Beidou positioning information processing method and system based on artificial intelligence

Through the Beidou positioning abnormal diagnosis method based on artificial intelligence, the segmentation and optimization technology of training data are used to solve the false alarm and missed response problems of Beidou positioning system abnormal diagnosis in the existing technology, the accuracy and efficiency of diagnosis are improved, and the System stability.

CN119089354AInactive Publication Date: 2024-12-06SHAOSHAN BANGSHENG BEIDOU TECHNOLOGY CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411187627.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-28
Publication Date
2024-12-06
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing Beidou positioning abnormal diagnosis methods rely on preset rules and thresholds, resulting in high false alarm rates and missed alarm rates in complex situations, and it is impossible to effectively identify and diagnose abnormalities in Beidou positioning system.

Method used

Using an artificial intelligence-based method, the Beidou positioning event sequence is obtained with different abnormal defect labels, segment and extract training data, generate target sample learning data sequences, and optimize the Beidou positioning abnormal diagnosis network, and generate the optimized network for Abnormal diagnosis.

Benefits of technology

It significantly improves the accuracy and efficiency of Beidou positioning abnormal diagnosis, improves the stability and reliability of the system, and reduces false alarms and missed response rates.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119089354A_ABST
    Figure CN119089354A_ABST
Patent Text Reader

Abstract

According to the Beidou positioning information processing method and system based on artificial intelligence provided by the invention, the event scales of the first and second training Beidou positioning event sequences are compared, and the extraction number of the subsequences is determined according to the event scales, so that the problem of data imbalance is effectively solved, training data are more representative, and the generalization ability of a model is improved. Through reasonable segmentation and extraction of the training data, the target sample learning data sequence is formed, and then parameter optimization is performed on the Beidou positioning anomaly diagnosis network, so that the model training efficiency is improved, and the anomaly diagnosis accuracy is greatly improved. By using the optimized Beidou positioning abnormity diagnosis network, the candidate Beidou positioning events can be subjected to fast and accurate abnormity diagnosis, the generated abnormity defect diagnosis data can effectively represent the abnormity defect labels of the candidate Beidou positioning events, and support is provided for subsequent troubleshooting and repair.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically, to a Beidou positioning information processing method and system based on artificial intelligence. Background Art

[0002] With the widespread application of the Beidou positioning system, the accuracy and stability of its positioning information are increasingly required. However, in the actual Beidou positioning process, due to various factors such as signal interference, equipment failure, etc., positioning events may be abnormal. In order to effectively identify and diagnose these anomalies and improve the reliability of the Beidou positioning system, the industry has been seeking more efficient and accurate anomaly diagnosis methods.

[0003] The traditional Beidou positioning anomaly diagnosis method mainly relies on preset rules and thresholds for judgment. Although this method is simple, it often seems powerless when faced with complex and changeable actual situations, and has high false alarm and missed alarm rates. Summary of the invention

[0004] In view of this, the purpose of this application is to provide a Beidou positioning information processing method and system based on artificial intelligence.

[0005] According to the first aspect of the present application, there is provided a Beidou positioning information processing method based on artificial intelligence, which is applied to a Beidou positioning platform system, and the method comprises: Acquire a first training Beidou positioning event sequence and a second training Beidou positioning event sequence, wherein the first training Beidou positioning event sequence and the second training Beidou positioning event sequence have different abnormal defect labels, and the event scale of the first training Beidou positioning event sequence is greater than the event scale of the second training Beidou positioning event sequence; Dividing the first training Beidou positioning event sequence into a plurality of training Beidou positioning event subsequences, and determining the extraction quantity of each of the training Beidou positioning event subsequences based on an event scale comparison parameter of the event scale of the first training Beidou positioning event sequence and the event scale of the second training Beidou positioning event sequence; Performing event extraction on each of the training Beidou positioning event subsequences based on the extraction quantity, outputting a third training Beidou positioning event sequence, and generating a target sample learning data sequence based on the third training Beidou positioning event sequence and the second training Beidou positioning event sequence, wherein an event scale comparison parameter of the event scale of the third training Beidou positioning event sequence and the event scale of the second training Beidou positioning event sequence is less than a set comparison parameter; Optimizing the parameters of the Beidou positioning anomaly diagnosis network based on the target sample learning data sequence to generate the optimized Beidou positioning anomaly diagnosis network; Based on the optimized Beidou positioning abnormality diagnosis network, abnormality diagnosis is performed on candidate Beidou positioning events to generate abnormal defect diagnosis data of the candidate Beidou positioning events, wherein the abnormal defect diagnosis data represents the abnormal defect label of the candidate Beidou positioning event.

[0006] In a possible implementation of the first aspect, acquiring a first training Beidou positioning event sequence and a second training Beidou positioning event sequence includes: Obtaining an initial positioning event carrying abnormal annotation data from a sample Beidou positioning event log, wherein the abnormal annotation data represents an abnormal defect label of the initial positioning event; Encoding the initial positioning event to generate a coding representation sequence of the initial positioning event; The first training Beidou positioning event sequence and the second training Beidou positioning event sequence are determined based on the coded representation sequence of the initial positioning event.

[0007] In a possible implementation of the first aspect, obtaining an initial positioning event carrying abnormal annotation data from a sample Beidou positioning event log includes: Acquiring positioning control path data of the initial positioning event; Determining screening knowledge points based on the positioning control path data of the initial positioning event; Based on the screening knowledge points and the screening strategy, the corresponding positioning event data is deleted from the initial positioning event to generate an updated initial positioning event.

[0008] In a possible implementation manner of the first aspect, encoding the initial positioning event to generate an encoding representation sequence of the initial positioning event includes: Constructing a positioning control vector of the initial positioning event according to a past positioning control process record of the initial positioning event, wherein the positioning control vector at least includes a positioning tag vector, a positioning time domain vector, and a positioning space domain vector; According to the positioning logic of the initial positioning event, construct a positioning application scenario vector of the initial positioning event; Performing time-sequential processing on the positioning control vector and the positioning application scenario vector of the initial positioning event according to a preset time window, so as to summarize the positioning control vectors and time-sequential positioning application scenario vectors of different time windows; Performing encoding conversion on at least one of the basic event features before time sequencing and the basic event features after time sequencing and a positioning application scenario vector to obtain first encoding data of the initial positioning event; Processing the persistence data in the first coded data of the initial positioning event to generate second coded data of the initial positioning event; The first coded data and the second coded data of the initial positioning event are merged to generate a coded representation sequence of the initial positioning event, wherein the coded representation sequence of the initial positioning event is a quantized coded representation sequence; In a possible implementation manner of the first aspect, encoding and converting at least one of the basic event features before time sequencing and the basic event features after time sequencing with a positioning application scenario vector to obtain first encoded data of the initial positioning event includes: Performing standardization rule conversion on at least one of the positioning control vector before timing and the positioning control vector after timing and the quantization vector in the positioning application scenario vector; and Performing bucket processing on at least one of the positioning control vector before timing and the positioning control vector after timing and a non-quantized vector in the positioning application scenario vector; In a possible implementation manner of the first aspect, the step of constructing the positioning control vector of the initial positioning event according to the past positioning control process record of the initial positioning event includes: Extracting key data from the past positioning control process records of the initial positioning event, wherein the key data includes positioning time, positioning location, positioning method, positioning accuracy, and positioning status; Clean the extracted key data, remove invalid, duplicate or erroneous data records, and generate cleaned key data; According to the positioning mode, a unique label value is assigned to each positioning mode, and the label value is used as a part of the positioning label vector; Generate event pattern features based on the generation sequence of positioning time as the positioning time domain vector; Extracting spatial features as the positioning airspace vector according to the positioning location, the spatial features including longitude and latitude, an area code of a geographic location, a distance from a specific location, and a moving speed and direction change of continuous positioning records; The positioning tag vector, the positioning time domain vector and the positioning space domain vector are merged to generate the positioning control vector, where the positioning control vector is used to describe the characteristics of the initial positioning event in time, space and mode; In a possible implementation manner of the first aspect, the step of constructing a positioning application scenario vector of the initial positioning event according to the positioning logic of the initial positioning event includes: Obtaining the positioning logic of the initial positioning event, the positioning logic including the conditions for triggering the positioning request, the purpose of the positioning data, and the key decision points in the positioning process; According to the positioning logic, different positioning application scenarios are identified; Describe each positioning application scenario in detail, and select key features related to the positioning application scenario, wherein the key features include user behavior data, application interface status, and device status; For each positioning application scenario, construct a positioning application scenario vector according to the extracted key features, wherein the positioning application scenario vector is used to reflect the key features in the positioning application scenario; The positioning application scenario vector is represented by binary encoding or one-hot encoding.

[0009] In a possible implementation of the first aspect, dividing the first training Beidou positioning event sequence into a plurality of training Beidou positioning event subsequences, and determining the extraction quantity of each of the training Beidou positioning event subsequences based on an event scale comparison parameter of an event scale of the first training Beidou positioning event sequence and an event scale of the second training Beidou positioning event sequence, comprises: When the event scale comparison parameter of the first training Beidou positioning event sequence and the second training Beidou positioning event sequence is greater than the set comparison parameter, the extraction quantity is determined based on the set comparison parameter, the event scale of the training Beidou positioning event subsequence and the event scale of the second training Beidou positioning event sequence.

[0010] In a possible implementation of the first aspect, the performing parameter optimization on the Beidou positioning abnormality diagnosis network based on the target sample learning data sequence to generate the optimized Beidou positioning abnormality diagnosis network includes: Selecting multiple feature vectors of each target sample learning data in the target sample learning data sequence based on a predefined selection strategy, loading them into a target deep learning network model, and outputting a weighted discriminant structure; Iteratively executing the step of selecting multiple feature vectors of each target sample learning data in the target sample learning data sequence based on a predefined selection strategy, loading them into the target deep learning network model, and outputting a weight discriminant structure until M weight discriminant structures are generated, where M is a positive integer greater than 1; The M weight discrimination structures are integrated to generate the Beidou positioning anomaly diagnosis network.

[0011] In a possible implementation of the first aspect, the performing parameter optimization on the Beidou positioning abnormality diagnosis network based on the target sample learning data sequence to generate the optimized Beidou positioning abnormality diagnosis network includes: Splitting the target sample learning data sequence into a learning subsequence and a verification subsequence; According to the initialization configuration data and the learning subsequence, the Beidou positioning anomaly diagnosis network is trained, wherein the initialization configuration data includes the scale of the training Beidou positioning event subsequence, M, and the scale of the feature vector of each target sample learning data in the target sample learning data sequence selected based on a predefined selection strategy; Performing parameter optimization search on the Beidou positioning anomaly diagnosis network; According to the network parameter information after the parameter optimization search, the parameters of the Beidou positioning abnormality diagnosis network are re-optimized to generate the optimized Beidou positioning abnormality diagnosis network, and the performance of the optimized Beidou positioning abnormality diagnosis network is tested based on the verification subsequence.

[0012] In a possible implementation of the first aspect, the performing abnormal diagnosis on a candidate Beidou positioning event based on the optimized Beidou positioning abnormality diagnosis network to generate abnormal defect diagnosis data of the candidate Beidou positioning event includes: Encoding the acquired candidate Beidou positioning event sequence to generate a coded representation sequence of the candidate Beidou positioning event; Loading the encoded representation sequence of the candidate Beidou positioning event into each of the weighted discriminant structures in the optimized Beidou positioning anomaly diagnosis network, so that each of the weighted discriminant structures generates discriminant data; The abnormality and defect diagnosis data is generated based on the discrimination data generated by each of the weight discrimination structures.

[0013] In a possible implementation of the first aspect, the discrimination data includes the confidence that the candidate Beidou positioning event corresponds to different abnormal defect labels, and the discrimination data generated based on each weighted discrimination structure generates the abnormal defect diagnosis data, including: Calculating the scale of the discrimination data whose confidence level of each abnormal defect label is greater than a set confidence level; The abnormal defect labels whose size is greater than half of the M are determined as the abnormal defect diagnosis data.

[0014] According to the second aspect of the present application, a Beidou positioning platform system is provided, which includes a processor and a readable storage medium, wherein the readable storage medium stores a program, which, when executed by the processor, implements the aforementioned Beidou positioning information processing method based on artificial intelligence.

[0015] According to the third aspect of the present application, a computer-readable storage medium is provided, in which computer-executable instructions are stored. When it is monitored that the computer-executable instructions are executed, the aforementioned Beidou positioning information processing method based on artificial intelligence is implemented.

[0016] According to any of the above aspects, the embodiment of the present application generates an optimized Beidou positioning abnormal diagnosis network by obtaining the first and second training Beidou positioning event sequences with different abnormal defect labels, and through a series of processing steps. Specifically, by comparing the event scales of the first and second training Beidou positioning event sequences, and determining the number of subsequences to be extracted accordingly, the data imbalance problem is effectively solved, making the training data more representative, thereby improving the generalization ability of the model. By reasonably segmenting and extracting the training data, a target sample learning data sequence is formed, and then the parameters of the Beidou positioning abnormal diagnosis network are optimized. This method not only improves the training efficiency of the model, but also greatly improves the accuracy of abnormal diagnosis. Using the optimized Beidou positioning abnormal diagnosis network, the candidate Beidou positioning event can be quickly and accurately diagnosed, and the generated abnormal defect diagnosis data can effectively characterize the abnormal defect label of the candidate Beidou positioning event, providing support for subsequent troubleshooting and repair. As a result, the accuracy and efficiency of Beidou positioning abnormal diagnosis are significantly improved, and the stability and reliability of the Beidou positioning system are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 A flowchart of a Beidou positioning information processing method based on artificial intelligence provided in an embodiment of the present application; Figure 2 A schematic diagram of the component structure of a Beidou positioning platform system provided in an embodiment of the present application for implementing the above-mentioned Beidou positioning information processing method based on artificial intelligence is shown. DETAILED DESCRIPTION

[0019] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present application.

[0020] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0021] Figure 1 The flowchart of the Beidou positioning information processing method based on artificial intelligence provided by the embodiment of the present application is shown. It should be understood that in other embodiments, the order of some steps of the Beidou positioning information processing method based on artificial intelligence can be interchanged according to actual needs, or some steps can be omitted or deleted. The detailed steps of the Beidou positioning information processing method based on artificial intelligence are described as follows.

[0022] Step S110, obtaining a first training Beidou positioning event sequence and a second training Beidou positioning event sequence, wherein the abnormal defect labels of the first training Beidou positioning event sequence and the second training Beidou positioning event sequence are different, and the event scale of the first training Beidou positioning event sequence is greater than the event scale of the second training Beidou positioning event sequence.

[0023] In this embodiment, the Beidou positioning platform system first accesses a database storing Beidou positioning event data. In this database, the Beidou positioning platform system finds two labeled Beidou positioning event sequence data sets, named "data set A" and "data set B" respectively. The Beidou positioning events in data set A are all labeled as "abnormal type 1", and the Beidou positioning events in data set B are all labeled as "abnormal type 2". According to statistics, data set A contains 100,000 Beidou positioning events, while data set B only has 50,000 Beidou positioning events. The Beidou positioning platform system uses data set A as the first training Beidou positioning event sequence and data set B as the second training Beidou positioning event sequence, and downloads them locally for subsequent processing.

[0024] Step S120: divide the first training Beidou positioning event sequence into multiple training Beidou positioning event subsequences, and determine the extraction quantity of each training Beidou positioning event subsequence based on the event scale comparison parameter of the event scale of the first training Beidou positioning event sequence and the event scale of the second training Beidou positioning event sequence.

[0025] In this embodiment, the Beidou positioning platform system then evenly divides the first training Beidou positioning event sequence (data set A) into 20 training Beidou positioning event subsequences, each of which contains 5,000 Beidou positioning events. Since the event scale of the first training Beidou positioning event sequence is twice that of the second training Beidou positioning event sequence, the Beidou positioning platform system decides to extract 2,500 Beidou positioning events from each training Beidou positioning event subsequence to ensure that the total number of extracted events is equivalent to that of the second training Beidou positioning event sequence, and the number of extracted events is determined based on the comparison parameters of the event scales of the two data sets.

[0026] Step S130, performing event extraction on each of the training Beidou positioning event subsequences based on the extraction quantity, outputting a third training Beidou positioning event sequence, and generating a target sample learning data sequence based on the third training Beidou positioning event sequence and the second training Beidou positioning event sequence, wherein an event scale comparison parameter of the event scale of the third training Beidou positioning event sequence and the event scale of the second training Beidou positioning event sequence is less than a set comparison parameter.

[0027] According to the extraction quantity determined in the previous step, the Beidou positioning platform system begins to extract events from each training Beidou positioning event subsequence. The process is random, but it ensures that 2,500 Beidou positioning events are extracted from each subsequence. After the extraction is completed, the Beidou positioning platform system merges these extracted events into a new sequence, named the third training Beidou positioning event sequence. The third training Beidou positioning event sequence contains 50,000 Beidou positioning events, which is comparable to the event scale of the second training Beidou positioning event sequence (dataset B). Subsequently, the Beidou positioning platform system combines the third training Beidou positioning event sequence with dataset B to form a target sample learning data sequence for subsequent network training.

[0028] Step S140, optimizing parameters of the Beidou positioning anomaly diagnosis network based on the target sample learning data sequence to generate the optimized Beidou positioning anomaly diagnosis network.

[0029] The BeiDou positioning platform system now has a target sample learning data sequence containing two different anomaly type labels. Next, the BeiDou positioning platform system inputs this target sample learning data sequence into a pre-built BeiDou positioning anomaly diagnosis network and begins to optimize the parameters of the BeiDou positioning anomaly diagnosis network. During this process, the BeiDou positioning anomaly diagnosis network will continuously learn and adjust its internal parameters to more accurately identify and classify two different anomaly types. After multiple rounds of training and verification, the performance of the BeiDou positioning anomaly diagnosis network gradually improved, and finally an optimized BeiDou positioning anomaly diagnosis network model was generated.

[0030] Step S150, performing abnormal diagnosis on the candidate Beidou positioning event based on the optimized Beidou positioning abnormal diagnosis network, generating abnormal defect diagnosis data of the candidate Beidou positioning event, wherein the abnormal defect diagnosis data represents the abnormal defect label of the candidate Beidou positioning event.

[0031] Once the Beidou positioning anomaly diagnosis network is optimized, the Beidou positioning platform system can start using the Beidou positioning anomaly diagnosis network to perform anomaly diagnosis on candidate Beidou positioning events. The Beidou positioning platform system receives a new candidate Beidou positioning event sequence from the real-time data stream. It first encodes and represents the candidate Beidou positioning event sequence to generate a coded representation sequence in the same format as the training data. Then, the Beidou positioning platform system inputs this coded representation sequence into the optimized Beidou positioning anomaly diagnosis network and outputs an abnormal defect diagnosis data, which indicates the abnormal defect label corresponding to each Beidou positioning event in the candidate Beidou positioning event sequence. The Beidou positioning platform system can take corresponding measures based on these labels, such as issuing an alarm or performing troubleshooting.

[0032] Based on the above steps, the embodiment of the present application generates an optimized Beidou positioning abnormal diagnosis network by obtaining the first and second training Beidou positioning event sequences with different abnormal defect labels, and through a series of processing steps. Specifically, by comparing the event scales of the first and second training Beidou positioning event sequences, and determining the number of subsequences to be extracted accordingly, the data imbalance problem is effectively solved, making the training data more representative, thereby improving the generalization ability of the model. By reasonably segmenting and extracting the training data, a target sample learning data sequence is formed, and then the parameters of the Beidou positioning abnormal diagnosis network are optimized. This method not only improves the training efficiency of the model, but also greatly improves the accuracy of abnormal diagnosis. Using the optimized Beidou positioning abnormal diagnosis network, the candidate Beidou positioning event can be quickly and accurately diagnosed, and the generated abnormal defect diagnosis data can effectively characterize the abnormal defect labels of the candidate Beidou positioning event, providing support for subsequent troubleshooting and repair. As a result, the accuracy and efficiency of Beidou positioning abnormal diagnosis are significantly improved, and the stability and reliability of the Beidou positioning system are improved.

[0033] In a possible implementation, step S110 includes: Step S111, obtaining an initial positioning event carrying abnormal annotation data from a sample Beidou positioning event log, wherein the abnormal annotation data represents an abnormal defect label of the initial positioning event.

[0034] In this embodiment, the Beidou positioning platform system first accesses the database storing sample Beidou positioning event logs, which record a large number of Beidou positioning events, some of which are marked as abnormal, and these abnormal marking data indicate the abnormal defect labels of the corresponding positioning events. For example, a Beidou positioning event may cause excessive positioning deviation due to signal interference, and this event will be marked as abnormal positioning accuracy.

[0035] The Beidou positioning platform system screens out these initial positioning events with abnormal annotations through query operations. These initial positioning events will become the basic data for subsequent processing and analysis.

[0036] Step S112: Encode the initial positioning event to generate a sequence of encoded representations of the initial positioning event.

[0037] In this embodiment, the Beidou positioning platform system will then encode the selected initial positioning events. The encoding process is to convert the various attributes of each initial positioning event (such as timestamp, longitude and latitude, signal strength, etc.) into a standardized digital format that can be easily used for mathematical operations and model processing.

[0038] For example, the Beidou positioning platform system can convert the timestamp into the number of seconds starting from a specific time point, convert the longitude and latitude into relative coordinate values ​​with a fixed point as the origin, and the signal strength may be directly converted into the corresponding numerical value. Through these conversions, each positioning event is represented as a numerical sequence, that is, its coded representation sequence.

[0039] Step S113: determining the first training Beidou positioning event sequence and the second training Beidou positioning event sequence based on the coded representation sequence of the initial positioning event.

[0040] After obtaining the coded representation sequences of all initial positioning events, the Beidou positioning platform system can determine two training sequences according to specific rules or algorithms: a first training Beidou positioning event sequence and a second training Beidou positioning event sequence.

[0041] Specifically, the Beidou positioning platform system can classify positioning events into two categories: normal and abnormal based on abnormal annotation data. Then, a certain number of events are extracted from each of these two categories and formed into two training sequences according to their time sequence or other relevant rules.

[0042] For example, the first training Beidou positioning event sequence may consist entirely of normal positioning events, which is used to train the model to recognize positioning patterns under normal conditions, while the second training Beidou positioning event sequence may contain positioning events of various abnormal conditions, which is used to train the model to recognize and predict abnormal conditions.

[0043] Through such processing, the Beidou positioning platform system prepares two training sequences, providing a data basis for the subsequent training of positioning algorithms or models.

[0044] In a possible implementation, step S111 includes: Step S1111, obtaining positioning control path data of the initial positioning event.

[0045] In this embodiment, the Beidou positioning platform system first extracts sample Beidou positioning event logs from the database, which contain a large number of initial positioning events. Each initial positioning event contains positioning control path data, which records various information in the positioning process in detail, such as the reception time, reception location, signal quality, etc. of the positioning signal.

[0046] For example, the Beidou positioning platform system can read a positioning control path data, showing that a certain positioning event was received through a specific Beidou satellite at a specific point in time, and also includes information such as the signal strength and error range at the time of reception.

[0047] Step S1112: determining screening knowledge points based on the positioning control path data of the initial positioning event.

[0048] Next, the Beidou positioning platform system can conduct in-depth analysis of these positioning control path data to determine screening knowledge points. Screening knowledge points are key indicators or features used to evaluate whether a positioning event is abnormal. These knowledge points may include but are not limited to positioning accuracy, signal stability, and the deviation between positioning time and expected time.

[0049] For example, the Beidou positioning platform system can find that in certain specific geographical areas, the stability of the positioning signal will be affected due to building obstruction or electromagnetic interference, so signal stability can be used as an important screening knowledge point.

[0050] Step S1113: based on the screening knowledge points and the screening strategy, deleting the corresponding positioning event data from the initial positioning event, and generating an updated initial positioning event.

[0051] After determining the screening knowledge points, the Beidou positioning platform system can formulate corresponding screening strategies, that is, to determine which positioning events are abnormal based on the specific values ​​or ranges of these knowledge points. Once an indicator of a positioning event exceeds the preset normal range, the event will be considered abnormal and deleted from the initial positioning event.

[0052] For example, the Beidou positioning platform system can set a threshold for signal stability, and positioning events below this threshold will be considered abnormal and deleted. After such a screening process, the Beidou positioning platform system will obtain an updated initial positioning event set, and the events in this set are relatively normal and reliable, which can be used for subsequent positioning algorithm training or data analysis.

[0053] Through such steps, the Beidou positioning platform system can effectively filter out high-quality data from a large number of positioning events and improve the accuracy and reliability of the positioning algorithm.

[0054] In a possible implementation, step S112 includes: Step S1121: construct a positioning control vector of the initial positioning event according to the past positioning control process record of the initial positioning event, wherein the positioning control vector at least includes a positioning tag vector, a positioning time domain vector and a positioning space domain vector.

[0055] In this embodiment, the Beidou positioning platform system can first review the past positioning control process records of the initial positioning event, which include various data in the positioning process, such as positioning tags, timestamps, and geographic locations. Based on these data, the Beidou positioning platform system can construct a positioning control vector, which includes three main parts: a positioning tag vector (recording the tags or identifiers used in the positioning process), a positioning time domain vector (reflecting the time and duration of the positioning event), and a positioning airspace vector (indicating the location and spatial distribution of the positioning event).

[0056] For example, for a specific initial positioning event, its positioning control vector may include a label indicating that the event is "outdoor walking", a timestamp accurate to seconds, and a latitude and longitude coordinate point.

[0057] Step S1122: constructing a positioning application scenario vector of the initial positioning event according to the positioning logic of the initial positioning event.

[0058] Next, the Beidou positioning platform system can build a positioning application scenario vector based on the positioning logic of the initial positioning event. The positioning application scenario vector describes the specific environment and context in which the positioning event occurs, such as whether it is in a city street, indoor shopping mall or mountainous area. This information is crucial for understanding the characteristics of the positioning event and possible problems.

[0059] For example, if the initial positioning event occurs in a high-rise building densely populated area in the city center, then its positioning application scenario vector may include a label such as “urban high-rise building environment”.

[0060] Step S1123: performing time-sequential processing on the positioning control vector and the positioning application scenario vector of the initial positioning event according to a preset time window, so as to summarize the positioning control vectors and time-sequential positioning application scenario vectors of different time windows.

[0061] The Beidou positioning platform system can perform time-series processing on positioning control vectors and positioning application scenario vectors according to preset time windows (such as every minute, every hour, etc.). This means that the Beidou positioning platform system can arrange these vectors in chronological order and summarize the data in different time windows. This helps to capture the changing trends and patterns of positioning events over time.

[0062] For example, the Beidou positioning platform system can find that within a certain period of time, both the positioning tag vector and the positioning application scenario vector have changed significantly, which may mean that the user has moved from indoors to outdoors during this period of time.

[0063] Step S1124: performing encoding conversion on at least one of the basic event features before time sequencing and the basic event features after time sequencing and the positioning application scenario vector to obtain first encoding data of the initial positioning event.

[0064] In this step, the Beidou positioning platform system can encode and convert the basic event features before time series (such as the label, time and location of a single positioning event) and the basic event features after time series (such as the changing trend of positioning events over a period of time) with the positioning application scenario vector. The conversion process converts these features into a standardized format to facilitate subsequent data processing and model training.

[0065] For example, the Beidou positioning platform system can convert the "outdoor walking" label into a specific digital code, convert the longitude and latitude coordinates into a two-dimensional vector, etc.

[0066] Step S1125 , processing the persistence data in the first coded data of the initial positioning event to generate second coded data of the initial positioning event.

[0067] After processing the first coded data, the BeiDou positioning platform system can pay special attention to the persistent data, such as continuous positioning timestamps or continuous geographic location changes, which are very important for capturing the dynamic characteristics of positioning events. The BeiDou positioning platform system can further process these data, such as calculating time intervals, distance changes, etc., to generate second coded data.

[0068] For example, the Beidou positioning platform system can calculate the time difference and distance difference between two consecutive positioning events and add this information to the second coded data.

[0069] Step S1126: merge the first coded data and the second coded data of the initial positioning event to generate a coded representation sequence of the initial positioning event, where the coded representation sequence of the initial positioning event is a quantized coded representation sequence.

[0070] Finally, the Beidou positioning platform system can fuse the first coded data and the second coded data to generate a quantized coded representation sequence, which integrates the static and dynamic characteristics of the positioning event, providing rich and comprehensive information for subsequent model training and data analysis.

[0071] For example, the Beidou positioning platform system can merge the label code, time vector in the first encoded data and the time difference, distance difference and other information in the second encoded data into a unified quantized coding representation sequence.

[0072] In a possible implementation, step S1124 may include: Step S1124-1, performing a standardization rule conversion on at least one of the positioning control vector before timing and the positioning control vector after timing and the quantization vector in the positioning application scenario vector. In this embodiment, the Beidou positioning platform system will first process those quantifiable data, that is, numerical data, which may exist in both the positioning control vector and the positioning application scenario vector. The quantized vector includes but is not limited to the positioning signal strength, positioning error range, positioning duration, etc.

[0073] For example, suppose the Beidou positioning platform system is processing an initial positioning event, in which the positioning control vector before timing contains a signal strength value (such as -80dBm), and the positioning application scenario vector contains a value representing the positioning error range (such as ±10 meters). In order to unify these quantitative data of different dimensions and value ranges into a common scale, the Beidou positioning platform system can use standardized rules for conversion.

[0074] The standardization rule transformation usually involves subtracting the mean of the data and dividing it by its standard deviation, thereby converting the data into a distribution with a mean of 0 and a standard deviation of 1. The advantage of this is that it can eliminate the deviations between different features due to different dimensions and value ranges, making subsequent data processing and model training more accurate and efficient.

[0075] In this example, the BeiDou positioning platform system can calculate the mean and standard deviation of the signal strength and positioning error range, respectively, and perform standard processing on these two features. The processed data will be included in the first coded data of the initial positioning event.

[0076] Step S1124-2: performing bucket processing on at least one of the positioning control vector before timing and the positioning control vector after timing and the non-quantized vector in the positioning application scenario vector.

[0077] In addition to quantitative data, the positioning control vector and positioning application scenario vector may also contain some non-quantitative data, such as positioning mode (such as GPS, GLONASS, etc.), positioning environment type (such as urban, suburban, indoor, etc.), etc. These data are usually discrete, unordered, and do not have direct numerical meaning.

[0078] For example, suppose that in the initial positioning event being processed by the Beidou positioning platform system, the positioning control vector before timing contains a positioning mode tag (such as "GPS"), and the positioning application scenario vector contains a positioning environment type tag (such as "city"). In order to convert these non-quantized label data into a format suitable for subsequent processing, the Beidou positioning platform system can use bucket processing.

[0079] Bucketing is the process of dividing data into a number of discrete buckets or categories and assigning a unique identifier to each bucket. In this example, the Beidou positioning platform system can create different buckets for different positioning modes and positioning environment types, and assign a digital identifier to each bucket. For example, "GPS" may be assigned "1" and "City" may be assigned "2".

[0080] After bucketing, these non-quantized label data are converted into quantitative data that can be used for subsequent data processing and model training, and are included in the first encoded data of the initial positioning event.

[0081] In a possible implementation, step S1121 may include: Step S1121-1, extracting key data from the past positioning control process records of the initial positioning event, the key data including positioning time, positioning location, positioning method, positioning accuracy, and positioning status.

[0082] In this embodiment, the Beidou positioning platform system first extracts key data from the past positioning control process records of the initial positioning event. These key data are the basis for constructing the positioning control vector, specifically including: Positioning time: records the specific time when the positioning event occurs, such as "2023-04-01 10:30:00".

[0083] Positioning location: records the specific location when the positioning event occurs, which can be a specific longitude and latitude coordinate, such as "39.908722 North Latitude, 116.397490 East Longitude".

[0084] Positioning method: record the positioning technology used, such as GPS, Wi-Fi positioning, base station positioning, etc.

[0085] Positioning accuracy: Indicates the accuracy of positioning, usually in meters, such as "10 meters".

[0086] Positioning status: indicates whether the positioning is successful, or other status information during the positioning process, such as "positioning successful", "positioning failed", etc.

[0087] Step S1121 - 2 , clean the extracted key data, remove invalid, duplicate or erroneous data records, and generate cleaned key data.

[0088] In this embodiment, the Beidou positioning platform system can clean the extracted key data to remove invalid, duplicate or erroneous data records. For example, records with obviously wrong timestamps (such as future time) or obviously abnormal location information (such as longitude and latitude outside the reasonable range) can be deleted. The cleaned data will be more accurate and reliable, providing a high-quality data foundation for subsequent steps.

[0089] Step S1121 - 3 , assigning a unique label value to each positioning mode according to the positioning mode, and using the label value as a part of the positioning label vector.

[0090] The Beidou positioning platform system can assign unique tag values ​​to different positioning methods, and these tag values ​​will be used as part of the positioning tag vector. For example, GPS positioning may be assigned a tag value of "1", Wi-Fi positioning is assigned a tag value of "2", and so on. The advantage of this is that different positioning methods can be converted into numerical forms that can be mathematically operated, which is convenient for subsequent data processing and model training.

[0091] Step S1121 - 4 , generating event pattern features based on the generation sequence of positioning time as the positioning time domain vector.

[0092] The Beidou positioning platform system can generate event pattern features based on the sequence of positioning time. These features may include the frequency and time interval of positioning events, which are used to describe the distribution and regularity of positioning events in the time dimension. These time domain features will constitute the positioning time domain vector.

[0093] Step S1121-5, extracting spatial features as the positioning airspace vector according to the positioning location, the spatial features including longitude and latitude, area code of the geographic location, distance from a specific location, and movement speed and direction changes of continuous positioning records.

[0094] The Beidou positioning platform system can extract spatial features based on the positioning location. These features include: Latitude and longitude: specific geographic location coordinates.

[0095] Area code of geographic location: coded information such as country, city, region, etc.

[0096] Distance from specific locations: such as the distance from places frequently visited by users, the distance from the city center, etc.

[0097] Continuous positioning records of movement speed and direction changes: Through continuous positioning records, the user's movement speed and direction changes can be calculated. These features are very important for understanding the user's movement behavior.

[0098] These spatial features will constitute the positioning spatial domain vector, which is used to describe the characteristics of the positioning event in the spatial dimension.

[0099] Step S1121-6, fusing the positioning tag vector, the positioning time domain vector and the positioning space domain vector to generate the positioning control vector, which is used to describe the characteristics of the initial positioning event in time, space and mode.

[0100] Finally, the BeiDou positioning platform system can fuse the positioning tag vector, the positioning time domain vector and the positioning space domain vector to generate a comprehensive positioning control vector, which integrates all the characteristics of the positioning event in time, space and mode, and provides a rich information basis for subsequent model training and data analysis. In this way, the BeiDou positioning platform system can more accurately understand and describe the characteristics and laws of each initial positioning event.

[0101] In a possible implementation, step S1122 may include: Step S1122-1, obtaining the positioning logic of the initial positioning event, wherein the positioning logic includes the conditions for triggering the positioning request, the purpose of the positioning data, and the key decision points in the positioning process.

[0102] In this embodiment, the Beidou positioning platform system will first obtain the positioning logic of the initial positioning event, which includes the conditions for triggering the positioning request (such as user active request, application automatic trigger, etc.), the purpose of positioning data (such as navigation, check-in, location sharing, etc.) and the key decision points in the positioning process (such as when to start positioning, which positioning technology to use, etc.). This information is crucial for understanding the purpose and context behind the positioning event.

[0103] For example, suppose a travel app automatically triggers a positioning request when a user opens the app and searches for nearby attractions to provide users with personalized recommendations. The positioning logic here may include: the user opens the app and searches for attractions as a trigger condition, and the positioning data is used to provide users with nearby attraction recommendations. Key decision points may include starting positioning immediately when the user searches, and giving priority to GPS positioning technology to ensure accuracy.

[0104] Step S1122-2: Identify different positioning application scenarios according to the positioning logic.

[0105] Based on the above positioning logic, the Beidou positioning platform system can identify different positioning application scenarios, which may vary due to different applications, different user behaviors or different positioning requirements.

[0106] For example, in the above travel application example, the Beidou positioning platform system can identify the following positioning application scenarios: "Nearby Attractions Recommendations": A location scenario triggered when a user opens the app and searches for nearby attractions.

[0107] “Navigate to a scenic spot”: This is the location scenario when the user selects a scenic spot and starts the navigation function.

[0108] "Check-in and share location": a location scenario in which a user checks in and shares their location after arriving at a scenic spot.

[0109] Step S1122-3, describe each positioning application scenario in detail, and select key features related to the positioning application scenario, wherein the key features include user behavior data, application interface status, and device status.

[0110] For each identified positioning application scenario, the Beidou positioning platform system can provide a detailed description and select key features related to the scenario. These key features may include user behavior data (such as search, click, slide, etc.), application interface status (such as the currently displayed page, the status of function buttons, etc.) and device status (such as the device's battery level, network connection status, etc.).

[0111] For example, in the positioning application scenario of "recommending nearby attractions", the Beidou positioning platform system can select the following key features: User behavior data: The user searched for "nearby attractions".

[0112] Application interface status: Currently displayed on the "Nearby Attractions" search results page.

[0113] Device status: The device is connected to the network and has sufficient battery power.

[0114] Step S1122-4: for each positioning application scenario, construct a positioning application scenario vector according to the extracted key features, wherein the positioning application scenario vector is used to reflect the key features in the positioning application scenario.

[0115] Based on the extracted key features, the Beidou positioning platform system can construct a positioning application scenario vector for each positioning application scenario. The positioning application scenario vector will reflect the key features of the scenario and help subsequent data analysis and model training.

[0116] For example, in the "nearby attractions recommendation" scenario, the positioning application scenario vector may include: A feature value indicating that "the user searched for nearby attractions" (e.g., 1 for searched, 0 for not searched).

[0117] A feature value indicating "currently displayed on the nearby attractions search results page" (e.g., 1 for that page, 0 for that page).

[0118] A characteristic value indicating that the device is connected to the network and has sufficient power (e.g., 1 if the condition is met, 0 if it is not).

[0119] Step S1122-5: Use binary encoding or one-hot encoding to represent the positioning application scenario vector.

[0120] Finally, the Beidou positioning platform system can use binary encoding or one-hot encoding to represent these positioning application scenario vectors. The advantage of this is that these vectors can be converted into a format that is easy for computers to process, facilitating subsequent data analysis and model training.

[0121] For example, in the "nearby attractions recommendation" scenario, if binary encoding is used, the above positioning application scenario vector may be represented as: [1, 1, 1]. If one-hot encoding is used (assuming we have three key features), the vector may be represented as: [1, 0, 0, 1, 0, 0, 1, 0, 0] (where each three bits represent a different state of a feature). In this way, the Beidou positioning platform system can accurately describe and identify different positioning application scenarios through the encoded vectors.

[0122] In a possible implementation, step S120 includes: When the event scale comparison parameter of the first training Beidou positioning event sequence and the second training Beidou positioning event sequence is greater than the set comparison parameter, the extraction quantity is determined based on the set comparison parameter, the event scale of the training Beidou positioning event subsequence and the event scale of the second training Beidou positioning event sequence.

[0123] In this embodiment, the Beidou positioning platform system first compares the event scale of the first training Beidou positioning event sequence and the second training Beidou positioning event sequence. The event scale may refer to the number of positioning events or the total amount of data contained in the sequence. The comparison is to determine the difference in quantity between the two sequences, so as to determine how to process the data later.

[0124] For example, suppose the first training BeiDou positioning event sequence contains 100,000 positioning events, and the second training BeiDou positioning event sequence contains 50,000 positioning events. The BeiDou positioning platform system can compare these two numbers and conclude that the event scale of the first sequence is twice that of the second sequence.

[0125] The Beidou positioning platform system can determine the relationship between the actual event scale comparison parameter (i.e. the ratio or difference between the scales of two sequence events, etc.) and the preset comparison parameter. The preset comparison parameter is a threshold set in advance to guide subsequent data processing strategies.

[0126] For example, if the comparison parameter is set to 1.5, it means that when the event scale of the first sequence is more than 1.5 times that of the second sequence, the Beidou positioning platform system can adopt a specific data processing strategy. In the previous example, the actual comparison parameter is 2 (100,000 / 50,000), which is greater than the set comparison parameter of 1.5.

[0127] When the actual event scale comparison parameter is greater than the set comparison parameter, the Beidou positioning platform system can determine the number of training Beidou positioning event subsequences extracted from the first training Beidou positioning event sequence based on the set comparison parameter, the event scale of the training Beidou positioning event subsequence and the event scale of the second training Beidou positioning event sequence.

[0128] For example, since the actual comparison parameter 2 is greater than the set comparison parameter 1.5, the Beidou positioning platform system now needs to determine how many subsequences to extract from the first sequence for subsequent processing. The determination of this number may be based on a variety of factors, such as the desire to keep the two sequences more balanced in the amount of data after processing, or to adjust according to specific business needs and resource limitations.

[0129] Suppose the BDS decides that the number of subsequences to be extracted from the first sequence should be such that the total number of events in these subsequences is close to the event size of the second sequence. In this case, the BDS can calculate an extraction quantity, such as extracting two subsequences containing 25,000 events from the first sequence, so that the total number of events in the two subsequences matches the 50,000 events in the second sequence.

[0130] This process ensures that the amount of data extracted from the two training Beidou positioning event sequences of different sizes can remain relatively balanced in subsequent processing, thereby avoiding problems such as model training bias or inefficient resource utilization due to large differences in data volume.

[0131] In a possible implementation, step S140 may include: Step S141, based on a predefined selection strategy, select multiple feature vectors of each target sample learning data in the target sample learning data sequence, load them into the target deep learning network model, and output a weighted discriminant structure.

[0132] In this embodiment, the Beidou positioning platform system will first select multiple feature vectors of each target sample learning data from the target sample learning data sequence according to a predefined selection strategy. These feature vectors may include positioning signal strength, positioning time, positioning accuracy and other aspects, which are crucial for diagnosing Beidou positioning anomalies.

[0133] After selecting the feature vectors, the Beidou positioning platform system can load these feature vectors into a pre-built deep learning network model, which can be a complex neural network with a multi-layer structure and multiple nodes, capable of learning and recognizing complex patterns.

[0134] After loading the feature vector, the deep learning model will perform forward propagation calculations and output a weighted discriminant structure based on the current parameters of the model. The weighted discriminant structure reflects the model's understanding and judgment of the input feature vector and is the basis for model learning and optimization.

[0135] For example, suppose the BeiDou positioning platform system is processing a sequence of 100 target sample learning data. For each sample, the BeiDou positioning platform system selects 5 feature vectors according to predefined strategies, such as signal strength, positioning time difference, positioning accuracy change, etc. Then, the BeiDou positioning platform system inputs these feature vectors into a deep learning model with 5 hidden layers. After calculation, the model outputs a weighted discriminant structure, which reflects the model's understanding of the current input features.

[0136] Step S142, iteratively executes the step of selecting multiple feature vectors of each target sample learning data in the target sample learning data sequence based on a predefined selection strategy, loading them into the target deep learning network model, and outputting a weight discriminant structure until M weight discriminant structures are generated, where M is a positive integer greater than 1.

[0137] The Beidou positioning platform system can repeatedly perform the above steps, that is, continuously select feature vectors from the target sample learning data sequence and load them into the deep learning model to generate a weighted discriminant structure. This process will be iterated multiple times until M weighted discriminant structures are generated.

[0138] Here, M is a positive integer greater than 1, and its specific value may be determined according to actual needs and the size of the data set. By generating multiple weighted discriminant structures through multiple iterations, the diversity and generalization ability of the model can be increased.

[0139] For example, continuing with the above example, the Beidou positioning platform system can repeat the above process 100 times (assuming M=100), each time selecting feature vectors from 100 target samples and inputting them into the model. In this way, the Beidou positioning platform system will eventually obtain 100 different weighted discriminant structures.

[0140] Step S143, integrating the M weight discrimination structures to generate the Beidou positioning anomaly diagnosis network.

[0141] When the Beidou positioning platform system obtains M weighted discriminant structures, it will perform an integration operation to merge these structures into a more robust and stable Beidou positioning anomaly diagnosis network. The integration process may include weight averaging, voting mechanisms or other complex integration strategies.

[0142] The purpose of integration is to improve the generalization performance and robustness of the model. By combining the judgment results of multiple weighted discriminant structures, the bias and variance of a single model can be reduced, thereby obtaining more accurate and reliable abnormal diagnosis results.

[0143] For example, after obtaining 100 weighted discriminant structures, the BeiDou positioning platform system begins to perform integration operations. It may use a simple weighted average method to weight the output of each structure; or use a voting mechanism to let multiple structures diagnose the same input and select the result with the most votes as the final output. Through such an integration process, the BeiDou positioning platform system obtains a more powerful and stable BeiDou positioning anomaly diagnosis network.

[0144] In a possible implementation, step S140 may further include: Step S144, splitting the target sample learning data sequence into a learning subsequence and a verification subsequence.

[0145] Step S145, training the Beidou positioning anomaly diagnosis network according to the initialization configuration data and the learning subsequence, wherein the initialization configuration data includes the scale of the training Beidou positioning event subsequence, M, and the scale of the feature vector of each target sample learning data in the target sample learning data sequence selected based on a predefined selection strategy.

[0146] Step S146, performing parameter optimization search on the Beidou positioning anomaly diagnosis network.

[0147] Step S147, re-optimize the parameters of the Beidou positioning abnormality diagnosis network according to the network parameter information after the parameter optimization search, generate the optimized Beidou positioning abnormality diagnosis network, and perform a performance test on the optimized Beidou positioning abnormality diagnosis network based on the verification subsequence.

[0148] In this embodiment, the Beidou positioning platform system first splits the target sample learning data sequence into a learning subsequence and a verification subsequence. The learning subsequence is used to train the Beidou positioning anomaly diagnosis network, while the verification subsequence is used to test and optimize the network performance test.

[0149] For example, suppose the BeiDou positioning platform system has a data sequence containing 1,000 target sample learning data. According to the preset ratio, such as 80% of the data is used for learning and 20% of the data is used for verification, the BeiDou positioning platform system can divide the first 800 data into learning subsequences and the last 200 data into verification subsequences.

[0150] Next, the BeiDou positioning platform system can use the learning subsequences and initialization configuration data to train the BeiDou positioning anomaly diagnosis network. The initialization configuration data includes the scale of the training BeiDou positioning event subsequences, the value of M (i.e., the number of weighted discriminant structures mentioned above), and the scale of the feature vector selected based on the predefined selection strategy.

[0151] For example, according to the initialization configuration data, the Beidou positioning platform system sets each training Beidou positioning event subsequence to contain 100 data, the value of M is 10, and the feature vector scale of each target sample learning data is 5. Then, the Beidou positioning platform system uses the learning subsequence (800 data) for training, taking 100 data as a subsequence each time, extracting 5 feature vectors of each data, and inputting them into the Beidou positioning anomaly diagnosis network for training. This process will be repeated many times until the network achieves a certain training effect.

[0152] During the training process, the Beidou positioning platform system can perform parameter optimization search on the Beidou positioning anomaly diagnosis network in order to find the parameter combination that can achieve the optimal network performance.

[0153] For example, the BeiDou positioning platform system can try different parameter combinations, such as learning rate, batch size, regularization coefficient, etc., to train the BeiDou positioning anomaly diagnosis network. By comparing the training and verification effects of the network under different parameter combinations, the BeiDou positioning platform system can find a set of optimal parameter combinations.

[0154] Finally, the Beidou positioning platform system can re-optimize the parameters of the Beidou positioning abnormality diagnosis network based on the network parameter information after the parameter optimization search to generate an optimized network. Then, the performance of the optimized network is tested using the verification subsequence.

[0155] For example, the BeiDou positioning platform system uses the optimal parameter combination found in the parameter optimization search to retrain the BeiDou positioning anomaly diagnosis network. After training, the BeiDou positioning platform system can use the verification subsequence (200 data) to test the performance of the network. By comparing the performance of the network on the verification subsequence, such as accuracy, recall rate and other indicators, the BeiDou positioning platform system can evaluate whether the performance of the optimized network has improved. If the performance improvement is obvious, it means that the parameter optimization is effective.

[0156] In a possible implementation, step S150 includes: Step S151, encoding and representing the acquired candidate Beidou positioning event sequence to generate an encoding and representation sequence of the candidate Beidou positioning event.

[0157] In this embodiment, after receiving the candidate Beidou positioning event sequence, the Beidou positioning platform system will first encode these candidate Beidou positioning event events. The purpose of encoding is to convert these candidate Beidou positioning event events into a format that can be understood and processed by the Beidou positioning abnormality diagnosis network.

[0158] For example, suppose the BeiDou positioning platform system receives a sequence of 10 candidate BeiDou positioning events. Each event contains information such as longitude and latitude, timestamp, signal strength, etc. The BeiDou positioning platform system can use a specific encoding method, such as one-hot encoding or embedding vector, to convert this information into a vector of fixed length, so that each candidate BeiDou positioning event is converted into a coded representation, and the entire sequence is converted into a coded representation sequence.

[0159] Step S152, loading the coded representation sequence of the candidate Beidou positioning event into each of the weighted discrimination structures in the optimized Beidou positioning anomaly diagnosis network, so that each of the weighted discrimination structures generates discrimination data.

[0160] Next, the Beidou positioning platform system can load the coded representation sequence into the optimized Beidou positioning anomaly diagnosis network, which is composed of multiple weighted discriminant structures, each of which processes the input coded representation sequence and generates discriminant data.

[0161] For example, the Beidou positioning platform system inputs the coded representation sequence generated in the previous step into the optimized Beidou positioning anomaly diagnosis network, which may contain multiple hidden layers and weighted discriminant structures, each of which performs complex calculations and processing on the input sequence. During the processing, each weighted discriminant structure will generate a corresponding discriminant data based on its own parameters and algorithms, which reflects the network's understanding and judgment of the input sequence.

[0162] Step S153, generating the abnormal defect diagnosis data based on the discrimination data generated by each of the weight discrimination structures.

[0163] Finally, the Beidou positioning platform system can generate final abnormal defect diagnosis data based on the discrimination data generated by each weighted discrimination structure. These data can help users understand whether there are abnormalities in the candidate Beidou positioning events, as well as the specific type and degree of the abnormality.

[0164] For example, after receiving the discrimination data generated by all weight discrimination structures, the Beidou positioning platform system will conduct a comprehensive analysis and processing of these data. For example, the Beidou positioning platform system can use voting mechanisms or weighted average methods to merge these discrimination data into a comprehensive diagnosis result, which will clearly indicate which candidate Beidou positioning events are abnormal, as well as the specific type and severity of the abnormality. At the same time, the Beidou positioning platform system will also generate a detailed abnormal defect diagnosis report for user reference and decision-making. This report may contain detailed information such as the location information, abnormality type, and occurrence time of the abnormal event, as well as suggestions and solutions for these abnormalities.

[0165] In a possible implementation manner, the discrimination data includes the confidence that the candidate Beidou positioning event corresponds to different abnormal defect labels, and step S153 includes: Step S1531, calculating the scale of the discrimination data whose confidence level of each abnormal defect label is greater than the set confidence level.

[0166] Step S1532, determining the abnormal defect labels whose scale is greater than half of M as the abnormal defect diagnosis data.

[0167] In this embodiment, after receiving the discrimination data generated by each weighted discrimination structure, the Beidou positioning platform system begins to analyze the discrimination data in detail. The discrimination data includes the confidence of the candidate Beidou positioning event corresponding to different abnormal defect labels, which indicates the possibility that the event belongs to a certain abnormal defect.

[0168] Assume that the Beidou positioning platform system receives discrimination data about the same candidate Beidou positioning event from three weighted discrimination structures, and these data contain five possible abnormal defect labels (such as signal interference, hardware failure, positioning deviation, data error, other abnormalities) and their corresponding confidence levels. The Beidou positioning platform system will first set a confidence threshold, such as 0.7, and then filter each label to retain only those discrimination data with a confidence level greater than 0.7.

[0169] Next, the Beidou positioning platform system can calculate the scale of the discriminant data with a confidence level greater than 0.7 under each abnormal defect label. For example, for the label "signal interference", if two weighted discriminant structures give a confidence level greater than 0.7, then the scale of the label is 2; for the label "hardware failure", if only one weighted discriminant structure has a confidence level greater than 0.7, then the scale of the label is 1, and so on.

[0170] After calculating the discriminant data size of each abnormal defect label, the Beidou positioning platform system can compare these sizes with half of M (M is the number of weighted discriminant structures mentioned earlier). If the discriminant data size of a label is greater than half of M, the Beidou positioning platform system will consider this label to be valid abnormal defect diagnosis data.

[0171] Continuing with the above example, assuming that the value of M is 3, then half of M is 1.5. For the "signal interference" tag, since its scale is equal to 2, which is greater than 1.5, the Beidou positioning platform system can determine it as valid abnormal defect diagnosis data. For the "hardware failure" tag, since its scale is only 1, which is less than 1.5, the Beidou positioning platform system will not regard it as valid abnormal defect diagnosis data.

[0172] In this way, the Beidou positioning platform system can accurately identify possible abnormal defects in candidate Beidou positioning events and provide users with targeted diagnostic results and suggestions.

[0173] Further, Figure 2 FIG. 1 is a schematic diagram of the hardware structure of a Beidou positioning platform system 100 for implementing the method provided in the embodiment of the present application. Figure 2As shown, the Beidou positioning platform system 100 may include at least one processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 and a controller 108 for communication functions. It can be understood by those skilled in the art that Figure 2 The structure shown is only for illustration and does not limit the structure of the Beidou positioning platform system 100. For example, the Beidou positioning platform system 100 may also include Figure 2 More or fewer components as shown, or with Figure 2 Different configurations are shown.

[0174] The memory 104 can be used to store software programs and modules of application software, such as program instructions corresponding to the above-mentioned method embodiments in the embodiments of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, realizing the above-mentioned Beidou positioning information processing method based on artificial intelligence. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the Beidou positioning platform system 100 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0175] The transmission device 106 is used to obtain or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the Beidou positioning platform system 100. In one example, the transmission device 106 includes a network adapter, which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency module, which is used to communicate with the Internet wirelessly.

[0176] It should be noted that the order of the above embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. The above describes specific embodiments of the embodiments of the present application. Other embodiments are within the scope of the attached claims. In some cases, the exceptions or steps recorded in the claims can be performed according to an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or can be advantageous.

[0177] Each embodiment in the embodiments of the present application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the above different embodiments, since they are basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0178] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program. The above program may be stored in a computer-readable storage medium, and the above storage medium may be a read-only memory, a disk or an optical disk, etc.

Claims

1. A Beidou positioning information processing method based on artificial intelligence, characterized in that: Applied to the Beidou positioning platform system, the method includes: Acquire a first training Beidou positioning event sequence and a second training Beidou positioning event sequence, wherein the first training Beidou positioning event sequence and the second training Beidou positioning event sequence have different abnormal defect labels, and the event scale of the first training Beidou positioning event sequence is greater than the event scale of the second training Beidou positioning event sequence; Dividing the first training Beidou positioning event sequence into a plurality of training Beidou positioning event subsequences, and determining the extraction quantity of each of the training Beidou positioning event subsequences based on an event scale comparison parameter of the event scale of the first training Beidou positioning event sequence and the event scale of the second training Beidou positioning event sequence; Performing event extraction on each of the training Beidou positioning event subsequences based on the extraction quantity, outputting a third training Beidou positioning event sequence, and generating a target sample learning data sequence based on the third training Beidou positioning event sequence and the second training Beidou positioning event sequence, wherein an event scale comparison parameter of the event scale of the third training Beidou positioning event sequence and the event scale of the second training Beidou positioning event sequence is less than a set comparison parameter; Optimizing the parameters of the Beidou positioning anomaly diagnosis network based on the target sample learning data sequence to generate the optimized Beidou positioning anomaly diagnosis network; Based on the optimized Beidou positioning abnormality diagnosis network, abnormality diagnosis is performed on candidate Beidou positioning events to generate abnormal defect diagnosis data of the candidate Beidou positioning events, wherein the abnormal defect diagnosis data represents the abnormal defect label of the candidate Beidou positioning event.

2. The Beidou positioning information processing method based on artificial intelligence according to claim 1 is characterized in that: The obtaining of the first training Beidou positioning event sequence and the second training Beidou positioning event sequence comprises: Obtaining an initial positioning event carrying abnormal annotation data from a sample Beidou positioning event log, wherein the abnormal annotation data represents an abnormal defect label of the initial positioning event; Encoding the initial positioning event to generate a coding representation sequence of the initial positioning event; The first training Beidou positioning event sequence and the second training Beidou positioning event sequence are determined based on the coded representation sequence of the initial positioning event.

3. The Beidou positioning information processing method based on artificial intelligence according to claim 2 is characterized in that: The obtaining of the initial positioning event carrying abnormal annotation data from the sample Beidou positioning event log includes: Acquiring positioning control path data of the initial positioning event; Determining screening knowledge points based on the positioning control path data of the initial positioning event; Based on the screening knowledge points and the screening strategy, the corresponding positioning event data is deleted from the initial positioning event to generate an updated initial positioning event.

4. The Beidou positioning information processing method based on artificial intelligence according to claim 2 is characterized in that: The encoding and representing the initial positioning event to generate an encoding and representing sequence of the initial positioning event includes: Constructing a positioning control vector of the initial positioning event according to a past positioning control process record of the initial positioning event, wherein the positioning control vector at least includes a positioning tag vector, a positioning time domain vector, and a positioning space domain vector; According to the positioning logic of the initial positioning event, construct a positioning application scenario vector of the initial positioning event; Performing time-sequential processing on the positioning control vector and the positioning application scenario vector of the initial positioning event according to a preset time window, so as to summarize the positioning control vectors and time-sequential positioning application scenario vectors of different time windows; Performing encoding conversion on at least one of the basic event features before time serialization and the basic event features after time serialization and a positioning application scenario vector to obtain first encoding data of the initial positioning event; Processing the persistence data in the first coded data of the initial positioning event to generate second coded data of the initial positioning event; The first coded data and the second coded data of the initial positioning event are merged to generate a coded representation sequence of the initial positioning event, wherein the coded representation sequence of the initial positioning event is a quantized coded representation sequence; The step of encoding and converting at least one of the basic event features before time sequencing and the basic event features after time sequencing with the positioning application scenario vector to obtain first encoded data of the initial positioning event includes: Performing standardization rule conversion on at least one of the positioning control vector before timing and the positioning control vector after timing and the quantization vector in the positioning application scenario vector; and Performing bucket processing on at least one of the positioning control vector before timing and the positioning control vector after timing and a non-quantized vector in the positioning application scenario vector; The step of constructing the positioning control vector of the initial positioning event based on the past positioning control process record of the initial positioning event includes: Extracting key data from the past positioning control process records of the initial positioning event, wherein the key data includes positioning time, positioning location, positioning method, positioning accuracy, and positioning status; Clean the extracted key data, remove invalid, duplicate or erroneous data records, and generate cleaned key data; According to the positioning mode, a unique label value is assigned to each positioning mode, and the label value is used as a part of the positioning label vector; Generate event pattern features based on the generation sequence of positioning time as the positioning time domain vector; Extracting spatial features as the positioning airspace vector according to the positioning location, the spatial features including longitude and latitude, an area code of a geographic location, a distance from a specific location, and a moving speed and direction change of continuous positioning records; The positioning tag vector, the positioning time domain vector and the positioning space domain vector are merged to generate the positioning control vector, where the positioning control vector is used to describe the characteristics of the initial positioning event in time, space and mode; The step of constructing a positioning application scenario vector of the initial positioning event according to the positioning logic of the initial positioning event includes: Obtaining the positioning logic of the initial positioning event, the positioning logic including the conditions for triggering the positioning request, the purpose of the positioning data, and the key decision points in the positioning process; According to the positioning logic, different positioning application scenarios are identified; Describe each positioning application scenario in detail, and select key features related to the positioning application scenario, wherein the key features include user behavior data, application interface status, and device status; For each positioning application scenario, construct a positioning application scenario vector according to the extracted key features, wherein the positioning application scenario vector is used to reflect the key features in the positioning application scenario; The positioning application scenario vector is represented by binary encoding or one-hot encoding.

5. The Beidou positioning information processing method based on artificial intelligence according to any one of claims 1 to 4, characterized in that: The step of dividing the first training Beidou positioning event sequence into a plurality of training Beidou positioning event subsequences, and determining the extraction quantity of each of the training Beidou positioning event subsequences based on an event scale comparison parameter of an event scale of the first training Beidou positioning event sequence and an event scale of the second training Beidou positioning event sequence, comprises: When the event scale comparison parameter of the first training Beidou positioning event sequence and the second training Beidou positioning event sequence is greater than the set comparison parameter, the extraction quantity is determined based on the set comparison parameter, the event scale of the training Beidou positioning event subsequence and the event scale of the second training Beidou positioning event sequence.

6. The Beidou positioning information processing method based on artificial intelligence according to any one of claims 1 to 4, characterized in that: The method of optimizing parameters of the Beidou positioning abnormality diagnosis network based on the target sample learning data sequence to generate the optimized Beidou positioning abnormality diagnosis network includes: Selecting multiple feature vectors of each target sample learning data in the target sample learning data sequence based on a predefined selection strategy, loading them into a target deep learning network model, and outputting a weighted discriminant structure; Iteratively executing the step of selecting multiple feature vectors of each target sample learning data in the target sample learning data sequence based on a predefined selection strategy, loading them into the target deep learning network model, and outputting a weight discriminant structure until M weight discriminant structures are generated, where M is a positive integer greater than 1; The M weight discrimination structures are integrated to generate the Beidou positioning anomaly diagnosis network.

7. The Beidou positioning information processing method based on artificial intelligence according to claim 6 is characterized in that: The method of optimizing parameters of the Beidou positioning abnormality diagnosis network based on the target sample learning data sequence to generate the optimized Beidou positioning abnormality diagnosis network includes: Splitting the target sample learning data sequence into a learning subsequence and a verification subsequence; According to the initialization configuration data and the learning subsequence, the Beidou positioning anomaly diagnosis network is trained, wherein the initialization configuration data includes the scale of the training Beidou positioning event subsequence, M, and the scale of the feature vector of each target sample learning data in the target sample learning data sequence selected based on a predefined selection strategy; Performing parameter optimization search on the Beidou positioning anomaly diagnosis network; According to the network parameter information after the parameter optimization search, the parameters of the Beidou positioning abnormality diagnosis network are re-optimized to generate the optimized Beidou positioning abnormality diagnosis network, and the performance of the optimized Beidou positioning abnormality diagnosis network is tested based on the verification subsequence.

8. The Beidou positioning information processing method based on artificial intelligence according to claim 6 is characterized in that: The abnormality diagnosis of the candidate Beidou positioning event based on the optimized Beidou positioning abnormality diagnosis network is performed to generate abnormal defect diagnosis data of the candidate Beidou positioning event, including: Encoding the acquired candidate Beidou positioning event sequence to generate a coded representation sequence of the candidate Beidou positioning event; Loading the encoded representation sequence of the candidate Beidou positioning event into each of the weighted discriminant structures in the optimized Beidou positioning anomaly diagnosis network, so that each of the weighted discriminant structures generates discriminant data; The abnormal defect diagnosis data is generated based on the discrimination data generated by each of the weight discrimination structures.

9. The Beidou positioning information processing method based on artificial intelligence according to claim 8 is characterized in that: The discrimination data includes the confidence of the candidate Beidou positioning event corresponding to different abnormal defect labels, and the abnormal defect diagnosis data is generated based on the discrimination data generated by each weighted discrimination structure, including: Calculating the scale of the discrimination data whose confidence level of each abnormal defect label is greater than a set confidence level; The abnormal defect labels whose size is greater than half of M are determined as the abnormal defect diagnosis data.

10. A Beidou positioning platform system, characterized in that: The Beidou positioning platform system includes a processor and a readable storage medium, wherein the readable storage medium stores a program, and when the program is executed by the processor, the Beidou positioning information processing method based on artificial intelligence as described in any one of claims 1-9 is implemented.

Citation Information

Cited By

  • Power terminal positioning method based on Beidou multi-frequency combination

    CN120294802A