A combined navigation information fusion method and system considering driving state deviations

Abstract

This invention provides a method and system for fusion of integrated navigation information considering driving state deviation, belonging to the field of positioning and navigation technology. It obtains the Kalman gain generated by a first filter over the past n periods, and uses a pre-trained LSTM to predict the position and speed differences between the INS and GPS during the current GPS signal interruption period, i.e., the vehicle driving state deviation. A second filter uses the predicted vehicle driving state deviation as input to update the filter gain. A pre-trained TL-LSTM predicts the difference between the outputs of the first and second filters, and the optimal estimate after additional correction is used as the final compensation to correct the INS navigation result. This invention, considering driving state deviation, combines a Long Short-Term Memory (LSTM) network with an Error State Kalman Filter (ESKF) to perform secondary correction on the INS navigation result to ensure the positioning accuracy of the integrated navigation system in the event of a lock-out.

Description

Technical Field

[0001] This invention relates to the field of positioning and navigation technology, and specifically to a method and system for fusing combined navigation information that takes into account driving state deviations. Background Technology

[0002] Accurate positioning is crucial for safe and efficient vehicle operation, enabling vehicles to determine their absolute position within their surroundings and operate according to predetermined driving patterns. As an independent unit, the inertial navigation system (INS) calculates navigation information such as position and velocity using acceleration and angular velocity data provided by the inertial measurement unit (IMU) without relying on external information. However, INS errors accumulate over time; while maintaining high accuracy in the short term, it's difficult to sustain this in the long term. To ensure vehicles receive accurate latitude and longitude data over extended periods, the Global Positioning System (GPS) is typically integrated with the INS to suppress INS divergence and provide better navigation accuracy. However, in environments where GPS signals are difficult to receive, such as tunnels and urban canyons, vehicles cannot obtain GPS information, forcing the integrated system to operate in pure INS mode. In this situation, the integrated system struggles to achieve effective information fusion, leading to reduced positioning accuracy and ultimately causing a discrepancy between the system's estimated vehicle driving state and the actual driving state.

[0003] Currently, there are some studies both domestically and internationally on GPS / INS integrated navigation information fusion methods during GPS outages, mainly focusing on hardware redundancy and software prediction. Hardware redundancy often involves adding extra sensors to provide missing navigation information such as position and velocity to correct the INS state output. While this method provides navigation information redundancy and positioning fault tolerance for the integrated navigation system, it also increases system cost, enhances the complexity of sensor connection and alignment, and makes it more susceptible to environmental constraints. Software prediction focuses on establishing the relationship between INS data and GPS navigation information through neural network methods. Taking advantage of the fact that INS operation is unaffected by external conditions, it predicts pseudo-GPS positions based on INS data during GPS outages, calculates the difference between the predicted position and the actual INS navigation result, and then corrects the INS state using a filter. However, this research has several problems. First, the integrated navigation system is highly nonlinear during vehicle movement. Commonly used Kalman filters are only suitable for linear systems, and suffer from insufficient filtering accuracy and error accumulation in the complex dynamic environment of vehicle movement. Second, most of these studies only use INS and GPS information from the current or previous moment, without fully considering the impact of historical vehicle information on positioning accuracy. Third, an unavoidable difference exists between the target output of the neural network during training and the actual output during prediction, leading to deviations between the predicted and actual vehicle trajectories during GPS outages. Fourth, while increasing the complexity of the neural network can improve prediction accuracy, it significantly reduces the algorithm's training efficiency. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for fusing combined navigation information that takes into account driving state deviations, so as to solve at least one of the technical problems existing in the background art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] In a first aspect, the present invention provides a method for fusing integrated navigation information considering driving state deviations, comprising:

[0007] Obtain the Kalman gain generated by the first filter in the past n periods, and use the pre-trained LSTM to predict the position difference and speed difference between INS and GPS in the current GPS signal interruption period, i.e., the vehicle driving state deviation.

[0008] The predicted vehicle driving state deviation is used as input to update the filter gain using the second filter;

[0009] The difference between the outputs of the first and second filters is predicted using a pre-trained TL-LSTM, and the optimal estimate after additional correction is used as the final compensation to correct the INS navigation results.

[0010] Furthermore, the Kalman gain of the first filter over the past n cycles is used as the input to the LSTM to predict the system observations for the current cycle, namely the position difference and velocity difference between INS and GPS, and the LSTM is trained accordingly.

[0011] Furthermore, after the LSTM training is completed, the predicted system observations are input into the second filter to obtain a new optimal estimate. The dataset used to train the LSTM is used as the source dataset, and the dataset used to train the TL-LSTM is used as the target dataset for transfer learning. The training input of the TL-LSTM is the Kalman gain of the current cycle, and the output is the difference between the optimal estimates of the two filters.

[0012] Furthermore, the vehicle's speed and position information are obtained through the specific force and angular velocity data provided by the onboard inertial measurement unit, including: using INS strapdown calculation technology, the dynamic equations of inertial mechanization are used to perform transformations and calculations between different coordinate systems to obtain the vehicle's position and speed data at each moment in the INS.

[0013] Furthermore, the LSTM model is trained by: obtaining the INS navigation results at each time step through INS strapdown computation, and then calculating the position and speed differences between INS and GPS, i.e., the vehicle driving state deviation; inputting the vehicle driving state deviation as the observation vector into the first error state Kalman filter ESKF-1, fusing the GPS and INS navigation information, and obtaining the optimal estimate of the Kalman gain and ESKF-1 at each time step; then using the Kalman gain of the previous time step and the previous m time steps as the model input, and the position and speed differences between INS and GPS at the current time step as the model output.

[0014] Furthermore, the training of the second LSTM model includes: predicting the driving state deviation at the current moment using the trained first LSTM model, inputting it as the observation vector into the second error state Kalman filter ESKF-2, obtaining the Kalman gain at the current moment and the optimal estimate of ESKF-2, using the Kalman gain at the current moment as the model input, and using the difference between the optimal estimates generated by ESKF-1 and ESKF-2 as the model output.

[0015] Secondly, the present invention provides a combined navigation information fusion system that considers driving state deviations, comprising:

[0016] The acquisition module is used to acquire the Kalman gain generated by the first filter in the past n cycles, and use the pre-trained LSTM to predict the position difference and speed difference between INS and GPS in the current GPS signal interruption cycle, i.e. the vehicle driving state deviation.

[0017] The update module is used to update the filter gain by taking the predicted vehicle driving state deviation as input from the second filter.

[0018] The correction module is used to predict the difference between the outputs of the first and second filters using a pre-trained TL-LSTM, and then use the additional corrected optimal estimate as the final compensation to correct the INS navigation results.

[0019] Thirdly, the present invention provides a non-transitory computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the combined navigation information fusion method considering driving state deviation as described in the first aspect.

[0020] Fourthly, the present invention provides a computer device including a memory and a processor, wherein the processor and the memory communicate with each other, the memory stores program instructions executable by the processor, and the processor invokes the program instructions to execute the combined navigation information fusion method considering driving state deviation as described in the first aspect.

[0021] Fifthly, the present invention provides an electronic device, comprising: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the combined navigation information fusion method considering driving state deviation as described in the first aspect.

[0022] The beneficial effects of this invention are as follows: It considers the GPS / INS integrated navigation information fusion based on driving state deviation, combines a Long Short-Term Memory (LSTM) network with an Error State Kalman Filter (ESKF), and performs secondary correction on the INS navigation results to ensure the positioning accuracy of the integrated navigation system in the unlocked state.

[0023] The advantages of additional aspects of the invention will be set forth more clearly in the following description or will be learned by practice of the invention. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart of the INS strapdown calculation according to an embodiment of the present invention.

[0026] Figure 2 This is a flowchart illustrating the workflow when GPS signals are available, as described in an embodiment of the present invention.

[0027] Figure 3 This is a flowchart illustrating the workflow when GPS signal is interrupted, as described in an embodiment of the present invention.

[0028] Figure 4 This is a graph showing how the performance of the LSTM model described in this embodiment of the invention changes with the input delay length.

[0029] Figure 5 This is a graph showing the change in training time as a function of input delay length according to an embodiment of the present invention.

[0030] Figure 6 This is a comparison result diagram of an embodiment of the present invention. Detailed Implementation

[0031] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0032] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0033] It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as here.

[0034] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, and / or groups thereof.

[0035] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0036] To facilitate understanding of the present invention, the present invention will be further explained and described below with reference to the accompanying drawings and specific embodiments, and the specific embodiments do not constitute a limitation on the embodiments of the present invention.

[0037] Those skilled in the art should understand that the accompanying drawings are merely schematic diagrams of embodiments, and the components in the drawings are not necessarily essential for implementing the present invention.

[0038] To improve the information fusion performance of GPS / INS integrated navigation systems during GPS signal interruptions, correct vehicle driving state deviations, and enhance positioning accuracy while maintaining training efficiency, this invention constructs an Error State Kalman Filter (ESKF)-Long Short-Term Memory (LSTM) neural network structure that considers historical vehicle dynamic information. By combining two LSTMs with the ESKF, the correction of vehicle driving state deviations and further optimization of the correction values ​​are achieved. Specifically, the ESKF fuses data from INS and GPS, outputting the navigation correction values ​​for the integrated system; the LSTMs establish the relationship between the filter's internal parameters and navigation information. Furthermore, the second LSTM is obtained through transfer learning from the first LSTM, thereby improving the training efficiency and robustness of the method.

[0039] Example 1

[0040] In this embodiment 1, a combined navigation information fusion system considering driving state deviation is first provided, including: an acquisition module, used to acquire the Kalman gain generated by the first filter in the past n periods, and use a pre-trained LSTM to predict the position difference and speed difference between INS and GPS in the current GPS signal interruption period, i.e., the vehicle driving state deviation; an update module, used to update the filter gain by using the predicted vehicle driving state deviation as input from the second filter; and a correction module, used to predict the difference between the outputs of the first filter and the second filter by using a pre-trained TL-LSTM, and use the additionally corrected optimal estimate as the final compensation to correct the INS navigation result.

[0041] In this embodiment, the above-described system is used to implement a combined navigation information fusion method that considers driving state deviations. The GPS / INS combined navigation information fusion method considering driving state deviations is divided into two cases: GPS signal availability and GPS interruption.

[0042] When GPS signals are available, the proposed method runs in training mode, consisting of two steps: initial optimization and additional correction.

[0043] In the initial optimization step, the core idea is to use the Kalman gain of ESKF-1 over the past n cycles (where n represents the input delay length m plus 1) as the input to the LSTM to predict the system observation value (i.e., the position difference O between INS and GPS) for the current cycle. t (p) and velocity difference O t (v) After LSTM training, the predicted system observations are input into ESKF-2 to obtain a new optimal estimate. The additional correction step begins after the initial optimization step. Transfer learning is performed using the dataset from the initial optimization step (where LSTM was trained) as the source dataset and the dataset from the additional correction step (where TL-LSTM was trained) as the target dataset. The training input to TL-LSTM is the Kalman gain for the current period, and the output is the difference between the optimal estimates of the two filters. This effectively models and compensates for errors generated by the filters. When GPS is interrupted, the proposed method operates in prediction mode, at which point ESKF-1 ceases operation. LSTM uses the Kalman gain generated by ESKF-1 over the past n periods as input to predict the necessary system observations for the current GPS signal interruption period. The prediction result is then input into ESKF-2 to update the filter gain. Finally, the difference between the outputs of the two filters, ESKF-1 and ESKF-2, is predicted using TL-LSTM, and the additionally corrected optimal estimate is used as the final compensation to correct the INS navigation results.

[0044] In this embodiment, the vehicle's speed and position information are obtained through specific force and angular velocity data provided by the onboard inertial measurement unit (INS). Specifically, using INS strapdown calculation technology and the dynamic equations of inertial mechanization, transformations and calculations are performed between different coordinate systems (such as inertial coordinate system, geocentric coordinate system, navigation coordinate system, and vehicle coordinate system), enabling the INS to provide the vehicle's position and speed data at every moment. This process provides the data foundation for subsequent GPS and INS data fusion.

[0045] For the construction of the LSTM-ESKF framework, the INS navigation result P at each time step is obtained through INS strapdown computation. INS (t) and V INS (t), and then further calculate the position difference O between INS and GPS.t (p) and velocity difference O t (v), representing the vehicle's driving state deviation, is input as the observation vector into the first error state Kalman filter (ESKF-1) to fuse GPS and INS navigation information, thereby obtaining the Kalman gain K at each time step. t (p) and K t (v) and the optimal estimate δ(p) of ESKF-1 final ) t and δ(v) final ) t Then, the Kalman gain (k) of the previous time step and its m previous time steps is calculated. t-m-1 (p),k t-m-1 (v); k t-m (p),k t-m (v); ...; k t-1 (p),k t-1 (v)) is used as the model input, and the position difference O between INS and GPS at the current time is used as the model input. t (p) and velocity difference O t (v) is used as the model output to train the LSTM model.

[0046] For the application of the dual LSTM-ESKF framework, the driving state deviation at the current moment is predicted by the trained LSTM model, and this deviation is used as the observation vector to be input into the second error state Kalman filter (ESKF-2) to obtain the Kalman gain K at the current moment. t (p) and K t (p) and the optimal estimate δ(p) of ESKF-2 final ) t 'and δ(v final ) t The Kalman gain at the current time step is used as the model input, and the difference between the optimal estimates generated by ESKF-1 and ESKF-2 is used as the model output. Based on the pre-trained LSTM model, a second LSTM model, called TL-LSTM, is further trained through transfer learning.

[0047] Example 2

[0048] like Figures 1 to 6 As shown, this embodiment provides a GPS / INS integrated navigation information fusion method that considers driving state deviations. When GPS lock-up may occur in complex situations, leading to a significant decrease in navigation accuracy, a Long Short-Term Memory (LSTM) network is combined with an Error State Kalman Filter (ESKF) to perform a secondary correction on the INS navigation results, ensuring the positioning accuracy of the integrated navigation system in the lock-up state, thus solving the problems existing in the background technology.

[0049] The method specifically includes the following steps:

[0050] 1. Based on the measurement data of the inertial measurement unit (IMU), perform INS strapdown calculation to obtain the position and velocity information of the INS at each moment.

[0051] 2. An LSTM-ESKF framework that considers historical vehicle dynamic information is established, which fully utilizes the linearization characteristics of ESKF in handling nonlinear problems and the powerful capabilities of LSTM in time series data modeling to perform initial corrections on INS navigation results.

[0052] 3. Building upon step 2, two LSTM-ESKF frameworks were used to correct vehicle driving state deviations and further optimize the correction values. Furthermore, the second LSTM was obtained through transfer learning from the first LSTM, thereby improving the training efficiency and robustness of the method.

[0053] 4. Determine whether the GPS / INS integrated navigation information fusion method, which considers driving state deviations, is in training mode or prediction mode based on whether the GPS signal is interrupted.

[0054] In step 1, the INS strapdown calculation uses inertial mechanization dynamic equations to calculate the vehicle's attitude, velocity, and position, which includes three main parts: common coordinate systems, coordinate system transformation, and inertial navigation system update algorithm.

[0055] (1) Common coordinate systems

[0056] Common coordinate systems include inertial coordinate systems, geocentric coordinate systems, navigation coordinate systems, and vehicle coordinate systems, as detailed below:

[0057] ① Inertial coordinate system (i-frame)

[0058] The origin of the inertial coordinate system is the Earth's center of mass. The x-axis points to the vernal equinox, the z-axis points to the North Pole, and the y-axis is perpendicular to the xoz plane, all in accordance with the right-hand rule. The inertial coordinate system is stationary relative to the stars, and the observations measured by inertial measurement devices are relative to the inertial coordinate system.

[0059] ② Geocentric-Earth-Fixed Coordinate System (e-system)

[0060] The origin of the geocentric coordinate system is the Earth's center of mass. The x-axis points to the zero meridian, the z-axis coincides with the Earth's rotation axis, the y-axis points to the 90° east longitude meridian, and the xoy plane coincides with the Earth's equatorial plane.

[0061] ③ Navigation coordinate system (n-system)

[0062] A navigation coordinate system is a geographic coordinate system. Typical geographic coordinate systems include the Northeast-Eastern (ENU) coordinate system and the Northeast-Eastern (NED) coordinate system. The Northeast-Eastern (ENU) coordinate system uses the center of mass of the vehicle as the origin, the due east direction as the x-axis, the due north direction as the y-axis, and the zenith direction as the z-axis.

[0063] ④ Carrier coordinate system (b system)

[0064] The carrier coordinate system has its origin at the carrier's center of mass. The y-axis points forward along the carrier's longitudinal axis, the x-axis points to the right along the carrier's transverse axis, and the z-axis points upward along the carrier. These three axes together constitute the right-hand rule.

[0065] (2) Coordinate system transformation

[0066] ① Transformation from inertial coordinate system to geocentric coordinate system

[0067] The rotation matrix for transforming from an inertial coordinate system to a geocentric coordinate system can be expressed as:

[0068]

[0069] Where t is time, ω ie This is the Earth's rotation speed.

[0070] ② Transformation from Earth-centered and Earth-fixed coordinate system to navigation coordinate system

[0071] The rotation matrix for transforming from the geocentric coordinate system to the navigation coordinate system can be expressed as:

[0072]

[0073] Where λ represents the latitude of the origin of the carrier, and L represents the longitude of the origin of the carrier.

[0074] ③ Transform from vehicle coordinate system to navigation coordinate system

[0075] The relative relationship between the vehicle coordinate system (b) and the navigation coordinate system (n) represents the vehicle's attitude. Define γ, θ, and ψ as the vehicle's roll, pitch, and yaw angles, respectively. The navigation coordinate system is transformed by rotating the navigation coordinate system around the z-axis by an angle γ, then around the x-axis by an angle θ, and finally around the y-axis by an angle ψ. The rotation matrices can be expressed as:

[0076]

[0077] (3) Inertial Navigation System Update Algorithm

[0078] The data output from the IMU sensor is processed by INS strapdown to obtain attitude, velocity, and position information. The specific processing method is as follows:

[0079] ① Attitude information calculation

[0080] Using the quaternion method, a quaternion is defined as:

[0081] Q = q0 + q1i + q2j + q3k (4) The quaternion describing the carrier attitude is The differential equation is expressed as follows:

[0082]

[0083] in The projection of the angular velocity of the carrier in the b-frame into the n-frame is expressed as:

[0084]

[0085] In the formula, The angular velocity of the carrier, Let be the attitude matrix. This is the projection of the Earth's rotational angular velocity onto the navigation coordinate system. Let the rotational angular velocity of the navigation coordinate system relative to the geocentric coordinate system satisfy:

[0086]

[0087] Among them, v n and v e R represents the northward and eastward velocities of the carrier, respectively. m R represents the radius of curvature of the meridional circle. n denoted by , where is the radius of curvature of the zonal circle, and h represents the elevation.

[0088] Solve the differential equation (6) and update Obtain the rotation matrix of the carrier:

[0089]

[0090] ② Calculation of velocity information

[0091] Let the velocity calculated from the downloaded volume in the navigation coordinate system be v. n Then the derivative of velocity with respect to time in the navigation coordinate system and the derivative of velocity with respect to time in the inertial coordinate system have the following relationship:

[0092]

[0093] The local gravity vector can be expressed as: Therefore:

[0094]

[0095] Among them, f bThe specific force information output by the accelerometer can be used to derive the differential equation for velocity update from equations (10) and (11):

[0096]

[0097] Among them, f n Specific force acceleration in the navigation coordinate system This is the Gothic correction term.

[0098] Discretizing equation (12) yields the velocity update equation:

[0099]

[0100] ③ Calculation of location information

[0101] In the navigation coordinate system, the position update equation can be obtained from the velocity integral:

[0102]

[0103] in, These represent the eastward velocity, northward velocity, and upward velocity in the navigation coordinate system, respectively.

[0104] In step 2, the LSTM-ESKF framework, which considers historical vehicle dynamic information, first obtains the INS navigation result P at each moment through INS strapdown calculation. INS (t) and V INS (t), and then further calculate the position difference O between INS and GPS. t (p) and velocity difference O t (v), representing the vehicle's driving state deviation, is input as an observation vector to ESKF-1 to fuse GPS and INS navigation information, thereby obtaining the Kalman gain K at each time step. t (p) and K t (v) and the optimal estimate δ(p) of ESKF-1 final ) t and δ(v) final ) t The Kalman gain (k) of the previous time step and its m previous time steps is used to calculate the Kalman gain (k). t-m-1 (p),k t-m-1 (v); k t-m (p),k t-m (v); ...; k t-1 (p),k t-1 (v)) is used as the model input, and the position difference O between INS and GPS at the current time is used as the model input. t (p) and velocity difference O t (v) is used as the model output to train the LSTM model.

[0105] Error State Kalman Filtering (ESKF) is an extension of the classic Kalman Filter (KF). It treats the INS as a motion model and GPS as an observation model, defining the true state variable as the sum of the nominal state variable and the error state variable. During the filtering process, only the error state variable is predicted and updated. The principle is as follows:

[0106] (1) System state variables

[0107] ESKF divides the system state into nominal state and error state, which together constitute the true state. The nominal state, without considering system noise and various disturbances, is obtained directly through integration of IMU data. The error state, however, includes various noises and disturbances and is predicted and updated using an extended Kalman filter (EKF). The nominal state x, error state δx, and true state are represented by these components. The definition is as follows:

[0108]

[0109] In equation (17), and These represent the position, velocity, and attitude quaternions, as well as the accelerometer bias and gyroscope bias, respectively. and This represents the vehicle's position error, velocity error, attitude error, accelerometer bias error, and gyroscope bias error. This represents generalized addition. This represents the multiplication operation of quaternions.

[0110] Because quaternions introduce parameter redundancy and the covariance matrix is ​​prone to singularity when representing attitude, the attitude is represented by the minimized parameter in the error state, i.e., the rotation error is expressed using δθ. The relationship between the rotation matrix δR, the axis-angle vector δθ, and the quaternion δq in the error state is as follows:

[0111]

[0112] In equation (18), Exp(·) represents the exponential mapping function, [·] × Represents the antisymmetric matrix of a vector.

[0113] (2) System kinematic model

[0114] During ESKF, the true state kinematic model of the IMU is as follows:

[0115]

[0116] Where a m and These represent the measured values ​​of acceleration and angular velocity, respectively. n and These represent white noise for acceleration and angular velocity measurements, respectively. ω and These represent the acceleration and angular velocity bias vectors, respectively. The nominal state kinematic model is as follows:

[0117]

[0118] From equations (19) and (20), the kinematic model of the error state can be obtained:

[0119]

[0120] Since equations (19)-(21) are all continuous-time equations, they need to be discretized. After discretization, the IMU at time t is obtained. k The nominal state kinematic model at time t is:

[0121]

[0122] In equation (22), Δt is the sampling interval of the IMU. The IMU's sampling interval at time t is obtained. k The error state kinematic model at time t is:

[0123]

[0124] In equation (23), w vk w represents Gaussian random impulse noise at velocity. θk w represents Gaussian random impulse noise under attitude. ak w represents Gaussian random impulse noise under acceleration bias. ωk This represents Gaussian random impulse noise with angular velocity bias. Their covariance is defined as follows:

[0125]

[0126] In equation (24), and The standard deviations of Gaussian white noise, representing acceleration and angular velocity respectively, are... and The standard deviations of Gaussian white noise represent the acceleration and angular velocity biases, respectively.

[0127] (3) ESKF prediction process

[0128] Under discrete conditions, the nominal state, error state, IMU measurement, and noise are defined as follows:

[0129]

[0130] Substituting the vector defined above into equation (22), the recursive formula for the nominal state vector at that moment is:

[0131] x k+1 =f(x) k ,u mk (26)

[0132] In equation (26), f(·) represents the recursive function of the nominal state variable. Combining equation (23), and linearizing it according to the Taylor formula, we obtain t. k+1 The recursive formula for the time error state vector is:

[0133]

[0134] In equation (27) f δ (·) represents the recursive function of the error state variable, F xk and F wk These are the Jacobian matrices corresponding to the error state and the noise state, respectively. The error state and covariance update process are then obtained:

[0135]

[0136] Q w The noise matrix has the following specific form:

[0137]

[0138] (4) ESKF observation process

[0139] Accelerometers and gyroscopes have bias errors, and simply integrating them using the IMU will cause pose estimation to drift, which will gradually increase over time. This paper uses position and velocity information obtained from GPS to correct the error state estimation of the ESKF. When GPS signals are available, its observation data is promptly incorporated into the filter for updating.

[0140] z k =h(x tk )+w mk (29)

[0141] In the formula z k It is the measurement signal vector, w mk Let h(·) represent random white noise representing the measured signal, and h(·) represent the nonlinear mapping function of the system state. The posterior calibration update equation for the error state is:

[0142]

[0143] According to the chain rule, the Jacobian matrix is ​​calculated as follows:

[0144]

[0145] in,

[0146]

[0147] (5) ESKF error state injection and zeroing

[0148] Based on the recursion of the prior system state and the calibration of the error state, the updated state is obtained:

[0149]

[0150] Each system state corresponds to the following:

[0151]

[0152] After injecting the error state into the nominal state, the prior error state and its corresponding covariance need to be reset. The specific steps are as follows:

[0153]

[0154] Where G is the reset function The linearized Jacobian matrix approximates the identity matrix, so it is usually replaced by the identity matrix in practical applications.

[0155] In step 3, the driving state deviation at the current moment is predicted by the trained LSTM model, and this deviation is input into ESKF-2 as an observation vector to obtain the Kalman gain K at the current moment. t (p) and K t (p) and the optimal estimate δ(p) of ESKF-2 final ) t 'and δ(v final ) t The Kalman gain at the current time step is used as the model input, and the difference between the optimal estimates generated by ESKF-1 and ESKF-2 is used as the model output. Based on the pre-trained LSTM model, a second LSTM model, called TL-LSTM, is further trained through transfer learning.

[0156] In step 4, the GPS / INS integrated navigation information fusion method considering driving state deviation is divided into two cases: GPS signal availability and GPS interruption.

[0157] (1) When GPS signal is available, the proposed method runs in training mode, which consists of two steps: initial optimization and additional correction.

[0158] ① Initial optimization steps

[0159] The core idea of ​​this step is to use the Kalman gain of ESKF-1 over the past n periods (where n represents the input delay length m plus 1) as the input to LSTM to predict the system observation value (i.e., the position difference O between INS and GPS) for the current period. t (p) and velocity difference O t (v)). After the LSTM is trained, the predicted system observations are input into ESKF-2 to obtain a new optimal estimate.

[0160] ② Additional correction steps

[0161] The additional correction step begins after the initial optimization step. Transfer learning is performed using the dataset from which the LSTM was trained in the initial optimization step as the source dataset and the dataset from which the TL-LSTM was trained in the additional correction step as the target dataset. The training input to the TL-LSTM is the Kalman gain of the current cycle, and the output is the difference between the optimal estimates of the two filters. In this way, the errors generated by the filters can be effectively modeled and compensated.

[0162] (2) When GPS is interrupted, the proposed method operates in prediction mode, at which time ESKF-1 no longer works.

[0163] The LSTM takes the Kalman gain generated by ESKF-1 over the past n periods as input to predict the necessary system observations for the current GPS signal outage period. The prediction is then input into ESKF-2 to update the filter gain. Finally, the difference between the outputs of the two filters, ESKF-1 and ESKF-2, is predicted using a TL-LSTM, and the additionally corrected optimal estimate is used as the final compensation to correct the INS navigation results.

[0164] In this embodiment, the proposed navigation strategy was verified through road testing. The experimental equipment used was the INS570D (a high-precision vehicle-mounted GPS / INS integrated navigation and positioning system), which includes an IMU570D with a sampling frequency of 100Hz and a satellite navigation board with an operating frequency of 10Hz. The specific specifications of the sensors used are shown in Table 1.

[0165] Table 1. Specifications of the sensors used in the experiment.

[0166]

[0167]

[0168] First, the proposed method is trained using 1800 seconds of available GPS signal data. Then, a 120-second GPS signal outage is simulated to verify the performance of the proposed scheme.

[0169] like Figure 4 As shown, when the input delay length m is 20, the RMSE position error reaches its minimum value of 5.85 meters, which is 35% higher than the positioning accuracy without input delay. However, as... Figure 5 The results show that when m exceeds 20, the improvement in model prediction accuracy is not significant; instead, it leads to a significant increase in training time (by 30%). Therefore, in this embodiment, the optimal value of the input delay length is determined to be 20 to ensure high positioning accuracy and keep the training time within a reasonable range.

[0170] like Figure 6 As shown, within 120 seconds of GPS interruption, the positioning error caused by integration drift in pure INS reached 23.80m, with a standard deviation of 44.45m. The method presented in this paper effectively limits the integration drift of INS by using two LSTM-ESKF frameworks, reducing the positioning error to 0.62m (a reduction of 97.39%) and the standard deviation to 0.33m (a reduction of 99.26%).

[0171] Example 3

[0172] This embodiment 3 provides a non-transitory computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, they implement the combined navigation information fusion method considering driving state deviations as described above. The method includes:

[0173] Obtain the Kalman gain generated by the first filter in the past n periods, and use the pre-trained LSTM to predict the position difference and speed difference between INS and GPS in the current GPS signal interruption period, i.e., the vehicle driving state deviation.

[0174] The predicted vehicle driving state deviation is used as input to update the filter gain using the second filter;

[0175] The difference between the outputs of the first and second filters is predicted using a pre-trained TL-LSTM, and the optimal estimate after additional correction is used as the final compensation to correct the INS navigation results.

[0176] Example 4

[0177] This embodiment 4 provides a computer device, including a memory and a processor, wherein the processor and the memory communicate with each other, and the memory stores program instructions executable by the processor. The processor calls the program instructions to execute the combined navigation information fusion method considering driving state deviation as described above, the method including:

[0178] Obtain the Kalman gain generated by the first filter in the past n periods, and use the pre-trained LSTM to predict the position difference and speed difference between INS and GPS in the current GPS signal interruption period, i.e., the vehicle driving state deviation.

[0179] The predicted vehicle driving state deviation is used as input to update the filter gain using the second filter;

[0180] The difference between the outputs of the first and second filters is predicted using a pre-trained TL-LSTM, and the optimal estimate after additional correction is used as the final compensation to correct the INS navigation results.

[0181] Example 5

[0182] This embodiment 5 provides an electronic device, including: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the combined navigation information fusion method considering driving state deviation as described above. The method includes:

[0183] Obtain the Kalman gain generated by the first filter in the past n periods, and use the pre-trained LSTM to predict the position difference and speed difference between INS and GPS in the current GPS signal interruption period, i.e., the vehicle driving state deviation.

[0184] The predicted vehicle driving state deviation is used as input to update the filter gain using the second filter;

[0185] The difference between the outputs of the first and second filters is predicted using a pre-trained TL-LSTM, and the optimal estimate after additional correction is used as the final compensation to correct the INS navigation results.

[0186] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0187] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0188] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0189] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment, whereby a series of operational steps are performed to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0190] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that, based on the technical solutions disclosed in the present invention, various modifications or variations that can be made by those skilled in the art without creative effort should be included within the scope of protection of the present invention.

Claims

1. A method for fusing integrated navigation information considering driving state deviations, characterized in that, include: Obtain the Kalman gain generated by the first filter in the past n periods, and use the pre-trained LSTM to predict the position difference and speed difference between INS and GPS in the current GPS signal interruption period, i.e., the vehicle driving state deviation. The predicted vehicle driving state deviation is used as input to update the filter gain using the second filter; The difference between the outputs of the first and second filters is predicted using a pre-trained TL-LSTM, and the optimal estimate after additional correction is used as the final compensation to correct the INS navigation results.

2. The method for fusing integrated navigation information considering driving state deviation according to claim 1, characterized in that, The Kalman gain of the first filter over the past n cycles is used as the input to the LSTM to predict the system observations for the current cycle, namely the position difference and velocity difference between INS and GPS, and the LSTM is trained accordingly.

3. The method for fusing integrated navigation information considering driving state deviation according to claim 2, characterized in that, After the LSTM is trained, the predicted system observations are input into the second filter to obtain a new optimal estimate. The dataset used to train the LSTM is used as the source dataset, and the dataset used to train the TL-LSTM is used as the target dataset for transfer learning. The training input of the TL-LSTM is the Kalman gain of the current cycle, and the output is the difference between the optimal estimates of the two filters.

4. The method for fusing integrated navigation information considering driving state deviation according to claim 1, characterized in that, The vehicle's speed and position information is obtained by using the specific force and angular velocity data provided by the onboard inertial measurement unit, including: using INS strapdown calculation technology, and utilizing the dynamic equations of inertial mechanization to perform transformations and calculations between different coordinate systems, to obtain the vehicle's position and speed data at each moment in the INS.

5. The method for fusing integrated navigation information considering driving state deviation according to claim 3, characterized in that, The LSTM model is trained by: obtaining the INS navigation results at each time step through INS strapdown computation, and then calculating the position and speed differences between INS and GPS, i.e., the vehicle driving state deviation; inputting the vehicle driving state deviation as the observation vector into the first error state Kalman filter ESKF-1, fusing the GPS and INS navigation information, and obtaining the optimal estimate of the Kalman gain and ESKF-1 at each time step; then using the Kalman gain of the previous time step and the previous m time steps as the model input, and the position and speed differences between INS and GPS at the current time step as the model output.

6. The method for fusing integrated navigation information considering driving state deviation according to claim 5, characterized in that, The training of the second LSTM model includes: predicting the driving state deviation at the current time using the trained first LSTM model, inputting it as the observation vector into the second error state Kalman filter ESKF-2, obtaining the Kalman gain at the current time and the optimal estimate of ESKF-2, using the Kalman gain at the current time as the model input, and using the difference between the optimal estimates generated by ESKF-1 and ESKF-2 as the model output.

7. A combined navigation information fusion system considering driving state deviation, characterized in that, include: The acquisition module is used to acquire the Kalman gain generated by the first filter in the past n cycles, and use the pre-trained LSTM to predict the position difference and speed difference between INS and GPS in the current GPS signal interruption cycle, i.e. the vehicle driving state deviation. The update module is used to update the filter gain by taking the predicted vehicle driving state deviation as input from the second filter. The correction module is used to predict the difference between the outputs of the first and second filters using a pre-trained TL-LSTM, and then use the additional corrected optimal estimate as the final compensation to correct the INS navigation results.

8. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the combined navigation information fusion method considering driving state deviation as described in any one of claims 1-6.

9. A computer device, characterized in that, The system includes a memory and a processor, which communicate with each other. The memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the combined navigation information fusion method considering driving state deviation as described in any one of claims 1-6.

10. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the combined navigation information fusion method considering driving state deviation as described in any one of claims 1-6.

Images