Robust estimation method for time-delay systems based on fusion state-augmented and dual polytopic filter

Abstract

The application discloses a time-delay system robust estimation method fusing state augmentation and double polytope filtering, and relates to the field of time-varying parameter system state observation. The method comprises the following steps: collecting load, vehicle speed and other data to construct an LPV polyhedron time-delay model, and eliminating time delay through double-layer state augmentation; the first layer polyhedron H∞ observation guarantees robustness in all working conditions, and the second layer fuses polyhedron Kalman filtering of plane projection and limited frequency domain KYP constraint to compress the estimation interval; decoupling obtains state and fault interval with error boundary, and the state and fault interval are fed to a fault-tolerant controller and are adaptively updated to observe constraint. The application can solve the problems of parameter perturbation drift of a traditional algorithm, failure positioning of a multi-actuator, and closed-loop control of estimation without confidence interval, and is suitable for real-time fault-tolerant observation of time-varying multi-branch equipment such as four-wheel electric vehicles and energy storage clusters.

Description

Technical Field

[0001] This invention relates to the field of robust state observation technology for time-varying parameter control systems, and more specifically, to a robust estimation method for time-delay systems that integrates state augmentation and dual multicell filtering. Background Technology

[0002] Linear variable-parameter (LPV) systems are widely used in practical engineering applications such as four-wheeled electric vehicles and energy storage battery clusters. Their state matrix varies over a wide range of time with parameters such as vehicle speed and load, and is often accompanied by input time delays and multi-actuator coupling characteristics. Traditional state observers (such as Romberg observers and standard Kalman filters) are usually designed based on steady-state models, which are difficult to adapt to conditions with drastic parameter perturbations, and are prone to state estimation drift or even divergence. To address the time delay problem, existing augmented state methods can convert time-delayed systems into time-delay-free forms, but the dimension increases sharply after augmentation, and conventional filtering algorithms face the dilemma of heavy computational burden and high conservatism in interval estimation.

[0003] In multi-actuator parallel systems, when a fault occurs in one branch, the fault residuals are coupled together, making it difficult for traditional residual thresholding methods to accurately locate the fault path. On-site maintenance often requires checking each branch one by one, which is inefficient. More importantly, most existing estimation algorithms only output state point estimates and lack quantifiable error boundaries, failing to provide safety constraints for chassis fault-tolerant controllers or battery management systems (BMS), thus exposing closed-loop control to uncertainty risks. Although some studies have attempted to introduce multi-cell filtering into state estimation, a single multi-cell layer either pursues global robustness at the expense of accuracy or pursues fine filtering at the expense of adaptability to a wide range of parameters, making it difficult to achieve both simultaneously.

[0004] Therefore, there is an urgent need for a joint estimation method that can maintain robust stability under all operating condition parameter perturbations, while also possessing multi-fault isolation capabilities and confidence boundaries. Summary of the Invention

[0005] In view of this, this invention proposes a robust estimation method for time-delay systems that integrates state augmentation and dual multi-cell filtering to address the problems of estimation drift under large parameter perturbations, difficulty in locating fault channels under coupled residuals of multiple parallel actuators, and lack of quantifiable error boundaries in the output of traditional LPV time-delay systems, which prevent direct use in closed-loop fault-tolerant control. The method includes: collecting multi-dimensional time-series data of system voltage, current, load, and vehicle speed; constructing an LPV linear variable-parameter time-delay model; extracting parameter extrema to generate a polyhedral convex hull covering all operating conditions; performing a two-level state augmentation transformation, with the first level fusing system state and fault variables, and the second level stacking full-delay history vectors to convert the time-varying time-delay model into a standard time-delay-free LPV high-dimensional augmented system; and solving the H∞ linear matrix inequality for all vertices of the convex hull to construct the second level... A global multi-cell observation layer is constructed, outputting a set of robust coarse-weighted estimation polyhedra across the entire domain. The cosine correlation coefficient between the residuals and the fault feature vectors of each actuator is calculated, and multi-branch fault source isolation is achieved based on the principle of planar projection. A second multi-cell Kalman fine-filter layer is constructed by introducing generalized KYP finite frequency domain constraints to compress the boundaries of the multi-cell intervals. The augmented multi-cells are split and optimized using a preset linear decoupling matrix, outputting the system state and fault-independent robust intervals respectively. An interval width threshold is set to complete the multi-cell boundary stability verification, filtering out abnormal expansion intervals caused by instantaneous impacts. The standardized encapsulated state, fault interval center point, and upper and lower limits form controllable estimates. The joint estimation message with error boundaries is sent to the vehicle chassis / energy storage BMS fault-tolerant controller. The controller tracks deviation feedback and adaptively adjusts the observer LMI constraint weights.

[0006] Furthermore, the construction of the LPV linear variable parameter time delay model and the generation of a polyhedral convexity includes: combining the system state matrix A(θ) and the time delay matrix... Input matrix Set it as a time-varying parameter matrix; extract the upper and lower extreme values ​​of load and vehicle speed parameters, construct the convex hull of a four-vertex polyhedron, and establish a time-delay state equation for each convex vertex.

[0007] Furthermore, the execution of the two-level state augmentation transformation includes: a first-order augmentation vector. Fusion of state and fault dimensions; generation of first-order vectors by stacking all time delays. Eliminate the time delay coupling term in the LPV model.

[0008] Furthermore, the step of constructing the first global multi-cell observation layer by solving the H∞ linear matrix inequality for all vertices of the convex hull includes: independently configuring H∞ interference suppression constraints and Lyapunov stability conditions for each set of parameter vertices of the convex hull; and merging the output states of each vertex after wrapping them with a multi-cell body to obtain a coarse estimation set for the entire domain.

[0009] Furthermore, the calculation of the cosine correlation coefficient between the residual and the fault feature vector of each actuator to isolate the fault source in multiple branches includes: solving for the cosine value of the angle between the residual vector and the preset fault feature vector of each branch. The branch with the highest correlation coefficient is selected as the fault occurrence channel.

[0010] Furthermore, the second multicell Kalman fine filter layer is constructed by introducing generalized KYP finite frequency domain constraints, including: dividing the low-frequency fault band and the high-frequency noise band, and configuring two sets of independent LMI constraints respectively; solving the time-varying Kalman gain based on the residual after fault isolation, and compressing the width of the multicell interval.

[0011] Furthermore, the process of splitting and optimizing the augmented multicell using a preset linear decoupling matrix includes: linearly decomposing the target augmented multicell vector using a fixed transformation matrix; and independently decoupling to generate system state multicell intervals and fault multicell intervals, each carrying upper and lower error boundaries.

[0012] Furthermore, the setting of the interval width threshold to complete the multi-cell boundary stability verification includes: setting a maximum allowable interval width threshold; performing smoothing correction when the multi-cell interval of a single frame exceeds the threshold, and eliminating the distorted interval caused by abrupt disturbances.

[0013] Furthermore, the standardized encapsulation state, the fault interval center point, and the upper and lower limit values ​​form a controllable estimate, including: extracting the center point of each multicell as the optimal estimate and synchronously binding the upper and lower error boundaries; and generating a unified estimation message in a formatted manner according to the controller communication protocol.

[0014] Furthermore, the receiver controller adaptively adjusts the observer LMI constraint weights based on the tracking deviation feedback, including: acquiring the system tracking deviation value output by the controller; increasing the H∞ interference suppression weights when the tracking deviation increases, and resolving the observation constraint matrix.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By extracting the upper and lower extreme values ​​of parameters such as load and vehicle speed, an LPV polyhedral convex hull is constructed to cover the entire operating parameter space. H∞ linear matrix inequality constraints are independently configured for each vertex to fundamentally suppress the estimation drift caused by parameter perturbation and ensure that the observer remains stable and convergent under extreme conditions such as rapid vehicle acceleration, heavy load climbing, or large current charging and discharging of battery clusters. 2. By utilizing the principle of planar projection, the cosine correlation coefficient between the estimated residual and the preset fault feature vector of each branch is calculated. The fault channel is determined by the maximum value of the correlation coefficient. This can effectively distinguish the coupled residual signals of multiple parallel branches (such as hub motors and battery clusters), realize the rapid and accurate location of the fault source, and greatly reduce the difficulty and time cost of on-site troubleshooting. 3. After a two-layer multi-cell filtering process (the first layer uses H∞ coarse estimation to ensure robustness, and the second layer uses generalized KYP finite-frequency domain Kalman filtering to compress the interval), the augmented state is split using a linear decoupling matrix, independently outputting the upper and lower bounds and center points of the system state and fault variables. This estimator with error boundaries can be directly used for amplitude limiting protection and safety constraints in fault-tolerant controllers, solving the problem that traditional point estimation cannot achieve closed-loop control and improving system operational safety. 4. By introducing the controller tracking deviation as a feedback signal, the H∞ disturbance suppression weight is automatically increased and the observation constraint matrix is ​​resolved when the deviation increases. This enables the observer to dynamically match the changing operating conditions of the equipment, achieving an adaptive balance between robustness and estimation accuracy, and avoiding performance degradation of fixed-weight design when operating conditions change. 5. Set the interval width threshold and perform frame-by-frame verification. Perform smooth correction on abnormally expanded multi-cell intervals to effectively filter out distorted estimates caused by sudden load changes, electromagnetic pulses and other instantaneous impacts, enhance the anti-interference capability of the engineering site, and ensure the continuous and reliable output of estimation messages. Attached Figure Description

[0016] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 The flowchart illustrates a robust estimation method for time-delay systems based on fused state augmentation and dual multicell filtering, as provided in an embodiment of the present invention. Detailed Implementation

[0017] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0018] See Figure 1 The robust estimation method for time-delay systems based on fusion state augmentation and dual multicell filtering provided by this invention can be applied to industrial control systems with large-scale time-varying parameters and communication delays, such as four-wheel independent drive electric vehicles, energy storage battery clusters, and multi-motor synchronous transmissions. The implementation of this invention will be described in detail below with reference to a specific application scenario (taking a four-wheel independent drive electric vehicle as an example).

[0019] Example 1: LPV time delay model construction and polyhedron convex hull generation In this embodiment, for the lateral stability control system of a four-wheel independent drive electric vehicle, multi-dimensional time-series data during system operation is first collected. Specifically, the phase current and terminal voltage of the four hub motors are collected in real time via the vehicle's CAN bus at a sampling frequency of 100Hz; the longitudinal vehicle speed and yaw rate are collected via the inertial measurement unit (IMU); and the sprung load change signal is collected via the suspension height sensor. All collected data are denoised by a first-order low-pass filter (cutoff frequency 50Hz) and then stored in a circular data buffer, caching the data of the most recent 500 sampling moments for parameter identification.

[0020] When constructing the LPV linear variable parameter time delay model, the system state vector is set as , , , These represent the sideslip angle, yaw rate, and tire sideslip angle, respectively; control input u This provides torque commands for the four hub motors. System state matrix. Time delay matrix and input matrix All parameters are set as time-varying parameter matrices that change with vehicle speed and load. The lower limit value of the vehicle speed parameter is extracted. Upper limit Lower limit of load parameters Upper limit The system consists of a two-dimensional parameter space comprised of two time-varying parameters: vehicle speed and load. Discrete time-delay state equations are established at the four vertices of this parameter space, resulting in a convex hull of a four-vertex polyhedron covering all operating conditions. The discrete time-delay model at each vertex is represented as follows: Where the subscripts i=1,2,3,4 correspond to the four convex hull vertices, and h is the maximum time delay step of the system (in this embodiment, h=3, corresponding to a communication time delay of 150ms). and These are process noise and measurement noise, respectively, and their amplitude boundaries are obtained through offline identification.

[0021] Example 2: Two-level state augmentation transformation To eliminate the coupling effect of time delay terms on the observer design, this invention performs a two-stage state augmentation transformation. The first-stage augmentation fuses the system state with actuator fault variables, defining an augmentation vector. ,in This represents the multiplicative fault coefficient of the four hub motors (normal value is 1, deviates from 1 during a fault).

[0022] The second-level augmented stack of first-order augmented vectors at all time delays defines the ultra-high-dimensional augmented state vector: After the above two-stage augmentation transformation, the original time-delay coupling terms are completely eliminated, resulting in a standard time-delay-free LPV high-dimensional augmented system: in, It is a high-dimensional matrix derived from the original system matrix through augmented transformation. This represents the process noise after augmentation. During the transformation, the stacking method of the time-delayed history vectors ensures that the dynamic characteristics of the original system are fully preserved in the augmented system. This augmentation transformation can be completed online within each sampling period, and the computational complexity is linearly related to h, meeting real-time requirements.

[0023] Example 3: Construction of the first global multicell H∞ observation layer For the four convex hull vertices generated in Example 1, H∞ disturbance suppression constraints and Lyapunov stability conditions are configured independently for each. For the i-th vertex, the following set of linear matrix inequalities (LMIs) is solved: in, >0 indicates a symmetric positive definite Lyapunov matrix. Let γ be the gain matrix of the observer to be determined, and let γ be the H∞ interference suppression index (γ=0.8 in this embodiment). Let be the perturbation input matrix. Solve the LMI above for all four vertices to obtain four sets of vertex observer gain matrices. .

[0024] In real-time operation, based on the time-varying parameters θ(k) (vehicle speed and load) at the current moment, a linear interpolation method is used to fuse the outputs of each vertex observer to construct a global coarse estimate of the multicell set: Among them, interpolation coefficients The value is determined based on the normalized distance between the current parameters and the parameters of the i-th vertex. This coarse estimate of the cellular set provides robust state estimation across the entire operating range, but its interval width is relatively large (typically about ±15% of the true value), requiring further compression by subsequent fine filtering layers.

[0025] Example 4: Multi-branch fault source isolation based on planar projection After obtaining the first coarse estimate, the system residual vector is calculated. To pinpoint which of the four hub motors is malfunctioning, this invention employs the planar projection correlation coefficient method.

[0026] Fault feature vectors of four actuators are pre-built offline. Each feature vector is obtained by injecting a typical fault (a 10% step deviation in amplitude) into the actuator and performing system identification. The residual vector r(k) and each fault feature vector are calculated in real time. The cosine correlation coefficient of the included angle: The branch corresponding to the highest correlation coefficient is selected as the fault occurrence channel, i.e.: At the same time, the judgment threshold η is set to 0.7. When the value is less than η, it is determined that no fault has occurred, thus avoiding false alarms. Experiments show that, under the condition that the signal-to-noise ratio is greater than 15dB, the fault location accuracy of this method is higher than 96%, and the fault detection delay is less than 2 sampling periods (40ms).

[0027] Example 5: Construction of the Second Multi-Cell Kalman Fine Filter Layer After isolating the fault source, a second layer of fine filtering is constructed using generalized KYP (Kalman-Yakubovich-Popov) finite frequency domain constraints. In this embodiment, frequency bands are divided according to the frequency domain characteristics of the fault and noise: typical actuator faults are characterized by low-frequency components (0.1-5Hz), while measurement noise is mainly characterized by high-frequency components (>50Hz).

[0028] Based on the generalized KYP lemma, two sets of independent LMI constraints are configured respectively: Low-frequency band (0.1-5Hz) constraint: To enhance sensitivity to low-frequency faults, a low-frequency weight matrix is ​​introduced into LMI. ; High-frequency (>50Hz) constraints: To suppress high-frequency measurement noise, a high-frequency weighting matrix is ​​introduced into LMI. .

[0029] Solving the two sets of LMIs above yields the fine filter gain. The time update and measurement update steps of multi-cell Kalman filtering are then performed: Updated in time: Measurement Update: Where Q(k) and R(k) are the adaptively updated noise covariance matrices. After fine filtering, the width of the multicell interval is compressed by about 65% compared with the first coarse estimation, the state estimation error is reduced from ±15% of the coarse estimation to within ±5%, and the fault amplitude estimation error is less than 3%.

[0030] Example 6: Multicellular Decoupling and Independent Region Output The augmented multicell output of the second-layer fine filter It simultaneously includes coupled information on system state and fault variables. This invention uses a pre-defined linear decoupling matrix T to achieve this decomposition: Where T is a fixed transformation matrix designed based on the inverse process of the augmented transformation, and its structure is as follows: That is, directly from ultra-high-dimensional augmented vectors Extract the current system state x(k) and fault variable f(k), discarding historical time delay terms. After decoupling, the system state is divided into multiple cell intervals. and faulty multicell interval Each carries upper and lower error boundaries: the state interval is represented as The fault range is represented as The interval widths reflect the confidence levels of the state estimation and fault estimation, respectively.

[0031] Example 7: Multicell Boundary Stability Verification and Adaptive Update To prevent abnormal expansion of the multicell interval due to sudden sensor impacts, communication packet loss, or other abnormal conditions, this invention sets a maximum allowable interval width threshold. For the state interval, the threshold is set to ±30% of the nominal value; for the fault interval (fault coefficient), the threshold is set to [0.5, 1.5].

[0032] In each sampling period, the width of each dimension's interval is first checked to see if it exceeds the threshold. If a cell's interval in a frame exceeds the threshold, a smoothing correction operation is performed. The correction result is obtained by taking the median of the current frame and the two previous frames, thus eliminating the distortion range caused by abrupt disturbances. This smooth correction strategy effectively avoids the impact of a single abnormal frame on subsequent control decisions. Experiments show that the error correction rate is less than 0.5% under random impulse noise interference.

[0033] Then, standardized encapsulation is performed: the center point of each multicell is extracted as the optimal estimate. and Synchronously bind upper and lower error boundaries and A unified estimation message is generated in accordance with the vehicle controller communication protocol (CAN2.0B extended frame). The message content includes: timestamp, optimal state value and upper and lower limits, optimal fault value and upper and lower limits, and interval width verification flag.

[0034] The joint estimation message with error boundaries is sent to the chassis domain fault-tolerant controller via the CAN bus. Upon receiving the estimation message, the controller performs conventional control law calculations using the center value and executes control quantity limiting protection using the error boundaries to prevent the actuators from exceeding the safety margin. Simultaneously, the controller collects the actual tracking deviation. (Including yaw rate tracking deviation and centroid sideslip angle tracking deviation), which are fed back to the observer via CAN.

[0035] After receiving tracking deviation feedback, the observer performs adaptive adjustment: when When the yaw rate deviation is greater than 0.1 rad / s, the tracking deviation is determined to have increased. The H∞ interference suppression weight γ is automatically increased (from 0.8 to 0.5), and the first-layer observation LMI constraint matrix is ​​resolved, updating the observer gain online. This adaptive mechanism ensures that the observer dynamically matches the optimal robust performance under different operating conditions.

[0036] The above scenarios are merely preferred embodiments of the present invention and are 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.

[0037] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program goods. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program goods embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0038] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program goods according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0039] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0040] These computer program instructions can also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0041] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A robust estimation method for time-delay systems that integrates state augmentation and dual multicell filtering, characterized in that, include: Collect multi-dimensional time-series data of system voltage, current, load, and vehicle speed, construct an LPV linear variable parameter time delay model, extract parameter extrema to generate a polyhedral convex hull covering all operating conditions; Perform a two-level state augmentation transformation: the first level fuses system state and fault variables, and the second level stacks full-delay history vectors to convert the time-varying time-delay model into a standard time-delay-free LPV high-dimensional augmented system. For each vertex of the convex hull, solve the H∞ linear matrix inequality to construct the first global multicell observation layer and output the global robust coarse estimation polyhedron set; calculate the cosine correlation coefficient between the residual and the fault feature vector of each actuator, and complete the multi-branch fault source isolation based on the plane projection principle; A second-layer multicell Kalman fine filter layer is constructed by introducing generalized KYP finite frequency domain constraints to compress the boundaries of the multicell interval; the augmented multicell is split and optimized by a preset linear decoupling matrix, and the system state and fault independent robust intervals are output respectively. Set an interval width threshold to complete the stability verification of the multicell boundary and filter out abnormal expansion intervals caused by instantaneous shocks; Standardized encapsulation states, fault interval center points, and upper and lower limits form controllable estimates; The joint estimation message with error boundaries is sent to the vehicle chassis / energy storage BMS fault-tolerant controller; the controller tracks the deviation feedback and adaptively adjusts the observer LMI constraint weights.

2. The robust estimation method for time-delay systems based on fused state augmentation and dual multicell filtering according to claim 1, characterized in that, The construction of the LPV linear variable-parameter time-delay model and the generation of a polyhedral convexity include: The system state matrix A(θ) and the time delay matrix Input matrix Set as a time-varying parameter matrix; Extract the upper and lower extreme values ​​of load and vehicle speed parameters, construct the convex hull of a four-vertex polyhedron, and establish a time-delay state equation for each convex vertex.

3. The robust estimation method for time-delay systems based on fused state augmentation and dual multicell filtering according to claim 1, characterized in that, The execution of the two-level state augmentation transformation includes: First-order augmented vector Fusion status and fault dimensions; Generation of first-order vectors at all time delays Eliminate the time delay coupling term in the LPV model.

4. The robust estimation method for time-delay systems based on fused state augmentation and dual multicell filtering according to claim 1, characterized in that, The process of constructing the first global multicell observation layer by solving the H∞ linear matrix inequalities for all vertices of the convex hull includes: For each set of parameters of the convex hull, independently configure H∞ disturbance suppression constraints and Lyapunov stability conditions for each vertex. The output states of each vertex are wrapped with multi-cell bodies and then merged to obtain a coarse global estimation set.

5. The robust estimation method for time-delay systems based on fused state augmentation and dual multicell filtering according to claim 1, characterized in that, The calculation of the residual and the cosine correlation coefficient of each actuator fault feature vector completes the isolation of multi-branch fault sources, including: Solve for the cosine of the angle between the residual vector and the preset fault feature vector of each branch. ; The branch with the highest correlation coefficient is selected as the fault occurrence channel.

6. The robust estimation method for time-delay systems based on fused state augmentation and dual multicell filtering according to claim 1, characterized in that, The construction of the second multicell Kalman fine filter layer by introducing generalized KYP finite frequency domain constraints includes: Divide the frequency band into low-frequency fault band and high-frequency noise band, and configure two sets of independent LMI constraints for each band. The time-varying Kalman gain is solved based on the residual after fault isolation, thus compressing the width of the multicell interval.

7. The robust estimation method for time-delay systems based on fused state augmentation and dual multicell filtering according to claim 1, characterized in that, The augmented multicell, optimized by splitting using a preset linear decoupling matrix, includes: The target augmented multicell vector is linearly decomposed using a fixed transformation matrix; Independent decoupling generates system state cell intervals and fault cell intervals, each carrying upper and lower error boundaries.

8. The robust estimation method for time-delay systems based on fused state augmentation and dual multicell filtering according to claim 1, characterized in that, The set interval width threshold is used to complete the multicell boundary stability verification, including: Preset maximum allowable width threshold for the interval; When the multi-cell interval of a single frame exceeds the threshold, a smoothing correction is performed to eliminate the distorted interval caused by abrupt disturbances.

9. The robust estimation method for time-delay systems based on fused state augmentation and dual multicell filtering according to claim 1, characterized in that, The standardized packaging state, fault interval center point, and upper and lower limit values ​​form the controllable estimates, including: Extract the center point of each multicell as the optimal estimate and simultaneously bind the upper and lower error boundaries; Generate a unified estimation message in accordance with the controller communication protocol.

10. The robust estimation method for time-delay systems based on fused state augmentation and dual multicell filtering according to claim 1, characterized in that, The receiver controller tracks the bias feedback to adaptively adjust the observer LMI constraint weights, including: The controller outputs the system tracking deviation value; When the tracking deviation increases, the H∞ interference suppression weight is increased, and the observation constraint matrix is ​​resolved.

Images