1 Introduction: The State-of-the-Art

In spite of the fact that the idea of Internet of Things (IoT) was born many years ago after Nikola Tesla‘s interview in Colliers magazine (1926, [1]), it was stated: “When wireless is perfectly applied the whole earth will be converted into a huge brain, which in fact it is, all things being particles of a real and rhythmic whole… and the instruments through which we shall be able to do this will be amazingly simple compared with our present telephone.” We were forced to wait approximately 100 years to be able to start the full-fledged implementation of the idea into our home ambience. A number of research publications regarding the solutions of IoT problems can be found today; for example, Google presents 39 600 000 citations in the year 2008, and 130 000 000—in the year 2013. Six commonly agreed areas of human life are considered as ready for IoT implementation: Smart manufactory, Smart transport, Smart health, Smart shopping, Smart city, and Smart home/office are on the top of the list of the broad spectrum of application of different IoT-computerized gadgets, artifacts, and services [2], [3]. The significance of the problem under consideration is recognized on a global level: this is indicated not only by annual IoT forums under the hospice of the IEEE (The IEEE Standards Association; an Internet of Things (IoT); 5–6 November 2013 in Silicon Valley, California [4], IEEE World Forum on Internet of Things (WF-IoT); 6–8 March 2014, Seoul, Korea [5] ), but also by a very serious attitude of the European Commission (EC) to the problem expressed by organizing the IERC—European Research Cluster on the IoT&S consisting of 33 research and development projects [6]. Currently, a European road map for the future research and innovations in the field of IoT and services for smart home environment is encouraged and supported in Horizon 2020 program as well [7–9]. All six areas of human activity mentioned above are intrinsically fuzzy. This is the reason why: (1) we have uncountable number of fuzzy logic-related patents (560,000 by 2014), and (2) a wide range of applications from smart manufactory control [10, 11] to hand-shake compensation and blood pressure monitoring gadgets [12, 13].

A thorough analysis of multiple sources focusing on architecture of IoT&S and environmental intelligence, like [14–16], shows that the main tendency to create a Smart Home Environment (SHE) is based on the usage of multi-agent systems. However, in this paper we advocate an approach different from [14–22] and many others; we claim that a significant difference exists between smart and intelligent things (and smart and intelligent agents as well). This new approach is widely presented in [23, 24]. According to this approach, a thing that can be electronically controlled and/or a “crisply programmed” agent, is called smart. However, we can call things and/or agents intelligent only if they are able to behave by modeling most simple features of human beings widely considered as his/her intellectual activity: (1) recognition and classification (of patterns, processes, situations), (2) behavior according to a set of fuzzy rules, and (3) functioning according to some prescribed tendency. In all these three activities, reasoning and control based on fuzzy logic play a decisive role.

2 SHE Intellectualized Control

The SHE itself must be considered not only as a certain set of complex sensors, actuators, means of communication, and smart things, but also as a complex sociotechnical fabric with the inherited mixture of what is really human, as well as elements of artificial intellect. Generally speaking, SHE is a fuzzy control system [25]. Its functional organization is presented in Fig. 1. Here, HE stands for the home environment functioning under the parameters of real ambient nature (NAT) and signals that express user’s personal wishes (PW). Fuzzy decision-making system (FDMS) must perform an intellectualized HE control. Its intellectual activity is based on a set of fuzzy rules that are created during the initial training, retraining (if user changes his/her wishes), and self-training (if there is no a priori information) processes that the system undergoes. The normalized (NORM) description \( \vec{x} \) of a concrete situation in the HE is fuzzified (F), processed according to a set of IF… THEN rules and converted by the defuzzifier (DF) into reasonable actions (A1–A3). By normalizing here must be understood procedures serving for the transformation of real physical data into form acceptable for computerized processing. All events and situation changes taking place in a home environment are collected and stored in a specialized data base (DB) which is further analyzed and evaluated by special expert system and/or by certain experts (ES). The results of this analysis serve as data for creating reasonable terms for verbalization of environmental parameters and for producing a proper list of IF–THEN type inference rules as well. The dotted line emphasizes an automatic online possibility (if it is necessary) to influence or even change the shapes of F terms and IF–THEN rules.

Fig. 1
figure 1

SHE intellectualized control system

Full size image

This research is focused on algorithmic technologies for the intellectualization of a smart home environment based on fuzzy logic and fuzzy rules processing.

A novel approach to fuzzy training, retraining, and self-training, as well as new algorithms for intellectualized home environment are proposed. All three fuzzy logic-based training algorithms take top-down hierarchical analysis of home situations under consideration to conquer the curse of the increasing number of rules. The paper is organized as follows: Sect. 2 presents’ the general description of intellectualization of the Smart Home Environment; Sects. 3–5 describe fuzzy algorithms for SHE training, retraining, and self-training, respectively. Results of modeling and simulation are presented in Sect. 6. Concluding remarks are presented at the end of the paper.

Note: all denotations of crisp and fuzzy variables that can be found in this paper are in normal style and their values are in italics.

3 SHE Initial Training

In general, SHE training process can be considered as a construction of a fuzzy rules set. An old-fashioned process of constructing fuzzy rules for SHE behavior, mostly based on the results of user’s interview, is considered as SHE training process. Usually a team of users, professional experts, and IT people accumulate, discuss, and prepare a set or requirements to be formalized and presented in the form of a special type of user cases and functional requirements for the system. Defined linguistic variables (x1, x2,…,xN) and types of their membership functions created rule set base (RB, consisting of general number I rules) and the identified defuzzification methods must be consider as the training result of a system.

Usually, if no special means are taken, the number of rules is expressed by the Descartes multiplication of all term numbers of all variables. So fighting the curse of dimensionality is the main problem of the initial training process. The goal of this section is to propose a hierarchical procedure to decrease the number of rules based on pragmatic causal interpretation of situations occurred in the HE.

In this paper, a simple SHE example (Fig. 2) is chosen to present the method used for minimizing the number of rules.

Fig. 2
figure 2

Simplified example of SHE

Full size image

Let us imagine the room (HE) with two working places (WP1, WP2) equipped with light sources (L1, L2). These light sources are controlled from FDMS using actions (A1, A2) which mean the status of voltage supplied to the corresponding light source, as “Increase”, “Decrease”, and “N” (do nothing).

Lighting (s1, s2) in each working place is measured by sensors (S1, S2).

It is considered that two users (ID1, ID2) may express their different wishes concerning the light intensity in different working places and corresponding to different users’ positions in the HE room. It must be emphasized that user’s presence in the room and his or her position in it is determined by corresponding space coordinates (xID1, yID1, and xID2, yID2).

A thorough analysis of the habits and wishes of users permits to create: (a) main terms for each variable under consideration, and (b) a set of rules to be performed in FDMS. The terms for space coordinates X and Y are shown in Fig. 3a, the terms for s1 and s2 lighting—in Fig. 3b and the output terms—in Fig. 3c.

Fig. 3
figure 3

Terms of Input a, b, and Output c Variables

Full size image

In general, the frame of every rule consists of concatenation of two strings: the string of antecedences x1, x2,…, xN in the IF part of the rule, and the string of consequences y1, y2,…, yM in the THEN part of the rule. So the r-th rule is written in the following way:

$$ {\text{IF }}\left\{ {{\text{x1}},{\text{ x2}}, \ldots ,{\text{xN}}} \right\}{\text{ THEN}}\;\{ {\text{y1}},{\text{ y2}}, \ldots ,{\text{yM}}\} $$

According to [26], the first row in the list of rules can be considered as a string of corresponding attributes assigned to the sensors and actuators of the environment under consideration. All other rows are defined as verbal antecedence connected by a fuzzy logic relation AND (it means min) in the IF part, and verbal consequence obtained by a fuzzy logic operation CoG in the THEN part. For example, in our case such a frame is shown in Fig. 4.

Fig. 4
figure 4

The frame of rules

Full size image

As it was mentioned above, the number of rules is expressed by the Descartes multiplication of all term numbers of all variables; in HE example, the number is 2 × 5 × 2 × 5 × 5 × 5 = 2500.

The goal of this section is to propose a hierarchical procedure to decrease the number of rules based on pragmatic causal interpretation of situations that occur in the HE.

There are several rule reduction methods. In [27], a method of merging the variables is presented. [28] is based on the usage of a multi-level logic. [29–32] consider natural variable separations based on the specificity of controller design. Coefficients evaluating the rule’s redundancy are successfully used in [33]. Some of these methods are very efficient, but to our view, they are too artificial and specified (suitable only for the problem under investigation); they require too many additional efforts.

This paper presents a new approach for the minimization of rules, and this approach is based on a hierarchical description of HE situations. The hierarchy covers the augmenting knowledge of the events occurring in the home environment that needs to be intellectualized. The sequence of events is presented by hierarchical decision tree, as it is shown in Fig. 5.

Fig. 5
figure 5

Top-down hierarchical decision tree based on fuzzy logic

Full size image

In the experimental model of HE (Fig. 2), an equipped room and its users with their different PW concerning the light intensity in working places, was investigated. For example, the analysis of situations that occur in the room starts from the question: “Is one of the ID1 or ID2 in, or are they both in?” This analysis permits to use information included in the identity cards of users and behave according to the real situation instead of processing full information, even if part of it is irrelevant/redundant to this particular situation. For the sake of simplicity, only a crisp analysis of a situation was presented in this paper, however, the fuzzy reasoning-based situation analysis can be applied., for example, the algorithms of persons'/users' identity recognition are possible as well.

After the crisp answers to those questions are received, the selection of a proper set of rules is usually made in the following hierarchical fuzzy blocks: “ID1 fuzzy position set”; “ID2 fuzzy position set”; “ID1 and ID2 fuzzy position set”, where the set of Takagi–Sugeno rules are used in the following way:

$$ \left( {\text{r}} \right){\text{ IF XID2}} = {\text{a}}_{\text{r}}\; {\text{and YID2}} = {\text{b}}_{\text{r}}\; {\text{THEN GO TO}}\mathop {{ \hbox{max} }\;z_{r} }\limits_{\forall r} ; $$

where zr = Kr μ(ar) × μ(br) for \( \forall r \); μ(*)—membership function of the corresponding value, Kr—positive weight coefficient, which emphasizes the significance of the r-th rule.

Frames of rules for sets of a fuzzy position are shown in Fig. 6.

Fig. 6
figure 6

Frames of rules for fuzzy position sets: for ID1-(a), for ID2-(b), and for ID1 & ID2-(c)

Full size image

In our example, blocks “ID1 fuzzy position set” and “ID2 fuzzy position set” contain 10 rules each, and “ID1 and ID2 fuzzy position set” contains a 100 rules.

After the user’s position in the room is determined, the levels of light intensity s1, s2 must be investigated. One or another set of rules from the block “Set of fuzzy decision rules’ subsets for light control” (from Fig. 5) starts to work and provides a certain decision. It should be emphasized that the frame of rules in each set for light control has the same frame, presented in Fig. 7.

Fig. 7
figure 7

Frame of rules for light control

Full size image

Each of 16 blocks of “Set of fuzzy decision rules’ subsets for light control” contains 25 rules. So, a total number of rules for light control in our pending example in FDMS, when a hierarchical decision tree is used, is 10 + 10 + 100 + 16*25 = 520. The efficiency gained in comparison with the initial case is 2500/520 = 4.8.

4 SHE Retraining

The trained SHE system has a rule set base (RB) consisting of I rules, as it was mentioned in Sect. 2. In this section a training algorithm is delivered for the general case according to the frame of rules presented in Fig. 4. It is obvious that in case of the hierarchical structure of a decision tree, the retraining process must be repeated for each structural block (Fig. 5).

SHE retraining process may be explained and understood by analyzing the changing wishes of user ID1, as given in Fig. 3b. It is shown that the former wish (Former brightness—Dusk) is changed into a new one—NPV (New Parameter Value—New brightness (Fig. 3b)). It means that from now on, the user ID1 will be satisfied with brightness of 270 lx at the working place WP1 (xID1 = 130, yID1 = 130), as it is shown in Fig. 3a.

The retraining may be considered as a construction of amendments in consequence of corresponding rules. So the first step is finding the rules that need to be changed.

According to F data of sensors that describe positions of users, a set of all possible antecedent parts R of rules is created (see Fig. 8a). For the sake of clarity, the truth values of corresponding terms (Fig. 3) are presented in the same figure as well. One antecedent part R(j) which corresponds to the highest product of membership functions of the corresponding values of parameters is left and used for further operations (Fig. 8b).

Fig. 8
figure 8

Generated set R of all possible rules (a) and the selected rule R(j) (b)

Full size image

An antecedent part of this rule is used to select all rules to be changed in Rule set base obtained during the initial training process. Consequent parts of the selected rules are changed according to the pseudo code and the algorithm presented in Fig. 9.

Fig. 9
figure 9

Pseudo code to determine output term A1 (a), a typical term T of the input variable s1 (b), and algorithm corresponding to the pseudo code (c)

Full size image

The whole process is organized as follows (Fig. 9): (1) the selected rule R(j) is consequently compared with the rule R(i) ∀i (here i is the number of the rule) from the Rule set base; (2) if the correspondence of A of the rule R(j) with the part B of the rule R(i) is determined, then the consequent A1 in rule R(i) must be changed to “Increase”, “Decrease”, or “N” (see Fig. 3c) according to the antecedent of s1 in this rule, as it is presented by the pseudo code (Fig. 9a and the block diagram in Fig. 9c). Here s1T[b] and s1T[c] are values s1 of the term T (Fig. 9b). The T is each possible term of variable S1—“Dark”, “Dusk”, “Light”, “V. Light”, “Dazzle” (Fig. 3b).

After the retraining process is finished, SHE will take decisions according to the new wish of user ID1.

5 SHE Self-training

In this section, the self-training process is considered as a SHE training “from zero”, i.e., when we have no priori information concerning the user’s wishes. At least two important cases must be distinguished: (1) terms of variables under consideration are known in advance; (2) only minimal and maximal values of variables are known. The main difference between these two cases is the need to build necessary terms from data produced at the moments when user expresses his wishes in case 2. Both cases are demonstrated in a simple SHE example, described in Sect. 3.

5.1 Self-training: Case 1

As it was mentioned before, in this case all terms of input and output variables are known and shown for our simplified example in Fig. 3. The frame of possible rules is considered as known as well, and is presented in Fig. 4. Let us imagine that the user ID1 while at a position xID1, yID1, as it is shown in Fig. 3a, expresses his first wish, for example, “New lighting” by adjusting the lighting 270 lx in the working place WP1 (Fig. 3b). The second user ID2 has not expressed any wish and at the moment is not taken into account, except for his position.

The self-training process in case 1 is organized as follows: (1) Set of all possible antecedent parts R of rules is created to determine the positions of ID1 and ID2 according to terms shown in Fig. 3a similarly as it was done in Fig. 8a. (2) One antecedent part R(j) corresponding to the highest product of membership functions of the corresponding parameters values is left for further operations (see Fig. 8b). (3) This left antecedent part R(j) is expanded by the burst B, consisting of K new rules with all possible combinations of five terms S1 and S2 (Fig. 10). As a matter of fact, the preliminary list of those 25 rules consists only of antecedents, and the next step serves for the determination of their consequents. (4) All those consequents are determined according to the pseudo code and algorithm depicted in Fig. 11. It must be emphasized that s1T[b] and s1T[c] are s1 values of S1 of the term T. The T is each possible term of variable S1 (like as S1 in Fig. 9b). Determined consequents are presented in the action column A1 (Fig. 10). A column A2 must be left unchanged because there is no information concerning brightness s2 at the working place WP2.

Fig. 10
figure 10

Burst B of generated rules

Full size image
Fig. 11
figure 11

Pseudo code to determine the output term A1 (a) and the algorithm corresponding to the pseudo code (b)

Full size image

Analogous steps must be performed in case if user ID1 expresses another wish that is different and/or the second user ID2 expresses any wish, and so on.

5.2 Self-training: Case 2

In case when only minimal and maximal values of input variables are known, the self-training procedure starts from building necessary terms from data produced at the moments when a user expresses his wishes. It must be stressed that terms expressing possible SHE actions are known in advance, as it was described in Sect. 3. By the way, rule generation procedure consists of the same steps as in case 1 (Sect. 4 A). The number of produced terms increases with every new wish the user makes. So in general, FDMS (Fig. 1, 2) may be considered as a complex of cumulative set of terms and cumulative list of rules.

The process of set generation for terms during the self-training in case 2 is organized as follows:

  1. (1)

    Starting with the first new wish NW, a corresponding term “Wish1” is introduced (Fig. 12b). It must be emphasized that the triangular term with a base from Min S till Max S is chosen only for the example under consideration. Many other types of terms (trapezium type, with different fixed base width and so on) may be used after consulting with the user and IT specialists.

    Fig. 12
    figure 12

    Description of term generation procedure in the self-training case

    Full size image
  2. (2)

    Two additional terms expressing the so-called contradicting wishes (C_Wish) for terms “Less” (C_WL) and “More” (C_WM) must be introduced as well. The shape of all those terms ready for computerized processing is described in Fig. 12. Here a, b, and c denote the values of parameters of a corresponding triangular term (Fig. 12c, d). In Fig. 12a, pseudo code is presented for the generation of terms, here Sterm[i] = {Wish[i], C_WL, C_WM}. In Fig. 12e, f, the process of generating new terms (Wish2, Wish3, Wish4) is presented.

  3. (3)

    Further, step 1 and step 2 have to be repeated every time when a new wish appears and the terms of wishes in the Figs. 12e are to be listed according to increasing values of the parameter b. During this process, [a, c] of all terms are subjected to the changes widely recommended in literature [34].

It is important to mention that in the process, we use the same output action terms in the self-training as it was done in training and retraining processes.

6 Modeling and Simulation Results

All training, retraining, and self-training procedures were simulated in a simple SHE test environment presented in Fig. 2. The simulation was performed on the SHE model developed as a special software/hardware system BIAsim shown in Fig. 13. The BIAsim permits to represent and visualize all gadgets, sensors, actuators, and moving SHE users who are to benefit from the comfort provided by intellectualized actions of the environment. Agent-based fuzzy intelligence is stored in the Fuzzy Decision Model library (FDM). Possible environmental situations are stored in the BiaTechSimDB. BiaTechFuzzyDB serves as a library for training, retraining, and self-training algorithms, input/output terms of variables, and sets of fuzzy rules. It must be noted that Excel format was used for initial presentation of fuzzy rules sets and input/output terms as well. Mamdani-type inference and decision-making procedures were performed in the FDMS for light control. Takagi–Sugeno type procedures were used for ID position setting. Corresponding actions were transferred into SHE model on the BIAsim screen for visualization of control effects.

Fig. 13
figure 13

A special software/hardware modeling system

Full size image

Experimental training process illustration was performed according to the requirements received from a project “Research of Smart Home Environment and Development of Intelligent Technologies (BIATech)” [project VP-1-3.1-ŠMM-10-V-02-020] by the team of IT specialists consisting of 8 persons. They created a list and a vocabulary of fuzzy variables, terms, and shapes of membership functions, satisfying the requirements of ordinary users. Using this information, 2500 rules were generated. The number of these rules was later minimized, as it was described in the Sect. 3. A randomly taken group of those rules constructed according to the frame (given in Fig. 4) is presented in Fig. 14a. Examples of rules that underwent the hierarchical minimization procedure (a frame Figs. 6a, 7) are given in Fig. 14b, c).

Fig. 14
figure 14

Fragment of rules for general training case (a), an example of rules in hierarchical case for ID1 position set (b), and fragment of rules for lighting control in a hierarchical case (c)

Full size image

Experimental simulation on the BIAsim has demonstrated and confirmed the behavior of the SHE model predicted in Sect. 3. Achieved simulation precision is higher than 1 lx.

An old-fashioned process of constructing fuzzy rules for SHE behavior, mostly based on the results of user’s interview, is considered as SHE training process. Usually, a team of users, professional experts, and IT specialists accumulate, discuss, and prepare a set or requirements to be formalized and presented in the form of a special type of user cases and functional requirements of the system. Defined linguistic variables, types of their membership functions, created rule base, and identified defuzzification methods must be considered as the training result of the system. Simplified results of such SHE training are described in Sect. 3.

Four experimental retraining cases were taken according to terms presented in Fig. 3 for user ID1 with coordinates xID1 = 130 and yID1 = 130. Previous lighting (the lighting satisfying users' primary wishes), a new one set by the user ID1, and the lighting supplied by the system after its retraining, are presented in Table 1.

Table 1 Retraining results for four cases
Full size table

The intensity of lighting in the working place reached by the SHE model on BIAsim is determined by the value b of the corresponding term if: (1) the new user’s wish belongs to the base of only one term, and (2) retraining starts from the darkened working place. The intensity of lighting in the working place reached by the SHE model on BIAsim is determined by the value c of the corresponding term if: (1) the new user’s wish belongs to the base of only one term, and (2) retraining starts from the dazzled working place. Otherwise, when the user’s new wish belongs to the bases of two neighboring terms, the intensity of lighting reached after retraining in the working place is determined by the point of crossing of these terms.

An example of rules for lighting control generated experimentally according to the proposed algorithms under consideration and used before retraining (initial rules) and rules after retraining are presented in Fig. 15a, b correspondingly. In Fig. 15, J402 stands for a lighting sensor S in the working place WP1.

Fig. 15
figure 15

Initial rules (a), and rules after retraining (b)

Full size image

Experiment for self-training in case 1 was made for users ID1 and ID2 separately with the same coordinates: xID = 130 and yID = 130. In this case, all terms are known and they are as presented in Fig. 3. But up until now, the user has not expressed any wishes concerning concrete level of lighting in the working place. When modeling the self-training process, the lighting desired and set by user ID1 is 210 lx, and the one set by user ID2—240. Five rules were generated for ID1 and five for ID2. After the self-training process, lighting in the working place WP1 for ID1 and ID2 was set to 200 lx and 220 lx accordingly. The concrete type of terms “Dusk” and “Light” (Fig. 3b) predetermine a precision of system’s response as described above. Here it must also be emphasized that modeling was performed when user starts acting in a darkened working place.

Experiment for self-training in case 2, when only Min and Max values of parameters are known (Min x = 0, Max x = 1200, Min y = 0, Max y = 900, Min S = 0 and Max S = 1000), was performed in two steps: (1) User ID1 at coordinates xID1 = 130 and yID1 = 130 expresses his wish (SWish1) and sets lighting to 238 lx. Three terms for x, three terms for y, three terms for S, and three rules were generated. After the self-training process, and the ID1 x and y coordinates were set, lighting supplied by the system was 231 lx. 2) User ID2 at coordinates xID1 = 130 and yID1 = 130 expresses his wish (SWish2) and sets lighting to 320 lx. The fourth term for S and four rules were generated. After the self-training process, when ID1 or ID2 position in terms of x and y coordinates were set, lighting achieved by the system was 231 and 318 lx correspondingly.

An example of modeling results with generated S terms according to the proposed algorithm is presented in Fig. 16a. Here, J402 corresponds to the lighting S in WP1. Generated rules for lighting ID1 in WP1 are presented in Fig. 16b.

Fig. 16
figure 16

BIAsim screenshot for visualization of two terms Wish1 and Wish2 generation (a) and created rules (b)

Full size image

This modeling and simulation confirms the vitality of the whole idea. Including the newly proposed algorithms for fuzzy training, retraining, and self-training for the intellectualization of home environment has expanded SHE’s possibilities. Efficiency of top-down hierarchical analysis of home situations under consideration when conquering the curse of increasing number of rules is demonstrated.

7 Concluding Remarks

The main research ideology was developed and implemented under the guidance of the COST Action IC0702 “Combining Soft Computing Techniques and Statistical Methods to Improve Data Analysis Solutions (SOFTSTAT)”. This research was supported and used in the EU Structural Funds projects: VP1—3.1-ŠMM-08-K-01-018 “Research and Development of Internet Infrastructure for IoT&S in the Smart Environment (IDAPI)” and VP1-3.1-ŠMM-10-V-02-020 “Research on Smart Home Environment and Development of Intelligent Technologies (BIATech)”.

In this paper, the opinion that a set of electronically controlled smart things and the whole home environment must be intellectualized using human-type reasoning to increase person's comfort was advocated.

A novel approach and new algorithms for a hierarchical fuzzy training, retraining, and self-training of an intellectualized home environment are proposed. Fuzzy logic-based training algorithms take top-down hierarchical analysis of home situations under consideration to conquer the curse of increasing number of rules. A successful combination of crisp algorithms for users’ presents/absence identification in the environment and the fuzzy logic-based algorithms enables significantly to decrease the number of necessary rules. In the paper, a case is presented where 2500 rules were decreased approximately 5 times.

For the first time the changes of users’ wishes are taken into account during the retraining process.

An entirely new ability of the system was investigated: a fuzzy logic-based algorithm to start the self-training process without any a priori information is developed. Vitality and efficiency of the proposed methodology was tested and simulated on a specialized virtual software/hardware modeling system. The modeling performed for the working place light control in conditions close to real home environment confirmed the possible practical efficiency of the investigated methodology.

Investigations of the control speed and accuracy, automation of the inclusion procedure of new variables, solving the scalability problems can be considered as topics for further research and development.