Robust filtering method for solving uncertainty of one-step random observation time delay
By designing a robust distributed weighted fusion Kalman estimator, the state estimation problem of multi-sensor network system under the uncertainty of random observation delays is solved, and the system robustness and estimation accuracy are improved.
Patent Information
- Application Number
- CN202510188910.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-20
- Publication Date
- 2025-06-06
AI Technical Summary
The prior art is difficult to provide stable state estimation and control when facing a variety of complex network uncertainties, especially in the case of random observation of delay uncertainties in step.
A robust filtering method is designed to construct three robust distributed weighted fusion Kalman estimators to transform robust local and fusion signal estimation problems into state estimation problems, combining the augmentation method, virtual noise method and the principle of extremely small robust estimation to ensure the accuracy of estimation and control.
It effectively improves the robustness of the multi-sensor network system, so that it can maintain stable estimation performance when facing various uncertainties, and verifies the universality and effectiveness of the method through application examples of ARMA signal processing.
Smart Images

Figure CN120110965A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of information fusion, and in particular to a robust filtering method for solving the uncertainty of one-step random observation delay. Background Art
[0002] A multi-sensor network system is a network consisting of multiple sensor nodes that can sense, process and transmit information in the environment. This system is usually used to monitor and control various physical, chemical or biological parameters, and is widely used in military, environmental monitoring, smart home, industrial automation and healthcare. If the sensor fails or is defective, it may cause the car to fail to correctly identify obstacles or road conditions, resulting in accidents. In the field of industrial automation, estimators need to monitor the status of production lines and predict equipment maintenance needs. In the information and communication industry, robust steady-state estimators can be used to monitor the network status and quickly detect and isolate fault points. This is essential to maintain the stability and robustness of the network.
[0003] As is well known, the classical Kalman filter is highly dependent on the system model parameters and noise variance being accurately known. In practical applications, however, the system model is often accompanied by uncertainty or bias, and the noise characteristics also exhibit complex and variable characteristics. Under these conditions, the performance of the classical Kalman filter may be significantly degraded or even completely ineffective. Therefore, the design of the robust Kalman filter aims to ensure that the filter error variance is maintained at a minimum upper bound in the face of all allowed uncertainties and noise characteristic changes, thereby providing stable and reliable state estimation and significantly enhancing the robustness of the system. When dealing with the uncertainty problem of networked systems, existing studies have explored systems with different situations including lost observations, packet loss, random measurement delays, and lost observations. However, these are limited by the failure to consider the problem of two or more complex network uncertainties at the same time. Different robust fusion strategies have different abilities to handle uncertainty. In a system environment with low uncertainty, a simple fusion method may meet the requirements. However, in a more complex environment, the robust fusion method must be more sophisticated and complex to ensure the accuracy of estimation and control. Summary of the invention
[0004] The purpose of the present invention is to provide a robust filtering method for solving the uncertainty of one-step random observation delay, construct three robust distributed weighted fusion Kalman estimators, transform the robust local and fusion signal estimation problems into state estimation problems for solution, and ensure the accuracy of estimation and control.
[0005] To achieve the above object, the present invention provides a robust filtering method for solving the uncertainty of one-step random observation delay, comprising the following steps:
[0006] Step 1: Establish a linear discrete-time uncertain multi-sensor network system, where the uncertainty induced by the network is expressed using two Bernoulli random variables ξ i (t) and γ i (t) to describe;
[0007] Step 2: Use the augmentation method to transform the model, calculate the second-order non-central moment and prove its accuracy relationship, convert the system in step 1 into an augmented multi-model system, and then introduce virtual noise to obtain an augmented multi-model correlated noise system;
[0008] Step 3: Based on the minimax robust estimation principle, a robust local steady-state Kalman predictor is derived, and then a robust local steady-state Kalman estimator is proposed;
[0009] Step 4: Based on the theory of robust local steady-state Kalman estimator, continue to advance the fusion robust estimator to obtain a robust weighted fusion estimator;
[0010] Step 5: Convert the ARMA signal processing problem into a state estimation problem and verify the universality of the robust weighted fusion estimator.
[0011] Optionally, the expression of the linear discrete-time uncertain multi-sensor network system is as follows:
[0012]
[0013] v i (t) = D i w(t)+η i (t),i=1,...,L
[0014] y i (t) = ξ i (t)z i (t)+(1-ξ i (t))(1-ξ i (t-1))z i (t-1)
[0015] Where: t is discrete time, x(t)∈R n is the state vector at time t, is the observation value of the ith sensor, is the measurement value received by the estimator after transmission, w(t)∈R n is the process noise, is the observation noise and is linearly related to w(t), β k (t)∈R 1 ,k=1,...,q is the state-dependent multiplicative noise. Φ∈R n×n ,Φ k ∈Rn×n ,Γ∈R n×r , and is a suitable constant matrix of known dimension; q is the amount of multiplicative noise, and L is the number of sensors.
[0016] Optionally, during the execution of step 2, the expression after conversion to the augmented multi-model system is as follows:
[0017] x ai (t+1)=Φ ai (t)x ai (t)+Γ ai (t)w ai (t)
[0018] y i (t) = Η ai (t)x ai (t)+D ai (t)w ai (t)
[0019] where Φ ai (t), x ai (t),Γ ai (t), w ai (t), Η ai (t), D ai (t) are the augmented constant matrix, augmented state vector, and augmented noise parameter, respectively, and correspond to the parameter matrix of the original system. The augmented multi-model correlated noise system obtained after the introduction of virtual noise is a time-invariant multi-model correlated noise system with only uncertain actual virtual noise variance, and the expression is as follows:
[0020]
[0021] definition and They are Φ ai (t),Γ ai (t),H ai (t) and D ai (t) is the average matrix.
[0022] Optionally, the robust local steady-state Kalman predictor in step 3 is expressed as follows:
[0023]
[0024] The robust local steady-state Kalman predictor is the basis of the robust local steady-state Kalman filter and smoother, and provides the initial prediction value and prediction error variance for subsequent filtering and fusion.
[0025] Optionally, the constraints of the robust weighted fusion estimator are:
[0026]
[0027] Where θ = CI, FCI1, FCI2 represent CI fusion with matrix weights and two FCI fusions respectively; matrix weights It is a robust local steady-state Kalma estimator for each original subsystem given; the fusion strategies of CI, FCI1 and FCI2 are all based on the minimax robust estimation principle. The CI method constructs the fusion estimate by finding the intersection of the local estimation error covariances to ensure the upper bound of the fusion estimation error variance.
[0028] Optionally, the ARMA signal in step 5 is a multi-sensor single-channel ARMA signal with random parameters, colored measurement noise and uncertain noise variance, and the expression is as follows:
[0029] A t (q -1 )s(t)=w(t),
[0030] z i (t) = γ i (t)s(t)+e i (t),i=1,K,L,
[0031] y i (t) = ξ i (t)z i (t)+(1-ξ i (t))(1-ξ i (t-1))z i (t-1),i=1,K,L,
[0032] A t (q -1 )=1+a 1 (t)q -1 +a 2 (t)q -1 +L+a n (t)q -1 ,a k (t) = a k +β k (t),k=1,K,n.
[0033] where s(t)∈R 1 is the scalar signal to be estimated, w(t)∈R 1 is the input noise, z i (t)∈R 1 is the measurement of the ith sensor, e i (t)∈R1 is the colored measurement noise, A t (q -1 ) is q -1 polynomial, where q -1 is the unit shift operator, i.e., q - 1 s(t)=s(t-1),a k (t), k = 1, K, n is the known mean a k and the random perturbation β k (t); w(t), e i (t) and β k (t) is uncorrelated white noise with zero mean and covariance.
[0034] The present invention provides a robust filtering method for solving the uncertainty of one-step random observation delay, taking into account the state-dependent multiplicative noise in the system state and observation matrix, one-step random delay, packet loss, continuous loss of observation and uncertain noise variance, so that the model is closer to the actual application scenario. Specifically based on the principle of maximum and minimum robust estimation, a CI fusion Kalman estimator (predictor, filter, smoother) and two FCI fusion Kalman estimators are proposed. The present invention combines the augmentation method and the virtual noise method, and introduces a robustness proof method based on the permutation matrix and the Lyapunov equation. By defining different permutation matrices, it is proved that the proposed fusion estimator is robust. In addition, through the application example of autoregressive moving average (ARMA) signal processing, it is verified that the robust local and fusion signal estimation problems can be solved by the state estimation method, thereby providing a solid practical application support for the results of system modeling. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0036] Figure 1 It is a schematic flow chart of the steps of a robust filtering method for solving the uncertainty of one-step random observation delay of the present invention.
[0037] Figure 2 It is a schematic diagram of the execution flow of the robust weighted fusion estimator in the present invention.
[0038] Figure 3 It is a schematic diagram for comparing an actual filtering error curve and its three times standard deviation boundary in a specific embodiment of the present invention.
[0039] Figure 4 It is a schematic diagram of the change of the fusion error curve in a specific embodiment of the present invention. DETAILED DESCRIPTION
[0040] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.
[0041] See also Figure 1 The present invention provides a robust filtering method for solving the uncertainty of one-step random observation delay, comprising the following steps:
[0042] S1: Establish a linear discrete-time uncertain multi-sensor network system, in which the uncertainty induced by the network is implemented using two Bernoulli random variables ξ i (t) and γ i (t) to describe;
[0043] S2: Use the augmentation method to transform the model, calculate the second-order non-central moment and prove its accuracy relationship, convert the system in step 1 into an augmented multi-model system, and then introduce virtual noise to obtain an augmented multi-model correlated noise system;
[0044] S3: Based on the minimax robust estimation principle, a robust local steady-state Kalman predictor is derived, and then a robust local steady-state Kalman estimator is proposed;
[0045] S4: Based on the theory of robust local steady-state Kalman estimator, we continue to advance the fusion robust estimator and obtain a robust weighted fusion estimator;
[0046] S5: Convert the ARMA signal processing problem into a state estimation problem and verify the universality of the robust weighted fusion estimator.
[0047] The following is further described in conjunction with specific execution steps and embodiments:
[0048] like Figure 2 It is a schematic diagram of the execution flow of the robust weighted fusion estimator in the present invention.
[0049] Step S1: Consider the following linear discrete-time uncertain multi-sensor network system:
[0050]
[0051] v i (t) = D i w(t)+η i(t),i=1,...,L
[0052] y i (t) = ξ i (t)z i (t)+(1-ξ i (t))(1-ξ i (t-1))z i (t-1)
[0053] Where: t is discrete time, x(t)∈R n is the state vector at time t, is the observation value of the ith sensor, is the measurement value received by the estimator after transmission, w(t)∈R n is the process noise, is the observation noise and is linearly related to w(t), β k (t)∈R 1 ,k=1,...,q is the state-dependent multiplicative noise. Φ∈R n×n ,Φ k ∈R n×n ,Γ∈R n×r , and is a suitable constant matrix of known dimension; q is the amount of multiplicative noise, and L is the number of sensors.
[0054] In the following analysis, the following assumptions are made:
[0055] Assumption 1.ξ i (t)∈R 1 ,γ i (t)∈R 1 , i = 1, L, L are independent Bernoulli random variables, taking values of 1 and 0 respectively, with known probability prob(ξ i (t) = 1) = ξ i ,prob(ξ i (t) = 0) = 1 - ξ i , and prob(γ i (t) = 1) = π i , prob(γ i (t) = 0) = 1 - π i , where π i and i are all known, satisfying 0≤π i ≤1, and assuming that ξ i (t) and γ i (t) are uncorrelated with other random signals.
[0056] The present invention uses two Bernoulli random variables ξ i (t) and γ i (t) is used to describe various uncertainties induced by the network, including packet loss, one-step random delay, and continuous loss of observations. It should be noted that lost observations and packet loss are two different concepts. Lost observations generally occur at sensor nodes. When this happens, the measurements obtained and packaged by the sensor only contain measurement noise v i (t) or v i (t-1), which indicates that the data collected by the sensor node is invalid. Data packet loss means that the sensor node has successfully obtained the correct data and encapsulated it in the data packet, but the data packet is lost during the data transmission process. In addition, no other data packet is lost at each sampling time t+1. It can be seen that ξ i (t)=0 then y i (t+1)=z i (t+1)(if ξ i (t+1)=1) or y i (t+1)=z i (t)(If ξ i (t+1)=0), so there will be no continuous packet loss.
[0057] Assumption 2. w(t),η i (t),i=1,...,L,and β k (t), k = 1, ..., q are all uncorrelated white noises with zero mean and covariance, where as well as They are white noise w(t),η i (t) and β k The uncertain actual variance of (t), δ kj represents the Kroneckerdelta function, δ kk =1,δ kj =0(k≠j).
[0058] Assumption 3. Initial states x(0) and w(t),η i (t) and β k (t) is uncorrelated, E[x(0)] = μ 0 where μ 0 is the mean of x(0). In addition, the variance of the initial state satisfies in is the unknown uncertain actual variance.
[0059] Hypothesis 4.Q w , and and P0 yes and as well as The known conservative upper limits satisfy:
[0060]
[0061] With unknown and uncertain actual variance and as well as The system with a known conservative upper bound Q is called the actual system, and w , and and P 0 The system is called the worst-case conservative system. The so-called "worst-case" system refers to the system where the noise and initial state variance are both maximized. The "minimum" variance estimator designed for this "worst-case" system is called the minimax robust estimator, which is the minimax robust estimation principle.
[0062] The present invention aims to design local and fusion robust steady-state Kalman estimators (predictors, filters and smoothers) for the state x(t) in an uncertain multi-sensor network system. So that under all allowed uncertainties, their actual steady-state estimation error variance With the corresponding minimum upper bound P θ (N), where θ = i, CI, FCI1, FCI2 represent the i-th local robust estimator, CI, FCI1 and FCI2 fuser weighted by the matrix, respectively. For N = -1, N = 0 and N > 0, they are called one-step predictor, filter and smoother, respectively.
[0063] In step S2, the virtual noise method is used to transform the model into a form with only uncertain noise variance, a new observation vector is introduced, and an augmented multi-model system is constructed.
[0064] Define a new observation vector
[0065]
[0066] Substituting the original formula into the original system, we can get the equivalent conversion of the original system into the augmented multi-model system, as follows:
[0067] x ai (t+1)=Φ ai (t)x ai (t)+Γ ai (t)w ai (t)
[0068] y i (t) = Η ai(t)x ai (t)+D ai (t)w ai (t)
[0069] where the definition and They are Φ ai (t),Γ ai (t),H ai (t) and D ai (t) the average matrix,
[0070]
[0071] Let ξ iz (t) = ξ i (t)-ξ i ,γ iz (t) = γ i (t)-π i and ψ io (t) = ξ iz (t)γ iz (t), we can get their statistical properties and easily prove that ξ iz (t), γ iz (t) and ψ io (t) are all white noise with a mean of zero. iz (t) = ξ i (t)-ξ i ,γ iz (t) = γ i (t)-π i and ψ io (t) = ξ iz (t)γ iz Substituting (t) into Φ ai (t),Γ ai (t),H ai (t) and D ai (t) is the random deviation term.
[0072] In addition, a multi-model system with a constant parameter matrix and multiplicative noise can be obtained:
[0073]
[0074] Among them are the definitions:
[0075] Using R i is m i ×m i The positive semidefinite matrix, that is, R i ≥ 0. R δ =diag(R 1,K,R L ) is also a positive semidefinite block diagonal matrix, that is, R δ =diag(R 1 ,K,R L )≥0. It can be concluded that
[0076] Furthermore, by defining the state error covariance equation and combining the actual and conservative state second-order non-central moments, the dynamic equation of the system state error can be obtained. By introducing the Lyapunov equation, the stability of the system state error covariance can be proved, and then linked with the matrix inequality, the non-negativity and stability of the virtual noise variance can be proved. This helps to verify the effectiveness of the virtual noise method.
[0077] According to Assumption 2, the actual augmented state equation is related to the actual noise variance and the uncorrelated x ai (t) and w ai (t). The actual augmented state x ai The actual second noncentral moment of (t)
[0078] Similarly, the conservative augmented state x ai The conservative second noncentral moment of (t) It can be calculated similarly.
[0079] Let M∈R m×m , and M is a semi-positive definite matrix, that is, M ≥ 0. For any C∈R m×m , we can get CMC T It is also a semi-positive matrix, namely CMC T ≥ 0. Definition and Get about ΔX ai Lyapunov equation for (t):
[0080]
[0081] get Through the semi-positive definiteness of the variance matrix, it can be concluded that and from and X ai (0)≥0, we can get In addition, it can be concluded Also available and From X ai (0) minus It can be concluded that ΔX ai (0)≥0.
[0082] Introducing the Matrix and It is defined as:
[0083]
[0084] in For multi-sensor network system, assuming The spectral radius is less than 1, that is Then, for any initial condition X ai ≥0, the solution of the generalized Lyapunov equation and X ai (t) converge to the unique positive semidefinite solution X of the following generalized steady-state Lyapunov equation ai and
[0085] Similarly, assuming The spectral radius is less than 1, that is Then, for any initial condition X aij ≥0, the solution of the generalized Lyapunov equation and X aij (t) converge to the unique positive semidefinite solution X of the following generalized steady-state Lyapunov equation aij and
[0086] The virtual process noise is introduced as follows, and the state equation can be rewritten as:
[0087]
[0088] It can be clearly seen that fi (t) is white noise with zero mean.
[0089] Will and Q fi They are defined as imaginary noise w fi The actual steady-state variance and conservative steady-state variance of (t) can be obtained as w fi (t) and w fj The actual steady-state and conservative steady-state cross covariance of (t). Introducing fictitious measurement noise, the measurement equation becomes:
[0090]
[0091] Will and They are defined as imaginary noise v fi The actual steady-state variance and conservative steady-state variance of (t) will be and S fi As w fi (t) and v fi(t) is the actual and conservative steady-state correlation matrix. Based on the above discussion, a time-invariant multi-model correlated noise system with only uncertain actual virtual noise variance is obtained. i )<1, we have This shows is a stable matrix and It is completely verifiable.
[0092] Assumption 5. Assumption is completely verifiable, and is completely correct, then
[0093] Step S4, designing a robust local steady-state Kalman predictor, which is used to predict the evolution of the system state and provide an initial state estimate.
[0094] For a conservative noise statistic Q fi ,R fi and S fi Based on the principle of minimax robust estimation, the standard Kalman filter algorithm is used to obtain the following conservative local optimal (linear minimum variance) steady-state Kalman predictor for the worst-case time-invariant augmented multi-model correlated noise system:
[0095]
[0096] Conservative local steady-state prediction error variance P ai (-1) satisfies the following steady-state Riccati equation:
[0097]
[0098] The worst-case system generates a conservative upper bound Q w , and P 0 The local measurement y i (t) is called conservative local measurement value is unknown. And the actual variance produced by the actual system and The measured value y i (t) is called the actual measurement value, which is known and obtained by the sensor.
[0099] The local steady-state forecast error is defined as The augmented noise λ fi (t),i=1,K,L definition The actual and conservative steady-state cross-covariance λ of the augmented noise can be obtained fi (t) and λ fj (t) respectively satisfy the following formula:
[0100]
[0101] Actual and conservative steady-state variance λ of the augmented noise fi (t) are defined as The actual and conservative local steady-state forecast error covariances satisfy the following Lyapunov equations:
[0102]
[0103] The actual and conservative local steady-state forecast error variance is:
[0104] Robust local steady-state Kalman filter and smoother. The filter is used to update the system estimate according to the current observation value and provide the most accurate estimation result. The smoother is used to post-process the estimate of the system state to eliminate errors and improve the estimation accuracy.
[0105] For the worst-case invariant augmented multi-model multi-sensor system, with known conservative noise statistics Q fi ,R fi and S fi , according to the actual local steady-state Kalman predictor: Practical Local Steady-State Kalman Filter (N=0) and Smoother for
[0106]
[0107] Furthermore, the robust fusion steady-state Kalman estimator is used to achieve multi-sensor data fusion. Through a reasonable weight allocation strategy, the estimator can assign different weights according to the credibility, accuracy and other factors of the observation data of each sensor, thereby optimizing the fusion result.
[0108] For the worst-case original multi-sensor network system, the conservative variance Q is known to be w , and P 0 ,Combining the matrix weighted CI and FCI fusion algorithms, the actual weighted state fusion time-varying Kalman estimator is given in a unified form:
[0109]
[0110] The constraints are:
[0111]
[0112] Where θ = CI, FCI1, FCI2 represent CI fuser with matrix weights and two FCI fusers respectively. The optimal weight coefficient is calculated by minimizing the nonlinear performance index
[0113]
[0114] This requires solving an L-dimensional nonlinear optimization problem, which can be solved using the 'fmincon' function in the MATLAB toolbox.
[0115] Calculate the weighting coefficients for the matrix weights of θ = FCI1 This can be simplified by computing the matrix trace.
[0116]
[0117] θ = matrix weights of FCI2 And the weighting coefficient This can be simplified by computing the determinant of the matrix.
[0118]
[0119] Two FCI weighting coefficients and It can be obtained by calculating the matrix trace or determinant, avoiding the solution of the L-dimensional linear optimal problem, thereby reducing the huge computational burden.
[0120] Conservative and practical fusion error variance P θ (N) and The unified framework:
[0121]
[0122] For mixed uncertain multi-sensor network systems, there is an accuracy relationship between the robust local and fusion steady-state Kalman estimators and the matrix inequality:
[0123]
[0124] Proof: The first inequality can be obtained directly, and the first two inequalities can be obtained by trace operations on the trace. and are respectively called the actual accuracy of the corresponding robust steady-state Kalman estimator, and trP θ (N),θ=CI,FCI1,FCI2 and trP i(N), 1, K, and L are called their robust accuracy, respectively. For all allowable uncertainties, the corresponding actual accuracy of each estimator is higher than or equal to its robust accuracy. The robust accuracy is the lowest actual accuracy, and the robust accuracy of all fusions is higher than the robust accuracy of each local estimator.
[0125] In step S5, the application of robust fusion Kalman filtering in multi-sensor ARMA signals with various uncertainties not only solves the ARMA signal estimation problem, especially the uncertain factors such as random parameters, colored measurement noise and uncertain noise variance in the model, but also verifies the universality of robust fusion Kalman filtering.
[0126] Consider the following multi-sensor single-channel ARMA signal with random parameters, colored measurement noise, and uncertain noise variance:
[0127] A t (q -1 )s(t)=w(t),
[0128] z i (t) = γ i (t)s(t)+e i (t),i=1,K,L,
[0129] y i (t) = ξ i (t)z i (t)+(1-ξ i (t))(1-ξ i (t-1))z i (t-1),i=1,K,L,
[0130] A t (q -1 )=1+a 1 (t)q -1 +a 2 (t)q -1 +L+a n (t)q -1 ,a k (t) = a k +β k (t),k=1,K,n.
[0131] where s(t)∈R 1 is the scalar signal to be estimated, w(t)∈R 1 is the input noise, z i (t)∈R 1 is the measurement of the ith sensor, e i (t)∈R 1 is the colored measurement noise, At (q -1 ) is q -1 polynomial, where q -1 is the unit shift operator, i.e., q - 1 s(t)=s(t-1),a k (t), k = 1, K, n is the known mean a k and the random perturbation β k (t); w(t), e i (t) and β k (t) is uncorrelated white noise with zero mean and covariance.
[0132] The purpose of this step is to design a robust local and fused steady-state Kalman estimator for the ARMA signal s(t). Note that the ARMA signal model with random parameters has a state space model equivalent to 7. So it has the following form:
[0133]
[0134] v i (t) = π i w(t)+η i (t),η i (t) = γ iz (t)w(t)+e i (t),γ iz (t) = γ i (t)-π i
[0135] y i (t) = ξ i (t)z i (t)+(1-ξ i (t))(1-ξ i (t-1))z i (t-1),i=1,K,L.
[0136] Therefore, the ARMA signal model is converted to a state space model with multiplicative noise, missing observations, packet loss and uncertain noise variance, which can be treated as special cases of the system. i =H,H ik =H k and D i =π i .This gives an important application background of the system considered.
[0137] definition and Based on the robust local signal estimator L can be obtained as a distributed fusion robust signal estimator and the actual and conservative fused steady-state estimate error variances and
[0138] Furthermore, it was demonstrated that FCI1 fusion homeostatic signaling filters The robustness of the noise is determined by selecting any three different actual noise variances. Satisfied, as shown in the following table:
[0139]
[0140] In this way, the corresponding three robust FCI1 fusion steady-state Kalman signal filters are obtained and the corresponding actual and conservative FCI1 fusion steady-state filter error variances and Figure 3 The corresponding actual filter error curves and their three times standard deviation boundaries of the three robust FCI1 fusion filters are shown, where the curve of the actual FCI1 fusion filter error is expressed in terms of implementation. The boundaries are indicated by dotted lines, and Boundaries are indicated by dashes.
[0141] A is the actual filter error curve of the FCI1 fusion signal filter and the corresponding error curve of group (1) and B is the actual filtering error curve of the FCI1 fusion signal filter and the corresponding and C is the actual filtering error curve of the FCI1 fusion signal filter and the corresponding and boundary.
[0142] According to probability theory, its actual standard deviation And its robust standard deviation from Figure 3 As can be seen in Figure 2, for each fusion error curve, more than 99% of the fusion filter error values are located at Within the borders, also located This demonstrates a robust FCI1 fusion homeostatic signaling filter The robustness and actual standard deviation of correctness.
[0143] To illustrate the multiplicative noise β h (t),h=1,2 pairs of robust fast covariance cross-fusion (FCI2) Kalman signal predictors The impact of robust accuracy is to conservatively reduce the multiplicative noise variance by a step size of 0.001. Increase from 0.003 to 0.03 and Increased from 0.002 to 0.02. Robust Accuracy for and Changes such as Figure 4 As shown in Figure 2, we can see that when the conservative multiplicative noise variance increases, This indicates that when the conservative multiplicative noise variance increases, the fast covariance intersection (FCI2) fused steady-state Kalman predictor The robust accuracy will be reduced.
[0144] In summary, compared with the prior art, the present invention has the following beneficial effects:
[0145] 1. By comprehensively considering factors such as uncertain noise variance, multiplicative noise, packet loss, continuous loss of observations and one-step random measurement delay, and designing a robust weighted fusion Kalman estimator, the robustness of the multi-sensor network system is effectively improved, enabling it to maintain stable estimation performance in the face of various uncertainties.
[0146] 2. By fusing multiple local estimators and optimizing them in combination with the accuracy relationship, the robust weighted fusion Kalman estimator can provide higher estimation accuracy than a single local estimator, thus better meeting application requirements.
[0147] 3. Compared with some complex robust fusion methods, the FCI fusion method simplifies the calculation of weighted coefficients by calculating matrices or determinants, which reduces the computational burden and makes it easier to implement in applications.
[0148] 4. By converting the ARMA signal processing problem into a state estimation problem and successfully applying the robust weighted fusion estimator, new ideas and methods are provided for this field and its application scope is expanded.
[0149] What is disclosed above is only one or more preferred embodiments of the present invention, which certainly cannot be used to limit the scope of rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of implementing the above embodiments and making equivalent changes according to the claims of the present invention still fall within the scope of the invention.
Claims
1. A robust filtering method for solving the uncertainty of one-step random observation delay, characterized in that: The following steps are involved: Step 1: Establish a linear discrete-time uncertain multi-sensor network system, where the uncertainty induced by the network is expressed using two Bernoulli random variables ξ i (t) and γ i (t) to describe; Step 2: Use the augmentation method to transform the model, calculate the second-order non-central moment and prove its accuracy relationship, convert the system in step 1 into an augmented multi-model system, and then introduce virtual noise to obtain an augmented multi-model correlated noise system; Step 3: Based on the minimax robust estimation principle, a robust local steady-state Kalman predictor is derived, and then a robust local steady-state Kalman estimator is proposed; Step 4: Based on the theory of robust local steady-state Kalman estimator, continue to advance the fusion robust estimator to obtain a robust weighted fusion estimator; Step 5: Convert the ARMA signal processing problem into a state estimation problem and verify the universality of the robust weighted fusion estimator.
2. The robust filtering method for solving the uncertainty of one-step random observation delay as claimed in claim 1, characterized in that: The expression of the linear discrete-time uncertain multi-sensor network system is as follows: v i (t)=D i w(t)+η i (t),i=1,...,L y i (t)=ξ i (t)z i (t)+(1-ξ i (t))(1-ξ i (t-1))z i (t-1) Where: t is discrete time, x(t)∈R n is the state vector at time t, is the observation value of the ith sensor, is the measurement value received by the estimator after transmission, w(t)∈R n is the process noise, is the observation noise and is linearly related to w(t), β k (t)∈R 1 , k=1,...,q is the state-dependent multiplicative noise, Φ∈R n×n ,Φ k ∈R n ×n ,Γ∈R n×r , and is a suitable constant matrix of known dimension; q is the amount of multiplicative noise, and L is the number of sensors.
3. The robust filtering method for solving the uncertainty of one-step random observation delay as claimed in claim 2, characterized in that: During the execution of step 2, the expression after conversion to the augmented multi-model system is as follows: x ai (t+1)=Φ ai (t)x ai (t)+Γ ai (t)w ai (t) y i (t)=Η ai (t)x ai (t)+D ai (t)w ai (t) where Φ ai (t), x ai (t),Γ ai (t), w ai (t), Η ai (t), D ai (t) are the augmented constant matrix, augmented state vector, and augmented noise parameter, respectively, and correspond to the parameter matrix of the original system. The augmented multi-model correlated noise system obtained after the introduction of virtual noise is a time-invariant multi-model correlated noise system with only uncertain actual virtual noise variance, and the expression is as follows: definition and They are Φ ai (t),Γ ai (t),H ai (t) and D ai (t) is the average matrix.
4. The robust filtering method for solving the uncertainty of one-step random observation delay as claimed in claim 3, characterized in that: The robust local steady-state Kalman predictor in step 3 is expressed as follows: The robust local steady-state Kalman predictor is the basis of the robust local steady-state Kalman filter and smoother, and provides the initial prediction value and prediction error variance for subsequent filtering and fusion.
5. The robust filtering method for solving the uncertainty of one-step random observation delay as claimed in claim 4, characterized in that: The constraints of the robust weighted fusion estimator are: Where θ = CI, FCI1, FCI2 represent CI fusion with matrix weights and two FCI fusions respectively; matrix weights It is a robust local steady-state Kalma estimator for each original subsystem given; the fusion strategies of CI, FCI1 and FCI2 are all based on the minimax robust estimation principle. The CI method constructs the fusion estimate by finding the intersection of the local estimation error covariances to ensure the upper bound of the fusion estimation error variance.
6. The robust filtering method for solving the uncertainty of one-step random observation delay as claimed in claim 5, characterized in that: The ARMA signal in step 5 is a multi-sensor single-channel ARMA signal with random parameters, colored measurement noise and uncertain noise variance, and the expression is as follows: A t (q -1 )s(t)=w(t), z i (t)=γ i (t)s(t)+e i (t),i=1,K,L, y i (t)=ξ i (t)z i (t)+(1-ξ i (t))(1-ξ i (t-1))z i (t-1),i=1,K,L, A t (q -1 )=1+a1(t)q -1 +a2(t)q -1 +L+a n (t)q -1 ,a k (t)=a k +β k (t),k=1,K,n. where s(t)∈R 1 is the scalar signal to be estimated, w(t)∈R 1 is the input noise, z i (t)∈R 1 is the measurement of the ith sensor, e i (t)∈R 1 is the colored measurement noise, A t (q -1 ) is q -1 polynomial, where q -1 is the unit shift operator, i.e., q -1 s(t)=s(t-1),a k (t), k = 1, K, n is the known mean a k and the random perturbation β k (t); w(t), e i (t) and β k (t) is uncorrelated white noise with zero mean and covariance.
Citation Information
Cited By
Interactive robust state estimator design method based on multiple adaptive factors
CN116186482A
