This means that the main fuzzy controller, considering the measured external and internal conditions and set point values as inputs, Fig. 3. Example of a 3D surface for non-linear

determines the set point mapping between inputs and output as fuzzy model

position of the roller blind as implemented in illumination fuzzy controller. output. PID controller is of type

PID/V, which means that it executes the velocity PID algorithm, and the output value defines the dimension for which the actuator must change its current position. In our case this means the alternation of the roller blind position. The input values for PID/V are: the desired position of the roller blind, which is defined as output signal of the main fuzzy controller and the temporary measured position of the roller blind. The output signal is calculated from the current difference between the desired and the measured roller blind position. The output signal provokes the appropriate movement of the actuator, i. e. roller blind. The main illumination fuzzy logic controller has two inputs: set point inside illumination and the difference between the inside illumination and the set point illumination. A decisive factor for window geometry alternations is "illumination” fuzzy controller with proper semantic background. It is also important to set properly the other parameters in the algorithm: parameters of the PID controller, filter time constants, sampling times and priorities of the loops.

Possible illumination oscillations are in the range of 1000 to 5000 lx or even more in short time periods. Therefore, it is more difficult to achieve efficient daylight regulation than thermal regulation. The two filters realized in filter blocks are included to damp the possible too fast and frequent oscillations of the roller blind movements caused when the external solar radiation is extremely changeable. Proper setting of the filter time constants means smoother roller blind movement. We want to exclude too frequent roller blind moving, since it is annoying to the occupants.

The first step in designing the fuzzy controller is to specify the control input and output variables and the domains for these variables. Fuzzy partitions including the corresponding linguistic terms have to be specified for these domains. This means that the unit intervals are completed with membership functions (fuzzy subsets), i. e. membership degrees are assigned to numerical values. The number and the shape of the membership functions for each variable must be defined. For the purpose of the control engineering the triangular membership functions are used. In our case the Sugeno type (IDR BLOCK

Fuzzy Logic Controller Designing Tool, 1999) of the controller is used, where fuzzy partition is done only for the input domains. These fuzzy partitions and the linguistic terms associated with the fuzzy sets and subsets represent the database of our knowledge base. The next step is to define the control rules using the linguistic terms associated with the fuzzy sets as they appear in the fuzzy partitions of the domains. On the basis of the preliminary experiments and observations of the optical process in the test chamber, the set of linguistic rules is designed for the control of the roller blind positioning to maintain the desired inside illumination. Methods to design the fuzzy controller are crisp; the obtained control function is always crisp.

The first approximation of the fuzzy controller does not result in an optimal control behavior. To improve the control behavior, tuning of the fuzzy controller through iterative procedure of experiments is necessary. The changes are considered depending on how well the fuzzy controller is able to handle the process. The possible modifications are: Redefining the domains of the variables. The adjustment of fuzzy sets offers several possibilities: changing the fuzzy partitions of the domains, adding and deleting membership functions, reshaping and rearranging membership functions. For each fuzzy variable up to seven memberships functions can be included.

• Alternating the rules in the set.

• Exchanging the logic operations in some rules, i. e. choosing other logic operators.

• Adjusting the consequences of the individual rules.

The redesign is necessary, when the controlled variable (in our case inside illumination) deviates too much from the set point values. With a trial-and-error optimization of the designed fuzzy controller the control performance is improved.