Design of NARMA L2 Neurocontroller for Nonlinear Dynamical System

Artificial Neural Network-based Neurocontroller for Hydropower Plant Control

“…The NARMA-L2 control technique, which is an artificial neural network (ANN)-based neuro-controller, offers an efficient solution to the problem via utilizing a backpropagation training algorithm. NARMA-L2 has been used as a controlling technique that linearizes the model inside the controller and cancels the disturbance dynamics to the input in the system, simultaneously aiming to maintain the original dynamic of the system [13]. The training process is performed offline and uses approximated models which represent the process dynamics.…”

“…Figure 2. NARMA-L2 neuro-controller structure [13] Mathematical model of the control law, with a structural representation as in Figure 2, is defined by the following discrete time characteristic equation, where in (1), ( 2) and (3), variables have the same meaning [16]:…”

Comparative analysis of PID and neural network controllers for improving starting torque of wound rotor induction motor

“…ANN is implemented in many applications such as electric power and there are promising in dentification/control system, the self-adapting and notable-rapid computing functions of ANN cause them to well desirable to deal with nonlinearities [25,26].This model consists of two nets, the first net implement is a controller and the other net simulates a version of the plant, NARMA-L2 model can be used to model the plant previously cited, using two distinct neural networks, it makes use of a nonlinear identification tool. The neuro controller is noted by special names: nonlinear auto-regressive moving average and feedback linearization control, while the plant model has a selected method (associate formulation), the plant model can be estimated by NARMA-l2 controller when the same formula, therefore the object of NARMA-L2 is to convert the system from nonlinear to linear dynamics system [27], when using the NARMA-L2 controller, the first step is identify the plant model, that is included of two functions f and g [28], these functions are representing the past values for each the output (y) and the control effort (u) determine by tapped delay lines (TDL), while the second step includes simply the rearrangement ofthe two sub networks f and g trained offline, at the result the computation time is decrease [29]. Figure 6 can be shown NARMA-L2 control structure, when using the NARMA-L2 controller, the first step is identify the plant model, that is included of two functions f and g [28], these functions are representing the past values for each the output (y) and the control effort (u) determine by tapped delay lines (TDL), while the second step includes simply the rearrangement of the two sub networks f and g trained offline, at the result the computation time is decrease [29].…”

“…The neuro controller is noted by special names: nonlinear auto-regressive moving average and feedback linearization control, while the plant model has a selected method (associate formulation), the plant model can be estimated by NARMA-l2 controller when the same formula, therefore the object of NARMA-L2 is to convert the system from nonlinear to linear dynamics system [27], when using the NARMA-L2 controller, the first step is identify the plant model, that is included of two functions f and g [28], these functions are representing the past values for each the output (y) and the control effort (u) determine by tapped delay lines (TDL), while the second step includes simply the rearrangement ofthe two sub networks f and g trained offline, at the result the computation time is decrease [29]. Figure 6 can be shown NARMA-L2 control structure, when using the NARMA-L2 controller, the first step is identify the plant model, that is included of two functions f and g [28], these functions are representing the past values for each the output (y) and the control effort (u) determine by tapped delay lines (TDL), while the second step includes simply the rearrangement of the two sub networks f and g trained offline, at the result the computation time is decrease [29]. Figure 7 shows the model which consists of WRIM with NARMA-L2 controller that includes two input and one output represents starting torque.…”

Design of Neural Network Controllers for the Horizontal Positioning of an Industrial Manipulator

“…This type of controller transforms a nonlinear system dynamic into a linear one. The first stage in using Narna-L2 control is to identify the plant model and to train the controller using input-output pairs, in order to obtain a controller that represent the dynamic of the system to be controlled [7]. After the controller is trained offline, two neural network approximation functions f and g result that represent the functions of the past values for the plant output y and the control effort u, values recorded by tapped delay lines (TDL) as presented in Fig.…”