阅读说明：本技术 一种基于北斗gnss的高精度定位方法、装置及系统 (High-precision positioning method, device and system based on Beidou GNSS ) 是由 宋欣 于 2020-07-02 设计创作，主要内容包括：本发明涉及一种基于北斗GNSS的高精度定位方法、装置及系统,该方法包括如下步骤：接收随机噪声,构建LSGAN神经网络；当北斗GNSS信号可用时,LSGAN神经网络处于训练模式：将惯性导航系统输出的三维位置信息以及经卡尔曼滤波器融合后输出的惯性导航系统误差补偿值作为LSGAN神经网络训练的输入样本对其训练；当北斗GNSS信号缺失时,LSGAN神经网络进入预测模式,利用训练后的LSGAN神经网络模型预测出惯性导航系统的误差值,利用该误差值得到修正后的三维位置信息。本发明提供的基于北斗GNSS的高精度定位装置及系统,通过多通道卷积网络更好的提取样本数据中的特征信息从而使网络具有较好的泛化能力,避免梯度消失现象,能够更精确地预测三维位置信息。(The invention relates to a high-precision positioning method, a device and a system based on Beidou GNSS, wherein the method comprises the following steps: receiving random noise and constructing an LSGAN neural network; when the big dipper GNSS signal is available, the LSGAN neural network is in a training mode: three-dimensional position information output by an inertial navigation system and an inertial navigation system error compensation value output after fusion by a Kalman filter are used as input samples for LSGAN neural network training to train the three-dimensional position information and the inertial navigation system error compensation value; when the Beidou GNSS signal is lost, the LSGAN neural network enters a prediction mode, an error value of the inertial navigation system is predicted by using the trained LSGAN neural network model, and the corrected three-dimensional position information is obtained by using the error value. According to the high-precision positioning device and system based on the Beidou GNSS, the characteristic information in the sample data is better extracted through the multi-channel convolution network, so that the network has better generalization capability, the gradient disappearance phenomenon is avoided, and the three-dimensional position information can be more accurately predicted.)

1. A high-precision positioning method based on Beidou GNSS is characterized by comprising the following steps:

receiving random noise and constructing an LSGAN neural network model;

when the big dipper GNSS signal is available, the LSGAN neural network is in a training mode: taking the three-dimensional position information output by the inertial navigation system and the inertial navigation system error compensation value output after the fusion of the Kalman filter as input samples of LSGAN neural network training; obtaining a real input sample, performing different amplitude transformation on the input sample, and storing the input sample and the transformation input sample to obtain a transformation input sample set; respectively inputting the transformation input samples in the transformation input sample set into each channel of a multi-channel convolution network of a discrimination network, and extracting and fusing features to obtain an output result;

when the Beidou GNSS signal is lost, the LSGAN neural network enters a prediction mode, navigation position information output by the inertial navigation system is used as input of a trained network, an error value of the inertial navigation system is predicted by using the trained LSGAN neural network model, and corrected three-dimensional position information is obtained by using the error value.

2. The Beidou GNSS based high precision positioning method according to claim 1, further comprising:

and when the error value between the actual output and the expected output sample of the LSGAN neural network is greater than the set threshold value, circularly utilizing the LSGAN neural network algorithm to obtain the updated value of the network weight value of the LSGAN neural network until the error value between the actual output and the expected output of the LSGAN neural network is less than the set threshold value.

3. The Beidou GNSS based high precision positioning method according to claim 1, wherein the LSGAN generation network comprises: nine layers including a first full-connection layer, a first micro-step winding layer, a second micro-step winding layer, a third micro-step winding layer, a fourth micro-step winding layer, a fifth micro-step winding layer, a sixth micro-step winding layer, a seventh micro-step winding layer and a second full-connection layer which are connected in sequence.

4. The Beidou GNSS based high precision positioning method according to claim 1, wherein the LSGAN discriminant network comprises: the device comprises a first feature extraction layer, a first feature mapping layer, a first acceptance layer, a second feature extraction layer, a second feature mapping layer, a second acceptance layer, a first full-connection layer and a least square loss calculation layer which are sequentially connected.

5. The Beidou GNSS based high precision positioning method according to any of the claims 1 to 4, wherein the process of transforming the input samples to obtain transformed output samples comprises: and respectively carrying out horizontal direction, forty-five degree direction, sixty degree direction, ninety degree direction, one hundred twenty degree direction, one hundred thirty-five degree direction and one hundred fifty degree direction on a course angle, a roll angle and a pitch angle in the inertial navigation error compensation value output by the Kalman filter after fusion to obtain a converted output sample.

6. The Beidou GNSS-based high-precision positioning method according to claim 5, wherein the process of inputting the transformation samples in the transformation sample set into each channel of a multi-channel convolution network of a discriminant network respectively and extracting and fusing features comprises: and correspondingly inputting the seven transformation samples in the transformation input sample set into seven channels in a multi-channel convolution network of a discrimination network according to gradient change, extracting the characteristics of the corresponding transformation samples in each channel, and performing characteristic fusion on the characteristics of the seven channels in a random form.

7. A high-precision positioning device based on Beidou GNSS comprises a generation module, a transformation module and a discrimination module,

the generating module is used for receiving noise, acquiring the current three-dimensional position and an inertial navigation error compensation value output after Kalman filter fusion, and generating three-dimensional information;

the transformation module is used for receiving the inertial navigation system error compensation value output after the fusion of the real Kalman filter, transforming the inertial navigation system error compensation value at different angles, and storing the inertial navigation system error compensation value and the transformation value thereof as input samples to obtain a transformation sample set;

and the judging module is used for respectively inputting the transformation input samples in the transformation input sample set into each channel of a multi-channel convolution network of the judging network by utilizing the LSGAN neural network, and extracting and fusing the characteristics to obtain an output result.

8. The Beidou GNSS-based high-precision positioning device according to claim 7, wherein the generation module comprises a generation network unit, and the generation network unit comprises nine layers including a first full connection layer, a first micro-step amplitude convolution layer, a second micro-step amplitude convolution layer, a third micro-step amplitude convolution layer, a fourth micro-step amplitude convolution layer, a fifth micro-step amplitude convolution layer, a sixth micro-step amplitude convolution layer, a seventh micro-step amplitude convolution layer and a second full connection layer which are connected in sequence.

9. The Beidou GNSS based high precision positioning device according to claim 7, wherein the discrimination module comprises a discrimination network, the discrimination network comprises: the device comprises a first feature extraction layer, a first feature mapping layer, a first acceptance layer, a second feature extraction layer, a second feature mapping layer, a second acceptance layer, a first full-connection layer and a least square loss calculation layer which are sequentially connected.

10. A high-precision positioning system based on the beidou GNSS, comprising the high-precision positioning device based on the beidou GNSS as claimed in any one of claims 7 to 9.

Technical Field

The invention relates to the field of satellite positioning and navigation, in particular to a high-precision positioning method, device and system based on a Beidou GNSS.

Background

At the end of 6 months in 2020, China completes 30 satellite launching networking and comprehensively builds a Beidou third system. The Beidou third system inherits two technical systems of active service and passive service, provides positioning navigation time service, global short message communication and international search and rescue service for global users, and can provide services such as satellite-based enhancement, foundation enhancement, precise single-point positioning, regional short message communication and the like for users in China and surrounding areas. The space signal precision of the Beidou third system is better than 0.5 meter; the global positioning precision is better than 10 meters, the speed measurement precision is better than 0.2 meter/second, and the time service precision is better than 20 nanoseconds; the positioning precision of the Asia-Pacific region is superior to 5 meters, the speed measurement precision is superior to 0.1 meter/second, the time service precision is superior to 10 nanoseconds, and the overall performance is greatly improved.

GAN (Generative adaptive Nets) inspired by two-person game in game theory, since iang goodfellow published the paper Generative adaptive Nets in 14 years, GAN is of great interest and has become a new favorite in recent years in the field of machine learning.

Disclosure of Invention

Aiming at the technical problems in the prior art, the first aspect of the invention provides a high-precision positioning method based on Beidou GNSS, which comprises the following steps: receiving random noise and constructing an LSGAN neural network model; when the big dipper GNSS signal is available, the LSGAN neural network is in a training mode: taking the three-dimensional position information output by the inertial navigation system and the inertial navigation system error compensation value output after the fusion of the Kalman filter as input samples of LSGAN neural network training; obtaining a real input sample, performing different amplitude transformation on the input sample, and storing the input sample and the transformation input sample to obtain a transformation input sample set; respectively inputting the transformation input samples in the transformation input sample set into each channel of a multi-channel convolution network of a discrimination network, and extracting and fusing features to obtain an output result; when the Beidou GNSS signal is lost, the LSGAN neural network enters a prediction mode, navigation position information output by the inertial navigation system is used as input of a trained network, an error value of the inertial navigation system is predicted by using the trained LSGAN neural network model, and corrected three-dimensional position information is obtained by using the error value.

In some embodiments of the invention, the method further includes when the error value between the actual output and the expected output sample of the LSGAN neural network is greater than the set threshold, recycling the LSGAN neural network algorithm to obtain an updated value of the network weight value thereof until the error between the actual output and the expected output of the LSGAN neural network is less than the set threshold.

In some embodiments of the invention, the network of LSGAN generation comprises: nine layers including a first full-connection layer, a first micro-step winding layer, a second micro-step winding layer, a third micro-step winding layer, a fourth micro-step winding layer, a fifth micro-step winding layer, a sixth micro-step winding layer, a seventh micro-step winding layer and a second full-connection layer which are connected in sequence.

In some embodiments of the invention, the discriminative network of LSGANs comprises: the device comprises a first feature extraction layer, a first feature mapping layer, a first acceptance layer, a second feature extraction layer, a second feature mapping layer, a second acceptance layer, a first full-connection layer and a least square loss calculation layer which are sequentially connected.

In the foregoing embodiment, the process of transforming the input samples to obtain transformed output samples includes: and respectively carrying out horizontal direction, forty-five degree direction, sixty degree direction, ninety degree direction, one hundred twenty degree direction, one hundred thirty-five degree direction and one hundred fifty degree direction on a course angle, a roll angle and a pitch angle in the inertial navigation error compensation value output by the Kalman filter after fusion to obtain a converted output sample.

Preferably, the process of inputting the transformation samples in the transformation sample set into each channel of a multi-channel convolution network of a discrimination network, and extracting and fusing features includes: and correspondingly inputting the seven transformation samples in the transformation input sample set into seven channels in a multi-channel convolution network of a discrimination network according to gradient change, extracting the characteristics of the corresponding transformation samples in each channel, and performing characteristic fusion on the characteristics of the seven channels in a random form.

The invention provides a high-precision positioning device based on the Beidou GNSS, which comprises a generation module, a transformation module and a discrimination module,

the generating module is used for receiving noise, acquiring the current three-dimensional position and an inertial navigation error compensation value output after Kalman filter fusion, and generating three-dimensional information; the transformation module is used for receiving the inertial navigation system error compensation value output after the fusion of the real Kalman filter, transforming the inertial navigation system error compensation value at different angles, and storing the inertial navigation system error compensation value and the transformation value thereof as input samples to obtain a transformation sample set; and the judging module is used for respectively inputting the transformation input samples in the transformation input sample set into each channel of a multi-channel convolution network of the judging network by utilizing the LSGAN neural network, and extracting and fusing the characteristics to obtain an output result.

In some embodiments of the invention, the generation module includes a generation network unit, and the generation network unit includes nine layers, namely, a first full connection layer, a first micro-step amplitude convolution layer, a second micro-step amplitude convolution layer, a third micro-step amplitude convolution layer, a fourth micro-step amplitude convolution layer, a fifth micro-step amplitude convolution layer, a sixth micro-step amplitude convolution layer, a seventh micro-step amplitude convolution layer, and a second full connection layer, which are connected in sequence.

In some embodiments of the invention, the discrimination module comprises a discrimination network comprising: the device comprises a first feature extraction layer, a first feature mapping layer, a first acceptance layer, a second feature extraction layer, a second feature mapping layer, a second acceptance layer, a first full-connection layer and a least square loss calculation layer which are sequentially connected.

The third aspect of the invention provides a high-precision positioning system based on the Beidou GNSS, which comprises the high-precision positioning device based on the Beidou GNSS.

Drawings

FIG. 1 is a schematic diagram of a generic GAN model structure;

FIG. 2 is a flow chart of a Beidou GNSS based high precision positioning method in some embodiments of the present invention;

fig. 3 is a diagram of an inclusion model in some embodiments of the invention;

fig. 4 is a block diagram of a high-precision positioning system based on the Beidou GNSS in some embodiments of the present invention.

Detailed Description

The principles and features of this invention are described below in conjunction with the following drawings, which are set forth by way of illustration only and are not intended to limit the scope of the invention.

First, some important terms in the present application are explained:

referring to fig. 1, the main inspiration of GAN comes from the thought of the game of zero sum in the game theory, and when applied to the deep learning neural network, the G learns the distribution characteristics of data by generating a network G (generator) and a discrimination network d (discriminator), and if the G is used for generating pictures, the G can generate a realistic image from a random number after training. G, D main functions are:

g is a generative network which receives a random noise z (random number) by which to generate an image; d is a discrimination network to discriminate whether a picture is "real". The input parameter is x, x represents a picture, and the output D (x) represents the probability that x is a real picture, if 1, 100% of the picture is real, and the output is 0, the picture cannot be real.

The above example is for the picture domain, but applications of GAN include, but are not limited to, the image processing domain.

LSGAN (Least square GAN) principle:

that is, a least squares loss function is used instead of the loss function of GAN, but with such a change, the problems of instability of GAN training and insufficient diversity of generated data are alleviated.

Decision boundary: in a statistical classification problem with two classes, the decision boundary or decision surface is a hypersurface that divides the basis vector space into two sets, one set. The classifier classifies all points on one side of the decision boundary as belonging to one class and all points on the other side as belonging to the other class.

Referring to fig. 2, a first aspect of the present invention provides a high-precision positioning method based on a Beidou GNSS, including the following steps: s1: receiving random noise, and constructing an LSGAN neural network (or called LSGAN neural network, or LSGAN for short); s2: determining whether the Beidou GNSS signals are available; s3: when the big dipper GNSS signal is available, the LSGAN neural network is in a training mode: taking the three-dimensional position information output by the inertial navigation system and the inertial navigation system error compensation value output after the fusion of the Kalman filter as input samples of LSGAN neural network training; obtaining a real input sample, performing different amplitude transformation on the input sample, and storing the input sample and the transformation input sample to obtain a transformation input sample set; respectively inputting the transformation input samples in the transformation input sample set into each channel of a multi-channel convolution network of a discrimination network, and extracting and fusing features to obtain an output result; s4: when the Beidou GNSS signal is lost, the LSGAN neural network enters a prediction mode, navigation position information output by the inertial navigation system is used as input of a trained network, an error value of the inertial navigation system is predicted by using the trained LSGAN neural network model, and corrected three-dimensional position information is obtained by using the error value.

In order to improve the fitting ability and prediction accuracy of the LSGAN samples, the data in the sample set is normalized:

suppose a GNSS failure time t₀The latitude and longitude at the previous moment is theta (t)₀-1)，(t₀-1), let T be the failure duration and f be the data acquisition frequency, then the predicted step number g ═ txf, the predicted trajectory and error are:

wherein:

predicting latitude and longitude increments for t moments output by the neural network;is a predicted latitude, longitude; e.g. of the type_θ(t₀+g)、e_s(t₀+ g) is the predicted longitude, latitude error; e (t)₀+ g) is the prediction error.

Taking a general motor vehicle as an example, the speed range is 0-120km/h, mainly concentrated in the range of 20-40km/h, about 50%: the course angle ranges from 0 degree to 360 degrees, mainly focusing in the ranges of 120 degrees to 160 degrees and 300 degrees to 340 degrees, and respectively accounting for 28 percent and 24 percent; the transverse rolling angle ranges from 0 to 5 degrees and is mainly concentrated at l to 3 degrees; the pitch angle ranges from-12 degrees to 12 degrees, and is mainly concentrated at-5 degrees to 5 degrees. (ii) a The acceleration range is-8-8 m/s²Normalizing said data to fall within the range [ -1,1 [ ]]It can also be according to the convergence of the actual LSGAN neural network as [ -10,10 ]]And adjusting within the interval.

In some embodiments of the invention, the method further includes when the error value between the actual output and the expected output sample of the LSGAN neural network is greater than the set threshold, recycling the LSGAN neural network algorithm to obtain an updated value of the network weight value thereof until the error between the actual output and the expected output of the LSGAN neural network is less than the set threshold.

In some embodiments of the invention, the network of LSGAN generation comprises: nine layers including a first full-connection layer, a first micro-step winding layer, a second micro-step winding layer, a third micro-step winding layer, a fourth micro-step winding layer, a fifth micro-step winding layer, a sixth micro-step winding layer, a seventh micro-step winding layer and a second full-connection layer which are connected in sequence.

Referring to fig. 3, in some embodiments of the invention, a discriminative network of LSGANs includes: the device comprises a first feature extraction layer, a first feature mapping layer, a first acceptance layer, a second feature extraction layer, a second feature mapping layer, a second acceptance layer, a first full-connection layer and a least square loss calculation layer which are sequentially connected.

In the foregoing embodiment, the process of transforming the input samples to obtain transformed output samples includes: and respectively carrying out horizontal direction, forty-five degree direction, sixty degree direction, ninety degree direction, one hundred twenty degree direction, one hundred thirty-five degree direction and one hundred fifty degree direction on a course angle, a roll angle and a pitch angle in the inertial navigation error compensation value output by the Kalman filter after fusion to obtain a converted output sample.

In the LSGAN neural network in the above embodiment, the distribution can be as close to the decision boundary as possible using least squares, and the loss function of the LSGAN neural network is defined as follows:

in the above formula, pdata (x) is real sample data, pz (x) is random noise, G is a generation network, D is a discrimination network, and V is_LSGAN(D) For discriminating the optimal objective function of the network, V_LSGAN(G) Generating an optimized objective function for the network; where a is set to 0, c is set to 1, b is set to 1, x represents sample data in the test sample set, and z is three-dimensional information. The sample data comprises the longitude and latitude, the speed, the acceleration, the heading angle, the roll angle and the pitch angle.

Preferably, the process of inputting the transformed samples in the transformed sample set into each channel of a multi-channel convolutional network of a decision network, and extracting and fusing features includes: and correspondingly inputting the seven transformation samples in the transformation input sample set into seven channels in a multi-channel convolution network of a discrimination network according to gradient change, extracting the characteristics of the corresponding transformation samples in each channel, and performing characteristic fusion on the characteristics of the seven channels in a random form. The characteristics are several data extracted from the aforementioned samples, preferably, the characteristics are heading angle, roll angle and pitch angle in the present application.

The invention provides a high-precision positioning device based on the Beidou GNSS, which comprises a generation module, a transformation module and a discrimination module,

the generating module is used for receiving noise, acquiring the current three-dimensional position and an inertial navigation error compensation value output after Kalman filter fusion, and generating three-dimensional information; the transformation module is used for receiving the inertial navigation system error compensation value output after the fusion of the real Kalman filter, transforming the inertial navigation system error compensation value at different angles, and storing the inertial navigation system error compensation value and the transformation value thereof as input samples to obtain a transformation sample set; and the judging module is used for respectively inputting the transformation input samples in the transformation input sample set into each channel of a multi-channel convolution network of the judging network by utilizing the LSGAN neural network, and extracting and fusing the characteristics to obtain an output result.

In some embodiments of the invention, the generation module includes a generation network unit, and the generation network unit includes nine layers, namely, a first full connection layer, a first micro-step amplitude convolution layer, a second micro-step amplitude convolution layer, a third micro-step amplitude convolution layer, a fourth micro-step amplitude convolution layer, a fifth micro-step amplitude convolution layer, a sixth micro-step amplitude convolution layer, a seventh micro-step amplitude convolution layer, and a second full connection layer, which are connected in sequence.

In some embodiments of the invention, the discrimination module comprises a discrimination network comprising: the device comprises a first feature extraction layer, a first feature mapping layer, a first acceptance layer, a second feature extraction layer, a second feature mapping layer, a second acceptance layer, a first full-connection layer and a least square loss calculation layer which are sequentially connected.

Referring to fig. 4, a third aspect of the present invention provides a high-precision positioning system based on a Beidou GNSS, including a GNSS receiver, an LSGAN neural network, a kalman filter model, an inertial navigation sensor, a GNSS antenna, and an RTK base station. The steps (i), (ii), (iii) and (iv) correspond to the steps (S1), (S2), (S3) and (S4), respectively, and are not described herein again.

Various aspects of the present disclosure are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the disclosure. 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-readable program instructions.

These computer-readable program instructions may be provided to a processing unit of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processing unit of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks. These computer-readable program instructions may also be stored in a computer-readable storage medium that can direct a computer, programmable data processing apparatus, and/or other devices to function in a particular manner, such that the computer-readable medium storing the instructions comprises an article of manufacture including instructions which implement the function/act specified in the flowchart and/or block diagram block or blocks.

The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer, other programmable apparatus or other devices implement the functions/acts specified in the flowchart and/or block diagram block or blocks.

The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

The foregoing description of the embodiments of the present disclosure has been presented for purposes of illustration and description, and is not intended to be exhaustive or limited to the embodiments disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen in order to best explain the principles of the embodiments, the practical application, or improvements made to the technology in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.