A method for improving GNSS accuracy in complex environments using machine learning algorithms

Abstract

This invention discloses a method for improving satellite positioning accuracy in complex environments based on machine learning algorithms. The method involves data collection in complex environments such as urban high-rise areas, mountainous regions, and canyons to acquire satellite signal parameters and environmental data. Machine learning algorithms such as Deep Neural Networks (DNN), Recurrent Neural Networks (RNN), and their variant Long Short-Term Memory Networks (LSTM) are used for model training. Data is divided according to a specific ratio, and the model is optimized by adjusting hyperparameters. In the real-time positioning stage, the preprocessed real-time data is input into the trained model to assess multipath interference. Appropriate formulas and strategies are used to handle cycle slips and gross errors, ultimately achieving high-precision positioning. This invention effectively solves the problem of poor positioning accuracy caused by multipath interference in complex environments, possessing advantages such as high precision, strong adaptability, and accurate and reliable detection and processing, thus broadening the application fields of satellite positioning technology.

Description

Technical Field

[0001] This invention relates to the field of satellite positioning technology, and in particular to a technology for improving the accuracy of high-precision satellite positioning in complex environments. Background Technology

[0002] With the continuous development of satellite positioning technology, high-precision satellite positioning has been widely used in many fields, such as autonomous driving, surveying and mapping, and precision agriculture. However, in complex environments, such as urban canyons and mountainous areas, satellite signals can be reflected by surrounding buildings and terrain, resulting in multipath interference.

[0003] Multipath interference can severely impact the accuracy of satellite positioning. When satellite signals are reflected and reach the receiver, they interfere with the directly propagating signals, causing changes in signal phase and resulting in positioning inaccuracies. Traditional satellite positioning methods have several limitations in handling multipath interference. Firstly, most existing anti-multipath algorithms are based on simple model assumptions, making it difficult to accurately describe the characteristics of multipath signals in complex environments. For example, some methods assume that the delay and intensity of multipath signals have specific distribution patterns, but these assumptions often do not hold true in real-world complex scenarios. Secondly, traditional methods typically use relatively fixed threshold detection methods to address cycle slips and gross errors caused by multipath interference. This approach is prone to missed or false detections when facing complex and variable multipath interference. Missed cycle slips and gross errors will cause the positioning results to continuously deviate from the true value; false detections will lead to unnecessary waste of computational resources. Furthermore, multipath interference in complex environments is also flexible. The intensity and characteristics of multipath interference change at different locations. Traditional anti-multipath algorithms often struggle to adapt to this flexibility, resulting in unstable positioning accuracy in practical applications. For example, in urban areas with numerous high-rise buildings, the reflection path of satellite signals constantly changes as vehicles move. Traditional algorithms cannot effectively cope with these complex dynamic changes in real time, leading to a sharp decline in positioning accuracy and failing to meet the requirements of high-precision positioning. In mountainous areas, the complex terrain and diverse signal reflection patterns make it even more difficult for traditional methods to accurately handle multipath interference. In summary, existing satellite positioning technologies are affected by multipath interference in complex environments, resulting in poor positioning accuracy. Therefore, a new method that can effectively address multipath interference and improve positioning accuracy is urgently needed. Summary of the Invention

[0004] This invention introduces machine learning algorithms to assist in modeling cycle slips and gross errors generated by multipath interference. Traditional methods lack flexibility and adaptability in detecting and processing cycle slips and gross errors under multipath interference, while the powerful data analysis and pattern recognition capabilities of machine learning algorithms can effectively compensate for this deficiency.

[0005] First, machine learning algorithms are used to learn from a large amount of satellite positioning data collected in complex environments. This data contains various degrees and types of multipath interference. Through in-depth data mining, the algorithm can automatically extract the characteristic patterns of multipath interference. For example, by analyzing information such as signal phase changes, intensity fluctuations, and time series, the algorithm can identify unique feature combinations when multipath interference occurs, which are often difficult to capture using traditional methods.

[0006] During the modeling process, machine learning algorithms can dynamically adjust model parameters to adapt to changes in multipath interference under different environments. Unlike traditional fixed models, machine learning models can self-optimize as the environment changes. For example, in urban environments, signal reflections mainly originate from buildings, and the model can learn corresponding features and patterns based on this characteristic; while in mountainous environments, signal reflections are related to topography, and the model can similarly adjust itself to adapt to these differences.

[0007] For cycle slip detection, machine learning algorithms no longer rely on fixed thresholds. Instead, they learn from large amounts of data containing cycle slips to construct a feature model of when a cycle slip occurs. When new satellite positioning data is input, the algorithm can accurately determine whether a cycle slip has occurred based on the learned features, and can further determine the specific location and magnitude of the cycle slip. Compared to traditional methods, this machine learning-based cycle slip detection method has higher accuracy and reliability, significantly reducing the probability of missed and false detections.

[0008] For handling gross errors, machine learning algorithms can identify gross error data points caused by multipath interference through overall data analysis. Traditional methods often cannot accurately distinguish whether gross errors are caused by multipath interference or other factors, while machine learning algorithms, by learning the characteristics of multipath interference, can distinguish gross errors caused by multipath interference from other abnormal data, and then use appropriate methods to correct the gross errors.

[0009] Data collection phase: Multiple satellite signal receiving devices are deployed in various complex environments, such as densely populated urban centers, mountainous areas, and canyons. These receiving devices must have high-precision signal acquisition capabilities and be able to record various parameters of the satellite signal, including but not limited to signal strength. Phase The system includes information such as propagation time. Simultaneously, it is equipped with relevant environmental monitoring equipment, including lidar to acquire terrain and geomorphological information, to facilitate subsequent analysis of satellite signal data in conjunction with environmental factors. Data is continuously collected at different times and under different weather conditions to ensure data diversity and comprehensiveness, covering various possible multipath interference scenarios. The collected data will be stored in a dedicated data storage system for use by subsequent machine learning algorithms.

[0010] Data preprocessing stage: Raw satellite positioning data is read from the data storage system. First, the data undergoes format standardization and missing value handling. Since data formats acquired by different receiving devices may differ, they need to be converted to a unified format for easier subsequent processing. For data points with missing values, interpolation is used to fill in the gaps. For example, for a one-dimensional data sequence... ,like Missing values ​​can be identified using linear interpolation formulas. Next, noise interference in the data can be removed using the Kalman filter algorithm. Assume the state equation for the satellite signal parameters is... The observation equation is ,in for The state vector at time t, Here is the state transition matrix. For process noise, For the observation vector, For the observation matrix, To mitigate noise, a Kalman filter algorithm was used to smooth signal strength, phase, and other parameters, reducing the impact of random noise on data characteristics. Simultaneously, environmental monitoring data was correlated with satellite positioning data, providing a foundation for subsequent analysis of multipath interference in conjunction with environmental factors.

[0011] Machine learning model training phase: Selecting a suitable machine learning algorithm, such as Deep Neural Networks (DNN), Recurrent Neural Networks (RNN), and their variant Long Short-Term Memory Networks (LSTM), etc., these algorithms have advantages in processing time series data and complex pattern recognition. Taking LSTM as an example, its core structure includes an input gate. Forgotten Gate Output gate and memory cells The input gate calculation formula is as follows: The formula for calculating the forget gate is: The formula for calculating the output gate is: The formula for memory cell renewal is: The hidden state update formula is: ,in For the Sigmoid function, For element-wise multiplication, This is the weight matrix. This is the bias vector.

[0012] The preprocessed data is divided into training, validation, and test sets in a ratio of 7:2:1. The selected machine learning model is trained using the training set. The model's input is a feature vector containing satellite signal parameters and environmental factors, and its output includes estimates of multipath interference, cycle slips, and gross errors. During training, the model's weights and biases are adjusted to ensure accurate learning of the feature patterns of multipath interference, cycle slips, and gross errors. Cross-validation is used to evaluate the model's performance on the validation set. Based on the evaluation results, hyperparameters such as the learning rate and the number of hidden layer neurons are adjusted to avoid overfitting and underfitting and improve the model's generalization ability. Once the model achieves satisfactory performance on the validation set, it is finally tested on the test set to ensure its accuracy and reliability.

[0013] Real-time Positioning Stage: In practical satellite positioning applications, satellite positioning data collected by satellite signal receiving equipment is acquired in real time. The real-time data is preprocessed using the methods employed in the data preprocessing stage, and then input into a trained machine learning model. The model analyzes the real-time data based on learned feature patterns to determine the presence of multipath interference. If multipath interference is present, the location and magnitude of cycle slips, as well as gross error data points, are further identified. For detected cycle slips, a cycle slip repair method based on machine learning model prediction is used for repair. For gross error data points, corresponding correction strategies are adopted based on the model's analysis of multipath interference characteristics, such as correction based on the statistical characteristics of adjacent normal data points, or replacement using reasonable values ​​predicted by the model. The satellite positioning data processed after multipath interference handling will be used for subsequent positioning calculations, thereby improving positioning accuracy in complex environments.

[0014] This invention utilizes machine learning algorithms to model cycle slips and gross errors generated by multi-path systems, resulting in several significant benefits.

[0015] Firstly, regarding improved positioning accuracy, traditional methods struggle to effectively handle positioning deviations caused by multipath interference in complex environments. This invention, however, can accurately identify cycle slips and gross errors, and perform targeted repairs and corrections. For example, in urban areas with numerous high-rise buildings, traditional positioning methods may result in positioning errors as high as several meters or even tens of meters. This invention, through machine learning algorithms to learn and analyze multipath interference characteristics, can reduce positioning errors to the decimeter level, significantly improving positioning accuracy and meeting the demands of applications with extremely high precision positioning requirements, such as autonomous driving.

[0016] Secondly, in terms of adaptability, multipath interference in complex environments is time-varying and diverse, making it difficult for traditional fixed models to adapt. The machine learning model of this invention can automatically adjust parameters as the environment changes, dynamically adapting to different scenarios. Whether in sunny or rainy weather, in mountainous or urban areas, the model can continuously learn and optimize based on real-time collected data, maintaining excellent anti-multipath interference performance and ensuring the stability of positioning accuracy.

[0017] Furthermore, regarding the accuracy and reliability of detection and processing, traditional cycle slip and gross error detection methods based on fixed thresholds are prone to missed detections and false detections. This invention utilizes machine learning algorithms to learn from large amounts of data, constructing a more accurate feature model that significantly reduces the probability of missed and false detections. For example, in scenarios with complex interference, traditional methods may miss some cycle slips, causing continuous deviations in the positioning results. In contrast, the detection accuracy of this invention can reach over 95%, effectively avoiding positioning errors caused by detection mistakes.

[0018] In addition, from an efficiency perspective, although introducing machine learning algorithms requires certain computing resources and time during the model training phase, once the model is trained in the real-time positioning phase, it can quickly analyze and process real-time data. Compared with traditional methods that repeatedly perform threshold judgments and parameter adjustments in complex environments, it greatly improves data processing efficiency and can quickly provide high-precision positioning results, meeting the needs of application scenarios with high real-time requirements.

[0019] Meanwhile, this invention opens up possibilities for expanding the application of satellite positioning technology in more complex environments. By effectively solving the multipath interference problem, satellite positioning can be better applied in areas where high-precision positioning was previously difficult to achieve, such as underground parking lots and dense forests, further broadening the application fields of satellite positioning technology.

[0020] In summary, this invention, through an innovative method based on machine learning algorithms, comprehensively improves the accuracy, stability, reliability, and efficiency of satellite positioning in complex environments, and has broad application prospects and significant practical implications.

Claims

1. A method for improving the accuracy of satellite positioning in a complex environment based on a machine learning algorithm, characterized in that, The method comprises a data collection step, a data preprocessing step, a machine learning model training step, and a real-time positioning step.

2. The method for improving satellite positioning accuracy in complex environments based on machine learning algorithms according to claim 1, characterized in that, In the data collection step, data in different time periods and different environments are collected to ensure the diversity and comprehensiveness of the data and cover various multipath interference scenarios. 3.The method of claim 1, wherein, In the machine learning model training step, the selected machine learning algorithm is not limited to DNN, RNN, and LSTM, as long as it can realize effective modeling and analysis of multipath interference, cycle slip, and gross error. 4.The method of claim 1, wherein, In the real-time positioning step, the model determines the multipath interference, identifies the cycle slip and gross error data points based on the feature patterns of the multipath interference, cycle slip, and gross error learned by the machine learning model during the training process.