Combined navigation positioning method for extreme weather rescue

Abstract

To provide a combined navigation positioning method for extreme weather rescue.SOLUTION: A method disclosed herein comprises: a step 1 of constructing an original training data set; a step 2 of adding incremental data set training samples on the basis of the original training data set, constructing an incremental grid real-time correction model, and obtaining a real-time dynamically updated multipath heat map; and a step 3 of assisting GNSS / IMU tight combination solution according to the real-time dynamically updated multipath heat map, and completing the combined navigation and positioning for extreme weather rescue.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a combination navigation positioning method, and more particularly to a combination navigation positioning method for extreme weather rescue. [Background technology]

[0002] This section provides only background information related to the present disclosure and is not necessarily prior art. Coastal cities are susceptible to typhoons and storms. Once a natural disaster strikes, emergency rescue of people in affected areas is crucial. High-precision, real-time, and reliable navigation and positioning technologies are also essential to improve rescue efficiency. However, signal attenuation caused by extreme weather and obstructions by tall buildings in complex urban environments make GNSS (Global Navigation Satellite System) signals susceptible to multipath interference and non-line-of-sight reception. This results in low-quality GNSS signals due to the multipath effect, severely limiting the measurement accuracy of GNSS and its combined navigation systems. Therefore, accurate positioning in complex urban environments with extreme weather is crucial for the evacuation of rescue vehicles and personnel after natural disasters.

[0003] To address the limitations of navigation and positioning technologies mentioned above, domestic and international researchers have conducted a great deal of valuable research, mainly categorized into signal processing methods, antenna design methods, and observation-based modeling methods. Signal processing-based methods cannot effectively identify and attenuate non-line-of-sight (NLOS) signals, and antenna line design methods are heavy and expensive, making these two methods unsuitable for widespread urban positioning deployment. Observation modeling-based error correction methods use mathematical statistics, machine learning, and other methods to establish a mathematical model between the original GNSS observations and multipath errors, obtain predicted values ​​for multipath errors, and then use the predicted values ​​to perform error correction and improve navigation and positioning accuracy. Currently, there are two main types of observation modeling-based error correction methods: control point data-based modeling and area grid-based modeling.

[0004] Sun proposed a training regression model based on GBDT (Gradient Boosting Decision Tree), which predicts and corrects multipath errors by comprehensively considering signal strength, satellite altitude angle, and pseudorange residuals. Positioning tests in complex urban environments showed that the 3D positioning accuracy of the algorithm based on multipath error correction was improved by more than 70% compared to conventional positioning algorithms. Qin proposed a multipath error prediction algorithm based on bagging trees that predicts and corrects multipath errors by considering signal strength and satellite altitude angle, primarily improving GNSS positioning accuracy in lush foliage environments, with the 3D RMSE of the positioning results improving by more than 80%. In addition to directly correcting errors using multipath prediction values, these predictions are also used in conjunction with Kalman filtering to adjust filter parameters. Zhang proposed a GNSS measurement noise covariance update strategy based on the average of the PDOP value and the predicted multipath error, using signal strength and satellite altitude angle as inputs to a bagging tree model. This strategy adaptively adjusts the Kalman filter parameters of the conventional GNSS / IMU loose combination algorithm. Compared with the conventional GNSS / IMU loose combination algorithm, this method improved the horizontal position accuracy by 28.15% and the three-dimensional position accuracy by 43.10%. Sun proposed a measurement noise covariance update scheme for the GNSS / IMU tight combination algorithm in urban areas by considering signal strength, satellite altitude angle, and coordinate information, using a combined bagged regression tree model to predict multipath error, and constructing an adaptive factor based on the predicted value of multipath error. This algorithm can improve the three-dimensional positioning accuracy by 9.21 m, which is 55% better than the traditional tight combination GNSS / IMU algorithm based on the extended Kalman filter and 15% better than the multipath error compensation method based on the extended Kalman filter, proving that the adaptive filtering algorithm based on multipath error prediction is effective in improving the positioning accuracy of combination GNSS / IMU navigation.However, methods based on modeling control point data still have problems: the same model predicts multipath errors for all urban scenarios where the localization scheme is not sophisticated enough, resulting in insufficient accuracy in multipath error prediction.

[0005] A navigation and positioning support method based on area-grid multipath modeling divides the positioning area into grid units, establishes a multipath error model for each grid unit, and precisely models the multipath effect. This effectively improves the accuracy and stability of navigation and positioning, while also offering the advantages of simple piggybacking and low cost. Sun proposes an area-grid-based multipath error prediction and correction strategy to improve GNSS positioning accuracy in urban environments. This method uses signal strength, satellite elevation angle, and a priori location information for the area as input features. It uses a machine learning random forest algorithm to train a regression model to predict multipath errors and then use them to correct pseudoranges. Dynamic testing in an urban environment demonstrated that this algorithm improved positioning accuracy by 41.50% horizontally and 63.38% three-dimensionally compared to traditional positioning methods. Lee introduced a nonlinear regression model to estimate high-elevation angle clock differences, and trained the extracted unbiased multipath using an SVM to generate a nonlinear multipath map model based on the azimuth and altitude angles of each satellite in the urban depth region. This model effectively improved positioning accuracy in the urban depth region by 58.4% and 77.7% in the vertical direction. Conventional grid and map models are divided into two-dimensional planes, making them difficult to accurately calibrate for users at different altitudes. Sun et al. extended the grid prediction model from a two-dimensional plane to a three-dimensional stereo layout. They proposed a navigation and positioning method based on three-dimensional grid multipath modeling for complex urban environments, fully considering the reflection environments at different altitudes and improving positioning accuracy in complex intersections and steep road sections.Although regional grid division modeling of multipath in complex urban environments can effectively improve positioning accuracy to a certain extent, this method still has limitations. Due to the inaccuracy of a priori location information caused by signal attenuation in complex urban environments, especially under the influence of extreme weather, it is easy to match with an incorrect grid prediction model. The current environment, which is affected by extreme weather, differs significantly from the pre-trained environment. The pre-trained multipath prediction model cannot dynamically adjust according to the signal characteristics of the current environment, and lacks adaptability to extreme environments and real-time positioning processing. This results in large positioning errors, makes it difficult to accurately position rescue vehicles, and reduces rescue efficiency in urban environments experiencing extreme weather.

[0006] In summary, existing technologies have the following drawbacks: (1) Existing GNSS / IMU combination navigation schemes based on machine learning methods for area grid multipath modeling have the problems of fixed pre-training datasets and a single sample library. Especially in complex urban environments under extreme weather conditions, signal attenuation is severe, and the degradation of training samples leads to deterioration of predictions. This makes it impossible to improve the positioning accuracy of combination navigation assisted by area grid multipath modeling, affecting the positioning accuracy of rescue vehicles and reducing rescue efficiency. (2) The existing grid multipath error prediction model built by the navigation and positioning scheme based on regional grid multipath modeling in complex urban environments is static, while the urban complex scenario in extreme weather is becoming increasingly large. Meanwhile, the multipath error prediction model trained based on offline data lacks adaptability to the current environment and the real-time nature of the positioning process, resulting in insufficient prediction accuracy of the model to ensure the reliability of the combined navigation and positioning scheme for extreme weather rescue. As a result, the model's prediction accuracy is insufficient to ensure the reliability of the combined navigation and positioning system used for extreme weather rescue. Please note that the information disclosed in the Background section above is intended to provide a better understanding of the context of the present disclosure and may therefore include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] Object of the invention: The technical problem to be solved by the present invention is to address the shortcomings of the prior art and provide a combined navigation and positioning method for extreme weather rescue.

[0008] In order to solve the above technical problems, the present invention provides a combination navigation and positioning method for extreme weather rescue, comprising the following steps: Step 1: Build an original training dataset. Step 2: Based on the original training dataset, an incremental dataset training sample is added, and an incremental grid real-time correction model is constructed to obtain a real-time dynamically updated multi-path heat map; Step 3: Assist GNSS / IMU tight combination calculation based on real-time dynamically updated multi-path heat map to complete combination navigation positioning for extreme weather rescue.

[0009] Furthermore, the step of constructing an original training dataset in step 1 specifically includes: Step 1-1: Calibrate the training dataset. Step 1-2, Build a region grid layout, Step 1-3: Dividing the sample data in the training dataset according to the satellite constellation types of GPS satellites and BeiDou satellites; Based on the corresponding reference position of the sample data, determine the grid layer to which the sample data belongs according to the three-dimensional stereo grid layout of step 1-2, and obtain the original training data set, which are respectively denoted as follows: GPS original training dataset

number

number

[0010] Furthermore, the step of calibrating the training dataset described in step 1-1 is specifically as follows: Step 1-1-1: A positioning carrier repeatedly drives a predetermined route in an urban area to collect a priori data, which is used to construct an original training dataset; Step 1-1-2: Perform differential positioning using the high-precision combination navigation positioning system and base station GNSS observation data to obtain reference coordinate information; Step 1-1-3: Collect the original observations output from the GNSS receiver, and extract the corresponding input features in the original observations, where the features include the pseudorange residual δ, the carrier-to-noise ratio C / N0, and the satellite elevation angle θ. e and satellite azimuth angle θ a and a set of feature variables is shown as follows:

number

number

[0011] Furthermore, the step of constructing a region grid layout described in step 1-2 specifically includes: Divide the two-dimensional plane grid, that is, based on the geographic information and GNSS position data of the urban area, select the smallest square A in the area containing the position data as the modeling area, one side of the square is parallel to the E-axis under the ENU coordinate system, divide a hexagonal grid within the square area, and divide the ENU coordinate where the hexagon is located as a unit, set the length of the hexagonal side to l, and set the center point of the hexagonal grid to 0; The divided two-dimensional planar grid is expanded in the height direction into grid rows, and each grid row is divided into n grid layers according to the selected height h, thereby completing the construction of a three-dimensional solid grid layout.

[0012] Furthermore, the step of constructing an incremental grid real-time correction model described in step 2 specifically includes: Step 2-1: Establish a GNSS / IMU combined navigation calculation result feedback update mechanism, which is specifically as follows: Match and call the existing three-dimensional grid prediction model to obtain the current calendar element multi-path error prediction value, and determine whether the following is satisfied:

number

[0013] Furthermore, the step of constructing a real-time dynamically updated multi-path heat map described in step 2-2 is specifically as follows: Step 2-2-1: Build an incremental dataset training sample, specifically as follows: According to the spatial coordinates of the current calendar element positioning carrier, search for the grid layer to which it belongs in the 3D grid layout, and obtain the pseudorange residual δ, the carrier-to-noise ratio C / N0, and the satellite elevation angle θ. e and satellite azimuth angle θ a Extract input features in the observations including

number

number

number

number

[0014] Let X be the incremental dataset training samples. C , Lagrange multiplier initialization condition Δa c = 0, and then use the current SVR prediction rule to determine whether it violates the KKT condition. Then, a new optimization problem is created to satisfy the KKT condition again, i.e., the current SVR prediction rule is updated. The update process is as follows: According to the KKT conditions, the deviation coefficient θ i and sampling error h(X i ) are defined as follows:

number

number

number

number

number

number

[0015] Furthermore, the SVR regression training for each grid layer described in step 2-2-2 specifically includes: Let N be the number of samples in the original training dataset, and ε be a predetermined deviation threshold between the predicted value of the prediction function f(X) and the true value of the multipath error Y. The goal of the SVR regression training is to minimize the error between the predicted value of the prediction function and the true value, which can be achieved by solving the following optimization problem:

number

number

number

number

number

number

number

number

[0016] Furthermore, the step of supporting GNSS / IMU tight combination calculation based on the real-time dynamically updated multi-path heat map described in step 3 specifically includes: Step 3-1: Under abnormal weather rescue conditions, extract input features and position accuracy factor (PDOP) values ​​from the original observations of the GNSS receiver, and obtain the initial position based on the IMU mechanical configuration; Step 3-2: Call the real-time dynamically updated multi-path heat map, match the initial position to the grid layer to which it belongs, perform position calculation, and obtain the specific position of the carrier in the abnormal weather rescue state.

[0017] Furthermore, the specific method of the position calculation described in step 3-2 is as follows: If the position accuracy factor PDOP value is equal to or greater than the threshold, no correction is performed, and the GNSS / IMU tight combination Kalman filtering is directly performed to calculate the position solution, and the position solution is taken as the final solution. If the PDOP value is less than the threshold, the pseudorange error is predicted using the grid prediction model, and the incremental grid real-time correction model established in step 2 is used to make the next correction.

[0018] Furthermore, the step of predicting the pseudorange error using the grid prediction model described in step 3-2 specifically includes: According to the grid prediction model, the multipath error values ​​obtained by the adjacent grids are extracted, and the weights W are assigned to the adjacent grids. n After the dynamic weighting process, the final multipath error prediction is

number

number

number

[0019] The present invention has the following beneficial effects. (1) This invention addresses the problem that single-sample degradation during offline learning affects the prediction performance of existing area grid multipath modeling methods based on machine learning methods. It constructs incremental dataset training samples and solves the sample degradation problem through incremental processing of training samples. When the current positioning result is in the same grid and belongs to the same constellation type as the previous incremental data, it uses the positioning results of multiple calendar elements to perform incremental SVR (Support Vector Regression) training on each grid with the original training dataset as batch training data samples, fully utilizing the signal features and multipath errors in the current positioning environment, and continuously receiving and accumulating new data to improve sample diversity. (2) Focusing on the problem that the existing multipath error prediction model based on machine learning is fixed, resulting in insufficient prediction accuracy and real-time performance, a feedback mechanism for updating the spatiotemporal information grid prediction model using the current calendar element positioning results is used. When the quality of the grid prediction model does not meet the requirements, the feedback mechanism obtains the positioning results of the calendar element, extracts the spatiotemporal information, and immediately corrects and optimizes the grid prediction model accordingly. At the same time, the grid model that is continuously updated and optimized during the model prediction process and the adjacent grids are dynamically weighted, converting single-grid model prediction into multiple-grid model prediction, thereby realizing adaptive updating of the grid prediction model. This improves the accuracy and adaptability of the machine learning-based grid multipath error prediction model in urban environments affected by extreme weather.

[0020] The advantages of the above and / or other aspects of the present invention will become more apparent as the present invention is more particularly described below in conjunction with the accompanying drawings and specific embodiments. [Brief explanation of the drawings]

[0021] [Figure 1] FIG. 1 is a schematic diagram of the general flow of the present invention. [Figure 2] FIG. 1 is a schematic diagram of incremental SVR training in the present invention. [Figure 3] FIG. 1 is a schematic diagram of the training sample grid membership determination method in the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0022] The core content of this invention is a combined navigation and positioning method for extreme weather rescue, which consists of three steps: (1) constructing an original training dataset, (2) constructing an incremental grid real-time correction model, and (3) supporting GNSS / IMU tight combination calculation with real-time dynamic updating of multipath heatmap. The general flow is shown in Figure 1.

[0023] (1) Construction of an original training dataset 1) Calibration of the training dataset The positioning carrier repeatedly drives a predetermined route in the urban area to collect a priori data and use it to construct an original training dataset. Then, the reference coordinate information is obtained from the high-precision combination navigation positioning system and the base station GNSS observation data differential positioning. At the same time, the original observation quantities output from the GNSS receiver are collected, and the pseudorange residual δ, carrier-to-noise ratio C / N0, and satellite height angle θ are calculated. e and satellite azimuth angle θ a Extract the corresponding input features in the original observations, including

number

[0024] The pseudorange error value obtained by inverting the reference coordinate information and the GNSS original observation amount and the extracted feature variables are calibrated, and the pseudorange error value is shown as follows: Δρ=c(Δδt u -Δδt s )+ΔI+ΔT+ε where c is the speed of light in a vacuum and Δδt s indicates the residual error after correcting the satellite clock difference, and Δδt u indicates the residual error after correcting the receiver clock difference, ΔI indicates the residual error due to insufficient correction for the ionospheric delay, and ΔT indicates the residual error due to insufficient correction for the troposphere delay. Since the error due to the multipath effect in ε is much larger than the other errors and the four residuals are negligible compared to the multipath error, the error due to the multipath effect in the pseudorange error can be obtained, that is, the calibration of the multipath error is completed and the above formula can be regarded as the multipath error. Each set of feature variables is calibrated with the corresponding multipath error, and labeled sample data is obtained, which is used to create the training dataset.

number

[0025] 2) Building a region grid layout Based on the geographic information and GNSS position data of the urban area, the smallest square A in the area containing the position data is taken as the modeling area, and one side of the square is parallel to the E axis under the ENU coordinate system. A hexagonal grid is divided into units of ENU coordinates where the hexagon is located within the square area, and the length of the side of the hexagon is set to l and the center point of the hexagonal grid is set to o according to the actual situation and positioning accuracy.

[0026] Because GNSS / IMU signal modeling based on a three-dimensional grid can fully account for reflection environments at different altitudes, adding the elevation dimension to the real-time, dynamically updated multipath grid proposed in this method can improve positioning performance in rugged urban road sections, complex interchanges, and UAV application scenarios. Therefore, it is also necessary to extend the divided two-dimensional planar grid into grid columns in the height direction, and then sequentially divide each grid column into n grid layers according to the selected height h.

[0027] Because GPS (Global Positioning System) and BDS (BeiDou Navigation Satellite System) have different orbital repetition periods, the training data needs to be divided according to the satellite constellation type. Based on this, it is also necessary to determine the grid to which the training sample belongs according to the divided grid layout based on the corresponding reference position of the sample. Finally, the original training data set is constructed, as shown below:

number

[0028] (2) Development of an incremental grid real-time correction model 1) GNSS / IMU combination navigation calculation result feedback update mechanism (IMU, Inertial Measurement Unit) Since the pseudorange error value based on the output of the three-dimensional grid prediction model can effectively reflect the current grid positioning quality, when matching and calling the existing three-dimensional grid prediction model step, the current calendar element error prediction value is extracted and compared with the average absolute pseudorange error value of the open-area GNSS data to evaluate the grid prediction model quality. If the calendar element error prediction value satisfies the following formula, it is determined that the calculated carrier position result needs to be fed back and the grid prediction model needs to be updated; otherwise, it is determined that the grid quality meets the condition, so there is no need to update the grid prediction model, and the original grid prediction model will continue to be used for the next positioning calculation.

number

[0029] Furthermore, if it is determined that the grid forecast model accuracy is insufficient, the time-space information included in the calendar element carrier position information is extracted, and the current calendar element T i and current position P i and the current position P i contains the coordinate information of the ENU coordinate system where the calendar element carrier is located, and is used as real-time data for correcting the grid forecast model at the current time point i.

[0030] 2) Modeling multipath errors for grid-wise incremental SVR 1. Construction of incremental dataset training samples using extracted spatiotemporal information First, according to the spatial coordinate data of the calendar element positioning carrier corresponding to the grid prediction model for positioning, the pseudorange residual δ, the carrier-to-noise ratio C / N0, the satellite altitude angle θ e and satellite azimuth angle θ a Extract input features in the observations containing

number

[0031] The spatial coordinate data of the render element positioning results are inverted to calibrate the pseudorange error values ​​and the extracted feature variables. Each set of feature variables and the corresponding multipath error are calibrated to generate incremental training samples.

number

[0032] Similarly, the incremental dataset also needs to be split according to the GPS and BDS constellation type and grid for the next training, so the incremental dataset is denoted as follows,

number

[0033] 2. Grid-wise incremental SVR multipath error model training Determine the grid layer to which the sample data in the original training dataset belongs. Let o be the hexagonal center point of the 3D grid layout. The x-axis and y-axis correspond to the E-axis and N-axis in the ENU coordinate system, respectively. Determine whether the sample data is inside the hexagonal grid based on the following two conditions:

number

[0034] For the original sample set, SVR can be understood as solving a convex quadratic programming problem with multiple coefficients equal to the number of training samples N. The resulting prediction function f(X) must be as close as possible to the true value Y. Let ε be the allowable deviation between f(X) and Y. In SVR, the value of Y is a continuous value. The corresponding optimization problem can be expressed as follows:

number

number

number

number

number

[0035] The following KKT conditions (Karush-Kuhn-Tucker) are satisfied, and the KKT conditions are sufficient conditions for solving convex quadratic programming problems:

number

number

number

[0036] Similarly, as shown in Figure 3, according to the grid layer to which the incremental dataset training samples belong, the prediction rules of the corresponding grid layer can be updated and optimized. For the batch incremental learning samples, the incremental data are matched one by one to the grid layer to which the incremental data is located, and incremental training is performed with the original training samples in the grid layer. c When arrives, first Δa c = 0, and then determine whether the current SVR prediction rule violates the KKT condition. If the KKT condition is not violated, i.e., it is determined that it does not affect the SVR prediction rule, there is no need to update the model. If the KKT condition is violated, the sample may cause the transfer of past sample sets, and finally the KKT condition is satisfied again, and the current SVR prediction model is updated accordingly. The update process is as follows:

[0037] The SVR binomial form is transformed into a Lagrangian form and the KKT conditions are optimized as follows:

number

number

number

number

number

number

number

[0038] When selecting and optimizing multiple Lagrangian multipliers, select the multiplier that most violates the KKT conditions and update the model parameters with the new Lagrangian multiplier values ​​as follows:

number

number

number

[0039] 3. Complete the construction of a multi-path heat map that dynamically updates in real time At this point, the construction of a real-time dynamically updated multipath heat map is completed, which includes the original grid multipath error prediction model and grid model prediction rules, and the optimized grid prediction model and rules, and the regional grid prediction model is dynamically updated in real time according to the positioning results on the original basis.

[0040] (3) Supporting GNSS / IMU combined navigation calculations with real-time dynamic updates of multipath heat maps Extract input features, PDOP (Position Dilution of Precision) values ​​from the original observations of the GNSS receiver, and call up a multi-path heat map dynamically updated in real time according to the initial position obtained from the IMU mechanical configuration, match the initial position to the grid layer it belongs to, and then match it to the corresponding three-dimensional grid prediction model.

[0041] The data quality is tested using the following formula to determine whether the grid model is valid. If the PDOP value is equal to or greater than the set threshold, no correction is performed, and the GNSS / IMU tight combination Kalman filtering is directly performed to calculate the position solution, which is then used as the final solution. If the PDOP value is less than the threshold, the grid multipath error model is used to predict the pseudorange error, and the next correction is performed. PDOP <k When the prediction step is performed, the multipath error values ​​obtained by the dynamic update of the prediction model for the adjacent grids are extracted and weighted to the adjacent grids. nThe final pseudorange error prediction value obtained after the dynamic weighting process is

number

[0042] The prediction results obtained by the area grid-based multipath error prediction model continue to change as the grid accuracy improves. Therefore, the pseudorange error prediction value is used to calculate the adaptive factor

number

number

[0043] In a specific embodiment, the present application provides a computer storage medium and a corresponding data processing unit, where the computer storage medium can store a computer program, and when the computer program is executed by the data processing unit, it can perform some or all of the steps in the inventive content and embodiments of the combination navigation and positioning method for extreme weather rescue provided by the present invention. The storage medium can be a magnetic disk, a compact disk, a read-only memory (ROM), a random access memory (RAM), etc.

[0044] It can be clearly understood by those skilled in the art that the technical solutions in the embodiments of the present invention can be implemented by a computer program and a corresponding general-purpose hardware platform. Based on this understanding, the technical solutions in the embodiments of the present invention, whether substantially or in part contributing to the prior art, may be embodied in the form of a computer program, i.e., a software product, which can be stored in a storage medium and includes a number of instructions that can cause a device with a data processing unit (such as a personal computer, a server, a microcontroller, an MCU, or a network device) to execute the methods described in each embodiment or part of the embodiments of the present invention.

[0045] The present invention provides an idea and a method for a combined navigation and positioning method for extreme weather rescue, and the specific technical solution can be realized by various methods and means. It should be understood that the above is only a preferred embodiment of the present invention, and that those skilled in the art can make many improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should be considered as part of the scope of protection of the present invention. Each component not explicitly defined in this embodiment can be realized by conventional technology.

Claims

1. It includes the following steps: Step 1: Build an original training dataset. Step 2: Add incremental dataset training samples based on the original training dataset, build an incremental grid real-time correction model, and obtain a multi-path heat map that is dynamically updated in real time; Step 3: Supporting GNSS / IMU tight combination calculation based on the dynamically updated multipath heat map in real time, and completing the combination navigation positioning for extreme weather rescue.

2. The step of constructing the original training dataset in step 1 specifically includes: Step 1-1: Calibrate the training dataset; Step 1-2: Build a region grid layout. Step 1-3: Dividing the sample data in the training dataset according to the satellite constellation types of GPS satellites and BeiDou satellites; According to the corresponding reference position of the sample data, determine the grid layer to which the sample data belongs according to the three-dimensional grid layout of step 1-2, and obtain the original training data set, which are respectively denoted as follows: GPS original training dataset [Number 56] and BeiDou original training dataset [Number 57] The combination navigation and positioning method for extreme weather rescue according to claim 1, wherein:

3. The step of calibrating the training dataset described in step 1-1 is specifically as follows: Step 1-1-1: A positioning carrier repeatedly travels a predetermined route in an urban area to collect a priori data, which is used to construct an original training dataset; Step 1-1-2: Perform differential positioning using the high-precision combination navigation positioning system and base station GNSS observation data to obtain reference coordinate information; Step 1-1-3: Collect original observations output from the GNSS receiver, and extract corresponding input features in the original observations, where the features include pseudorange residual δ, carrier-to-noise ratio C / N ０ , satellite altitude angle θ ｅ and satellite azimuth angle θ ａ and a set of feature variables [Number 58] year, Step 1-1-4: Invert the reference coordinate information obtained in step 1-1-2 and the GNSS original observation amount collected in step 1-1-3 to obtain the pseudorange error value Δρ as follows: Dr=c(Δδ､ ｕ -Dst ｓ )+ΔI+ΔT+e where c is the speed of light in a vacuum and Δδt ｓ indicates the residual error after the satellite clock difference correction, and Δδt ｕ indicates the residual error after correcting the receiver clock difference, ΔI indicates the residual error after the ionospheric delay has not been fully corrected, ΔT indicates the residual error after the tropospheric delay has not been fully corrected, The pseudorange error is regarded as a multipath error, Step 1-1-5: Calibrate the pseudorange error value obtained in step 1-1-4 and the feature variables extracted in step 1-1-3, obtain labeled sample data, and create a training dataset. [Number 59] 3. The combination navigation and positioning method for extreme weather rescue according to claim 2, further comprising:

4. The step of constructing the region grid layout described in step 1-2 specifically includes: Divide the two-dimensional plane grid, that is, based on the geographic information and GNSS position data of the urban area, the smallest square A in the area containing the position data is taken as the modeling area, one side of the square is parallel to the E axis under the ENU coordinate system, divide a hexagonal grid inside the square area, and divide the ENU coordinate where the hexagon is located as a unit, set the side length of the hexagon as l, and set the center point of the hexagonal grid as 0; The combined navigation and positioning method for extreme weather rescue according to claim 3, further comprising: extending the divided two-dimensional planar grid into grid rows in a height direction; and dividing each grid row into n grid layers according to a selected height h, thereby completing the construction of a three-dimensional solid grid layout.

5. The step of constructing an incremental grid real-time correction model described in step 2 specifically includes: Step 2-1: Establish a GNSS / IMU combined navigation calculation result feedback update mechanism, which is specifically as follows: Match and call the existing three-dimensional grid prediction model to obtain the current calendar element multi-path error prediction value, and determine whether the following is satisfied: [Number 60] denotes the mean absolute pseudorange error value of the open-area GNSS data, If yes, the original grid prediction model and prediction rules obtained through training in step 2-2-2 are fed back and updated, specifically as follows: The time-space information of the current calendar element positioning carrier is respectively represented as the current calendar element T ｉ and current position P ｉ where the current position P ｉ The current calendar element includes coordinate information of the ENU coordinate system where the positioning carrier is located, and is used as real-time data for correcting the grid forecast model at the current time point i; If not, the update will not be performed. Step 2-2: based on the three-dimensional stereo grid layout, perform incremental SVR multipath error modeling for each grid, i.e., construct a real-time dynamically updated multipath heat map. The combination navigation and positioning method for extreme weather rescue according to claim 4, characterized in that

6. The step of constructing a real-time dynamically updated multipath heatmap described in step 2-2 is specifically as follows: Step 2-2-1: Construct an incremental dataset training sample, specifically as follows: According to the spatial coordinates of the current calendar element positioning carrier, search for the grid layer to which it belongs in the three-dimensional grid layout, and obtain the pseudorange residual δ, the carrier-to-noise ratio C / N ０ , satellite altitude angle θ ｅ and satellite azimuth angle θ ａ Extract input features in the observations including [Number 61] and Inversion and calibration are performed according to the methods in steps 1-1-4 and 1-1-5 to obtain incremental data sets; [Number 62] and According to the method in steps 1-3, the incremental dataset is divided according to the GPS and BDS constellation type and grid, and a GPS incremental dataset and a BeiDou incremental dataset are obtained, which are respectively denoted as follows: [Number 63] Step 2-2-2: Incremental SVR multipath error training is performed for each grid to obtain the original grid prediction model, which specifically includes: Determine the grid layer to which the sample data in the original training dataset belongs, where the center point of the hexagon in the three-dimensional solid grid layout is set to 0, and the x-axis and y-axis correspond to the E-axis and N-axis in the ENU coordinate system, respectively, and determine according to the following two conditions: [Number 64] Only when both of the above two equations are satisfied, the position of the sample data (x ０ , y ０ ) is in a grid column, Iteratively process each sample in the original training dataset, divide it into corresponding grid columns, and divide it into the grid layers to which it belongs according to the height data of each sample in the original training dataset; perform SVR regression training for each grid layer; and obtain an SVR prediction rule, i.e., the original grid model prediction rule; Step 2-2-3: Update the original grid prediction model and prediction rule, i.e., according to the grid layer to which the incremental dataset training samples described in step 2-2-1 belong, update and optimize the SVR prediction rule of the corresponding grid layer, i.e., match the incremental dataset training samples one by one to the grid layer to which they belong, introduce multiple new training samples and perform incremental training together with the original training samples in the grid layer, to obtain a dynamically updated SVR prediction rule, specifically as follows: Let X be the incremental dataset training sample. Ｃ Then, the Lagrange multiplier initialization condition Δac=0 is set, and the current SVR prediction rule is used to determine whether the KKT condition is violated. A new optimization problem is created to satisfy the KKT condition again, i.e., the current SVR prediction rule is updated. The update process is as follows: According to the KKT condition, the deviation coefficient θ ｉ and sampling error h(X ｉ ) are defined as follows: [Number 65] Combined with the current SVR prediction rule, the incremental dataset training samples are divided into three parts based on the presence or absence of support vectors and the presence or absence of boundary support vectors; E set, i.e., the set of boundary support vectors: [Number 66] S set, i.e., the set of support vectors: [Number 67] R set, i.e., the set of non-support vectors: [Number 68] During the incremental training process, the above conditions are used to select new samples. If the new sample belongs to the R set, the sample is discarded directly; if the sample belongs to the E set or the S set, the sample is retained; Finally, the optimized SVR prediction rule is obtained as follows: [Number 69] where f′ is the updated prediction function, and is the final regression curve determined by the support vectors obtained in the balance determination. [Number 70] is the multi-pass error prediction after being updated by the batch incremental model, J is the number of features, Step 2-2-4: completing the construction of a real-time dynamically updated multi-path heat map, wherein the real-time dynamically updated multi-path heat map includes the original grid prediction model and prediction rule, and the updated grid prediction model and prediction rule.

7. The step of performing SVR regression training for each grid layer described in step 2-2-2 specifically includes: Let N be the number of samples in the original training data set, and ε be a preset deviation threshold between the predicted value of the prediction function f(X) and the true value Y of the multipath error. The goal of the SVR regression training is to minimize the error between the predicted value of the prediction function and the true value, which is achieved by solving the following optimization problem: [Number 71] where C is the normalization coefficient and the loss function is [Number 72] and X ｉ denotes the i-th set of input features, [Number 73] Y ｉ denotes the true value of the multipath error of the corresponding i-th set of features, [Number 74] denotes the relaxation variables introduced, J denotes the number of features, ω denotes the weight vector, b denotes the bias, and sum denotes iterative processing over all data points. Introducing the Lagrange coefficient a, the above optimization problem can be converted into a binomial form as follows: [Number 75] According to the following KKT conditions: [Number 76] The SVR prediction rule after nonlinear mapping is obtained as follows: [Number 77] where K(,) denotes the kernel function operation, [Number 78] The combination navigation and positioning method for extreme weather rescue according to claim 6, wherein is a Lagrange multiplier.

8. The step of supporting GNSS / IMU tight combination calculation based on the real-time dynamically updated multi-path heat map described in step 3 specifically includes: Step 3-1: Under the abnormal weather rescue condition, extract the input features and the position accuracy factor PDOP value from the original observation of the GNSS receiver, and obtain the initial position based on the IMU mechanical configuration; Step 3-2: Calling the real-time dynamically updated multi-path heat map, matching the initial position to the grid layer to which it belongs, performing position calculation, and obtaining the specific position of the carrier in the abnormal weather rescue state. The combination navigation positioning method for abnormal weather rescue, as described in claim 7, characterized in that

9. The specific method for calculating the position described in step 3-2 is as follows: If the position accuracy factor PDOP value is equal to or greater than the threshold, no correction is performed, and the GNSS / IMU tight combination Kalman filtering is directly performed to calculate the position solution, and the position solution is set as the final solution; 9. The combination navigation and positioning method for extreme weather rescue according to claim 8, wherein, when the PDOP value is less than the threshold, the grid prediction model is used to predict the pseudorange error, and the incremental grid real-time correction model established in step 2 is used to make the next correction.

10. The step of predicting the pseudorange error using the grid prediction model described in step 3-2 specifically includes: According to the grid prediction model, the multipath error values ​​obtained by the adjacent grids are extracted, and the weights W are assigned to the adjacent grids. ｎ After the dynamic weighting process, the final multipath error prediction is [Number 79] where n denotes the grid prediction model number of the adjacent grid, m denotes the number of grid prediction models of the adjacent grid, Multipath error estimates are used to estimate the adaptive factor [Number 80] and construct the GNSS / IMU tight combination measurement noise matrix R ｋ The specific adjustments are as follows: [Number 81] Measurement noise matrix R ｋ 10. The combination navigation positioning method for extreme weather rescue according to claim 9, wherein after adjusting, the carrier position is obtained according to Kalman filtering calculation.

Images