Method and apparatus for predicting positioning model error

Abstract

The present disclosure relates to a method and apparatus for predicting positioning model errors. A method for training an error prediction model is provided, wherein the error prediction model is used to predict the error between the predicted positioning of data to be predicted using a positioning prediction model and the actual positioning, wherein the positioning prediction model is trained using a training set. The method comprises: constructing an error prediction feature, wherein the error prediction feature is associated with the training set; constructing an error prediction label, wherein the error prediction label comprises: using the positioning prediction model to predict the positioning of data in a validation set, and calculating the error between the predicted positioning of the data in the validation set and the actual positioning, and obtaining an error prediction label based on the error; and using the error prediction label as supervision, using the error prediction feature to train the error prediction model.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence, and more specifically to a method and device for predicting positioning model errors. Background Art

[0002] As location services become increasingly necessary and widespread, and machine learning algorithms and big data technologies mature, the use of machine learning algorithms for wireless network positioning has gradually become one of the key technologies for wireless network positioning, both in academia and in the design of commercial Internet solutions.

[0003] The application scenarios of location-based services are often closely related to positioning accuracy, so it is necessary to fully utilize positioning data of various accuracy levels. For example, data with high positioning accuracy can be selected and added to the training set to increase its richness. For example, for problems with limited training set coverage, semi-supervised learning can be used for iterative learning.

[0004] Currently, widely used positioning models use the mathematical principle of regression and machine learning algorithms to perform positioning based on big data. After training, they can be used to predict data without location labels. Because the predicted data lacks location labels and the problem to be solved is a regression task, it is impossible to characterize the prediction error for each sample to be predicted.

[0005] The current mainstream method is to use the error level of the entire validation set in the pre-divided validation set to approximately characterize the overall error of the sample to be predicted, such as Figure 1 This characterization has at least the following flaws:

[0006] 1. The characterization is valid only when the validation set and the samples to be predicted follow the same distribution. However, in many practical application scenarios, due to various issues such as sampling methods, the distributions of the two are often different.

[0007] 2. Indicators such as mean absolute error (MAE), mean square error (MSE), and root mean square error (RMSE) can only evaluate the accuracy and versatility of the model as a whole based on its performance on the test set. They cannot provide an interval or characterization of the prediction error for each sample to be predicted.

[0008] Currently, existing network MR (Measurement Report) data suffers from varying positioning errors due to factors such as a lack of AGPS (Assisted Global Positioning System) tags and uneven sampling coverage in some cells. Furthermore, MR data is feature-rich, and its wireless positioning accuracy exhibits nonlinear correlations with multiple features. Therefore, predicting wireless network positioning errors is both necessary and feasible. Summary of the Invention

[0009] A brief overview of the present disclosure is provided below to provide a basic understanding of some aspects of the present disclosure. However, it should be understood that this overview is not an exhaustive overview of the present disclosure. It is not intended to identify key or important parts of the present disclosure, nor is it intended to limit the scope of the present disclosure. Its purpose is simply to present certain concepts of the present disclosure in a simplified form as a prelude to the more detailed description that will be given later.

[0010] According to a first aspect of the present disclosure, a method for training an error prediction model is provided, wherein the error prediction model is used to predict the error between the predicted positioning of data to be predicted obtained by using a positioning prediction model and the actual positioning, and the positioning prediction model is obtained by training using a training set. The method comprises: constructing an error prediction feature, wherein the error prediction feature is associated with the sampling density and data quality of the training set; constructing an error prediction label, comprising: using the positioning prediction model to predict the positioning of data in a validation set, and calculating the error between the predicted positioning of the data in the validation set and the actual positioning, and obtaining an error prediction label based on the error; and using the error prediction label as supervision to train the error prediction model using the error prediction feature.

[0011] According to the second aspect of the present disclosure, a method for predicting the error of a positioning prediction model is provided, which is used to predict the error between the predicted positioning of data to be predicted obtained using the positioning prediction model and the actual positioning. The method comprises: training the error prediction model using the method described in the first aspect of the present disclosure; and predicting the error of the data to be predicted using the error prediction model.

[0012] According to a third aspect of the present disclosure, a device for training an error prediction model is provided, wherein the error prediction model is used to predict the error between the predicted positioning of data to be predicted obtained by using a positioning prediction model and the actual positioning, and the positioning prediction model is obtained by training using a training set. The device includes: a feature construction module, wherein the feature construction module is configured to construct an error prediction feature, and the error prediction feature is associated with the sampling density and data quality of the training set; a label construction module, wherein the label construction module is configured to construct an error prediction label as follows: using the positioning prediction model to predict the positioning of data in a validation set, and calculating the error between the predicted positioning of the data in the validation set and the actual positioning, and obtaining an error prediction label based on the error; and a training module, wherein the training module is configured to train the error prediction model using the error prediction feature with the error prediction label as supervision.

[0013] According to the fourth aspect of the present disclosure, a device for predicting the error of a positioning prediction model is provided, which is used to predict the error between the predicted positioning of the data to be predicted obtained using the positioning prediction model and the actual positioning. The device includes: the device according to the third aspect of the present disclosure; a prediction module, which uses the error prediction model obtained by the device according to the third aspect of the present disclosure to predict the error of the data to be predicted.

[0014] According to a fifth aspect of the present disclosure, a non-transitory computer-readable storage medium is provided, on which a program is stored, characterized in that when the program is executed by a computer, the computer is caused to execute the method according to the first aspect of the present disclosure.

[0015] According to a sixth aspect of the present disclosure, a non-transitory computer-readable storage medium is provided, on which a program is stored, characterized in that when the program is executed by a computer, the computer is caused to execute the method according to the second aspect of the present disclosure.

[0016] According to the seventh aspect of the present disclosure, a device for training an error prediction model is provided, comprising a memory and a processor, wherein the memory is communicatively coupled to the processor, and a program is stored in the memory, and when the program is executed by the processor, the processor executes the method described in the first aspect of the present disclosure.

[0017] According to an eighth aspect of the present disclosure, a device for predicting the error of a positioning prediction model is provided, comprising a memory and a processor, wherein the memory is communicatively coupled to the processor, and a program is stored in the memory, and when the program is executed by the processor, the processor executes the method described in the second aspect of the present disclosure.

[0018] According to a ninth aspect of the present disclosure, a computer program product is provided, comprising a computer program, which implements the method according to the first aspect of the present disclosure when executed by a processor.

[0019] According to a tenth aspect of the present disclosure, a computer program product is provided, comprising a computer program, which implements the method according to the second aspect of the present disclosure when executed by a processor.

[0020] The method and device proposed in this disclosure provide a characterization of positioning error. Compared with the traditional performance evaluation of the model on the validation set based on indicators such as MAE and MSE, the error prediction effect can be improved when the validation set and the data to be predicted have different distributions, providing strong support for the subsequent evaluation and application of wireless network positioning algorithm models.

[0021] Other features and advantages of the present disclosure will become more apparent from the following detailed description of exemplary embodiments of the present disclosure with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0023] The present disclosure may be more clearly understood from the following detailed description with reference to the accompanying drawings, in which:

[0024] Figure 1 A schematic diagram showing a positioning model error prediction method in the prior art is shown;

[0025] Figure 2 A schematic diagram illustrating a method for constructing error prediction features and error prediction according to an embodiment of the present disclosure is shown;

[0026] Figure 3 A schematic diagram of an error prediction system according to an embodiment of the present disclosure is shown;

[0027] Figure 4 A schematic diagram of system data partitioning iteration is shown;

[0028] Figure 5 A flowchart of a positioning model error prediction method according to an embodiment of the present disclosure is shown;

[0029] Figure 6 An exemplary configuration of a computing device that can implement embodiments according to the present disclosure is shown. DETAILED DESCRIPTION

[0030] The following detailed description is made with reference to the accompanying drawings, and the following detailed description is provided to assist in a comprehensive understanding of the various exemplary embodiments of the present disclosure. The following description includes various details to assist in understanding, but these details are considered to be examples only and are not intended to limit the present disclosure, which is defined by the appended claims and their equivalents. The words and phrases used in the following description are intended only to enable a clear and consistent understanding of the present disclosure. In addition, descriptions of well-known structures, functions, and configurations may be omitted for clarity and brevity. Those skilled in the art will recognize that various changes and modifications may be made to the examples described herein without departing from the spirit and scope of the present disclosure.

[0031] Figure 2 A schematic diagram of a method for constructing error prediction features and error prediction according to an embodiment of the present disclosure is shown.

[0032] like Figure 2As shown, in an embodiment according to the present disclosure, feature engineering is performed on data 201 with a positioning tag to extract positioning features 202, wherein the positioning tag can indicate the actual positioning of the data, for example, the positioning tag can be an AGPS tag; the data is divided into a training set 203 and a validation set 204 according to a certain ratio, and the data of the training set 203 is used to train a positioning model 205; the positioning model 205 is used to perform positioning prediction on the data of the validation set 204, the obtained predicted value is compared with the actual value, and the error of the positioning prediction is calculated to obtain an error prediction label 206; an error prediction feature 208 is constructed, and the error prediction feature 208 includes a feature 211 of the data to be predicted obtained based on the data to be predicted 210 and a local density feature 207 of the training set obtained based on the training set 203, which will be described in detail below; with the error prediction label 206 as supervision, an error prediction model 209 is trained based on the error prediction feature 208, and the error prediction model 209 can be used to predict the error of the positioning prediction result of the data to be predicted 210.

[0033] The error prediction method according to an embodiment of the present disclosure is described in detail below.

[0034] 1. Data partitioning

[0035] For example, according to the embodiment of the present disclosure, multiple days of data of the same cell can be taken as data with positioning tags. Figure 4 In the embodiment shown, data can be divided by day, and the MR data of the first 7 days can be used as data with positioning labels 201 to train and verify the positioning model, and the MR data of the 8th day of the same cell can be used as data to be predicted 210.

[0036] like Figure 4 As shown, sliding window iteration can be used to iteratively perform training and prediction.

[0037] 2. Construct positioning features

[0038] 1) To keep the predicted data consistent with the training set, the collected MR data are grouped according to the base station ID (enodeb_id) and cell ID (cell_id) to obtain several data sets.

[0039] 2) In each data set divided in 1), the frequency of occurrence of the FCN (Frequency Channel Number) and PCI (Physical Cell Identifier) of neighboring cells in all MR data is counted. The top k cell numbers are selected, ranked from highest to lowest in frequency. k can be set as needed, for example, k = 100. The RSRP (Reference Signal Received Power) value of each MR data entry for that neighboring cell is then indexed using these top k cell numbers. If no RSRP value exists for that neighboring cell, the RSRP value is left blank. This forms the positioning feature of [primary cell RSRP value, primary cell Time Advanced (TA) value, top 1 neighboring cell RSRP value, …, top k neighboring cell RSRP values]. All MR data for the same cell share the same top k neighboring cell index, ensuring feature alignment.

[0040] 3) To improve positioning accuracy, in embodiments of the present disclosure, the longitude and latitude of MR data belonging to the same cell can be clustered using DBSCAN. The grids of the MR data belonging to the same cluster are then connected to form irregular polygons, and a model is trained for the MR data within each polygon. When performing positioning prediction on MR data belonging to a specific cell, the polygon to which it belongs is first determined, and then the positioning model corresponding to that polygon is used for positioning. In embodiments of the present disclosure, the polygon number of the MR data can be used as a feature to describe local dispersion for error prediction.

[0041] 3. Construct quality characteristics of the data to be predicted

[0042] 1) Evaluate the data quality of the predicted data and count missing important positioning feature fields. In an embodiment of the present disclosure, the number of missing RSRP values for k neighboring cells in the constructed positioning features [primary cell RSRP value, primary cell TA value, top 1 neighboring cell RSRP value, ..., top k neighboring cell RSRP values] can be counted. In an embodiment of the present disclosure, whether the primary cell TA and primary cell RSRP are missing can also be counted, with the values set to 0 if missing and 1 otherwise.

[0043] 2) To describe the complexity of wireless network coverage around the data to be predicted, in an embodiment according to the present disclosure, the primary area TA value and / or RSRP value may also be included in the characteristics of the data to be predicted.

[0044] 4. Construct local density features of the training set

[0045] 1) For the data to be predicted, the positioning model can be used to predict its positioning, and then the distribution of the MR data in the training set can be statistically analyzed around the positioning. Specifically, the distance between the actual positioning (for example, longitude and latitude) of the MR data in the training set and the predicted positioning of the data to be predicted can be divided according to a certain distance standard to obtain multiple subsets. For example, according to the needs of subsequent business applications, the training set can be divided into four subsets based on the distance between the actual positioning of the MR data in the training set and the predicted positioning of the data to be predicted in the four ranges of 0-20m, 20-60m, 60-100m, and above 100m. This subset division operation will be performed once for each sample to be predicted. Note that the above numerical ranges are non-limiting examples, and the criteria for dividing the training set into subsets are not limited to this.

[0046] 2) Statistical features of the actual positioning of the MR data in each subset of the training set are respectively counted, such as the longitude mean, longitude variance, longitude range, and longitude sample number, as well as the latitude mean, latitude variance, latitude range, and latitude sample number, and these statistical features are included in the local density features of the training set.

[0047] 3) Count the frequency of occurrence of the primary region enodeb_id (base station identifier) and cell_id (cell identifier) numbers of the MR data in each subset of the training set. Select the top n primary regions from highest to lowest frequency, where n can be set as needed, for example, n = 10. Count the frequency of occurrence, primary region TA mean, primary region TA variance, primary region RSRP mean, and primary region RSRP variance of each primary region in each subset as corresponding fields and store them in a temporary primary region dictionary. If the primary region number of the data to be predicted is included in the temporary primary region dictionary, that is, the primary region of the data to be predicted is included in the top n primary regions, then index the temporary primary region dictionary and include the corresponding field in the local density feature of the training set. Otherwise, leave it blank.

[0048] 4) Count the frequency of neighbor PCI and FCN neighbor numbers for the MR data in each subset of the training set. Select the top m neighbors from highest to lowest frequency, where m can be set as needed, for example, m = 15. Count the frequency of each neighbor in each subset, the mean RSRP value, and the variance of the RSRP value, and store them in a temporary neighbor dictionary. If the neighbor number of the data to be predicted is included in the temporary neighbor dictionary, that is, the neighbor of the data to be predicted is included in the top m neighbors, then index the temporary neighbor dictionary and include the corresponding field in the local density feature of the training set. Otherwise, leave it blank.

[0049] 5. Construct error prediction labels

[0050] 1) For the validation set, use the positioning prediction model to predict the positioning of the data in the validation set, and calculate the error between the predicted positioning and the actual positioning.

[0051] 2) Based on business needs, in embodiments of the present disclosure, for example, the error can be divided into four ranges: 0-20m, 20-60m, 60-100m, and greater than 100m, thereby maintaining consistency with the training set division criteria described above. Note that the above numerical ranges are non-limiting examples, and the criteria for dividing the error are not limited to these.

[0052] 3) At this point, the error prediction can be converted into a multi-classification task. For example, when the error prediction is divided into 4 ranges, it can be converted into a 4-classification task.

[0053] In addition, depending on the business scenario, the division in 2) may not be performed, and the task is converted into a regression task.

[0054] In an embodiment of the present disclosure, when the data is located in longitude and latitude, error prediction features and error labels can be constructed using both longitude and latitude, and training and prediction can be performed using single-output machine learning. In an embodiment of the present disclosure, error prediction features and error labels can also be constructed separately for the longitude model and the latitude model to obtain a more accurate model.

[0055] 6. Error prediction model training and prediction

[0056] 1) Use the features constructed in steps 2, 3, and 4 as error prediction features. Using these features and the error prediction labels constructed in step 5, you can train a classification model based on a machine learning algorithm, such as the LightGBM model. Since only the validation set has AGPS labels, you can complete this step on the validation set and evaluate the performance of the error prediction model.

[0057] 2) Use the error prediction model obtained in 1) to predict the positioning error of the data to be predicted without AGPS tags.

[0058] Figure 3 FIG. 1 shows a schematic diagram of an error prediction system according to an embodiment of the present disclosure. Figure 3 As shown, the system includes an error prediction feature construction module and an error prediction label construction module.

[0059] The error prediction label construction module can be configured to determine the error between the predicted value and the actual value based on the actual AGPS labels of the validation set and the positioning model, and to generate an error prediction label based on the error. The binning module can be configured to bin the error, for example, into four ranges: 0-20m, 20-60m, 60-100m, and above 100m.

[0060] The error prediction feature construction module can be configured to construct error prediction features. The error prediction features may include features of the data to be predicted and local density features of the training set. For example, Figure 3 As shown, the features of the data to be predicted may include values indicating whether the main area parameters and neighboring area parameters in the data to be predicted are missing, such as vacant values, TA missing, Nan neighboring areas, etc.; the local density features of the training set may include corresponding fields obtained from the temporary dictionary index, such as the main area RSRP+TA value, the neighboring area RSRP value, etc.; the local density features of the training set may also include statistical features of the actual positioning of the MR data in each subset, such as the number of LNG (longitude) items, the number of LAT (latitude) items, the LNG variance, the LAT variance, the LNG range, the LAT range, etc.

[0061] The system may further include a training module, which may be configured to train the error prediction model using the error prediction features with the error prediction labels as supervision.

[0062] like Figure 3 As shown, the error prediction model can be an error prediction LGB classifier, clustering features clustering polygon statistical features, and the classifier can be trained using error prediction features with error prediction labels as supervision. In addition, in embodiments according to the present disclosure, the error prediction model can also be cell-specific.

[0063] Figure 5 A flowchart illustrating a positioning model error prediction method according to an embodiment of the present disclosure is shown.

[0064] At S501 , an error prediction feature may be constructed. The error prediction feature is associated with a training set, specifically, with a sampling density and data quality of the training set.

[0065] At S502 , an error prediction label may be constructed, including: using a positioning prediction model to predict the positioning of data in a validation set, and calculating the error between the predicted positioning and the actual positioning of the data in the validation set, and obtaining an error prediction label based on the error.

[0066] At S503 , the error prediction model may be trained using the error prediction features with the error prediction labels as supervision.

[0067] Figure 6 An exemplary configuration of a computing device 600 capable of implementing embodiments according to the present disclosure is shown.

[0068] Computing device 600 is an example of a hardware device to which the above-described aspects of the present disclosure can be applied. Computing device 600 can be any machine configured to perform processing and / or computations. Computing device 600 can be, but is not limited to, a workstation, a server, a desktop computer, a laptop computer, a tablet computer, a personal data assistant (PDA), a smartphone, an in-vehicle computer, or a combination thereof.

[0069] like Figure 6 As shown, the computing device 600 may include one or more components that can be connected or communicated with the bus 602 via one or more interfaces. The bus 602 may include, but is not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MCA) bus, an Enhanced ISA (EISA) bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus. The computing device 600 may include, for example, one or more processors 604. The one or more processors 604 may be any type of processor and may include, but is not limited to, one or more general-purpose processors or special-purpose processors (such as dedicated processing chips). The processor may be configured to implement, for example, Figure 5 The method shown.

[0070] The computing device 600 may also include or be connected to a non-transitory storage device 614, which may be any non-transitory storage device that can store data and may include, but is not limited to, a disk drive, an optical storage device, a solid-state memory, a floppy disk, a flexible disk, a hard disk, magnetic tape, or any other magnetic medium, a compact disk or any other optical medium, a cache memory, and / or any other memory chip or module, and / or any other medium from which a computer can read data, instructions, and / or code. The computing device 600 may also include random access memory (RAM) 610 and read-only memory (ROM) 612. The ROM 612 may store programs, utilities, or processes to be executed in a non-volatile manner. The RAM 610 may provide volatile data storage and store instructions related to the operation of the computing device 600.

[0071] In summary, according to the first aspect of the present disclosure, a method for training an error prediction model is provided, wherein the error prediction model is used to predict the error between the predicted positioning of the data to be predicted obtained using the positioning prediction model and the actual positioning, and the positioning prediction model is obtained by training using a training set. The method includes: constructing an error prediction feature, and the error prediction feature is associated with the sampling density and data quality of the training set; constructing an error prediction label, including: using the positioning prediction model to predict the positioning of the data in the validation set, and calculating the error between the predicted positioning of the data in the validation set and the actual positioning, and obtaining an error prediction label based on the error; and using the error prediction label as supervision to train the error prediction model using the error prediction feature.

[0072] In an embodiment according to the present disclosure, constructing error prediction features includes: constructing features of data to be predicted; and constructing local density features of a training set.

[0073] In an embodiment of the present disclosure, the characteristics of the data to be predicted include: a value indicating whether the main area parameter and the neighboring area parameter in the data to be predicted are missing; and the main area parameter in the data to be predicted.

[0074] In an embodiment of the present disclosure, constructing the local density feature of the training set includes: judging whether the main area number of the data to be predicted is included in the temporary main area dictionary of each subset of the training set; if the main area number of the data to be predicted is included in the temporary main area dictionary of the subset, indexing the temporary main area dictionary and including the corresponding field of the temporary main area dictionary in the local density feature of the training set; judging whether the neighboring area number of the data to be predicted is included in the temporary neighboring area dictionary of each subset of the training set; if the neighboring area number of the data to be predicted is included in the temporary neighboring area dictionary of the subset, indexing the temporary neighboring area dictionary and including the corresponding field of the temporary neighboring area dictionary in the local density feature of the training set; and calculating the statistical features of the actual positioning of the data in each subset, and including the calculated statistical features in the local density feature of the training set; wherein the subset is obtained by dividing the training set according to the distance standard between the actual positioning of the data of the training set and the predicted positioning of the data to be predicted.

[0075] In an embodiment according to the present disclosure, the temporary main area dictionary of each subset of the training set is constructed as follows: the frequency of occurrence of the main area number of the data in the subset is counted, multiple main areas are selected from high to low frequency, and the statistical characteristics of the main area parameters of the data in the subset are used as corresponding fields, wherein the numbers of the multiple main areas selected and the corresponding fields constitute a temporary main area dictionary; wherein, the temporary neighboring area dictionary of each subset of the training set is constructed as follows: the frequency of occurrence of the neighboring area number of the data in the subset is counted, multiple neighboring areas are selected from high to low frequency, and the statistical characteristics of the neighboring area parameters of the data in the subset are used as corresponding fields, wherein the numbers of the multiple neighboring areas selected and the corresponding fields constitute a temporary neighboring area dictionary.

[0076] In an embodiment according to the present disclosure, the error prediction labels are divided according to the same distance criterion as that used to divide the training set into subsets.

[0077] According to the second aspect of the present disclosure, a method for predicting the error of a positioning prediction model is provided, which is used to predict the error between the predicted positioning of data to be predicted obtained using the positioning prediction model and the actual positioning. The method comprises: training the error prediction model using the method described in the first aspect of the present disclosure; and predicting the error of the data to be predicted using the error prediction model.

[0078] According to a third aspect of the present disclosure, a device for training an error prediction model is provided, wherein the error prediction model is used to predict the error between the predicted positioning of data to be predicted obtained by using a positioning prediction model and the actual positioning, and the positioning prediction model is obtained by training using a training set. The device includes: a feature construction module, wherein the feature construction module is configured to construct an error prediction feature, and the error prediction feature is associated with the sampling density and data quality of the training set; a label construction module, wherein the label construction module is configured to construct an error prediction label as follows: using the positioning prediction model to predict the positioning of data in a validation set, and calculating the error between the predicted positioning of the data in the validation set and the actual positioning, and obtaining an error prediction label based on the error; and a training module, wherein the training module is configured to train the error prediction model using the error prediction feature with the error prediction label as supervision.

[0079] In an embodiment according to the present disclosure, constructing error prediction features includes: constructing features of data to be predicted; and constructing local density features of a training set.

[0080] In an embodiment of the present disclosure, the characteristics of the data to be predicted include: a value indicating whether the main area parameter and the neighboring area parameter in the data to be predicted are missing; and the main area parameter in the data to be predicted.

[0081] In an embodiment of the present disclosure, constructing the local density feature of the training set includes: judging whether the main area number of the data to be predicted is included in the temporary main area dictionary of each subset of the training set; if the main area number of the data to be predicted is included in the temporary main area dictionary of the subset, indexing the temporary main area dictionary and including the corresponding field of the temporary main area dictionary in the local density feature of the training set; judging whether the neighboring area number of the data to be predicted is included in the temporary neighboring area dictionary of each subset of the training set; if the neighboring area number of the data to be predicted is included in the temporary neighboring area dictionary of the subset, indexing the temporary neighboring area dictionary and including the corresponding field of the temporary neighboring area dictionary in the local density feature of the training set; and calculating the statistical features of the actual positioning of the data in each subset, and including the calculated statistical features in the local density feature of the training set; wherein the subset is obtained by dividing the training set according to the distance standard between the actual positioning of the data of the training set and the predicted positioning of the data to be predicted.

[0082] In an embodiment according to the present disclosure, the temporary main area dictionary of each subset of the training set is constructed as follows: the frequency of occurrence of the main area number of the data in the subset is counted, multiple main areas are selected from high to low frequency, and the statistical characteristics of the main area parameters of the data in the subset are used as corresponding fields, wherein the numbers of the multiple main areas selected and the corresponding fields constitute a temporary main area dictionary; wherein, the temporary neighboring area dictionary of each subset of the training set is constructed as follows: the frequency of occurrence of the neighboring area number of the data in the subset is counted, multiple neighboring areas are selected from high to low frequency, and the statistical characteristics of the neighboring area parameters of the data in the subset are used as corresponding fields, wherein the numbers of the multiple neighboring areas selected and the corresponding fields constitute a temporary neighboring area dictionary.

[0083] In an embodiment according to the present disclosure, the error prediction labels are divided according to the same distance criterion as that used to divide the training set into subsets.

[0084] According to the fourth aspect of the present disclosure, a device for predicting the error of a positioning prediction model is provided, which is used to predict the error between the predicted positioning of the data to be predicted obtained using the positioning prediction model and the actual positioning. The device includes: the device according to the third aspect of the present disclosure; a prediction module, which uses the error prediction model obtained by the device according to the third aspect of the present disclosure to predict the error of the data to be predicted.

[0085] According to a fifth aspect of the present disclosure, a non-transitory computer-readable storage medium is provided, on which a program is stored, characterized in that when the program is executed by a computer, the computer is caused to execute the method according to the first aspect of the present disclosure.

[0086] According to a sixth aspect of the present disclosure, a non-transitory computer-readable storage medium is provided, on which a program is stored, characterized in that when the program is executed by a computer, the computer is caused to execute the method according to the second aspect of the present disclosure.

[0087] According to the seventh aspect of the present disclosure, a device for training an error prediction model is provided, comprising a memory and a processor, wherein the memory is communicatively coupled to the processor, and a program is stored in the memory, and when the program is executed by the processor, the processor executes the method described in the first aspect of the present disclosure.

[0088] According to an eighth aspect of the present disclosure, a device for predicting the error of a positioning prediction model is provided, comprising a memory and a processor, wherein the memory is communicatively coupled to the processor, and a program is stored in the memory, and when the program is executed by the processor, the processor executes the method described in the second aspect of the present disclosure.

[0089] According to a ninth aspect of the present disclosure, a computer program product is provided, comprising a computer program, which implements the method according to the first aspect of the present disclosure when executed by a processor.

[0090] According to a tenth aspect of the present disclosure, a computer program product is provided, comprising a computer program, which implements the method according to the second aspect of the present disclosure when executed by a processor.

[0091] This disclosure provides a method for characterizing the prediction error of unlabeled data when a supervised machine learning model processes a regression task. This characterization is performed for each sample, and a prediction label application strategy can be formulated based on the characterization. This greatly improves the waste caused by discarding all prediction labels due to poor overall prediction results in multi-cell and multi-model solutions for wireless network positioning. The methods, devices, non-transitory computer-readable storage media, computer program products, etc. according to the embodiments of the present disclosure have at least the following advantages:

[0092] 1. Through supervised learning, a certain relationship is established between the sampling density of the training set, the quality of the sample data to be predicted, and the prediction error, so as to predict the error level without using prior knowledge to infer the error model;

[0093] 2. The accuracy of the error prediction model is improved by constructing the sampling density of the training set data, the distribution statistics of the important positioning features of the training set data, the quality of the sample data to be predicted, and the feature engineering of the sample data to be tested;

[0094] 3. It provides a method for predicting the error of each sample for machine learning regression tasks, rather than using the overall performance of the model on a test set to evaluate unlabeled samples to be predicted. This method is more accurate and allows for the refined use of localized samples.

[0095] 4. Compared with modeling based on the root cause of error formation, modeling based on measurement periodic error is more universal, simple to deploy in different scenarios, more lightweight, and has low modeling cost.

[0096] The present disclosure provides a method for characterizing the prediction error of unlabeled data when a supervised machine learning model processes a regression task. This characterization is for each sample, and a prediction label application strategy can be formulated based on the characterization, which greatly improves the waste caused by abandoning all prediction labels due to poor overall prediction effect in multi-cell and multi-model solutions in wireless network positioning.

[0097] The subject matter of the present disclosure is provided as an example of devices, systems, methods, and programs for performing the features described in the present disclosure. However, in addition to the features described above, other features or variations are also contemplated. It is contemplated that any emerging technology may be used to implement the components and functions of the present disclosure in place of any of the technologies described above.

[0098] In addition, the above description provides examples and does not limit the scope, applicability or configuration set forth in the claims. Without departing from the spirit and scope of this disclosure, the function and arrangement of the elements discussed can be changed. Various embodiments can appropriately omit, replace or add various processes or components. For example, features described with respect to certain embodiments can be combined in other embodiments.

[0099] Similarly, while operations are depicted in a particular order in the drawings, this should not be understood as requiring that such operations be performed in the particular order shown or in sequential order, or that all illustrated operations be performed to achieve desired results. In certain circumstances, multitasking and parallel processing can be advantageous.

Claims

1. A method for training an error prediction model, wherein the error prediction model is used to predict the error between a predicted location obtained using a positioning prediction model and an actual location of data to be predicted, wherein the positioning prediction model is trained using a training set, the method comprising: Constructing error prediction features, including constructing features of the data to be predicted and constructing local density features of the training set, wherein the error prediction features are associated with the sampling density and data quality of the training set; Construct error prediction labels, including: Using the positioning prediction model to predict the positioning of the data in the validation set, and Calculating the error between the predicted location and the actual location of the data in the validation set, and obtaining an error prediction label based on the error; and Using the error prediction labels as supervision, the error prediction features are used to train the error prediction model; The characteristics of the data to be predicted include: a value indicating whether the main area parameter and the neighboring area parameter in the data to be predicted are missing, and the main area parameter in the data to be predicted; Constructing local density features of the training set includes: If the main region number of the data to be predicted is included in the temporary main region dictionary of each subset of the training set, indexing the temporary main region dictionary and including the corresponding field of the temporary main region dictionary in the local density feature of the training set; If the neighboring area number of the data to be predicted is included in the temporary neighboring area dictionary of each subset of the training set, indexing the temporary neighboring area dictionary and including the corresponding field of the temporary neighboring area dictionary in the local density feature of the training set; and Calculate the statistical features of the actual positioning of the data in each subset, and include the calculated statistical features in the local density features of the training set; The subsets are obtained by dividing the training set according to a distance standard based on the distance between the actual positioning of the data in the training set and the predicted positioning of the data to be predicted.

2. The method according to claim 1, in, The temporary main region dictionary of each subset of the training set is constructed as follows: Count the frequency of occurrence of the main area number of the data in the subset. Select multiple main areas according to the frequency from high to low, and use the statistical characteristics of the main area parameters of the data in the subset as the corresponding fields, The numbers and corresponding fields of the selected primary areas constitute a temporary primary area dictionary; The temporary neighbor dictionary of each subset of the training set is constructed as follows: The frequency of the neighboring area numbers of the data in the statistical subset, Select multiple neighboring areas according to the frequency from high to low, and use the statistical characteristics of the neighboring area parameters of the data in the subset as the corresponding fields, The numbers and corresponding fields of the selected multiple neighboring areas constitute a temporary neighboring area dictionary.

3. The method according to claim 1, wherein The error prediction labels are divided according to the same distance criterion as that used to divide the training set into subsets.

4. A method for predicting the error of a positioning prediction model, for predicting the error between a predicted positioning obtained using a positioning prediction model and an actual positioning of data to be predicted, the method comprising: Training an error prediction model using the method according to any one of claims 1 to 3; The error prediction model is used to predict the error of the data to be predicted.

5. A device for training an error prediction model, the error prediction model being used to predict the error between a predicted location obtained using a positioning prediction model and an actual location of data to be predicted, the positioning prediction model being trained using a training set, the device comprising: A feature construction module, wherein the feature construction module is configured to construct error prediction features, including constructing features of the data to be predicted and constructing local density features of the training set, wherein the error prediction features are associated with the sampling density and data quality of the training set; A label construction module is configured to construct an error prediction label as follows: Using the positioning prediction model to predict the positioning of the data in the validation set, and Calculate the error between the predicted location and the actual location of the data in the validation set, and obtain an error prediction label based on the error; as well as A training module, wherein the training module is configured to train an error prediction model using error prediction features with error prediction labels as supervision; The characteristics of the data to be predicted include: a value indicating whether the main area parameter and the neighboring area parameter in the data to be predicted are missing, and the main area parameter in the data to be predicted; Constructing local density features of the training set includes: If the main region number of the data to be predicted is included in the temporary main region dictionary of each subset of the training set, indexing the temporary main region dictionary and including the corresponding field of the temporary main region dictionary in the local density feature of the training set; If the neighboring area number of the data to be predicted is included in the temporary neighboring area dictionary of each subset of the training set, indexing the temporary neighboring area dictionary and including the corresponding field of the temporary neighboring area dictionary in the local density feature of the training set; and Calculate the statistical features of the actual positioning of the data in each subset, and include the calculated statistical features in the local density features of the training set; The subsets are obtained by dividing the training set according to a distance standard based on the distance between the actual positioning of the data in the training set and the predicted positioning of the data to be predicted.

6. The device according to claim 5, in, The temporary main region dictionary of each subset of the training set is constructed as follows: Count the frequency of occurrence of the main area number of the data in the subset. Select multiple main areas according to the frequency from high to low, and use the statistical characteristics of the main area parameters of the data in the subset as the corresponding fields, The numbers and corresponding fields of the selected primary areas constitute a temporary primary area dictionary; The temporary neighbor dictionary of each subset of the training set is constructed as follows: The frequency of the neighboring area numbers of the data in the statistical subset, Select multiple neighboring areas according to the frequency from high to low, and use the statistical characteristics of the neighboring area parameters of the data in the subset as the corresponding fields, The numbers and corresponding fields of the selected multiple neighboring areas constitute a temporary neighboring area dictionary.

7. The device according to claim 5, wherein The error prediction labels are divided according to the same distance criterion as that used to divide the training set into subsets.

8. A device for predicting the error of a positioning prediction model, for predicting the error between a predicted positioning obtained using a positioning prediction model and an actual positioning of data to be predicted, the device comprising: The device according to any one of claims 5 to 7; A prediction module, wherein the prediction module predicts the error of the data to be predicted by using the error prediction model obtained by the device according to any one of claims 5-7.

9. A non-transitory computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a computer, the computer is caused to execute the method according to any one of claims 1 to 3.

10. A non-transitory computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a computer, the computer is caused to execute the method according to claim 4 .

11. A device for training an error prediction model, comprising a memory and a processor, wherein the memory is communicatively coupled to the processor, and a program is stored in the memory, and when the program is executed by the processor, the processor executes the method according to any one of claims 1-3.

12. A device for predicting the error of a positioning prediction model, comprising a memory and a processor, wherein the memory is communicatively coupled to the processor, and a program is stored in the memory, and when the program is executed by the processor, the processor executes the method according to claim 4.

13. A computer program product comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 3.

14. A computer program product comprising a computer program, which, when executed by a processor, implements the method according to claim 4.

Images