Diffraction error modeling method and device based on deep learning
By using a deep learning-based diffraction error modeling method and training a satellite signal diffraction error model using an LSTM network, the problem of decreased positioning accuracy of GNSS technology in heavily obstructed environments is solved, achieving more efficient GNSS signal processing and more reliable positioning results.
Patent Information
- Application Number
- CN202511916157.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-18
- Publication Date
- 2026-02-06
AI Technical Summary
Existing GNSS technology struggles to effectively eliminate diffraction errors in heavily obstructed environments, leading to decreased positioning accuracy and reliability, and failing to meet the demands for high-precision measurements.
A deep learning-based approach is adopted to establish a diffraction error model by acquiring raw GNSS observation data and position coordinates. An LSTM network is used to train and predict the diffraction error of the satellite, and the satellite signal is weighted to improve positioning accuracy.
It improves the accuracy and efficiency of GNSS signal processing in highly obstructed environments, providing more reliable high-precision positioning results, especially for tasks such as landslide detection.
Smart Images
Figure CN121479237A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of surveying and mapping, and in particular relates to a method and apparatus for diffraction error modeling based on deep learning. Background Technology
[0002] Due to its advantages such as high automation, all-weather operation, and high measurement accuracy, Global Navigation Satellite System (GNSS) positioning technology has become an important means for surveying and mapping, safety monitoring of civil and water conservancy infrastructure (such as bridges, dams, and high-rise buildings), and geological disaster monitoring (such as slopes, landslides, and ground subsidence). In environments with a wide field of view, when data is collected throughout the day and applied using short baseline static data processing methods, the measurement accuracy of GNSS can even reach 1 mm in the horizontal direction and better than 2 mm in the vertical direction.
[0003] With rapid urbanization and the growth of dense forests in natural valleys, numerous deformation monitoring stations for landslides, slopes, dams, and bridges are located in highly obstructed environments with dense buildings, trees, and hillsides. This high obstruction reduces the number of visible GNSS satellites and leads to multipath effects and non-line-of-sight (NLOS, including diffraction errors). These errors are difficult to parameterize and eliminate using differential or universal modeling methods, severely reducing positioning accuracy and reliability. In this context, frequent outliers and significant noise make GNSS technology unsuitable for the needs of surveying, civilian, and water infrastructure safety monitoring, becoming a major problem limiting the application of GNSS technology in high-precision positioning.
[0004] Current research on diffraction errors is insufficient. Diffraction error elimination can be broadly categorized into three types: signal-to-noise ratio (SNR) weighted methods, auxiliary sensor methods, geographic elevation masking methods, and machine learning methods. SNR weighted methods utilize the low SNR characteristic of GNSS diffraction or reflection signals, using a phase observation weighting model to reduce the observation weights that include diffraction errors. Auxiliary sensor methods primarily utilize other sensors to detect diffraction signals, then reduce or eliminate the corresponding observations, such as fisheye cameras, laser scanners, 3D city models, or array antennas. Geographic elevation masking methods eliminate diffraction errors by analyzing posterior residuals or SNR time series, which has proven effective for static high-precision positioning.
[0005] However, how to provide a new method to further improve the accuracy and efficiency of GNSS signal processing is a problem that urgently needs to be solved. Summary of the Invention
[0006] In view of this, this application provides a deep learning-based diffraction error modeling method and apparatus, which aims to improve the accuracy and efficiency of GNSS signal processing.
[0007] Firstly, this application provides a deep learning-based method for modeling diffraction errors, including: Acquire raw GNSS observation data of satellites observed by monitoring stations and reference stations under historical conditions, as well as the position coordinates of the observation stations and reference stations; The diffraction error of the satellite is calculated based on the raw GNSS observation data, the position coordinates of the observation station and the reference station; The first dataset is established based on the satellite's diffraction error, the satellite's position relative to the monitoring station, and the signal strength of the satellite obtained by the monitoring station. Based on the first dataset, the first diffraction error prediction model for the monitoring station was trained.
[0008] Optionally, the raw GNSS observation data includes phase observations, pseudorange observations, and signal-to-noise ratio observations.
[0009] Optionally, the step of calculating the satellite's diffraction error based on the raw GNSS observation data and the position coordinates of the observation station and the reference station includes: Based on the raw GNSS observation data, a continuous time period with unchanged ambiguity is determined, and the ambiguity value within the continuous time period is calculated and fixed. Calculate the double-difference carrier phase residual based on the ambiguity value, the position coordinates of the observation station and the reference station; The site where the diffraction effect occurs is identified based on the error, and the diffraction error of the site where the diffraction effect occurs is determined from the double-difference carrier phase residual.
[0010] Optionally, the step of determining a continuous time period with unchanged ambiguity based on the raw GNSS observation data, and calculating and fixing the ambiguity value within the continuous time period includes: Based on the raw GNSS observation data, the HMW combination and GF combination were calculated respectively. The ambiguity is detected by combining HMW and GF, and then the continuous time period in which the ambiguity remains unchanged is determined. Based on the double difference variance formula, the floating-point ambiguity and variance are calculated using the least squares method. Then, the integer ambiguity is calculated using the Lambda method based on the floating-point ambiguity and variance, and the ambiguity value within the continuous time period is fixed.
[0011] Optionally, the steps of identifying the sites where the diffraction effect occurs based on the error, and determining the diffraction error of the sites where the diffraction effect occurs from the double-difference carrier phase residual, include: If the phase residual of the double-difference carrier is greater than zero and monotonically increasing, then the monitoring station has experienced a diffraction effect. If the phase residual of the double-difference carrier is less than zero and decreases monotonically, then the reference station has experienced a diffraction effect.
[0012] Optionally, it also includes: A second dataset is established based on the satellite's diffraction error, the satellite's position relative to the reference station, and the satellite's signal strength obtained by the reference station. Based on the second dataset, the satellite's position information relative to the monitoring station and the satellite's diffraction error are fitted to obtain the second diffraction error prediction model for the reference station.
[0013] Optionally, it also includes: Based on the first and second diffraction error models, satellite signals exhibiting diffraction effects are determined in real time, and weighted for the satellite signals received by the reference station and the monitoring station, respectively.
[0014] Secondly, this application provides a deep learning-based diffraction error modeling device, comprising: The acquisition module is used to acquire the raw GNSS observation data of the satellite observed by the monitoring station and the reference station under historical conditions, as well as the position coordinates of the observation station and the reference station; The calculation module is used to calculate the diffraction error of the satellite based on the raw GNSS observation data and the position coordinates of the observation station and the reference station. A module is established to create the first dataset based on the satellite's diffraction error, the satellite's position relative to the monitoring station, and the signal strength of the satellite obtained by the monitoring station. The training module is used to train the first diffraction error prediction model of the monitoring station based on the first dataset.
[0015] Thirdly, this application provides an electronic device, including the deep learning-based diffraction error modeling device described above.
[0016] Fourthly, this application provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores at least one piece of program code, which is executed by a processor to implement the deep learning-based diffraction error modeling method as described in any of the preceding claims.
[0017] The beneficial effects of the technical solution provided in this application include: This application provides a deep learning-based diffraction error modeling method. Based on LSTM, a diffraction error model of the satellite is established to achieve accurate weighting of each satellite, thereby improving the accuracy and reliability of GNSS positioning and providing more reliable results for high-precision tasks such as landslide detection. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0019] Figure 1 A flowchart illustrating a deep learning-based diffraction error modeling method provided in an embodiment of this application; Figure 2 A flowchart of a deep learning-based diffraction error modeling method provided in another embodiment of this application; Figure 3 A flowchart illustrating a deep learning-based diffraction error modeling method provided in an embodiment of this application; Figure 4 A structural block diagram of a deep learning-based diffraction error modeling device provided in an embodiment of this application; Figure 5 This is a structural block diagram of an electronic device provided in an embodiment of this application.
[0020] The attached figures are labeled as follows: 11: Acquisition Module; 12: Calculation Module; 13: Building Module; 14: Training Module; 21: Processor; 22: Memory. Detailed Implementation
[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0022] In recent years, machine learning has been widely applied to GNSS error processing due to its potential in handling complex pattern recognition and signal processing tasks. Machine learning methods can effectively identify non-line-of-sight signals and improve positioning accuracy. Many studies have shown that machine learning technology has great potential in improving the accuracy and efficiency of GNSS signal processing.
[0023] Figure 1 A flowchart illustrating a deep learning-based diffraction error modeling method provided in one embodiment of this application. See also... Figure 1 ,include: S101: Obtain raw GNSS observation data of the satellite observed by the monitoring station and reference station under historical conditions, as well as the position coordinates of the observation station and reference station.
[0024] In some examples, raw GNSS observation data includes phase observations. Pseudo-distance observation Signal-to-noise ratio (SNR) observation.
[0025] S102, the diffraction error of the satellite is calculated based on the original GNSS observation data, the position coordinates of the observation station and the reference station.
[0026] In some examples, step S102 includes: S1021, Based on the original GNSS observation data, determine the continuous time period in which the ambiguity remains unchanged, and calculate and fix the ambiguity value within the continuous time period.
[0027] In some examples, step S1021 includes: Step 1: Based on the raw GNSS observation data, calculate the HMW combination (geometrically independent combination) and GF combination (geometrically free combination).
[0028] In some examples, the formulas for calculating the HMW and GF combinations are as follows:
[0029] In the formula, This represents the observations of HMW (Hatch–Melbourne–Wübbena). This indicates observations without geometric combinations. This represents the clutter observation value at frequency point 1. This represents the clutter observation value at frequency point 2. This indicates the frequency of frequency point 1. This indicates the frequency of frequency point 2. This represents the pseudorange observation value at frequency point 1. This represents the pseudorange observation value at frequency point 2.
[0030] Step 2: Detect whether the ambiguity changes based on the HMW combination and GF combination, and then determine the continuous time period in which the ambiguity remains unchanged.
[0031] In some examples, a cycle slip is detected if the difference HMW and GF combination between adjacent epochs is greater than 0.5 cycles and 0.25 cycles, respectively, which means that one continuous arc segment ends and another begins.
[0032] Step 3: Calculate the floating-point ambiguity and variance using the least squares method according to the double-difference variance formula, and then calculate the integer ambiguity using the Lambda method (least squares ambiguity decorrelation adjustment) based on the floating-point ambiguity and variance, and then fix the ambiguity value within the continuous time period.
[0033] In some examples, the ambiguity value is calculated as follows:
[0034]
[0035]
[0036]
[0037] in, It is frequency The wavelength. It is the geometric distance from the satellite to the antenna. It is an integer ambiguity. This represents the unmodeled residuals, including multipath errors, diffraction errors, and carrier phase measurement noise. (Subscript) Indicates a monitoring station. Indicates a reference station. Superscript Indicates satellite . Indicates satellite .
[0038] Indicates monitoring station Received satellite of Frequency point carrier phase observation value. Indicates monitoring station Received satellite of Frequency point carrier phase observation value. Indicates monitoring station Received satellite of Frequency point carrier phase observation value. Indicates monitoring station Received satellite of Frequency point carrier phase observation value.
[0039] Indicates satellite With monitoring station The distance between the antennas. Indicates satellite Reference Station The distance between the antennas. Indicates satellite With monitoring station The distance between the antennas. Indicates satellite Reference Station The distance between the antennas.
[0040] Indicates monitoring station Received satellite of Frequency point integer ambiguity. Indicates monitoring station Received satellite of Frequency point integer ambiguity. Indicates monitoring station Received satellite of Frequency point integer ambiguity. Indicates monitoring station Received satellite of Frequency point integer ambiguity.
[0041] This represents the double difference operator, and the specific calculation method is as described above.
[0042] Indicates to , , , Perform double difference operations. Indicates to , , , Perform double difference operations. Indicates to , , , Perform double difference operations.
[0043] The satellite position can be calculated from the broadcast ephemeris. The positions of the reference station and the monitoring station are known. The floating-point ambiguity N and variance can be calculated using the least squares method, and the ambiguity can be fixed by applying the Lambda method.
[0044] S1022, Calculate the double-difference carrier phase residual based on the ambiguity value, the position coordinates of the observation station and the reference station.
[0045] In some examples, the calculation process for the double-difference carrier phase residual is as follows:
[0046] Since the satellite with the highest elevation angle is usually chosen as the master satellite, single-difference observations from the master satellite are unlikely to be affected by diffraction, and the noise in the observations can be ignored. Therefore, the characteristics of the double-difference carrier phase residuals depend to a large extent on the single-difference observations from non-master satellites.
[0047] In some examples, if the double-difference carrier phase residual is greater than zero and monotonically increasing, the monitoring station exhibits a diffraction effect; if the double-difference carrier phase residual is less than zero and monotonically decreasing, the reference station exhibits a diffraction effect.
[0048] S1023, Identify the site where the diffraction effect occurs based on the error, and determine the diffraction error of the site where the diffraction effect occurs from the double-difference carrier phase residual.
[0049] Multipath errors are typically less than 1 / 4 of the carrier phase period (close to 5 cm) and exhibit obvious periodicity. However, diffraction errors show a clear trend and, as shown, can accumulate to more than 5 cm or even decimeter levels in a short period of time. They usually occur at the two ends of the time series and can be separated by differential methods or filtering techniques (such as low-pass filtering).
[0050] S103. Establish the first dataset based on the satellite's diffraction error, the satellite's position relative to the monitoring station, and the signal strength of the satellite obtained by the monitoring station.
[0051] S104 trains the first diffraction error prediction model for the monitoring station based on the first dataset.
[0052] In some examples, during deep learning training, the extracted feature data, such as diffraction error, signal-to-noise ratio (i.e., satellite signal strength), satellite elevation angle, and azimuth angle, are first divided into training and validation sets. Then, the training set is fed into a Long Short-Term Memory (LSTM) network, and predictions are calculated through forward propagation. The mean squared error (MSE) or mean absolute error (MAE) is used as the loss function to measure the difference between the predictions and the actual diffraction errors. Next, the Adam optimization algorithm is used to update the weights and biases of the LSTM network through backpropagation to minimize the loss function. During training, appropriate training epochs and batch sizes are set, and the validation set is used to monitor for overfitting. If necessary, early stopping or hyperparameter adjustment is used to optimize model performance.
[0053] In some examples, after establishing the first diffraction error prediction model, the diffraction error of the satellite signal received by the monitoring station is determined in real time using the first diffraction error prediction model, and the satellites are weighted according to the diffraction error model:
[0054] In the formula, Indicates satellite diffraction error, Indicates satellite The weight.
[0055] Figure 2 A flowchart illustrating a deep learning-based diffraction error modeling method provided in another embodiment of this application. See also... Figure 2 ,include: S201: Obtain raw GNSS observation data of the satellite observed by the monitoring station and reference station under historical conditions, as well as the position coordinates of the observation station and reference station.
[0056] See step S101.
[0057] S202, the diffraction error of the satellite is calculated based on the original GNSS observation data and the position coordinates of the observation station and the reference station.
[0058] See step S102.
[0059] S203. The first dataset is established based on the satellite's diffraction error, the satellite's position relative to the monitoring station, and the signal strength of the satellite obtained by the monitoring station.
[0060] See step S103.
[0061] S204. Based on the first dataset, the first diffraction error prediction model for the monitoring station is trained.
[0062] See step S104.
[0063] S205. A second dataset is established based on the satellite's diffraction error, the satellite's position relative to the reference station, and the satellite's signal strength obtained by the reference station.
[0064] In some examples, the process of creating the second dataset for the reference station is the same as that of creating the first dataset for the monitoring station, which will not be elaborated here.
[0065] S206. Based on the second dataset, fit the satellite's position information relative to the monitoring station and the satellite's diffraction error to obtain the second diffraction error prediction model for the reference station.
[0066] In some examples, the training process of the second diffraction error prediction model of the reference station is the same as that of the first diffraction error prediction model of the monitoring station, which will not be elaborated here.
[0067] S207, based on the first diffraction error model and the second diffraction error model, determines in real time the satellite signals exhibiting diffraction effects, and weights the satellite signals received by the reference station and the monitoring station respectively.
[0068] In some examples, the weighting methods for the reference station and the monitoring station are the same, and will not be elaborated here. The satellite signals received by the monitoring station are weighted according to the diffraction error predicted by the first diffraction error model, and the satellite signals received by the reference station are weighted according to the diffraction error predicted by the second diffraction error model.
[0069] See Figure 3 , Figure 3 A flowchart of a deep learning-based diffraction error modeling method provided in an embodiment of this application.
[0070] Figure 4 A structural block diagram of a deep learning-based diffraction error modeling device provided in an embodiment of this application. See also... Figure 4 ,include: The acquisition module 11 is used to acquire the raw GNSS observation data of the satellite observed by the monitoring station and the reference station under historical conditions, as well as the position coordinates of the observation station and the reference station; Calculation module 12 is used to calculate the diffraction error of the satellite based on the original GNSS observation data, the position coordinates of the observation station and the reference station; Module 13 is established to create a first dataset based on the satellite's diffraction error, the satellite's position relative to the monitoring station, and the signal strength of the satellite obtained by the monitoring station. Training module 14 is used to train the first diffraction error prediction model of the monitoring station based on the first dataset.
[0071] Figure 5 This is a structural block diagram of an electronic device provided according to an embodiment of this application. See also... Figure 5 Electronic devices may include Figure 4The aforementioned deep learning-based diffraction error modeling device. Typically, the electronic device includes a processor 21 and a memory 22. The processor 21 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 21 can be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 21 may also include a main processor and a coprocessor. The main processor is used to process data in the wake-up state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor used to process data in the standby state. The memory 22 may include one or more computer-readable storage media, which may be non-transitory. The memory 22 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer-readable storage medium in memory 22 is used to store at least one instruction, which is executed by processor 21 to implement the deep learning-based diffraction error modeling method executed by an electronic device provided in the method embodiments of this application.
[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A deep learning-based diffraction error modeling method, characterized in that, include: Acquire raw GNSS observation data of satellites observed by monitoring stations and reference stations under historical conditions, as well as the position coordinates of the observation stations and reference stations; The diffraction error of the satellite is calculated based on the raw GNSS observation data, the position coordinates of the observation station and the reference station; The first dataset is established based on the satellite's diffraction error, the satellite's position relative to the monitoring station, and the signal strength of the satellite obtained by the monitoring station. Based on the first dataset, the first diffraction error prediction model for the monitoring station was trained.
2. The deep learning-based diffraction error modeling method according to claim 1, characterized in that, Raw GNSS observation data includes phase observations, pseudorange observations, and signal-to-noise ratio observations.
3. The deep learning-based diffraction error modeling method according to claim 1, characterized in that, The steps for calculating the satellite's diffraction error based on the raw GNSS observation data and the position coordinates of the observation station and reference station include: Based on the raw GNSS observation data, a continuous time period with unchanged ambiguity is determined, and the ambiguity value within the continuous time period is calculated and fixed. Calculate the double-difference carrier phase residual based on the ambiguity value, the position coordinates of the observation station and the reference station; The site where the diffraction effect occurs is identified based on the error, and the diffraction error of the site where the diffraction effect occurs is determined from the double-difference carrier phase residual.
4. The deep learning-based diffraction error modeling method according to claim 3, characterized in that, Based on raw GNSS observation data, the steps of determining a continuous time period with unchanged ambiguity, and calculating and fixing the ambiguity value within that continuous time period include: Based on the raw GNSS observation data, the HMW combination and GF combination were calculated respectively. The ambiguity is detected by combining HMW and GF, and then the continuous time period in which the ambiguity remains unchanged is determined. Based on the double difference variance formula, the floating-point ambiguity and variance are calculated using the least squares method. Then, the integer ambiguity is calculated using the Lambda method based on the floating-point ambiguity and variance, and the ambiguity value within the continuous time period is fixed.
5. The deep learning-based diffraction error modeling method according to claim 3, characterized in that, The steps for identifying the sites where diffraction effects occur based on errors, and determining the diffraction errors of these sites from the double-difference carrier phase residuals, include: If the phase residual of the double-difference carrier is greater than zero and monotonically increasing, then the monitoring station has experienced a diffraction effect. If the phase residual of the double-difference carrier is less than zero and decreases monotonically, then the reference station has experienced a diffraction effect.
6. The deep learning-based diffraction error modeling method according to claim 1, characterized in that, Also includes: A second dataset is established based on the satellite's diffraction error, the satellite's position relative to the reference station, and the satellite's signal strength obtained by the reference station. Based on the second dataset, the satellite's position information relative to the monitoring station and the satellite's diffraction error are fitted to obtain the second diffraction error prediction model for the reference station.
7. The deep learning-based diffraction error modeling method according to claim 6, characterized in that, Also includes: Based on the first and second diffraction error models, satellite signals exhibiting diffraction effects are determined in real time, and weighted for the satellite signals received by the reference station and the monitoring station, respectively.
8. A diffraction error modeling device based on deep learning, characterized in that, include: The acquisition module is used to acquire the raw GNSS observation data of the satellite observed by the monitoring station and the reference station under historical conditions, as well as the position coordinates of the observation station and the reference station; The calculation module is used to calculate the diffraction error of the satellite based on the raw GNSS observation data and the position coordinates of the observation station and the reference station. A module is established to create the first dataset based on the satellite's diffraction error, the satellite's position relative to the monitoring station, and the signal strength of the satellite obtained by the monitoring station. The training module is used to train the first diffraction error prediction model of the monitoring station based on the first dataset.
9. An electronic device, characterized in that, Includes the deep learning-based diffraction error modeling device as described in claim 8.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one piece of program code, which is executed by a processor to implement the deep learning-based diffraction error modeling method as described in any one of claims 1 to 7.