Information Technology Reference
In-Depth Information
generation block changes to a weighted sum of six components, then the candidate
solution will be considered as the best solution in solution space. In the third phase
the optimization process will be continued to reduce error.
4.1
Controller Structure Details
To design a controller which satisfies different objectives at first we assume that
these objectives can be decoupled and then one separate controller should be de-
signed to fullfill each of which. Finally, outputs of different controllers will be fused
together to obtain the desired control signal. In our proposed controller, two BEL-
BICs have been assigned to our objectives: position tracking for cart, and angle reg-
ulating for pendulum. As described in previous sections, two major inputs should
be provided for BELBIC, which are sensory and stress inputs. Under inspiration of
[11] the cart position error and its first derivation are defined as sensory signals for
one Belbic; more over, the pendulum angle and its first derivation are defined as
sensory signals for the other one. Note that each BELBIC controller has two neu-
rons for sensory inputs. Fig. 3 shows the design and structure of proposed controller
in Simulink environment. As indicated in this figure, just a summation operator is
used to combine the output of controllers. Decoupling of controllers is compensated
by fusing the objectives in reward or stress functions. This approach has two major
benefits: first, reduction of fusion cost, and second, not changinge the structure of
controller between learning and optimization phases.
Fig. 3
Design and structure of proposed controller in Simulink environment
5
Stress Generation
5.1
Imitation Phase
In imitation phase, BELBIC learns the control signal produced by the initial basic
controller; therefore we need to consider the control signal error to generate stress
Search WWH ::
Custom Search