A GPS Failure Tolerance Processing Method for Integrated Navigation

Abstract

The present invention discloses a GPS failure fault-tolerant processing method for integrated navigation, comprising four steps: Step 1: When GPS is valid, samples are collected to establish a weak predictor based on Gaussian process regression; Step 2: Calculate the training error of the weak predictor; Step 3: Set the weights of each weak predictor; Step 4: When GPS fails, the weak predictors are combined into a strong predictor, which outputs an estimated value of GPS measurement information for integrated navigation. The present invention has the advantages of high prediction accuracy and effectively extends the time to withstand GPS failure, thus ensuring high-reliability integrated navigation.

Description

Technical Field

[0001] The present invention relates to the field of integrated navigation fault-tolerant processing, and in particular to a GPS failure fault-tolerant processing method for integrated navigation. Background Art

[0002] Inertial navigation systems (INS) are a commonly used navigation system due to their independence and lack of reliance on external information. However, their calculation errors accumulate over time, leading to the emergence of integrated navigation. Integrated navigation integrates different navigation sensors or systems to obtain more reliable and accurate navigation information. The Global Positioning System (GPS) is widely used in various integrated navigation systems due to its high precision and efficiency. However, GPS receivers are susceptible to environmental factors when receiving satellite signals. For example, high-rise buildings and tunnels can block satellite signals. These factors can cause short-term GPS failures, during which the integrated navigation accuracy decreases. Therefore, it is particularly important to design a GPS failure-tolerant method for integrated navigation.

[0003] Currently, there are two types of fault-tolerant processing methods to resist GPS failure. One is hardware redundancy, which is to compensate for speed, position and other information by installing sensors such as odometers and speedometers. However, this method increases the hardware cost of the navigation system. The second type is software redundancy. By establishing a software predictor, during the GPS failure period, the predictor is used to estimate the inertial navigation system error or directly estimate the GPS measurement information to compensate for the navigation error.

[0004] Numerous algorithms are used to build software predictors, including neural networks, support vector regression, partial least squares regression, and random forest regression. However, most existing algorithms for combating GPS failures focus on building a single prediction model, and their accuracy decreases rapidly as the failure time increases.

[0005] Based on this, researching a GPS failure fault-tolerant processing method with higher prediction accuracy and longer prediction accuracy time has become the direction of industry development. Summary of the Invention

[0006] Purpose of the invention: To address the problems of short GPS failure resistance and low accuracy in the prior art, the present invention proposes a GPS failure fault-tolerant processing method for integrated navigation. Gaussian process regression is used to establish multiple weak predictors, and weights are assigned based on their training errors, thereby combining them into a strong predictor. This method has higher estimation accuracy, effectively extends the GPS failure resistance time, and overcomes the shortcomings of the prior art.

[0007] Technical solution: A GPS failure fault-tolerant processing method for integrated navigation, characterized in that the specific steps of the fault-tolerant processing method are as follows:

[0008] Step 1) When GPS is valid, collect the measurement information of the inertial measurement unit in the integrated navigation system X = {X k |k=1,2,…n} and the corresponding GPS measurement information Y={Y k |k=1,2,…n} constitute the training set, and Gaussian process regression is used to establish weak predictors with different kernel functions.

[0009] in:

[0010] X k is a vector containing the gyroscope and accelerometer measurement data;

[0011] Y k is a vector containing GPS position measurement information;

[0012] n is the number of samples in the training set;

[0013] Step 2) Calculate the training error δ of the weak predictor t t ,

[0014]

[0015] in:

[0016] T is the number of weak predictors;

[0017] H t (X k ) is to convert X k Input the predicted output vector obtained by weak predictor t;

[0018] ||H t (X k )-Y k ||2 is the calculation of H t (X k )-Y k The second norm of ;

[0019] Step 3) Set the weight β of each weak predictor according to the training error t ,

[0020]

[0021] Step 4) If GPS fails, the weak predictors are combined into a strong predictor.

[0022]

[0023] in:

[0024] X * A vector containing gyroscope and accelerometer measurement data when GPS fails;

[0025] Y * is the estimated value of GPS measurement information output by the strong predictor;

[0026] The estimation results of the strong predictor are used for integrated navigation information fusion to achieve high-reliability integrated navigation in the event of GPS failure.

[0027] Specifically, in step 1), when GPS is valid, the inertial measurement unit measurement information X in the integrated navigation system is collected. k |k=1,2,…n} and the corresponding GPS measurement information Y={Y k |k=1,2,...n} constitute the training set, and the collected measurement information is continuously updated as the navigation process progresses.

[0028] Specifically, in step 1), Gaussian process regression is used to establish weak predictors of different kernel functions. The optional kernel function is any function that satisfies Mercer's theorem, including but not limited to square exponential covariance function, periodic covariance function, linear covariance function, quadratic rational kernel function, etc.

[0029] Beneficial effects:

[0030] (1) The present invention uses Gaussian process regression to train weak predictors. Compared with algorithms such as neural networks and support vector regression, the choice of kernel functions is wider and the model parameters are reduced.

[0031] (2) The present invention combines multiple weak predictors into a strong predictor, effectively improving the accuracy of the predictor, thereby extending the time it takes for the integrated navigation system to resist GPS failure.

[0032] (3) The present invention adopts the Gaussian process regression method, which has good adaptability to processing nonlinear and small sample situations. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 A flowchart of a GPS failure tolerance processing method for integrated navigation is provided;

[0034] Figure 2 This is a latitude error simulation curve diagram after adopting the fault-tolerant processing method proposed by the present invention;

[0035] Figure 3 This is a simulation curve diagram of the longitude error after adopting the fault-tolerant processing method proposed in the present invention. DETAILED DESCRIPTION

[0036] The present invention will be further explained below with reference to the accompanying drawings.

[0037] The present invention can be better understood with reference to the following examples. Figure 1As shown, the present invention is a GPS failure fault tolerance processing method for integrated navigation, the specific steps are as follows:

[0038] Step 1) When GPS is valid, collect the measurement information of the inertial measurement unit in the integrated navigation system X = {X k |k=1,2,…n} and the corresponding GPS measurement information Y={Y k |k=1,2,…n} constitute the training set. The collected measurement information is continuously updated with the navigation process. Gaussian process regression is used to establish weak predictors with different kernel functions. The optional kernel function is any function that satisfies Mercer's theorem, including but not limited to the squared exponential covariance function, periodic covariance function, linear covariance function, quadratic rational kernel function, etc.

[0039] in:

[0040] X k is a vector containing the gyroscope and accelerometer measurement data;

[0041] Y k is a vector containing GPS position measurement information;

[0042] n is the number of samples in the training set;

[0043] Step 2) Calculate the training error δ of the weak predictor t t ,

[0044]

[0045] in:

[0046] T is the number of weak predictors;

[0047] H t (X k ) is to convert X k Input the predicted output vector obtained by weak predictor t;

[0048] ||H t (X k )-Y k ||2 is the calculation of H t (X k )-Y k The second norm of ;

[0049] Step 3) Set the weight β of each weak predictor according to the training error t ,

[0050]

[0051] Step 4) If GPS fails, the weak predictors are combined into a strong predictor.

[0052]

[0053] in:

[0054] X * A vector containing gyroscope and accelerometer measurement data when GPS fails;

[0055] Y * is the estimated value of GPS measurement information output by the strong predictor;

[0056] The estimation results of the strong predictor are used for integrated navigation information fusion to achieve high-reliability integrated navigation in the event of GPS failure.

[0057] The feasibility of the present invention is verified by the following simulation:

[0058] (1) GPS-assisted inertial navigation system to form integrated navigation;

[0059] (2) Gyroscope random constant drift 0.03° / h, random white noise The accelerometer has a random constant bias of 0.2 mg, a random white noise of 0.2 mg, and a GPS position measurement error of 10 m.

[0060] (3) The inertial measurement unit data update period is 10ms, the GPS update period is 1s, the integrated navigation filter period is 1s, and the simulation time is 1600s;

[0061] (4) When GPS is valid, a training set is collected, with n = 25, and the samples in the training set are continuously updated as the navigation progresses;

[0062] (5) Four weak predictors are established using Gaussian process regression, i.e., T = 4, whose kernel functions are square exponential covariance function, periodic covariance function, linear covariance function, and quadratic rational kernel function respectively;

[0063] (6) In the time period from 800s to 980s, the GPS is disabled for 180s.

[0064] Through computer simulation, the latitude error simulation curve after adopting the fault tolerance processing method proposed by the present invention is as follows Figure 2 As shown, the longitude error simulation curve after adopting the fault tolerance processing method proposed by the present invention is as follows Figure 3 As shown. Figure 2 、 Figure 3The comparison curves in the figure show that when dealing with a GPS failure lasting up to 180 seconds, the position error of the system without fault tolerance (i.e., pure inertial navigation) accumulates rapidly, while the position error of the system using the GPS failure fault tolerance method proposed in this invention is smaller. Compared with the normal GPS stage, the position error only increases slightly at the beginning of the failure and then maintains a normal error level.

[0065] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A GPS failure tolerance processing method for integrated navigation, characterized in that: The specific steps of this fault-tolerant processing method are as follows: Step 1) When GPS is valid, collect the measurement information of the inertial measurement unit in the integrated navigation system X = {X k |k=1,2,...n} and the corresponding GPS measurement information Y={Y k |k=1,2,...n} constitute the training set, and Gaussian process regression is used to establish weak predictors with different kernel functions. in: X k is a vector containing the gyroscope and accelerometer measurement data; Y k is a vector containing GPS position measurement information; n is the number of samples in the training set; Step 2) Calculate the training error δ of the weak predictor t t , in: T is the number of weak predictors; H t (X k ) is to convert X k Input the predicted output vector obtained by weak predictor t; ||H t (X k )-Y k ||2 is the calculation of H t (X k )-Y k The second norm of ; Step 3) Set the weight β of each weak predictor according to the training error t , Step 4) If GPS fails, the weak predictors are combined into a strong predictor. in: X * A vector containing gyroscope and accelerometer measurement data when GPS fails; Y * is the estimated value of GPS measurement information output by the strong predictor; The estimation results of the strong predictor are used for integrated navigation information fusion to achieve high-reliability integrated navigation in the event of GPS failure.

2. The GPS failure tolerance processing method for integrated navigation according to claim 1, characterized in that: In the step 1), when GPS is valid, the measurement information of the inertial measurement unit in the integrated navigation system is collected. k |k=1,2,...n} and the corresponding GPS measurement information Y={Y k |k=1,2,...n} constitute the training set, and the collected measurement information is continuously updated as the navigation process progresses.

3. The GPS failure tolerance processing method for integrated navigation according to claim 1, characterized in that: In the step 1), Gaussian process regression is used to establish weak predictors of different kernel functions, where the kernel function is any function that satisfies Mercer's theorem, including square exponential covariance function, periodic covariance function, linear covariance function, and quadratic rational kernel function.

Images