SINS / DVL Seamless Integrated Navigation Method Assisted by Transfer Gaussian Process Regression Model

Abstract

A SINS / DVL seamless integrated navigation method assisted by a transfer Gaussian process regression model belongs to the field of navigation technology and solves the problem of low reliability of SINS / DVL seamless integrated navigation technology. The method of the present invention includes: generating DVL-related data sets based on a simulation platform, a trajectory generator, and a navigation system model; powering on the vehicle and the navigation system, and collecting DVL-related data sets during the effective period of the DVL; during the failure period of the DVL, using the generated DVL-related data sets as the source domain data sets and the collected DVL-related data sets as the target domain data sets to train the hyperparameters of the transfer Gaussian process regression model; using two transfer Gaussian process regression models to predict the DVL measurement values and inertial navigation errors respectively, and obtaining the corresponding variances; converting the predicted values of both models into the velocity estimates of the vehicle in the navigation system, and fusing the information to obtain the velocity calculation results based on the two models. The present invention is applicable to maintaining high-precision calculation of the SINS / DVL integrated navigation system in the case of long-term DVL failure.

Description

Technical Field

[0001] This application relates to the field of navigation technology, and particularly to the information fusion of navigation systems. Background Art

[0002] A navigation system with high precision and strong reliability is a prerequisite for the successful completion of tasks by a vehicle. The common navigation system in the marine environment is the strapdown inertial navigation system (SINS) / Doppler velocity log (DVL) integrated navigation system. Under the interference of the complex marine environment, the DVL sensor may be interrupted at any time due to its dependence on environmental information, leading to the divergence of the positioning error of the navigation system, and then affecting the operation of the vehicle. In this case, users need a set of intelligent seamless navigation algorithms for the situation of DVL interruption, which can suppress the divergence of the error of the navigation system when the DVL is interrupted.

[0003] Regarding the failure problem of the DVL sensor, some scholars have used intelligent technologies such as machine learning for research. The specific idea is to use the powerful fitting ability of machine learning algorithms to assist in constructing a virtual DVL or constructing the solution result of the navigation system. Existing technologies such as the "DVL modeling method based on fuzzy multi-output least squares support vector machine" disclosed in the Chinese patent with the publication number CN115099134A and the publication date of September 23, 2022, extend the single-output least squares support vector machine model to a multi-output least squares support vector machine model, and use this model to fit the output of the Doppler velocity log in the body coordinate system; another example is the "hybrid processing method for DVL failure in integrated navigation" disclosed in the Chinese patent with the bulletin number CN106840150B and the bulletin date of October 15, 2019, which uses a partial least squares model combined with a support vector regression model to predict the DVL measurement information. However, the above inventions can only provide DVL-related estimates and cannot provide the corresponding variances, and due to being completely based on the historical data of the navigation system, they are usually only applicable to seamless navigation under short-term DVL interruption, and the algorithms usually fail under long-term DVL interruption. Summary of the Invention

[0004] The purpose of the present invention is to solve the problem of low reliability of the existing SINS / DVL seamless integrated navigation technology, and provide a SINS / DVL seamless integrated navigation method assisted by a transfer Gaussian process regression model.

[0005] The present invention is realized through the following technical solutions. On the one hand, the present invention provides a SINS / DVL seamless integrated navigation method assisted by a transfer Gaussian process regression model, and the method includes:

[0006] Step 1: Use a trajectory generator in a simulation platform to generate a simulated trajectory including various maneuvering states of the vehicle, and establish a SINS / DVL integrated navigation system model;

[0007] Step 2: Generate the direct output of the gyroscope device in the SINS based on the simulated trajectory and the SINS / DVL integrated navigation system model The direct output f of the accelerometer device b (t), and the direct output of the DVL beam The historical velocity solution result of the SINS / DVL integrated navigation system in the navigation coordinate system And the difference between the velocity solution results of the standalone SINS and the SINS / DVL integrated navigation system in the navigation coordinate system

[0008] Step 3: Construct two sets of datasets based on the data generated in Step 2 Wherein represents the input dataset, represents the output dataset;

[0009] Step 4: Power on the vehicle and the SINS / DVL integrated navigation system

[0010] Step 5: When both the SINS and DVL in the SINS / DVL integrated navigation system are working properly, the SINS / DVL integrated navigation system uses Kalman filtering to fuse the SINS and DVL sensor information to generate the navigation information attitude velocity and position solution results

[0011] Meanwhile, the SINS / DVL integrated navigation system collects the direct output of the gyroscope device in the SINS The direct output f of the accelerometer device b (t), the historical velocity solution result of the SINS / DVL integrated navigation system in the navigation coordinate system and the velocity solution error of the pure inertial navigation

[0012] Construct two sets of datasets based on the collected data Wherein represents the input dataset, represents the output dataset;

[0013] Step 6: When the DVL is abnormal, use the dataset constructed in Step 3 as the source domain dataset and the dataset constructed in Step 5 as the target domain dataset to train the hyperparameters of the transfer Gaussian process regression model

[0014] Use to train the hyperparameters of the transfer Gaussian process regression model 1: TGPR-1, and use to train the hyperparameters of the transfer Gaussian process regression model 2: TGPR-2

[0015] Step 7: After the training hyperparameters are completed, TGPR-1 uses the f b (t), and estimate to obtain and the corresponding variance TGPR-2 uses the f b (t), and to estimate and obtain and the corresponding variance

[0016] Step 8: The SINS / DVL integrated navigation system uses the output by TGPR-1 as the measurement of the DVL to perform information fusion with the output of the SINS, and obtains the navigation system speed estimate of the integrated navigation system based on TGPR-1. At the same time, the SINS / DVL integrated navigation system uses the output by TGPR-2 to correct the speed estimate of the pure inertial navigation system and obtains the solution speed estimate

[0017] Step 9: Use and the uncertainties corresponding to these two speed estimates to perform information fusion on these two speed estimates, and obtain the navigation system speed estimate

[0018] of the SINS / DVL integrated navigation system at the current moment based on TGPR-1 and TGPR-2. In a second aspect, the present invention provides a computer device, including a memory and a processor. A computer program is stored in the memory. When the processor runs the computer program stored in the memory, it executes the steps of a SINS / DVL seamless integrated navigation method assisted by a transfer Gaussian process regression model as described above.

[0019] In a third aspect, the present invention provides a computer-readable storage medium, in which multiple computer instructions are stored. The multiple computer instructions are used to cause a computer to execute a SINS / DVL seamless integrated navigation method assisted by a transfer Gaussian process regression model as described above.

[0020] In a fourth aspect, the present invention provides an electronic device, including:

[0021] at least one processor; and,

[0022] a memory communicatively connected to the at least one processor; wherein,

[0023] the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute a SINS / DVL seamless integrated navigation method assisted by a transfer Gaussian process regression model as described above.

[0024] Advantages of the present invention:

[0025] The present invention provides a SINS / DVL seamless integrated navigation method assisted by a transfer Gaussian process regression model, which can suppress the error divergence of the navigation system under long-term DVL failure conditions and improve the reliability of the integrated navigation system.

[0026] The present invention discloses a SINS / DVL seamless integrated navigation method assisted by a transfer Gaussian process regression model, designs a Gaussian process regression model equipped with a transfer learning mechanism, and suppresses the error divergence of the navigation system when the DVL sensor fails. The steps are as follows: Step 1: Generate DVL-related data sets based on a simulation platform, a trajectory generator, and a navigation system model; Step 2: Power on the vehicle and the navigation system, and collect DVL-related data sets during the effective period of the DVL; Step 3: During the DVL failure period, use the data set in Step 1 as the source domain data set, and use the data set in Step 2 as the target domain data set to train the hyperparameters of the transfer Gaussian process regression model; Step 4: Use two transfer Gaussian process regression models to predict the DVL measurement values and inertial navigation errors respectively, and obtain the corresponding variances; Step 5: Convert the predicted values of the two models in Step 4 into the velocity estimates of the vehicle in the navigation system, and fuse the information to obtain the velocity calculation results based on the two models.

[0027] The present invention designs a seamless navigation method based on a transfer Gaussian process regression model for the seamless navigation problem of the SINS / DVL integrated navigation system. Compared with traditional machine learning algorithms, the Gaussian process regression model introduced in this method can provide both fitting results and corresponding uncertainties, which is suitable for the information fusion process. The invention includes two Gaussian process regression models. Both Gaussian process regression models take the device outputs of the SINS and the historical calculation results of the integrated navigation system as inputs. One Gaussian process regression model takes the original measurements of the DVL as the output, and the other Gaussian process regression model takes the calculation error of the pure inertial navigation as the output. The final navigation calculation result is obtained by fusing information based on the results of the two Gaussian process regression models. At the same time, a transfer learning coefficient λ is introduced into the above two Gaussian process regression models. The present invention can utilize virtual data containing rich maneuvering characteristics to assist in training the model during the long-term interruption of the DVL when the vehicle undergoes variable maneuvers, enabling the model to successfully complete the seamless navigation task and suppressing the error accumulation of the navigation system.

[0028] The present invention is applicable to maintaining high-precision calculation of the SINS / DVL integrated navigation system in the case of long-term failure of the DVL. Brief Description of the Drawings

[0029] In order to more clearly illustrate the technical solutions of the present application, the accompanying drawings required for use in the embodiments will be briefly introduced below. Obviously, for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative efforts.

[0030] Figure 1 It is a flowchart for implementing the method of the present invention;

[0031] Figure 2 It is a structural diagram of the seamless navigation system of the present invention in the normal working mode of the DVL;

[0032] Figure 3 It is a structural diagram of the seamless navigation system of the present invention in the failure mode of the DVL;

[0033] Figure 4 It is a speed error curve graph under the method proposed by the present invention. Detailed Embodiments

[0034] The following details the embodiments of the present invention. The examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present invention and should not be construed as limiting the present invention.

[0035] Embodiment 1. A SINS / DVL seamless integrated navigation method assisted by a transfer Gaussian process regression model, the method comprising:

[0036] Step 1: Use a trajectory generator on a simulation platform to generate a simulated trajectory including various maneuvering states of the vehicle, and establish a SINS / DVL integrated navigation system model;

[0037] Step 2: Generate the direct output of the gyroscope device in the SINS based on the simulated trajectory and the SINS / DVL integrated navigation system model The direct output f b (t) of the accelerometer device, the direct beam output of the DVL The historical velocity calculation result of the SINS / DVL integrated navigation system in the navigation coordinate system And the difference between the velocity calculation results of the independent SINS and the SINS / DVL integrated navigation system in the navigation coordinate system

[0038] Step 3: Construct two sets of data sets according to the data generated in Step 2 Wherein Represents the input data set, Represents the output data set;

[0039] Step 4: Power on the vehicle and power on the SINS / DVL integrated navigation system;

[0040] Step 5: When both the SINS and DVL in the SINS / DVL integrated navigation system are working properly, the SINS / DVL integrated navigation system uses Kalman filtering to fuse the SINS and DVL sensor information to generate the navigation information attitude Velocity And position calculation results

[0041] At the same time, the SINS / DVL integrated navigation system collects the direct output of the gyroscope device in the SINS The direct output f b (t) of the accelerometer device, the historical velocity calculation result of the SINS / DVL integrated navigation system in the navigation coordinate system And the velocity calculation error of the pure inertial navigation

[0042] Construct two sets of data sets based on the collected data Wherein Represents the input data set, Represents the output data set;

[0043] Step 6: When the DVL is abnormal, use the dataset constructed in Step 3 as the source domain dataset and the dataset constructed in Step 5 as the target domain dataset to train the hyperparameters of the transfer Gaussian process regression model;

[0044] Use to train the hyperparameters of the transfer Gaussian process regression model 1: TGPR-1, and use to train the hyperparameters of the transfer Gaussian process regression model 2: TGPR-2;

[0045] Step 7: After training the hyperparameters, TGPR-1 uses the f b (t), and to estimate and obtain and the corresponding variance TGPR-2 uses the f b (t), and to estimate and obtain and the corresponding variance

[0046] Step 8: The SINS / DVL integrated navigation system uses the output by TGPR-1 as the measurement of the DVL and performs information fusion with the output of the SINS to obtain the navigation system velocity estimate of the integrated navigation system based on TGPR-1. At the same time, the SINS / DVL integrated navigation system uses the output by TGPR-2 to correct the velocity estimate of the pure inertial navigation system to obtain the solution velocity estimate

[0047] Step 9: Use and the uncertainties corresponding to these two velocity estimates to perform information fusion on these two velocity estimates to obtain the navigation system velocity estimate

[0048] In this embodiment, a seamless navigation method based on a transfer Gaussian process regression model is designed for the seamless navigation problem of the SINS / DVL integrated navigation system. Compared with traditional machine learning algorithms, the Gaussian process regression model introduced in this method can provide both fitting results and corresponding uncertainties, which is suitable for the information fusion process. The invention includes two Gaussian process regression models. Both Gaussian process regression models take the device outputs of the SINS and the historical calculation results of the integrated navigation system as inputs. One Gaussian process regression model takes the original measurements of the DVL as the output, and the other Gaussian process regression model takes the calculation error of the pure inertial navigation as the output. The final navigation calculation result is obtained by further fusing the information based on the results of the two Gaussian process regression models. At the same time, a transfer learning coefficient λ is introduced into the above two Gaussian process regression models, enabling the method to utilize virtual data containing rich maneuvering characteristics to assist in training the model under the condition of variable maneuvers of the carrier during a long-term interruption of the DVL, so that the model can successfully complete the seamless navigation task and suppress the error accumulation of the navigation system.

[0049] Embodiment 2. This embodiment further limits a SINS / DVL seamless integrated navigation method assisted by a transfer Gaussian process regression model described in Embodiment 1. In this embodiment, the transfer Gaussian process regression model is further limited, specifically including:

[0050] The kernel function matrix of the transfer Gaussian process regression model is K, where the element in the i-th row and j-th column of K is K i,j = λk(x {i} , x {j} ), where k is the Gaussian kernel function, x {i} , x {j} are the input navigation data, and λ is a parameter to be determined; for the two TGPR models, when x {i} , x {j} are both experimental data of navigation sensors or both simulation data of a trajectory generator, λ is 1, otherwise λ is one of the hyperparameters to be trained, and 0 < λ < 1.

[0051] In this embodiment, a transfer learning coefficient λ is introduced, and the value of λ is related to the data source, enabling the method to utilize virtual data containing rich maneuvering characteristics to assist in training the model under the condition of variable maneuvers of the carrier during a long-term interruption of the DVL, so that the model can successfully complete the seamless navigation task and suppress the error accumulation of the navigation system.

[0052] Embodiment 3. This embodiment further limits a SINS / DVL seamless integrated navigation method assisted by a transfer Gaussian process regression model described in Embodiment 1. In this embodiment, the TGPR model is further limited, specifically including:

[0053] The TGPR model optimizes hyperparameters by maximizing the target domain conditional probability, that is, by maximizing log(P(Y T |X T ,X S ,Y S ))

[0054] The Gaussian process regression model introduced in this embodiment can simultaneously provide fitting results and corresponding uncertainties, and is applicable to the information fusion process of the navigation system.

[0055] Embodiment 4 further limits a SINS / DVL seamless integrated navigation method assisted by a transfer Gaussian process regression model described in Embodiment 1. In this embodiment, the uncertainty in Step 9 is further limited, specifically including:

[0056] In Step 9, the uncertainty is obtained by taking the variance output by TGPR-1 as the measurement noise variance matrix and through the Kalman filter information fusion process. the uncertainty is the variance

[0057] output by TGPR-2.

[0058] Embodiment 5 further limits a SINS / DVL seamless integrated navigation method assisted by a transfer Gaussian process regression model described in Embodiment 4. In this embodiment, the navigation system velocity estimation based on TGPR-1 and TGPR-2 in Step 9 is further limited, specifically including:

[0059] In Step 9, the navigation system velocity estimation based on TGPR-1 and TGPR-2 at the current moment of the integrated navigation system

[0060]

[0061] This embodiment further fuses the navigation system velocity estimation based on TGPR-1 and the navigation system velocity estimation

[0062] Embodiment 6 is a further limitation on a SINS / DVL seamless integrated navigation method assisted by a transfer Gaussian process regression model described in Embodiment 1. In this embodiment, the SINS / DVL integrated navigation system model is further limited, specifically including:

[0063] The SINS / DVL integrated navigation system model is set as follows: Adopting a tightly integrated combination method, the state equation and measurement equation of the SINS / DVL navigation system are obtained. The state equation: X(k) = φ(k,k - 1)X(k - 1)+W(k - 1) is derived from the error equation of the inertial navigation system, where X(k) is the state quantity of the navigation system at time k, X = [φ e ,φ n ,φ u ,δv e ,δv n ,δv u ,δL,δλ,δh,ε x ,ε y ,ε z ,▽ x ,▽ y ,▽ z T , where [φ e ,φ n ,φ u is the attitude angle error of the navigation system in the northeast - sky direction, [δv e ,δv n ,δv u is the velocity error of the navigation system in the northeast - sky direction, [δL,δλ,δh] is the position error of the longitude, latitude, and altitude of the navigation system, [ε x ,ε y ,ε z is the zero - bias of the three axes of the gyroscope, [▽ x ,▽ y ,▽ z is the zero - bias of the three axes of the accelerometer, φ(k,k - 1) represents the state transition matrix of the system from time k - 1 to time k, obtained from the error equation of the inertial navigation system, and W is the system noise; The measurement equation: Z(k)=H(k)X(k)+V(k), and the measured quantity in the measurement equation is constructed from the measured ground velocity of the four beams of the DVL and the corresponding values calculated by the inertial navigation system: where Z is the measured quantity of the navigation system, represents the projection component of the velocity calculated by the inertial navigation on the i - th beam direction of the Doppler sensor, and H SINS represents the height measurement result of the inertial navigation system, ​represents the velocity measurement result of the i-th beam of the Doppler sensor, H(k) is the measurement matrix at time k, and V is the measurement noise;

[0064] The information fusion method applied in the SINS / DVL integrated navigation system is the Kalman filtering algorithm, which includes time update and parameter update, as shown in the formula:

[0065]

[0066] P(k,k - 1) = φ(k,k - 1)P(k - 1)φ(k,k - 1) T +Q(k - 1)

[0067] K(k) = P(k,k - 1)H(k) T (H(k)·P(k,k - 1)H(k) T +R(k)) -1

[0068]

[0069] P(k) = (I - K(k)H(k))P(k,k - 1)

[0070] Among them, is the state update value from time k - 1 to time k, is the state estimate value at time k - 1, P(k,k - 1) is the updated value of the covariance matrix from time k - 1 to time k, P(k) is the covariance matrix at time k, Q(k) is the system noise matrix at time k, R(k) is the system measurement noise matrix at time k, and K(k) is the gain matrix at time k.

[0071] Embodiment 7. This embodiment is an embodiment of a SINS / DVL seamless integrated navigation method assisted by a transfer Gaussian process regression model as described above, as Figure 1 shown, specifically including:

[0072] Step 1: Use a trajectory generator on the simulation platform to generate a carrier simulation trajectory including various maneuvering states of the carrier;

[0073] Step 2: Build a SINS / DVL integrated navigation system model on the simulation platform. It includes:

[0074] Obtain the state equation and measurement equation of the SINS / DVL navigation system,

[0075] State equation: X(k) = φ(k,k - 1)X(k - 1) + W(k - 1) is derived from the error equation of the inertial navigation system,

[0076] where X(k) is the state quantity of the navigation system at time k, X = [φe , φ n , φ u , δv e , δv n , δv u , δL, δλ, δh, ε x , ε y , ε z , ▽ x , ▽ y , ▽ z T , where [φ e , φ n , φ u is the attitude angle error of the navigation system in the northeast - sky direction, [δv e , δv n , δv u is the velocity error of the navigation system in the northeast - sky direction, [δL, δλ, δh] is the position error of the longitude, latitude and altitude of the navigation system, [ε x , ε y , ε z is the three - axis zero bias of the gyroscope, [▽ x , ▽ y , ▽ z is the three - axis zero bias of the accelerometer, φ(k, k - 1) represents the state transition matrix of the system from time k - 1 to time k, obtained from the error equation of the inertial navigation system, and W is the system noise;

[0077] Measurement equation: Z(k) = H(k)X(k)+V(k). The measured quantity in the measurement equation is constructed from the measured ground velocity of the four beams of the DVL and the corresponding values calculated by the inertial navigation system: where Z is the measured quantity of the navigation system, represents the projection component of the velocity calculated by the inertial navigation in the direction of the i - th beam of the Doppler sensor, represents the velocity measurement result of the i - th beam of the Doppler sensor, H(k) is the measurement matrix at time k, and V is the measurement noise;

[0078] The information fusion method applied in the SINS / DVL integrated navigation system is the Kalman filter algorithm, which includes time update and parameter update, as shown in the formula:

[0079]

[0080] P(k, k - 1) = φ(k, k - 1)P(k - 1)φ(k, k - 1) T +Q(k - 1)

[0081] K(k) = P(k, k - 1)H(k) T ​(H(k)·P(k,k - 1)H(k) T +R(k)) -1

[0082]

[0083] P(k) = (I - K(k)H(k))P(k,k - 1)

[0084] where is the state update value from time k - 1 to time k, is the state estimate value at time k - 1, P(k,k - 1) is the updated value of the covariance matrix from time k - 1 to time k, P(k) is the covariance matrix at time k, Q(k) is the system noise matrix at time k, R(k) is the system measurement noise matrix at time k, and K(k) is the gain matrix at time k;

[0085] Step 3: Generate based on the trajectory generated in Step 1 and the integrated navigation system model built in Step 2

[0086] The direct output of the gyroscope device in SINS:

[0087] The direct output of the accelerometer device: f b (t),

[0088] The direct beam output of DVL:

[0089] The historical velocity calculation result of the SINS / DVL integrated navigation system in the navigation coordinate system: where

[0090] and the difference between the velocity calculation results of the independent SINS and the integrated navigation system in the navigation coordinate system: And collect the above data;

[0091] Step 4: Construct two sets of data sets based on the above data where represents the input data set, represents the output data set. Where: f b , are the direct inputs of the gyroscope, accelerometer, and DVL respectively. is the velocity calculation result of the integrated navigation system, while is the difference between the velocity calculation of the inertial navigation system independently and the velocity calculation of the integrated navigation system, both of which need to be obtained based on the state equation and measurement equation in Step 1 and through Kalman filtering;

[0092] Step 5: Power on the carrier and the SINS / DVL navigation system;

[0093] Step 6: As Figure 2 shown, when both the SINS and DVL are working properly, the navigation system uses Kalman filtering to fuse the information of the SINS and DVL sensors to generate the navigation information attitude, velocity, and position calculation results at time t: Meanwhile, the navigation system collects the direct output of the gyroscope components in the SINS: The direct output of the accelerometer components: f b (t), and the historical velocity calculation results of the integrated navigation system in the navigation coordinate system: where and the difference between the velocity calculation results of the independent SINS and the integrated navigation system in the navigation coordinate system: Based on the above data, two data sets are constructed where represents the input data set, represents the output data set;

[0094] Step 7: When the DVL is abnormal, use the data set constructed in Step 4 as the source domain data set and the data set collected in Step 6 as the target domain data set to train the hyperparameters of the transfer Gaussian process regression model. Use to train the hyperparameters of the Gaussian process regression model 1: TGPR-1, and use to train the hyperparameters of the Gaussian process regression model 2: TGPR-2. It includes: the kernel function matrix of the transfer Gaussian process regression model is K, where the element in the i-th row and j-th column of K is K i,j = λk(x {i} , x {j} ), where k is the Gaussian kernel function, x is the input navigation data, and λ is a parameter to be determined. For the two TGPR models, when x {i} , x {i} are both source domain data or both target domain data, λ is 1, otherwise λ is one of the hyperparameters to be trained, 0 < λ < 1; the TGPR model optimizes the hyperparameters by maximizing the target domain conditional probability, that is, maximizing log(P(Y T |X T , X S , Y S ))), where Y T is Y 1 T or Y S is Y 1 S or

[0095] Step 8: Figure 3 As shown, TGPR-1 uses the current time of the navigation system f b (t), and estimate get and the corresponding variance TGPR-2 uses the current time of the navigation system f b (t), and estimate get and the corresponding variance

[0096] Step 9: Figure 3 As shown, Instead of constructing the measurement vector from the DVL output, the variance of the TGPR-1 output As the measurement noise variance matrix, and through Kalman filtering for information fusion, the velocity estimation of the integrated navigation system based on the TGPR-1 navigation system is obtained. and the corresponding uncertainty At the same time, the integrated navigation system uses the output of TGPR-2 Correcting the velocity estimate of a pure inertial navigation system Get the velocity estimate of the integrated navigation system based on the TGPR-2 navigation system And the corresponding uncertainty is

[0097] Step 10: Figure 3 As shown, based on The uncertainties corresponding to these two velocity estimates are The two velocity estimates are fused to obtain the navigation velocity estimate of the integrated navigation system based on TGPR-1 and TGPR-2 at the current moment.

[0098] At this point, the intelligent seamless navigation algorithm of the SINS / DVL integrated navigation system has been completed.

[0099] The present invention combines the navigation data collected by the experiment to illustrate the effectiveness of the algorithm. The experimental parameters are set as follows: the gyroscope zero bias of each axis is 0.01° / h, the angle random walk is 0.001° / h, the accelerometer zero bias is 500μg, and the speed random walk is The random noise of the Doppler velocimeter is 0.05 m / s. The total duration of the experiment is 1000 s. The measurement of the Doppler velocimeter was interrupted artificially during 500 - 800 s by Figure 4 It can be seen that the intelligent seamless navigation algorithm of the present invention reduces the divergence of the navigation system solution error during the interruption of the Doppler velocimeter.

[0100] The technical problem to be solved by the present invention is to provide a seamless integrated navigation method of SINS / DVL assisted by a transfer Gaussian process regression model, which can suppress the divergence of the navigation system error under the condition of long-term failure of DVL and improve the reliability of the integrated navigation system.

[0101] The protection scope of the present invention is not limited to the above application examples, and also includes various replacements or changes that can be easily made by those skilled in the art within the core idea of the present invention.

Claims

1. A SINS / DVL seamless integrated navigation method based on the assistance of migration Gaussian process regression model, characterized in that: The method comprises: Step 1: Generate simulated trajectories containing various maneuvering states of the carrier using the trajectory generator on the simulation platform and establish a SINS / DVL integrated navigation system model; Step 2: Generate direct output of the gyroscopic device in the SINS based on the simulated trajectory and the SINS / DVL integrated navigation system model The direct output of the accelerometer device is f b (t)′, beam direct output of DVL Historical velocity solution results of SINS / DVL integrated navigation system in navigation coordinate system The difference between the velocity solution results of the independent SINS and the SINS / DVL integrated navigation system in the navigation coordinate system Step 3: Construct two sets of data based on the data generated in step 2 in represents the input dataset, Represents the output dataset; Step 4: Start the carrier and the SINS / DVL integrated navigation system; Step 5: When both SINS and DVL of the SINS / DVL integrated navigation system are working normally, the SINS / DVL integrated navigation system uses Kalman filtering to fuse the SINS and DVL sensor information to generate the navigation information attitude at time t speed And position solution results At the same time, the SINS / DVL integrated navigation system collects the direct output of the gyro device in the SINS The direct output of the accelerometer device is f b (t), beam direct output of DVL Historical velocity solution results of SINS / DVL integrated navigation system in navigation coordinate system The difference between the velocity solution results of the independent SINS and the SINS / DVL integrated navigation system in the navigation coordinate system Construct two sets of data based on the collected data in represents the input dataset, Represents the output dataset; Step 6: When the DVL is abnormal, the dataset constructed in step 3 is used as the source domain dataset, and the dataset constructed in step 5 is used as the target domain dataset to train the hyperparameters of the migration Gaussian process regression model; use Training transfer Gaussian process regression model 1: Hyperparameters of TGPR-1, using Training Transfer Gaussian Process Regression Model 2: Hyperparameters of TGPR-2; Step 7: After training the hyperparameters, TGPR-1 uses the current state of the SINS / DVL integrated navigation system. f b (t) and Estimating the beam direct output of DVL get and the corresponding variance TGPR-2 uses the SINS / DVL integrated navigation system to f b (t) and Estimate the difference in velocity solution between independent SINS and SINS / DVL integrated navigation system in the navigation coordinate system get and the corresponding variance Step 8: The SINS / DVL integrated navigation system converts the TGPR-1 output The DVL measurement is used to fuse information with the output of SINS to obtain the navigation system velocity estimate based on TGPR-1 for the integrated navigation system. At the same time, the SINS / DVL integrated navigation system uses the output of TGPR-2 Correcting the velocity estimate of a pure inertial navigation system Get the solution speed estimate of SINS / DVL integrated navigation system based on TGPR-2 Step 9: Exploitation The uncertainties corresponding to these two velocity estimates are The two velocity estimates are fused to obtain the navigation system velocity estimate based on TGPR-1 and TGPR-2 at the current moment of the SINS / DVL integrated navigation system.

2. The SINS / DVL seamless integrated navigation method based on the migration Gaussian process regression model according to claim 1 is characterized in that: The kernel function matrix of the migration Gaussian process regression model is K, where the i-th row and j-th column element of K is K i,j =λk(x {i} ,x {j} ), where k is the Gaussian kernel function, x {i} , x {j} is the input navigation data, λ is an undetermined parameter; for the two TGPR models, when x {i} 、x {j} When all data are experimental data of navigation sensors or all data are simulated data of trajectory generators, λ is 1. Otherwise, λ is one of the hyperparameters to be trained, 0<λ<1.

3. The SINS / DVL seamless integrated navigation method based on the migration Gaussian process regression model according to claim 1 is characterized in that: The TGPR model optimizes the hyperparameters by maximizing the conditional probability of the target domain, that is, maximizing log(P(Y T |X T ,X S ,Y S )), where Y T Y1 T or Y S Y1 S or 4. The SINS / DVL seamless integrated navigation method based on the assistance of the migration Gaussian process regression model according to claim 1 is characterized in that: In step 9, Uncertainty By dividing the variance of the TGPR-1 output As the measurement noise variance matrix, it is obtained through the Kalman filter information fusion process. Uncertainty That is the variance of TGPR-2 output 5. The SINS / DVL seamless integrated navigation method based on the assistance of the migration Gaussian process regression model according to claim 4 is characterized in that: In step 9, the integrated navigation system estimates the navigation system velocity based on TGPR-1 and TGPR-2 at the current moment for:

6. The SINS / DVL seamless integrated navigation method based on the assistance of the migration Gaussian process regression model according to claim 1 is characterized in that: In step 1, a SINS / DVL integrated navigation system model is established, including: Obtain the state equations and measurement equations of the SINS / DVL navigation system; The information fusion method used in the SINS / DVL integrated navigation system is the Kalman filter algorithm, which includes time update and parameter update.

7. The SINS / DVL seamless integrated navigation method based on the migration Gaussian process regression model according to claim 6 is characterized in that: The time update and parameter update are specifically as follows: P(k,k-1)=φ(k,k-1)P(k-1)φ(k,k-1) T +Q(k-1) K(k)=P(k,k-1)H(k) T (H(k)·P(k,k-1)H(k) T +R(k)) -1 P(k)=(IK(k)H(k))P(k,k-1) in, is the state update value from time k-1 to time k, is the state estimate at time k-1, P(k,k-1) is the updated value of the covariance matrix from time k-1 to time k, P(k) is the covariance matrix at time k, Q(k) is the system noise matrix at time k, R(k) is the system measurement noise matrix at time k, K(k) is the gain matrix at time k, φ(k,k-1) represents the state transfer matrix of the system from time k-1 to time k, H(k) is the measurement matrix at time k, Z(k) is the measurement equation, and the measurement in the measurement equation is constructed by the actual ground velocity measurements of the four beams of DVL and the corresponding values ​​solved by the inertial navigation system.

8. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor runs the computer program stored in the memory, the steps of the method according to any one of claims 1 to 7 are performed.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a plurality of computer instructions, and the plurality of computer instructions are used to enable a computer to execute the method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: include: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 7.

Images