CN110516198B - Distributed nonlinear Kalman filtering method - Google Patents
Distributed nonlinear Kalman filtering method Download PDFInfo
- Publication number
- CN110516198B CN110516198B CN201910645592.5A CN201910645592A CN110516198B CN 110516198 B CN110516198 B CN 110516198B CN 201910645592 A CN201910645592 A CN 201910645592A CN 110516198 B CN110516198 B CN 110516198B
- Authority
- CN
- China
- Prior art keywords
- node
- function
- posterior distribution
- distributed
- time
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/12—Discovery or management of network topologies
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Physics (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Databases & Information Systems (AREA)
- Operations Research (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Algebra (AREA)
- Evolutionary Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Complex Calculations (AREA)
Abstract
The invention belongs to the field of signal processing, in particular to a distributed nonlinear Kalman filtering method, which describes the real state of a target by using a posterior distribution function, and approximates the posterior distribution function by using a function distributed by an exponential family; firstly, the invention obtains the intermediate state estimation by optimizing the backward KL divergence between the intermediate posterior distribution approximate function and the real posterior distribution function of each node; then, the final state estimation result of each node is calculated by optimizing the convex combination of the forward KL divergence between the final posterior distribution approximation function of each node and the intermediate posterior distribution approximation function of the neighboring node. The invention has high convergence speed, adopts a mode that a plurality of nodes work simultaneously, can effectively save the calculation cost and greatly improve the calculation efficiency, and is widely applied to the fields of positioning, target tracking and the like of a distributed infinite sensor and a nonlinear power system.
Description
Technical Field
The invention belongs to the field of signal processing, relates to a target tracking problem in the field of signal processing, particularly relates to a target tracking problem on a distributed wireless sensor network, and particularly relates to a distributed nonlinear Kalman filtering method.
Background
Because the Kalman filtering is very suitable for real-time signal processing and large state space models, the Kalman filtering has wide application in the fields of communication systems, power systems, aerospace, environmental pollution control, industrial control, radar signal processing and the like since the time comes, and a plurality of achievements are obtained. In recent years, with the explosive growth of sequences and streaming data, kalman filters have been widely used in machine learning.
Kalman filtering is an algorithm for optimally estimating the state of a system by inputting observation data through the system by using a linear system state equation. However, the typical kalman filtering is only suitable for linear signal models, and in real life, more nonlinear environments are often encountered. At present, the research on the general nonlinear filtering problem is quite active, and the common nonlinear filtering is Extended Kalman Filtering (EKF), insensitive kalman filtering (UKF) and the like; however, these algorithms perform linear approximation processing on the nonlinear kalman filter system, and then perform processing by using the method of the linear kalman filter system.
Common centralized computing needs to consume a large amount of time and resources, and a distributed method decomposes problems into a plurality of small parts which are distributed to a plurality of computers for processing, so that the computing cost can be saved, and the computing efficiency is greatly improved. In recent years, cooperative flooding processing over distributed wireless sensor networks has become an effective data processing technique; the processing mode greatly improves the expandability and the flexibility of the network and is widely applied to the fields of environment monitoring, disaster relief management, parameter estimation, target tracking and the like. However, most of the existing distributed non-linear kalman filtering is based on extended kalman or insensitive kalman filtering, and linear approximation processing is performed on the non-linear kalman when the state estimation of each node is calculated.
Disclosure of Invention
The invention aims to provide a distributed nonlinear Kalman filtering method, which adopts a distribution function q (x) of an index family to approximate real posterior distribution, measures the difference between the distribution function q (x) and the real posterior distribution by using KL divergence, optimizes the KL divergence to be minimum, and ensures that the q (x) is closest to the real posterior distribution p (x | y).
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
a distributed nonlinear Kalman filtering method comprises the following steps:
Step 2, setting a sampling suggestion distribution function at the time tSampling according to the proposed distribution function such that each particle is independently co-distributed to the proposed distribution:
wherein S is the number of input sampling particles,a state vector representing the s-th particle sampled at time t by node k;
and 3, step 3: and calculating the particle weight of the node k according to the following formula, and performing normalization processing:
wherein the content of the first and second substances,represents the weight of the s-th sampled particle at time t of node k, W k,t Representing the sum of the weights of all the sampled particles at time t, y, of node k l,t Representing the observed value of the neighbor node l of the node k at the time t; />Representing neighbor node l pairs &>Prediction of the observed value; />Representing node k to a particle s state vector>Predicting;
and 4, step 4: calculating an intermediate posterior distribution approximation function based on the particle weights calculated by the above equationIn (d) is based on the mean value>And covariance matrix>
and 5: in a distributed network, the intermediate posterior distribution approximation function obtained by each nodeDiffusion is performed according to an intermediate posterior distribution approximation function of each neighbor node of node k>Calculating to obtain the final state estimation q of the node k k (x k,t ):
Wherein, a l,k The weight of the neighbor node l to the current node k is satisfiedAnd calculate q k (x k,t ) Mean value of (a) k,t Sum covariance matrix sigma k,t :
μ k,t As a state estimation vector for node k.
The invention has the beneficial effects that:
the nonlinear distributed Kalman filtering method provided by the invention has the following advantages:
1. the algorithm provided by the invention is suitable for a nonlinear environment, and has a wider application range compared with a classical linear Kalman filtering algorithm;
2. the distributed algorithm provided by the invention only needs to obtain the observation information of the neighbor nodes at each node, allows each node to process simultaneously, does not need to send all the observation information to the fusion center for processing, requires less communication energy and has higher operation efficiency;
3. the distributed algorithm provided by the invention has stronger robustness compared with a corresponding centralized algorithm. For the centralized type, when the processing center has a problem, the whole system can not work normally, and the distributed algorithm can effectively avoid the risk of crash of the whole system when the fusion center has a problem;
4. when the intermediate state estimation of each node is calculated, the backward KL divergence between the approximate distribution and the real posterior distribution is directly optimized based on the unbiased algorithm directly optimized by the Monte Carlo technology, and the linear approximation of a nonlinear function is not needed;
5. the method is realized by diffusing the intermediate state estimation result of each node in the neighborhood. When the number of nodes in the topological structure of the distributed network and the number of neighbor nodes corresponding to each node are large, the method provided by the invention has smaller steady-state error fluctuation than that of a corresponding centralized method.
Drawings
Fig. 1 is a schematic flow diagram of each node in the distributed nonlinear kalman filtering method according to the present invention.
Fig. 2 shows a distributed network topology (taking 10 nodes in the network as an example) adopted in the embodiment.
Fig. 3 is a graph of the tracking result of a node in a centralized monte carlo experiment according to the embodiment of the present invention.
Fig. 4 is a diagram comparing MSE to centralized location in accordance with the present invention.
Fig. 5 is a diagram comparing MSE with centralized speed in the example of the invention.
Detailed Description
The invention is further illustrated by the following figures and examples.
In the present invention, consider thatDistributed network architecture of nodes, each node being treated with a distribution function of an exponential family>To approximate the a posteriori distribution function &'s for the node>It is called as middleAn approximation function of the posterior distribution, the invention takes>The difference between the two is measured by KL divergence as a Gaussian function; defining the backward KL divergence expression for each node as:
where k denotes the kth node of the overall topology, N k A neighborhood network (including itself) representing node k; x is a radical of a fluorine atom k,t Representing the state vector at the time t of the kth node; y is k,t Representing observation data of the node k at the time t; suppose k's neighbor nodes are respectively usedIndicates, then>Representing that all the neighbor nodes of k obtain an observation data set at the moment t:y 1:t represents a set of observation information from time 1 to t; />Set for representing observation information of all neighbor nodes of node k at 1-t
It is assumed that the observed information of each neighboring node is independent of each other, i.e.Are independent of each other;the expression node k is obtained according to the accumulated observation information of all the neighbor nodes at the time from 1 to tEstimating the state of the last posteriori; />
Optimizing the backward KL divergence of each node to minimize the backward KL divergence to obtain an intermediate posterior distribution approximation function closest to the true posterior distribution in the sense of the backward KL divergenceBecause of the selected->The function being Gaussian distributed, by pairsAnd &>Performing moment matching to obtain: />E represents expectation; calculated to be->I.e. get>Mean and covariance matrix of (a), i.e. can be determined>A function.
The present embodiment provides a distributed non-linear kalman filtering method, a flow of which is shown in fig. 1; in this embodiment, a state estimation result is obtained by calculating each node in the distributed network;
the method comprises the following specific steps:
step 2, setting a sampling suggestion distribution function at the time tSampling according to the proposed distribution function such that each particle is independently identically distributed to the proposed distribution:
wherein S is the number of input sampling particles,a state vector representing the s-th particle sampled by node k at time t;
and 3, step 3: and calculating the particle weight of the node k according to the following formula, and performing normalization processing:
wherein the content of the first and second substances,representing the weight, W, of the s-th sampled particle at time t of node k k,t Represents the sum of the weights of all the sampled particles at time t of node k, y l,t Representing the observed value of the neighbor node l of the node k at the time t; />Represents neighbor node/pair->Predicting the observation value by calculating an observation equation; />Represents the status vector ≥ of node k for particle s>The prediction is obtained by calculating a state transition equation;
Therefore, the temperature of the molten steel is controlled,due to being ^ based>A Gaussian distribution function, i.e. get->The distribution function of (c) is as follows:
and 5: in a distributed network, each willIntermediate posterior distribution approximation function obtained by nodePerforming diffusion according to the intermediate posterior distribution approximation function of each neighbor node of the node k>Correcting the state estimation of the node k to obtain the final state estimation q of the node k k (x k,t );
Defining the forward KL divergence between the final state estimation of the node k and the intermediate posterior distribution approximation function of the neighbor nodes as:
Minimizing the forward KL divergence, we get:because in which>All obey a Gaussian distribution, so q k (x k,t ) Q, also following a Gaussian distribution, calculated by k (x k,t ) Mean value μ k,t Sum covariance matrix Σ k,t
Resulting in a final state estimate: q. q of k (x k,t )~N(μ k,t ,Σ k,t ),q k (x k,t ) The mean value is the state estimation vector of the node;
step 6: and carrying out time iteration to obtain a state estimation result at the next moment.
Simulation conditions
Simulation experiment: the method provided by the invention is used for target tracking of a distributed network, and is compared With a centralized method provided by an article 'Nonlinear Kal-man Filtering With university Minimization'. The topological structure of the whole distributed network is 10 nodes, the topological structure is shown in fig. 2, one sensor at each node respectively tracks a target, and the observed noise power of each node is the same; the state space model of the object motion is as follows:
x t =F t x t-1 +w t ,w t ~N(0,Q t )
y t =h(x t )+v t ,v t ~N(0,R t )
wherein, F t Represents the state transition matrix, h (x) t ) Representing an observation function, w t Is state noise, v t Is the observation noise, w t And v t Mean value is zero and covariance matrix is Q t And R t White gaussian noise of (1):
wherein s represents position information of the sensor; in the present invention, a constant measurement rate is adopted, let Δ t =1,σ CV =10 -2 (ii) a Covariance matrix of observed noise isσ R =20. The state vector is four-dimensional: />Position information including two directions, the second dimension and the fourth dimension representing velocity information of the two directions; the number of particles was sampled 500 times, iterated 300 times, and subjected to 100 monte carlo experiments. The simulation results of the method of the invention and the centralized nonlinear Kalman filtering are shown in FIGS. 3, 4 and 5.
FIG. 3 shows that the distributed nonlinear Kalman filtering method provided by the invention can effectively track the target. Fig. 4 and 5 show that compared With the centralized method (labeled as "centralized" in the figure) proposed by "Nonlinear Kalman Filtering With conversion Minimization", the convergence speed of the MSE curve for position and speed estimation is the same, but the steady-state error fluctuation of the method proposed by the present invention is smaller, and the steady-state performance of the method is even slightly better than that of the centralized algorithm.
While the invention has been described with reference to specific embodiments, any feature disclosed in this specification may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise; all of the disclosed features, or all of the method or process steps, may be combined in any combination, except mutually exclusive features and/or steps.
Claims (1)
1. A distributed nonlinear Kalman filtering method comprises the following steps:
step 1, aiming at a node k, according to a posterior distribution function of the t-1 moment of the node kThe prior distribution function at time t is calculated with the state transition equation>
Step 2, setting a sampling suggestion distribution function at the time tSampling according to the proposed distribution function such that each particle is independently co-distributed to the proposed distribution:
wherein S is the number of input sampling particles,a state vector representing the s-th particle sampled at time t by node k;
and step 3: and calculating the particle weight of the node k according to the following formula, and performing normalization processing:
wherein the content of the first and second substances,representing the weight, W, of the s-th sampled particle at time t of node k k,t Representing the sum of the weights of all the sampled particles at time t, y, of node k l,t Representing the observed value of the neighbor node l of the node k at the time t; />Represents a neighbor node l pairPrediction of the observed value; />Representing node k versus a particle s state vector->Predicting;
and 4, step 4: calculating the intermediate posterior distribution approximation function according to the particle weights obtained by the above formulaIs based on the mean value->And covariance matrix ≥>
and 5: in a distributed network, the intermediate posterior distribution approximation function obtained by each nodePerforming diffusion according to the intermediate posterior distribution approximation function of each neighbor node of the node k>Calculating to obtain the final state estimation q of the node k k (x k,t ):
Wherein, a l,k The weight of the neighbor node l to the current node k is satisfiedAnd calculate q k (x k,t ) Mean value of (a) k,t Sum covariance matrix Σ k,t :/>
μ k,t As a state estimation vector for node k.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910645592.5A CN110516198B (en) | 2019-07-17 | 2019-07-17 | Distributed nonlinear Kalman filtering method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910645592.5A CN110516198B (en) | 2019-07-17 | 2019-07-17 | Distributed nonlinear Kalman filtering method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110516198A CN110516198A (en) | 2019-11-29 |
CN110516198B true CN110516198B (en) | 2023-04-07 |
Family
ID=68622990
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910645592.5A Active CN110516198B (en) | 2019-07-17 | 2019-07-17 | Distributed nonlinear Kalman filtering method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110516198B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110895332B (en) * | 2019-12-03 | 2023-05-23 | 电子科技大学 | Distributed tracking method for extended target |
CN111211760B (en) * | 2020-01-15 | 2023-04-11 | 电子科技大学 | Feedback particle filtering method based on distributed diffusion strategy |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101505532A (en) * | 2009-03-12 | 2009-08-12 | 华南理工大学 | Wireless sensor network target tracking method based on distributed processing |
CN102082560A (en) * | 2011-02-28 | 2011-06-01 | 哈尔滨工程大学 | Ensemble kalman filter-based particle filtering method |
JP2013149203A (en) * | 2012-01-23 | 2013-08-01 | Nippon Telegr & Teleph Corp <Ntt> | Optimal model estimation device, method and program |
WO2014023607A1 (en) * | 2012-08-06 | 2014-02-13 | Agt Group (R&D) Gmbh | System and method for updating a data structure with sensor measurement data |
CN108566178A (en) * | 2018-04-16 | 2018-09-21 | 武汉理工大学 | A kind of random opportunistic network characteristic value filtering method of unstable state |
CN108599737A (en) * | 2018-04-10 | 2018-09-28 | 西北工业大学 | A kind of design method of the non-linear Kalman filtering device of variation Bayes |
CN109671100A (en) * | 2018-11-30 | 2019-04-23 | 电子科技大学 | A kind of distributed variable diffusion direct tracking of combination coefficient particle filter |
CN109710978A (en) * | 2018-11-30 | 2019-05-03 | 电子科技大学 | A kind of direct tracking of distributed heterogeneous adaptive particle filter |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7289906B2 (en) * | 2004-04-05 | 2007-10-30 | Oregon Health & Science University | Navigation system applications of sigma-point Kalman filters for nonlinear estimation and sensor fusion |
-
2019
- 2019-07-17 CN CN201910645592.5A patent/CN110516198B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101505532A (en) * | 2009-03-12 | 2009-08-12 | 华南理工大学 | Wireless sensor network target tracking method based on distributed processing |
CN102082560A (en) * | 2011-02-28 | 2011-06-01 | 哈尔滨工程大学 | Ensemble kalman filter-based particle filtering method |
JP2013149203A (en) * | 2012-01-23 | 2013-08-01 | Nippon Telegr & Teleph Corp <Ntt> | Optimal model estimation device, method and program |
WO2014023607A1 (en) * | 2012-08-06 | 2014-02-13 | Agt Group (R&D) Gmbh | System and method for updating a data structure with sensor measurement data |
CN108599737A (en) * | 2018-04-10 | 2018-09-28 | 西北工业大学 | A kind of design method of the non-linear Kalman filtering device of variation Bayes |
CN108566178A (en) * | 2018-04-16 | 2018-09-21 | 武汉理工大学 | A kind of random opportunistic network characteristic value filtering method of unstable state |
CN109671100A (en) * | 2018-11-30 | 2019-04-23 | 电子科技大学 | A kind of distributed variable diffusion direct tracking of combination coefficient particle filter |
CN109710978A (en) * | 2018-11-30 | 2019-05-03 | 电子科技大学 | A kind of direct tracking of distributed heterogeneous adaptive particle filter |
Non-Patent Citations (3)
Title |
---|
DISTRIBUTED NONLINEAR KALMAN FILTERINGWITH APPLICATIONS TO WIRELESS LOCALIZATION;Federico S. Cattivelli等;《IEEE》;20100628;全文 * |
Minimising the Kullback–Leibler Divergence for Model Selection in Distributed Nonlinear Systems;Oliver M. Cliff等;《Entropy》;20180123;全文 * |
非线性系统的一种分布式状态估计算法及其稳定性分析;杨姝;《电子测量与仪器学报》;20190331;第33卷(第3期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN110516198A (en) | 2019-11-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108304612B (en) | Iterative square root CKF (tracking of target) automobile radar target tracking method based on noise compensation | |
Kantas et al. | Distributed maximum likelihood for simultaneous self-localization and tracking in sensor networks | |
Dong et al. | Robust consensus nonlinear information filter for distributed sensor networks with measurement outliers | |
CN111178385A (en) | Target tracking method for robust online multi-sensor fusion | |
CN110504676B (en) | Power distribution network state estimation method based on APSO-BP | |
CN111127523B (en) | Multi-sensor GMPHD self-adaptive fusion method based on measurement iteration update | |
CN112884237A (en) | Power distribution network prediction auxiliary state estimation method and system | |
CN110516198B (en) | Distributed nonlinear Kalman filtering method | |
CN109687845B (en) | Robust cluster sparse regularization multitask adaptive filter network | |
CN110601777A (en) | Method for estimating satellite-ground downlink co-channel interference under low-orbit mobile satellite constellation | |
CN108566178B (en) | Unsteady random opportunity network characteristic value filtering method | |
CN114626307B (en) | Distributed consistent target state estimation method based on variational Bayes | |
CN111798494A (en) | Maneuvering target robust tracking method under generalized correlation entropy criterion | |
CN109115228B (en) | Target positioning method based on weighted least square volume Kalman filtering | |
Jenabzadeh et al. | Distributed consensus filter for a class of continuous‐time nonlinear stochastic systems in sensor networks | |
CN110532517A (en) | Gas pipeline method for parameter estimation based on improved ARUKF | |
CN110300380B (en) | Target tracking method for balancing system energy consumption and tracking precision in mobile WSN (wireless sensor network) | |
CN110649911B (en) | Distributed nonlinear Kalman filtering method based on alpha divergence | |
George | Distributed maximum likelihood using dynamic average consensus | |
Lee et al. | Design of an alpha-beta filter by combining fuzzy logic with evolutionary methods | |
CN110794676A (en) | CSTR process nonlinear control method based on Hammerstein-Wiener model | |
Kwon et al. | Implementation of computation-efficient sensor network for Kalman filter-based intelligent position-aware application | |
CN115775038A (en) | Short-term load prediction method based on IGWO optimization LSSVM | |
CN109474892B (en) | Strong robust sensor network target tracking method based on information form | |
Liu et al. | The covariance intersection fusion estimation algorithm weighted by diagonal matrix based on genetic simulated annealing algorithm and machine learning |
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 |