Many recent studies have focused on various aspects of Intelligent Spaces (IS) [1–3]. The IS was proposed by Lee [2] as an environmental system that is able to support humans in informative and physical ways, and as a space that contains human and artificial systems. Thus, an IS could utilize computer monitors to provide information to humans, and robots could be used to provide physical services to humans as physical agents.

Over the past few years, several definitions of ISs have been proposed; it is important to acknowledge the definitions that were presented in [4,5], in which ISs are described as spaces with many embedded and networked sensors and actuators. Their essential functions are the following: (a) to observe the space using distributed sensors; (b) to extract useful information from the obtained data; and (c) to fuse the information acquired by each sensor and to share it with other devices efficiently. Taking into account these properties, several studies [6–8] have been devoted to set up ISs over WSNs.

Over the past few years, Wireless Sensor Networks (WSNs) [9,10] have received the attention of many researchers. WSNs are defined [11] as networks that are composed of a large number of sensor nodes, and they can be conceived of as small computers with extremely basic interfaces and components. Each node consists of a processing unit with limited computational capability and memory, sensors, a communication device and a limited power source, usually in the form of a battery.

WSNs provide perfect scenarios for sensor collaboration for a global purpose. In this context, the term collaborative involves communication of individual measurements among sensors to accomplish their tasks and to achieve a common goal.

Despite the wide interest in sensors and collaborative WSNs, little attention has been paid to optimizing the integration of collaborative Fuzzy Rule-Based Systems (FRBSs) into WSNs. FRBSs [12] are considered an extension of classical rule-based systems because they deal with “IF-THEN” rules whose antecedents and consequents are composed of fuzzy logic statements (fuzzy rules) [13] instead of classical logic statements. One of the main characteristics of these integrated systems is the capacity of each node to incorporate facilities for simple sensing, actuation, communication and computation; however, the full capabilities of such networks are reached only by the cooperation of all of the nodes. Because of these qualities, these networks can efficiently provide all of the services necessary to implement an IS.

To optimize the implementation of an IS by means of a WSN, this paper presents a distributed architecture proposal for collaborative FRBSs embedded in a WSN. The main contribution of the design that is presented is to make it possible to decrease the consumption of the required resources [(a) memory; (b) CPU; and (c) battery] by an FRBS embedded in a node of a WSN, without a substantial decrease in the accuracy of the inferred values.

To achieve this goal, we include the following in the proposed architecture:

A design of the inference engine that is based on the approximate Mamdani-type [12], to achieve a high level of accuracy in complex problem domains.

The remainder of this paper is organized as follows. Section 2 addresses related work and motivation. Section 3 shows the modular architecture proposed to optimize the integration of collaborative FRBSs embedded in a WSN, which includes a description of the following: an inference engine design, visual interface facilities, algorithms proposed to make rule reduction in KBs and to evaluate KB performance, and the facilities that incorporate by a module of transformation of the basis of the rules and the communications protocol. Section 4 shows the experimental results that are obtained to evaluate the performances of the proposed architecture applied to the problem of modeling the two plagues of the olive tree: prays and repilo. Finally, conclusions are drawn in Section 5.

Lee proposed the concept of the IS as an environmental system able to support humans in informative and physical ways and to contain human and artificial systems. These human and artificial systems become clients of the IS and, simultaneously, the artificial systems become agents of the IS. Because the whole space is an intelligent system, the IS, as a spatial system, is able to easily monitor and provide services to clients. Specific tasks that cannot be achieved by the IS alone are accomplished by utilizing its clients [2].

Investigators [1] have proposed that this research field is not necessarily closely related to robotics; however, they believe that robots under ISs have many interesting features. Thus, an IS could utilize computer monitors to provide information to humans, and robots could be used to provide physical services to humans acting as physical agents. If necessary, robots as well as humans are supported by an IS. When a robot lacks the sensors to navigate around an IS, the robot is treated as a client of the IS and the information lacked is provided to the robot by the IS.

As described in [2], an IS has two roles with respect to a robot working in it. One role is the enlargement of ability, and the other role is resource sharing. Thus, an IS has a role as an extended sensor for the robot and enhances the ability of the robot to receive a request from a distant location. Further resource sharing occurs when more than one robot uses the resources of an IS, and robots can thus decrease their common resources. However, an IS does not aim to dispense with sensors or robots’ autonomy; rather, it supports a robot by providing the resources it lacks to act as a normal robot, while helping a robot with adequate resources to act as an even better robot.

Intelligent environments were described in [3] as complex systems (i.e., rooms) in which humans and machines collaborate to accomplish a task. From such a perspective, intelligent environments can also be considered as a novel human-machine interface, and the overall goal of the research will be to design and develop integrated sensor-based systems that allow natural and efficient mechanisms for human-computer interactions in places where humans work, learn, relax, and play.

To have a well-accepted definition of an IS, Mohan proposed four requirements that a physical space needs to possess in order to be called intelligent:

ISs are designed for humans, and they should facilitate normal human activities taking place in these spaces.

In [4], smart environments that are also called ISs are described as spaces with many embedded and networked sensors and actuators, where the essential functions of these smart environments are the following:

To observe the space using distributed sensors.

Other properties of the ISs are described in [4,5] as the following:

The system must have a function that fuses the information acquired by each sensor and that shares it with other devices efficiently.

In this proposed concept of an IS, sensor nodes are also distributed in the space because it is necessary to reduce the network load in a large-scale network, and it can be built to process raw data at each sensor node before collecting the information. To realize detailed observation over the whole space, it is necessary for networked sensors to cooperate with each other.

The field of research on ISs that are based on distributed sensor systems has been expanding recently [14], and many researchers have developed designs of systems for providing informative services to end users (e.g., support during a meeting [15], health care [16], support of the elderly [17], information display using a pan-tilt projector [18], support of human activities [19] and robot localization [20,21]).

On the other hand, WSNs are defined [9] as networks that are composed of a large number of sensor nodes and that can be conceived of as small computers with extremely basic interfaces and components. Each node consists of a processing unit with limited computational capability and memory, sensors, a communication device and a limited power source, usually in the form of a battery.

These properties of the sensor nodes of the WSNs allow them to take advantage of their process and communications capacities to implement the main characteristics of an IS, which are the following: to act as a server or client and to provide services to the clients; to fuse information; to share information and resources; to capture and dynamically maintain an awareness of the events; to be responsive to specific events and triggers; to be robust and adaptive to dynamic changes; and to have flexibility and scalability. On the other hand, the capacities of WSNs enable them to implement the distribution of a control algorithm of a rule-based system [22].

Because of these qualities, several studies have recently been devoted to set up ISs over WSNs, such as the following:

In [6], an implementation example for setting up an IS concept is described; this method is implemented with several communications technologies, including WSNs. As indicated in [6], WSNs are the foothold of the future generation of IS systems. They serve as acquisition tools in ambient circumstances to replace or upgrade users’ experiences. Successful ambient adaptation leads to automatic execution of users’ activities and needs.

As has been described above, an IS implementation, at minimum, requires the system to have a function that fuses the information acquired by each sensor and shares it with other devices efficiently. However, none of the schemes presented above describes the algorithm used to perform the tasks of information fusion and sharing.

The study in [23] shows how FRBSs can control artificial and natural lighting automatically, in an intelligent residential lighting control system based on a ZigBee wireless sensor network.

To control artificial and natural lighting automatically, an intelligent residential lighting control system with a ZigBee wireless sensor network is used to collect the environmental data and transmit the data to a PC, where a fuzzy system is executed to generate the lighting control signal, which is transmitted through the WSN. The execution of these tasks requires a significant amount of resources in terms of wireless communications’ bandwidth and battery consumption.

To improve the above-mentioned problem, a choice to take into account is the use of a fuzzy rule-based collaborative and cooperative approach.

Next, the definitions of the terms collaborative and cooperative are presented:

The term collaborative involves communication of individual measures among sensors to accomplish their tasks and to achieve a common goal [24].

Therefore, the integration of FRBS into WSNs is typical in examples of such collaborative and cooperating objects. These networks consist of objects that are individually capable of simple sensing, actuation, communication and computation, but the full capabilities of such networks are reached only by the cooperation of all of these objects. Because of these qualities, these networks can efficiently provide all of the services necessary to implement an IS. To the best of our knowledge, two collaborative and cooperative approaches that integrate fuzzy rule-based systems into WSNs have been presented:

The approach titled D-FLER has been presented in [27]. This system incorporates distributed and embedded collaborative mechanisms for reasoning about the observed data and making decisions or taking actions in a coordinated manner. It uses two types of inputs: individual observations (sensor readings of the current node) and neighborhood observations (fuzzified sensor data from the neighboring nodes). However, D-FLER has some limitations, derived from the need to transmit a significant amount of data to neighboring nodes and to process a significant amount of data.

In addition, this last proposal can achieve these mentioned advantages with a decrease in the amount of data to process and transmit.

Some scenarios that could benefit from the proposed approach are ISs that perform tasks of controlling or modeling in which sensor inferences may be affected by measures or inferences of neighboring sensors, e.g., event detections (fire detection, alarms), environmental monitoring (water quality, pest detection), industrial process control, and robotics.

However, these proposals also present some drawbacks associated with the following limitations:

The scheme does not include a human-machine interface that makes it easy to obtain the necessary knowledge and the visualization of data.

To make it easy and efficient to develop and set up an IS with WSNs and embedded FRBSs, in this work we propose the use of the following:

A fuzzy rule-based collaborative and a cooperative approach.

On the other hand, the use of an interface makes this work easy but does not guarantee an optimal KB generator. In addition, this new scheme introduces some additional modules proposed to optimize these KBs by means of rule redundancy reductions and rule adaptations to specific modeling problems.

A modular architecture has been proposed to implement the following tasks, which were mentioned above: KB construction, rule reduction, rule adaptation, KB transmission and obtaining the operations results.

This architecture (Figure 1) provides the following: (1) In the Personal Computer: (a) a visual interface for easy fuzzy rule development; (b) a module for rule reduction to optimize the generated KB; (c) a module for the analysis of error, to evaluate the accuracy reduction; (d) a module for rule transformation that facilitates the adaptation of rules to optimize the work of the FRBS inference engine; and (e) the communications protocol, which is being designed to manage multipurpose sensor networks, to transmit this KB to the desired nodes and to receive data from the sensor nodes. (2) In the sensor node: the basic elements of the FRBS (Figure 2).

To illustrate the proposed methodology of rule reduction and the evaluation of performances, three experiments have been performed to model, with simple KBs, the behavior of two plagues of the olive tree: prays (Prays oleae Bern.) and repilo (Spilocaea oleagina). The life cycles of these plagues depend on the humidity, temperature and photoperiod, but they are not the same [51]. The common steps of each experiment consist of the following:

KB generation in the visual interface.

Each experiment has a specific role. In Experiment 1, an approximate KB is composed of an RB with two groups of rules (GR) (Tables 1 and 2) and their associated specific fuzzy set (FS) definitions (Figures 3 and 4).

In Experiment 2, an approximate KB is composed of a RB with two GR (Tables 3 and 4) and the associated specific FS definitions (Figures 5 and 6). In this case, the rules and FS definitions have been obtained using the redundancy reduction module. As can be observed, this algorithm decreases twelve rules for prays RB and eleven rules for the repilo KB (Tables 3 and 4) and generates, by means of fusion, new FSs (six for prays and four for repilo) (Figures 5 and 6).

With the purpose of illustrating the rule reduction process, an example of reduction is shown. In the GR used to model prays (Table 1), there is a set of three rules with the same consequent and the same second and third proposition in their antecedents:

R1: If Humidity is S and Temperature is VS and Photoperiod is S Then Prays Alert is VS

Applying the proposed algorithm, these original rules can be simplified into a single rule (New R1).

New R1: If Humidity is S&M&L and Temperature is VS and Photoperiod is S Then Prays Alert is VS

Except for the first proposition in the antecedent, the rule “New R1” is equal to the three originals. Now, the FS associated with the “Humidity” variable is “S&M&L” (Figure 5), which has been obtained by means of composition of the three original FS (Figure 3).

In Experiment 3, a descriptive KB is composed of an RB with two GRs (Tables 1 and 2) and the definition of linguistic labels associated with each linguistic variable (Figure 7). In this case, the use of linguistic labels prevents specific FS definitions for each GR.

To evaluate the accuracy of the KBs used in the second and third experiments, we calculate the parameters AE, SAE and RE, which are explained in Section 3.2.3. In this case, the calculation of these parameters has been performed by taking into account that System 1 is an FRBS with the original approximate KB and System 2 is an FRBS with the following:

A descriptive KB, if we are considering the second experiment.

To show the results of this evaluation, we have obtained the following indicators:

(a) The “Sum of the Absolute Errors” (SAE) and the “Relative Error” (RE) have been calculated considering a wide set of system states (400,000) in the space of the photoperiod, temperature and humidity variables.

To calculate these parameters, we have used Equations (2) and (3), taking into account that we now have three input variables: photoperiod, temperature and humidity and L₁ = 100, L₂ = 40, L₃ = 100.

Considering that the proposed FRBS is used to model the behavior of two plagues (prays and repilo) of the olive tree, their global SAE must be calculated as the sum of the SAEs for prays and repilo. (4) SAE Global = SAE Prays + SAE Repilo

In the same way, the global RE must be calculated as the average of the REs for prays and repilo. (5) RE Global = ( RE Prays + RE Repilo ) 2

In Tables 5 and 6, we can see the values taken by these accuracy parameters.

As can be observed in Table 6, in the case of modeling with a reduced approximate KB, the parameters RE_Glogal and SAE_Global take low values (relatively close to zero), mainly if they are compared with their values in the case of modeling with a descriptive KB (see Table 5), so much so that they are more than fifty nine times greater. Therefore, the use of the reduced approximate KB entails a small decrease in the accuracy of the modeling process, while the use of the descriptive KB entails an unacceptable degradation of the accuracy.

(b). In each of the second and third experiments, we calculated 400,000 values of the AE parameter. Because this number is too large to be shown in a table, to be informative about the values of this parameter, we present the following:

First is the evolution of the sum of the AE for each state of a wide set of values (100) in the space of the photoperiod variable:

In the case of modeling prays.

Second, for some values of the photoperiod variable, the evolution of the AE for a wide set of states (4,000) in the space of temperature and humidity variables are shown:

In the case of modeling prays.

As can be observed in the previous figures, the use of the reduced approximate KB proposed entails a small decrease in the accuracy, which is limited to small regions of the input variable space (see Figures 9, 11, 13 and 15); otherwise, the use of the descriptive KB proposed entails a highest decrease in the accuracy that is widespread on extensive regions of the input variable space (see Figures 8, 10, 12 and 14). To evaluate the effect on battery consumption, two sets of measurements have been performed over each one of the FRBS proposed in the three experiments:

The run time necessary to execute 400,000 continued inferences. The results obtained (Table 7) show that the use of the approximate reduced KB versus the original KB saves approximately 25.64% in battery consumption. These rule reductions have been obtained without an appreciable lack of accuracy in modeling (RE = 0.47%).

The results obtained (Table 8) show that the use of the approximate reduced KB versus the original KB makes it possible to have an increase of 50% in the number of inferences.

From the analysis of the experimental results obtained, we notice that it has been possible to compose an optimized KB, adapted to model olive plagues, which generates a decrease in the FRBS run time (an increase in the inference rate) and a decrease in battery consumption (a decrease in the time of discharge in the battery sensor), without substantially decreasing the accuracy.

This work has presented a distributed architecture proposal for collaborative FRBSs embedded in WSNs, which has been designed to optimize the implementation of IS by means of WSNs. This architecture includes the following:

A design of the inference engine adapted to achieve a high level of accuracy by means of the use of an approximate KB.

This architecture, the proposed methodologies of rule reduction and the evaluation of performances have been applied to the problem of modeling two plagues of the olive tree: the Prays and the Repilo.

The results have shown the effectiveness of the proposed structure for optimizing the execution of an FRBS embedded into a sensor. In this sense, we can observe that the use of a reduced approximate KB, obtained according to the proposed architecture, entails the following:

A small decrease in the accuracy of the modeling process, and this decrease is limited to small regions of the input variable space.

These results suggest that the architecture presented in this paper allows a decrease in the consumption of the resources required (memory, CPU and battery) by an FRBS embedded in a node of a WSN, without a substantial decrease in the accuracy of the inferred values.