User positioning method and device based on communication big data, equipment, medium and product

Abstract

The invention discloses a user positioning method and device based on communication big data, equipment, a medium and a product, and belongs to the technical field of wireless communication and location services, and the method comprises the steps: obtaining MDT sampling point data; executing a multi-stage cleaning process on the MDT data, and performing density anomaly detection by adopting an LOF (Local Outlier Factor) algorithm; rasterizing the cleaned data and classifying the data into buildings / roads / other grids; performing normalization processing on the signal features of the adjacent cells, screening high-frequency adjacent cells, and constructing a uniform feature vector containing a main cell and the first 30 adjacent cells; a high-precision positioning model is constructed by adopting a LightGBM machine learning algorithm, and hyper-parameter adjustment and optimization are carried out through cross validation and early stop strategy optimization training and by utilizing grid search; and calling a corresponding model according to the ECI of the main service cell of MR data, normalizing according to a uniform feature sequence, and then executing positioning calculation. According to the method, the data quality and model precision of the positioning fingerprint database are remarkably improved, and the method is suitable for a high-precision user positioning scene under a large-scale communication network.

Description

Technical Field

[0001] This application relates to the field of wireless communication and location services, and in particular to a user positioning method, apparatus, device, medium and product based on communication big data. Background Technology

[0002] With the rapid development of mobile communication technology, especially the arrival of the 5G era, the demand for accurate user location information is growing. High-precision positioning technology is not only crucial in consumer applications such as emergency rescue and navigation services, but also plays an irreplaceable role in operator and enterprise applications such as network optimization, passenger flow analysis, and regional insights.

[0003] In recent years, positioning methods based on big data in communications, especially those utilizing measurement report data to construct fingerprint databases, have shown great potential. The core idea of ​​this method is to first collect massive amounts of data within a specific area, containing wireless signal characteristics (such as Reference Signal Received Power (RSRP) and Reference Signal Received Quality (RSRQ)) and their corresponding real-world geographic coordinates, to construct a "location fingerprint database." Then, the signal characteristics measured in real-time by the terminal to be located are matched against the fingerprint database to estimate its location.

[0004] However, existing fingerprint database-based localization methods using MR data still face a series of serious challenges in achieving high-precision localization: Data Quality Challenges: The MDT data used to build the fingerprint database contains a large amount of invalid and noisy data during the collection process. For example, sampling points from high-speed mobile users have significant location uncertainty; outliers caused by device malfunctions or signal interference can severely pollute the quality of the fingerprint database. Traditional data cleaning methods are relatively simple and cannot effectively identify and remove these complex outliers, resulting in inherently insufficient accuracy of the constructed fingerprint database.

[0005] Model efficiency and accuracy challenges: A fingerprint database is essentially a complex mapping model from multidimensional signal features to geographic coordinates. Traditional machine learning algorithms (such as support vector machines and simple decision trees) often encounter bottlenecks in training efficiency, memory consumption, and model accuracy when processing large-scale, high-dimensional MR data. They struggle to learn accurate and complex mapping relationships from massive amounts of data within an acceptable timeframe, limiting positioning accuracy and system practicality.

[0006] Feature engineering challenges: MR data contains rich signal features from the serving cell and multiple neighboring cells. A key issue is how to effectively filter out the most critical signals for location determination from these high-dimensional, sparse features and construct a unified feature vector. Inappropriate feature processing can lead to difficulties in model training and poor generalization ability, directly affecting the stability and reliability of the final positioning results.

[0007] Therefore, there is an urgent need in this field for a systematic, high-precision positioning method that can overcome the aforementioned shortcomings. This method needs to possess efficient and accurate data cleaning capabilities, be able to construct robust positioning models using high-performance machine learning algorithms, and have a complete feature engineering and model management process, ultimately achieving high-precision, high-availability user positioning based on big data communication data. Summary of the Invention

[0008] The purpose of this application is to provide a user positioning method, device, equipment, medium and product based on big data communication, which can significantly improve positioning accuracy and system availability through multi-level data cleaning, refined feature engineering and efficient machine learning modeling.

[0009] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a user location method based on big data communication, including: Acquire MDT sampling point data; the data includes the primary serving cell identifier (ECI), neighboring cell signal characteristics, and cell GPS latitude and longitude. Perform multi-level cleaning on the MDT sampling point data; The sampled point data after multi-level cleaning is rasterized, and the raster is classified into building raster, road raster, or other raster types in combination with the building layer; Using the community as a unit, the LOF (Local Outlier Factor) algorithm is applied to the classified sampling point data to detect density anomalies and remove outlier sampling points with LOF values ​​greater than a preset threshold. Feature engineering is performed on the sampled data after anomaly removal, and normalization is performed to obtain normalized feature vectors; Based on the normalized feature vectors and the corresponding GPS coordinates, a high-precision positioning model is trained using the LightGBM machine learning algorithm. Latitude and longitude are used as the regression target, mean square error (MSE) is used as the loss function, and cross-validation and early stopping strategies are combined for training optimization. If the positioning accuracy of the high-precision positioning model on the validation set does not meet the preset index, then perform grid search optimization on the LightGBM hyperparameters and retrain the high-precision positioning model using the optimal parameters to obtain a well-trained high-precision positioning model. Based on the trained high-precision positioning model, the user positioning result is output for real-time MR data.

[0010] Optionally, multi-level cleaning is performed on the MDT sampling point data, specifically including the following steps: The location of the cell center is estimated based on the sampling points with TA=0, and sampling points that are more than twice the ideal coverage radius are removed; Group users by user identifier and sort them by time; identify user motion state; impose time continuity and minimum sampling point constraints on motion state groups; and remove invalid groups. Calculate the average speed of the motion groups and remove groups whose average speed exceeds the preset high-speed threshold; Based on motion direction consistency detection, abnormal sampling points with sudden changes in direction are eliminated.

[0011] Optionally, on a cell-by-cell basis, the LOF (Local Outlier Factor) algorithm is applied to the classified sampling point data for density anomaly detection, removing outlier sampling points with LOF values ​​greater than a preset threshold. This specifically includes the following steps: The sampling point data are grouped according to ECI; Define the k-distance for each sampling point p; The k-distance neighborhood of sampling point p is determined based on the k-distance of the sampling point p; Calculate the reachability distance between each sampling point p and its k neighbors; The local reachability density of sampling point p is calculated based on the k-distance neighborhood of sampling point p and the reachability distance between sampling point p and its k neighbors; The local anomaly factor of sampling point p is calculated based on the local reachability density of the sampling point p. If the local anomaly factor is greater than a preset threshold, it is determined to be an abnormal sampling point and the abnormal sampling point is removed.

[0012] Optionally, feature engineering is performed on the sampled data after anomaly removal to normalize it, resulting in a normalized feature vector. This process includes the following steps: Filter out neighboring cells with a frequency of less than 1%, sort the remaining neighboring cells by average field strength and select the top 30, and construct a unified feature vector containing the main cell's RSRP, RSRQ, tadv and the RSRP and RSRQ of each of the 30 neighboring cells.

[0013] Optionally, the expression for the loss function is as follows: ; in, Is it the actual longitude or latitude? These are model predictions. This represents the number of sampling points used in model training.

[0014] Optionally, the LightGBM hyperparameters include: learning rate, maximum number of bins (max_bin), row sampling ratio (bagging_fraction), column sampling ratio (feature_fraction), and minimum data size in leaf nodes (min_data_in_leaf).

[0015] Secondly, this application provides a user positioning device based on communication big data, comprising: The data acquisition module is used to acquire MDT sampling point data; the data includes the primary serving cell identifier (ECI), neighboring cell signal characteristics, and cell GPS latitude and longitude. A multi-level cleaning module is used to perform multi-level cleaning on the MDT sampling point data; The raster classification module is used to rasterize the multi-level cleaned sampling point data and classify the raster into building raster, road raster, or other raster in combination with the building layer; The anomaly detection module is used to perform density anomaly detection on the classified sampling point data at the cell level, using the LOF local outlier factor algorithm to remove outlier sampling points with LOF values ​​greater than a preset threshold. The normalization module is used to perform feature engineering on the sampled point data after anomaly removal, and to perform normalization processing to obtain the normalized feature vector; The model training module is used to train a high-precision positioning model based on the normalized feature vectors and corresponding GPS coordinates using the LightGBM machine learning algorithm. It uses latitude and longitude as the regression target, the mean square error (MSE) as the loss function, and combines cross-validation and early stopping strategies for training optimization. The parameter optimization module is used to perform grid search tuning on the LightGBM hyperparameters if the positioning accuracy of the high-precision positioning model on the validation set does not meet the preset index, and to retrain the high-precision positioning model using the optimal parameters to obtain a trained high-precision positioning model. The user location result prediction module is used to output user location results based on the trained high-precision positioning model from real-time MR data.

[0016] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the user positioning method based on communication big data as described above.

[0017] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the user location method based on communication big data described above.

[0018] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the user location method based on communication big data described above.

[0019] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a user positioning method, apparatus, device, medium, and product based on big data communication, which has the following significant advantages: Significantly improved data quality: Through four-level rule cleaning based on motion state, distance, speed, and direction, combined with LOF density anomaly detection, complex noise and outliers are effectively removed, providing high-quality samples for model training; High positioning accuracy: It adopts the LightGBM high-efficiency gradient boosting framework, which can handle 63-dimensional high-dimensional sparse features and achieve high-precision regression under large-scale data. The measured positioning success rate at 50 meters is significantly better than that of traditional methods. Feature engineering standardization: By filtering neighboring regions by frequency and using fixed sorting, a unified dimension feature vector is constructed to ensure feature consistency between the training and inference stages and improve the model's generalization ability. Efficient and stable model training: Combining GPU acceleration, early stopping mechanism and grid search optimization, it balances training efficiency and model performance, and avoids overfitting; The system is highly scalable: it models and manages at the cell level, supports dynamic updates of the entire network model and regional assessments, and is suitable for carrier-grade deployment; Wide applicability: Through grid classification (buildings / roads / others), it can provide differentiated positioning support for different scenarios and enhance business adaptability. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart illustrating a user location method based on big data communication, provided as an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] In one exemplary embodiment, such as Figure 1 As shown, a user location method based on big data communication is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, it includes the following steps: Step 101: Obtain MDT sampling point data; the data includes the primary serving cell identifier (ECI), neighboring cell signal characteristics, and cell GPS latitude and longitude.

[0025] Step 102: Perform multi-level cleaning on the MDT sampling point data.

[0026] The MDT cleaning process is as follows: The MDT sampling point data was grouped by user number, sorted by time, and then motion state recognition was performed. Preliminary cleaning was then carried out according to the following rules, the process of which is as follows: 1) Distance cleaning between the sampling point and the cell. First, estimate the latitude and longitude of the cell based on the average latitude and longitude of the sampling points with TA=0. Then, calculate the distance between each sampling point and the average latitude and longitude, and remove sampling points that are too far away and are more than twice the ideal coverage radius of the cell.

[0027] 2) Continuous cleaning. Further grouping of sampling points by user and sorting by time determines the user's motion state, distinguishing between moving and stationary states, and grouping by motion state. For data in the same motion state group, a secondary grouping is performed according to a time continuity threshold (e.g., 5 minutes). If a secondary group has fewer than 3 sampling points, the grouped data is deemed invalid, and the entire secondary group is discarded.

[0028] 3) Motion speed cleaning. Each user's motion state is grouped twice. Based on the distance and time between the start and end points, the average motion speed of the group is calculated, and ultra-high speed motion groups with an average motion speed greater than a threshold (e.g., 120 km / h) are removed.

[0029] 4) Motion direction cleaning. For each motion state of the user, the motion direction of the sampling point is calculated in a secondary group. If the motion direction of the sampling point is opposite to the directions of the two points before and after it, it means that the direction of the sampling point is unidirectional and needs to be cleaned.

[0030] Step 103: Rasterize the multi-level cleaned sampling point data and classify the raster into building raster, road raster, or other raster based on the building layer.

[0031] Specifically, the effective sampling points after cleaning are rasterized into 20-meter grids, assigned a raster ID, and associated with the 20-meter raster layer of the building layer.

[0032] The sampling point grids are classified. If the sampling points of the stationary group fall within the building grid, they are recorded as building grids; the moving group with an average moving speed exceeding the high-speed moving threshold (e.g., 15 kilometers per hour) is recorded as road grids; sampling points that do not belong to building grids or road grids are marked as other grids.

[0033] Step 104: Using the cell as a unit, apply the LOF (Local Outlier Factor) algorithm to the classified sampling point data to detect density anomalies and remove outlier sampling points with LOF values ​​greater than a preset threshold.

[0034] For the MDT data after initial cleaning, further cleaning is required using the Local Outlier Detection (LOF) algorithm. LOF is a density-based anomaly detection algorithm. Its basic idea is: given a sample dataset, if a point in the dataset has a dense neighborhood of points, then that point is likely a normal data point; conversely, if a point is far from its neighbors (i.e., the density of points in its local neighborhood is low), then that point may be an outlier. Therefore, the LOF algorithm can be used to clean out outlier sampling points with anomalous latitude and longitude based on the latitude and longitude of the sampling points. The relevant definitions of the LOF algorithm are as follows: k-distance (k-distance) For an object p, its k-distance(p) is defined as: -p is the distance to the k-th nearest neighbor; -Recorded as: ; k-distance neighborhood The k-distance neighborhood of object p includes all objects whose distance from p does not exceed k-distance(p): ; in: - : The k-distance neighborhood of object p; - : The distance between objects p and q; - k-distance of object p; reachability distance The reachable distance from object p to object o is defined as: ; Local Reachability Density (LRD) Formula for calculating the local reachability density of object p: ; in: - The number of objects in the k-distance neighborhood of object p; - The reachable distance from object p to object o; Local Outlier Factor (LOF) The formula for calculating the local anomaly factor of object p is as follows: ; Abnormal judgment criteria LOF 1: The local density of object p is similar to that of its neighbors, so it is not an outlier; LOF 1: If the local density of object p is lower than that of its neighbors, it may be an outlier. LOF 1: If the local density of object p is higher than that of its neighbors, it may be the core point of a dense region.

[0035] LOF is performed on a cell-by-cell basis, and the algorithm process is as follows: The sampling points are grouped according to ECI, and subsequent calculations are performed according to the ECI grouping; Based on the specified k value (e.g., 5), calculate the distance k-distance from each sampling point p to the nearest Kth sampling point; Calculate the reach-distance between each sampling point p and its k neighbors; Calculate the local reachability density (LRD) for each sampling point p; Calculate the Local Outlier Factor (LOF) for each sampling point p; For a given anomaly threshold T (e.g., 2), filter out anomaly sampling points with LOF greater than T.

[0036] Step 105: Perform feature engineering on the sampled data after anomaly removal and normalize it to obtain the normalized feature vector.

[0037] After removing outlier samples, the neighboring region indices of the samples need to be normalized to meet the modeling requirements of the lightGbm algorithm. Filter out neighboring cells that appear less than 1% of the total number of occurrences; The average field strength of each neighboring cell is taken, and the top 30 neighboring cells are sorted from largest to smallest as training features. Then the neighboring cell configuration is saved, and the neighboring cell features need to be unified for subsequent test data.

[0038] Ultimately, the normalized features of the sampling points include three features (RSRP, RSRQ, and TADV) of the primary serving cell and two features (RSRP and RSRQ) of 30 neighboring cells, for a total of 63 features: scrsrp, scrsrq, TADV, nb_rsrp1, nb_rsrq1, nb_rsrp2, nb_rsrq2, nb_rsrp3, nb_rsrq3, nb_rsrp4, nb_rsrq4, nb_rsrp5, nb_rsrq5, nb_rsrp6, nb_rsrq6, nb_rsrp7, nb_rsrq ...q1, nb_rsrq1, nb_rsrq1, nb_rsrq1, nb_rsrq1, nb_rsrq1, nb_rsrq1, nb_rsrq1, nb_rsrq1, nb_ rq7, nb_rsrp8, nb_rsrq8, nb_rsrp9, nb_rsrq9, nb_rsrp10, nb_rsrq10, nb_rsrp11, nb_rsr q11, nb_rsrp12, nb_rsrq12, nb_rsrp13, nb_rsrq13, nb_rsrp14, nb_rsrq14, nb_rsrp15, nb_ rsrq15, nb_rsrp16, nb_rsrq16, nb_rsrp17, nb_rsrq17, nb_rsrp18, nb_rsrq18, nb_rsrp19 , nb_rsrq19, nb_rsrp20, nb_rsrq20, nb_rsrp21, nb_rsrq21, nb_rsrp22, nb_rsrq22, nb_rsr p23, nb_rsrq23, nb_rsrp24, nb_rsrq24, nb_rsrp25, nb_rsrq25, nb_rsrp26, nb_rsrq26, nb_ rsrp27, nb_rsrq27, nb_rsrp28, nb_rsrq28, nb_rsrp29, nb_rsrq29, nb_rsrp30, nb_rsrq30.

[0039] Step 106: Based on the normalized feature vectors and the corresponding GPS coordinates, a high-precision positioning model is trained using the LightGBM machine learning algorithm, with latitude and longitude as the regression target, mean square error (MSE) as the loss function, and training optimization is performed by combining cross-validation and early stopping strategies.

[0040] After data cleaning and neighbor feature preprocessing, the data is divided into training and validation sets in an 8:2 ratio. Initial model parameters are set, and the training set is used as input for LightGBM model building. The core of LightGBM is an ensemble learning method using gradient boosting decision trees, rather than the traditional hierarchical structure of neural networks. LightGBM employs a serial structure of multiple trees, each outputting a prediction value. The results from all trees are then summed to obtain the final output value. During training, the training of a new tree aims to fit the residual of the previous tree. That is, it doesn't completely trust every residual tree, but rather assumes that each tree only learns a small portion of the truth. By learning from multiple trees, the deficiencies are compensated for, ultimately achieving a fit to the training samples.

[0041] The model's input feature vector contains radio signal measurements from the main cell and neighboring cells, including over sixty features such as rsrp, rsrq, sinr, and tadv. The optimization objective for model training is a standard regression loss function: Mean Aquared Error (MSE). ; in, These are the actual coordinates (longitude or latitude). These are model predictions. The training process utilizes Light GBM's GPU support and GPU histogram algorithm to accelerate training efficiency.

[0042] Step 107: If the positioning accuracy of the high-precision positioning model on the validation set does not reach the preset index, then perform grid search optimization on the LightGBM hyperparameters and retrain the high-precision positioning model using the optimal parameters to obtain the trained high-precision positioning model. During training, to determine the optimal number of iterations and avoid overfitting, the model employs a strategy combining cross-validation and early stopping. Performance on the training set is evaluated using K-Fold cross-validation; training automatically stops if the MSE (Mean Sequence Equation) on the validation set does not improve over twenty consecutive iterations. After training, the model is tested on an independent validation set, and the Geodesic Distance Error (GDI) between the predicted and actual latitude and longitude is calculated. The percentage of this error within a 50-meter or 100-meter range is used as the final positioning accuracy metric.

[0043] When the initial model's localization accuracy fails to meet preset requirements, this application initiates a hyperparameter tuning process based on grid search. This process systematically iterates through different combinations of key parameters to ensure the model achieves optimal performance. The core hyperparameters for tuning include: learning rate, maximum bin size, row sampling fraction, column sampling fraction, and minimum data size in leaf nodes. After determining the optimal parameter combination through grid search, this set of parameters will be used for final model training, and the model file will be saved, completing the entire training process.

[0044] Step 108: Based on the trained high-precision positioning model, output the user positioning result for the real-time MR data.

[0045] Using a pre-trained high-precision positioning model, the system performs positioning calculations on real-time MR data, outputs positioning results, and obtains the user's accurate location.

[0046] Based on the primary serving cell ECI of the sampling points, the corresponding LightGbm localization model file of the cell is found. The sampling points are normalized according to the neighbor cell order in the model file. Then, the LightGbm localization algorithm is called to locate the sampling points and output the localization results.

[0047] Based on the same inventive concept, this application also provides a user positioning device based on communication big data for implementing the user positioning method based on communication big data described above. The solution provided by this device is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more user positioning device embodiments based on communication big data provided below can be found in the limitations of the user positioning method based on communication big data described above, and will not be repeated here.

[0048] In one exemplary embodiment, a user positioning device based on communication big data is provided, comprising: The data acquisition module is used to acquire MDT sampling point data; the data includes the primary serving cell identifier (ECI), neighboring cell signal characteristics, and cell GPS latitude and longitude. A multi-level cleaning module is used to perform multi-level cleaning on the MDT sampling point data; The raster classification module is used to rasterize the multi-level cleaned sampling point data and classify the raster into building raster, road raster, or other raster in combination with the building layer; The anomaly detection module is used to perform density anomaly detection on the classified sampling point data at the cell level, using the LOF local outlier factor algorithm to remove outlier sampling points with LOF values ​​greater than a preset threshold. The normalization module is used to perform feature engineering on the sampled point data after anomaly removal, and to perform normalization processing to obtain the normalized feature vector; The model training module is used to train a high-precision positioning model based on the normalized feature vectors and corresponding GPS coordinates using the LightGBM machine learning algorithm. It uses latitude and longitude as the regression target, the mean square error (MSE) as the loss function, and combines cross-validation and early stopping strategies for training optimization. The parameter optimization module is used to perform grid search tuning on the LightGBM hyperparameters if the positioning accuracy of the high-precision positioning model on the validation set does not meet the preset index, and to retrain the high-precision positioning model using the optimal parameters to obtain a trained high-precision positioning model. The user location result prediction module is used to output user location results based on the trained high-precision positioning model from real-time MR data.

[0049] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 2 As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores user location data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network. When the computer program is executed by the processor, it implements a user location method based on large amounts of communication data.

[0050] Those skilled in the art will understand that Figure 2The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0051] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0052] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0053] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0054] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0055] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0056] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0057] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A user positioning method based on communication big data, characterized in that, The user positioning method based on communication big data comprises the following steps: Obtain MDT sampling point data; the data contains ECI, adjacent area signal characteristics and cell GPS longitude and latitude; Perform multi-level cleaning on the MDT sampling point data; Grid the sampling point data after multi-level cleaning, and classify the grid into building grid, road grid or other grid in combination with the building layer; Perform density anomaly detection on the classified sampling point data by using LOF local outlier factor algorithm, and remove outlier sampling points with LOF value greater than a preset threshold; Perform feature engineering on the sampling point data after anomaly removal, and perform normalization processing to obtain a normalized feature vector; Train a high-precision positioning model based on the normalized feature vector and corresponding GPS coordinates by using LightGBM machine learning algorithm, take longitude and latitude as the regression target, use MSE as the loss function, and combine cross-validation and early stopping strategy for training optimization; If the positioning accuracy of the high-precision positioning model on the validation set does not reach the preset index, perform grid search optimization on the LightGBM hyperparameters, and retrain the high-precision positioning model using the optimal parameters to obtain a trained high-precision positioning model; Output the user positioning result based on the trained high-precision positioning model for real-time MR data.

2. The method of claim 1, wherein, Perform multi-level cleaning on the MDT sampling point data, which comprises the following steps: Estimate the cell center position based on the sampling point with TA=0, and remove the sampling points with a distance greater than 2 times the ideal coverage radius; Group by user identifier and sort by time, identify user motion state, apply time continuity and minimum sampling point number constraint to motion state group, and remove invalid groups; Calculate the average speed of the motion group, and remove the group with average speed greater than the preset high-speed threshold; Remove abnormal sampling points with direction mutation based on motion direction consistency detection.

3. The method of claim 1, wherein, Perform density anomaly detection on the classified sampling point data by using LOF local outlier factor algorithm, and remove outlier sampling points with LOF value greater than a preset threshold, which comprises the following steps: Group the sampling point data according to ECI; Define the k-distance of each sampling point p; Determine the k-distance neighborhood of the sampling point p based on the k-distance of the sampling point p; Calculate the reachable distance of each sampling point p and its k neighbors; Calculate the local reachable density of the sampling point p based on the k-distance neighborhood of the sampling point p and the reachable distance of the sampling point p and its k neighbors; Calculate the local anomaly factor of the sampling point p based on the local reachable density of the sampling point p; If the local anomaly factor is greater than a preset threshold, it is determined as an abnormal sampling point, and the abnormal sampling point is removed.

4. The method of claim 1, wherein, Perform feature engineering on the sampling point data after anomaly removal, and perform normalization processing to obtain a normalized feature vector, which comprises the following steps: Filter the adjacent areas with an appearance frequency lower than 1%, sort the remaining adjacent areas by average field strength to take the top 30, and construct a unified feature vector containing the RSRP, RSRQ, tadv of the main cell and the RSRP and RSRQ of the 30 adjacent areas.

5. The method of claim 1, wherein, The expression of the loss function is as follows: ； wherein, is the true longitude or latitude, is the model prediction, is the number of sampling points involved in the model training.

6. The method of claim 1, wherein, The LightGBM hyperparameters include: learning rate, maximum number of buckets max_bin, row sampling ratio bagging_fraction, column sampling ratio feature_fraction, and minimum data amount in leaf min_data_in_leaf. 7.A user positioning device based on communication big data, characterized in that, The user positioning device based on communication big data comprises: A data acquisition module is configured to acquire MDT sampling point data; the data comprises ECI (Elevation Cell Identifier), adjacent cell signal characteristics, and cell GPS longitude and latitude; A multi-level cleaning module is configured to perform multi-level cleaning on the MDT sampling point data; A grid classification module is configured to grid the sampling point data after multi-level cleaning and classify the grid into a building grid, a road grid, or other grids in combination with a building layer; An anomaly detection module is configured to perform density anomaly detection on the classified sampling point data by using a LOF (Local Outlier Factor) algorithm in a cell unit, and eliminate outlier sampling points with a LOF value greater than a preset threshold; A normalization module is configured to perform feature engineering on the sampling point data after anomaly elimination, perform normalization processing, and obtain a normalized feature vector; A model training module is configured to train a high-precision positioning model by using a LightGBM machine learning algorithm based on the normalized feature vector and corresponding GPS coordinates, take longitude and latitude as a regression target, use MSE (Mean Squared Error) as a loss function, and perform training optimization in combination with a cross-validation and early stopping strategy; A parameter optimization module is configured to perform grid search optimization on LightGBM hyperparameters if the positioning accuracy of the high-precision positioning model on a validation set does not reach a preset index, retrain the high-precision positioning model by using optimal parameters, and obtain a trained high-precision positioning model; A user positioning result prediction module is configured to output a user positioning result based on the trained high-precision positioning model for real-time MR data.

8. A computer device comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that the processor executes the computer program to implement the user positioning method based on communication big data according to any one of claims 1-6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the user positioning method based on communication big data according to any one of claims 1-6.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the user positioning method based on communication big data according to any one of claims 1-6.

Images