Smart rail train positioning method, device, electronic equipment and storage medium

Abstract

The present invention provides a method, device, electronic device, and storage medium for positioning a smart rail train, relating to the field of rail transit technology. The method comprises: obtaining inertial navigation information, wheel speed information, and a previous state vector of the smart rail train at a previous moment; determining a predicted state vector at the current moment based on the inertial navigation information and the previous state vector at the previous moment; inputting the inertial navigation information and the wheel speed information at the previous moment into an error time series model, and outputting error posture information at the current moment; the error time series model is trained based on posture correction information corresponding to each historical timestamp after posture graph optimization correction; and determining corrected positioning information of the smart rail train based on the error posture information and the predicted state vector. The present invention can achieve error correction of positioning information, thereby improving positioning accuracy, route accuracy, and driving safety.

Description

Technical Field

[0001] The present invention relates to the field of rail transit technology, and in particular to a method, device, electronic equipment and storage medium for positioning a smart rail train. Background Art

[0002] When the smart rail train is traveling on public roads, it is separated from the fixed track and its movement on the road shows characteristics similar to that of moving in a two-dimensional plane. Therefore, it is necessary to use the various sensors it carries to make real-time predictions of its own driving position and automatically adjust the driving route and speed of the vehicle body based on the real-time prediction values ​​to ensure the accuracy of the predetermined driving route and driving safety.

[0003] In the existing technology, without relying on external perception sensors such as cameras and lidar and prior information, the inertial measurement unit (IMU) is generally combined with the global navigation satellite system (GNSS). The IMU is used to obtain speed and displacement, and the GNSS is used to receive low-frequency satellite position signals to locate the position of the smart rail train in real time. However, the accuracy of the automotive-grade IMU is low, and the error accumulates exponentially in the process of obtaining speed and displacement. The GNSS measurement data also has errors, and the signal is intermittent and unavailable when driving within the city, resulting in large errors in the predicted positioning information, which in turn leads to low accuracy of the driving route and low driving safety. Summary of the Invention

[0004] The present invention provides a smart rail train positioning method, device, electronic device and storage medium to address the defects in the prior art of large errors in the predicted positioning information, which in turn leads to low accuracy of the driving route and low driving safety. The method realizes error correction of the positioning information and improves positioning accuracy, accuracy of the driving route and driving safety.

[0005] The present invention provides a method for positioning a smart rail train, comprising:

[0006] Obtain the inertial navigation information, wheel speed information and previous state vector of the smart rail train at the previous moment;

[0007] Determining a predicted state vector at a current moment based on the inertial navigation information at a previous moment and the previous state vector;

[0008] Inputting the inertial navigation information and the wheel speed information at the previous moment into an error time series model, and outputting the error posture information at the current moment; the error time series model is trained based on the posture correction information corresponding to each historical time stamp after the posture graph is optimized and corrected;

[0009] Based on the error posture information and the predicted state vector, the corrected positioning information of the smart rail train is determined.

[0010] According to the smart rail train positioning method provided by the present invention, the posture correction information corresponding to each historical timestamp is obtained based on the following steps:

[0011] Acquire first posture information corresponding to the first ground transponder and second posture information corresponding to the second ground transponder;

[0012] Determining, based on the inertial navigation history information and the first posture information, posture prediction information corresponding to each of the historical timestamps between the first ground transponder and the second ground transponder;

[0013] A pose graph is optimized based on each of the pose prediction information and the second pose information to obtain pose correction information corresponding to each of the historical timestamps after correction.

[0014] According to the smart rail train positioning method provided by the present invention, the pose graph optimization is performed based on each of the pose prediction information and the second pose information to obtain the pose correction information corresponding to each of the corrected historical timestamps, including:

[0015] Determining posture change information between adjacent historical timestamps based on each of the posture prediction information;

[0016] Determine accumulated information of posture changes between the posture prediction information corresponding to the last historical timestamp and the second posture information;

[0017] Determining a cumulative error based on the accumulated pose change information, the second pose information, and pose prediction information corresponding to the last historical timestamp;

[0018] Taking minimization of the cumulative error as the optimization goal, each of the posture prediction information and each of the posture change information is iteratively optimized to obtain the posture correction information corresponding to each of the historical timestamps after correction.

[0019] According to the smart rail train positioning method provided by the present invention, determining the posture prediction information corresponding to each historical timestamp between the first ground transponder and the second ground transponder based on the inertial navigation history information and the first posture information includes:

[0020] Performing Kalman filtering based on the inertial navigation history information and the first pose information to predict and obtain first pose prediction information corresponding to each of the historical timestamps;

[0021] For each of the historical timestamps between the first ground transponder and the second ground transponder, if there is Global Positioning System (GNSS) historical information in the historical timestamp, determining the pose prediction information corresponding to the historical timestamp based on the GNSS historical information and the first pose prediction information;

[0022] When the GNSS historical information does not exist at the historical timestamp, the first posture prediction information is determined as the posture prediction information corresponding to the historical timestamp.

[0023] According to the smart rail train positioning method provided by the present invention, determining the predicted state vector at the current moment based on the inertial navigation information at the previous moment and the previous state vector includes:

[0024] Performing Kalman filtering based on the inertial navigation information at the previous moment and the previous state vector to predict a first state vector at the current moment;

[0025] Upon receiving the GNSS measurement information, determining a predicted state vector at the current moment based on the first state vector and the GNSS measurement information;

[0026] In a case where the GNSS measurement information is not received, the first state vector is determined as the predicted state vector at the current moment.

[0027] According to the smart rail train positioning method provided by the present invention, determining the predicted state vector at the current moment based on the first state vector and GNSS measurement information includes:

[0028] Determining a state covariance matrix corresponding to the first state vector;

[0029] determining a Kalman gain based on the state covariance matrix and a GNSS noise covariance matrix corresponding to the GNSS measurement information;

[0030] determining an error state vector based on the Kalman gain, the GNSS measurement information, and the first state vector;

[0031] Based on the error state vector, the predicted state vector at the current moment is determined.

[0032] According to the smart rail train positioning method provided by the present invention, determining the state covariance matrix corresponding to the first state vector includes:

[0033] Get the previous covariance matrix, rotation matrix and noise covariance matrix of the previous moment;

[0034] Determining a state transition matrix at the previous moment based on a time interval between the current moment and the previous moment, the inertial navigation information at the previous moment, and the rotation matrix;

[0035] A state covariance matrix corresponding to the first state vector is determined based on the noise propagation matrix at a previous moment, the state transfer matrix, the previous covariance matrix, and the noise covariance matrix.

[0036] The present invention also provides a smart rail train positioning device, comprising:

[0037] The acquisition module is used to obtain the inertial navigation information, wheel speed information and previous state vector of the smart rail train at the previous moment;

[0038] a prediction module, configured to determine a predicted state vector at a current moment based on the inertial navigation information at a previous moment and the previous state vector;

[0039] An output module, configured to input the inertial navigation information and the wheel speed information at the previous moment into an error time series model, and output error posture information at the current moment; the error time series model is trained based on posture correction information corresponding to each historical time stamp after posture graph optimization correction;

[0040] A positioning module is used to determine the corrected positioning information of the smart rail train based on the error posture information and the predicted state vector.

[0041] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the smart rail train positioning method as described above is implemented.

[0042] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements any of the above-mentioned smart rail train positioning methods.

[0043] The smart rail train positioning method, device, electronic device and storage medium provided by the present invention obtain the inertial navigation information, wheel speed information and previous state vector of the smart rail train at the previous moment, and then predict the predicted state vector at the current moment based on the inertial navigation information and the previous state vector. The error timing model learns the errors of each historical time stamp obtained by the posture correction information corresponding to each historical time stamp after posture graph optimization correction. After the inertial navigation information and wheel speed information are input into the error timing model, the error posture information at the current moment is output. Through the error posture information, the predicted state vector at the current moment is corrected in real time in a data-driven manner to obtain the corrected positioning information at the current moment, thereby improving the positioning accuracy of the smart rail train, and thereby improving the accuracy of the smart rail train's route and the driving safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0045] Figure 1 This is one of the flow charts of the smart rail train positioning method provided by an embodiment of the present invention;

[0046] Figure 2 This is a schematic diagram of a segmented travel section of a smart rail train provided by an embodiment of the present invention;

[0047] Figure 3 This is the second flow chart of the smart rail train positioning method provided by an embodiment of the present invention;

[0048] Figure 4 1 is a schematic diagram of a process for pose graph optimization according to an embodiment of the present invention;

[0049] Figure 5 is a schematic diagram of an example structure of an initial pose graph provided by an embodiment of the present invention;

[0050] Figure 6 1 is a schematic structural diagram of a smart rail train positioning device provided by an embodiment of the present invention;

[0051] Figure 7 It is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0052] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0053] In view of the problem that the positioning information predicted in the prior art has large errors, which leads to low accuracy of the driving route and low driving safety, the embodiment of the present invention provides a method for positioning a smart rail train. Figure 1 This is one of the flow charts of the smart rail train positioning method provided by an embodiment of the present invention. Figure 1 As shown, the method includes:

[0054] Step 110: Obtain the inertial navigation information, wheel speed information, and previous state vector of the smart rail train at the previous moment.

[0055] It should be noted that the appearance of the smart rail train is very similar to that of a tram. It integrates the advantages of trams and buses, uses road markings instead of tracks, and uses automatic tracking and trajectory following technology to control the smart rail train to travel intelligently along the marking lines. It has the characteristics of low cost, zero emissions, intelligence and high efficiency. Figure 2 This is a schematic diagram of the segmented travel of the smart rail train provided by an embodiment of the present invention, such as Figure 2 As shown in the figure, ground transponders are deployed along the route of the smart rail train. The ground transponders, onboard host and communication equipment together constitute the Balise Transmission Module (BTM). In addition, there is a public section between adjacent ground transponders. The smart rail train travels in a nearly two-dimensional plane on the public section. Since it is separated from the fixed track, the route of the smart rail train in the public section is not fixed. The position of the smart rail train in real time needs to be determined by various sensors on the vehicle to ensure the accuracy of the route and driving safety.

[0056] Optionally, the smart rail train is equipped with an inertial measurement unit (IMU), a global navigation satellite system (GNSS), and a wheel speed sensor. The inertial navigation information measured by the IMU at the previous moment may include the three-axis angular velocity (ω x ,ω y ,ω z ) and the three-axis acceleration (f x ,f y ,f z) By integrating the three-axis angular velocity, the previous moment's position information can be obtained. By integrating the three-axis acceleration, the previous moment's velocity and displacement information can be obtained. GNSS can receive low-frequency position signals from satellites and express them in the form of longitude and latitude. Wheel speed sensors can be installed in the bogies of smart rail trains. Using the grating principle, wheel speed information can be measured. This wheel speed information can be interpreted as the real-time rotational speed of the wheel axle, facilitating the determination of the speed information of the smart rail train.

[0057] In addition, smart rail trains can also be equipped with cameras and lidar sensors. The cameras can capture road features such as traffic signals, traffic congestion, and road markings to assist in the precise positioning of the smart rail train. Lidar sensors can be used to measure the distance between the smart rail train and other targets in the surrounding environment, such as other vehicles, pedestrians, and buildings, to identify and avoid obstacles during driving and ensure driving safety. It should be noted that the embodiments of the present invention are applied to real-time positioning scenarios that do not rely on external perception sensors such as cameras and lidars and prior information.

[0058] It should be noted that, in the embodiment of the present invention, the body coordinate system of the smart rail train is overlapped with the coordinate system of the IMU to obtain the coordinate system s.

[0059] Optionally, the previous state vector of the smart rail train at the previous moment can be s t-1 =[p t-1 ,v t-1 ,q t-1 ] T , where s t-1 represents the previous state vector, p t-1 Indicates the position information of the previous moment, v t-1 Indicates the velocity information at the previous moment, q t-1 Represents the posture vector at the previous moment.

[0060] Step 120: Determine a predicted state vector at the current moment based on the inertial navigation information at the previous moment and the previous state vector.

[0061] Specifically, after obtaining the inertial navigation information and the previous state vector at the previous moment, the predicted state information at the current moment can be predicted based on the kinematic relationship and the Kalman filter.

[0062] Furthermore, the determining the predicted state vector at the current moment based on the inertial navigation information at the previous moment and the previous state vector includes:

[0063] Performing Kalman filtering based on the inertial navigation information at the previous moment and the previous state vector to predict a first state vector at the current moment;

[0064] Upon receiving the GNSS measurement information, determining a predicted state vector at the current moment based on the first state vector and the GNSS measurement information;

[0065] In a case where the GNSS measurement information is not received, the first state vector is determined as the predicted state vector at the current moment.

[0066] Specifically, after obtaining the inertial navigation information and the previous state vector at the previous moment, the time interval between the previous moment and the current moment is determined, and the rotation matrix corresponding to the conversion from the coordinate system s to the coordinate system n in the northeast sky form is determined. Then, using formula (1), the first state vector at the current moment is obtained according to the kinematic relationship. Formula (1) is:

[0067]

[0068] Among them, Δt represents the time interval between the previous moment and the current moment, which is determined by the IMU measurement frequency. Indicates the rotation matrix corresponding to the coordinate system n converted from the coordinate system s to the northeast sky form at the previous moment, g n represents the gravitational acceleration in the northeastern sky coordinate system n, f t-1 Represents the linear acceleration in the inertial navigation information at the previous moment, ω t-1 Indicates the angular velocity in the inertial navigation information of the previous moment, Represents quaternion multiplication, q(*) represents the generation of quaternion representation from the rotation axis angle representation, p t1 Represents the first position information in the first state vector at the current moment, v t1 Represents the first velocity information in the first state vector at the current moment, q t1 represents the first posture vector in the first state vector at the current moment, [p t1 ,v t1 ,q t1 ] T Represents the first state vector at the current moment.

[0069] In addition, since the measurement frequency of IMU is high frequency and the measurement frequency of GNSS is low frequency, there are some timestamps that can receive both inertial navigation information and GNSS measurement information, and there are some timestamps that only receive inertial navigation information. If GNSS measurement information is not received, the first state vector determined above can be determined as the predicted state vector at the current moment. If GNSS measurement information is received, since the measurement accuracy of the automotive-grade IMU is low and the error is large, the GNSS measurement information can be used to correct the first state vector, thereby obtaining a more accurate predicted state vector, which can be expressed as s t =[pt ,v t ,q t ] T , s t Represents the predicted state vector at the current moment, p t Represents the predicted position information in the predicted state vector, v t Represents the predicted speed information in the predicted state vector, q t represents the predicted pose vector in the predicted state vector.

[0070] Furthermore, determining the state covariance matrix corresponding to the first state vector includes:

[0071] Get the previous covariance matrix, rotation matrix and noise covariance matrix of the previous moment;

[0072] Determining a state transition matrix at the previous moment based on a time interval between the current moment and the previous moment, the inertial navigation information at the previous moment, and the rotation matrix;

[0073] A state covariance matrix corresponding to the first state vector is determined based on the noise propagation matrix at a previous moment, the state transfer matrix, the previous covariance matrix, and the noise covariance matrix.

[0074] Specifically, based on the idea of ​​Kalman filter, while calculating the first state vector, formula (2) can be used to calculate the linear acceleration f in the inertial navigation information of the previous moment according to the time interval Δt and the previous moment. t-1 , rotation matrix and the unit matrix, the state transfer matrix after linearization of the motion equation in the error Kalman filter at the previous moment is calculated, formula (2) is:

[0075]

[0076] Among them, F t-1 represents the state transfer matrix at the previous moment, I represents the identity matrix, and (*)^ represents the calculation of the antisymmetric matrix.

[0077] Afterwards, the noise propagation matrix at the previous moment is determined using the identity matrix. The noise propagation matrix at the previous moment can be expressed as Among them, L t-1 represents the noise propagation matrix at the previous moment.

[0078] Then, using formula (3), the state covariance matrix corresponding to the first state vector is calculated based on the noise propagation matrix, state transfer matrix, previous covariance matrix and noise covariance matrix at the previous moment. Formula (3) is:

[0079] Pt1=Ft-1 P t-1 F t-1 T +L t-1 Q t-1 L t-1 T

[0080] Among them, P t1 represents the state covariance matrix corresponding to the first state vector, P t-1 represents the previous covariance matrix at the previous moment, Q t-1 Represents the noise covariance matrix in the motion equation at the previous moment. The motion equation is related to the kinematic relationship of the vehicle, and the noise covariance matrix is ​​related to the properties of the IMU itself.

[0081] Furthermore, determining the predicted state vector at the current moment based on the first state vector and GNSS measurement information includes:

[0082] Determining a state covariance matrix corresponding to the first state vector;

[0083] determining a Kalman gain based on the state covariance matrix and a GNSS noise covariance matrix corresponding to the GNSS measurement information;

[0084] determining an error state vector based on the Kalman gain, the GNSS measurement information, and the first state vector;

[0085] Based on the error state vector, the predicted state vector at the current moment is determined.

[0086] Specifically, if GNSS measurement information is received, the GNSS measurement information can be converted to obtain the GNSS measurement value y t =[y x ,y y ,y z ], which is used to represent the GNSS measurement components of the three axes in three-dimensional space. Then, using formula (4), the Kalman gain is calculated according to the state covariance matrix corresponding to the first state vector and the GNSS noise covariance matrix. Formula (4) is:

[0087] K t =P t1 H t (H t P t1 H t T +R) -1

[0088] Among them, K t represents the Kalman gain, H tis determined based on the identity matrix, and H t Indicated as H t =[I, 0, 0] T ,R represents the GNSS noise covariance matrix corresponding to the GNSS measurement value, and the GNSS noise covariance matrix is ​​related to the GNSS own properties.

[0089] After the Kalman gain is calculated, the error state vector can be calculated using formula (5) according to the state covariance matrix corresponding to the Kalman gain, GNSS measurement value and the first state vector. Formula (5) is:

[0090] δs t =K t (y t -P t1 )

[0091] Among them, δs t represents the error state vector, and the error state vector δs t The error position information δp is included t , error velocity information δv t and the error posture vector δq t , that is, δs t =[δp t ,δv t ,δq t ].

[0092] After obtaining the error state vector, the predicted state vector can be determined based on the operation of the error state vector and the corresponding elements in the first state vector. The predicted state vector can be calculated based on formula (6), which is:

[0093]

[0094] in, The quaternion extracted from the error state vector is converted into an axis angle (rotation vector) representation.

[0095] In addition, after obtaining the predicted state vector, the predicted state covariance matrix P corresponding to the predicted state vector can be calculated using formula (7): t , to update the state covariance matrix P corresponding to the first state vector t1 , formula (7) is:

[0096] P t =(IK t H t )P t1

[0097] Step 130: Input the inertial navigation information and the wheel speed information at the previous moment into an error time series model, and output the error posture information at the current moment; the error time series model is trained based on the posture correction information corresponding to each historical time stamp after the posture graph is optimized and corrected.

[0098] Specifically, due to the existence of errors in inertial navigation information, GNSS measurements, and system modeling, and the accumulation of these errors over time, there are still errors in the determined predicted state vector. Figure 3 This is a second flow chart of the method for locating a smart rail train provided by an embodiment of the present invention. Figure 3 As shown, in this embodiment of the present invention, a data-driven approach is used to construct a deep learning-based error time series model. This error time series model learns the error between the predicted state vector output by the Kalman filter and the true position value. After the error time series model completes, the inertial navigation information and wheel speed information from the previous moment are input into the error time series model, and the error pose information at the current moment is output. This error pose information facilitates correction of the predicted state vector to improve the positioning accuracy of the smart rail train.

[0099] It should be noted that, in order to prevent the error time series model from being overfitted or underfitted due to its high complexity, in the embodiment of the present invention, for the error posture information, the error time series model only outputs the position error information p err And heading angle error information θ err , that is, the error pose information may include position error information p err And heading angle error information θ err .

[0100] Optionally, during the positioning process of the smart rail train, since the error posture information has time series correlation, that is, the error posture information gradually accumulates with time, and the error posture information at the current moment is affected by multiple time steps in the past, an LSTM (Long Short-Term Memory Network) can be used to construct an initial error time series model. After training the initial error time series model, an error time series model is obtained. By using time series correlation, the error posture information can be more accurately predicted, thereby achieving more precise positioning.

[0101] After constructing the initial error time series model, the inertial navigation history information and wheel speed history information are obtained and arranged in chronological order to construct a training data set. At the same time, when the smart rail train passes the ground transponder, the ground transponder can provide the position truth value to the onboard host of the smart rail train. The position truth value is converted into (x BTM ,y BTM ,θBTM ), where (x BTM ,y BTM ) represents the two-dimensional coordinate information of the ground transponder, θ BTM Indicates the heading angle of the smart rail train when it passes the ground transponder. Since the position truth value is the absolute position truth value, in the embodiment of the present invention, the position truth value provided by the ground transponder is used as the pose truth value, and the initial error time series model is trained using the pose truth value and the above-mentioned training data set to obtain the error time series model. At the same time, the smart rail train positioning method provided in the embodiment of the present invention is applied to positioning within the common section between adjacent ground transponders.

[0102] Furthermore, the posture correction information corresponding to each historical timestamp is obtained based on the following steps:

[0103] Acquire first posture information corresponding to the first ground transponder and second posture information corresponding to the second ground transponder;

[0104] Determining, based on the inertial navigation history information and the first posture information, posture prediction information corresponding to each of the historical timestamps between the first ground transponder and the second ground transponder;

[0105] A pose graph is optimized based on each of the pose prediction information and the second pose information to obtain pose correction information corresponding to each of the historical timestamps after correction.

[0106] Since the actual layout of ground transponders on the route of the smart rail train is non-continuous, only two position truth values ​​can be provided between adjacent ground transponders as pose truth values ​​for model training, that is, the acquisition of position truth values ​​is intermittent, low-frequency and non-equidistant. The embodiment of the present invention provides a smart rail train positioning method that is applied to the common section between adjacent ground transponders, resulting in the loss of position truth values ​​in the common section. Therefore, in the embodiment of the present invention, it is necessary to restore the high-frequency position truth values ​​between adjacent ground transponders based on the low-frequency position truth values ​​provided by the ground transponders, so as to train the initial error time series model in conjunction with the training data set. Specifically, Figure 4 : is a flow chart of pose graph optimization provided by an embodiment of the present invention, such as Figure 4As shown, for any pair of ground transponders, the above-mentioned Kalman filtering process is started when the smart rail train passes the first ground transponder, and the first position true value corresponding to the first ground transponder is obtained. The first position true value is converted to obtain the first pose information. The first pose information is used as the initial state value at t=0 in the Kalman filter. Based on the first pose information and the inertial navigation history information, the Kalman filter is performed to obtain the pose prediction information corresponding to each historical timestamp. Each pose prediction information is between the first ground transponder and the second ground transponder. An initial pose graph is constructed based on each pose prediction information, and the pose prediction information in the initial pose graph is iteratively optimized based on the second ground transponder. The pose prediction information corresponding to each historical timestamp predicted by the Kalman filter is corrected in the form of pose graph optimization (PGO). Since the pose graph optimization has a good effect, the corrected pose information obtained after correction can be used as the position true value to train the initial error time series model to solve the problem of sparse position true value determined by the ground transponder.

[0107] It should be noted that since the pose graph optimization process is performed offline, the pose correction information calculated in this process can only be used to train the initial error timing model and cannot meet the real-time positioning requirements.

[0108] Furthermore, the pose graph optimization is performed based on each of the pose prediction information and the second pose information to obtain the pose correction information corresponding to each of the corrected historical timestamps, including:

[0109] Determining posture change information between adjacent historical timestamps based on each of the posture prediction information;

[0110] Determine accumulated information of posture changes between the posture prediction information corresponding to the last historical timestamp and the second posture information;

[0111] Determining a cumulative error based on the accumulated pose change information, the second pose information, and pose prediction information corresponding to the last historical timestamp;

[0112] Taking minimization of the cumulative error as the optimization goal, each of the posture prediction information and each of the posture change information is iteratively optimized to obtain the posture correction information corresponding to each of the historical timestamps after correction.

[0113] Specifically, after obtaining the pose prediction information between each historical timestamp, the pose change information between adjacent historical timestamps is determined based on the pose prediction information corresponding to each adjacent historical timestamp, and the first pose information, the second pose information, and each pose prediction information are used as nodes, and the pose change information is used as the measurement edge between adjacent nodes. The initial pose graph is constructed based on each node and each measurement edge. In the initial pose graph, the first pose information is used as the first node, the second pose information is used as the last node, and the pose change accumulation information between the pose prediction information corresponding to the last historical timestamp and the second pose information is used as the last measurement edge in the initial pose graph. For example, Figure 5 is a schematic diagram of an example structure of the initial pose graph provided by an embodiment of the present invention, such as Figure 5 As shown, taking the first ground transponder as node A and the second ground transponder as node B as an example, the posture prediction information corresponding to the last historical timestamp is node B'. The error in the initial posture graph is caused by the two nodes of the last measured edge, that is, the error is caused by the posture prediction information and the second posture information corresponding to the last historical timestamp at both ends of the posture change accumulation information. Therefore, after determining the posture change accumulation information, formula (8) is used to calculate the cumulative error based on the posture change accumulation information, the posture prediction information corresponding to the last historical timestamp, and the second posture information. Formula (8) is:

[0114]

[0115] in, and Both represent cumulative errors, M represents the total number of nodes, Indicates the pose prediction information corresponding to the last historical timestamp, Represents the second pose information, T represents the transformation matrix in SE3, which can be obtained by predicting the position information p t and the predicted pose vector q t Build obtained.

[0116] After the cumulative error is calculated, ideally, the posture prediction information corresponding to the last historical timestamp should coincide with the second posture information, that is, ideally, the cumulative error is 0. Therefore, in an embodiment of the present invention, minimizing the cumulative error is the optimization goal, that is, the posture prediction information corresponding to the last historical timestamp is brought as close to the second posture information as possible, that is, node B' is continuously moved closer to node B, and in the process of approaching, each node and each measurement edge is continuously adjusted so that the above-mentioned cumulative error is evenly distributed to the posture prediction information corresponding to each node, ensuring that the posture prediction information corresponding to each historical timestamp between the first ground transponder and the second ground transponder is optimal. In the iterative optimization process, the LM (Levenberg-Marquardt) algorithm can be used to solve the optimal solution for each posture prediction information between the first ground transponder and the second ground transponder, that is, the posture correction information corresponding to each historical timestamp is obtained as the true position value under each historical timestamp to train the initial error time series model. In addition, after obtaining the posture correction information corresponding to each historical timestamp, only the position component and heading angle component are extracted from the posture correction information, and the position component and heading angle are input into the initial error time series model together with the above-mentioned training data set for supervised training to obtain a trained error time series model.

[0117] Furthermore, determining the pose prediction information corresponding to each of the historical timestamps between the first ground transponder and the second ground transponder based on the inertial navigation history information and the first pose information includes:

[0118] Performing Kalman filtering based on the inertial navigation history information and the first pose information to predict and obtain first pose prediction information corresponding to each of the historical timestamps;

[0119] For each of the historical timestamps between the first ground transponder and the second ground transponder, if there is Global Positioning System (GNSS) historical information in the historical timestamp, determining the pose prediction information corresponding to the historical timestamp based on the GNSS historical information and the first pose prediction information;

[0120] When the GNSS historical information does not exist at the historical timestamp, the first posture prediction information is determined as the posture prediction information corresponding to the historical timestamp.

[0121] Specifically, after performing Kalman filtering based on the inertial navigation history information and the first pose information to obtain the first pose prediction information corresponding to each timestamp, if there is GNSS history information for the historical timestamp corresponding to the node, the first pose prediction information corresponding to the historical timestamp can be preliminarily corrected based on the GNSS history information. The steps of correcting the first pose prediction information based on the GNSS history information are the same as the steps of correcting the first state vector based on the GNSS measurement information, and the embodiments of the present invention are not repeated here. If there is no GNSS history information for the historical timestamp corresponding to the node, the first pose prediction information corresponding to the historical timestamp can be determined as the pose prediction information.

[0122] Step 140: Determine the corrected positioning information of the smart rail train based on the error posture information and the predicted state vector.

[0123] Specifically, after determining the error posture information and predicted state vector at the current moment, the predicted attitude vector in the predicted state vector can be converted to obtain the predicted heading angle at the current moment, and then the predicted position information and the predicted heading angle are corrected according to the error posture information to obtain the corrected positioning information of the smart rail train, so that the smart rail train can determine the precise driving route based on the corrected positioning information.

[0124] The smart rail train positioning method provided by an embodiment of the present invention obtains the inertial navigation information, wheel speed information and previous state vector of the smart rail train at the previous moment, and then predicts the predicted state vector at the current moment based on the inertial navigation information and the previous state vector. The error time series model learns the errors of each historical time stamp obtained by the posture correction information corresponding to each historical time stamp after posture graph optimization correction. After the inertial navigation information and wheel speed information are input into the error time series model, the error posture information at the current moment is output. Through the error posture information, the predicted state vector at the current moment is corrected in real time in a data-driven manner to obtain the corrected positioning information at the current moment, thereby improving the positioning accuracy of the smart rail train, and thereby improving the accuracy of the smart rail train's route and driving safety.

[0125] The smart rail train positioning device provided by the present invention is described below. The smart rail train positioning device described below and the smart rail train positioning method described above can be referenced to each other.

[0126] The embodiment of the present invention also provides a smart rail train positioning device, Figure 6 FIG. 1 is a schematic diagram of the structure of the smart rail train positioning device provided by an embodiment of the present invention. Figure 6 As shown, the smart rail train positioning device 600 includes: an acquisition module 610, a prediction module 620, an output module 630 and a positioning module 640, wherein:

[0127] An acquisition module 610 is used to obtain the inertial navigation information, wheel speed information, and previous state vector of the smart rail train at the previous moment;

[0128] A prediction module 620, configured to determine a predicted state vector at a current moment based on the inertial navigation information at a previous moment and the previous state vector;

[0129] An output module 630 is configured to input the inertial navigation information and the wheel speed information at the previous moment into an error time series model, and output error pose information at the current moment; the error time series model is trained based on pose correction information corresponding to each historical time stamp after pose graph optimization correction;

[0130] The positioning module 640 is used to determine the corrected positioning information of the smart rail train based on the error posture information and the predicted state vector.

[0131] The smart rail train positioning device provided by an embodiment of the present invention obtains the inertial navigation information, wheel speed information and previous state vector of the smart rail train at the previous moment, and then predicts the predicted state vector at the current moment based on the inertial navigation information and the previous state vector. The error timing model learns the errors of each historical time stamp obtained by the posture correction information corresponding to each historical time stamp after posture graph optimization correction. After the inertial navigation information and wheel speed information are input into the error timing model, the error posture information at the current moment is output. Through the error posture information, the predicted state vector at the current moment is corrected in real time in a data-driven manner to obtain the corrected positioning information at the current moment, thereby improving the positioning accuracy of the smart rail train, and thereby improving the accuracy of the smart rail train's route and driving safety.

[0132] Optionally, the prediction module 620 is specifically configured to:

[0133] Performing Kalman filtering based on the inertial navigation information at the previous moment and the previous state vector to predict a first state vector at the current moment;

[0134] Upon receiving the GNSS measurement information, determining a predicted state vector at the current moment based on the first state vector and the GNSS measurement information;

[0135] In a case where the GNSS measurement information is not received, the first state vector is determined as the predicted state vector at the current moment.

[0136] Optionally, the prediction module 620 is specifically configured to:

[0137] Determining a state covariance matrix corresponding to the first state vector;

[0138] determining a Kalman gain based on the state covariance matrix and a GNSS noise covariance matrix corresponding to the GNSS measurement information;

[0139] determining an error state vector based on the Kalman gain, the GNSS measurement information, and the first state vector;

[0140] Based on the error state vector, the predicted state vector at the current moment is determined.

[0141] Optionally, the prediction module 620 is specifically configured to:

[0142] Get the previous covariance matrix, rotation matrix and noise covariance matrix of the previous moment;

[0143] Determining a state transition matrix at the previous moment based on a time interval between the current moment and the previous moment, the inertial navigation information at the previous moment, and the rotation matrix;

[0144] A state covariance matrix corresponding to the first state vector is determined based on the noise propagation matrix at a previous moment, the state transfer matrix, the previous covariance matrix, and the noise covariance matrix.

[0145] Optionally, the smart rail train positioning device 600 further includes a training module, which is specifically used to:

[0146] Acquire first posture information corresponding to the first ground transponder and second posture information corresponding to the second ground transponder;

[0147] Determining, based on the inertial navigation history information and the first posture information, posture prediction information corresponding to each of the historical timestamps between the first ground transponder and the second ground transponder;

[0148] A pose graph is optimized based on each of the pose prediction information and the second pose information to obtain pose correction information corresponding to each of the historical timestamps after correction.

[0149] Optional training module, specifically for:

[0150] Determining posture change information between adjacent historical timestamps based on each of the posture prediction information;

[0151] Determine accumulated information of posture changes between the posture prediction information corresponding to the last historical timestamp and the second posture information;

[0152] Determining a cumulative error based on the accumulated pose change information, the second pose information, and pose prediction information corresponding to the last historical timestamp;

[0153] Taking minimization of the cumulative error as the optimization goal, each of the posture prediction information and each of the posture change information is iteratively optimized to obtain the posture correction information corresponding to each of the historical timestamps after correction.

[0154] Optional training module, specifically for:

[0155] Performing Kalman filtering based on the inertial navigation history information and the first pose information to predict and obtain first pose prediction information corresponding to each of the historical timestamps;

[0156] For each of the historical timestamps between the first ground transponder and the second ground transponder, if there is Global Positioning System (GNSS) historical information in the historical timestamp, determining the pose prediction information corresponding to the historical timestamp based on the GNSS historical information and the first pose prediction information;

[0157] When the GNSS historical information does not exist at the historical timestamp, the first posture prediction information is determined as the posture prediction information corresponding to the historical timestamp.

[0158] Figure 7 Schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. Figure 7 As shown, the electronic device may include: a processor 710, a communication interface 720, a memory 730, and a communication bus 740, wherein the processor 710, the communication interface 720, and the memory 730 communicate with each other via the communication bus 740. The processor 710 may call the logic instructions in the memory 730 to execute the smart rail train positioning method, which includes:

[0159] Obtain the inertial navigation information, wheel speed information and previous state vector of the smart rail train at the previous moment;

[0160] Determining a predicted state vector at a current moment based on the inertial navigation information at a previous moment and the previous state vector;

[0161] Inputting the inertial navigation information and the wheel speed information at the previous moment into an error time series model, and outputting the error posture information at the current moment; the error time series model is trained based on the posture correction information corresponding to each historical time stamp after the posture graph is optimized and corrected;

[0162] Based on the error posture information and the predicted state vector, the corrected positioning information of the smart rail train is determined.

[0163] In addition, the logic instructions in the above-mentioned memory 730 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0164] On the other hand, the present invention further provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the smart rail train positioning method provided by the above methods, which includes:

[0165] Obtain the inertial navigation information, wheel speed information and previous state vector of the smart rail train at the previous moment;

[0166] Determining a predicted state vector at a current moment based on the inertial navigation information at a previous moment and the previous state vector;

[0167] Inputting the inertial navigation information and the wheel speed information at the previous moment into an error time series model, and outputting the error posture information at the current moment; the error time series model is trained based on the posture correction information corresponding to each historical time stamp after the posture graph is optimized and corrected;

[0168] Based on the error posture information and the predicted state vector, the corrected positioning information of the smart rail train is determined.

[0169] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for positioning a smart rail train provided by the above methods is implemented. The method includes:

[0170] Obtain the inertial navigation information, wheel speed information and previous state vector of the smart rail train at the previous moment;

[0171] Determining a predicted state vector at a current moment based on the inertial navigation information at a previous moment and the previous state vector;

[0172] Inputting the inertial navigation information and the wheel speed information at the previous moment into an error time series model, and outputting the error posture information at the current moment; the error time series model is trained based on the posture correction information corresponding to each historical time stamp after the posture graph is optimized and corrected;

[0173] Based on the error posture information and the predicted state vector, the corrected positioning information of the smart rail train is determined.

[0174] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0175] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0176] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for positioning a smart rail train, characterized in that: include: Obtain the inertial navigation information, wheel speed information and previous state vector of the smart rail train at the previous moment; Determining a predicted state vector at a current moment based on the inertial navigation information at a previous moment and the previous state vector; Inputting the inertial navigation information and the wheel speed information at the previous moment into an error time series model, and outputting the error posture information at the current moment; The error timing model is trained based on the pose correction information corresponding to each historical timestamp after pose graph optimization and correction; The pose graph is optimized to predict the pose information corresponding to each historical timestamp based on the true position value provided by the adjacent ground transponder. Offline Perform iterative optimization; the posture prediction information corresponding to each historical timestamp is between the adjacent ground transponders; Based on the error posture information and the predicted state vector, the corrected positioning information of the smart rail train is determined.

2. The smart rail train positioning method according to claim 1, characterized in that: The posture correction information corresponding to each historical timestamp is obtained based on the following steps: Acquire first posture information corresponding to the first ground transponder and second posture information corresponding to the second ground transponder; Determining, based on the inertial navigation history information and the first posture information, posture prediction information corresponding to each of the historical timestamps between the first ground transponder and the second ground transponder; A pose graph is optimized based on each of the pose prediction information and the second pose information to obtain pose correction information corresponding to each of the historical timestamps after correction.

3. The smart rail train positioning method according to claim 2, characterized in that: The performing pose graph optimization based on each of the pose prediction information and the second pose information to obtain pose correction information corresponding to each of the corrected historical timestamps includes: Determining posture change information between adjacent historical timestamps based on each of the posture prediction information; Determine accumulated information of posture changes between the posture prediction information corresponding to the last historical timestamp and the second posture information; Determining a cumulative error based on the accumulated pose change information, the second pose information, and pose prediction information corresponding to the last historical timestamp; Taking minimization of the cumulative error as the optimization goal, each of the posture prediction information and each of the posture change information is iteratively optimized to obtain the posture correction information corresponding to each of the historical timestamps after correction.

4. The smart rail train positioning method according to claim 2, characterized in that: The determining, based on the inertial navigation history information and the first posture information, posture prediction information corresponding to each of the historical timestamps between the first ground transponder and the second ground transponder includes: Performing Kalman filtering based on the inertial navigation history information and the first pose information to predict and obtain first pose prediction information corresponding to each of the historical timestamps; For each of the historical timestamps between the first ground transponder and the second ground transponder, if there is Global Positioning System (GNSS) historical information in the historical timestamp, determining the pose prediction information corresponding to the historical timestamp based on the GNSS historical information and the first pose prediction information; When the GNSS historical information does not exist at the historical timestamp, the first posture prediction information is determined as the posture prediction information corresponding to the historical timestamp.

5. The smart rail train positioning method according to any one of claims 1 to 4, characterized in that: The determining of the predicted state vector at the current moment based on the inertial navigation information at the previous moment and the previous state vector includes: Performing Kalman filtering based on the inertial navigation information at the previous moment and the previous state vector to predict a first state vector at the current moment; Upon receiving the GNSS measurement information, determining a predicted state vector at the current moment based on the first state vector and the GNSS measurement information; In a case where the GNSS measurement information is not received, the first state vector is determined as the predicted state vector at the current moment.

6. The smart rail train positioning method according to claim 5, characterized in that: The determining, based on the first state vector and the GNSS measurement information, the predicted state vector at the current moment includes: Determining a state covariance matrix corresponding to the first state vector; determining a Kalman gain based on the state covariance matrix and a GNSS noise covariance matrix corresponding to the GNSS measurement information; determining an error state vector based on the Kalman gain, the GNSS measurement information, and the first state vector; Based on the error state vector, the predicted state vector at the current moment is determined.

7. The smart rail train positioning method according to claim 6, characterized in that: The determining of a state covariance matrix corresponding to the first state vector includes: Get the previous covariance matrix, rotation matrix and noise covariance matrix of the previous moment; Determining a state transition matrix at the previous moment based on a time interval between the current moment and the previous moment, the inertial navigation information at the previous moment, and the rotation matrix; A state covariance matrix corresponding to the first state vector is determined based on the noise propagation matrix at a previous moment, the state transfer matrix, the previous covariance matrix, and the noise covariance matrix.

8. A smart rail train positioning device, characterized in that: include: The acquisition module is used to obtain the inertial navigation information, wheel speed information and previous state vector of the smart rail train at the previous moment; a prediction module, configured to determine a predicted state vector at a current moment based on the inertial navigation information at a previous moment and the previous state vector; An output module, configured to input the inertial navigation information and the wheel speed information at the previous moment into an error time series model, and output error posture information at the current moment; The error timing model is trained based on the pose correction information corresponding to each historical timestamp after pose graph optimization and correction; The pose graph is optimized to predict the pose information corresponding to each historical timestamp based on the true position value provided by the adjacent ground transponder. Offline Perform iterative optimization; the posture prediction information corresponding to each historical timestamp is between the adjacent ground transponders; A positioning module is used to determine the corrected positioning information of the smart rail train based on the error posture information and the predicted state vector.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the program, the smart rail train positioning method as described in any one of claims 1-7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the smart rail train positioning method as described in any one of claims 1 to 7 is implemented.

Images