Mobile communication big data positioning method and device, equipment and storage medium

Abstract

The invention discloses a mobile communication big data positioning method, apparatus and device, and a storage medium. The positioning method comprises the steps of obtaining data sent to a base station by a mobile terminal to form a measurement report; performing preprocessing and data filling on the measurement report data; establishing a local coordinate system taking a main service base station as an original point, and finishing positioning from a high-dimensional signal space to a two-dimensional geographic space by combining a single feature in measurement report data with an improved deep learning model; and the positioning precision is evaluated by using two indexes of the mean absolute error and the root-mean-square error, so that high-precision positioning of the mobile terminal is realized. According to the method, the missing data is effectively filled, the mapping between the high-dimensional signal space and the two-dimensional geographic coordinates is realized, and the positioning precision of the terminal is enhanced. A missing value filling method taking hierarchical clustering as a core is provided, and on this basis, accurate positioning of the mobile terminal under a small sample condition is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of positioning and navigation, and specifically relates to a method, device, equipment and storage medium for positioning mobile communication big data. Background Art

[0002] With the continuous development of mobile cellular networks, location-based services (LBS) have attracted the attention of many practitioners and researchers. Compared with the positioning services provided by GPS, the data-driven mobile cellular network side positioning technology has more advantages. All mobile terminals can receive cellular signals, which provides a ubiquitous global positioning service, and this positioning method can be carried out on almost any mobile terminal without consuming extra energy. However, due to a large number of missing values of neighbor cell dimension level values, too few splines containing location information, etc., the positioning performance is not good, mainly manifested as large computational complexity and low positioning accuracy.

[0003] Therefore, it is crucial to develop a fast, reliable and robust method to accurately locate mobile terminals. Due to the differences of mobile devices and complex communication environments, existing research cannot effectively repair a large number of missing values in MR (measurement reports), and most of the observation data in LTE commercial networks only contains signal strength information from serving cells. The existence of this problem greatly limits the positioning accuracy of cellular networks. In addition, due to reasons such as privacy, a large amount of GPS data information in MR data is missing, and only a small amount of data contains coordinates, which makes the data-driven machine learning single-point positioning method have certain limitations. How to complete the positioning of all terminals in the serving cell with a small number of samples is the current key concern. Among them, during each session and call, measurement data related to the mobile device is collected by the network. The present invention refers to these measurements as measurement reports. Summary of the Invention

[0004] Aiming at the above problems existing in the prior art, the purpose of the present invention is to provide a method, device, equipment and storage medium for positioning mobile communication big data, which can effectively fill in missing data, reduce the data missing rate to less than 3%, solve the data dimensionality disaster, and improve the Transformer, realize the mapping between the high-dimensional signal space and the two-dimensional geographical coordinates, and enhance the positioning accuracy of the terminal. Taking the missing values in MR as the starting point, based on the mobile user portrait technology, a missing value filling method with hierarchical clustering as the core is proposed. On this basis, accurate positioning of mobile terminals under small sample conditions is realized.

[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is: A method for positioning mobile communication big data, comprising: Obtain the data sent by the mobile terminal to the base station to form a measurement report; Preprocess and fill in the data of the measurement report; Establish a local coordinate system with the serving base station as the origin, and use a single feature in the measurement report data combined with an improved deep learning model to complete the positioning from the high-dimensional signal space to the two-dimensional geographical space; Use two indicators, mean absolute error and root mean square error, to evaluate the positioning accuracy, and achieve high-precision positioning of the mobile terminal.

[0006] As a further improvement of the above technical solution: When completing the positioning from the high-dimensional signal space to the two-dimensional geographical space, construct a mapping relationship from the high-dimensional signal space to the two-dimensional geographical space, use the path loss and the local coordinate values of the cell as a triple positioning relationship model, and combine the improved deep learning model to complete the precise positioning of the terminal.

[0007] The preprocessing of the measurement report data includes data standardization and data compression to reduce the data scale and computational complexity; When filling in the data, introduce a hierarchical clustering algorithm to perform self-filling and mutual-filling in the same-class cluster splines, and complete the filling of the mobile user portrait through multiple iterations; The single feature used in the measurement report data is the reference signal received power value.

[0008] After one clustering, self-filling fills the splines with the same user ID in the same class, and after one clustering, mutual-filling fills the splines with different user IDs in one class.

[0009] When establishing the local coordinate system, convert the coordinates of all base stations and splines, calculate the Gauss plane coordinates from the geodetic coordinates, set the coordinates of the serving base station as the origin in the two-dimensional plane coordinate system, calculate the distance and angle between the sample point and the origin, and obtain the two-dimensional plane coordinates of the sample and the base station in the local coordinate system to complete the construction of the local coordinate system.

[0010] The terminal receives the signal values from m cells and reports them. Use m cells to construct a high-dimensional signal space, characterize the spline features through the reported multiple reference signal received power values, and find m mapping relationships.

[0011] The deep learning model used is the Transformer model. The Transformer architecture stacks multiple identical segments of the encoder and decoder together. Both the encoder and decoder modules are constructed using multi-head self-attention units and position feed-forward networks.

[0012] The positioning method includes the following steps: S1. Standardize and compress the measurement report data to reduce the data scale and computational complexity; S2. Based on the mobile user portrait filling method, handle the problem of a large number of missing data in the level value data of multiple neighboring cell dimensions in the measurement report data; S3. Introduce the Chebyshev distance hierarchical clustering algorithm to perform self-filling and mutual filling in the same cluster splines, and iterate multiple times to complete the mobile user portrait filling; S4. Unify the coordinates of the terminal user and the base station, and establish a local coordinate system with the serving base station as the origin; S5. Use the reference signal received power value in the measurement report and the base station parameters related to the location to construct a mapping relationship from the high-dimensional signal space to the two-dimensional geographical space; S6. Based on the positioning relationship model with path loss and cell local coordinate values as triples, combined with the improved Transformer model, complete the precise positioning of the terminal.

[0013] Step S2 includes the following steps: S21. Self-filling: Perform self-filling on the splines of the same user, and fill in the non-zero average value; S22. For the remaining missing values after self-filling, temporarily fill them with an artificially set value, and then return to modify the filled values in this part during the subsequent clustering operation.

[0014] Step S3 includes the following steps: S31. In the initial stage, artificially select a threshold for clustering; S32. Perform self-filling: Fill the splines with the same user ID in the same cluster; S33. Perform mutual filling: Fill the splines with different user IDs in the same cluster; S34. Enlarge the threshold by 1.05 times and perform hierarchical clustering again, repeating the iteration until the termination condition is reached.

[0015] A mobile communication big data positioning device includes: A mobile terminal for uploading data including current location information to the base station; A base station providing wireless coverage for connecting the mobile terminal to the Internet; A radio access network element management system to which the measurement report data of the mobile terminal is reported for storage; A server that obtains multiple measurement report data from the radio access network element management system and parses the multiple measurement report data.

[0016] A positioning device includes a memory and a processor. A computer program is stored in the memory, and the processor is configured to execute the computer program to implement the positioning method.

[0017] A storage medium carries at least one computer program. When the computer program is executed by an electronic device, the electronic device implements the positioning method described above.

[0018] The beneficial effects of the present invention are as follows: (1) Effectively fill in missing data, reducing the data missing rate to within 3%, solving the data dimensionality disaster, improving the Transformer, realizing the mapping between the high-dimensional signal space and the two-dimensional geographical coordinates, and enhancing the positioning accuracy of the terminal.

[0019] (2) Taking the missing values in the MR as the starting point, based on the mobile user profiling technology, a missing value filling method with hierarchical clustering as the core is proposed. On this basis, precise positioning of mobile terminals under small sample conditions is realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is a principle block diagram of the mobile terminal positioning method based on measurement reports provided by the present invention; Figure 2 is a visualization picture of MR data provided by the present invention; Figure 3 is an iterative flowchart of clustering filling provided by the present invention; Figure 4 is a schematic diagram of the local coordinate system provided by the present invention; Figure 5 is a structural diagram of the Transformer encoding layer provided by the present invention; Figure 6 is a comparison picture of the positioning error and the performance of other algorithms provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] The following will describe the specific embodiments of the present invention in detail with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for the purpose of illustration and explanation of the present invention, and are not intended to limit the present invention.

[0022] For ease of description, spatial relative terms such as "above", "over", "on the upper surface", "upper" etc. can be used here to describe the spatial positional relationship of a device or feature shown in the figure with other devices or features. It should be understood that spatial relative terms are intended to encompass different orientations in use or operation in addition to the orientation depicted in the figure for the device. For example, if the device in the attached drawing is inverted, a device described as "above or over other devices or structures" will then be positioned "below or under other devices or structures". Thus, the exemplary term "above" can include both the orientations of "above" and "below". The device can also be positioned in other different ways (rotated 90 degrees or in other orientations), and corresponding interpretations are made for the spatial relative descriptions used here.

[0023] A mobile communication big data positioning method, as Figure 1 shown, includes the following steps: Step S1: Perform data standardization and data compression on measurement report (MR) data to reduce the data scale and lower the computational complexity.

[0024] The measurement report contains status information when the terminal is connected to the network, such as signal strength, connection time, base station parameters, etc.

[0025] The measurement report (MR) can specifically include the following categories: the sample measurement time (TimeStamp) of the sample data measurement object, the unique identifier (MmeUeSIapID) of the UE (user) at the S1 interface on the MME (network node) side, the MME group identifier (MmeGroupID), the MME code (Mmecode), the reference signal received power of the primary serving cell (MR.LteScRSRP), the reference signal received power of the neighboring cell (MR.LteNcRSRP), the time advance of the primary serving cell (MR.LteScTadv), and the carrier number of the primary serving cell (MR.LteScEarfcn).

[0026] The standardization, compression, and data label selection of the original MR data mainly include the extraction of neighboring cells for data structuring, and data compression for data redundancy problems, enabling the data to be statistically analyzed from a unified perspective, and the selection of data feature labels to prepare for subsequent model positioning.

[0027] For the extraction of neighboring cells, due to the stacking of neighboring cells in the original data, when processing this part of the data, the cell is used as the processing unit to extract all the neighboring cells received by the terminal user under this serving cell and perform data standardization operations.

[0028] Regarding the data compression, due to reasons such as the measurement reporting mechanism, the same terminal will report multiple splines within 1s. Usually, only one of them has neighbor cell measurement values reported, and the measurement information of the remaining samples is empty, resulting in a large data scale. In this case, it is necessary to perform data compression while ensuring that valid data values are not discarded.

[0029] Regarding the selection of data feature labels, the present invention fills data guided by improving the positioning accuracy of user terminals, and uses the geographical location coordinates of users as the labels to be predicted.

[0030] In this embodiment, when extracting neighbor cells, in the lte_ncell_eNodeBid_Cellid (neighbor cell identifier) column of the splines reported by a single user at a certain moment, the neighbor cells detected by the terminal at that moment will be reported, and there will be corresponding cell RSRP values (reference signal receiving power) in the subsequent lte_ncell_rsrp (neighbor cell reference signal receiving power value) column. Extract all the cells in the lte_ncell_eNodeBid_Cellid column and the RSRP values of the corresponding cells in the lte_ncell_rsrp column. For the sample with Cellid of 506, the reported neighbor cells are CELL_1019 and CELL_626, and the corresponding RSRP values of these two neighbor cells are -82 and -87. In this operation, CELL_1019 and CELL_626 are used as the column names, and the corresponding RSRP values are placed under this sample.

[0031] Z-score normalization processing is performed on the unique identifiers of users and cells, and its conversion function is as follows: ; where x is the reference signal receiving power RSRP. is the mean of the original data, is the standard deviation. The processed data is visualized as Figure 2 shown. The z-axis is the normalized user ID, the y-axis is the normalized cell ID, and the x-axis is the RSRP value received by the user in the corresponding cell. It can be seen from the figure that the RSRP values are concentrated in a certain numerical range.

[0032] Taking the cell CELL_992 as an example, considering the limited coverage of the base station and the influence of noise, 700m is selected as the engineering parameter. The neighboring cells within a radius of 700m centered on the serving cell are selected as the feature dimensions, and the number of neighboring cells within a radius of 700m is 70. Taking the intersection of the reported neighboring cells and the neighboring cells within the radius, there are 24 neighboring cells in total. Taking the cell CELL_1019 as an example, all the neighboring cells reported by users in CELL_1019 within 24 hours are recorded as set A, and the cells within the coverage of the base station to which CELL_1019 belongs are recorded as set B. Take the intersection of set A and set B. Set A: There are 112 neighboring cells reported by users in CELL_1019 in total. Set B: Taking the coordinates of the base station to which CELL_1019 belongs as the coordinate origin, 31 cells within a radius of 700m are selected. Taking the intersection of A and B, there are 16 cells in total.

[0033] Step S2: Based on the mobile user portrait filling method, handle the problem of a large number of missing values in the level value data of multiple neighboring cell dimensions in the MR data.

[0034] This step is the filling before hierarchical clustering and can be called preliminary filling.

[0035] The main reasons for data missing include: physical failure of the data receiving end, data loss during transmission, and improper acquisition process; the two stages of data missing refer to the batch or unit stage and the data item stage.

[0036] Mobile user portrait filling requires selecting user tags. Combining the communication mechanism with the reported MR data of users, the feature tags are finally selected to accurately reflect user characteristics.

[0037] In this step, the filling method adopted includes two steps: Step S21: Self-filling. Specifically, each spline reported by the UE only reports a few or one neighboring cell. If directly clustering and using different clustering thresholds under the Euclidean distance for hierarchical clustering, multiple splines reported by the same user at the same geographical location will be clustered into different classes. Therefore, in this step, before clustering, the splines of the same UE are self-filled with non-zero average values.

[0038] Step S22: For the remaining missing values after self-filling, a manually set value is temporarily used for filling. When performing clustering operations later, the filled values in this part are returned for modification. In this step, -145 is temporarily used for filling.

[0039] Step S3: Introduce the Chebyshev distance hierarchical clustering algorithm to perform self-filling and mutual filling in the splines of the same class cluster, and iterate multiple times to complete the mobile user portrait filling.

[0040] The MR data filling uses a hierarchical clustering iterative algorithm, and the filling includes self-filling and mutual filling. After one clustering, for self-filling, the splines with the same user ID in the same class are filled, and for mutual filling, the splines with different user IDs in one class are filled, both of which are non-zero mean filling.

[0041] In hierarchical clustering, considering that MR data may fail in the high-dimensional signal space, various distance metrics are considered, and finally the Chebyshev distance is selected for clustering, and the Calinski-Harabasz (CH) index is used to measure the clustering quality. The distance between sample points in the present invention is calculated in the high-dimensional signal space. Suppose there are two sets of data in the high-dimensional signal space , .

[0042] The Chebyshev distance d and CH can be described by the following expressions: ; ; is the within-class scatter, represents the number of MR samples, is the number of clusters obtained after hierarchical clustering, is the covariance matrix between the classes of each class, represents the trace of the matrix. The larger the CH value, the more dispersed the clusters tend to be.

[0043] The hierarchical clustering algorithm is used to accurately complete the characterization of mobile user features, and the specific steps are as follows: 1) In the initial stage, a smaller threshold is selected for clustering.

[0044] 2) Perform self-filling: Fill the splines with the same user ID in the same cluster.

[0045] 3) Perform mutual filling: Fill the splines with different user IDs in the same cluster.

[0046] 4) Enlarge the threshold by 1.05 times and perform hierarchical clustering again, repeating iteratively until the termination condition is reached.

[0047] In this embodiment, the scipy (scipy is an open-source Python algorithm library and mathematical toolkit) library is used for hierarchical clustering. For this hierarchical clustering, the main parameters to be selected are: Linkage criterion: Select Single (a linkage criterion in the scipy library), and use the minimum distance between samples in different classes as the criterion; Distance threshold: If the linkage distance is greater than this threshold, these two classes will not be merged; Measurement standard for calculating connections: Select "chebyshev" (Chebyshev distance); In the initial stage, a small number of splines are experimented with, and using some industrial parameters, the clustering threshold is debugged, and the CH (Calinski-Harabasz) index is used to measure the clustering quality. The CH index is used when the true labels of the data are unknown to evaluate the goodness of the clustering. The higher the score of the CH index, the better the clustering effect. Specifically, this index is the ratio of the between-class scatter to the within-class scatter, and the calculation formula is as follows: ; After one clustering, the splines with the same user ID in the same class are filled, and here the non-zero mean of all RSRP values under the same neighbor cell dimension (feature) is filled.

[0048] After one clustering, the splines with different user IDs in a class are filled. The mutual filling is carried out after the sample self-filling. Combining the clustering results and the self-filling data, non-zero mean filling is also carried out.

[0049] The threshold is magnified by 1.05 times, and hierarchical clustering is performed again, and the missing values are filled. This is repeated iteratively until the termination condition is reached, that is, whether the threshold reaches 24. The flowchart of this algorithm is as Figure 3 shown.

[0050] In the result of the initial hierarchical clustering using the Chebyshev distance, when the distance threshold is 35 dB, all samples can be clustered into one class. After the clustering and filling, it is found that by continuously performing iterative filling, when the threshold reaches 24 dB, most of the missing data can be filled.

[0051] Step S4: Unify the coordinates of the terminal user and the base station, and establish a local coordinate system with the serving base station as the origin.

[0052] The local coordinate system is independent of the global coordinate system and is a standard coordinate system established within the service range of the serving base station. Moreover, the cell association of each base station can be carried out through the Gauss plane coordinates.

[0053] The terminal will receive the signal values from multiple cells and report them. Through the mapping relationship between m cells (including the main cell) and the terminal, the position of the jth terminal is calculated: ; where, is the coordinate of the base station corresponding to the i-th cell received by the terminal within the coverage range of the current serving cell, that is, the true coordinate of the terminal. is the th cell and the The path loss between terminals.

[0054] In the above formula, represents the mapping function.

[0055] In this embodiment, the local coordinate system is as Figure 4 shown. This local coordinate system includes the x-axis and the y-axis. In the figure, the red represents the base station, and the blue represents the spline. Among them, both axes are constructed in meters, and the terminal coordinates and the base station coordinates are both GPS coordinates.

[0056] Convert the coordinates of all base stations and splines from geodetic coordinates , and calculate the Gauss plane coordinates The forward Gauss formula is as follows: ; ; In the formula is the meridian arc length from the equator to the projection point; is the radius of the prime vertical; ; is the longitude difference; is the longitude of the central meridian.

[0057] , and used below are the first eccentricity, the semi-major axis, and the semi-minor axis, which are common geometric parameters in the earth ellipsoid, respectively.

[0058] is the square term of the second eccentricity.

[0059] The calculation formula of is as follows: Among them: ; ; After converting to the Gauss plane coordinates, set the coordinates of the serving base station as the origin (0, 0) in the two-dimensional plane coordinate system, calculate the distance and angle between the sample point and the origin, and obtain the two-dimensional plane coordinates of the sample and the base station in the local coordinate system, thus completing the construction of the local coordinate system.

[0060] Step S5: Use the RSRP (Reference Signal Received Power) value in MR and the base station parameters related to the location to construct the mapping relationship from the high-dimensional signal space to the two-dimensional geographical space.

[0061] In this step, the mapping between the high-dimensional signal space and the two-dimensional geographical coordinates is realized by combining the improved neural network model, which improves the positioning accuracy of the terminal.

[0062] In this embodiment, given the coordinates and corresponding transmit powers of the base stations, the path loss of the i-th under the current serving cell is described as: ; where is the path loss, represents the reference signal transmit power of the base station corresponding to the -th cell, represents the reference signal received power value received by the terminal from the -th cell. On the premise of ensuring the positioning accuracy, a relationship between the signal propagation distance and the corresponding path loss for the -th signal under the current cell coverage is proposed and described as follows: ; where is the coordinate of the base station corresponding to the -th cell received by the terminal within the coverage of the current serving cell, i.e., the true coordinate of the terminal.

[0063] ; is the path loss between the -th cell and the -th terminal. By calculating the triplets of these cells, they are used as the input for the deep learning model (i.e., the Transformer model). Among them, the triplet is the basic unit of knowledge representation, in the form of (subject, relation, object), which is used to describe the association between two entities (or an entity and an attribute).

[0064] After data cleaning and feature selection, cells are used to construct the high-dimensional signal space. By accurately characterizing the spline features with multiple reported RSRP values, the mapping relationships are found, and finally the position of the terminal is accurately located. The objective function is defined as follows: ; Minimize the mapping errors of the cells of the

[0065] Step S6: Propose a positioning relationship model with path loss and cell local coordinate values as triplets, and combine the improved Transformer model to complete the precise positioning of the terminal.

[0066] The deep learning-based wireless positioning method uses an improved Transformer model. When modeling, only the Encoder part of the original model is used, and the Decoder part is directly changed to a fully connected layer. Finally, the tensor is projected into the form of [batch_size, output_len].

[0067] In this embodiment, the Transformer architecture stacks N identical segments of the following two units together: namely, the encoder and the decoder.

[0068] The encoder and decoder modules are mainly constructed using the multi-head self-attention unit and the position feed-forward network. The first sub-layer is the multi-head self-attention module, which allows the encoder to focus on the most relevant information in the sequence. The output vector of the MHSA (multi-head self-attention) is added to the position embedding tensor through a residual connection and further processed through a normalization layer, as shown in the following formula: ; ; represents layer normalization, which feeds the normalized residual MHSA output vector into the position feed-forward network and normalizes it to produce the output of the encoder, as follows: ; ; FFN represents the point-wise feed-forward network. represents the output of the FFN module. represents the encoder. The Transformer encoder is constructed by repeatedly stacking N sub-layers of MHSA and FFN, as well as residual connections and normalization. The Transformer encoding layer structure is as follows Figure 5 shown, Figure 5 in which, , , , are the input signals of each encoder respectively, , , , are the output signals of each encoder respectively.

[0069] Similar to the encoder, the decoder is also composed of N identical blocks, which are connected by three sub-layers. The first sub-layer is the masked Multi-Head Self-Attention, whose operation is similar to the MHSA module discussed above, except that future positions in the sequence are masked.

[0070] The triple features are normalized, and then the data is fed into the model. In this experiment, Min-Max normalization is used to process the data, which maps the data values to the range [0,1]. The processing function is as follows: ; where is the minimum value of the feature (the triple feature in the current cell dimension), and is the maximum value of this feature.

[0071] Experiments were conducted comparing traditional SVR (Support Vector Regression), KNN (K-Nearest Neighbor), a regression localization reconstruction algorithm improved based on AdaBoost (Adaptive Boosting), and the CRL (Conditional Random Field) algorithm. Using the triple feature as the input, the algorithm comparison results are as Figure 6 shown. Figure 6 In

[0072] The missing value filling method based on hierarchical clustering in the present invention reduces the data in the original ultra-high dimensional signal space to a lower level, avoiding the curse of dimensionality of the data; introduces the Chebyshev distance as a measure of the inter-cluster distance, and based on the hierarchical clustering result, fills the data within the cluster, and iteratively magnifies the clustering threshold successively to complete the filling of the missing data; thus, in the local coordinate system, based on data-driven, using a single feature of RSRP in the MR data and a modified Transformer model, the positioning from the high-dimensional signal space to the two-dimensional geographical space is completed. Therefore, the present invention improves the positioning ability of the mobile terminal.

[0073] Finally, it is necessary to state here that the above embodiments are only used to further illustrate the technical solutions of the present invention, and should not be construed as limiting the protection scope of the present invention. Some non-essential improvements and adjustments made by those skilled in the art based on the above content of the present invention all fall within the protection scope of the present invention.

Claims

1. A method for positioning mobile communication big data, characterized in that, including: Obtain the data sent by the mobile terminal to the base station to form a measurement report; Preprocess and fill in the data of the measurement report; Establish a local coordinate system with the serving base station as the origin, and use a single feature in the measurement report data combined with an improved deep learning model to complete the positioning from the high-dimensional signal space to the two-dimensional geographical space; Use two indicators, the mean absolute error and the root mean square error, to evaluate the positioning accuracy, and achieve high-precision positioning of the mobile terminal.

2. The positioning method according to claim 1, characterized in that: When completing the positioning from the high-dimensional signal space to the two-dimensional geographical space, construct a mapping relationship from the high-dimensional signal space to the two-dimensional geographical space, use the path loss and the local coordinate value of the cell as a positioning relationship model of a triple, and combine the improved deep learning model to complete the precise positioning of the terminal.

3. The positioning method according to claim 2, wherein: Preprocessing the measurement report data includes performing data standardization and data compression to reduce the data scale and computational complexity; When filling in the data, introduce a hierarchical clustering algorithm to perform self-filling and mutual-filling in the same-class cluster splines, and iterate multiple times to complete the filling of the mobile user portrait; The single feature used in the measurement report data is the reference signal received power value.

4. The positioning method according to claim 3, characterized in that: In self-filling, after one clustering, the splines with the same user ID in the same class are filled, and in mutual-filling, after one clustering, the splines with different user IDs in one class are filled.

5. The positioning method according to claim 1, wherein: When establishing the local coordinate system, convert the coordinates of all base stations and splines, calculate the Gauss plane coordinates from the geodetic coordinates, set the coordinates of the serving base station as the origin in the two-dimensional plane coordinate system, calculate the distance and angle between the sample point and the origin, and obtain the two-dimensional plane coordinates of the sample and the base station in the local coordinate system to complete the construction of the local coordinate system.

6. The positioning method according to claim 5, wherein: The terminal receives the signal values from m cells and reports them, constructs a high-dimensional signal space with m cells, depicts the spline features through the reported multiple reference signal received power values, and finds m mapping relationships.

7. The positioning method according to claim 3, wherein: The deep learning model adopted is the Transformer model. The Transformer architecture stacks multiple identical segments of the encoder and the decoder together, and both the encoder and decoder modules are constructed using multi-head self-attention units and position feed-forward networks.

8. A mobile communication big data positioning device, characterized in that, including: A mobile terminal for uploading data including current location information to the base station; A base station providing wireless coverage for connecting the mobile terminal to the Internet; A radio access network element management system to which the measurement report data of the mobile terminal is reported for storage; A server that obtains multiple measurement report data from the radio access network element management system and analyzes the multiple measurement report data.

9. A positioning device, characterized in that, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is used to execute the computer program to implement the positioning method according to any one of claims 1 to 7.

10. A storage medium carrying at least one computer program, characterized in that, When the computer program is executed by an electronic device, the electronic device implements the positioning method according to any one of claims 1 to 7.

Images