ij ij ij

(4)

ij vij

and uij

as 1。 As seen from Fig。 2, the reverse

that the system approached to the sliding surface   would

transmission error of the output layer is described as

also be correspondingly reduced and the time to reach

M M

( f (4) (I (4) ))

the sliding surface would be longer。 In order to meet   the

(4) j j

  

j (I (4) )



( f (4) (I (4) ))



(I (4) )

rapid nature of the system and reduce the chattering,  the

j  j j j

particle  swarm  optimization  algorithm  is  adopted    to

(4) (4)

d j     Oj

As  w(3)  1,

(15)

the  rule-layer  had  only   the   reverse

optimize the values of cx, cl, cθ, α, ε1, k1, ε2 and k2。

Similar   to   other   evolutionary   algorithms,  PSO

algorithm  works  well  with  the  fitness  value  of   each

transmission error。

(3) M

  M

( f (3) (I (3) ))

particle based on the notations of “group” and “evolution”  [16]。  In  the  original  PSO,  the  position of

j        

(I (3) ) ( f (3) (I (3) ))



(I (3) )

j  j j j

each particle in the swarm represents a possible solution。

(4) (4)

k jk

k

(16)

The position and velocity of particle i at iteration n are denoted  as  xi, d(n)  and  vi, d(n),  respectively。  The    new

The degree of membership layer anti-pass error is velocity at the next iteration, vi,d(n+1), is calculated by

M M

( f (2) (I (2) ))

using its current velocity vi,d(n), the distance between the

(2) j j

j  

(I (2) )



( f (2) (I (2) ))



(I (2) )

particle’s best previous position pi,d(n) and xi,d(n), as well