PCA (Principal Component Analysis) and RF (Radio Frequency) combined underwater gravity matching aided navigation adaptation area selection method and system

Abstract

The invention belongs to the field of underwater navigation and surveying and mapping, and particularly discloses a PCA (Principal Component Analysis) and RF (Radio Frequency) combined underwater gravity matching aided navigation adaptation area selection method and system, and the system comprises a data processing unit which is used for receiving underwater gravity characteristic data acquired by an underwater vehicle and executing principal component analysis processing to obtain a data processing result; extracting main features and compressing data; the random forest classifier is used for analyzing the processed data and identifying an adaptive area; and the evaluation module is used for determining a navigation adaptive area and a non-adaptive area of each to-be-tested sample. Compared with the prior art, the invention provides a strategy fusing principal component analysis and a random forest algorithm, and feature extraction and dimension reduction are performed on original data through principal component analysis so as to reduce the burden of subsequent model training; and on the basis of data features after dimension reduction, a random forest algorithm is used for evaluating the suitability of a research area, and an adaptive area and a non-adaptive area are accurately distinguished, so that the overall performance of gravity matching navigation is optimized.

Description

Technical Field

[0001] The present invention relates to the fields of underwater navigation and surveying, and in particular to a method and system for selecting an adaptation zone for combined PCA and RF underwater gravity matching assisted navigation. Background Art

[0002] In underwater environments, inertial navigation systems are affected by cumulative errors and require auxiliary positioning compensation. Gravity matching is an effective compensation method, and selecting a reasonable adaptation area is crucial. With the advancement of technology and the expansion of application scenarios, the demand for adaptation area selection and optimization methods is increasing. Therefore, establishing a more scientific and efficient adaptation area automatic selection method is crucial to improving the overall performance of gravity matching assisted inertial navigation.

[0003] Chinese patent application number CN202111143171.6 discloses a method for optimizing underwater gravity adaptation areas based on a support vector machine based on a convolution slope eigenvalue, including: using a Sobel operator to perform a convolution operation on a gravity anomaly map to obtain a convolution feature map; performing a single-line slope solution based on the convolution feature map and the gravity anomaly map to determine the convolution slope parameters; determining the difference between convolution rows and columns and the convolution variance of the convolution feature map, and the pooling difference and range of the gravity anomaly map; standardizing the convolution slope parameters, the difference between convolution rows and columns, and the pooling difference and range of the convolution variance to obtain a standardized processing result; based on the standardized processing result, constructing a classification discriminant function based on a support vector machine algorithm; based on the classification discriminant function, discriminating the area to be identified, and determining the gravity adaptation area and non-adaptation area of ​​the area to be identified. This method uses convolution slope eigenvalues ​​and support vector machines to select underwater gravity adaptation areas. Although it can achieve high accuracy, the use of the Sobel operator for convolution operations and subsequent slope solution steps increases the complexity of the calculation, especially when dealing with large-scale data sets; at the same time, the performance of the support vector machine depends to a large extent on the selection of parameters, such as regularization parameters, kernel functions and their parameters; inappropriate parameter selection may cause the model to overfit or underfit, affecting the accuracy of classification. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a method and system for selecting an adaptation area for combined PCA and RF underwater gravity matching assisted navigation.

[0005] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0006] One of the purposes of the present invention is to provide a method for selecting a combined PCA and RF underwater gravity matching assisted navigation adaptation area, comprising the following steps:

[0007] S1. Acquire underwater gravity characteristic data through an underwater vehicle;

[0008] S2. Standardize the gravity characteristic data;

[0009] S3, the gravity characteristic data matrix Z after standardization is processed = [z1, z2, ···, z M ] Construct the sample correlation coefficient matrix R, calculate the p eigenvectors of the sample correlation coefficient matrix R, and calculate the principal component matrix F from the eigenvectors;

[0010] S4. Calculate the variance contribution rate of each principal component and cumulative variance contribution If the cumulative variance contribution rate a(N) of the current N principal components reaches the preset threshold, it is considered that the selected principal component can better obtain the original information;

[0011] S5, take the first N principal components to construct a new data set Y = [F1, F2, ..., F N ], use the Bootstrap method to resample the data set Y and randomly generate K training sets, denoted as

[0012] S6. Use the generated K training sets to construct multiple independent classification and regression trees CART to form a random forest;

[0013] S7. Use the generated random forest to perform classification tests on new or unknown samples. Each decision tree gives a prediction result, and finally summarizes these results to the next step;

[0014] S8. Summarize the classification results of all decision trees, and determine the navigation adaptation area and non-adaptation area of ​​each sample to be tested by voting method.

[0015] Furthermore, the gravity characteristic data in step S1 include: gravity anomaly standard deviation, roughness, correlation coefficient, kurtosis coefficient, skewness coefficient, slope standard deviation, difference constant entropy and gradient.

[0016] Furthermore, in step S2, the gravity characteristic data are standardized using a z-score normalization method, and the mean of each gravity characteristic data is converted to 0, and the standard deviation is converted to 1.

[0017] Furthermore, in the step S6, in the process of constructing multiple independent classification and regression trees CART, gravity feature data is randomly selected from the gravity feature data, and for each node, the best split point that maximizes the purity of the child node is calculated; the purity of the node is usually evaluated by the Gini index, and the calculation formula of the Gini index is: Among them, P j is the frequency of occurrence of category j in S.

[0018] The second object of the present invention is to provide a system for selecting an adaptation area for combined PCA and RF underwater gravity matching assisted navigation, comprising:

[0019] A data processing unit, for receiving underwater gravity characteristic data acquired by the underwater vehicle, and performing principal component analysis processing, extracting main features and compressing data;

[0020] a random forest classifier to analyze the processed data and identify the adaptation areas;

[0021] and an evaluation module for determining a navigation adaptation area and a non-adaptation area for each sample to be tested.

[0022] Compared with the prior art, the present invention proposes a strategy that integrates principal component analysis and random forest algorithm. Principal component analysis is used to extract features and reduce the dimension of the original data to reduce the burden of subsequent model training. Based on the data features after dimensionality reduction, the random forest algorithm is used to evaluate the adaptability of the study area, accurately distinguish between the adapted area and the non-adapted area, thereby optimizing the overall performance of gravity matching navigation. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 The present invention is a flowchart of the method for selecting the adaptation area of ​​the combined PCA and RF underwater gravity matching assisted navigation.

[0024] Figure 2 This is a block diagram of the system for selecting the adaptation zone of the combined PCA and RF underwater gravity matching assisted navigation of the present invention.

[0025] Figure 3 It is the track of Z1 area in the simulation example.

[0026] Figure 4 It is the track of Z2 area in the simulation example.

[0027] Figure 5 Predetermine the classification results in the simulation instance.

[0028] Figure 6 This is the classification result of the method for selecting the adaptation area of ​​the combined PCA and RF underwater gravity matching assisted navigation of the present invention. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments. The specific embodiments described herein are only used to explain the present invention and are not used to limit the invention.

[0030] Example 1

[0031] like Figure 1As shown, this embodiment exemplarily shows a method for selecting a joint PCA and RF underwater gravity matching assisted navigation adaptation area, including the following steps:

[0032] S1. Obtain underwater gravity characteristic data through underwater vehicles. In order to comprehensively analyze the gravity field data, the following gravity characteristic data are selected: gravity anomaly standard deviation, roughness, correlation coefficient, peak coefficient, skewness coefficient, slope standard deviation, difference constant entropy and gradient;

[0033] S2. Use the z-score regularization method to standardize the gravity characteristic data, convert the mean of each gravity characteristic data into 0, and convert the standard deviation into 1;

[0034] S3, the gravity characteristic data matrix Z after standardization is processed = [z1, z2, ···, z M ] Construct the sample correlation coefficient matrix R, calculate the p eigenvectors of the sample correlation coefficient matrix R, and calculate the principal component matrix F from the eigenvectors;

[0035] S4. Calculate the variance contribution rate of each principal component and cumulative variance contribution If the cumulative variance contribution rate a(N) of the current N principal components reaches the preset threshold, it is considered that the selected principal component can better obtain the original information;

[0036] S5, take the first N principal components to construct a new data set Y = [F1, F2, ..., F N ], use the Bootstrap method to resample the data set Y and randomly generate K training sets, denoted as

[0037] S6. Use the generated K training sets to construct multiple independent classification and regression trees CART to form a random forest. In the process of constructing multiple independent classification and regression trees CART, randomly select gravity feature data from the gravity feature data, and for each node, calculate the best split point that maximizes the purity of the child node. The purity of the node is usually evaluated by the Gini index, and the calculation formula of the Gini index is: Among them, P j is the occurrence frequency of category j in S; the integrated learning method of multi-classification regression tree CART can significantly improve the accuracy and stability of prediction.

[0038] S7. Use the generated random forest to perform classification tests on new or unknown samples. Each decision tree gives a prediction result, and finally summarizes these results to the next step;

[0039] S8. Summarize the classification results of all decision trees and determine the navigation adaptation area and non-adaptation area of ​​each sample to be tested by voting. This method determines the classification results by majority vote, which enhances the bias and uncertainty that may be brought by a single model.

[0040] Example 2

[0041] like Figure 2 As shown, this embodiment provides a system for selecting an adaptation area for combined PCA and RF underwater gravity matching assisted navigation, including:

[0042] A data processing unit 1, for receiving underwater gravity characteristic data acquired by the underwater vehicle, and performing principal component analysis processing, extracting main features and compressing data;

[0043] a random forest classifier 2 to analyze the processed data and identify the adaptation area;

[0044] and an evaluation module 3 for determining the navigation adaptation area and the non-adaptation area of ​​each sample to be tested.

[0045] Simulation Example

[0046] In order to verify the feasibility of the present invention, this embodiment selected the gravity anomaly data from the University of California, San Diego (http: / / topex.ucsd.edu) for experiment. The resolution of the experimental data is 1'×1', and the longitude and latitude range of the data is 130° to 136° east longitude and 10° to 13° north latitude. In this area, the maximum value of the gravity anomaly is 121.90mGal, the minimum value is -57.60mGal, the average value is 7.66mGal, and the variance is 12.34mGmal. For analysis, we interpolated the original data, converted it into a grid resolution of 0.125'×0.125', and divided the area into 1024 sub-areas, each of which contains 90×90 grids.

[0047] Since the PCA-RF combination method needs to solve the classification model based on the total characteristic parameters and categories of the samples, we need to calibrate the adaptation area in advance. The parameters of the calibration experiment are as follows: the constant drift of the gyroscope is set to 0.05° / h, and the constant error of the accelerometer is set to 0.05m / s 2 , the simulation speed is set to 10m / s, the initial position error is set to 500m, the number of sampling points is 75, and the sampling period is 10 seconds. The random error of the sampling value with a standard deviation of 1mGal is used as the real-time measurement data of the gravimeter, and the Tercom algorithm is used for trajectory matching. A total of 4 tracks are set in the experiment, and the starting point and the end point are represented by a five-pointed star and a circle respectively. The four tracks in the Z1 and Z2 areas are as follows Figure 3 and Figure 4 As shown in the figure, the calibration experiment results show that the average positioning error of 1000 times on four routes is higher than The area is marked as the adaptation area, represented by a yellow rectangular frame, and the rest of the area is marked as the non-adaptation area, such as Figure 4 As shown:

[0048] The 1024 sub-region samples were randomly divided into a training set (820 samples) and a test set (204 samples). The random forest used 100 decision trees, the minimum number of samples for a leaf node was 5, and the experimental environment was MATLAB R2022a. Figure 5 The results of the PCA-RF adaptation zone division are shown. In the test set, the classification accuracy of PCA-RF is 94.12%, the recall rate of the non-adaptation zone is 96.97%, and the recall rate of the adaptation zone is 82.05%. In all samples, the classification accuracy is 96%.

[0049] The above trained classification model was applied to other waters in the South China Sea, with a range of 127°E to 130°E east longitude and 16°N to 19°N north latitude. The minimum gravity anomaly in this area is -21.07mGal, the minimum is 56.55mGal, the average is 13.88mGal, and the variance is 10.30mGal. After linear interpolation, the spatial resolution of the image is 0.125'×0.125', and the area is divided into 256 regions. The classification results of PCA-RF are shown in Figure 2. Figure 6 As shown, the area indicated by the rectangular box is the adaptation area. Figure 6 It can be seen that the suitable area is mainly concentrated in the southwest of the region.

[0050] To further verify Figure 6 The adaptability of the area shown in the figure is shown in the figure. Four directions are randomly selected in all the adaptation areas for navigation 1000 times, and the average positioning error and accuracy (defined as the positioning error less than the length of the grid diagonal in 1000 experiments) are recorded. The PCA-RF method can maintain the average positioning error (274.11m) and standard deviation (36.56m) in the adaptation area. The accuracy is 88.8%. The results show that the method adopted in this paper has good applicability and can effectively judge the applicability of unknown areas, which is convenient for subsequent path planning.

[0051] The technical solution of the present invention is not limited to the above-mentioned specific embodiments. All technical variations made according to the technical solution of the present invention fall within the protection scope of the present invention.

Claims

1. A method for selecting an adaptation area for combined PCA and RF underwater gravity matching assisted navigation, characterized in that: The following steps are involved: S1. Acquire underwater gravity characteristic data through an underwater vehicle; S2. Standardize the gravity characteristic data; S3, constructing a sample correlation coefficient matrix R for the gravity characteristic data matrix Z=[z1,z2,···,zM] after the standardized processing, and calculating p eigenvectors of the sample correlation coefficient matrix R, and calculating the principal component matrix F from the eigenvectors; S4. Calculate the variance contribution rate of each principal component and cumulative variance contribution If the cumulative variance contribution rate a(N) of the current N principal components reaches the preset threshold, it is considered that the selected principal component can better obtain the original information; S5. Take the first N principal components to construct a new data set Y = [F1, F2, ..., FN], use the Bootstrap method to resample in the data set Y, and randomly generate K training sets, denoted as S6. Use the generated K training sets to construct multiple independent classification and regression trees CART to form a random forest; S7. Use the generated random forest to perform classification tests on new or unknown samples. Each decision tree gives a prediction result, and finally summarizes these results to the next step; S8. Summarize the classification results of all decision trees, and determine the navigation adaptation area and non-adaptation area of ​​each sample to be tested by voting method.

2. The method for selecting the adaptation area of ​​the combined PCA and RF underwater gravity matching assisted navigation according to claim 1 is characterized in that: The gravity characteristic data in step S1 include: gravity anomaly standard deviation, roughness, correlation coefficient, kurtosis coefficient, skewness coefficient, slope standard deviation, difference constant entropy and gradient.

3. The method for selecting the adaptation area of ​​the combined PCA and RF underwater gravity matching assisted navigation according to claim 1 is characterized in that: In step S2, the gravity characteristic data are standardized using the z-score normalization method, and the mean of each gravity characteristic data is converted to 0, and the standard deviation is converted to 1.

4. The method for selecting the adaptation area of ​​the combined PCA and RF underwater gravity matching assisted navigation according to claim 1 is characterized in that: In the step S6, during the construction of multiple independent classification and regression trees CART, gravity feature data is randomly selected from the gravity feature data, and for each node, the best split point that maximizes the purity of the child node is calculated; the purity of the node is usually evaluated by the Gini index, and the calculation formula of the Gini index is: Among them, P j is the occurrence frequency of category j in S.

5. A system for selecting an adaptation zone for combined PCA and RF underwater gravity matching assisted navigation, characterized in that: include: A data processing unit, for receiving underwater gravity characteristic data acquired by the underwater vehicle, and performing principal component analysis processing, extracting main features and compressing data; a random forest classifier to analyze the processed data and identify the adaptation areas; and an evaluation module for determining a navigation adaptation area and a non-adaptation area for each sample to be tested.

Images