A positioning and pose online learning method and system based on big data information mining

Abstract

The application discloses a positioning and pose online learning method and system based on big data information mining, and the technical key points are as follows: the method comprises the following steps: acquiring online data; constructing multi-dimensional features according to the online data, using a method combining density-based clustering and principal component analysis for big data information mining, and screening out data with good observation quality and stable observation noise; using the screened data to combine decision tree and neural network algorithms for further parameter optimization, updating model parameters in the positioning and pose algorithm, and obtaining more accurate results; combining the obtained results with user needs, and outputting or performing secondary calculation. The application can meet the positioning accuracy needs of daily travel; but for more stringent positioning requirements, the application can also meet the online customization of parameters for the universal model, so that the error is further reduced, and the positioning and orientation accuracy is improved.

Description

Technical Field

[0001] This invention relates to the fields of big data mining and navigation positioning, specifically to an online learning method and system for positioning and attitude determination based on big data information mining. Background Technology

[0002] Data mining is the process of searching for hidden information within large amounts of data using algorithms, and it is currently widely used in various data processing applications. Achieving precise positioning and attitude determination requires both high-quality input data and highly adaptable positioning and attitude determination processing parameters. Currently, domestic technologies for precise positioning have seen development in multiple directions, including the refinement of data acquisition instruments, the accuracy of raw data storage, and the comprehensiveness of positioning error correction. However, there are still gaps in big data information mining of large amounts of multi-epoch raw data, and in online learning of positioning and attitude determination processing parameters. Improving data acquisition instruments requires extensive material adaptation and refinement. While the accuracy of raw data storage currently meets the needs of precise positioning and attitude determination, its impact on improving the accuracy of the final positioning result is relatively small. Existing models for known positioning errors have achieved significant error correction, meeting the positioning accuracy requirements for daily travel. However, for positioning requirements under more stringent conditions, online customized parameter correction of general models is needed to further reduce errors and improve positioning and orientation accuracy. Summary of the Invention

[0003] This invention addresses the technical problems existing in the prior art by proposing an online learning method and system for positioning and orientation determination based on big data information mining.

[0004] According to a first aspect of the present invention, a method for online learning of localization and orientation based on big data information mining is provided, the method comprising the following steps:

[0005] Obtain online data;

[0006] Multidimensional features are constructed based on online data, and big data information is mined using a combination of density-based clustering and principal component analysis to select data with good observation quality and stable observation noise.

[0007] The filtered data is used in conjunction with decision trees and neural network algorithms to further optimize the parameters and update the model parameters in the localization and orientation algorithm to obtain more accurate results.

[0008] The results are then combined with user needs for output or secondary calculation.

[0009] Based on the above technical solution, the present invention can also be improved as follows.

[0010] Preferably, acquiring online data includes acquiring online data through a network connection module, wherein the data includes satellite navigation data or inertial navigation data.

[0011] Preferably, the step of constructing multidimensional features based on online data and using a combination of density-based clustering and principal component analysis for big data information mining includes:

[0012] For massive and constantly updated epochal observation data, a combination of density-based clustering and principal component analysis is used to cluster similar data along the feature dimension and remove epochal data in the database that deviates from the cluster center by more than a threshold. For diverse raw data inputs, principal component analysis is used to reduce the dimensionality of the raw data, and an algorithm combining decision trees and neural networks is used to learn the positioning and orientation processing parameters.

[0013] Preferably, the method combining density-based clustering and principal component analysis includes the following steps: judging the integrity of online data; if incomplete, discarding the data; if complete, using the OPTICS clustering algorithm and outputting the cluster ranking; judging whether the distance between the cluster ranking and the cluster center is less than a set threshold; if so, discarding the data; if not, performing principal component analysis.

[0014] Preferably, before performing principal component analysis, the method further includes: determining whether all dimensions of the original data have been evaluated; if so, the data is retained; otherwise, the method returns to the OPTICS clustering algorithm.

[0015] Preferably, the step of further parameter optimization using the filtered data combined with decision tree and neural network algorithms includes:

[0016] The decision tree divides the data through hierarchical attribute judgments; for a given neural network algorithm type, the positioning and attitude determination processing parameters to be updated are used as neural network inputs, and the processing results are used as the latest processing parameters to participate in the positioning and attitude determination calculation.

[0017] Preferably, the decision tree divides the data through hierarchical attribute judgment, including: for the processing parameters in positioning and orientation, the decision tree selects multiple attributes such as the initial confidence of the parameters, the confidence of the result feedback, and the neural network parameters, and makes a decision after inputting new observation data to determine the type of neural network algorithm to be selected.

[0018] Preferably, the further parameter optimization using the filtered data combined with decision trees and neural network algorithms also includes:

[0019] Based on the feedback from the positioning and attitude determination results calculated by the neural network algorithm, the decision coefficients of the decision tree are adjusted; using the neural network algorithm of decision tree decision, one of the GA-BP neural network algorithm and the GA-Elman neural network algorithm is selected for parameter learning.

[0020] According to a second aspect of the present invention, a localization and orientation online learning system based on big data information mining is proposed, comprising a data input and storage module, a data analysis and mining module, a parameter learning module, and a data service module; wherein,

[0021] The data input and storage module is used to acquire online data;

[0022] The data analysis and mining module is used to construct multidimensional features based on input data, and combine density-based clustering and principal component analysis methods to mine a large amount of data and screen out data with good observation quality and stable observation noise.

[0023] The parameter learning module is used to combine the data mined by information mining with decision trees and neural network algorithms to further optimize the parameters, update the model parameters in the localization and orientation algorithm, and obtain more accurate results.

[0024] The data service module is used to combine the obtained results with user needs and output or perform secondary calculations.

[0025] The advantages of this invention are:

[0026] 1. A method combining density-based clustering and principal component analysis is used for big data information mining to improve the stability and integrity of the data.

[0027] 2. An algorithm combining decision trees and neural networks is used to learn and update the localization and attitude determination processing parameters online, thereby obtaining the most reasonable correction results of the current processing parameters in a timely manner and applying them to the localization and attitude determination algorithm to improve the accuracy of localization and attitude determination.

[0028] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0029] Figure 1 This is a flowchart of the online learning method for localization and orientation determination based on big data information mining according to the present invention.

[0030] Figure 2 This is a flowchart of the big data information mining process of the present invention;

[0031] Figure 3This is a flowchart illustrating the online learning process for the processing parameters of this invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] This invention provides an online learning method for localization and orientation determination based on big data information mining, such as... Figure 1 As shown, the method includes the following steps:

[0034] Obtain online data;

[0035] Multidimensional features are constructed based on online data, and big data information is mined using a combination of density-based clustering and principal component analysis to select data with good observation quality and stable observation noise.

[0036] By combining the data mined from information with decision trees and neural network algorithms to optimize parameters and update the model parameters in the localization and orientation algorithm, more accurate results can be obtained.

[0037] The accurate results obtained are then combined with user needs for output or secondary calculation.

[0038] It should be noted that this invention employs a combination of density-based clustering and principal component analysis for big data information mining, improving data stability and integrity. It also uses an algorithm combining decision trees and neural networks for online learning and updating of positioning and orientation processing parameters, thereby obtaining the most reasonable parameter correction results in a timely manner and applying them to the positioning and orientation algorithm to improve positioning and orientation accuracy. Furthermore, for positioning requirements under more stringent conditions, it performs online customized parameter corrections for general models, further reducing errors and improving positioning and orientation accuracy.

[0039] Specifically, acquiring online data includes acquiring online data through a network connection module. The data mainly includes satellite navigation data or inertial navigation data, specifically including one or more raw data from inertial navigation three-axis gyroscopes, accelerometers, satellite navigation ephemeris, carrier phase observations, and pseudorange observations.

[0040] Specifically, the process of constructing multidimensional features based on online data and using a combination of density-based clustering and principal component analysis for big data information mining to select data with good observation quality and stable observation noise includes:

[0041] For massive and constantly updated epochal observation data, a combination of density-based clustering and principal component analysis is used to cluster similar data along the feature dimension and remove epochal data in the database that deviates from the cluster center by more than a threshold. For diverse raw data inputs, principal component analysis is used to reduce the dimensionality of the raw data, and an algorithm combining decision trees and neural networks is used to learn the positioning and orientation processing parameters.

[0042] Methods combining density-based clustering and principal component analysis, such as... Figure 2 As shown. The method includes the following steps: acquiring raw data, judging the integrity of the raw data, discarding the data if it is incomplete; if it is complete, using the OPTICS clustering algorithm and outputting the cluster ranking; judging whether the distance between the cluster ranking and the cluster center is less than a set threshold, discarding the data if it is, and performing principal component analysis if it is not.

[0043] Before principal component analysis, it is determined whether all dimensions of the original data have been evaluated. If so, the data is retained; otherwise, it is returned to the OPTICS clustering algorithm.

[0044] It should be noted that, to overcome the subjectivity of global parameter selection, the OPTICS clustering algorithm is employed. This algorithm overcomes the drawbacks of using global parameters and is a density-based clustering algorithm based on parameter order. The OPTICS algorithm outputs a cluster order, i.e., a linear list of output data, representing the density order of the data's cluster structure. The cluster structure corresponding to this order contains cluster information at each level; therefore, cluster information similar to cluster centers or cluster structures of arbitrary shapes can be extracted based on the cluster order. When new observation data is input, it can be filtered and cleaned based on its distance from the cluster centers and a set threshold, continuously enabling information mining of big data.

[0045] Since the raw data includes various sources such as inertial navigation triaxial gyroscopes, accelerometers, satellite ephemeris data, carrier phase observations, and pseudorange observations, principal component analysis is required to process the filtered results. Principal component analysis reduces the data dimensionality and minimizes the loss of data information.

[0046] Specifically, the step of further optimizing parameters by combining the filtered data with decision trees and neural network algorithms to update the model parameters in the localization and orientation algorithm and obtain more accurate results includes:

[0047] Decision trees partition data through hierarchical attribute judgments. For a given neural network algorithm type, the localization and attitude determination processing parameters to be updated are used as input to the neural network. After processing by the neural network, the results are used as the latest processing parameters in the localization and attitude determination calculation. For the processing parameters in localization and attitude determination, the decision tree selects multiple attributes such as initial parameter confidence, result feedback confidence, and neural network parameters. After inputting new observation data, it makes a decision to determine the type of neural network algorithm to be selected. Based on the localization and attitude determination results calculated by the neural network algorithm, the decision tree's decision coefficients are adjusted. Using the neural network algorithm determined by the decision tree, one of the GA-BP neural network algorithm and the GA-Elman neural network algorithm is selected for parameter learning.

[0048] Specifically, online learning methods for processing parameters based on a combination of decision trees and neural networks, such as... Figure 3 As shown, the decision tree algorithm is a predictive analysis model expressed in the form of a tree structure, mainly composed of nodes and directed edges. Each decision point implements a test function with discrete output, denoted as a branch. There are two types of nodes: internal nodes and leaf nodes; internal nodes represent a feature or attribute, while leaf nodes represent a class.

[0049] Decision trees divide samples through hierarchical attribute judgments. For the processing parameters in localization and pose determination, the decision tree selects multiple attributes such as initial parameter confidence, result feedback confidence, and neural network parameters. After inputting new observation data, it makes a decision to determine the type of neural network algorithm to be selected.

[0050] For a given neural network algorithm type, the localization and pose determination parameters to be updated are used as input to the neural network. After processing by the neural network, the processing result is used as the latest processing parameters for localization and pose determination. The neural network can be either a GA-BP network or a GA-Elman network. These are high-accuracy networks obtained by combining a genetic (GA) algorithm with a BP neural network and an Elman neural network, respectively. A BP neural network is a feedforward neural network that exhibits excellent nonlinear mapping capabilities, self-learning and adaptability, generalization ability, and fault tolerance in various applications. An Elman neural network is a dynamic recurrent neural network that adapts to time-varying characteristics, enhancing the network's global stability and comprehensive computational power. The genetic algorithm is an adaptive global optimization search algorithm that uses a fitness function for regulation during the search process. In the optimization calculation, the fitness function measures the probability of each individual in the population reaching the optimal solution, thereby adaptively controlling the search process and obtaining the optimal solution. Therefore, both neural network algorithms can utilize optimal network weights and thresholds to overcome the tendency to get trapped in local minima and significantly improve the model's accuracy. In different scenarios, the accuracy of the two neural network algorithms will differ to some extent, while the decision tree can autonomously determine a better solution to maximize the accuracy of positioning and orientation.

[0051] Finally, the results are combined with user needs to output or perform secondary calculations, providing users with personalized result data.

[0052] According to another aspect of the present invention, an online learning system for positioning and orientation determination based on big data information mining is proposed, comprising a data input and storage module, a data analysis and mining module, a parameter learning module, and a data service module; wherein,

[0053] The data input and storage module is used to acquire online data in blocks;

[0054] The data analysis and mining module is used to construct multidimensional features based on input data, and combine density-based clustering and principal component analysis methods to mine a large amount of data and screen out data with good observation quality and stable observation noise.

[0055] The parameter learning module is used to combine the data mined by information mining with decision trees and neural network algorithms to further optimize the parameters, update the model parameters in the localization and orientation algorithm, and obtain more accurate results.

[0056] The data service module is used to combine the obtained results with user needs, output or perform secondary calculations, and provide personalized result data requirements.

[0057] It is understood that the positioning and attitude determination online learning system based on big data information mining provided by this invention corresponds to the positioning and attitude determination online learning method based on big data information mining provided in the foregoing embodiments. The relevant technical features of the positioning and attitude determination online learning system based on big data information mining can be referred to the relevant technical features of the positioning and attitude determination online learning method based on big data information mining, and will not be repeated here.

[0058] In summary, this invention improves the stability and integrity of big data information by employing a combination of density-based clustering and principal component analysis. Simultaneously, it uses an algorithm combining decision trees and neural networks to learn and update positioning and orientation processing parameters online, thereby obtaining the most reasonable parameter correction results in a timely manner and applying them to the positioning and orientation algorithm to improve positioning and orientation accuracy.

[0059] The scope of protection of this invention is not limited to the examples described above. Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its scope and spirit. If such modifications and variations fall within the scope of the claims of this invention and their equivalents, then the intent of this invention also includes these modifications and variations.

Claims

1.A positioning and orientation online learning method based on big data information mining, the method comprising the following steps: acquiring online data; constructing multi-dimensional features according to the online data, performing big data information mining by a method combining density-based clustering and principal component analysis, and screening data with good observation quality and stable observation noise; including: for a large and constantly updated epoch observation data, a method combining density-based clustering and principal component analysis is used to cluster the same data in feature dimensions, and epoch data deviating from the cluster center by more than a threshold value in the database is removed; for diversified original data input, principal component analysis is used to reduce the dimension of the original data, and a decision tree combined with a neural network algorithm is used to learn the positioning and orientation processing parameters; the method combining density-based clustering and principal component analysis includes: judging the integrity of the online data, if not complete, discarding the data; if complete, performing an OPTICS clustering algorithm and outputting cluster sorting; judging whether the distance between the cluster sorting and the cluster center is less than the set threshold value, if yes, discarding the data; if not, performing principal component analysis; further parameter optimization is performed by using the screened data combined with the decision tree and neural network algorithm, including: the decision tree divides the data by hierarchical progressive attribute judgment; for a determined neural network algorithm type, each positioning and orientation processing parameter to be updated is taken as the neural network input, and after the neural network processing, the processing result is taken as the latest processing parameter to participate in the positioning and orientation solution, the model parameters in the positioning and orientation algorithm are updated, and more accurate results are obtained; the obtained results are combined with the needs of users, and output or secondary calculation is performed. 2.The positioning and pose orientation online learning method based on big data information mining of claim 1, wherein: The online data is acquired through a network connection module; wherein the data includes satellite navigation data or inertial navigation data. 3.The positioning and pose orientation online learning method based on big data information mining of claim 1, wherein: Before the principal component analysis, it further includes: judging whether all dimensions of the original data have been judged, if yes, retaining the data, if not, returning to the OPTICS clustering algorithm. 4.The positioning and mapping online learning method based on big data information mining of claim 1, wherein: The decision tree divides the data by hierarchical progressive attribute judgment, including: For the processing parameters in positioning and orientation, the decision tree selects multiple attributes such as initial parameter reliability, result feedback reliability, and neural network parameters, and makes decisions after inputting new observation data to determine the type of neural network algorithm to be selected. 5.The positioning and mapping online learning method based on big data information mining of claim 1, wherein: The further parameter optimization by using the screened data combined with the decision tree and neural network algorithm further includes: According to the positioning and orientation result feedback calculated by the neural network algorithm, the decision tree decision coefficient is adjusted, the neural network algorithm determined by the decision tree is used, and one of the GA-BP neural network algorithm and the GA-Elman neural network algorithm is selected for processing parameter learning. 6.A positioning and pose orientation online learning system based on big data information mining, characterized in that, including a data input and storage module, a data analysis and mining module, a parameter learning module, and a data service module; wherein, the data input and storage module is used to acquire online data; The data analysis mining module is used for constructing multi-dimensional features according to input data, mining a large amount of data by combining a density-based clustering method with a principal component analysis method, and screening out data with good observation quality and stable observation noise; the data analysis mining module comprises: for a large and constantly updated epoch observation data, a method combining a density-based clustering method with a principal component analysis method is used to cluster the same type of data in the feature dimension, and epoch data deviating from the clustering center by more than a threshold value in the database is removed; for diversified original data input, a principal component analysis method is used to reduce the dimension of the original data, and a decision tree and neural network combined algorithm is used to learn positioning and pose processing parameters; The method combining the density-based clustering method with the principal component analysis method comprises: judging the integrity of online data, discarding the data if the data is not complete, and if the data is complete, performing an OPTICS clustering algorithm and outputting cluster sorting; judging whether the distance between the cluster sorting and the clustering center is less than a set threshold value, discarding the data if the distance is less than the threshold value, and if the distance is not less than the threshold value, performing principal component analysis; The parameter learning module is used for further parameter optimization of the data after information mining by combining a decision tree and a neural network algorithm, and the parameter learning module comprises: a decision tree divides data by layer-by-layer attribute judgment; for a determined neural network algorithm type, each positioning and pose processing parameter to be updated is taken as a neural network input, and after neural network processing, the processing result is taken as the latest processing parameter to participate in positioning and pose solution, model parameters in the positioning and pose algorithm are updated, and more accurate results are obtained; The data service module is used for outputting or secondary calculation of each result obtained in combination with user needs.

Images