Method for detecting fault of positioning base station and storage medium

Abstract

A method for detecting a fault of a positioning base station and a storage medium, relating to the technical field of navigation positioning, the method comprising: acquiring real-time observation data and ephemeris data of a target base station, the target base station being a fault positioning base station; taking the real-time observation data and the ephemeris data as input of a pre-trained deep learning model, performing reasoning by using the deep learning model, and outputting a prediction result; and determining a fault type of the target base station and a corresponding repair measure based on the prediction result.

Description

Technical Field

[0001] The present invention relates to the field of navigation and positioning, and in particular to a method and storage medium for detecting positioning base station failures. Background Art

[0002] High-precision positioning refers to obtaining high-precision location information of the target in the geographic coordinate system through positioning technology, and the location accuracy is required to be controlled at the meter level, decimeter level or even centimeter level. High-precision positioning not only includes the location information of the target, but also involves information such as the target's moving speed and direction, as well as the target's arrival and retention time at different locations. The application fields of high-precision positioning technology are very wide, such as vehicle networking, smart transportation, emergency rescue, vehicle navigation, counter-terrorism and stability maintenance. With the development of technologies such as the Internet of Things, 5G communications, artificial intelligence, and cloud services, high-precision positioning technology provides more possibilities in logistics management, trajectory tracking, personal inquiries, and the Internet of Everything.

[0003] In high-precision positioning systems, positioning base stations play a vital role. For example, in RTK (Real-Time Kinematic) positioning technology, the three-dimensional coordinate information of the positioning base station is usually known, and it can be used as a reference point for a fixed position to provide accurate positioning data. In addition, the positioning base station can observe and receive satellite data, and then send the observation data to the mobile station (i.e., the mobile device that needs to be positioned) in real time through a radio station (data link). This process is a key step in achieving millimeter-level high-precision three-dimensional real-time positioning in RTK positioning technology.

[0004] For another example, in PPP-RTK (Precise Point Positioning with Real-Time Kinematic) positioning technology, the positioning base station is the basis for calculating the SSR (State Space Representation) correction number. The clock correction number, orbit correction number, phase difference correction number, atmospheric troposphere correction number, and ionosphere correction number all rely on accurate positioning base station observation data and broadcast ephemeris data. If the positioning base station fails, it will cause accuracy loss in calculating the SSR correction number, and even cause potential risks such as service collapse. It can be seen that fault detection and repair of positioning base stations are of great significance to maintaining the normal operation of the positioning system.

[0005] In related technologies, positioning base stations are usually detected through professional software (such as Teqc), and the status of positioning base stations is preliminarily screened and detected based on the detected fixed types of information (such as ionospheric observations, ionospheric tube observation value variability, etc.). However, the information detected in this way is less targeted, and only a given fault can be detected through numerical statistical calculations and linear combinations. Sometimes the detection information is normal but the positioning base station fails. In addition, no measures to repair the fault can be provided to the operation and maintenance personnel, who can only rely on their own professional skills and experience to repair the fault. Summary of the invention

[0006] Embodiments of the present application provide a method and storage medium for detecting a positioning base station failure.

[0007] In a first aspect, an embodiment of the present application provides a method for detecting a positioning base station failure, comprising: obtaining real-time observation data and ephemeris data of a target base station, where the target base station is a positioning base station that has a failure; using the real-time observation data and ephemeris data as input to a pre-trained deep learning model, performing reasoning using the deep learning model, and outputting a prediction result, where the deep learning model is a model with memory processing capabilities; based on the prediction result, determining the fault type of the target base station and the corresponding repair measures.

[0008] In a second aspect, an embodiment of the present application provides a non-transitory computer storage medium, which stores a computer program. When the computer program is executed by a processor, the method for detecting a positioning base station failure in the above embodiment is implemented.

[0009] The method for detecting and positioning base station faults in the embodiments of the present application obtains real-time observation data and ephemeris data of the target base station where the fault occurs, and then uses a deep learning model with memory processing capabilities for reasoning to predict the type of fault that occurs in the target base station and the corresponding repair measures. On the one hand, based on reasoning and prediction of the real-time observation data and ephemeris data of the target base station, abnormal data can be identified more sensitively, which helps to improve the reliability and stability of fault detection; on the other hand, repair measures corresponding to the fault type can be provided, which helps to reduce the technical burden of operation and maintenance personnel and the probability of misjudgment caused by subjective factors, thereby improving the reliability, convenience and repair efficiency of positioning base station fault detection.

[0010] Other features and advantages of the present application will be described in the following description, and partly become apparent from the description, or be understood by implementing the present application. Other advantages of the present application can be realized and obtained by the schemes described in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The accompanying drawings are used to provide an understanding of the technical solution of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the technical solution of the present application and do not constitute a limitation on the technical solution of the present application.

[0012] Figure 1 A flowchart of an embodiment of a method for detecting a positioning base station failure in an embodiment of the present application; Figure 2 This is a flow chart of another embodiment of a method for detecting a positioning base station failure in an embodiment of the present application; Figure 3 This is a schematic diagram of a process of using a deep learning model to perform reasoning in an embodiment of a method for detecting and positioning a base station failure in an embodiment of the present application; Figure 4 This is a schematic diagram of a process of training a deep learning model in one embodiment of a method for detecting and positioning a base station failure in an embodiment of the present application; Figure 5 The present invention is a flowchart of constructing training samples in one embodiment of a method for detecting and positioning base station failures in an embodiment of the present application. DETAILED DESCRIPTION

[0013] The embodiments of the present application will be described in detail below in conjunction with the accompanying drawings. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

[0014] The present application describes multiple embodiments, but the description is exemplary rather than restrictive, and it is obvious to those skilled in the art that there may be more embodiments and implementations within the scope of the embodiments described in the present application. Although many possible feature combinations are shown in the drawings and discussed in the specific embodiments, many other combinations of the disclosed features are also possible. Unless specifically limited, any feature or element of any embodiment may be used in combination with any other feature or element in any other embodiment, or may replace any other feature or element in any other embodiment.

[0015] The present application includes and contemplates combinations of features and elements known to those of ordinary skill in the art. The embodiments, features and elements disclosed in the present application may also be combined with any conventional features or elements to form a unique invention scheme. Any features or elements of any embodiment may also be combined with features or elements from other invention schemes to form another unique invention scheme. Therefore, it should be understood that any feature shown and / or discussed in the present application may be implemented individually or in any appropriate combination. Therefore, except for the limitations made according to the attached claims and their equivalents, the embodiments are not subject to other restrictions. In addition, various modifications and changes may be made within the scope of protection of the attached claims.

[0016] In addition, when describing representative embodiments, the specification may have presented the method and / or process as a specific sequence of steps. However, to the extent that the method or process does not rely on the specific order of the steps described herein, the method or process should not be limited to the steps of the specific order described. As will be understood by those of ordinary skill in the art, other sequences of steps are also possible. Therefore, the specific sequence of the steps set forth in the specification should not be interpreted as a limitation to the claims. In addition, the claims for the method and / or process should not be limited to the steps of performing them in the order written, and those skilled in the art can easily understand that these sequences can be changed and still remain within the spirit and scope of the embodiments of the present application.

[0017] Figure 1 A flowchart of an embodiment of a method for detecting a positioning base station failure of the present application is shown. Figure 1 As shown, the process may include the following steps.

[0018] Step 110: Acquire real-time observation data and ephemeris data of the target base station.

[0019] The target base station is a positioning base station that has a fault.

[0020] In this embodiment, the execution entity of the method for detecting and positioning base station failure (for example, a terminal device or a server) can obtain real-time observation data and ephemeris data from the target base station through a serial port or other communication interface through the target base station, or can also upload it to a cloud server (for example, a data center) through a network ntrip protocol.

[0021] Here, the observation data and ephemeris data may include observation data and ephemeris corresponding to different satellites observed by the positioning base station at different frequency points.

[0022] Step 120: Use the real-time observation data and ephemeris data as inputs to a pre-trained deep learning model, use the deep learning model to perform reasoning, and output prediction results.

[0023] Among them, the deep learning model is a model with memory processing capabilities, such as the Transformer model, Long Short-Term Memory (LSTM), Memory-Augmented Neural Networks (MANNs), etc. In order to use the memory processing capabilities of the deep learning model to capture the correlation features between data, thereby improving the prediction accuracy of the deep learning model.

[0024] In this embodiment, the deep learning model can characterize the corresponding relationship between the real-time observation data and the ephemeris data and the prediction results. The prediction results can characterize the fault type and its corresponding repair measures.

[0025] Step 130: Based on the prediction result, determine the fault type of the target base station and the corresponding repair measures.

[0026] The method for detecting and positioning base station faults in this embodiment obtains real-time observation data and ephemeris data of the target base station where the fault occurs, and then uses a deep learning model for reasoning to predict the type of fault that occurs in the target base station and the corresponding repair measures. On the one hand, based on the reasoning and prediction of the real-time observation data and ephemeris data of the target base station, compared with judging based on fixed types of information, it can more sensitively identify abnormal data, which helps to improve the reliability and stability of fault detection; on the other hand, it can provide repair measures corresponding to the fault type, which helps to reduce the technical burden of operation and maintenance personnel and the probability of misjudgment caused by subjective factors, thereby improving the reliability, convenience and repair efficiency of positioning base station fault detection.

[0027] In some embodiments, the deep learning model may include a Transformer model that employs a multi-attention head mechanism.

[0028] As an example, a deep learning model may include a Transformer model and an output layer, wherein the Transformer model may include an encoding layer, a pooling layer, and a decoding layer, and one or more of the encoding layer, the pooling layer, and the decoding layer may be provided with multiple attention head modules, so that each attention head module extracts the correlation information between the input data (i.e., real-time observation data and ephemeris data) from different dimensions, so that the Transformer model can perform reasoning based on the correlation information between the data, and the memory length of the Transformer model can be increased, thereby improving the reliability of positioning base station fault detection. The output layer may be connected to the decoding layer of the Transformer model, and is used to perform feature mapping according to the decoded data output by the Transformer model, and determine the type label corresponding to the input data.

[0029] Next reference Figure 2 , Figure 2 A flowchart of selecting a deep learning model in an embodiment of a method for detecting and positioning a base station failure of the present application is shown. Figure 2 As shown, the process may include the following steps.

[0030] Step 210: Acquire real-time observation data and ephemeris data of the target base station.

[0031] Step 220: Obtain the real location of the target base station.

[0032] Step 230: Perform positioning calculation based on the real-time observation data and ephemeris data to determine the predicted position of the target base station.

[0033] Step 240: Determine the difference between the actual position and the predicted position.

[0034] Step 250: When the difference is not greater than a preset threshold, perform inference based on the second model.

[0035] Step 260: When the difference is greater than a preset threshold, perform inference based on the first model.

[0036] In this embodiment, the deep learning model may include a first model and a second model, wherein the parameter amount of the first model is greater than the parameter amount of the second model.

[0037] As an example, both the first model and the second model may include a Transformer model and an output layer that adopts a multi-attention head mechanism. The parameters of the model may include the number of attention heads, the number of hidden layers, etc. Compared with the second model, the first model may include more attention heads and hidden layers.

[0038] Generally, the larger the number of parameters in a deep learning model, the higher the accuracy of the reasoning. Correspondingly, the greater the amount of computation required for the reasoning process, which is suitable for reasoning tasks with higher accuracy requirements. For reasoning tasks with lower accuracy requirements, a deep learning model with a smaller number of parameters can be selected for reasoning to reduce the amount of computation and increase the reasoning speed.

[0039] Usually, the real position of the positioning base station is known, and the execution subject can directly obtain the real position of the target base station. Then, the predicted position of the target base station is solved based on the real-time observation data and ephemeris data, and the difference between the real position and the predicted position is determined. The difference can characterize the positioning error of the target base station. When the difference is greater than the preset threshold (for example, it can be 2 meters), it means that the positioning error of the target base station is large, and the need for troubleshooting is high. At this time, the first model with a larger parameter amount can be used for full detection to improve the accuracy of fault detection; when the difference is not greater than the preset threshold, it means that the positioning error of the target base station is small, and the need for troubleshooting is small. At this time, the second model with a smaller parameter amount can be selected for simplified detection to reduce the amount of calculation and improve the reasoning speed.

[0040] In this embodiment, the predicted position of the target base station can be calculated based on the observed data of the target base station, and the positioning error of the target base station can be represented by the difference between the predicted position and the actual position, so as to evaluate the severity of the fault of the target base station. When the difference is small, the second model with a smaller parameter amount can be used for reasoning to simplify the detection process. Under the premise of ensuring accuracy, the computing resources required for detection can be reduced; when the difference is large, the first model with a larger parameter amount can be used for reasoning to ensure the accuracy of fault detection. Selecting the corresponding deep learning model according to actual needs can balance the efficiency and detection quality of positioning base station fault detection.

[0041] Figure 3 FIG. 1 is a flowchart showing a method for detecting and locating a base station failure in an embodiment of the present application using a deep learning model for reasoning, such as Figure 3 As shown, the process may include the following steps.

[0042] Step 310: extract the correlation information between each data in the real-time observation data and the ephemeris data based on the multi-attention head mechanism.

[0043] Step 320: Based on the multi-attention head mechanism, the real-time observation data and ephemeris data are fused with the correlation information, and bias processing and normalization processing are performed to obtain encoded data.

[0044] As an example, multiple attention head modules can be used to number the input real-time observation data and ephemeris data to form an input data sequence, and then each attention head module calculates the attention weight of each data in the real-time observation data and ephemeris data according to different dimensions (that is, the correlation information between each data in the real-time observation data and ephemeris data). After that, the correlation information output by multiple attention head modules is fused with the real-time observation data and ephemeris data, and then bias processing and normalization processing are performed to obtain the encoded data.

[0045] Step 330: Perform pooling processing on the encoded data based on the multi-attention head mechanism to obtain pooled data.

[0046] In this embodiment, through pooling processing, the adverse impact of abnormal data on the reasoning process can be reduced, which helps to improve the robustness of the deep learning model.

[0047] Step 340: decode the pooled data based on the multi-attention head mechanism to obtain decoded data.

[0048] Step 350: feature mapping is performed on the decoded data, and the type identification corresponding to the decoded data is output as the prediction result corresponding to the real-time observation data and the ephemeris data.

[0049] The type identifier represents the fault type and the corresponding repair measures.

[0050] In a specific example, a deep learning model may include a Transformer model and an output layer, wherein the Transformer model includes an encoding layer, a pooling layer, and a decoding layer. The encoding layer may encode the input real-time observation data and ephemeris data based on a multi-attention head mechanism to extract the correlation information between the real-time observation data and the ephemeris data and fuse them with the input real-time observation data and ephemeris data, and then obtain the encoded data after bias and normalization processing, so as to extract the feature information of each dimension from the real-time observation data and the ephemeris data; then the pooling layer is used to perform pooling processing on the encoded data based on the multi-attention head mechanism to reduce the adverse effect of abnormal data on reasoning; then, the pooled data is input into the decoding layer for decoding processing to obtain the decoded data; finally, the output layer performs feature mapping on the decoded data to determine the type identification corresponding to the decoded data as the prediction result corresponding to the real-time observation data and the ephemeris data.

[0051] In this embodiment, the deep learning model can use a multi-attention head mechanism to encode the input real-time observation data and ephemeris data to extract correlation information, and then encode, pool and decode based on the multi-attention head mechanism, and use the type identification corresponding to the real-time observation data and ephemeris data predicted by feature mapping to determine the fault type of the target base station and the corresponding repair measures, which helps to improve the accuracy of fault detection.

[0052] In some optional implementations of this embodiment, step 350 may include: pre-determining the correspondence between the type identifier and the fault type and the corresponding repair measure, determining the fault type and the corresponding repair measure corresponding to the type identifier included in the prediction result as the fault type and the corresponding repair measure of the target base station.

[0053] As an example, the number can be used as a type identifier, and each number corresponds to a fault type and a corresponding repair measure, so as to construct a corresponding relationship between the type identifier, the fault type, and the corresponding repair measure. For example, if the number is 1, the corresponding fault type is that there is an obstruction in a certain direction, and the repair measure is to be checked by the operation and maintenance personnel; if the number is 2, there is a multipath at the frequency point L of the target base station, and the repair measure is restart.

[0054] Optionally, the deep learning model can also predict the severity of the fault, and accordingly, the output prediction result can also characterize the severity of the fault. For example, for the same type of fault, different severity levels can correspond to different numbers. The number is 3, and the corresponding fault type is a minor fault inside the target base station, and the repair measure is to recommend replacing the faulty part; the number is 4, and the corresponding error type is a major fault inside the target base station, and the repair measure is to replace the faulty part in time.

[0055] In this embodiment, by constructing a correspondence between the type identifier and the fault type and the corresponding repair measures, combined with the type identifier output by the deep learning model, the operation and maintenance personnel can intuitively and conveniently determine the fault type of the target base station and the repair measures that need to be taken. There is no need for the operation and maintenance personnel to make judgments based on their own professional skills and experience, which can reduce the technical burden of the operation and maintenance personnel and the probability of misjudgment caused by subjective factors, thereby improving the reliability and convenience of positioning base station fault detection.

[0056] Reference below Figure 4 , Figure 4 The process of obtaining a deep learning model in one embodiment of the method for detecting and positioning a base station failure of the present application is shown, such as Figure 4 As shown, the process may include the following steps.

[0057] Step 410: Obtain a training set and an initial deep learning model to be trained.

[0058] Each training sample in the training set includes sample observation data and ephemeris data with marked type identification.

[0059] In this embodiment, the initial deep learning model represents the deep learning model to be trained.

[0060] Step 420: Input the training set into the initial deep learning model, use the initial deep learning model to perform reasoning, and predict the type identification corresponding to each training sample.

[0061] Step 430: determine a loss value based on a preset loss function and the type identification of each training sample label and the predicted type identification, and adjust the parameters of the initial deep learning model based on the loss value.

[0062] As an example, the cross entropy function can be used as the loss function. After obtaining the loss value, the back propagation characteristics of the deep learning model are utilized to use the stochastic gradient descent method to update the parameters of the initial deep learning model.

[0063] Step 440: When the training termination condition is met, a trained deep learning model is obtained.

[0064] As an example, the training termination condition may be that the number of iterations reaches a preset number or the loss function converges.

[0065] pass Figure 4 The described embodiments can obtain, through training, the deep learning model required for detecting and positioning base station failures in any of the above embodiments.

[0066] In some optional implementations of this embodiment, the initial deep learning model includes a coding layer, a pooling layer, a decoding layer and an output layer; wherein the coding layer is configured to: encode the sample observation data and ephemeris data based on a multi-attention head mechanism to obtain the output data of the coding layer; the pooling layer is configured to: generate random noise, and merge the random noise with the output data of the coding layer and then perform pooling based on a multi-attention head mechanism to obtain the output data of the pooling layer; the decoding layer is configured to: decode the output data of the pooling layer based on a multi-attention head mechanism to obtain the output data of the coding layer; the output layer is configured to: perform feature mapping on the output data of the coding layer, and predict the type identifier corresponding to the training sample as the output result of the initial deep learning model.

[0067] In this embodiment, the encoding layer, pooling layer and decoding layer all adopt a multi-attention head mechanism, so that the deep learning model can extract feature information of each dimension in the input data based on the multi-attention head mechanism, and use it for reasoning, which helps to improve the prediction accuracy of the deep learning model; in addition, the pooling layer adopts a multi-attention head mechanism for pooling processing, which can reduce the adverse effects of abnormal data on the reasoning process, and help to improve the robustness of the deep learning model.

[0068] In an optional example of this embodiment, the output layer can be a fully connected layer, which is used to match the decoded data with the type identifier, so as to simplify the classification processing at the output, avoid transferring the training center to the initial output layer, and improve the overall performance of the deep learning model.

[0069] In some optional implementations of this embodiment, Figure 4 The illustrated embodiment may also include: step 450, determining the deep learning model as the first model; step 460, adjusting the hyperparameters of the deep learning model to reduce the number of parameters of the deep learning model, and determining the adjusted deep learning model as the second model.

[0070] As an example, after obtaining the trained deep learning model through the above steps 410 to 440, two model files of the deep learning model can be obtained, and one of the model files can be stored or deployed as the first model; at the same time, the hyperparameters of the model can be adjusted based on the other model file, for example, the number of multi-attention heads and / or the number of hidden layers can be reduced to reduce the number of parameters of the model, and the adjusted model can be stored or deployed as the second model.

[0071] In this embodiment, after the deep learning model is obtained through training, the deep learning model that retains all parameters can be used as the first model, and the deep learning model with reduced parameters can be used as the second model, so that different models can be selected for reasoning according to needs in actual applications.

[0072] In some optional implementations of this embodiment, the training samples can be Figure 5 The process shown is as follows: Figure 5 As shown, the process may include the following steps.

[0073] Step 510: Obtain historical data when a positioning base station fails.

[0074] The historical data includes observation data and ephemeris data when the positioning base station fails, the type of failure that occurred, and the repair measures used to repair the failure.

[0075] Step 520: Based on the correspondence between the type identifier and the fault type and the corresponding repair measure, determine the type identifier corresponding to the fault type and the repair measure in the historical data.

[0076] Step 530: Use the observation data and ephemeris data in the historical data as sample observation data and ephemeris data, and label the sample observation data and ephemeris data based on the determined type identifier to obtain the labeled sample observation data and ephemeris data as training samples.

[0077] In this embodiment, training samples can be constructed based on historical data when a positioning base station fails, so as to train a deep learning model, so that the deep learning model can learn the correspondence between the observation data and ephemeris data when a failure occurs and the fault type and repair measures, so that it can be used to detect the failure of the positioning base station.

[0078] An embodiment of the present application further provides a non-transitory computer storage medium, which stores a computer program. When the computer program is executed by a processor, the method for detecting a positioning base station failure in any of the above embodiments is implemented.

[0079] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0080] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A method for detecting a positioning base station failure, characterized in that: include: Acquire real-time observation data and ephemeris data of a target base station, where the target base station is a positioning base station that has a fault; The real-time observation data and ephemeris data are used as inputs of a pre-trained deep learning model, and the deep learning model is used for reasoning to output a prediction result, wherein the deep learning model is a model with memory processing capability; Based on the prediction result, the fault type of the target base station and the corresponding repair measures are determined.

2. The method according to claim 1, characterized in that The deep learning model includes a Transformer model that adopts a multi-attention head mechanism.

3. The method according to claim 2, characterized in that The deep learning model includes a first model and a second model, wherein the parameter amount of the first model is greater than the parameter amount of the second model; Performing reasoning using the deep learning model includes: Obtaining the real position of the target base station; Performing positioning calculation based on the real-time observation data and the ephemeris data to determine the predicted position of the target base station; determining a difference between the actual position and the predicted position; When the difference is not greater than a preset threshold, performing reasoning based on the second model; When the difference is greater than the preset threshold, reasoning is performed based on the first model.

4. The method according to claim 3, characterized in that: The real-time observation data and ephemeris data are used as inputs of a pre-trained deep learning model, and reasoning is performed using the deep learning model, including: Extracting correlation information between each data in the real-time observation data and the ephemeris data based on a multi-attention head mechanism; Based on a multi-attention head mechanism, the real-time observation data and the ephemeris data are fused with the correlation information, and bias processing and normalization processing are performed to obtain encoded data; Performing pooling processing on the encoded data based on a multi-attention head mechanism to obtain pooled data; Decoding the pooled data based on a multi-attention head mechanism to obtain decoded data; Feature mapping is performed on the decoded data, and a type identifier corresponding to the decoded data is output as a prediction result corresponding to the real-time observation data and the ephemeris data, wherein the type identifier represents a fault type and a corresponding repair measure.

5. The method according to claim 4, characterized in that Determining the fault type of the target base station and corresponding repair measures based on the prediction result includes: A predetermined correspondence between a type identifier and a fault type and a corresponding repair measure is determined, and the fault type and the corresponding repair measure corresponding to the type identifier included in the prediction result are determined as the fault type and the corresponding repair measure of the target base station.

6. The method according to claim 4, characterized in that The deep learning model is obtained in the following way: Acquire a training set and an initial deep learning model to be trained, wherein each training sample in the training sample set includes sample observation data and ephemeris data marked with a type identifier; Inputting the training set into the initial deep learning model, performing reasoning using the initial deep learning model, and predicting the type identification corresponding to each of the training samples; Determine a loss value based on a preset loss function and the type identification of each training sample label and the predicted type identification, and adjust the parameters of the initial deep learning model based on the loss value; When the training termination condition is met, the trained deep learning model is obtained.

7. The method according to claim 6, characterized in that After obtaining the trained deep learning model, the method further includes: Determine the deep learning model as the first model; The hyperparameters of the deep learning model are adjusted to reduce the number of parameters of the deep learning model, and the adjusted deep learning model is determined as the second model.

8. The method according to claim 6, characterized in that The initial deep learning model includes a coding layer, a pooling layer, a decoding layer and an output layer; wherein, The encoding layer is configured to: encode the sample observation data and ephemeris data based on a multi-attention head mechanism to obtain output data of the encoding layer; The pooling layer is configured to: generate random noise, and combine the random noise with the output data of the encoding layer and then perform pooling processing based on a multi-attention head mechanism to obtain the output data of the pooling layer; The decoding layer is configured to: decode the output data of the pooling layer based on a multi-attention head mechanism to obtain the output data of the encoding layer; The output layer is configured to: perform feature mapping on the output data of the encoding layer, and predict the type identifier corresponding to the training sample as the output result of the initial deep learning model.

9. The method according to claim 6, characterized in that The output layer is a fully connected layer.

10. The method according to claim 6, characterized in that The training samples are obtained in the following way: Acquire historical data when a positioning base station fails, the historical data including observation data and ephemeris data when the positioning base station fails, the type of failure that occurs, and repair measures adopted to repair the failure; Based on the correspondence between the type identifier and the fault type and the corresponding repair measure, determine the type identifier corresponding to the fault type and the repair measure in the historical data; The observation data and ephemeris data in the historical data are used as sample observation data and ephemeris data, and the sample observation data and ephemeris data are marked based on the determined type identifier to obtain the marked sample observation data and ephemeris data as the training sample.

11. A non-transitory computer storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for detecting a positioning base station failure according to any one of claims 1 to 10 is implemented.

Images