Recursive parameter identification of constrained systems: anapplication to electrically stimulated muscle

In the application of real-time identification methods for diagnosis or adaptive control of biomedical systems, there is often known model information that is ignored. Constraints on the allowable values of parameters, which may be based on physical considerations, are often neglected because the information does "fit" easily into commonly used parameter-identification algorithms. In this paper a method of incorporating constraints on model parameters is developed. This method is applicable to most recursive parameter-identification algorithms. It enforces linear equality constraints on identified parameters. The use of this method for the real-time identification of autoregressive moving-average-type time series models, subject to parameter constraints, is described in detail. These constraints may be time varying. At each time step, the parameter estimate obtained by a recursive least squares estimator is orthogonally projected onto the constraint surface. This simple idea, when appropriately executed, enhances the output prediction accuracy of estimated parameters. Using constraint information in this way is important when we do not wish to destroy a "natural" parameterization of the model (by an initial projection to incorporate equality constraints), or when we cannot use a single initial model simplification (because the constraints are time varying or involve inputs and outputs). Because it improves output prediction at future times, this method is advantageous for use in predictive adaptive controllers. The use of this algorithm is demonstrated in the identification of electrically stimulated quadriceps muscles in paraplegic human subjects, using percutaneous intramuscular electrodes. The nonlinear steady-state force versus pulsewidth recruitment characteristic of the electrode-muscle system is identified simultaneously with the input-output muscle response dynamics, using a Hammerstein-type model. Knowledge of the recruitment curve's shape is translated into constraints on the identified parameters. This information improves the experimental predictive quality of the identified model.     

Muscle-joint models incorporating activation dynamics,moment-angle, and moment-velocity properties

Muscle input/output models incorporating activation dynamics, moment-angle, and moment-velocity factors are commonly used to predict the moment produced by muscle during nonisometric contractions: the three factors are generally assumed to be independent. The authors examined the ability of models with independent factors, as well as models with coupled factors, to fit input/output data measured during simultaneous modulation of the fraction of muscle stimulated (recruitment) and joint angle inputs. The models were evaluated in stimulated cat soleus muscles producing ankle extension moment, with regard to their potential applications in neuroprostheses with either fixed parameters or parameter adaptation. Both uncoupled and coupled models predicted the output moment well for random angle perturbation sizes ranging from 10° to 30°. For the uncoupled model, the best parameter values depended on the range of perturbations and the mean angle. Introducing coupling between activation and velocity in the model reduced this parameter sensitivity; one set of model parameter values fit the data for all perturbation sizes and also fit the data under isometric or constant stimulation conditions. Thus, the coupled model would be the most appropriate for applications requiring fixed parameter values. In contrast, with continuous parameter adaptation, errors due to changing test conditions decreased more quickly for the uncoupled model, suggesting that it would perform well in adaptive control of neuroprostheses     

Model requirements for simulation of low-voltage MOSFET in automotive applications

This paper focuses on the modeling of low-voltage automotive power electronic circuits to obtain accurate system simulation, including estimation of losses. The aim is to compare several metal-oxide semiconductor field-effect transistor (MOSFET) models to find out which can be used for low-voltage, high-current automotive converter simulations. As these models are intended for system simulation, only analytical models are addressed as they may be implemented into any circuit simulator. The different modes of operation of the switches are described (commutation, synchronous rectification, avalanche...), and several models of the power MOSFET transistor, allowing for simulation in these modes, are presented. Special care is given to the parameter extraction methods and to the interconnection model of the commutation cell. The four test circuits used to identify the low-voltage power MOSFET model parameters are presented. Comparison between simulations and measurements obtained with a calorimeter are then detailed. This measurement method is accurate and offers a simple way to prove the quality of simulation results. It is shown that the parameter identification is of major concern to achieve high accuracy, as classical Spice models can give good results, providing the model parameters are correctly set.     

Thermal parameter estimation using recursive identification

A novel method that converts a semiconductor transient thermal impedance curve (TTIC) into an equivalent thermal RC network model is presented. Thermal resistance (R) and thermal capacitance (C) parameters of the model are identified using manufacturer's data and offline recursive least square techniques. Relevant estimation theory concepts and the formulation of an appropriate model for the identification process are given. Model synthesis is illustrated using an isolated base power transistor module. The application of time decoupled theory for high order thermal models is outlined. Simulation of junction temperature responses using model and manufacturer TTICs are compared. Estimated parameter validity is further confirmed by parameter calculation obtained from module physical dimensions     

电动汽车锂离子电池模型参数辨识和荷电状态估算

针对电动汽车锂离子电池荷电状态在线估算准确率低、实时性差等问题,文中建立一种精确在线估算荷电状态的有效方法,采用MAFF-RLS和EKF对荷电状态进行估算。建立锂离子电池的等效电路模型,将MAFF-RLS应用在电池等效电路模型的参数辨识上,可以有效在线辨识模型参数。在模型参数辨识的基础上,将辨识出的模型参数作为荷电状态估算的输入,采用EKF估算动力电池实时荷电状态。经过实验仿真发现,采用MAFF-RLS和EKF联合估算荷电状态能够提高估算精确度,估算误差仅在2%以内。     

Using principal-components regression to stabilize EMG-muscle forceparameter estimates of torso muscles

Models for estimating muscle force from surface electromyographic (EMG) recordings require parameter estimates with low intertrial variability. The inclusion of multiple muscles in multivariate statistical models can lead to multicollinearity, especially when there are significant correlations between synergist muscles. One result of multicollinearity is that parameter estimates are very sensitive to changes in the independent variables. This study compared the parameter variability of multiple regression and principal-components regression techniques when applied to a six muscle EMG analysis of the lumbar region of the torso. Nine subjects participated, Twenty-three percent of the traditional multiple-regression parameters had incorrect signs, but none of the principal-components regression parameters did. The principal components regression technique also produced parameter estimates having an order of magnitude smaller parameter variability. It was concluded that principal-components regression is an effective method of mitigating the effect of multicollinearity in torso EMG models     

基于回归算法的证件物品识别器参数辨识模型

证件物品管理识别器参数辨识存在局部最优现象,噪声干扰下辨识精度下降,提出基于回归算法的证件物品管理识别器参数辨识模型.将证件物品管理识别参数输出误差平方和,代入粒子群算法适应度函数,通过粒子群优化算法实时更新粒子个体最优值以及全局最优值,初步辨识证件物品管理识别器参数,并将所获取结果作为支持向量回归算法迭代初始值,利用...     

Joint Estimation of States and Parameters for an Input Nonlinear State-Space System with Colored Noise Using the Filtering Technique

This paper concerns the state and parameter estimation problem for an input nonlinear state-space system with colored noise. By using the data filtering and the over-parameterization technique, we transform the original nonlinear state-space system into two identification models with filtered states: one containing the system parameters and the other containing the noise model’s parameters. A combined state and parameter estimation algorithm is developed for identifying the state-space system. The key is that the estimation of system parameters uses the estimated states, and the estimation of states uses the preceding parameter estimates. A simulation example is provided to show that the proposed algorithm can work well.     

Using evoked EMG as a synthetic force sensor of isometricelectrically stimulated muscle

A method for the estimation of the force generated by electrically stimulated muscle during isometric contraction is developed here. It is based upon measurements of the evoked electromyogram (EMG) [EEMG] signal. Muscle stimulation is provided to the quadriceps muscle of a paralyzed human subject using percutaneous intramuscular electrodes, and EEMG signals are collected using surface electrodes. Through the use of novel signal acquisition and processing techniques, as well, as a mathematical model that reflects both the excitation and activation phenomena involved in isometric muscle force generation, accurate prediction of stimulated muscle forces is obtained for large time horizons. This approach yields synthetic muscle force estimates for both unfatigued and fatigued states of the stimulated muscle. In addition, a method is developed that accomplishes automatic recalibration of the model to account for day-to-day changes in pickup electrode mounting as well as other factors contributing to EEMG gain variations. It is demonstrated that the use of the measured EEMG as the input to a predictive model of muscle torque generation is superior to the use of the electrical stimulation signal as the model input. This is because the measured EEMG signal captures all of the neural excitation, whereas stimulation-to-torque models only reflect that portion of the neural excitation that results directly from stimulation. The time-varying properties of the excitation process cannot be captured by existing stimulation-to-torque models, but they are tracked by the EEMG-to-torque models that are developed here. This work represents a promising approach to the real-time estimation of stimulated muscle force in functional neuromuscular stimulation applications     

Identification of induction motor parameters from transient statorcurrent measurements

This paper describes three methods for estimating the lumped model parameters of an induction motor using startup transient data. A three-phase balanced induction motor is assumed. Measurements of the stator currents and voltages are required for the identification procedure, but no measurements from the motor shaft are needed. The first method presented applies simple models with limited temporal domains of validity and obtains parameter estimates by extrapolating the model error bias to zero. This method does not minimize any specific error criterion and is presented as a means of finding a good initial guess for a conventional iterative maximum-likelihood or least-squares estimator. The second method presented minimizes equation errors in the induction motor model in the least-square sense using a Levenburg-Marquardt iteration. The third identification method is a continuation of the Levenburg-Marquardt method, motivated by observed properties of some pathological loss functions. The third method minimizes errors in the observations in the least-squared sense and is, therefore, a maximum-likelihood estimator under appropriate conditions of normality. The performance of the identification schemes is demonstrated with both simulated and measured data, and parameters obtained using the methods are compared with parameters obtained from standard tests     

Recursive Extended Least Squares Parameter Estimation for Wiener Nonlinear Systems with Moving Average Noises

Many control algorithms are based on the mathematical models of dynamic systems. System identification is used to determine the structures and parameters of dynamic systems. Some identification algorithms (e.g., the least squares algorithm) can be applied to estimate the parameters of linear regressive systems or linear-parameter systems with white noise disturbances. This paper derives two recursive extended least squares parameter estimation algorithms for Wiener nonlinear systems with moving average noises based on over-parameterization models. The simulation results indicate that the proposed algorithms are effective.     

Method for photovoltaic parameter extraction according to a modified double‐diode model

The variables presented in the current–voltage equation of a photovoltaic (PV) device are usually called PV parameters. There are several different methods for PV parameter extraction from measured data according to different models. However, many of these methods provide results that do not represent I–V curves of thin films devices correctly. This can occur because either the applied model or the PV parameter extraction methods are not suitable. It is also possible that the extracted parameters provide a good mathematical representation of the curves but without physical meaning (e.g. negative series resistance). This work presents a method for PV parameter extraction based on a modified double‐diode model. In this model, the ideality factor related to the recombination of the charge carriers in the space‐charge region is assumed as a variable. This method has been tested for different I–V curves of different PV module technologies providing very good results and parameters with physical meaning in all the cases. Copyright © 2012 John Wiley & Sons, Ltd.     

非最小相位ARMA模型的一种自适应辨识算法

本文提出了一种加性有色高斯噪声中因果非最小相位ARMA模型的自适应辨识算法。模型输入假定为非高斯独立同分布随机过程。算法只利用了观测信号的高阶统计量。在每次迭代中,先估计AR参数,再估计MA参数,但不用计算残差序列。在参数递推中采用了随机梯度法。仿真实验证实了本文算法的有效性。     

Pulse Pile-up Correction by Particle Swarm Optimization with Double-layer Parameter Identification Model in X-ray Spectroscopy

In X-ray spectrum analysis, the pulse pile-up is a long-standing issue which deteriorates the energy resolution and count rates of the radiation detection systems. In this study, a novel pulse pile-up identification method based on particle swarm optimization and double-layer parameter identification model (PSO-DLPIM) is proposed. Different Gaussian pile-up waveforms are realized by exponential pulse through Sallen-Key (S-K) low-pass filtering. Then, the proposed model recognizes the parameters of each sub-Gaussian pulse. Especially, it can be used to modelling the pulse indirectly without a certain model parameter and overcomes the model mismatch troubles. Finally, computer simulations and experimental tests are carried out and the results show that this method has higher accuracy for the recognition of pile-up pulses. The example shows that the minimum distance between pulses that can be identified by this method is 0.05 μs. And when the pulse generation time is known and the environmental noise is low, the relative error of the amplitude of pulse pile-up recognition is as low as 0.15%. Therefore, this method can greatly improve the resolution of the X-ray spectrum.

     

Wavelet-Based SAR Image Despeckling and Information Extraction, Using Particle Filter

This paper proposes a new-wavelet-based synthetic aperture radar (SAR) image despeckling algorithm using the sequential Monte Carlo method. A model-based Bayesian approach is proposed. This paper presents two methods for SAR image despeckling. The first method, called WGGPF, models a prior with Generalized Gaussian (GG) probability density function (pdf) and the second method, called WGMPF, models prior with a Generalized Gaussian Markov random field (GGMRF). The likelihood pdf is modeled using a Gaussian pdf. The GGMRF model is used because it enables texture parameter estimation. The prior is modeled using GG pdf, when texture parameters are not needed. A particle filter is used for drawing particles from the prior for different shape parameters of GG pdf. When the GGMRF prior is used, the particles are drawn from prior in order to estimate noise-free wavelet coefficients and for those coefficients the texture parameter is changed in order to obtain the best textural parameters. The texture parameters are changed for a predefined set of shape parameters of GGMRF. The particles with the highest weights represents the final noise-free estimate with corresponding textural parameters. The despeckling algorithms are compared with the state-of-the-art methods using synthetic and real SAR data. The experimental results show that the proposed despeckling algorithms efficiently remove noise and proposed methods are comparable with the state-of-the-art methods regarding objective measurements. The proposed WGMPF preserves textures of the real, high-resolution SAR images well.     

Nonlinear parametric identification of magnetic bearings

This paper proposes a new electromagnetic force model and its parameter identification method. As a case study, the parameters of the proposed model for an experimental electromagnetic bearing system are obtained using extended Kalman filter (EKF). The experimental setup includes a symmetric rigid rotor which is disturbed by the electromagnet of a magnetic bearing. Experimental results show that the system response to harmonic excitation includes super-harmonic terms which are not shown by the well-known conventional electromagnetic force model. This shortcoming necessitates an investigation to propose a more realistic electromagnetic force model. Based on the observations of the system response, a novel parametric model is presented in the form of a nonlinear Mathieu–Duffing equation with unknown coefficients. Then in the operating frequency range, a random input is synthesized and applied to the experimental system as a persistent excitation and the response of the system is recorded. In order to estimate the states and parameters of the model, the EKF method has been applied to the recorded input–output data. To validate the identification results the outputs of estimated and experimental models are compared in time and frequency domains. The results show a notable improvement in modeling of magnetic force. The proposed model and the method for identifying its parameters are applicable for all magnetic fields.     

Extraction of Passive Device Model Parameters Using Genetic Algorithms

The extraction of model parameters for embedded passive components is crucial for designing and characterizing the performance of multichip module (MCM) substrates. In this paper, a method for optimizing the extraction of these parameters using genetic algorithms is presented. The results of this method are compared with optimization using the Levenberg‐Marquardt (LM) algorithm used in the HSPICE circuit modeling tool. A set of integrated resistor structures are fabricated, and their scattering parameters are measured for a range of frequencies from 45 MHz to 5 GHz. Optimal equivalent circuit models for these structures are derived from the s‐parameter measurements using each algorithm. Predicted s‐parameters for the optimized equivalent circuit are then obtained from HSPICE. The difference between the measured and predicted s‐parameters in the frequency range of interest is used as a measure of the accuracy of the two optimization algorithms. It is determined that the LM method is extremely dependent upon the initial starting point of the parameter search and is thus prone to become trapped in local minima. This drawback is alleviated and the accuracy of the parameter values obtained is improved using genetic algorithms.     

Parameter and State Estimation Algorithm for a State Space Model with a One-unit State Delay

This paper derives a state estimation based parameter identification algorithm for state space systems with a one-unit state delay. We derive the identification model of an observability canonical state space system with a one-unit state delay. The key is to replace the unknown states in the parameter estimation algorithm with their state estimates and to identify the parameters of the state space models. Finally, two illustrative examples are given to show the effectiveness of the proposed algorithm.     

A Two-Stage Identification Scheme for the Determination of the Parameters of a Model of Left Heart and Systemic Circulation

An identification scheme is developed for the determination of several parameters of a modified "Windkessel" model of the systemic arterial system for an individual patient undergoing cardiac catheterization. The scheme utilizes a modification of the Prony method [10], [11] as a "starter method" to determine good nominal values for the model parameters being varied. These values then serve as input to a well-known iterative nonlinear least-squares identification method (Marquardt method [14]) which then converges rapidly to frmal values of the parameters. Solution of the model equations with these parameter values yields the best fit in a least-squares sense of model-generated and observed aortic and brachial artery pressures. This two stage or sequential Prony-Marquardt technique represents an extension of our previous work associated with the analysis of multiexponential decay curves [18], and is applied here to the identification of parameters associated with the humam arterial system. When coupled with a method of determining the contractile mechanics of the left ventricle (eg., the ventricular elastance concept [l]-[5]), this identification scheme permits a functional characterization of the hemodynamic properties of the left ventricle and its systemic load, for an individual subject.     

Optimal identification of regression-type processes underadaptively controlled observations

The paper presents a new approach for the design and analysis of adaptive systems for the optimal identification of physical processes and signals described by linear regression-type equations. Contrary to the traditional methods, a compound model of the observed process is proposed. This model describes an unobservable process that is subject to identification and the observing device (sensor) separately. The introduced adaptive model of the sensor with bounded linear range of its characteristic is more general and adequate than the commonly used ones. It is shown that optimal adaptive control of the sensor parameters and its fit to the statistics of the identified process significantly improve the accuracy of the parameter estimates and increase their convergence rate. Results of the theoretical part of the paper are illustrated by a simple analytic example and confirmed via simulation