Deep learning model training method and positioning method

Abstract

The invention discloses a training method of a deep learning model, and the method comprises the steps: obtaining an observation pseudo-range and a theoretical pseudo-range of a target GNSS satellite at a target historical moment, and determining a target measurement noise value as a label of a training sample based on the observation pseudo-range and the theoretical pseudo-range, namely, a true value; and then, training the deep learning model by using the training sample and taking the target measurement noise value as a loss truth value until a predicted measurement noise value of the target GNSS satellite output by the deep learning model and a loss value output by a loss function corresponding to the target measurement noise value meet a preset condition, and completing training of the deep learning model. The theoretical pseudo-range (pseudo-range true value) is calculated based on the positioning result with the positioning precision meeting the preset precision requirement, so that the measured noise true value is constructed based on the observation pseudo-range and the theoretical pseudo-range of the target GNSS satellite at the target historical moment, reliable true value data can be provided for training a deep learning model, and the accuracy of the deep learning model is improved. And the accuracy of the measurement noise output by the trained deep learning model is ensured.

Description

Technical Field

[0001] This application relates to the field of localization, specifically to a training method and localization method for a deep learning model. Background Technology

[0002] Kalman filter-based positioning algorithms, when performing positioning based on Global Navigation Satellite System (GNSS) observations, need to obtain the measurement noise of each observation to form a measurement noise matrix for updating the filtering algorithm. Currently, the mainstream measurement noise estimation method is to obtain it through nonlinear transformation based on fields such as signal-to-noise ratio (SNR) in GNSS observations. However, on the one hand, the definition of SNR varies depending on the model and manufacturer; on the other hand, a single feature is difficult to accurately measure the current measurement noise.

[0003] The inventors of this application have discovered that the academic community has attempted to use trained measurement noise models to predict GNSS satellite measurement noise in real time in order to dynamically update the measurement noise matrix. However, due to the difficulty in obtaining true measurement noise data, existing measurement noise models have low accuracy in predicting measurement noise, affecting positioning accuracy and hindering their application in industry. Summary of the Invention

[0004] In view of this, this application provides a training method and a localization method for a deep learning model, which can provide the optimal estimation of measurement noise for the Kalman filter-based localization algorithm and ensure localization accuracy.

[0005] To solve the above problems, the technical solution provided in this application is as follows:

[0006] In a first aspect of this application, a method for training a deep learning model is provided, the method comprising:

[0007] Obtain the observed pseudorange and theoretical pseudorange of the target GNSS satellite at historical moments.

[0008] The target measurement noise value is determined based on the observed pseudorange and theoretical pseudorange of the target GNSS satellite;

[0009] Using training samples and the target measurement noise value as the ground truth loss, the deep learning model is trained until the predicted measurement noise value of the target GNSS satellite output by the deep learning model and the loss value output by the loss function corresponding to the target measurement noise value satisfy a preset condition. The training samples are a set of arbitrary satellite observations at any historical time. The input of each training sample includes at least: target GNSS satellite features and all GNSS satellite features. The output of each training sample is the measurement noise value of the target GNSS satellite.

[0010] In a second aspect of this application, a positioning method is provided, the method comprising:

[0011] Acquire the current GNSS measurement data observed by the target device, wherein the current GNSS measurement data includes: the target GNSS satellite characteristics observed by the target GNSS satellite at the current time and the total GNSS satellite characteristics observed by all GNSS satellites at the current time;

[0012] The current GNSS measurement data is input into a deep learning model to obtain the measurement noise value of the target GNSS satellite. The deep learning model is obtained by training using the method described in the first aspect.

[0013] The measurement noise matrix in the Kalman filter algorithm is updated based on the measured noise value to obtain the positioning result of the target device.

[0014] In a third aspect of this application, a training apparatus for a deep learning model is provided, the apparatus comprising:

[0015] The acquisition unit acquires the observed pseudorange and theoretical pseudorange of the target GNSS satellite at historical moments.

[0016] The determination unit determines the target measurement noise value based on the observed pseudorange and theoretical pseudorange of the target GNSS satellite;

[0017] The training unit is used to train the deep learning model using training samples and the target measurement noise value as the ground truth loss value, until the predicted measurement noise value of the target GNSS satellite output by the deep learning model and the loss value output by the loss function corresponding to the target measurement noise value satisfy a preset condition. The training samples are a set of arbitrary satellite observations at any historical time. The input of each training sample includes at least: the target GNSS satellite features and all GNSS satellite features. The output of each training sample is the measurement noise value of the target GNSS satellite.

[0018] In a fourth aspect of this application, a positioning device is provided, the device comprising:

[0019] The acquisition unit is used to acquire the current GNSS measurement data observed by the target device. The current GNSS measurement data includes: the target GNSS satellite characteristics observed by the target GNSS satellite at the current time and the total GNSS satellite characteristics observed by all GNSS satellites at the current time.

[0020] The acquisition unit is further configured to input the current GNSS measurement data into a deep learning model to obtain the measurement noise value of the target GNSS satellite, wherein the deep learning model is obtained by training using the method described in the first aspect;

[0021] The positioning unit is used to update the measurement noise matrix in the Kalman filter algorithm based on the measurement noise value, so as to obtain the positioning result of the target device.

[0022] In a fifth aspect of this application, an electronic device is provided, comprising:

[0023] One or more processors;

[0024] A memory, on which one or more programs are stored.

[0025] When the one or more programs are executed by the one or more processors, the one or more processors implement the training method as described in the first aspect, or the localization method as described in the second aspect.

[0026] In a sixth aspect of this application, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the training method as described in the first aspect, or the localization method as described in the second aspect.

[0027] In a seventh aspect of this application, a computer program product is provided that, when executed by a processor, implements the training method as described in the first aspect, or the localization method as described in the second aspect.

[0028] Therefore, this application has the following beneficial effects:

[0029] This application provides a method for training a deep learning model. In its implementation, the observed pseudorange and theoretical pseudorange of the target GNSS satellite at historical moments are obtained, and the target measurement noise value is determined as the label (or ground truth) of the training sample based on these two values. Then, using the training samples and the target measurement noise value as the ground truth loss, the deep learning model is trained until the predicted measurement noise value of the target GNSS satellite output by the deep learning model and the loss value output by the loss function corresponding to the target measurement noise value satisfy a preset condition, thus completing the training of the deep learning model. Each training sample includes features of the target GNSS satellite and features of all GNSS satellites. The deep learning model outputs the measurement noise value corresponding to the target GNSS satellite for each training sample. In the technical solution provided in this application, since the theoretical pseudorange is calculated in advance based on the positioning results that meet the preset accuracy requirements, it can be understood as the pseudorange true value. Therefore, constructing the measurement noise true value (target measurement noise value) based on the observed pseudorange and theoretical pseudorange of the target GNSS satellite at the target's historical moment can provide reliable true value data for training the deep learning model, ensuring the accuracy of the measurement noise output by the trained deep learning model, and enabling the real-time prediction of measurement noise through the deep learning model to be implemented in the industry.

[0030] Furthermore, based on the deep learning model trained above, this application also provides a positioning method. When positioning a target device, the method acquires the current GNSS measurement data observed by the target device. This current GNSS measurement data includes the target GNSS satellite features observed by the target GNSS satellite at the current time and the features of all GNSS satellites observed by all GNSS satellites at the current time. The acquired current GNSS measurement data is input into the trained deep learning model. The deep learning model is used to obtain the measurement noise value corresponding to the target GNSS satellite, and the measurement noise matrix in the Kalman filter algorithm is updated based on this measurement noise value to obtain the positioning result of the target device. Since the deep learning model provided in this application can output accurate measurement noise values ​​in real time, updating the measurement noise matrix in the Kalman filter algorithm based on these measurement noise values ​​can ensure the positioning accuracy of the positioning result. Attached Figure Description

[0031] Figure 1 A flowchart illustrating a training method for a deep learning model provided in an embodiment of this application;

[0032] Figure 2 This is a schematic diagram of the structure of a deep learning model provided in an embodiment of this application;

[0033] Figure 3 A flowchart of a positioning method provided in an embodiment of this application;

[0034] Figure 4 A schematic diagram of the structure of a training device for a deep learning model provided in an embodiment of this application;

[0035] Figure 5 This is a schematic diagram of the structure of a positioning device provided in an embodiment of this application;

[0036] Figure 6 This is a schematic diagram of the basic structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0037] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the embodiments of this application will be further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0038] To facilitate understanding and explanation of the technical solutions provided in the embodiments of this application, the background technology of this application will be described first.

[0039] The Kalman filter-based positioning algorithm (KALMA) can significantly improve the smoothness and accuracy of GNSS positioning, but it also has problems such as difficulty in filtering gross errors, resulting in unreliable confidence of positioning results. To solve this problem, existing technologies have proposed a measurement noise adaptive method to predict the measurement noise of GNSS satellites in order to dynamically update the measurement noise matrix in the KALMA algorithm.

[0040] However, the inventors of this application have discovered that due to differences in how different types (models) of equipment process the same GNSS physical quantities, significant differences in the measured noise values ​​obtained by different types of equipment in the same environment occur. Solving the problem of large differences in measured noise values ​​caused by model differences requires relying on the true noise data of different equipment models. Because true noise data is difficult to obtain, the adaptive measurement noise method cannot be used on a large scale in industry. The so-called true value refers to the actual value exhibited by the equipment under certain time and space conditions. For example, noise affecting GNSS positioning accuracy includes system noise and environmental noise; therefore, it is necessary to obtain the true system noise and environmental noise values ​​of the equipment under certain time and space conditions.

[0041] To address the issues of low accuracy and limited large-scale industrial application of existing measurement noise models, this application provides a deep learning model training method. Specifically, during deep learning model training, ground truth noise values ​​are constructed based on the observed pseudoranges and theoretical pseudoranges of the target GNSS satellite at historical times. These ground truth noise values ​​are then used as labels for training samples. The deep learning model is trained based on these training samples and labels until the predicted measurement noise value of the target GNSS satellite output by the deep learning model and the loss value corresponding to the ground truth noise value satisfy a preset condition. This yields a deep learning model capable of optimally predicting / estimating measurement noise. In other words, this application constructs ground truth measurement noise values ​​using the observed pseudoranges and theoretical pseudoranges of the target GNSS satellite at historical times. Since the theoretical pseudorange is pre-calculated based on positioning results that meet preset accuracy requirements, it is equivalent to the pseudorange ground truth, providing ground truth data and a large number of training samples for subsequent training. This ensures the accuracy of the measurement noise output by the trained deep learning model and enables its large-scale industrial application.

[0042] The deep learning models in this application include, but are not limited to, CNN, LSTM, etc.

[0043] To facilitate understanding of this application, the training process for training a measurement noise model provided in this application will be described in detail below with reference to the accompanying drawings.

[0044] See Figure 1This is a flowchart illustrating a training method for a deep learning model provided in an embodiment of this application. This method trains a deep learning model into a measurement noise model capable of predicting measurement noise in real time. The method includes:

[0045] S101: Obtain the observed pseudorange and theoretical pseudorange of the target GNSS satellite at the target's historical moment.

[0046] In GNSS positioning, pseudorange refers to the distance between a GNSS receiver (referred to as the receiver) and a GNSS satellite (referred to as the satellite). Observed pseudorange is the distance obtained by multiplying the propagation time of the ranging code signal transmitted by the satellite to the receiver by the speed of light during GNSS positioning. However, due to factors such as atmospheric delay and multipath effects, observed pseudorange usually deviates from the true distance between the satellite and the receiver. Therefore, observed pseudorange cannot be used as the true value of pseudorange.

[0047] For devices supporting GNSS positioning, all are equipped with the aforementioned receivers. These receivers can receive satellite signals from satellite positioning systems such as BeiDou (BDS), GPS, and GLONASS. Therefore, pseudorange can also be considered the distance between the device and the satellite. Theoretical pseudorange can be understood as the true pseudorange value, calculated in advance based on the known positions of the satellites and the device's known positioning results. It is important to note that to ensure the accuracy of the theoretical pseudorange, the positioning accuracy of the device's results needs to meet preset accuracy requirements. Whether the accuracy requirement is at the centimeter, decimeter, or meter level can be selected by those skilled in the art according to the needs of the specific business scenario; this application does not impose any restrictions.

[0048] Specifically, the process of obtaining the theoretical pseudorange includes:

[0049] Based on the historical positioning results corresponding to the target device's historical time, the theoretical pseudorange of the target GNSS satellite at the target historical time is obtained. For example, the historical positioning results can be positioning results obtained in advance using a high-precision positioning algorithm, and then generated after quality control and other methods to represent the true position of the target device. The target device can be any device equipped with the aforementioned receiver, such as a smartphone, tablet, aircraft, car, robot, etc., and this application does not impose any restrictions on this. The target device supports obtaining positioning results through high-precision positioning algorithms (such as RTK positioning algorithms).

[0050] It should be noted that in this embodiment, it is assumed that the measurement noise of GNSS satellites is only related to the hardware settings of the target device and is not related to the positioning algorithm. Based on this assumption, the positioning results obtained by the high-precision positioning algorithm can be used to construct the theoretical pseudorange of the ground value data for training the deep learning model.

[0051] S102: Determine the target measurement noise value based on the observed pseudorange and theoretical pseudorange of the target GNSS satellite.

[0052] In this embodiment, after obtaining the observed pseudorange and theoretical pseudorange of the target GNSS satellite, the ground truth of the measurement noise can be constructed using both. Specifically, the square of the difference between the observed pseudorange and the theoretical pseudorange of the target GNSS satellite is used as the target measurement noise value. During subsequent deep learning model training, this target measurement noise value is used as the label of the training sample, which makes the training process easier to converge.

[0053] S103: Using training samples, with the target measurement noise value as the true loss value, train the deep learning model until the predicted measurement noise value of the target GNSS model output by the deep learning model and the loss value output by the loss function corresponding to the target measurement noise value satisfy the preset conditions.

[0054] The training samples consist of a dataset of satellite observations at any historical moment. Each training sample includes at least the characteristics of the target GNSS satellite and the characteristics of all GNSS satellites. When a training sample is input into the deep learning model, the model uses it to predict the measurement noise value of the target GNSS satellite. In other words, when training the deep learning model, not only the characteristics of the target GNSS satellite communicating with the target device are referenced, but also the characteristics of other GNSS satellites, providing more reference information for the training of the deep learning model and improving the accuracy of its predictions.

[0055] Specifically, when training a deep learning model, the target measurement noise value is used as the ground truth measurement noise value of the training samples and input into the deep learning model. This yields the predicted measurement noise value output by the deep learning model for the training samples. Based on the loss value output by the loss function corresponding to the ground truth measurement noise value and the predicted measurement noise value of the training samples, it is determined whether to proceed to the next round of training. This process continues until the loss value between the predicted measurement noise value and the target measurement noise value output by the deep learning model meets a preset condition. This preset condition can be set according to the actual application, such as the loss value reaching a set threshold or the number of training iterations reaching a preset number.

[0056] Specifically, the target GNSS satellite features may include the actual observation data of the target GNSS satellite at the target historical time, the all GNSS satellite features include the actual observation data of each GNSS satellite at any historical time, and the all GNSS satellites also include the target GNSS satellite, and any historical time includes the target historical time.

[0057] The actual observation data may include any one of the following or other combinations thereof: observation data related to satellite positioning quality, observation data related to atmospheric error estimation quality, observation type, verification relationship between different observations, and observation data related to residuals.

[0058] Observational data related to satellite positioning quality can be read through the relevant interfaces of the target equipment, including but not limited to satellite elevation angle, signal-to-noise ratio, and carrier continuous observation duration.

[0059] Observational data related to the quality of atmospheric error estimation can be obtained through atmospheric error models, including but not limited to ionospheric error confidence and tropospheric error confidence.

[0060] The types of observations may include, but are not limited to, carrier wave observations, L5 observations, and GPS observations.

[0061] The verification relationships between observations include, but are not limited to, code-minus-carrier (CMC) verification and pseudorange rate smoothing error. Through the verification relationships, the consistency characteristics between different observations can be obtained.

[0062] The residual-related observation data are obtained based on the intermediate calculation results of the positioning algorithm, including but not limited to pseudorange residuals, pseudorange rate residuals, and phase residuals.

[0063] Furthermore, considering that different target devices may exhibit significant differences in noise measurement under the same environment due to model variations, this application provides another embodiment of a method for training a deep learning model. This embodiment uses the device model as a feature in the training samples to obtain a noise measurement model corresponding to that device model. The difference between this method embodiment and the previous embodiment is that, before inputting the training samples into the deep learning model, the method further includes: obtaining the device model features based on the device model parameters.

[0064] In this scenario, each training sample input to the deep learning model also includes aircraft type features concatenated with the target GNSS satellite features and all GNSS satellite features, respectively. That is, a training sample includes either GNSS satellite features + aircraft type features or all GNSS satellite features + aircraft type features. The pseudorange of the target GNSS satellites at historical target times originates from the target equipment.

[0065] Specifically, device model features can be obtained through the embedding layer of a deep learning model. The device model parameters are used as input to the embedding layer, which then extracts the device model features. In other words, the embedding layer learns about device model differences to increase the universality and robustness of the deep learning model.

[0066] The following combination Figure 2 The structure of the deep learning model in this application is described below, which should be understood as follows: Figure 2 The structure shown is merely an example; any deep learning model that can achieve the effects of this application falls within the scope of this application.

[0067] like Figure 2 As shown, the deep learning model includes an embedding layer, a first fully connected layer, a convolutional layer, a pooling layer, a concatenation layer, and a second fully connected layer. Specifically, the first fully connected layer is connected to the concatenation layer, and the concatenation layer is connected to the second fully connected layer; the convolutional layer is connected to the pooling layer, and the pooling layer is connected to the concatenation layer.

[0068] During training, the aircraft type features are extracted using the embedding layer, and these features, along with the target GNSS satellite features, are input into the first fully connected layer to obtain the observation features of the target GNSS satellite. All GNSS satellite features and the aircraft type features are then input into a convolutional layer, followed by pooling processing to obtain the environmental features. Finally, the observation features of the target GNSS satellite and the environmental features are input into a stitching layer, and then passed through a second fully connected layer connected to this stitching layer to obtain the predicted measurement noise value for the target GNSS satellite.

[0069] In other words, when training the deep learning model, aircraft model features and GNSS satellite features will be utilized to obtain the noise measurement model that best fits the current aircraft model. This reduces the problem of erroneous convergence caused by differences in aircraft models, thereby providing the optimal noise estimation to improve positioning accuracy. The deep learning model can be a convolutional neural network, a long short-term memory network, etc.

[0070] As can be seen, the technical solution provided in this application constructs ground truth values ​​for measurement noise based on the observed pseudorange and theoretical pseudorange of the target GNSS satellite when training the deep learning model. This provides reliable sample labels for training the deep learning model and improves the accuracy of noise prediction. Furthermore, when training the deep learning model, the model information of the target device can be added, enabling the trained deep learning model to be applied on a large scale and improving its applicability.

[0071] After training the deep learning model using the methods described above, it can be used to obtain the optimal noise estimation during actual localization. The application of the deep learning model will be explained below with reference to the accompanying diagram.

[0072] See Figure 3 The figure is a flowchart of a positioning method provided in an embodiment of this application, as shown below. Figure 3 As shown, the method includes:

[0073] S301: Acquire the current GNSS measurement data observed by the target device.

[0074] The current GNSS measurement data includes: the target GNSS satellite characteristics observed by the target GNSS satellite at the current time and the total GNSS satellite characteristics observed by all GNSS satellites at the current time.

[0075] S302: Input the current GNSS measurement data into the deep learning model to obtain the measurement noise value of the target GNSS satellite.

[0076] In passing Figure 1 After training the deep learning model, the method shown obtains the target GNSS satellite features observed by the target GNSS satellite corresponding to the target device at the current time, as well as the GNSS satellite features observed by all GNSS satellites at the current time, when performing real-time positioning of the target device. The obtained data is then input into the deep learning model to obtain the optimal measurement noise estimate of the target GNSS at the current time.

[0077] S303: Update the measurement noise matrix in the Kalman filter algorithm based on the measurement noise value to obtain the positioning result of the target device.

[0078] Those skilled in the art can implement the Kalman filter algorithm by referring to existing technologies, which will not be described in detail in this application. The core innovation of this application is to provide a scheme for training deep learning models. The measurement noise values ​​generated by the deep learning model trained using the scheme of this application can be used to update the measurement noise matrix in the Kalman filter algorithm.

[0079] As mentioned above, when training a deep learning model, the device model parameters can be further considered. Therefore, in practical applications, in order to obtain the optimal measurement noise estimation for the target device, the positioning method provided in this application can further obtain the device model parameters, obtain the model features based on the device model parameters, and input the model features into the deep learning model to obtain a noise estimation that conforms to the device model, thereby improving the positioning accuracy.

[0080] Specifically, the aircraft model parameters and current GNSS measurement data are input into a deep learning model. The embedding layer in this deep learning model, based on the aircraft model parameters, obtains the aircraft model characteristics of the target device. The aircraft model characteristics and the target GNSS satellite characteristics included in the current GNSS measurement data are input into the first fully connected layer of the deep learning model to obtain the observation characteristics of the target GNSS satellite. Simultaneously, the aircraft model characteristics and all GNSS satellite characteristics included in the current GNSS measurement data are input into the convolutional and pooling layers of the deep learning model to obtain environmental characteristics. The observation characteristics of the target GNSS satellite and the environmental characteristics are input into the stitching layer of the deep learning model. Through the second fully connected layer connected to the stitching layer in the deep learning model, the measurement noise value of the target GNSS satellite is obtained.

[0081] Since the deep learning model trained using the scheme provided in this application can output accurate measurement noise values ​​in real time, updating the measurement noise matrix in the Kalman filter algorithm based on these measurement noise values ​​can ensure the positioning accuracy of the positioning results. Therefore, the deep learning model and its training scheme provided in this application can be used on a large scale in the industry.

[0082] Based on the above method embodiments, this application also provides a training device and a localization device for a deep learning model, which will be described below with reference to the accompanying drawings.

[0083] See Figure 4 This figure is a structural diagram of a training device for a deep learning model provided in an embodiment of this application, as shown below. Figure 4 As shown, the device 400 includes: an acquisition unit 401, a determination unit 402, and a training unit 403.

[0084] Obtain the observed pseudorange and theoretical pseudorange of the target GNSS satellite at historical moments.

[0085] The target measurement noise value is determined based on the observed pseudorange and the theoretical pseudorange of the target GNSS satellite;

[0086] Using training samples and the target measurement noise value as the true loss value, the deep learning model is trained until the predicted measurement noise value of the target GNSS satellite output by the deep learning model and the loss value output by the loss function corresponding to the target measurement noise value satisfy a preset condition. The training samples are a set of any satellite observations at any historical time. The input of each training sample includes at least: the target GNSS satellite features and all GNSS satellite features. The output of each training sample is the measurement noise value of the target GNSS satellite.

[0087] In some implementations, the pseudorange of the target GNSS satellite at a target historical moment is obtained from the target device, and the acquisition unit 401 is further used to acquire the model characteristics of the target device based on the model parameters of the target device;

[0088] The input to each training sample further includes: the aircraft model features concatenated with the target GNSS satellite features and all GNSS satellite features respectively.

[0089] In some implementations, the deep learning model includes an embedding layer, and the acquisition unit 401 is specifically used to take the model parameters of the target device as input to the embedding layer and extract the model features of the target device through the embedding layer.

[0090] In some implementations, the training unit 403 is specifically used to input the target GNSS satellite features and the aircraft type features into the first fully connected layer of the deep learning model to obtain the observation features of the target GNSS satellite; input all GNSS satellite features and the aircraft type features into the convolutional layer and pooling layer of the deep learning model to obtain environmental features; input the observation features of the target GNSS satellite and the environmental features into the stitching layer of the deep learning model, and obtain the predicted measurement noise value of the target GNSS satellite through the second fully connected layer connected to the stitching layer in the deep learning model.

[0091] In some embodiments, the determining unit 402 is specifically used to take the square of the difference between the observed pseudorange of the target GNSS satellite and the theoretical pseudorange as the target measurement noise value.

[0092] In some implementations, the target GNSS satellite characteristics include: actual observation data of the target GNSS satellite at the target's historical time.

[0093] The characteristics of all GNSS satellites include: actual observation data of all GNSS satellites at any historical moment, wherein all GNSS satellites include the target GNSS satellite, and any historical moment includes the target historical moment.

[0094] In some implementations, the actual observation data includes at least one of the following data or a combination thereof:

[0095] Observational data related to satellite positioning quality, observational data related to atmospheric error estimation quality, observation types, verification relationships between different observations, and observational data related to residuals.

[0096] In some embodiments, the acquisition unit 401 is further configured to obtain the theoretical pseudorange of the target GNSS satellite at the target historical time based on the historical positioning results corresponding to the historical time of the target, wherein the positioning accuracy of the historical positioning results meets the preset accuracy requirements.

[0097] See Figure 5 This figure is a structural diagram of a positioning device provided in an embodiment of this application, as shown below. Figure 5 As shown, the device 500 includes an acquisition unit 501 and a positioning unit 502.

[0098] The acquisition unit 501 is used to acquire the current GNSS measurement data observed by the target device. The current GNSS measurement data includes: the target GNSS satellite characteristics observed by the target GNSS satellite at the current time and the total GNSS satellite characteristics observed by all GNSS satellites at the current time.

[0099] The acquisition unit 501 is further configured to input at least the current measurement data into a deep learning model to obtain the measurement noise value of the target GNSS satellite, wherein the deep learning model is obtained through... Figure 2 The method described is used for training;

[0100] The positioning unit 502 is used to update the measurement noise matrix in the Kalman filter algorithm based on the measurement noise value in order to obtain the positioning result of the target device.

[0101] In some embodiments, the acquisition unit 501 is further configured to acquire the model parameters of the target device; input the model parameters and the current GNSS measurement data into the deep learning model, and the embedding layer of the deep learning model obtains the model features of the target device based on the model parameters; input the target GNSS satellite features and the model features into the first fully connected layer of the deep learning model to obtain the observation features of the target GNSS satellite; input all GNSS satellite features and the model features into the convolutional layer and pooling layer of the deep learning model to obtain environmental features; input the observation features of the target GNSS satellite and the environmental features into the stitching layer of the deep learning model, and obtain the measured noise value of the target GNSS satellite through the second fully connected layer connected to the stitching layer in the deep learning model.

[0102] It should be noted that the specific implementation of each unit in the above device embodiments can be found in the relevant descriptions in the above method embodiments. The division of units in this application embodiment is illustrative and only represents a logical functional division; in actual implementation, there may be other division methods. The functional units in this application embodiment can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. For example, in the above embodiments, the processing unit and the sending unit can be the same unit or different units. The integrated unit can be implemented in hardware or as a software functional unit.

[0103] Based on the deep learning model training or localization method provided in the above method embodiments, this application also provides an electronic device, including: one or more processors; a storage device storing one or more programs thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any of the above embodiments.

[0104] The following is for reference. Figure 6 This document illustrates a structural schematic diagram of an electronic device 600 suitable for implementing embodiments of this application. The terminal devices in these embodiments may include, but are not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Android Devices), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs (televisions), desktop computers, etc. Figure 6 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0105] like Figure 6 As shown, electronic device 600 may include a processing device (e.g., a central processing unit, a graphics processor, etc.) 601, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 602 or a program loaded from storage device 608 into random access memory (RAM) 603. RAM 603 also stores various programs and data required for the operation of electronic device 600. Processing device 601, ROM 602, and RAM 603 are interconnected via bus 604. Input / output (I / O) interface 605 is also connected to bus 604.

[0106] Typically, the following devices can be connected to I / O interface 605: input devices 606 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 607 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 608 including, for example, magnetic tapes, hard disks, etc.; and communication devices 609. Communication device 609 allows electronic device 600 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 6 An electronic device 600 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0107] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 609, or installed from a storage device 608, or installed from a ROM 602. When the computer program is executed by the processing device 601, it performs the functions defined in the methods of the embodiments of this application.

[0108] The electronic device provided in this application embodiment belongs to the same inventive concept as the deep learning model training method or localization method provided in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0109] Based on the deep learning model training or localization method provided in the above method embodiments, this application provides a computer-readable medium storing a computer program thereon, wherein the program, when executed by a processor, implements the deep learning model training or localization method as described in any of the above embodiments.

[0110] It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium, a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0111] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0112] The aforementioned computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the training method or localization method of the aforementioned deep learning model.

[0113] The units described in the embodiments of this application can be implemented in software or in hardware. The name of the unit / module does not necessarily limit the unit itself; for example, a voice data acquisition module can also be described as a "data acquisition module".

[0114] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0115] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems or apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.

[0116] It should be understood that in this application, "at least one (item)" means one or more, and "more than one" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0117] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0118] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for training a deep learning model, characterized in that, The method includes: Obtain the observed pseudorange and theoretical pseudorange of the target GNSS satellite at historical moments. The target measurement noise value is determined based on the observed pseudorange and theoretical pseudorange of the target GNSS satellite; Using training samples and the target measurement noise value as the true loss value, the deep learning model is trained until the predicted measurement noise value of the target GNSS satellite output by the deep learning model and the loss value output by the loss function corresponding to the target measurement noise value satisfy a preset condition. The training samples are a set of any satellite observations at any historical time. The input of each training sample includes at least: the target GNSS satellite features and all GNSS satellite features. The output of each training sample is the measurement noise value of the target GNSS satellite.

2. The method according to claim 1, characterized in that, The pseudorange of the target GNSS satellite at a historical moment is obtained from the target equipment, and the method further includes: Based on the model parameters of the target device, the model characteristics of the target device are obtained; The input to each training sample further includes: the aircraft model features concatenated with the target GNSS satellite features and all GNSS satellite features respectively.

3. The method according to claim 2, characterized in that, The deep learning model includes an embedding layer, and the acquisition of the target device's model features specifically involves: The model parameters of the target device are used as input to the embedding layer, and the model features of the target device are extracted through the embedding layer.

4. The method according to claim 3, characterized in that, The step of training the deep learning model using training samples and the target measurement noise as the true loss value specifically includes: The target GNSS satellite features and the aircraft type features are input into the first fully connected layer of the deep learning model to obtain the observation features of the target GNSS satellite. The environmental features are obtained by inputting all the GNSS satellite features and the aircraft model features into the convolutional and pooling layers of the deep learning model. The observation features of the target GNSS satellite and the environmental features are input into the stitching layer of the deep learning model. The predicted measurement noise value of the target GNSS satellite is obtained through the second fully connected layer in the deep learning model that is connected to the stitching layer.

5. The method according to any one of claims 1-4, characterized in that, The step of determining the target measurement noise value based on the observed pseudorange and theoretical pseudorange of the target GNSS satellite includes: The square of the difference between the observed pseudorange and the theoretical pseudorange of the target GNSS satellite is taken as the target measurement noise value.

6. The method according to any one of claims 1-4, characterized in that, The target GNSS satellite characteristics include: the actual observation data of the target GNSS satellite at the target's historical time. The characteristics of all GNSS satellites include: actual observation data of all GNSS satellites at any historical moment, wherein all GNSS satellites include the target GNSS satellite, and any historical moment includes the target historical moment.

7. The method according to claim 6, characterized in that, The actual observation data shall include at least one of the following data or a combination thereof: Observational data related to satellite positioning quality, observational data related to atmospheric error estimation quality, observation types, verification relationships between different observations, and observational data related to residuals.

8. The method according to any one of claims 1-4, characterized in that, The method further includes: Based on the historical positioning results corresponding to the historical time of the target, the theoretical pseudorange of the target GNSS satellite at that historical time is obtained, and the positioning accuracy of the historical positioning results meets the preset accuracy requirements.

9. A positioning method, characterized in that, The method includes: Acquire the current GNSS measurement data observed by the target device, wherein the current GNSS measurement data includes: the target GNSS satellite characteristics observed by the target GNSS satellite at the current time and the total GNSS satellite characteristics observed by all GNSS satellites at the current time; The current GNSS measurement data is input into a deep learning model trained by any one of claims 1-8 to obtain the measurement noise value of the target GNSS satellite; The measurement noise matrix in the Kalman filter algorithm is updated based on the measurement noise value to obtain the positioning result of the target device.

10. The method according to claim 9, characterized in that, The method further includes: Obtain the model parameters of the target device; The step of inputting the current GNSS measurement data into a deep learning model trained by any one of claims 1-8 to obtain the measurement noise value of the target GNSS satellite specifically includes: The aircraft model parameters and the current GNSS measurement data are input into the deep learning model. The embedding layer of the deep learning model obtains the aircraft model characteristics of the target device based on the aircraft model parameters. The target GNSS satellite features and the aircraft type features are input into the first fully connected layer of the deep learning model to obtain the observation features of the target GNSS satellite. The environmental features are obtained by inputting all the GNSS satellite features and the aircraft model features into the convolutional and pooling layers of the deep learning model. The observation characteristics of the target GNSS satellite and the environmental characteristics are input into the stitching layer of the deep learning model. The measured noise value of the target GNSS satellite is obtained through the second fully connected layer in the deep learning model that is connected to the stitching layer.

Images