Neural Network PID Decoupling Control Based on Chaos Particle
Swarm Optimization
TENG Wei-feng, PAN Hai-peng, REN Jia
College of Mechanical Engineering and Automation, Zhejiang Sci-Tech University, Hangzhou 310018, P. R. China
E-mail: tengweifengdeyou@163.com
Abstract: As a new kind of neural network model, Neural network PID (PIDNN) combines the advantages of PID and neural
network. However, the error back propagation algorithm (BP) limits the performance of PIDNN. In order to realize effective
control of nonlinear, large delay and strong coupling system, this paper proposes a neural network PID control method based
on chaos particle swarm optimization. Using chaos particle swarm algorithm to replace the reverse pass algorithm of original
PID neural network, adjusting the weights of PIDNN between each neuron, the algorithm achieved rapid decoupling control
effect. The simulation results show that the proposed method in this paper, compared with the original BP algorithm, has more
excellent dynamic and steady-state performance.
Key words: Neural network PID; Chaos particle swarm optimization; Decoupling control.
1. Introduction
The structure of industrial control system with
multivariate is always complex. All variables of the system
are mutual influence and there exists serious coupling
between them. Changes in input variables usually cause
changes in multiple output variables, thus affecting the
control effect. If the coupling between the variables of the
system is very serious, the system will not work properly.
So far, decoupling method of multivariable system
mainly includes: traditional decoupling, adaptive
decoupling and intelligent decoupling. With the
development of control theory, applications of intelligent
decoupling are increasingly popular. Because of the unique
advantages of the neural network decoupling in solving
nonlinear problems, it gets widespread concern in the
decoupling control of nonlinear system
[1]
.Paper[2]
proposed two methods for decoupling control, one method
is to design a neural network compensator for making the
Bristol first coefficient matrix including neural networks
compensation as a diagonal matrix; The other method is to
design a neural network decoupling compensation device
by neural network series, parallel and feedback operation.
However, due to the short shortcoming of slow learning
rate and poor dynamic performance of neural network, the
method can not eliminate the static coupling.
Paper[3],combining genetic algorithm with neural network
PID, designed a fitness function based on performance
indicators. By the calculations of adaptive crossover
probability and mutation probability, the method changed
the network weights, ensured to get the optimal parameter,
eliminated the coupling between variables and realized the
decoupling control of the system. Based on the
optimization features of the Particle Swarm Optimization,
paper[4] designed a PID neural network decoupling
controller based on particle swarm optimization, and made
it be used for multivariable
nonlinear coupling control. Due
to the weak point of PSO which is easy to fall into local
This thesis is supported by the National Natural Science Fund
(61203177)
extreme, slow convergence in later evolution, and poor
stability. The initial weights which is optimized by PSO,
are not necessarily the optimal value. This will affect the
dynamic and static performance of the control system.
Based on the above literaturesˈthis paper designed a neural
network PID controller by using the chaotic particle swarm
optimization algorithm to optimize the neural network
weights of each layer, eliminating the coupling of multi
input multi output system, achieving the control objectives
of fast response, small overshoot, and high accuracy,
obtaining good performance of dynamic and steady.
2. Neural Network PID Decoupling Control
Based on Chaos Particle Swarm Optimization
2.1 Neural Network PID
PID neural network (PIDNN) is to embed the PID
control into the neural network whose hidden layers are
consists of proportion, integral, differential neurons. The
structure of double input and double output coupled system
is shown in Fig 1.It is a three layer feed forward neural
network. and it is constituted by a plurality of parallel
sub-network. The structure of each sub-network
is
231uu
.The input layer has two sub-networks of neurons,
they are given value of the input and feedback value of the
output. There are three neurons in hidden layerˈthey are
proportional, integral and differential neurons.
Their state transition functions are respectively
proportional, integral and derivative functions which are
limiting. If there are
n
control variable, you can also build
a decoupling control system consists of
n
sub-networks.
The traditional PIDNN adjust the inertia weight by training
the network to obtain the minimum value of the objective
function, at the same time to complete the decoupling of
multivariable control system
[5]
Proceedings of the 33rd Chinese Control Conference
Jul
y
28-30, 2014, Nan
j
in
g
, China
5017
资源评论
