High-availability terminal differential positioning method and device

Abstract

The invention relates to the technical field of positioning, and particularly provides a differentiated positioning method and device for a high-availability terminal, and the method comprises the following steps: S1, data preprocessing; s2, layered multi-modal feature calculation is carried out; s3, performing cross-scale information flow feature fusion; s4, training a high-availability positioning model; and S5, real-time positioning and feedback are carried out. Compared with the prior art, the method has the advantages that a high-precision and high-robustness positioning function can be realized, and the generalization performance and the practicability of a positioning system are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of positioning technology, and specifically provides a high-availability terminal differential positioning method and device. Background Art

[0002] In current positioning technologies, affected by numerous factors such as complex indoor and outdoor environments, electromagnetic interference, and re-deployment of positioning base stations, and combined with device heterogeneity, positioning signals exhibit dynamic time-variation. These factors often cause the signal distribution collected during the positioning stage to deviate from the signal distribution constructed during the offline stage, thereby seriously affecting the positioning accuracy. Traditional positioning methods often struggle to provide stable and high-precision positioning results when faced with these challenges.

[0003] To address this issue, the industry has proposed various positioning methods based on machine learning. However, most of these methods focus on single positioning feature extraction or single positioning model training, and fail to fully utilize multi-scale local spatio-temporal features and cross-scale information flow features, resulting in limitations in the generalization performance and practicality of the positioning system. Summary of the Invention

[0004] The present invention aims at the deficiencies of the above-mentioned prior art and provides a highly practical high-availability terminal differential positioning method.

[0005] A further technical task of the present invention is to provide a highly available terminal differential positioning device with reasonable design, safety, and applicability.

[0006] The technical solution adopted by the present invention to solve its technical problems is as follows:

[0007] A high-availability terminal differential positioning method includes the following steps:

[0008] S1. Data preprocessing;

[0009] S2. Hierarchical multi-modal feature calculation;

[0010] S3. Cross-scale information flow feature fusion;

[0011] S4. High-availability positioning model training;

[0012] S5. Real-time positioning and feedback.

[0013] Further, in step S1, it includes:

[0014] S1-1. Data collection;

[0015] S1-2. Data preprocessing;

[0016] S1-3. Feature extraction.

[0017] Further, in step S1-1, in the offline phase, a large amount of positioning signal data is collected under various environments and when using different terminals. In the online phase, positioning signal data from the terminal is collected in real time;

[0018] In step S1-2, the data is cleaned to remove outliers and noise, and normalized;

[0019] In step S1-3, according to the characteristics of the positioning signal, multi-scale and multi-modal features are extracted. The features include time-domain features, frequency-domain features, and spatial-domain features. The time-domain features are to extract the change trend of signal strength over time and the stability of signal arrival time:

[0020] The frequency-domain features are to extract the spectral characteristics of the signal through the Fourier transform method;

[0021] The spatial-domain features are to analyze the intensity distribution of the signal in different directions and the distance relationship between the signal and the base station.

[0022] Further, in step S2, it includes:

[0023] S2-1. Using a convolutional neural network deep learning model to perform hierarchical processing on the extracted features to capture multi-scale local spatio-temporal features;

[0024] S2-2. Introducing a feature selection algorithm to screen the extracted features and remove redundant and irrelevant features.

[0025] Further, in step S3, it includes:

[0026] S3-1. Using the methods of splicing and weighted summation to fuse the features under different sampling scales to achieve information complementarity; using multi-scale convolutional kernels or pyramid structures to capture multi-scale features;

[0027] S3-2. Using self-attention mechanism or cross-attention mechanism, calculating the importance score of each feature, and then performing weighted summation on the features to obtain the fused features;

[0028] S3-3. Introducing a feature fusion strategy to fuse different modal features through splicing and weighted fusion methods to make full use of the complementarity between different modalities.

[0029] Further, in step S4, it includes:

[0030] S4-1. For large-scale datasets, use deep learning models; for small-scale datasets, use traditional machine learning models to build highly available positioning models;

[0031] S4-2. Adopt a cross-validation strategy, divide the data collected in the offline phase into a training set and a validation set, train the model and verify its performance.

[0032] S4-3. Use the data in the online positioning phase to evaluate the performance of the trained model, propose a model update strategy, and use an online learning algorithm or an incremental learning algorithm to continuously update and optimize the model according to the positioning effect in actual applications to adapt to the positioning requirements in different environments.

[0033] Further, in step S5, it includes:

[0034] S5-1. Integrate the preprocessing, feature extraction, and trained machine learning model to build a positioning system, deploy the system on a positioning server or a terminal device, and process positioning requests in real time;

[0035] S5-2. Preprocess and extract features from the data to obtain positioning features, input them into the trained machine learning model, output the positioning result, and perform real-time navigation and obstacle avoidance operations on the vehicle or terminal according to the positioning result;

[0036] S5-3. Provide feedback on the system according to the positioning effect in actual applications, optimize the system according to the feedback information, and regularly evaluate and maintain the performance of the system.

[0037] A highly available terminal differential positioning device includes: at least one memory and at least one processor;

[0038] The at least one memory is used to store machine-readable programs;

[0039] The at least one processor is used to call the machine-readable program to execute a highly available terminal differential positioning method.

[0040] Compared with the prior art, a highly available terminal differential positioning method and device of the present invention have the following outstanding beneficial effects:

[0041] (1) Through hierarchical multi-modal feature calculation and cross-scale information flow feature fusion, and by selecting a suitable machine learning model, the present invention can make full use of multi-scale local spatio-temporal features and cross-scale information flow features to more accurately capture useful information in the positioning signal and achieve high-precision positioning.

[0042] (2) Facing the problem of dynamic time-variation of positioning signals caused by factors such as complex indoor and outdoor environments, electromagnetic interference, re-deployment of positioning base stations, and device heterogeneity, the method of the present invention enhances the robustness of the model by introducing strategies such as feature selection, regularization, dropout, and technologies such as cross-scale information flow feature fusion and attention mechanism. This enables the system to maintain stable positioning performance under different environments, times, and terminals, reducing positioning errors caused by environmental changes or terminal differences.

[0043] (3) By introducing machine learning technology, the present invention can be adaptively adjusted according to the positioning requirements and data characteristics in actual applications. By continuously updating and optimizing the model, the system can adapt to different positioning scenarios and conditions, improving the flexibility and practicality of positioning.

[0044] (4) Compared with traditional positioning methods, the positioning method based on machine learning does not require expensive hardware devices or complex layout designs, reducing the system cost. At the same time, by optimizing the algorithm and model, the positioning efficiency can be improved, further reducing the cost.

[0045] (5) The present invention can be easily extended to other positioning technologies and application scenarios. For example, this method can be applied to satellite positioning systems such as GPS and Beidou, or combined with other sensor data (such as inertial sensors, cameras, etc.) for multi-source fusion positioning to further improve positioning accuracy and robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0047] Attached Figure 1 is a flow schematic diagram of a highly available terminal differential positioning method. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] In order to enable those skilled in the art to better understand the solution of the present invention, the following will further describe the present invention in detail in combination with specific embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0049] The following gives a best embodiment:

[0050] Such as Figure 1As shown in the figure, a high-availability terminal differential positioning method in this embodiment has the following steps:

[0051] S1. Data preprocessing;

[0052] Including:

[0053] S1-1. In the offline stage, collect a large amount of positioning signal data under various environments (such as indoor, outdoor, urban, rural, etc.), different time periods (daytime, night, seasonal changes, etc.), and using different terminals (mobile phones, tablets, dedicated positioning devices, etc.). In the online stage, collect positioning signal data from the terminal in real time.

[0054] S1-2. Clean the data, remove outliers and noise, and perform normalization processing to ensure data quality.

[0055] S1-3. According to the characteristics of the positioning signal, extract multi-scale and multi-modal features.

[0056] These features include time-domain features: extract the change trend of signal strength over time, the stability of signal arrival time, etc.

[0057] Frequency-domain features: Through methods such as Fourier transform, extract the spectral characteristics of the signal, such as main frequency, bandwidth, etc.

[0058] Spatial-domain features: Analyze the intensity distribution of the signal in different directions, and the distance relationship between the signal and the base station, etc.

[0059] S2. Hierarchical multi-modal feature calculation;

[0060] Including:

[0061] S2-1. Use deep learning models such as convolutional neural networks (CNNs) to perform hierarchical processing on the extracted features to capture multi-scale local spatio-temporal features. Specifically, first extract low-level features such as edges and textures through a shallow network, and then gradually deepen to extract higher-level features such as signal patterns and structures using a deep network;

[0062] S2-2. In order to further improve the accuracy of feature extraction, the present invention also introduces a feature selection algorithm to screen the extracted features and remove redundant and irrelevant features.

[0063] S3. Cross-scale information flow feature fusion;

[0064] Including:

[0065] S3-1. Use methods such as splicing and weighted summation to fuse features at different sampling scales to achieve information complementarity. Use multi-scale convolutional kernels or pyramid structures to better capture multi-scale features.

[0066] S3-2. Introduce an attention mechanism to perform weighted processing on the fused features to highlight important features, suppress irrelevant features, and capture feature expressions with higher detail. Specifically, use a self-attention mechanism or a cross-attention mechanism. By calculating the importance scores of each feature and then performing weighted summation on the features, the fused features are obtained.

[0067] S3-3. To further enhance the robustness of the model, the present invention also introduces a feature fusion strategy to fuse features of different modalities through methods such as concatenation and weighted fusion to make full use of the complementarity between different modalities.

[0068] S4. Training of a highly available positioning model;

[0069] including:

[0070] S4-1. Based on the above feature extraction and fusion methods, select a suitable machine learning model. For large-scale datasets, consider using a deep learning model; for small-scale datasets, a traditional machine learning model can be used to build a highly available positioning model.

[0071] S4-2. Adopt strategies such as cross-validation to divide the data collected in the offline stage into a training set and a validation set, train the model and verify its performance. By adjusting model parameters, optimizer settings, etc., improve the generalization performance and positioning accuracy of the model. By introducing strategies such as regularization and dropout, prevent the model from overfitting;

[0072] S4-3. Use the data in the online positioning stage to evaluate the performance of the trained model to ensure that it meets the positioning requirements of high precision and high robustness. At the same time, the present invention also proposes a model update strategy. Use an online learning algorithm or an incremental learning algorithm to continuously update and optimize the model according to the positioning effect in actual applications to adapt to the positioning requirements in different environments.

[0073] S5. Real-time positioning and feedback;

[0074] including:

[0075] S5-1. Integrate modules such as preprocessing, feature extraction, and the trained machine learning model to build a complete highly robust precise positioning system. Deploy the system on a positioning server or a terminal device to facilitate real-time processing of positioning requests.

[0076] S5-2. In the online positioning stage, collect positioning signal data from the terminal. Preprocess and extract features from the data to obtain positioning features, input them into the trained machine learning model, and output the positioning result. According to the positioning result, perform real-time navigation and obstacle avoidance operations on the vehicle or terminal;

[0077] S5-3. Collect real-time positioning data, including positioning results, positioning times, environmental information, etc. Provide feedback on the system according to the positioning effect in actual applications. Optimize the system according to the feedback information. Regularly perform performance evaluation and maintenance on the system to ensure its long-term stable operation.

[0078] Based on the above method, a highly available terminal differential positioning device in this embodiment includes: at least one memory and at least one processor;

[0079] The at least one memory is used to store machine-readable programs;

[0080] The at least one processor is used to call the machine-readable program and execute a highly available terminal differential positioning method.

[0081] The above specific embodiments are only specific cases of the present invention. The patent protection scope of the present invention includes but is not limited to the above specific embodiments. Any technical solution that conforms to the technical solutions described in the above specific embodiments of the present invention and any appropriate changes or substitutions made by those of ordinary skill in the art shall fall within the patent protection scope of the present invention.

[0082] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A high-availability terminal differentiated positioning method, characterized in that: The steps are as follows: S1, data preprocessing; S2, hierarchical multimodal feature calculation; S3, cross-scale information flow feature fusion; S4, high availability positioning model training; S5. Real-time positioning and feedback.

2. A high-availability terminal differentiated positioning method according to claim 1, characterized in that: In step S1, it includes: S1-1. Data collection; S1-2, data preprocessing; S1-3. Feature extraction.

3. A high-availability terminal differentiated positioning method according to claim 2, characterized in that: In step S1-1, in an offline phase, a large amount of positioning signal data is collected in a variety of environments and using different terminals, and in an online phase, positioning signal data from the terminal is collected in real time; In step S1-2, the data is cleaned, outliers and noise are removed, and normalization is performed; In step S1-3, according to the characteristics of the positioning signal, multi-scale and multi-modal features are extracted, and the features include time domain features, frequency domain features and space domain features. The time domain features are used to extract the trend of signal strength changes over time and the stability of signal arrival time: The frequency domain feature is to extract the spectrum characteristics of the signal through the Fourier transform method; The spatial domain features are to analyze the intensity distribution of the signal in different directions and the distance relationship between the signal and the base station.

4. A high-availability terminal differentiated positioning method according to claim 3, characterized in that: In step S2, it includes: S2-1. Using the convolutional neural network deep learning model, the extracted features are processed hierarchically to capture multi-scale local spatiotemporal features. S2-2. Introduce feature selection algorithm to screen the extracted features and remove redundant and irrelevant features.

5. A high-availability terminal differentiated positioning method according to claim 4, characterized in that: In step S3, it includes: S3-1. Use the concatenation and weighted summation methods to fuse features at different sampling scales to achieve information complementarity; use multi-scale convolution kernels or pyramid structures to capture multi-scale features; S3-2, use the self-attention mechanism or cross-attention mechanism to calculate the importance score of each feature, and then perform weighted summation on the features to obtain the fused features; S3-3. Introduce feature fusion strategy to fuse the features of different modalities through splicing and weighted fusion methods, and make full use of the complementarity between different modalities.

6. A high-availability terminal differentiated positioning method according to claim 5, characterized in that: In step S4, it includes: S4-1. For large-scale data sets, use deep learning models; for small-scale data sets, use traditional machine learning models to build a highly available positioning model; S4-2. Using a cross-validation strategy, the data collected in the offline phase is divided into a training set and a validation set to train the model and verify its performance. S4-3. Use the data from the online positioning phase to evaluate the performance of the trained model, propose a model update strategy, and use online learning algorithms or incremental learning algorithms to continuously update and optimize the model based on the positioning effect in actual applications to adapt to positioning requirements in different environments.

7. A high-availability terminal differentiated positioning method according to claim 6, characterized in that: In step S5, it includes: S5-1. Integrate preprocessing, feature extraction and trained machine learning models to build a positioning system, deploy the system on a positioning server or terminal device, and process positioning requests in real time; S5-2, preprocessing and feature extraction of data to obtain positioning features, input into the trained machine learning model, output positioning results, and perform real-time navigation and obstacle avoidance operations on the vehicle or terminal based on the positioning results; S5-3. Provide feedback to the system based on the positioning effect in actual applications, optimize the system based on the feedback information, and regularly evaluate and maintain the system performance.

8. A high-availability terminal differentiated positioning device, characterized in that: include: at least one memory and at least one processor; The at least one memory is used to store a machine-readable program; The at least one processor is configured to call the machine-readable program to execute the method according to any one of claims 1 to 7.

Images