NARMA Model Research Articles

Learning reservoir dynamics with temporal self-modulation

Reservoir computing (RC) can efficiently process time-series data by mapping the input signal into a high-dimensional space via randomly connected recurrent neural networks (RNNs), which are referred to as a reservoir. The high-dimensional representation of time-series data in the reservoir simplifies subsequent learning tasks. Although this simple architecture allows fast learning and facile physical implementation, the learning performance is inferior to that of other state-of-the-art RNN models. In this study, to improve the learning ability of RC, we propose self-modulated RC (SM-RC) that extends RC by adding a self-modulation mechanism. SM-RC can perform attention tasks where input information is retained or discarded depending on the input signal. We find that a chaotic state can emerge as a result of learning in SM-RC. Furthermore, we demonstrate that SM-RC outperforms RC in NARMA and Lorenz model tasks. Because the SM-RC architecture only requires two additional gates, it is physically implementable as RC, thereby providing a direction for realizing edge artificial intelligence.

Bort dinamik sistemlərin identifikasiyası və nəzarətində neyron şəbəkələrindən istifadə İ.M. İsmayılov, T.V. Binnətəliyeva

The article presents the principles of creating a predictive control strategy based on a neural network model of aviation systems. The use of neural networks for identifying and controlling dynamic systems is substantiated, including the use of model predictive control for predicting the state (parameters) of an aircraft as a dynamic system, more suitable among neural network architectures. For model predictive control, a model of an object is used to predict the future behavior of the object, and an optimization algorithm is used to select a control input that optimizes the future operation of the object. To reflect the dynamics of the system in real time, the issues of training a neural network were considered, for which a standard model, the NARMA model, was chosen as the structure of the model. Keywords: flight safety, flight dynamics, aviation intelligent control systems, predictive control, neural networks, static methods, regression analysis.

The performance of classification and forecasting Dong Nai River water quality for sustainable water resources management using neural network techniques

This study has classified and predicted the water quality for the Dong Nai river at different times. The data mining techniques were used to screen out unreliable data. The neural network model FFNN in the form of architecture I(8)-HL(9)-O(1) has been successfully developed to classify the water quality of the Dong Nai river during the period from 2012 to 2019. For this model, the RMSE values for the training, validation, and test sets are equal to 0.0392, 0.0411, and 0.0423, respectively. Also, the MAPE values for those data sets are 1.241, 1.444, and 1.511, respectively. The models ARIMA(0,1,3)(1,1,1), NAR, NAR-MA, LSTM, and LSTM-MA were used to forecast the water quality according to different time series (months) k = 12 and k = 24 at the monitoring points. The LSTM-MA hybrid model provided faster training time and more reliable forecasts than the ARIMA, NAR, NAR-MA, and LSTM models. The RMSE and MAPE values of the LSTM-MA hybrid model for predicting the forecast set are smaller than those from other models. The LSTM-MA hybrid model was used to classify the water quality over the years 2020 to 2022. The water quality classification derived from the LSTM-MA hybrid model in close agreement with the actual monitoring data. This model becomes reliable and useful for the water quality classification of the Dong Nai River.

Online Support Vector Regression Based Adaptive NARMA-L2 Controller for Nonlinear Systems

NARMA model is a simple and effective way to represent nonlinear systems, based on the NARMA model, NARMA-L2 controller is designed and has been successfully applied in the literature. Success of NARMA-L2 controller is directly related to the precision with which controlled systems’ dynamics can be estimated. In this paper, online SVR is utilized to obtain controlled plant’s subdynamics and consequently this information is used in the construction of NARMA-L2 controller. Hence functionality of NARMA-L2 controllers and high generalization capability of SVR are combined. Also, SVR formulates a convex optimization problem and therefore guarantees global optimum solution. The proposed method is assessed by performing simulations on a nonlinear CSTR system, the robustness of the designed controller is also tested under noisy and uncertainty conditions.

Nonlinear Autoregressive Moving Average-L2 Model Based Adaptive Control of Nonlinear Arm Nerve Simulator System

This paper considers the trouble of the usage of approximate strategies for realizing the neural controllers for nonlinear SISO systems. In this paper, we introduce the nonlinear autoregressive moving average (NARMA-L2) model which might be approximations to the NARMA model. The nonlinear autoregressive moving average (NARMA-L2) model is an precise illustration of the input–output behavior of finite-dimensional nonlinear discrete time dynamical systems in a neighborhood of the equilibrium state. However, it isn't always handy for purposes of neural networks due to its nonlinear dependence on the manipulate input. In this paper, nerves system based arm position sensor device is used to degree the precise arm function for nerve patients the use of the proposed systems. In this paper, neural network controller is designed with NARMA-L2 model, neural network controller is designed with NARMA-L2 model system identification based predictive controller and neural network controller is designed with NARMA-L2 model based model reference adaptive control system. Hence, quite regularly, approximate techniques are used for figuring out the neural controllers to conquer computational complexity. Comparison were made among the neural network controller with NARMA-L2 model, neural network controller with NARMA-L2 model system identification based predictive controller and neural network controller with NARMA-L2 model reference based adaptive control for the preferred input arm function (step, sine wave and random signals). The comparative simulation result shows the effectiveness of the system with a neural network controller with NARMA-L2 model based model reference adaptive control system. Index Terms--- Nonlinear autoregressive moving average, neural network, Model reference adaptive control, Predictive controller DOI: 10.7176/JIEA/10-3-03 Publication date: April 30 th 2020

Application of affine NARMA model to design of adaptive power system stabilizer

Application of affine NARMA model to design of adaptive power system stabilizer

Network Traffic Anomaly Detection Based on N-ARMA Model

Network Traffic Anomaly Detection Based on N-ARMA Model

Implementation of NARMA-L2 Controller for Shell and Tube Heat Exchanger Temperature Process

In this research, the authors present the tuning and implementation of the nonlinear moving average autoregressive-linearization level 2 (NARMA-L2) controller for a shell and tube heat exchanger (STHE) temperature process. A mathematical model of the process plant is considered for simulation, and the model is discretized for the controller operation. The NARMA-L2 controller computes a sequence of control variable (the cold water inflow rate) adjustments in order to optimize the future performance of STHE process. The selected controller is trained to provide a quick control over the process variable in the nominal operating range of STHE by selecting appropriate constraints for NARMA models. The paper compares the performance of the selected NARMA-L2 controller with that of a benchmark controller, and the comparative results are presented.

Decoupling control in statistical sense: minimised mutual information algorithm

This paper presents a novel concept to describe the couplings among the outputs of the stochastic systems which are represented by NARMA models. Compared with the traditional coupling description, the presented concept can be considered as an extension using statistical independence theory. Based on this concept, the decoupling control in statistical sense is established with the necessary and sufficient conditions for complete decoupling. Since the complete decoupling is difficult to achieve, a control algorithm has been developed using the Cauchy-Schwarz mutual information criterion. Without modifying the existing control loop, this algorithm supplies a compensative controller to minimise the statistical couplings of the system outputs and the local stability has been analysed. In addition, a further discussion illustrates the combination of the presented control algorithm and data-based mutual information estimation. Finally, a numerical example is given to show the feasibility and efficiency of the proposed algorithm.

Support kernels regression for NARMA system identification (SKR<SUB align="right">NARMA)

This paper deals with the identification of nonlinear systems using multi-kernel approach. First, we have improved the support vector regression (SVR) method in order to identify nonlinear complex system. Our idea consists of dividing the regressor vector in several blocks, and, for each one a kernel function is used. This blockwise SVR approach is called support kernel regression (SKR). Furthermore, we have proposed two methods, SKR(rbf-lin) and SKR(rbf-rbf). Second, the SKR approach is improved to deal with the problem of NARMA system identification. Therefore, a new method called support kernel regression for NARMA model (SKRNARMA) is suggested. The basic idea is to consider the terms of auto-correlation and cross-correlation of the nonlinearity of input output discrete time processes, and for every term a kernel function is used. An example of MIMO system is presented for qualitative comparison with the classical SVR approach based on a single kernel function. The results reveal the accuracy and the robustness of the obtained model based on our proposed (SKRNARMA)-based approach.

Monopole-Gear Design Based on Neural Network and Modified Particle Swarm Optimization

In order to raise the design efficiency and get the most excellent design effect, this paper combined Particle Swarm Optimization (PSO) algorithm and put forward a new kind of neural network, which based on PSO algorithm, and the implementing framework of PSO and NARMA model. It gives the basic theory, steps and algorithm; The test results show that rapid global convergence and reached the lesser mean square error MSE) when compared with Genetic Algorithm, Simulated Annealing Algorithm, the BP algorithm with momentum term.

A New Kind of Optimization Method for Gear Design

In order to raise the design efficiency and get the most excellent design effect, this paper combined Particle Swarm Optimization (PSO) algorithm and put forward a new kind of neural network, based on PSO algorithm and NARMA model. It gives the basic theory, steps and algorithm; The test results show that rapid global convergence and reached the lesser mean square error MSE) when compared with Genetic Algorithm, Simulated Annealing Algorithm, the BP algorithm with momentum term.

Choosing the parameters of the NARMA model implemented with the recurrent perceptron for speech prediction

Speech signals have statistically nonstationary properties and cannot be processed properly by means of classical linear parametric models (AR, MA, ARMA). The neural network approach to time series prediction is suitable for learning and recognizing the nonlinear nature of the speech signal. We present a neural implementation of the NARMA model (nonlinear ARMA) and test it on a class of speech signals, spoken by both men and women in different dialects of the English language. The Akaike’s information criterion is proposed for the selection of the parameters of the NARMA model.

Model-Free Adaptive Control for a Class of Nonlinear Discrete-Time Systems Based on the Partial Form Linearization

The NARMA model is an exact representation of the input-output behavior of finite-dimensional nonlinear discrete-time dynamical systems in the neighborhood of the equilibrium state. However, it is inconvenient for purposes of adaptive control due to its nonlinear dependence on the control input, even by using the neural network method. In this paper, we introduce a so called model-free adaptive control (MFAC) method, which is based on some new dynamical linearization model and concept, the partial form linearization (PFL) and the pseudo-partial derivative (PPD) of a SISO nonlinear discrete-time system. The model-free means that the controller design is only based on the I/O data of the controlled plant, no training process, no structure information and no model are needed. Rigorous analysis and extensive simulations have shown that it has BIBO stability and performs very well.

Robust adaptive control of nonaffine nonlinear plants with small input signal changes.

Assuming small input signal magnitudes, ARMA models can approximate the NARMA model of nonaffine plants. Recently, NARMA-L1 and NARMA-L2 approximate models were introduced to relax such input magnitude restrictions. However, some applications require larger input signals than allowed by ARMA, NARMA-L1 and NARMA-L2 models. Under certain assumptions, we recently developed an affine approximate model that eliminates the small input magnitude restriction and replaces it with a requirement of small input changes. Such a model complements existing models. Using this model, we present an adaptive controller for discrete nonaffine plants with unknown system equations, accessible input-output signals, but inaccessible states. Our approximate model is realized by a neural network that learns the unknown input-output map online. A deadzone is used to make the weight update algorithm robust against modeling errors. A control law is developed for asymptotic tracking of slowly varying reference trajectories.

A systematic method to identify nonlinear dynamics of BWR by using the reactor noise

For the identification of the dynamics of the Vermont Yankee BWR with the reactor noise, different parametric models have been tested. The widely used ARMA model is unable to identify the nonlinearity in the noise data. A systematic method by using the NARMA model, which takes advantage of both the ANN and ARMA, is developed. Comparisons are made between the identification results with ARMA and NARMA model. The advantages of identification with NARMA model over ARMA model are demonstrated. The linear-kernels of the identified NARMA models are extracted so that the natural frequency, damping ratio and time constants of the BWR are obtained. The values of those characteristics are well corresponded with the eigenvalues calculated by the differential equations of the Vermont Yankee BWR. The damping ratio with negative value is found to be a criterion for the existence of limit-cycle, which can be seen from the impulse response on the (Xt, Xt−1) plane, in stable nonlinear system.

Inverse-NARMA: a robust control method applied to SI engine idle-speed regulation

This paper presents a control technique based on a directly identified non-linear inverse of a stable system. A NARMA model of the inverse of the uncontrolled system is identified using reversed input–output data collected experimentally from the plant. Linear identification is carried out on the plant which is pre-compensated by the identified inverse, sufficiently time delayed to be causal. This achieves a reduction in the uncertainty representation necessary to model the non-linearity. Subsequent robust linear controller design results in an improvement in robust performance when using the inverse pre-compensator. The method is applied to the automotive SI engine idle-speed control problem using a parameter space linear robust control design technique. The result is experimentally validated using an electric dynamometer.

Two Approaches for State Space Realization of NARMA Models: Bridging the Gap

This paper studies the necessary and sufficient conditions for observable realization of a general class of nonlinear high-order input-output difference equations. In particular, it proves the equivalence of the two seemingly different existing approaches in the literature. The paper also provides a subclass of NARMA input-output models that are guaranteed to have an observable realization. It is shown that this class covers several important subclasses of existing NARMA models.

INPUT-OUTPUT DECOUPLING OF NONLINEAR DISCRETE-TIME SYSTEMS BY STATIC OUTPUT FEEDBACK

The input-output decoupling problem for discrete-time nonlinear systems by static output feedback is studied in this paper. The discrete-time nonlinear system is described either by state equations or by high order input-output difference equations (NARMA model). In both cases the necessary and sufficient conditions are given for the solution of the problem.

A direct adaptive neural-network control for unknown nonlinear systems and its application

In this paper a direct adaptive neural-network control strategy for unknown nonlinear systems is presented. The system considered is described by an unknown NARMA model, and a feedforward neural network is used to learn the system. Taking the neural network as a neural model of the system, control signals are directly obtained by minimizing either the instant difference or the cumulative differences between a set point and the output of the neural model. Since the training algorithm guarantees that the output of the neural model approaches that of the actual system, it is shown that the control signals obtained can also make the real system output close to the set point. An application to a flow-rate control system is included to demonstrate the applicability of the proposed method and desired results are obtained.