CN108759846B - Method for establishing self-adaptive extended Kalman filtering noise model - Google Patents

Method for establishing self-adaptive extended Kalman filtering noise model Download PDF

Info

Publication number
CN108759846B
CN108759846B CN201810527432.6A CN201810527432A CN108759846B CN 108759846 B CN108759846 B CN 108759846B CN 201810527432 A CN201810527432 A CN 201810527432A CN 108759846 B CN108759846 B CN 108759846B
Authority
CN
China
Prior art keywords
value
noise model
neural network
attitude
extended kalman
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201810527432.6A
Other languages
Chinese (zh)
Other versions
CN108759846A (en
Inventor
刘昊
沈晔星
黄成�
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Southeast University
Original Assignee
Southeast University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Southeast University filed Critical Southeast University
Priority to CN201810527432.6A priority Critical patent/CN108759846B/en
Publication of CN108759846A publication Critical patent/CN108759846A/en
Application granted granted Critical
Publication of CN108759846B publication Critical patent/CN108759846B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
    • G01C21/26—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)
  • Automation & Control Theory (AREA)
  • Navigation (AREA)

Abstract

The invention discloses a method for establishing a noise model of adaptive extended Kalman filter, which comprises the following steps: acquiring sensor data, recording a difference value between an observed value and a predicted value in each extended Kalman filtering calculation and a change value of a course angle in the attitude after the filtering calculation, and calculating a correlation coefficient between the difference value between the observed value and the predicted value and the change value of the course angle in the attitude after the filtering calculation; establishing an observation noise model according to a correlation coefficient between a difference value between the observation value and the predicted value and a change value of a course angle in the attitude after filtering calculation; and taking parameters required by the noise model as input of the neural network, taking the target value as output corresponding to the neural network, and performing neural network model training to obtain the optimal parameters for observing the noise model. And extracting the relation between the external environment influence and the sensor data, and searching for the optimal model parameter through a neural network, so that the influence of the external environment on the walking track calculation is minimized.

Description

Method for establishing self-adaptive extended Kalman filtering noise model
Technical Field
The invention relates to the technical field of self-adaptive extended Kalman filtering, in particular to a method for establishing a self-adaptive extended Kalman filtering noise model for walking dead reckoning.
Background
The walking track estimation is one of the important components of the navigation system, and the relative position of the pedestrian can be accurately positioned by means of inertial navigation in areas without satellite signals or weak signals, such as indoor areas, jungles and the like. The existing walking track calculation algorithm generally uses sensors such as an accelerometer, a gyroscope, a magnetometer and the like to judge the attitude so as to complete track calculation, in the attitude judgment process, an extended Kalman filtering algorithm is the key point for judging the attitude by fusing sensor data, the main process is to use gyroscope data to predict the attitude, and then acceleration and magnetometers data are used to correct the predicted attitude to a certain extent with a fixed confidence degree. Therefore, the accuracy of attitude determination directly affects the accuracy of dead reckoning. The existing extended Kalman filtering algorithm for judging the attitude of the sensor usually uses fixed observation noise, which is unreasonable in actual complex and changeable external environments, and the fixed observation noise for a certain sensor can be perfectly used in certain environments, but under the condition that the data of the sensor is greatly influenced by the outside, the dead reckoning precision can be greatly influenced.
Disclosure of Invention
In order to solve the technical problems, the invention aims to provide a method for establishing a self-adaptive extended Kalman filtering noise model, which is used for extracting the relation between the external environment influence and sensor data and searching for an optimal model parameter through a neural network so as to minimize the influence of the external environment on walking track calculation.
The technical scheme of the invention is as follows:
a method for establishing an adaptive extended Kalman filter noise model comprises the following steps:
s01: acquiring sensor data, recording a difference value between an observed value and a predicted value in each extended Kalman filtering calculation and a change value of a course angle in the attitude after the filtering calculation, and calculating a correlation coefficient between the difference value between the observed value and the predicted value and the change value of the course angle in the attitude after the filtering calculation;
s02: establishing an observation noise model according to a correlation coefficient between a difference value between the observation value and the predicted value and a change value of a course angle in the attitude after filtering calculation;
s03: and taking parameters required by the noise model as input of the neural network, taking the target value as output corresponding to the neural network, and performing neural network model training to obtain the optimal parameters for observing the noise model.
In a preferred technical solution, the observation noise model is:
STDEVmag＝k×e+b
wherein e is the difference between the observed value and the predicted value, and k and b are parameters.
In a preferred technical scheme, the neural network model comprises 2 layers, wherein the first layer is a Log-Sigmoid layer and comprises s1A neuron, the second layer being a linear layer.
In a preferred embodiment, the target value in step S03 is a desired value, and the formula is:
t＝k1×STDθ+k2×eθ
wherein STDθStandard deviation of advancing direction, eθFor the offset angle of the back-and-forth path, k1And k is2The track direction consistency and direction accuracy are weighted.
Compared with the prior art, the invention has the advantages that:
1. according to the method, the confidence coefficient of the observation data is dynamically adjusted through the correlation between the sensor data, and the influence of interference data caused by the external environment on attitude judgment is reduced, so that the accuracy of walking track calculation is improved, and the best tracking and positioning effect is achieved.
2. And training the noise model parameters of the sensor data by using a neural network to find the noise model parameters which best meet the actual application expectation, so that the adaptive extended Kalman filtering can achieve a better state estimation effect. The method has the advantages that the navigation direction is ensured to be accurate, the positioning navigation error caused by the interference of the external environment on the sensor is effectively reduced, and compared with the walking track calculation effect of fixed observation noise, the anti-interference performance is improved by about 20% while the direction accuracy is ensured.
Drawings
The invention is further described with reference to the following figures and examples:
FIG. 1 is a flow chart of the adaptive extended Kalman filter noise model building method of the present invention;
FIG. 2 is a schematic diagram of a neural network architecture of the present invention;
FIG. 3 is a schematic diagram of neural network parameter adjustment according to the present invention;
FIG. 4 is a performance surface at the end of training of the present invention;
FIG. 5 is a plot of sum of squares error versus iteration number for the present invention;
FIG. 6 is a graph comparing fixed observation noise and adaptive observation noise traces according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the accompanying drawings in conjunction with the following detailed description. It should be understood that the description is intended to be exemplary only, and is not intended to limit the scope of the present invention. Moreover, in the following description, descriptions of well-known structures and techniques are omitted so as to not unnecessarily obscure the concepts of the present invention.
Example (b):
as shown in fig. 1, a method for building an adaptive extended kalman filter noise model for walking dead reckoning includes the following steps:
the first step is as follows: the method includes collecting straight-line walking data used for training a neural network, using a predicted pose calculated from gyroscope data, and using an accelerometer and a gyroscope for observing the pose. Recording the difference value between the observed value and the predicted value in each time of extended Kalman filtering in the walking process and the change value of the course angle in the attitude after filtering calculation, carrying out correlation analysis on the data, and calculating the correlation coefficient between the difference value between the observed value and the predicted value and the change value of the course angle in the attitude after filtering calculation.
The second step is that: and establishing an observation noise model according to the correlation between the difference value between the observation value and the predicted value and the change value of the course angle in the attitude after filtering calculation, and establishing a linear or nonlinear noise model, wherein the linear model is adopted in the example.
The linear model was established as follows:
STDEVmag＝k×e+b
where e is the difference between the magnetometer data and the predicted estimate, and k and b are the coefficients to be trained. Namely, the most suitable k and b are searched, so that the straight-line walking track is more accurate.
The third step: the method comprises the steps of giving a noise model and different parameters, substituting the model into PDR extended Kalman filtering for calculation to obtain a simulated track, recording track direction accuracy and direction consistency, giving weights with different direction accuracy and consistency according to actual application requirements, and adding the weights to obtain a target value.
The desired outputs are as follows:
t＝k1×STDθ+k2×eθ
STD in the above formulaθStandard deviation of advancing direction, eθIs the offset angle of the back and forth path. k is a radical of1And k is2The track direction consistency and direction accuracy are weighted.
The fourth step: and taking parameters required by the noise model as input of the neural network, taking the target value as corresponding output of the neural network, and carrying out neural network model training until the model error is smaller than a threshold value.
A relatively general two-layer neural network model is adopted to design a network structure, a two-layer neural network structure diagram is shown in fig. 2, and as shown in fig. 2, the two-layer neural network comprises 2 inputs and 1 output, the inputs are coefficients k and b, and the output is expected t. The first layer of the neural network is a Log-Sigmoid layer which contains s1Individual neuron, usually by taking s12-8, the second layer is a linear layer and is also an output layer. For the consideration of minimizing network system parameters, 2 descending experiments are carried out in sequence from 8 hidden neurons, and the minimum system model is selected under the condition that the satisfied error is small enough.
As shown in fig. 3, the above neural network model is trained using the BP algorithm, and for a two-layer neural network, the first step of the algorithm is to propagate the input forward through the network:
p＝[ki bi]T
a0＝p
am+1＝fm+1(wm+1am+bm+1),m＝0,1
a＝a2
where p is the input vector, aiIs the output of the ith layer and a is the network output.
The second step is to propagate the sensitivity back through the network:
Figure BDA0001676298060000041
Figure BDA0001676298060000042
wherein s isiSensitivity to i-th layer back propagation.
The first layer is the Log-Sigmoid layer, so
Figure BDA0001676298060000043
The second layer is a linear layer that is,
Figure BDA0001676298060000044
the last step is to update the weight and the offset value
Wm(k+1)＝Wm(k)-αsm(am-1)T,m＝2,1
bm(k+1)＝bm(k)-αsm
Wherein WiAnd biThe weight matrix and the bias value matrix of the ith layer.
Taking k value from 0 to 10, step 0.5, b value from 0 to 5, step 0.5, and PDR calculation, and corresponding output t is 0.2 × STDθ+1×eθAnd recording and circularly and sequentially substituting the data into a BP algorithm for calculation, and circulating for a plurality of times until the square sum error is smaller than a threshold value, namely the training is successful. Note that the initial values of W and b need to be randomly selected many times to avoid falling into the local optimal solution, and the three-layer neural network model BP training method is similar to the two-layer method.
The training effect is as follows:
in the case of 8 neurons in the first layer, the trained performance surface is shown in FIG. 4 below:
in fig. 4, the deeper the black color, the smaller the value, i.e., the error is the minimum, and the k and b values corresponding to the point are the coefficients required to minimize the model error. It is obvious that the minimum value is obtained when k is 1.4 and b is 2.7, and the standard deviation of the back-and-forth path in each step direction is 18.600 and 20.74, which are respectively reduced by 14.1% and 28.5% compared with the standard deviation of the PDR algorithm with fixed confidence, and it is seen that the anti-interference performance of the PDR algorithm to the external environment is better improved.
FIG. 5 is a graph of the square sum error decrease with iteration number, and it can be seen that after 1 × 10^4 iterations at a learning rate of 0.01, the error is less than the threshold and the model substantially converges to achieve the desired effect. Through experiments, if the learning rate is increased, certain oscillation occurs.
When the model training is finished, a parameter value corresponding to the minimum target value is searched, the parameter value is the optimal parameter of the required noise model, and the optimal parameter obtained by the neural network training is used for track tracking and an original fixed noise track diagram, for example, as shown in fig. 6, it is obvious that the anti-interference performance of the adaptive observation noise PDR track tracking is improved to a certain extent. Table 1 lists the positioning error and the anti-interference performance of the improved adaptive extended kalman filter walking track reckoning algorithm compared with the original algorithm under three paths, and it can be seen from the table that the improved algorithm has greater advantages.
TABLE 1 improved Algorithm Performance comparison Table
Figure BDA0001676298060000051
It is to be understood that the above-described embodiments of the present invention are merely illustrative of or explaining the principles of the invention and are not to be construed as limiting the invention. Therefore, any modification, equivalent replacement, improvement and the like made without departing from the spirit and scope of the present invention should be included in the protection scope of the present invention. Further, it is intended that the appended claims cover all such variations and modifications as fall within the scope and boundaries of the appended claims or the equivalents of such scope and boundaries.

Claims (2)

1. A method for establishing a noise model of adaptive extended Kalman filter is characterized by comprising the following steps:
s01: acquiring sensor data, recording a difference value between an observed value and a predicted value in each extended Kalman filtering calculation and a change value of a course angle in the attitude after the filtering calculation, and calculating a correlation coefficient between the difference value between the observed value and the predicted value and the change value of the course angle in the attitude after the filtering calculation;
s02: establishing an observation noise model according to a correlation coefficient between a difference value between the observation value and the predicted value and a change value of a course angle in the attitude after filtering calculation;
the observation noise model is as follows:
STDEVmag＝k×e+b
wherein e is the difference between the observed value and the predicted value, and k and b are parameters;
s03: taking parameters required by the noise model as input of the neural network, taking the target value as output corresponding to the neural network, and performing neural network model training to obtain optimal parameters for observing the noise model;
the target value in S03 is desired, and the formula is:
t＝k1×STDθ+k2×eθ；
wherein STDθStandard deviation of advancing direction, eθFor the offset angle of the back-and-forth path, k1And k is2The track direction consistency and direction accuracy are weighted.
2. The adaptive extended Kalman filter noise model building method of claim 1, characterized in that the neural network model comprises 2 layers, the first layer is a Log-Sigmoid layer and comprises s1A neuron, the second layer being a linear layer.
CN201810527432.6A 2018-05-29 2018-05-29 Method for establishing self-adaptive extended Kalman filtering noise model Active CN108759846B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810527432.6A CN108759846B (en) 2018-05-29 2018-05-29 Method for establishing self-adaptive extended Kalman filtering noise model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810527432.6A CN108759846B (en) 2018-05-29 2018-05-29 Method for establishing self-adaptive extended Kalman filtering noise model

Publications (2)

Publication Number Publication Date
CN108759846A CN108759846A (en) 2018-11-06
CN108759846B true CN108759846B (en) 2021-10-29

Family

ID=64003209

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810527432.6A Active CN108759846B (en) 2018-05-29 2018-05-29 Method for establishing self-adaptive extended Kalman filtering noise model

Country Status (1)

Country Link
CN (1) CN108759846B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2022126397A1 (en) * 2020-12-15 2022-06-23 深圳市大疆创新科技有限公司 Data fusion method and device for sensor, and storage medium
CN113899362B (en) * 2021-09-09 2023-09-22 武汉大学 Pedestrian dead reckoning method with uncertainty evaluation based on residual network
CN114046785A (en) * 2021-11-10 2022-02-15 广东微电科技有限公司 Magnetic detection signal linear noise filtering method and system, computer readable storage medium, magnetic navigation sensor and AGV trolley

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102692223A (en) * 2012-06-27 2012-09-26 东南大学 Control method of multilevel non-linear filters for wireless sensor network (WSN)/inertial navigation system (INS) combination navigation
CN103149580A (en) * 2013-02-04 2013-06-12 东南大学 Global position system (GPS)/inertial navigation system (INS) combined navigation method based on strong tracking kalman filter (STKF) and wavelet neural network (WNN)
CN103148855A (en) * 2013-02-27 2013-06-12 东南大学 INS (inertial navigation system)-assisted wireless indoor mobile robot positioning method
CN103983263A (en) * 2014-05-30 2014-08-13 东南大学 Inertia/visual integrated navigation method adopting iterated extended Kalman filter and neural network
WO2016198933A1 (en) * 2015-06-11 2016-12-15 SenionLab AB Determining sensor orientation in indoor navigation
CN106969784A (en) * 2017-03-24 2017-07-21 中国石油大学(华东) It is a kind of concurrently to build figure positioning and the combined error emerging system of inertial navigation

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102692223A (en) * 2012-06-27 2012-09-26 东南大学 Control method of multilevel non-linear filters for wireless sensor network (WSN)/inertial navigation system (INS) combination navigation
CN103149580A (en) * 2013-02-04 2013-06-12 东南大学 Global position system (GPS)/inertial navigation system (INS) combined navigation method based on strong tracking kalman filter (STKF) and wavelet neural network (WNN)
CN103148855A (en) * 2013-02-27 2013-06-12 东南大学 INS (inertial navigation system)-assisted wireless indoor mobile robot positioning method
CN103983263A (en) * 2014-05-30 2014-08-13 东南大学 Inertia/visual integrated navigation method adopting iterated extended Kalman filter and neural network
WO2016198933A1 (en) * 2015-06-11 2016-12-15 SenionLab AB Determining sensor orientation in indoor navigation
CN106969784A (en) * 2017-03-24 2017-07-21 中国石油大学(华东) It is a kind of concurrently to build figure positioning and the combined error emerging system of inertial navigation

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
人工神经网络改进的AHRS/GPS紧耦合滤波算法;夏琳琳等;《弹箭与制导学报》;20091215;第29卷(第06期);第199-203页 *
神经 网络辅助 卡尔曼滤波技术在组合导航系统中的应用研究;白宇骏;《中国惯性技术学报》;20030430;第40-43页 *
神经网络校正的卡尔曼滤波算法在车辆GPS/DR组合定位系统中的应用;郑定富 等;《科技尚品》;20160331;第194-195页 *
组合导航中低成本磁航向系统的神经网络补偿;刘育浩等;《传感技术学报》;20081120;第21卷(第11期) *

Also Published As

Publication number Publication date
CN108759846A (en) 2018-11-06

Similar Documents

Publication Publication Date Title
Shen et al. Seamless GPS/inertial navigation system based on self-learning square-root cubature Kalman filter
Dai et al. An INS/GNSS integrated navigation in GNSS denied environment using recurrent neural network
CN106714110B (en) Wi-Fi position fingerprint map automatic construction method and system
CN110118560B (en) Indoor positioning method based on LSTM and multi-sensor fusion
CN108759846B (en) Method for establishing self-adaptive extended Kalman filtering noise model
CN106093849B (en) A kind of Underwater Navigation method based on ranging and neural network algorithm
CN111340868B (en) Unmanned underwater vehicle autonomous decision control method based on visual depth estimation
CN112613532B (en) Moving target tracking method based on radar and cyclic neural network complement infrared fusion
CN113268074B (en) Unmanned aerial vehicle flight path planning method based on joint optimization
CN110401978A (en) Indoor orientation method based on neural network and particle filter multi-source fusion
CN114199248B (en) AUV co-location method for optimizing ANFIS based on mixed element heuristic algorithm
CN110906933A (en) AUV (autonomous Underwater vehicle) auxiliary navigation method based on deep neural network
CN111913175A (en) Water surface target tracking method with compensation mechanism under transient failure of sensor
CN109916388A (en) Fiber Optic Gyroscope Temperature Drift compensation method based on wavelet de-noising and neural network
Jwo et al. Applying back-propagation neural networks to GDOP approximation
CN110346821A (en) A kind of SINS/GPS integrated attitude determination localization method solving the problems, such as GPS long-time losing lock and system
Zhang et al. Application of modified ekf based on intelligent data fusion in auv navigation
CN114819068A (en) Hybrid target track prediction method and system
Dai et al. Robust adaptive UKF based on SVR for inertial based integrated navigation
CN105424043A (en) Motion state estimation method based on maneuver judgment
CN110132287A (en) A kind of satellite high-precision joint method for determining posture based on extreme learning machine network building out
CN113156473A (en) Self-adaptive discrimination method for satellite signal environment of information fusion positioning system
Zheng et al. An rnn-based learnable extended kalman filter design and application
Vertzberger et al. Adaptive attitude estimation using a hybrid model-learning approach
RU2685767C1 (en) Method and device for strap down inertial navigation

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant