Adaptive Single Neuron Anti-Windup PID Controller Based on the Extended Kalman Filter Algorithm

Hernández-Barragán, Jesús; Rios, Jorge D.; Alanis, Alma Y.; López-Franco, Carlos; Gómez-Avila, Javier; Arana‐Daniel, Nancy

doi:10.3390/electronics9040636

Cited by 11 publications

(6 citation statements)

References 49 publications

(80 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For example, Arec et al proposed a dendrite morphological neural network with an efficient training algorithm for some synthetic problems and a real-life problem [30]. Hernandez-Barragan et al proposed an adaptive single-neuron proportional-integral-derivative controller to manipulate a four-wheeled omnidirectional mobile robot [31]. Luo et al proposed a decision-tree-initialized dendritic neuron model for fast and accurate data classification [32].…”

Section: Introductionmentioning

confidence: 99%

A Neuron Model with Dendrite Morphology for Classification

et al. 2021

View full text Add to dashboard Cite

Recent neurological studies have shown the importance of dendrites in neural computation. In this paper, a neuron model with dendrite morphology, called the logic dendritic neuron model (LDNM), is proposed for classification. This model consists of four layers: a synaptic layer, a dendritic layer, a membrane layer, and a soma body. After training, the LDNM is simplified by proprietary pruning mechanisms and is further transformed into a logic circuit classifier. Moreover, to address the high-dimensional challenge, feature selection is employed as the dimension reduction method before training the LDNM. In addition, the effort of employing a heuristic optimization algorithm as the learning method is also undertaken to speed up the convergence. Finally, the performance of the proposed model is assessed by five benchmark high-dimensional classification problems. In comparison with the other six classical classifiers, LDNM achieves the best classification performance on two (out of five) classification problems. The experimental results demonstrate the effectiveness of the proposed model. A new perspective for solving classification problems by the proposed LDNM is provided in the paper.

show abstract

Section: Introductionmentioning

confidence: 99%

A Neuron Model with Dendrite Morphology for Classification

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Some of these techniques require access to the complete state of the system and information on its uncertainties and delays, and usually complex calculations (Tahoun 2015(Tahoun , 2017c(Tahoun , 2020. Among these techniques, neural networks stand out; their characteristics allow the implementation of easy, fast, and robust PID controllers known as neural PID controllers, which vary mainly on architecture and training methodology (Rios et al, 2020b;Tahoun & Arafa, 2020;Hernandez-Barragan et al, 2020).…”

Section: Introductionmentioning

confidence: 99%

“…But the inputs can also include the proportional, derivative, and integral errors (Sento & Kitjaidure, 2016). Adaptive neural PID controllers trained with the extended Kalman filter (EKF) algorithm based algorithms have proved to show faster learning speed rates and convergence time than adaptive neural PID based on backpropagation training methods, which makes EKF training based neural PID controller more suitable for experimental and real-time tests (Hernandez-Barragan et al, 2020). Also, training algorithms based on Extended Kalman filter (EKF) for neural networks have proven to reliable for recurrent and feedforward neural networks for control applications, presenting real-time applications (Haykin, 2004;Sanchez, Alanis & Loukianov, 2010;Alanis, Arana-Daniel & Lopez-Franco, 2019;Rios et al, 2020a).…”

Section: Introductionmentioning

confidence: 99%

“…Besides the previously mentioned withdraws of PID controllers, a common problem is the windup effect, which is the result of accumulative error action due to the integral part of the controller. This effect produces saturation on actuators and contributes to lowperformance, overshoot, high settling time, and instability, losing controllability (Visioli, 2006;Kumar & Negi, 2012;Hernandez-Barragan et al, 2020), which is the reason why anti-windup strategies are important when using PID controllers Tahoun (2017a). Among the proposed anti-windup strategies in the literature are limiter integrator, back-calculation, and observer approach (Visioli, 2006;Kumar & Negi, 2012;Kheirkhahan, 2017;Angel, Viola & Paez, 2019).…”

Section: Introductionmentioning

confidence: 99%

“…However, it requires and anti-wind up methodology. The contribution of this work is the proposal of adaptive neural PD controllers trained with extended Kalman Filter-based algorithms; the adaptive properties of these controllers allow them to compensate for the missing integral part, which can cause the windup effect; in this way, there is no need for an anti-windup methodology as in previous work (Hernandez-Barragan et al, 2020) and at the same time getting a good performance. The proposed EKF-based trained PD controllers, single neuron and multilayer adapt their weights online, eliminating the steady-state error; moreover, oscillations and overshoot are highly suppressed.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Adaptive neural PD controllers for mobile manipulator trajectory tracking

Hernández-Barragán¹,

Rios²,

Gómez-Avila³

et al. 2021

PeerJ Computer Science

Self Cite

View full text Add to dashboard Cite

Artificial intelligence techniques have been used in the industry to control complex systems; among these proposals, adaptive Proportional, Integrative, Derivative (PID) controllers are intelligent versions of the most used controller in the industry. This work presents an adaptive neuron PD controller and a multilayer neural PD controller for position tracking of a mobile manipulator. Both controllers are trained by an extended Kalman filter (EKF) algorithm. Neural networks trained with the EKF algorithm show faster learning speeds and convergence times than the training based on backpropagation. The integrative term in PID controllers eliminates the steady-state error, but it provokes oscillations and overshoot. Moreover, the cumulative error in the integral action may produce windup effects such as high settling time, poor performance, and instability. The proposed neural PD controllers adjust their gains dynamically, which eliminates the steady-state error. Then, the integrative term is not required, and oscillations and overshot are highly reduced. Removing the integral part also eliminates the need for anti-windup methodologies to deal with the windup effects. Mobile manipulators are popular due to their mobile capability combined with a dexterous manipulation capability, which gives them the potential for many industrial applications. Applicability of the proposed adaptive neural controllers is presented by simulating experimental results on a KUKA Youbot mobile manipulator, presenting different tests and comparisons with the conventional PID controller and an existing adaptive neuron PID controller.

show abstract