Robust fusion estimation for systems with colored noises, observation lags and missing data
By introducing Bernoulli white noise sequences and state augmentation techniques into a multi-sensor network system, a robust fusion estimation method is constructed, which solves the problem of unstable state estimation caused by colored noise and observation time delay, and achieves high-precision and robust state estimation.
Patent Information
- Application Number
- CN202611174066.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-04
- Publication Date
- 2026-11-10
AI Technical Summary
Existing technologies struggle to effectively handle complex interference caused by colored noise and observation delays in multi-sensor network systems, leading to unstable state estimation and increased errors. In particular, traditional methods cannot guarantee the stability and optimality of multi-sensor fusion results under conditions of communication link uncertainty and data loss.
By introducing uncorrelated Bernoulli white noise sequences and state augmentation techniques, colored noise, observation delay, and packet loss are uniformly incorporated into the system model, constructing a robust fusion estimation method. The upper bound of the error is established using matrix augmentation and Lyapunov steady-state equations, and global multi-sensor data fusion is performed in combination with matrix weighted optimality criteria to improve estimation accuracy and robustness.
It achieves high-precision state estimation of multi-sensor network systems in complex communication environments, dynamically adapts to complex communication scenarios, has superior fault tolerance and anti-interference capabilities, and ensures global optimal fusion accuracy.
Smart Images

Figure CN123067112A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of information engineering technology and automatic control theory, specifically to a state estimation method for multi-sensor networked systems, and particularly to a robust fusion Kalman estimation method for systems with colored noise, observation delays, and complex packet loss. Background Technology
[0002] With the widespread application of multi-sensor network systems in intelligent control, navigation and positioning, and industrial monitoring, achieving high-precision state estimation in complex environments has gradually become a research hotspot. In practical systems, information collected by sensors typically needs to be transmitted through networks. Due to the uncertainty of communication links and energy limitations, measurement data often inevitably suffers from observation delays and data loss. Furthermore, noise in practical systems often exhibits time correlation, displaying colored noise characteristics rather than the ideal white noise process.
[0003] The shortcomings of existing technologies lie in the fact that time delays cause the current observation information to become out of sync with the system state, and colored noise introduces time correlations, violating the independent and identically distributed assumptions of classical Kalman filtering, resulting in more complex stochastic characteristics in the system model. Filters designed for single uncertainties are difficult to apply directly, and the stability and optimality of multi-sensor fusion results cannot be guaranteed. How to integrate colored noise and one-step observation time delays into the system modeling framework, construct a multi-source covariance consistency processing system through state dimension expansion, and thus achieve robustness assurance, has become a key issue with significant engineering value and theoretical challenges. Summary of the Invention
[0004] The purpose of this invention is to provide a robust fusion estimation method for systems with colored noise, observation delay, and complex packet loss. This method comprehensively considers colored noise, multiplicative noise, random delay of observations, and packet loss under different conditions, thereby improving the reliability and estimation accuracy of the robust weighted steady-state Kalman estimator.
[0005] This invention provides a robust fusion estimation method. Its detailed solution and gains include: a system architecture that comprehensively covers multiple network constraints, and the introduction of uncorrelated Bernoulli white noise sequences to provide a high-precision mathematical probability characterization of packet reception and delay states. A state augmentation technique is introduced, using matrix augmentation to absorb colored noise and multi-step delay terms into a unified, time-delay-free extended-dimensional state equation. Based on this, virtual process noise and measurement noise are extracted to construct a conservative Lyapunov steady-state equation for the worst-case system. Subsequently, matrix trace inequalities and non-negative definite matrix factorization techniques are used to establish the steady-state upper bound of the local estimator. Finally, a matrix-weighted optimality criterion based on covariance is applied to perform global multi-sensor data fusion, significantly reducing the fusion error tracking region and ensuring estimation performance.
[0006] The technical solution of this invention to solve the above-mentioned technical problems is as follows: A robust fusion estimation method for a system with colored noise, observation delay, and packet loss, comprising the following steps:
[0007] Step 1: Construct a basic multi-sensor networked system model with colored noise, observation delay, and complex packet loss.
[0008] Step 2: Based on the original system model in Step 1, Bernoulli variables are introduced to model time delays and packet loss, and the state augmentation method is used to transform it into an augmented stochastic parameter multi-model system with multiplicative noise.
[0009] Step 3: Based on the augmented multi-model system and augmented noise conservative covariance output in Step 2, and according to the minimax robust estimation principle, derive the complete formulas for the robust local steady-state Kalman predictor, filter, and smoother in sequence.
[0010] Step 4: Based on the local robust estimators of all sensors obtained in Step 3, a robust matrix-weighted fusion steady-state Kalman estimator is constructed using the matrix-weighted optimal fusion rule. The upper bound of the steady-state error covariance is derived to complete the theoretical proof of robustness.
[0011] Step 5: Based on the complete robust fusion Kalman estimation algorithm obtained in Step 4, build a multi-sensor simulation system with colored noise, random time delay, and complex packet loss, and verify the correctness and effectiveness of the algorithm through simulation experiments.
[0012] In some possible implementations, step 1 specifically involves:
[0013] Step 1.1: Construct the system's five-layer equations, retaining all model formulas completely:
[0014] ① Equation of state: ;
[0015] ② Sensor with time delay measurement equation: ;
[0016] ③ Measurement of white noise coupling equation: ;
[0017] ④ Colored noise recursive expression: ;
[0018] ⑤ Colored noise operator expansion: ;
[0019] ⑥ Time-varying coefficient perturbation model: ;
[0020] ⑦ Sensor receiver output equation with double packet loss and delay buffer:
[0021] ;
[0022] in, It is discrete time. Let be the state vector of the system. It is the first Observations from each sensor It is the measurement value received by the estimator after being transmitted over the network. It's process noise. For fixed-length delay, It is observation noise. It's white noise. , and Bernoulli white noise, It is state-dependent multiplicative noise. It is a known constant matrix with appropriate dimensions. It is the amount of multiplicative noise. It is the number of sensors. It is colored noise. Unit lag operator;
[0023] Step 1.2, define three classes of 0-1 Bernoulli random variables to characterize packet loss, transmission failure, and time delay triggering, given prior probabilities:
[0024] , ;
[0025] , ;
[0026] , ;
[0027] constraint ;set up , and They are independent of each other and unrelated to other noise;
[0028] Step 1.3, Set the initial state Independent of all noise, given an initial mean Uncertain actual initial variance ,in, yes The mean, It is an unknown, uncertain actual variance; given a conservative upper bound matrix. and And satisfy:
[0029] ;
[0030] in, and It is zero-mean and uncorrelated white noise. as well as They are white noise and The actual variance is uncertain.
[0031] Step 1 integrates four types of network interference—multiplicative disturbance, colored noise, one-step delay, and double-layer packet loss—through a unified set of equations. It quantifies the random characteristics of transmission using Bernoulli variables, distinguishes between the actual variance of noise and the conservative upper bound, and constructs a complete mathematical model. This model can adapt to complex communication scenarios such as navigation and industrial monitoring in one go, solving the problem of poor applicability of traditional single disturbance models. It can also provide a unified calculation benchmark for subsequent filtering and fusion, lock the worst boundary of disturbance in advance, avoid estimation divergence from the source, and intuitively relate to engineering adjustable parameters such as communication reception and delay.
[0032] In some possible implementations, step 2 specifically involves:
[0033] Step 2.1, Variable substitution simplifies the cached observation equation: Definition ,Will Rewritten as:
[0034] ;
[0035] Derivation , and Product constraints and first- and second-order expectations ;
[0036] Step 2.2, Colored Noise State Dimension Expansion: Establishing Colored Noise Equivalent state-space model:
[0037] ;
[0038] ;
[0039] Give , , The complete block matrix expression; where, yes OK A matrix of columns, for OK A matrix of columns, for OK A matrix of columns;
[0040] Step 2.3: Construct the single-sensor augmented state vector Augmented noise vector Define the block matrix , Establish the time-varying augmented state equation and observation equation:
[0041] ;
[0042] ;
[0043] Step 2.4, Zero-mean perturbation decomposition: Define bias noise:
[0044]
[0045] Cross-coupling disturbance Derive the variance of each bias noise:
[0046] ;
[0047] Step 2.5, Separation of the mean of the time-varying matrix: Split into mean constant matrix Combined with the superposition of all multiplicative noise disturbance terms, the complete multiplicative noise expansion of the augmented state and observation equations is output, giving... , , , , And the expressions for all other perturbation block matrices;
[0048] Step 2.6, Construct augmented composite noise Derivation , Complete expression; calculate the actual cross-covariance and conservative cross-covariance of the augmented noise respectively:
[0049] ;
[0050] Give , , Summation formula; proof that it satisfies for all allowable uncertainties. ;
[0051] Step 2.7, Construction of the global augmented state vector: Calculate the actual second-order noncentral moments Conservative second-order noncentral moments ,prove By quantifying the uncertainty perturbation behavior of the system through second-order non-central moments, augmented noise is introduced to improve the robustness of the filter.
[0052] Step 2 simplifies the packet loss observation equation based on the original model in Step 1, expands the time delay and colored noise into augmented states, and transforms the complex coupled system into a standard linear model containing only multiplicative noise. Simultaneously, it calculates the actual and conservative second moments and noise covariance to define the upper bound of the error. This not only eliminates the theoretical defects of the classical Kalman model in handling time delay and colored noise, but also unifies various disturbances to simplify subsequent derivations. Furthermore, it enables independent parallel operation of sensors, and the standardized augmented framework can be extended to adapt to other estimation algorithms, thus building a robust foundation at the model level.
[0053] In some possible implementations, step 3 specifically involves:
[0054] Step 3.1, Recursive formula for the robust locally steady-state Kalman predictor:
[0055] ① Status one-step forecast: ;
[0056] ②Forecast residuals: ;
[0057] ③ Definitions of forecast transition matrix, forecast gain, and residual covariance:
[0058] ;
[0059] in, For the conservative steady-state error covariance of local forecasts;
[0060] Step 3.2, the unified expression for the robust locally time-varying Kalman filter and smoother is:
[0061] ;
[0062] ;
[0063] .
[0064] Step 3, based on the augmented system and conservative noise parameters from Step 2, solves for the complete set of recursive gains for steady-state prediction, filtering, and smoothing with the worst-case perturbation as a constraint. One algorithm covers three types of time-series requirements: prediction, real-time monitoring, and backtracking analysis. Relying on the conservative upper bound constraint to prevent estimation errors from diverging, each sensor operates independently and has single-node fault tolerance capability. The steady-state gain reduces the consumption of embedded real-time computing power, and the residual correction mechanism can quickly offset the instantaneous deviations caused by the network and noise.
[0065] In some possible implementations, step 4 specifically involves:
[0066] Step 4.1, Unified expression for multi-sensor fusion estimator:
[0067] ;
[0068] Weight normalization constraint: ;
[0069] Step 4.2, the formula for calculating the optimal matrix weights is: ;in, This is the block covariance matrix formed by the conservative error covariances of each local estimator. It is a column vector consisting entirely of 1s;
[0070] Step 4.3, performance comparison of three weighting methods: matrix weighting, diagonal weighting, and scalar weighting; computational cost: matrix weighting > diagonal weighting > scalar weighting; estimation error trace: scalar weighting > diagonal weighting > matrix weighting, proving that matrix weighting is globally optimal;
[0071] Step 4.4, Proof of robustness:
[0072] ① For the local estimator: Based on the augmented noise conservative covariance and Lyapunov steady-state equation, prove the local actual error covariance. ;
[0073] ② For the fusion estimator: Based on the order-preserving property of matrix weighted linear transformation and the non-negative definite matrix factorization method, the actual fusion error covariance is derived. ;
[0074] ③ Under a unified framework, we present a unified form of the actual / conservative error covariance for the three types of fusion estimators: forecasting, filtering, and smoothing, and complete the proof of all robustness theorems.
[0075] Step 4 uses the local estimates from each source as input, derives the optimal matrix weighting formula using linear minimum variance, and builds a unified global fusion prediction / filtering / smoothing architecture. Combining matrix factorization and Lyapunov equations, it is proved that the fusion error has a conservative upper bound, and the fusion estimation accuracy is better than any single sensor. The weights automatically reduce the proportion of high-fault sensors, inherit all local robustness characteristics, and the modular architecture facilitates code encapsulation, completely solving the suboptimal divergence problem of multi-source fusion and achieving global optimal estimation.
[0076] In some possible implementations, step 5 specifically involves:
[0077] Step 5.1, Build the simulation numerical system: Set Multiplicative noise term Road sensor, given discrete state matrix, observation matrix, and noise coupling matrix. Colored noise operator coefficients , Given Bernoulli probability Actual variance of all noise The variance is compared with the conservative upper bound.
[0078] Step 5.2, run the algorithm: synchronously execute Step 3 (local robust prediction / filtering / smoothing) and Step 4 (matrix weighted fusion algorithm), and record the true state. Local estimates, global fusion estimates, real-time estimation errors, and conservative error covariance traces;
[0079] Step 5.3, Visualization and analysis of multi-dimensional simulation results:
[0080] ① Compare the actual state with the tracking curves of local prediction, filtering, one-step smoothing, and two-step smoothing to verify the tracking fit of the single-sensor local estimator;
[0081] ② Actual error curve of fusion filtering Actual error boundary, Conservative error boundary comparison verifies that the error is always constrained to the conservative upper bound corridor, with no drift or divergence.
[0082] ③ The conservative error variance trace of the fusion forecaster varies with the data packet reception rate Delay probability By varying the three-dimensional surface, it is verified that the higher the receiver rate, the higher the fusion estimation accuracy, and the algorithm adapts to changes in communication quality.
[0083] Step 5: Based on the complete algorithm, build a multi-sensor simulation platform containing colored noise, time delay, and packet loss. It traverses multiple sets of communication and noise parameters, outputs tracking curves, error boundaries, and a three-dimensional surface of variance. This intuitively verifies the tracking accuracy under complex interference and that the error is always limited to a conservative boundary. The three-dimensional surface can quantify and guide communication deployment strategies. The standardized simulation process provides a reference for hardware porting, completing the verification loop from theory to engineering implementation.
[0084] The beneficial effects of this invention are: 1. Dynamically adapting to complex communication-constrained scenarios; 2. Exhibiting superior fault-tolerant and anti-interference tracking performance under mixed uncertain interference; 3. Ensuring an extremely high lower limit of global optimization fusion accuracy. Attached Figure Description
[0085] Figure 1 is a schematic flowchart of the robust fusion estimation method of the present invention for dealing with two-step random time delay and uncertainty of lost observations.
[0086] Figure 2 is a schematic diagram of the formation and robustness proof process of the local and fusion Kalamn estimator in this invention.
[0087] Figure 3 is a simulation comparison of the actual state and the tracking effect of a local Kalman predictor, filter, one-step smoothing and two-step smoothing in a system with colored noise, random observation time delay and data packet loss in a specific embodiment of the present invention.
[0088] Figure 4 shows the dynamic curve of the actual distributed state fusion filtering error in a specific embodiment of the present invention, and the corresponding... and Comparison chart of boundary convergence characteristics;
[0089] Figure 5 illustrates the conservative error variance trace of the global fusion forecaster in a specific embodiment of the present invention. With the increase in sensor node data acceptance rate ( ) and transmission latency ( The relationship between three-dimensional curved surfaces with varying parameters such as ) Detailed Implementation
[0090] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.
[0091] Please see Figure 1 This invention provides a robust fusion estimation method to address two-step random time delays and uncertainties in lost observations, comprising the following steps:
[0092] S1: The original system is a multi-sensor networked system model with colored noise, observation delay, and complex packet loss.
[0093] S2: Introduce Bernoulli random variables to model delay and packet loss, and apply state augmentation methods to transform the original system into an augmented stochastic parameter multi-model system with multiplicative noise;
[0094] S3: Based on the minimax robust estimation principle, a locally optimal steady-state Kalman predictor is obtained, and then a robust locally steady-state Kalman filter and a smoother are designed respectively.
[0095] S4: Under a unified framework, a matrix-weighted optimal fusion algorithm is applied to propose a robust matrix-weighted fusion steady-state Kalman estimator, and the upper bound condition of the steady-state error covariance is derived to prove its robustness.
[0096] S5: A simulation example using a multi-sensor network system with colored noise, random observation delays, and data loss demonstrates this.
[0097] The following provides further explanation in conjunction with the specific implementation steps:
[0098] In step S1, a reasonable system model is designed for the multi-sensor networked system that meets the conditions, and appropriate assumptions and definitions are made about the variables therein.
[0099] Specifically, the qualified multi-sensor networked system model is as follows:
[0100]
[0101]
[0102]
[0103]
[0104]
[0105]
[0106] .
[0107] in, It is discrete time. Let be the state vector of the system. It is the first Observations from each sensor It is the measurement value received by the estimator after being transmitted over the network. It's process noise. For fixed-length delay, It is observation noise. It's white noise. , and Bernoulli white noise, It is state-dependent multiplicative noise. It is a known constant matrix with appropriate dimensions. It is the amount of multiplicative noise. It is the number of sensors. It is colored noise. Unit lag operator.
[0108] And it satisfies the following assumptions:
[0109] It is an uncorrelated Bernoulli random variable taking values of 1 or 0, and has a known probability:
[0110] ,satisfy At the same time, assume , and They are all uncorrelated with other random signals. and It is zero-mean and uncorrelated white noise. as well as They are white noise and The uncertain actual variance. That is, satisfying:
[0111] ; This refers to the Kronecker delta function. .
[0112] initial state and and Irrelevant ,in yes The mean of the initial state. Furthermore, the variance of the initial state satisfies... ,in It is the unknown, uncertain actual variance.
[0113] and They are and The known conservative upper bound.
[0114] Based on the above assumptions, the problem addressed in this invention is to design robust local and fusion time-varying Kalman estimators (predictors, filters, smoothers) for assumed states in uncertain multi-sensor networked systems. This makes the variance of their actual steady-state estimation error constant for all permissible uncertainties. It has a corresponding minimum upper bound ,Right now:
[0115]
[0116] in, Let represent the i-th local robust estimator and the matrix-weighted fusion estimator, respectively.
[0117] The purpose of step S2 is to transform the system model into a multi-model, multi-sensor system containing only uncertain noise variance. First, the packet loss equation is transformed as follows:
[0118] make ,but:
[0119] .
[0120] Depend on Definition and and The properties of the Bernoulli distribution can be obtained as follows: Therefore, we can obtain , , , .
[0121] It is known that colored noise satisfies the following spatial state equation:
[0122]
[0123]
[0124] in:
[0125]
[0126] In the formula yes OK A matrix of columns, for OK A matrix of columns, for OK A matrix of columns, The The list is The other columns are 0.
[0127] The following multi-model systems exist:
[0128]
[0129]
[0130] in:
[0131]
[0132]
[0133] Define noise:
[0134]
[0135] Furthermore, the statistical characteristics of this noise are as follows:
[0136] .
[0137] Through the above transformations, we can obtain the following equivalent multi-model system with constant parameter matrices and multiplicative noise.
[0138] ;
[0139] ;
[0140] in:
[0141]
[0142]
[0143]
[0144] ;
[0145] in They are respectively The mean matrix.
[0146] The augmentation method used in step S2 is state augmentation. By adding system delay measurements or other uncertainties as additional state variables to the state vector, the original Kalman filtering problem is transformed into an augmented state-space problem. Augmentation methods can fuse data from multiple sensors or information sources, improving the accuracy and reliability of measurements. It is suitable for dynamically changing systems, enabling real-time updates of state estimates to adapt to system changes. This method preserves the state-space representation and simplifies the extension of delay handling methods to other methods, such as moving horizon estimation (MHE), nonlinear dynamic data coordination (NDDR), unscented Kalman filters (UKF), and particle filters.
[0147] Furthermore, augmented noise and augmented state vectors are introduced. With augmented noise The definition is as follows:
[0148]
[0149] in:
[0150] ;
[0151] ;
[0152] According to and The augmented noise is obtained by calculating the actual and conservative cross-covariance. and The actual and conservative cross-covariances are as follows:
[0153] ;
[0154] in:
[0155] ;
[0156] ;
[0157] .
[0158] Will and Use respectively and Replacement will yield the corresponding result. .
[0159] The actual and conservative steady-state variances of augmented noise are defined as follows: For all permissible uncertain actual variances and ,have .
[0160] To demonstrate the robustness of the robust fusion steady-state Kalman estimator, an augmented state vector is introduced. It is defined as It is easy to see its conservatism and actual second-order non-centrality. and They are respectively
[0161] ;
[0162] and .
[0163] The second-order noncentral moment is an eigenvalue describing the characteristics of a random variable. It provides information about the shape of the random variable's distribution, particularly its distribution characteristics beyond the mean and variance. Calculating the actual and conservative second-order noncentral moments helps us to understand the statistical properties of random variables more comprehensively. The actual second-order noncentral moment of the augmented state vector can be used to analyze the stability of a system. By considering the noncentral moment, we can more accurately assess the system's behavior in the face of uncertainty and external disturbances, thereby improving the system's robustness.
[0164] For all permissible uncertain actual variances and ,have .
[0165] The formation and proof process of the weighted fusion time-varying Kalamn estimator during the execution of steps S3 and S4 is as follows: Figure 2 As shown.
[0166] The expression for the locally optimal time-varying Kalman forecast is as follows:
[0167] ;
[0168] ;
[0169] .
[0170] The robust locally time-varying Kalman filter and smoother are as follows:
[0171] ;
[0172] ;
[0173] .
[0174] Using the matrix-weighted optimal fusion rule, the robust matrix-weighted fusion steady-state Kalman estimator for a multi-sensor system is given as follows:
[0175] ;
[0176] .
[0177] The weighted fusion rules in step S4 are as follows:
[0178] First, the optimal weighted fusion rule is a method used in multi-sensor data fusion. Its core idea is to obtain more accurate fusion results by assigning different weights to data from different sensors. It improves the overall prediction accuracy by combining the prediction results of multiple models, and the weights of each model can be determined during training through cross-validation. In multi-sensor systems, the allocation of weights has a significant impact on the fusion effect. Proper weight allocation results in good fusion; improper allocation does not significantly improve the system's accuracy and reliability.
[0179] Under the linear minimum variance principle, weighted fusion has three weighting methods: 1. matrix weighting; 2. scalar weighting; 3. diagonal weighting. The advantages and disadvantages of these three weighting methods are mainly analyzed from the perspectives of computational complexity and error. Computational complexity: matrix weighting > diagonal weighting > scalar weighting; Error: scalar weighting > diagonal weighting > matrix weighting.
[0180] In matrix-weighted fusion, the weights are matrices, meaning that more factors are considered when calculating the weight matrix. The fused state values are also composed of combinations of multiple state values, thus this method yields estimates with relatively small errors. This method is suitable for situations with complex state-space models and correlations between states. While the computational cost is relatively high due to matrix operations, it provides relatively accurate fusion results.
[0181] The optimal matrix weighting is:
[0182] .
[0183] Through inference and analysis, and by utilizing the properties of matrix traces, the core performance criterion relationship is derived: That is, the global estimation system after matrix weighted fusion always has a lower bound of the mean square trace of error when dealing with the worst perturbation that is lower than or equal to the performance of the best individual in the system, thus solving the suboptimal divergence problem of multi-source fusion.
[0184] Step S5: Instance Verification and Steady-State Characteristic Analysis In the simulation implementation phase, a specific... And clearly define the network probability settings (such as...) The data scenario is described. Because the designed algorithm dynamically adapts to the collapse of the observation dimension caused by packet loss, the generated tracking curve exhibits a very high temporal fit to the true state value. Even when faced with severe colored noise fluctuations, the estimation bias does not drift but rather converges within a certain range. Within the safety grace corridor, the variance trace surface visually illustrates the feedback mechanism by which improvements in network communication quality lead to a rapid decrease in steady-state trace indicators.
[0185] To better verify the proposed technology, the inventors provide relevant simulation results of the system state, as follows:
[0186] The following is a mixed uncertainty system with uncertain variance and colored noise:
[0187] ;
[0188] ;
[0189] ;
[0190] ;
[0191] ;
[0192]
[0193] .
[0194] In the simulation experiment, the value is:
[0195]
[0196] The following simulation results were obtained.
[0197] Easy to obtain Figure 1 This is a schematic flowchart of the steps of a robust fusion estimation method for dealing with two-step random time delays and uncertainty of lost observations according to the present invention. Figure 2 This is a schematic diagram illustrating the formation and robustness proof process of the local and fusion Kalamn estimator in this invention. Figure 3 The diagram shows the tracking results of the local Kalman predictor, filter, one-step smoothing, and two-step smoothing, with the black line representing the system. The true value is shown in the image, and the red line represents the corresponding tracking curve. This demonstrates that the model performs well under these parameters. Figure 4 The paper presents three sets of actual filtering error curves for the corresponding robust fusion filter and their three-standard-deviation boundaries, with the actual fusion filtering error curves represented by solid lines. The boundary is indicated by a red dotted dashed line, while The boundary is indicated by a short blue line. Figure 5 To fuse the error variance of the forecaster and and The relationship diagram shows that when The error variance of the time fusion predictor varies with the receiver rate. The increase in the fusion rate leads to differences in the fusion process. Clearly, for higher receive rates, the proposed fusion filter exhibits better performance.
[0198] In summary, the following conclusions are drawn: A robust fused Kalman estimator problem was studied for multi-sensor network systems with colored noise, one-step observation delay, and complex packet loss. By introducing Bernoulli random variables to model the delay and packet loss, and utilizing state dimension expansion to transform the original system into an augmented multi-model system with multiplicative noise, a local and fused steady-state estimator was derived based on a matrix weighting strategy. Furthermore, it was proven that the estimator error variance has a convergent steady-state solution and possesses a minimum upper bound property. System modeling and simulation verified the correctness and effectiveness of the designed robust fused Kalman estimator under conditions of colored noise, random delay, and data loss.
[0199] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A robust fusion estimation method for systems with colored noise, observation delay, and packet loss, characterized in that, The steps include the following: Step 1: Construct a basic multi-sensor networked system model with colored noise, observation delay, and complex packet loss. Step 2: Based on the original system model in Step 1, Bernoulli variables are introduced to model time delays and packet loss, and the state augmentation method is used to transform it into an augmented stochastic parameter multi-model system with multiplicative noise. Step 3: Based on the augmented multi-model system and augmented noise conservative covariance output in Step 2, and according to the minimax robust estimation principle, derive the complete formulas for the robust local steady-state Kalman predictor, filter, and smoother in sequence. Step 4: Based on the local robust estimators of all sensors obtained in Step 3, a robust matrix-weighted fusion steady-state Kalman estimator is constructed using the matrix-weighted optimal fusion rule. The upper bound of the steady-state error covariance is derived to complete the theoretical proof of robustness. Step 5: Based on the complete robust fusion Kalman estimation algorithm obtained in Step 4, build a multi-sensor simulation system with colored noise, random time delay, and complex packet loss, and verify the correctness and effectiveness of the algorithm through simulation experiments.
2. The robust fusion estimation method for a system with colored noise, observation delay, and packet loss as described in claim 1, characterized in that, Step 1 is as follows: Step 1.1: Construct the system's five-layer equations, retaining all model formulas completely: ① Equation of state: ; ② Sensor with time delay measurement equation: ; ③ Measurement of white noise coupling equation: ; ④ Colored noise recursive expression: ; ⑤ Colored noise operator expansion: ; ⑥ Time-varying coefficient perturbation model: ; ⑦ Sensor receiver output equation with double packet loss and delay buffer: ; in, It is discrete time. Let be the state vector of the system. It is the first Observations from each sensor It is the measurement value received by the estimator after being transmitted over the network. It's process noise. For fixed-length delay, It is observation noise. It's white noise. , and Bernoulli white noise, It is state-dependent multiplicative noise. It is a known constant matrix with appropriate dimensions. It is the amount of multiplicative noise. It is the number of sensors. It is colored noise. Unit lag operator; Step 1.2, define three classes of 0-1 Bernoulli random variables to characterize packet loss, transmission failure, and time delay triggering, given prior probabilities: , ; , ; , ; constraint ;set up , and They are independent of each other and unrelated to other noise; Step 1.3, Set the initial state Independent of all noise, given an initial mean Uncertain actual initial variance ,in, yes The mean, It is an unknown, uncertain actual variance; given a conservative upper bound matrix. and And satisfy: ; in, and It is zero-mean and uncorrelated white noise. as well as They are white noise and The actual variance is uncertain.
3. The robust fusion estimation method for a system with colored noise, observation delay, and packet loss as described in claim 1, characterized in that, Step 2 is as follows: Step 2.1, Variable substitution simplifies the cached observation equation: Definition ,Will Rewritten as: ; Derivation , and Product constraints and first- and second-order expectations ; Step 2.2, Colored Noise State Dimension Expansion: Establishing Colored Noise Equivalent state-space model: ; ; Give , , The complete block matrix expression; where, yes OK A matrix of columns, for OK A matrix of columns, for OK A matrix of columns; Step 2.3: Construct the single-sensor augmented state vector Augmented noise vector Define the block matrix , Establish the time-varying augmented state equation and observation equation: ; ; Step 2.4, Zero-mean perturbation decomposition: Define bias noise: Cross-coupling disturbance Derive the variance of each bias noise: ; Step 2.5, Time-varying matrix mean separation: [The text abruptly ends here, likely due to an incomplete sentence or a formatting error.] Split into mean constant matrix Combined with the superposition of all multiplicative noise disturbance terms, the complete multiplicative noise expansion of the augmented state and observation equations is output, giving... , , , , And the expressions for all other perturbation block matrices; Step 2.6, Construct augmented composite noise Derivation , Complete expression; calculate the actual cross-covariance and conservative cross-covariance of the augmented noise respectively: ; Give , , Summation formula; prove that it satisfies the following for all permissible uncertainties. ; Step 2.7, Construction of the global augmented state vector: Calculate the actual second-order noncentral moments Conservative second-order noncentral moments ,prove By quantifying the uncertainty perturbation behavior of the system through second-order non-central moments, augmented noise is introduced to improve the robustness of filtering.
4. The robust fusion estimation method for a system with colored noise, observation delay, and packet loss as described in claim 1, characterized in that, Step 3 specifically involves: Step 3.1, Recursive formula for the robust locally steady-state Kalman predictor: ① Status one-step forecast: ; ②Forecast residuals: ; ③ Definitions of forecast transition matrix, forecast gain, and residual covariance: ; in, For the conservative steady-state error covariance of local forecasts; Step 3.2, the unified expression for the robust locally time-varying Kalman filter and smoother is: ; ; 。 5. The robust fusion estimation method for a system with colored noise, observation delay, and packet loss as described in claim 1, characterized in that, Step 4 is as follows: Step 4.1, Unified expression for multi-sensor fusion estimator: ; Weight normalization constraint: ; Step 4.2, the formula for calculating the optimal matrix weights is: ;in, This is the block covariance matrix formed by the conservative error covariances of each local estimator. It is a column vector consisting entirely of 1s; Step 4.3, performance comparison of three weighting methods: matrix weighting, diagonal weighting, and scalar weighting; computational cost: matrix weighting > diagonal weighting > scalar weighting; estimation error trace: scalar weighting > diagonal weighting > matrix weighting, proving that matrix weighting is globally optimal; Step 4.4, Proof of robustness: ① For the local estimator: Based on the augmented noise conservative covariance and Lyapunov steady-state equation, prove the local actual error covariance. ; ② For the fusion estimator: Based on the order-preserving property of matrix weighted linear transformation and the non-negative definite matrix factorization method, the actual fusion error covariance is derived. ; ③ Under a unified framework, we present a unified form of the actual / conservative error covariance for the three types of fusion estimators: forecasting, filtering, and smoothing, and complete the proof of all robustness theorems.
6. The robust fusion estimation method for a system with colored noise, observation delay, and packet loss as described in claim 1, characterized in that, Step 5 specifically involves: Step 5.1, Build the simulation numerical system: Set Multiplicative noise term Road sensor, given discrete state matrix, observation matrix, and noise coupling matrix. Colored noise operator coefficients , ; Given Bernoulli probability Actual variance of all noise The variance is compared with the conservative upper bound. Step 5.2, run the algorithm: synchronously execute Step 3 (local robust prediction / filtering / smoothing) and Step 4 (matrix weighted fusion algorithm), and record the true state. Local estimates, global fusion estimates, real-time estimation errors, and conservative error covariance traces; Step 5.3, Visualization and analysis of multi-dimensional simulation results: ① Compare the actual state with the tracking curves of local prediction, filtering, one-step smoothing, and two-step smoothing to verify the tracking fit of the single-sensor local estimator; ② Actual error curve of fusion filtering Actual error boundary, Conservative error boundary comparison verifies that the error is always constrained to the conservative upper bound corridor, with no drift or divergence. ③ The conservative error variance trace of the fusion forecaster varies with the data packet reception rate Delay probability By varying the three-dimensional surface, it is verified that the higher the receiver rate, the higher the fusion estimation accuracy, and the algorithm adapts to changes in communication quality.
