A vehicle positioning method and device based on an informer network and a computer

Abstract

This invention relates to a vehicle positioning method, apparatus, and computer based on Informer networks. The method includes: determining whether the vehicle's satellite positioning function is normal; clustering the vehicle's traffic environment image data using a Gaussian mixture model to obtain multiple traffic environment cluster datasets; sorting the vehicle's motion feature data under each traffic condition according to the vehicle's travel time sequence to obtain a motion feature time series dataset; training Informer neural network models using the motion feature time series datasets corresponding to multiple traffic conditions; determining the Informer neural network prediction model using a bulldozer distance analysis method when the vehicle's satellite positioning function is abnormal; and predicting the vehicle's trajectory using the Informer neural network prediction model. This invention uses Informer networks to predict vehicle trajectories, effectively solving the problem of how to accurately predict vehicle positions when GPS fails for extended periods.

Description

Technical Field

[0001] This invention relates to the field of vehicle navigation and positioning technology, and specifically to a vehicle positioning method, device, and computer based on Informer networks. Background Technology

[0002] Vehicle navigation and positioning technology is a key technology in intelligent transportation systems (ITS), and many ITS applications and location-based services (LBS) require information about vehicle location. Global Positioning System (GPS) motion relative positioning technology, as a high-precision positioning method, is widely used for vehicle location. However, traditional GPS pure motion relative positioning technology often fails to provide accurate positioning or even loses its positioning capability in certain constrained urban road environments, such as roads with tall buildings, roads surrounded by dense forests, and tunnels, due to satellite signal blockage or instability. This results in the inability to provide reliable and accurate positioning for vehicles, leading to missing vehicle tracks. Summary of the Invention

[0003] To address the technical problems in vehicle navigation and positioning technology, such as the inability to provide reliable positioning when satellite signals are blocked or unstable, this invention provides a vehicle positioning method, device, and computer based on Informer networks.

[0004] The technical solution of the present invention to solve the above-mentioned technical problems is as follows:

[0005] Determine if the vehicle's satellite positioning function is working properly;

[0006] When the vehicle's satellite positioning function is normal, the traffic environment image data of the vehicle is clustered using a Gaussian mixture model to obtain a multi-class traffic environment cluster dataset; wherein, each traffic environment cluster dataset corresponds to a traffic condition.

[0007] The multiple motion feature data of the vehicles corresponding to each type of traffic condition are sorted according to the vehicle's travel time to obtain a multi-type motion feature time series dataset; wherein, each type of motion feature time series dataset corresponds to one type of traffic condition.

[0008] Informer neural network models are trained using motion feature time series data from each of the aforementioned motion feature time series datasets to obtain multiple Informer neural network prediction models.

[0009] When the vehicle's satellite positioning function is abnormal, the bulldozer distance analysis method is used to compare the vehicle's current traffic environment image data with the traffic environment image data in multiple traffic environment clustering datasets to determine the vehicle's current traffic condition category, and select the corresponding Informer neural network prediction model based on the determined current traffic condition category.

[0010] The selected Informer neural network prediction model is used to predict the vehicle's trajectory when the satellite positioning function malfunctions.

[0011] The beneficial effects of this invention are as follows: By employing an Informer network to predict vehicle trajectories, this invention effectively solves the problem of accurately predicting vehicle positions when satellite positioning functions are unavailable for extended periods. In the method of judging traffic conditions using a combination of Gaussian mixture models and KL divergence, the bulldozer distance analysis method is used instead of KL divergence, solving the problem of KL divergence asymmetry and demonstrating superior judgment performance compared to KL divergence when measuring non-overlapping distributions. By applying the Informer network model to predict vehicle trajectories, the self-attention distillation technique, probabilistic sparse self-attention mechanism, and generative decoder of the Informer network model are used as the core technologies of the basic network, improving training and inference speeds, reducing network memory overhead, and enhancing prediction accuracy.

[0012] Based on the above technical solution, the present invention can be further improved as follows.

[0013] Furthermore, the traffic environment image data of the vehicles is clustered using a Gaussian mixture model to obtain a multi-class traffic environment clustering dataset, including the following steps:

[0014] Select a feature vector from the traffic environment image data;

[0015] The probability distribution of the feature vector in the traffic environment image data of the vehicle is calculated using a Gaussian mixture model;

[0016] All the traffic environment image data are clustered according to the corresponding probability distribution to obtain a multi-class traffic environment cluster dataset.

[0017] Furthermore, the probability distribution of the feature vector in the traffic environment image data of the vehicle is calculated using a Gaussian mixture model, including the following steps:

[0018] Using formula Calculate the probability distribution in the traffic environment image data of the vehicle; where y represents the feature vector, p(y) represents the probability distribution, and N(y|μ) k ,∑ k) represents the k-th component of the Gaussian mixture model, π k This represents the weight of each component in the Gaussian mixture model.

[0019] Furthermore, the bulldozer distance analysis method is used to classify all the traffic environment cluster datasets to obtain multi-class traffic environment cluster datasets, including the following steps:

[0020] The current traffic environment image data of the vehicle is compared with traffic environment image data in multiple traffic environment clustering datasets using the bulldozer distance analysis method to determine the current traffic condition category of the vehicle, including the following steps:

[0021] Using the bulldozer distance analysis method, the current traffic condition category of the vehicle is determined based on the distance between the target probability distribution and the pre-stored probability distribution; wherein, the target probability distribution is the probability distribution of the feature vector in the current traffic environment image data of the vehicle, and the pre-stored probability distribution is the probability distribution of the feature vector in the traffic environment image data of each traffic environment cluster dataset.

[0022] Furthermore, using the bulldozer distance analysis method, the current traffic condition category of the vehicle is determined based on the distance value between the target probability distribution and the pre-stored probability distribution, including the following steps:

[0023] The distance between the target probability distribution and the pre-stored probability distribution is calculated using the bulldozer distance calculation formula; the bulldozer distance calculation formula is as follows:

[0024] Calculate the distance between the target probability distribution and the pre-stored probability distribution; where s represents the target probability distribution of the current traffic situation, and s(x) and It is defined in the complex plane R n probability distribution on, It is R n ×R n Joint distribution on, For s and The set of all joint distributions γ that are combined, s and yes The marginal distributions are obtained by sampling (x,y)~y from the joint distribution γ. It is the kernel density function with respect to x and y. For all t > 0, t = 1 / α, α is 1, and p is 2. Indicates s and The distance between the corresponding pre-stored probability distributions;

[0025] Minimum The corresponding current traffic conditions are categorized as follows: kind; This indicates the traffic condition category corresponding to the pre-stored traffic environment image data of the vehicle.

[0026] The beneficial effects of adopting the above-mentioned further technical solution are that the improved bulldozer distance analysis method replaces KL divergence, which solves the problem of KL divergence asymmetry. Moreover, when measuring non-overlapping distributions, the judgment effect is better than KL divergence. Compared with the general bulldozer distance method, the solution time can be shortened while improving the solution accuracy.

[0027] Furthermore, the multiple motion feature data of the vehicles corresponding to each type of traffic condition are sorted according to the vehicle's travel time order to obtain a multi-type motion feature time series dataset, including the following steps:

[0028] The multiple motion feature data of the vehicles corresponding to each type of traffic condition are sorted according to the driving time of the vehicles to obtain multiple initial time series datasets.

[0029] Empty and abnormal data are removed from each type of initial time series dataset to obtain multiple types of motion feature time series datasets.

[0030] Furthermore, an Informer neural network model is trained using the motion feature time series data from each type of motion feature time series dataset to obtain multiple Informer neural network prediction models, including the following steps:

[0031] All motion feature time series data in each type of motion feature time series dataset are normalized to obtain multiple normalized datasets; wherein each normalized dataset corresponds to one type of traffic condition.

[0032] The Informer neural network model is trained using data from multiple normalized datasets to obtain multiple Informer neural network prediction models.

[0033] The Informer neural network prediction model is validated and / or tested using a portion of the normalized dataset.

[0034] Furthermore, the vehicle's motion characteristic data includes the vehicle's driving timestamp, heading angle, pitch angle, latitude, longitude, altitude, and speed.

[0035] To address the aforementioned technical problems, this invention also provides a vehicle positioning device based on an Informer network, the specific technical solution of which is as follows:

[0036] A vehicle positioning device based on an Informer network includes:

[0037] The function judgment module is used to determine whether the vehicle's satellite positioning function is normal;

[0038] The model training module is used to cluster the traffic environment image data of the vehicle using a Gaussian mixture model when the vehicle's satellite positioning function is normal, to obtain a multi-class traffic environment clustering dataset; to sort the multiple motion feature data of the vehicle corresponding to each traffic condition according to the vehicle's travel time order, to obtain a multi-class motion feature time series dataset; and to train an Informer neural network model using the motion feature time series data in each of the motion feature time series datasets, to obtain multiple Informer neural network prediction models; wherein, each traffic environment clustering dataset corresponds to one traffic condition.

[0039] The trajectory determination module is used to compare the current traffic environment image data of the vehicle with the traffic environment image data in multiple traffic environment clustering datasets using the bulldozer distance analysis method when the vehicle's satellite positioning function is abnormal, determine the current traffic condition category of the vehicle, and select the corresponding Informer neural network prediction model based on the determined current traffic condition category; and use the selected Informer neural network prediction model to predict the vehicle's driving trajectory when the satellite positioning function is abnormal.

[0040] To address the aforementioned technical problems, this invention also provides a vehicle positioning device based on an Informer network, the specific technical solution of which is as follows:

[0041] A computer includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the above-described vehicle positioning method based on an Informer network. Attached Figure Description

[0042] Figure 1 This is a flowchart of a vehicle localization method based on an Informer network according to an embodiment of the present invention. Detailed Implementation

[0043] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0044] like Figure 1 As shown, this embodiment provides a vehicle positioning method based on an Informer network, including the following steps:

[0045] S1. Determine if the vehicle's satellite positioning function is normal; the determination of satellite positioning function can be made by determining whether the GPS signal is available.

[0046] S2. When the vehicle's satellite positioning function is normal, real-time traffic environment image data and vehicle motion characteristic data are collected; the traffic environment image data of the vehicle are clustered using a Gaussian mixture model to obtain multiple traffic environment cluster datasets; each traffic environment cluster dataset corresponds to a traffic condition; the vehicle's motion characteristic data includes at least the vehicle's driving timestamp t, heading angle ψ, pitch angle θ, latitude B, longitude L, altitude H, and driving speed v.

[0047] The process involves using a Gaussian mixture model to cluster the traffic environment image data of the vehicles, resulting in a multi-class traffic environment clustering dataset, including the following steps:

[0048] S20. Select a feature vector from the traffic environment image data of the vehicle;

[0049] S21. Calculate the probability distribution of the feature vector in the traffic environment image data of the vehicle using a Gaussian mixture model; the specific steps are as follows:

[0050] The feature vectors are calculated using a Gaussian mixture model in the traffic environment image data of the vehicle. The probability distribution in the image data; where y represents the feature vector, p(y) represents the probability distribution, and N(y|μ) k ,∑ k ) represents the k-th component of the Gaussian mixture model, π k This represents the weight of each component in the Gaussian mixture model.

[0051] S22. Cluster all the traffic environment image data of the vehicles according to the corresponding probability distribution to obtain a multi-class traffic environment cluster dataset.

[0052] S3. Sort the multiple motion feature data of the vehicles corresponding to each type of traffic condition according to the vehicle's travel time order to obtain a multi-type motion feature time series dataset. The specific steps are as follows:

[0053] S30. Normalize all the motion feature time series data in each type of motion feature time series dataset to obtain multiple normalized datasets; wherein, each normalized dataset corresponds to one type of traffic condition.

[0054] The specific calculation formula for data normalization is as follows:

[0055]

[0056] In the formula, μ represents the mean of each data point in the time series dataset, σ represents the standard deviation of each data point in the time series dataset, and x represents each data point in the time series dataset. * This represents the normalized values ​​of each data point in the time series dataset; the normalized values ​​are then included in the normalized dataset.

[0057] S31. Train the Informer neural network model using data from multiple normalized datasets to obtain multiple Informer neural network prediction models; specifically, select 80% of the data from the normalized time series datasets as training data to train the Informer neural network model to obtain multiple Informer neural network prediction models.

[0058] S32. Validate and / or test the Informer neural network prediction model using a portion of the normalized dataset. Select 10% of the data in the normalized time series dataset as validation data, and select the remaining 10% of the data in the normalized time series dataset as test data. The validation data is used to validate the trained Informer neural network model, and the test data is used to test the accuracy of the trained Informer neural network model.

[0059] S4. Train an Informer neural network model using the motion feature time series data in each type of motion feature time series dataset to obtain multiple Informer neural network prediction models.

[0060] After the data has been processed in the previous step, the input sequence is set to x1, x2, ..., x3. T The sequence is multidimensional data, including the vehicle's travel timestamp t, heading angle ψ, pitch angle θ, latitude B, longitude L, altitude H, and travel speed v. The initial parameter settings are as follows: encoder input dimension 7, encoder output dimension 7, model size 512, encoder layers 2, decoder layers 1, activation function GeLU, learning rate 0.0001, and batch size of input training data 32. The model training process is as follows:

[0061] The model input consists of filtered and smoothed feature scalars It consists of a local timestamp (PE) and a global timestamp (SE); the conversion formula is:

[0062]

[0063] In the formula: i∈{1,...,L}x}, where α is a factor of the size between balancing scalar mapping and local / global embedding.

[0064] The formula corresponding to the characteristic scalar The specific operation involves converting the i-dimensional vector to a 512-dimensional vector using Conv1D. The local timestamp uses PositionalEmbedding from Transformer, calculated using the following formula:

[0065]

[0066]

[0067] Where d model Given the input feature dimension, the global timestamp uses a fully connected layer to map the input timestamp to a 512-dimensional embedding.

[0068] Specific methods for generating encoders include:

[0069] Unified input The input to the encoder part of the model first undergoes sparse self-attention computation in the attention module. Each key only focuses on u main queries, where Q is the query vector, K is the key vector, and V is the value vector. The calculation formula is shown below:

[0070]

[0071] in, It is a sparse matrix of the same size as Q and contains only the sparse metric M(q). i The top-u queries under (K). Add a sampling factor c, setting u = clnLq. First, randomly sample c × lnL keys for each query and calculate the sparsity score M(q) for each query. i ,K). Q i ,K i V i The i-th row represents Q, K, and V respectively, and d represents q. i The dimension, and L k =L q =L, sparsity metric M(q) i The approximate formula for calculating K is:

[0072]

[0073] Then, select the N queries with the highest sparsity scores, where N is c×lnL by default. Only calculate the dot product of the N queries and the key, and do not calculate the remaining LN queries.

[0074] The output after sparse self-attention computation contains redundant combinations of V values. Therefore, a distillation operation is needed to assign higher weights to the dominant features with primary characteristics, and to generate a focused self-attention feature map in the next layer. This is specifically achieved through four Convld convolutional layers and one max pooling layer.

[0075] After multiple sparse self-attention layer calculations and distillation operations, the input to the Informer neural network model's decoder is obtained. The decoder used in the Informer neural network model is similar to a traditional decoder; to generate long sequence outputs, the decoder requires the following input:

[0076]

[0077] in, The input sequence to the Decoder is the original sequence. The sequence to be predicted (padded with 0s) is then passed through a mask-based sparsity self-attention layer, which prevents each position from focusing on future positions, thus avoiding autoregression. The output of this layer, along with the output of the encoder, is then passed to a multi-head attention layer for computation. Finally, a fully connected layer is used to obtain the final output. The predicted output and the true value are compared using the loss function MSE, calculated as follows:

[0078]

[0079] Where n is the number of samples, y i For real data, The model is used to predict data. It iterates continuously until the training conditions terminate, ultimately generating the desired model. Specifically, training termination occurs when the model reaches the required number of iterations or when an early stopping mechanism is triggered because the MSE (Mean Sequence Size) fails to decrease. Finally, the Informer prediction model with the minimum loss function is obtained.

[0080] S5. When the vehicle's satellite positioning function is abnormal, the current traffic environment image data of the vehicle is compared with the traffic environment image data in the multiple traffic environment clustering datasets using the bulldozer distance analysis method to determine the current traffic condition category of the vehicle, and the corresponding Informer neural network prediction model is selected based on the determined current traffic condition category of the vehicle.

[0081] The specific steps are as follows: Using the bulldozer distance analysis method, the current traffic condition category of the vehicle is determined based on the distance value between the target probability distribution and the pre-stored probability distribution; wherein, the target probability distribution is the probability distribution of the feature vector in the current traffic environment image data of the vehicle, and the pre-stored probability distribution is the probability distribution of the feature vector in the traffic environment image data of each traffic environment cluster dataset.

[0082] The method of using bulldozer distance analysis to determine the current traffic condition category of the vehicle based on the distance between the target probability distribution and the pre-stored probability distribution includes the following steps:

[0083] Using the bulldozer distance calculation formula

[0084] Calculate the distance between the target probability distribution and the pre-stored probability distribution; where s represents the target probability distribution of the current traffic situation, and s(x) and It is defined in the complex plane R n probability distribution on, It is R n ×R n Joint distribution on, For s and The set of all joint distributions γ that are combined, s and yes The marginal distributions are obtained by sampling (x,y)~y from the joint distribution γ. It is the kernel density function with respect to x and y. This function is positive definite for all cases t > 0. In practical solutions, it is generally set to t = 1 / α. In this invention, the empirical value of α is 1, and the empirical value of p is 2. Indicates s and The distance between the corresponding pre-stored probability distributions;

[0085] Minimum The corresponding current traffic conditions are categorized as follows: kind; This indicates the traffic condition category corresponding to the pre-stored traffic environment image data of the vehicle. They are respectively as well as Represents the pre-stored probability distribution of low-complexity traffic condition classes. Represents the pre-stored probability distribution of a medium-complexity traffic condition class. This represents the pre-stored probability distribution of high-complexity traffic condition classes. When the traffic environment in which a vehicle is located changes, the vehicle motion feature data collected during this period will be packaged, identified, and categorized into the vehicle motion feature dataset corresponding to the traffic condition.

[0086] S6. Predict the vehicle's trajectory when the satellite positioning function malfunctions using the selected Informer neural network prediction model.

[0087] This invention utilizes an Informer network to predict vehicle trajectories, effectively addressing the problem of accurately predicting vehicle positions when satellite positioning is unavailable for extended periods. In the method of judging traffic conditions using a combination of Gaussian mixture models and KL divergence, an improved bulldozer distance analysis method replaces KL divergence, resolving the asymmetry problem of KL divergence. Furthermore, when measuring non-overlapping distributions, it outperforms KL divergence in judgment. Compared to general bulldozer distance methods, the improved bulldozer distance method used in this invention can shorten the solution time while improving accuracy. By applying an Informer network model to predict vehicle trajectories, the self-attention distillation technique, probabilistic sparse self-attention mechanism, and generative decoder of the Informer network model are used as the core technologies of the basic network, improving training and inference speeds, reducing network memory overhead, and enhancing prediction accuracy.

[0088] Example 2

[0089] Based on Embodiment 1, this embodiment provides a vehicle positioning device based on an Informer network, comprising:

[0090] The function judgment module is used to determine whether the vehicle's satellite positioning function is normal;

[0091] The model training module is used to cluster the traffic environment image data of the vehicle using a Gaussian mixture model when the vehicle's satellite positioning function is normal, to obtain a multi-class traffic environment clustering dataset; to sort the multiple motion feature data of the vehicle corresponding to each traffic condition according to the vehicle's travel time order, to obtain a multi-class motion feature time series dataset; and to train an Informer neural network model using the motion feature time series data in each of the motion feature time series datasets, to obtain multiple Informer neural network prediction models; wherein, each traffic environment clustering dataset corresponds to one traffic condition.

[0092] The trajectory determination module is used to compare the vehicle's current traffic environment image data with traffic environment image data in multiple traffic environment clustering datasets using a bulldozer distance analysis method when the vehicle's satellite positioning function is abnormal. This determines the vehicle's current traffic condition category, and selects the corresponding Informer neural network prediction model based on the determined category. The selected Informer neural network prediction model is then used to predict the vehicle's trajectory when the satellite positioning function is abnormal. In this embodiment, the function determination module, model training module, and trajectory determination module can be computer function system modules, specific computer hardware with corresponding functions, or specific computers with corresponding functions.

[0093] This invention utilizes an Informer network to predict vehicle trajectories, effectively solving the problem of accurately predicting vehicle location when GPS is unavailable for extended periods. Simultaneously, by employing Gaussian Mixture Model (GMM) and IMP-Wasserstein distance methods, different traffic conditions can be categorized. The aim is to differentiate the trained model according to different traffic conditions. When GPS is unavailable, selecting the appropriate model based on the current traffic environment significantly improves accuracy. Here, GMM stands for Gaussian Mixture Model, and Wasserstein distance is also known as Earth Mover's distance (EMD). IMP-Wasserstein is an improved bulldozer distance; its specific calculation formula is shown in Example 1. By employing self-attention distillation, probabilistic sparse self-attention mechanisms, and a generative decoder as the core technologies of the Informer network, training and inference speeds are improved, network memory overhead is reduced, and prediction accuracy is enhanced. This invention can be used to assist in vehicle localization and improve vehicle localization capabilities in complex environments.

[0094] Example 3

[0095] Based on Embodiment 1, this embodiment provides a computer, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the vehicle positioning method based on the Informer network in Embodiment 1. The memory in this embodiment can be internal computer memory, external computer storage device, hard disk, mobile storage device, or cloud storage, etc.; by implementing the vehicle positioning method based on the Informer network using a computer program, computational efficiency is improved.

[0096] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the concept and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A vehicle localization method based on Informer networks, characterized in that, Includes the following steps: Determine if the vehicle's satellite positioning function is working properly; When the vehicle's satellite positioning function is normal, the traffic environment image data of the vehicle is clustered using a Gaussian mixture model to obtain a multi-class traffic environment cluster dataset; wherein, each traffic environment cluster dataset corresponds to a traffic condition. The motion feature data of multiple vehicles corresponding to each type of traffic condition are sorted according to the driving time of the vehicles to obtain a multi-type motion feature time series dataset; wherein, each type of motion feature time series dataset corresponds to one type of traffic condition. Informer neural network models are trained using motion feature time series data from each of the aforementioned motion feature time series datasets to obtain multiple Informer neural network prediction models. When the vehicle's satellite positioning function is abnormal, the bulldozer distance analysis method is used to compare the vehicle's current traffic environment image data with the traffic environment image data in multiple traffic environment clustering datasets to determine the vehicle's current traffic condition category, and select the corresponding Informer neural network prediction model based on the determined current traffic condition category. The selected Informer neural network prediction model is used to predict the vehicle's trajectory when the satellite positioning function malfunctions.

2. The vehicle positioning method based on Informer networks according to claim 1, characterized in that, The traffic environment image data of the vehicles is clustered using a Gaussian mixture model to obtain a multi-class traffic environment clustering dataset, including the following steps: Select a feature vector from the traffic environment image data; The probability distribution of the feature vector in the traffic environment image data is calculated using a Gaussian mixture model. All the traffic environment image data are clustered according to the corresponding probability distribution to obtain a multi-class traffic environment cluster dataset.

3. The vehicle positioning method based on Informer networks according to claim 2, characterized in that, The probability distribution of the feature vector in the traffic environment image data of the vehicle is calculated using a Gaussian mixture model, including the following steps: Using formula Calculate the probability distribution in the traffic environment image data of the vehicle; where y represents the feature vector, p(y) represents the probability distribution, and N(y|μ) k ,∑ k ) represents the k-th component of the Gaussian mixture model, π k This represents the weight of each component in the Gaussian mixture model.

4. The vehicle positioning method based on an Informer network according to claim 2, characterized in that, The current traffic environment image data of the vehicle is compared with traffic environment image data in multiple traffic environment clustering datasets using the bulldozer distance analysis method to determine the current traffic condition category of the vehicle, including the following steps: Using the bulldozer distance analysis method, the current traffic condition category of the vehicle is determined based on the distance between the target probability distribution and the pre-stored probability distribution; wherein, the target probability distribution is the probability distribution of the feature vector in the current traffic environment image data of the vehicle, and the pre-stored probability distribution is the probability distribution of the feature vector in the traffic environment image data of each traffic environment cluster dataset.

5. The vehicle positioning method based on Informer networks according to claim 4, characterized in that, Using the bulldozer distance analysis method, the current traffic condition category of the vehicle is determined based on the distance value between the target probability distribution and the pre-stored probability distribution, including the following steps: The distance between the target probability distribution and the pre-stored probability distribution is calculated using the bulldozer distance calculation formula; the bulldozer distance calculation formula is as follows: Where s represents the target probability distribution of the current traffic situation, and s(x) and It is defined in the complex plane R n probability distribution on, It is R n ×R n Joint distribution on, For s and The set of all joint distributions γ that are combined, s and yes The marginal distributions are obtained by sampling (x,y)~y from the joint distribution γ. It is the kernel density function with respect to x and y. For all t > 0, t = 1 / α, α is 1, and p is 2. Indicates s and The distance between the corresponding pre-stored probability distributions; Minimum The corresponding current traffic conditions are categorized as follows: kind; This indicates the traffic condition category corresponding to the pre-stored traffic environment image data of the vehicle.

6. The vehicle positioning method based on Informer networks according to claim 1, characterized in that, The multiple motion feature data of the vehicles corresponding to each type of traffic condition are sorted according to the vehicle's travel time order to obtain a multi-type motion feature time series dataset, including the following steps: The multiple motion feature data of the vehicles corresponding to each type of traffic condition are sorted according to the driving time of the vehicles to obtain multiple initial time series datasets. Empty and abnormal data are removed from each type of initial time series dataset to obtain multiple types of motion feature time series datasets.

7. The vehicle localization method based on Informer networks according to claim 6, characterized in that, Train an Informer neural network model using the motion feature time series data from each of the aforementioned motion feature time series datasets to obtain multiple Informer neural network prediction models, including the following steps: All motion feature time series data in each type of motion feature time series dataset are normalized to obtain multiple normalized datasets; wherein each normalized dataset corresponds to one type of traffic condition; The Informer neural network model is trained using data from multiple normalized datasets to obtain multiple Informer neural network prediction models. The Informer neural network prediction model is validated and / or tested using a portion of the normalized dataset.

8. The vehicle positioning method based on Informer networks according to claim 1, characterized in that, The vehicle's motion characteristic data includes the vehicle's driving timestamp, heading angle, pitch angle, latitude, longitude, altitude, and speed.

9. A vehicle positioning device based on an Informer network, characterized in that, include: The function judgment module is used to determine whether the vehicle's satellite positioning function is normal; The model training module is used to cluster the traffic environment image data of the vehicle using a Gaussian mixture model when the vehicle's satellite positioning function is normal, to obtain a multi-class traffic environment clustering dataset; to sort the multiple motion feature data of the vehicle corresponding to each traffic condition according to the vehicle's travel time order, to obtain a multi-class motion feature time series dataset; and to train an Informer neural network model using the motion feature time series data in each of the motion feature time series datasets, to obtain multiple Informer neural network prediction models; wherein, each traffic environment clustering dataset corresponds to one traffic condition. The trajectory determination module is used to compare the current traffic environment image data of the vehicle with the traffic environment image data in multiple traffic environment clustering datasets using the bulldozer distance analysis method when the vehicle's satellite positioning function is abnormal, determine the current traffic condition category of the vehicle, and select the corresponding Informer neural network prediction model based on the determined current traffic condition category; and use the selected Informer neural network prediction model to predict the vehicle's driving trajectory when the satellite positioning function is abnormal.

10. A computer, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method as described in any one of claims 1-8.

Images