of the satellite signal power to the noise power, usually in decibels (dB); the carrier-to-noise ratio refers to the ratio of the carrier power to the noise power. These two parameters are direct indicators of the quality of satellite signals, and the higher the value, the better the signal quality. The system obtains the SNR and CNR of each visible satellite through the original observation data of the satellite navigation receiver. For GPS satellites, the system obtains the SNR of the L1 and L2 frequency bands; for Beidou satellites, the system obtains the SNR of the B1 and B2 frequency bands; for Galileo satellites, the system obtains the SNR of the E1 and E5a frequency bands. The CNR data is usually directly provided by the receiver or extracted from the original signal through signal processing algorithms.

[0069] Next, the system calculates the signal strength index based on the signal-to-noise ratio and carrier-to-noise ratio data. The signal strength index is a comprehensive index that reflects the overall quality of satellite signals. The system uses a weighted average method to calculate the signal strength index, and the formula is: Signal strength index = w1 × average signal-to-noise ratio + w2 × average carrier-to-noise ratio, where w1 and w2 are weight coefficients, set through experience, usually w1 = 0.6 and w2 = 0.4. For multi-band receivers, the system first calculates the signal strength index for each band, and then takes the minimum value as the final index for the satellite. The purpose of doing this is to ensure that the signals in each band meet the quality requirements. The system also sets a signal strength threshold, usually 25 dB. When the signal strength index is lower than this threshold, it is considered that the signal strength is insufficient, which may affect the positioning accuracy.

[0070] Then, the system analyzes the satellite signal quality based on the signal strength index to obtain the signal reliability evaluation value. The signal reliability evaluation not only considers the signal strength, but also the stability and consistency of the signal. The system evaluates the stability of the signal by calculating the variance of the signal strength index over a period of time (usually 10 - 30 seconds) in the past; and evaluates the consistency of the signal by comparing the signal strength differences between different satellites. The system comprehensively considers the three factors of signal strength, stability, and consistency, and calculates the signal reliability evaluation value. The value range of this value is 0 - 1, and the larger the value, the more reliable the signal. The specific calculation formula is: Signal reliability evaluation value = (Signal strength index / Maximum signal strength) × (1 - Standardized variance) × Consistency factor. Among them, the consistency factor is calculated based on the standard deviation of the signal strengths of different satellites. The smaller the standard deviation, the better the consistency, and the closer the factor value is to 1.

[0071] Next, the system counts the number and distribution of currently visible satellites to obtain the satellite geometric structure parameters. The satellite geometric structure parameters describe the distribution of visible satellites in space and are important factors affecting positioning accuracy. The system first counts the number of visible satellites with reliable signals. Usually, at least four satellites are required for three-dimensional positioning. Then, the system calculates the Position Dilution of Precision (PDOP), Horizontal Dilution of Precision (HDOP), and Vertical Dilution of Precision (VDOP). The Position Dilution of Precision is an index that measures the impact of satellite geometric distribution on positioning accuracy. The smaller the value, the better the geometric distribution and the higher the positioning accuracy. The Horizontal Dilution of Precision reflects the impact of satellite distribution on horizontal positioning accuracy; the Vertical Dilution of Precision reflects the impact of satellite distribution on height measurement accuracy. These dilution of precision factors are calculated through the geometric relationship between the satellite positions and the receiver position. The system also analyzes the distribution uniformity of satellites in the sky by calculating the standard deviation of the azimuth angles of visible satellites and the coverage range of elevation angles.

[0072] Finally, the system conducts a comprehensive analysis based on the signal reliability evaluation value and satellite geometric structure parameters to obtain the GNSS signal availability result. The GNSS signal availability result is a binary judgment indicating whether the current GNSS signal is reliable for navigation. The system sets a series of judgment criteria: the signal reliability evaluation value must be greater than 0.7; the number of available satellites must be greater than or equal to 5; the position dilution of precision must be less than 6; the horizontal dilution of precision must be less than 4; the vertical dilution of precision must be less than 8; the satellite distribution uniformity must satisfy that the azimuth standard deviation is greater than 60 degrees and at least one satellite has an elevation angle greater than 60 degrees. Only when all these conditions are met will the system determine that the GNSS signal is available. Otherwise, the system will determine that the GNSS signal is unavailable and the backup navigation mode needs to be activated.

[0073] Based on the above embodiments, as an alternative embodiment, the quality assessment of the satellite navigation data to obtain the GNSS signal availability result includes:

[0074] S301, obtain the signal-to-noise ratio and carrier-to-noise ratio data of the satellite navigation data, and calculate a signal strength index based on the signal-to-noise ratio and the carrier-to-noise ratio data;

[0075] Specifically, obtain the signal-to-noise ratio and carrier-to-noise ratio data of the satellite navigation data, and calculate a signal strength index based on the signal-to-noise ratio and the carrier-to-noise ratio data. The system obtains the signal-to-noise ratio and carrier-to-noise ratio of each visible satellite through a satellite navigation receiver. The signal-to-noise ratio is the ratio of the satellite signal power to the noise power, usually expressed in decibels (dB), reflecting the strength of the signal relative to the background noise; the carrier-to-noise ratio is the ratio of the carrier power to the noise power, which is another important signal quality indicator. For multi-system multi-band receivers, the system obtains the signal-to-noise ratio and carrier-to-noise ratio of different frequency bands (such as L1 / L2, B1 / B2, E1 / E5a) of different satellite systems (such as GPS, Beidou, Galileo) respectively. The system calculates the signal strength index using the weighted average method. The formula is: signal strength index = α × normalized signal-to-noise ratio + β × normalized carrier-to-noise ratio, where α and β are weight coefficients, and the optimal values are determined through experiments. For multi-band signals, the minimum value of the signal strength index of each frequency band is taken as the final signal strength index of the satellite, which can ensure that the signals of each frequency band meet the quality requirements. The calculation of the signal strength index provides a quantitative basis for subsequent signal quality analysis.

[0076] S302, analyze the satellite signal quality based on the signal strength index to obtain a signal reliability evaluation value;

[0077] Specifically, the satellite signal quality is analyzed based on the signal strength index to obtain a signal reliability evaluation value. The signal reliability evaluation value is a quantitative representation of the overall reliability of the satellite signal, ranging from 0 to 1, and the larger the value, the more reliable the signal. The system first sets a basic signal strength threshold (usually 25 dB), compares the signal strength index of each satellite with this threshold, and calculates the signal strength ratio. Then, the system analyzes the signal stability by calculating the variance of the signal strength index within a past time window (usually 30 seconds). The smaller the variance, the more stable the signal. The system normalizes the variance to obtain a signal stability index. The system also analyzes the dynamic characteristics of the signal, including Doppler frequency shift and carrier phase change, which reflect the changes in the signal propagation path. Finally, the system comprehensively considers the signal strength ratio, signal stability index, and dynamic characteristics, and calculates the signal reliability evaluation value by weighted summation. This multi-factor evaluation method can comprehensively reflect the quality of the satellite signal and provide a reliable basis for subsequent comprehensive analysis.

[0078] S303. Statistically analyze the number and distribution of currently visible satellites to obtain satellite geometric structure parameters;

[0079] Specifically, the system statistically analyzes the number and distribution of currently visible satellites to obtain satellite geometric structure parameters. The satellite geometric structure parameters describe the distribution of visible satellites in space and are key factors affecting positioning accuracy. The system first counts the number of visible satellites with reliable signals (i.e., the number of satellites whose signal reliability evaluation value exceeds the set threshold), and then calculates the geometric dilution of precision based on the spatial positions of these satellites, including the position dilution of precision (PDOP), horizontal dilution of precision (HDOP), and vertical dilution of precision (VDOP). The position dilution of precision is the square root of the sum of the squares of the diagonal elements after inverting the matrix composed of the unit vectors from the receiver to the satellites, reflecting the impact of the satellite geometric distribution on three-dimensional positioning accuracy; the horizontal dilution of precision and vertical dilution of precision respectively reflect the impact of satellite distribution on horizontal positioning and altitude measurement. The system also analyzes the uniformity of satellite distribution in the sky by calculating the standard deviation of the azimuth angles of visible satellites and the elevation coverage. The larger the azimuth standard deviation, the more uniform the satellite distribution in the horizontal direction; the wider the elevation coverage, the more reasonable the satellite distribution in the vertical direction. These satellite geometric structure parameters jointly reflect the impact of the current satellite distribution on positioning accuracy and provide an important basis for judging the availability of GNSS signals.

[0080] S304. Conduct a comprehensive analysis based on the signal reliability evaluation value and satellite geometric structure parameters to obtain the GNSS signal availability result.

[0081] Specifically, through comprehensive analysis of the signal reliability evaluation value and satellite geometric structure parameters, the availability result of the GNSS signal is obtained. The availability result of the GNSS signal is a binary judgment indicating whether the current GNSS signal is reliable for navigation. The system sets a series of judgment conditions: the signal reliability evaluation value must be greater than a preset threshold (usually 0.8); the number of reliable satellites must meet the minimum requirement (usually more than 5); the dilution of precision (DOP) must be less than the upper limit (usually 6); the horizontal dilution of precision (HDOP) must be less than the upper limit (usually 4); the vertical dilution of precision (VDOP) must be less than the upper limit (usually 8); the satellite distribution uniformity must meet the condition that the azimuth standard deviation is greater than the lower limit (usually 60 degrees) and at least one satellite elevation angle is greater than the upper limit (usually 60 degrees). The system designs a fuzzy logic decision-making model to map the satisfaction degree of these conditions to the interval [0,1], and then obtains the final availability score through weighted summation. If the availability score exceeds the set threshold (usually 0.85), the system determines that the GNSS signal is available; otherwise, it determines that the GNSS signal is unavailable. This comprehensive multi-factor fuzzy decision-making method is more robust than simple threshold judgment, can more accurately evaluate the actual availability of the GNSS signal, provides a reliable basis for switching the working mode of the navigation system, and thus ensures the continuity and reliability of navigation in complex environments.

[0082] S104, when the availability result of the GNSS signal indicates that the GNSS information is valid: input the multi-modal feature data into the federated filtering algorithm, perform integrated navigation solution on the inertial navigation data and the GNSS information to obtain an initial navigation result; estimate and model the error amount in the atmospheric information based on the initial navigation result to obtain an atmospheric information error model;

[0083] Specifically, when the availability result of the GNSS signal indicates that the GNSS information is valid, the system first inputs the multi-modal feature data into the federated filtering algorithm, performs integrated navigation solution on the inertial navigation data and the GNSS information to obtain an initial navigation result; then estimates and models the error amount in the atmospheric information based on the initial navigation result to obtain an atmospheric information error model. This step is a key link executed under the condition of good GNSS signals, which not only provides high-precision navigation results but also prepares for the backup navigation mode when the GNSS signal fails.

[0084] Based on the above embodiments, as an alternative embodiment, the step of inputting the multi-modal feature data into the federated filtering algorithm, performing integrated navigation solution on the inertial navigation data and the GNSS information to obtain an initial navigation result includes:

[0085] S401, separate the multi-modal feature data according to the data source to obtain subsystem observation data, and respectively construct filtering models for the subsystem observation data to obtain subsystem state equations;

[0086] Specifically, the multi-modal feature data is separated according to the data source to obtain subsystem observation data, and filtering models are respectively constructed for the subsystem observation data to obtain subsystem state equations. The multi-modal feature data is a comprehensive feature representation obtained through deep learning feature extraction, including features from multiple data sources such as inertial navigation, satellite navigation, and atmospheric information. Separating these features according to the data source is the first step of the federated filtering algorithm, aiming to achieve distributed processing and improve the computational efficiency and robustness of the system. The system first projects the multi-modal feature data into different feature subspaces using a feature mapping matrix to obtain feature subsets corresponding to inertial navigation and satellite navigation. Then, the system constructs an inertial subsystem model for the features related to inertial navigation, including the state equation of a 15-dimensional state vector (3-dimensional attitude error, 3-dimensional velocity error, 3-dimensional position error, 3-dimensional gyroscope zero bias, and 3-dimensional accelerometer zero bias). This state equation uses the inertial navigation error propagation model to describe the evolution law of the error state over time. For the features related to satellite navigation, the system constructs a satellite navigation subsystem model, including position and velocity observation equations, which describe the mapping relationship between the observed quantities and the state vector. These two subsystem models together form the basis of the federated filtering algorithm, enabling the system to independently process different data sources and improving the flexibility and fault tolerance of the algorithm.

[0087] S402, perform state estimation based on the subsystem state equation to obtain a locally optimal estimated value, and calculate the confidence level of the locally optimal estimated value to obtain an information fusion weight;

[0088] Specifically, perform state estimation based on the subsystem state equation to obtain a locally optimal estimated value, and calculate the confidence level of the locally optimal estimated value to obtain an information fusion weight. In this step, the system uses an extended Kalman filter to independently perform state estimation for each subsystem. The extended Kalman filter is an effective tool for dealing with nonlinear systems. It linearly approximates the nonlinear system near the current operating point through linearization techniques and then applies the standard Kalman filtering algorithm. For the inertial subsystem, the filter performs two stages: time update and measurement update. In the time update stage, the state and error covariance at the next moment are predicted through the state equation. In the measurement update stage, the prediction result is corrected using the observed data, the filtering gain is calculated, and the state estimation and error covariance are updated. Similarly, the satellite navigation subsystem also performs the same filtering process. After completing the local state estimation, the system calculates the confidence level of the estimation result of each subsystem, that is, the information fusion weight. The calculation method is based on the error covariance matrix of the subsystem. The inverse matrix of the covariance matrix reflects the estimation accuracy. The larger the value, the higher the accuracy and the corresponding weight is also larger. The system uses the information matrix (the inverse of the covariance matrix) as the basis for the weight and combines a dynamic weight adjustment strategy to dynamically adjust the weight according to the working state and environmental conditions of the subsystem, improving the self-adaptability of the fusion.

[0089] S403. Perform global state estimation according to the information fusion weights to obtain an initial navigation result.

[0090] Specifically, perform global state estimation according to the information fusion weights to obtain an initial navigation result. Under the framework of the federated filtering algorithm, global state estimation is a process of fusing the local estimation results of each subsystem into a unified global optimal estimation. The system adopts a weighted information fusion method. Based on the information fusion weights calculated in the previous step, it fuses the state estimations and error covariances of each subsystem. Specifically, the global state estimation value is equal to the weighted average of the state estimation values of each subsystem, and the weights are the corresponding information fusion weights; the global error covariance is equal to the inverse of the weighted sum of the error covariances of each subsystem. This fusion method theoretically ensures the optimality of the global estimation, can make full use of the advantages of each subsystem, and improve the estimation accuracy. In addition, the system also implements a feedback mechanism, feeding back the global fusion result to each subsystem as the initial value for filtering at the next moment, maintaining the consistency of the entire system. In this way, the system obtains an initial navigation result including position, velocity, and attitude. This result fully integrates the short-term high accuracy of inertial navigation and the long-term stability of satellite navigation, significantly improving the navigation accuracy and reliability while meeting the real-time requirements. Experiments show that compared with the traditional loose-coupling integrated navigation method, the position accuracy of this method is improved by about 30%, and the attitude accuracy is improved by about 20%, showing obvious advantages especially in the case of unstable satellite signals.

[0091] Based on the above embodiments, as an alternative embodiment, estimating and modeling the error amount in the atmospheric information data based on the initial navigation result to obtain an atmospheric information error model includes:

[0092] S501. Use the initial navigation result as a reference value to extract the atmospheric parameters at the corresponding space-time points to obtain reference data;

[0093] Specifically, taking the initial navigation result as the reference value, the atmospheric parameters at the corresponding space-time points are extracted to obtain reference data. The initial navigation result is high-precision navigation information obtained by combining inertial navigation data and satellite navigation information through a federated filtering algorithm, which includes parameters such as position, velocity, and attitude. Since the availability result of the satellite navigation signal indicates that the satellite navigation information is valid, the initial navigation result has high accuracy and reliability and can be used as a benchmark for estimating the error of atmospheric information. The system extracts the theoretical atmospheric parameters at the corresponding space-time points from the standard atmospheric model according to the position information (longitude, latitude, altitude) and time information in the initial navigation result. The standard atmospheric model is a physical model that describes the variation law of atmospheric parameters with height and geographical location. It is established based on a large number of historical observation data and includes the standard distribution of parameters such as air pressure, temperature, and humidity. The standard atmospheric model adopted by the system combines the International Standard Atmosphere (ISA) and regional meteorological data and can provide more accurate theoretical values. In addition, the system also considers the influence of factors such as seasonal variation, diurnal variation, and local terrain on atmospheric parameters and improves the accuracy of theoretical values by correcting the model. In this way, the system obtains the theoretical atmospheric parameters corresponding to the current position and time as reference data, providing a benchmark for subsequent error analysis.

[0094] S502, calculate the deviation between the atmospheric information data and the reference data to obtain an error sample sequence, and perform statistical feature analysis on the error sample sequence to obtain the error distribution characteristics;

[0095] Specifically, the deviation between the atmospheric information data and the reference data is calculated to obtain an error sample sequence, and the statistical characteristics of the error sample sequence are analyzed to obtain the error distribution characteristics. The system compares the actually observed atmospheric information data with the reference data obtained in step S501, calculates the difference between the two, and obtains the error samples of the atmospheric information. These error samples reflect the deviation between the actual atmospheric conditions and the theoretical model, and contain the comprehensive influence of factors such as local meteorological conditions, environmental disturbances, and sensor errors. The system collects the error samples within a period of time through the sliding window technique to form an error sample sequence. The window size is usually set to 300 - 600 seconds, which can capture the time-varying characteristics of the errors. For the collected error sample sequence, the system conducts a comprehensive statistical characteristic analysis to extract the statistical laws of the errors. The analysis content includes: calculating basic statistics such as the mean, variance, and standard deviation to describe the central tendency and dispersion degree of the errors; performing probability distribution fitting to determine whether the errors conform to the normal distribution or other probability distributions; calculating the autocorrelation function and partial autocorrelation function to analyze the time correlation of the errors; conducting power spectrum analysis to study the frequency domain characteristics of the errors; analyzing the correlation between the errors and factors such as location, altitude, and time to establish a spatial distribution model of the errors. Through these analyses, the system obtains comprehensive error distribution characteristics, reveals the statistical laws and variation characteristics of the atmospheric information errors, and lays a foundation for establishing an error compensation function.

[0096] S503. Based on the error distribution characteristics, an error compensation function is established to obtain an atmospheric information error model.

[0097] Specifically, an error compensation function is established based on the error distribution characteristics to obtain an atmospheric information error model. The error compensation function is a mathematical expression that describes the relationship between the atmospheric information error and various influencing factors and is the core of realizing the atmospheric information error compensation. The system designs a suitable form of the error compensation function according to the error distribution characteristics obtained in step S502. Usually, the system divides the error compensation function into two parts: a deterministic component and a random component. The deterministic component is used to describe the systematic errors related to factors such as position and altitude and is represented by a polynomial function or a spline function. For example, for the pressure error, the system establishes a two-dimensional polynomial model based on altitude and latitude; for the temperature error, the system uses a radial basis function (RBF) network to establish a spatial distribution model. The random component is used to describe the time-varying random errors and is usually represented by a time series model, such as an autoregressive moving average model (ARMA) or an autoregressive integrated moving average model (ARIMA). The system estimates the model parameters through optimization methods such as the least squares method and the maximum likelihood estimation to minimize the difference between the error values predicted by the model and the actual error samples. To improve the generalization ability of the model, the system uses cross-validation techniques to evaluate the model performance and uses regularization methods to prevent overfitting. Finally, the system obtains a complete atmospheric information error model, which can predict the error amount of the atmospheric information based on the current position, altitude, and time, providing a basis for subsequent atmospheric information compensation when the satellite navigation signal is invalid. Experimental verification shows that the prediction accuracy of the model for the atmospheric information error reaches more than 85%, which can effectively improve the usability of the atmospheric information in navigation and provide strong support for backup navigation when the satellite navigation signal fails.

[0098] S105. When the satellite navigation signal availability result indicates that the satellite navigation information is invalid: use the atmospheric information error model to compensate the real-time atmospheric information to obtain the compensated atmospheric information; fuse and position the compensated atmospheric information with the inertial navigation data to obtain a high-precision navigation result.

[0099] Specifically, when the satellite navigation signal availability result indicates that the satellite navigation information is invalid, the system executes a backup navigation strategy to ensure continuous provision of reliable navigation information in the event of satellite navigation failure. This situation usually occurs when an unmanned aerial vehicle (UAV) flies in an urban canyon, a dense forest, or an area affected by human interference, where the satellite signal is severely blocked or interfered, resulting in the satellite navigation system being unable to provide accurate position information. In this case, relying on a traditional inertial navigation system will lead to error accumulation, and the position accuracy will rapidly decline over time. Therefore, other auxiliary information needs to be used for navigation enhancement.

[0100] Based on the above embodiments, as an alternative embodiment, the fusing and positioning the compensated atmospheric information with the inertial navigation data to obtain a high-precision navigation result includes:

[0101] S601. Perform parameter conversion on the compensated atmospheric information, extract the components related to position, velocity, and attitude, and obtain the atmospheric navigation parameters;

[0102] Specifically, perform parameter conversion on the compensated atmospheric information, extract the components related to position, velocity, and attitude, and obtain the atmospheric navigation parameters. The compensated atmospheric information is high-precision atmospheric data obtained by compensating the error of real-time atmospheric information using the atmospheric information error model, including parameters such as air pressure, temperature, and humidity. Although these original atmospheric parameters are related to geographical location, they cannot be directly used for navigation positioning and need to be converted into parameters related to navigation. The system first converts the compensated air pressure data into air pressure altitude using the air pressure altitude formula. The air pressure altitude formula is based on the law of the change of atmospheric pressure with altitude, and its expression is h = c × T × ln(P0 / P), where h is the altitude, P is the current air pressure, P0 is the standard sea-level air pressure (1013.25 hPa), T is the average temperature, and c is a constant. The system corrects the formula using the compensated temperature data to improve the accuracy of altitude calculation. Then, the system estimates the horizontal position change by analyzing the spatial gradients of air pressure, temperature, and humidity. The system differentiates the atmospheric parameters at adjacent times and combines the spatial distribution model of atmospheric parameters to calculate the position change amount. In addition, the system also analyzes the wind direction and wind speed using the characteristics of the airflow field to assist in estimating the flight speed and heading. Through these conversions and analyses, the system extracts the navigation parameters related to position, velocity, and attitude from the compensated atmospheric information to form an atmospheric navigation parameter set. Although the accuracy of these parameters is not as high as that of direct navigation measurements, they can provide valuable navigation reference information in the case of the failure of satellite navigation signals.

[0103] S602. Combine the atmospheric navigation parameters and inertial navigation data, establish an error state equation and an observation equation to obtain a fusion navigation model, and design a federated Kalman filter according to the fusion navigation model to obtain an optimal state estimator;

[0104] Specifically, by combining atmospheric navigation parameters and inertial navigation data, an error state equation and an observation equation are established to obtain a fusion navigation model, and a federated Kalman filter is designed based on the fusion navigation model to obtain an optimal state estimator. In the case where GNSS signals are invalid, the system needs to make full use of the available inertial navigation data and atmospheric navigation parameters, describe the relationship between them through a mathematical model, and achieve data fusion. The system first establishes an inertial navigation error state equation, which describes the evolution law of the error state of the inertial navigation system over time. The system selects a 15-dimensional state vector, including attitude errors (3 dimensions), velocity errors (3 dimensions), position errors (3 dimensions), gyroscope biases (3 dimensions), and accelerometer biases (3 dimensions). The error state equation is represented in continuous time form as X'(t) = F(t)X(t) + G(t)W(t), where X(t) is the state vector, F(t) is the system matrix, G(t) is the noise distribution matrix, and W(t) is the system noise. The system matrix F(t) is derived based on the inertial navigation error propagation theory and includes the effects of factors such as the Earth's rotation and Coriolis force. Then, the system establishes an observation equation, which describes the relationship between the atmospheric navigation parameters and the state vector, expressed as Z(t) = H(t)X(t) + V(t), where Z(t) is the observation vector, H(t) is the observation matrix, and V(t) is the observation noise. The observation vector includes the difference between the atmospheric navigation parameters and the inertial navigation output, reflecting the errors of the inertial navigation. Based on these two equations, the system obtains a complete fusion navigation model, which organically combines atmospheric information and inertial navigation, providing a theoretical basis for subsequent state estimation.

[0105] Based on the integrated navigation model, the system designs a federated Kalman filter to obtain an optimal state estimator. The federated Kalman filter is a distributed estimation algorithm, which is particularly suitable for dealing with multi-source information fusion problems. The system designs a federated architecture including a main filter and multiple sub-filters. Among them, the inertial sub-filter processes inertial navigation data, and the atmospheric sub-filter processes atmospheric navigation parameters. The system optimally designs the parameters of these filters, including the state transition matrix, the observation matrix, the process noise covariance matrix, and the measurement noise covariance matrix. The state transition matrix is obtained by discretizing the continuous-time system matrix F(t); the observation matrix is determined according to the relationship between the atmospheric navigation parameters and the state vector; the process noise covariance matrix is set according to the performance parameters of the inertial sensors; the measurement noise covariance matrix is determined according to the uncertainty of the atmospheric navigation parameters. The system assigns information weights to each sub-filter through an information distribution strategy, and adopts an adaptive weight calculation method to dynamically adjust the weights according to the working state and data reliability of the subsystems. The system also designs an information feedback mechanism to realize the information interaction between the sub-filter and the main filter and ensure global consistency. Through these designs, the system obtains a high-performance optimal state estimator, which can effectively fuse inertial navigation data and atmospheric navigation parameters and provide accurate state estimation.

[0106] S603, using the optimal state estimator to correct the inertial navigation data to obtain a high-precision navigation result.

[0107] Specifically, use the optimal state estimator to correct the inertial navigation data to obtain a high-precision navigation result. In this step, the system applies the output of the optimal state estimator to the inertial navigation system to correct its cumulative error and improve the navigation accuracy. Specifically, the system first obtains the error state estimation values obtained by the optimal state estimator, including attitude error, velocity error, position error, etc. Then, the system subtracts these error estimation values from the original output of the inertial navigation system to achieve error compensation. For the attitude error, the system uses the quaternion correction method for correction to ensure the orthogonality of the attitude representation; for the velocity and position errors, the system directly performs algebraic correction. The system also uses the estimated gyroscope zero bias and accelerometer zero bias to perform real-time compensation on the output of the inertial measurement unit to reduce the error source. In addition, the system implements an error feedback mechanism to feedback the estimated error information to the inertial navigation algorithm to optimize the navigation solution process. Through these corrections and optimizations, the system significantly improves the accuracy and stability of inertial navigation, effectively suppressing the problem of error accumulation. The finally obtained high-precision navigation result includes position, velocity, and attitude information, which can meet the navigation requirements of the UAV in a complex environment. Experimental verification shows that in the case of complete failure of the satellite navigation signal, compared with pure inertial navigation, the cumulative position error within 30 minutes is reduced by more than 65%, providing reliable navigation guarantee for the UAV and greatly improving the adaptability and safety of the system in a complex environment.

[0108] The above satellite-aided atmospheric fusion positioning system based on an inertial navigation system may include:

[0109] A data acquisition module 1, configured to acquire inertial navigation data, satellite navigation data, and atmospheric information data to obtain multi-source navigation raw data;

[0110] A feature extraction module 2, configured to perform deep learning feature extraction on the multi-source navigation raw data to obtain multi-modal feature data;

[0111] A quality assessment module 3, configured to perform quality assessment on the satellite navigation data to obtain a satellite-aided signal availability result;

[0112] An atmospheric error model construction module 4, configured to: when the satellite-aided signal availability result indicates that the satellite-aided information is valid, input the multi-modal feature data into a federated filtering algorithm, perform integrated navigation solution on the inertial navigation data and the satellite-aided information to obtain an initial navigation result; estimate and model the error amount in the atmospheric information based on the initial navigation result to obtain an atmospheric information error model;

[0113] A fusion positioning module 5, configured to: when the satellite-aided signal availability result indicates that the satellite-aided information is invalid, compensate the real-time atmospheric information using the atmospheric information error model to obtain compensated atmospheric information; perform fusion positioning on the compensated atmospheric information and the inertial navigation data to obtain a high-precision navigation result.

[0114] It should be noted that: when the system provided in the above embodiment realizes its functions, only the division of the above functional modules is used for illustration. In actual applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be elaborated here.

[0115] The above are only exemplary embodiments of the present disclosure, and the scope of the present disclosure cannot be limited thereby. That is, any equivalent changes and modifications made in accordance with the teachings of the present disclosure still fall within the scope covered by the present disclosure. Those skilled in the art will easily think of other implementation schemes of the present disclosure after considering the specification and the disclosure of the practical truth.

[0116] This application aims to cover any variations, uses, or adaptive changes of the present disclosure, which follow the general principles of the present disclosure and include common general knowledge or conventional technical means in the technical field not recorded in the present disclosure. The specification and embodiments are only regarded as exemplary, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A satellite navigation atmosphere fusion positioning method based on an inertial navigation system, characterized in that: The method comprises: Acquire inertial navigation data, satellite navigation data and atmospheric information data to obtain multi-source navigation raw data; Performing deep learning feature extraction on the multi-source navigation raw data to obtain multimodal feature data; Performing a quality assessment on the satellite navigation data to obtain a satellite navigation signal availability result; The performing quality assessment on the satellite navigation data to obtain a satellite navigation signal availability result includes: obtaining signal-to-noise ratio and carrier-to-noise ratio data of the satellite navigation data, and calculating a signal strength index based on the signal-to-noise ratio and the carrier-to-noise ratio data; analyzing satellite signal quality based on the signal strength index to obtain a signal reliability assessment value; counting the number and distribution of currently visible satellites to obtain satellite geometric structure parameters; and performing a comprehensive analysis based on the signal reliability assessment value and the satellite geometric structure parameters to obtain a satellite navigation signal availability result; When the satellite navigation signal availability result indicates that the satellite navigation information is valid: inputting the multimodal feature data into a federated filtering algorithm, performing combined navigation solution on the inertial navigation data and the satellite navigation information, and obtaining an initial navigation result; estimating and modeling the error amount in the atmospheric information based on the initial navigation result, and obtaining an atmospheric information error model; When the satellite navigation signal availability result indicates that the satellite navigation information is invalid: the real-time atmospheric information is compensated by using the atmospheric information error model to obtain compensated atmospheric information; the compensated atmospheric information is fused with the inertial navigation data for positioning to obtain a high-precision navigation result.

2. The method according to claim 1, characterized in that: Acquiring the atmospheric information data includes: Collect atmospheric pressure, temperature and humidity data at different spatial locations to obtain raw atmospheric data; Establishing an airflow field model based on the original atmospheric data, and performing characteristic analysis to obtain airflow field characteristic data; Using the airflow field characteristic data to perform turbulence feature recognition to obtain turbulence pattern characteristic data; Inputting the airflow field characteristic data and turbulence pattern characteristic data into a dynamic modeling algorithm, establishing an atmospheric parameter time-varying model, and obtaining atmospheric dynamic characteristic data; The atmospheric dynamic characteristic data is used as a part of the multi-source navigation raw data.

3. The method according to claim 1, characterized in that The deep learning feature extraction is performed on the multi-source navigation raw data to obtain multimodal feature data, including: Performing standardization processing on the multi-source navigation raw data to obtain a standardized data sequence; Using a convolutional neural network to perform multi-scale feature extraction on the standardized data sequence to obtain initial feature data, and performing time series modeling based on the initial feature data to obtain a time series feature sequence; Anomaly pattern recognition is performed on the time series feature sequence to obtain anomaly detection results, and the time series feature sequence and the anomaly detection results are fused to obtain multimodal feature data.

4. The method according to claim 1, characterized in that The step of inputting the multi-modal feature data into a federated filtering algorithm, performing combined navigation calculation on the inertial navigation data and the satellite navigation information, and obtaining an initial navigation result includes: Separating the multimodal feature data according to data sources to obtain subsystem observation data, and constructing filter models for the subsystem observation data to obtain subsystem state equations; Performing state estimation based on the subsystem state equation to obtain a local optimal estimation value, and calculating the confidence of the local optimal estimation value to obtain an information fusion weight; A global state estimation is performed according to the information fusion weight to obtain an initial navigation result.

5. The method according to claim 1, characterized in that The estimating and modeling the error amount in the atmospheric information data based on the initial navigation result to obtain the atmospheric information error model comprises: Taking the initial navigation result as a reference value, extracting the atmospheric parameters of the corresponding time and space points to obtain reference data; Calculating the deviation between the atmospheric information data and the reference data to obtain an error sample sequence, and performing statistical characteristic analysis on the error sample sequence to obtain an error distribution characteristic; An error compensation function is established based on the error distribution characteristics to obtain an atmospheric information error model.

6. The method according to claim 1, characterized in that The method of fusing the compensated atmospheric information with the inertial navigation data to obtain a high-precision navigation result includes: Perform parameter conversion on the compensated atmospheric information, extract position, velocity and attitude related components, and obtain atmospheric navigation parameters; Combining the atmospheric navigation parameters and inertial navigation data, establishing an error state equation and an observation equation to obtain a fusion navigation model, and designing a federal Kalman filter based on the fusion navigation model to obtain an optimal state estimator; The inertial navigation data is corrected using the optimal state estimator to obtain a high-precision navigation result.

7. The method according to claim 6, characterized in that The method of designing a federated Kalman filter according to the fusion navigation model to obtain an optimal state estimator includes: Analyzing the error state equation and the observation equation in the fusion navigation model, and constructing a federated filtering system structure based on the error state equation and the observation equation to obtain a main filter and a plurality of sub-filters; According to the processing requirements of the sub-filter, a state transfer matrix and an observation matrix are designed to obtain a system dynamic description and an observation relationship; According to the error characteristics of the atmospheric navigation parameters and the inertial navigation data, a process noise covariance matrix and a measurement noise covariance matrix are set to obtain a noise parameter model; According to the system dynamic description and the noise parameter model, an information allocation strategy is designed to obtain a weight allocation scheme; According to the weight allocation scheme, a global consistency constraint is determined, and according to the federated filtering system structure, the system dynamic description, the noise parameter model, the weight allocation scheme and the global consistency constraint, the optimal state estimator is constructed.

8. The method according to claim 5, characterized in that The step of establishing an error compensation function based on the error distribution characteristics comprises: Performing spectrum analysis on the error distribution characteristics to separate systematic error components and random error components to obtain an error decomposition result; Establishing a parameterized model for the system error component to obtain a deterministic error compensation function; Establishing a probability statistical model for the random error component to obtain a random error correction function; The deterministic error compensation function and the random error correction function are integrated to obtain a complete error compensation function.

9. A satellite-guided atmospheric fusion positioning system based on an inertial navigation system, characterized in that: The system comprises: The data acquisition module is used to acquire inertial navigation data, satellite navigation data and atmospheric information data to obtain multi-source navigation raw data; A feature extraction module is used to perform deep learning feature extraction on the multi-source navigation raw data to obtain multimodal feature data; The quality assessment module is used to perform quality assessment on the satellite navigation data to obtain a satellite navigation signal availability result; the quality assessment on the satellite navigation data to obtain the satellite navigation signal availability result includes: obtaining signal-to-noise ratio and carrier-to-noise ratio data of the satellite navigation data, and calculating a signal strength index based on the signal-to-noise ratio and the carrier-to-noise ratio data; analyzing satellite signal quality based on the signal strength index to obtain a signal reliability assessment value; counting the number and distribution of currently visible satellites to obtain satellite geometric structure parameters; performing a comprehensive analysis based on the signal reliability assessment value and the satellite geometric structure parameters to obtain a satellite navigation signal availability result; The atmospheric error model building module is used for: when the satellite navigation signal availability result indicates that the satellite navigation information is valid, inputting the multimodal feature data into a federated filtering algorithm, performing combined navigation solution on the inertial navigation data and the satellite navigation information, and obtaining an initial navigation result; estimating and modeling the error amount in the atmospheric information based on the initial navigation result, and obtaining an atmospheric information error model; The fusion positioning module is used for compensating the real-time atmospheric information by using the atmospheric information error model to obtain compensated atmospheric information when the satellite navigation signal availability result indicates that the satellite navigation information is invalid; fusing the compensated atmospheric information with the inertial navigation data to obtain a high-precision navigation result.

Images