基于信息融合智能家居控制系統(tǒng)和它的應(yīng)用-外文翻譯（英文）

1、L. Zhang et al.: Information Fusion Based Smart Home Control System and Its Application Contributed Paper Manuscript received June 25, 2008 0098 3063/08/$20.00 © 2008 I

2、EEE 1157Information Fusion Based Smart Home Control System and Its Application Lan Zhang, Henry Leung, and Keith C. C. Chan Abstract — This paper presents an information fusion based smart home control system. The pro

3、posed control framework includes an internet access and information acquisition module, in-house networking services with bluetooth wireless connectivity, an information fusion based controller using fuzzy logic and

4、fuzzy neural network, and embedded computational units in home appliances. It integrates information from multiple sources to control home appliances to create the smart home environment. The detailed description of

5、the proposed system, from design to implementation, is given in this paper1. Index Terms — Information Fusion, Fuzzy Logic, Fuzzy Neural Network, Smart Home I. INTRODUCTION Smart Home refers to a domestic environment t

6、hat applies knowledge to improve the quality of resident’s life by facilitating a flexible, comfortable, healthy and efficient environment. Currently, smart home systems are being extensively integrated with smart de

7、vices and enhanced services and can be found in aspects related to environment, security, entertainment, communication, assistive and healthcare systems, and electronic home appliances. However, existing designs

8、focus on developing intelligent applications limited to relatively simple remote monitoring or pre- programmed control [1 – 4]. Some of the existing challenges include the lack of a comprehensive infrastructure for sma

9、rt homes to integrate old appliances into a new design, the lack of ability to thoroughly perceive internal and external situations by combining multiple information streams. This paper presents a smart home design ba

10、sed on information fusion. The proposed design consists of an information acquisition module, a fusion based controller with multi-functional service interfaces, communication networks, and smart home appliances. The

11、 objective is to sense and gather information from multiple sources (sensors, internet etc.), combine the information and use the same to intelligently control internal and external environments. Fig.1 illustrates th

12、e proposed system architecture, where appliances are controlled through wired (power line) and/or wireless links in order to realize the idea of a smart home system. Lan Zhang and Henry Leung are with the Department of

13、 Electrical and Computer Engineering, University of Calgary, Calgary, AB, Canada. E-mail: {zlan , leungh} @ ucalgary.ca Keith C. C. Chan is the Head of the Department of Computing, Hong Kong Polytechnic University,

14、 Kowloon, Hong Kong. E-mail: cskcchan@inet.polyu.edu.hk). Fig. 1. System architecture of smart home control system The core of the proposed smart home system is an information fusion based controller. Fuzzy logic

15、(FL) is employed here to combine the heterogeneous source of information to execute an inference mechanism in terms of IF- THEN rules [5]. However, certain characteristics of fuzzy logic, such as the heuristic nature

16、of membership functions and the fuzzy rules defined by expert experiences limit the system performances in real world applications. To handle this issue, a fuzzy neural network (FNN) [6] is developed here as the fusi

17、on model with learning ability for the proposed smart home control. The proposed system is applied to control an alarm clock in order to generate a suitable alarm lead time. It acquires information from the internet

18、concerning the external factors such as the current weather and traffic conditions. Fusion algorithms based on FL and FNN are used to calculate and to determine a suitable alarm time according to the available extern

19、al information. Different scenarios are used in our simulation to evaluate the system performance. In this control application, the alarm lead-time generated by FNN is derived and compared to that generated by FL. The

20、 accuracy, reliability and stability of the FNN are indicative of the improved performance when compared to the FL model. Complete electronic design of the proposed system has been developed and a hardware implementa

21、tion of this smart alarm clock is reported in this paper. The paper is organized as follows. Section 2 presents the design of information fusion based smart home control system and functionalities of the individual

22、system components. The smart alarm clock and its electronic design are presented in Section 3. Following that, a thorough control performance under different scenarios is given in Section 4. Concluding remarks are gi

23、ven in Section 5. Alarm Clock Traffic internet Weather Local PCWireless ConnectionDatabase Power-line ConnectionSprinkler Physical Sensors Soil moisture L. Zhang et al.: Information Fusion Based Smart Home Control System

24、 and Its Application 1159variables to fuzzy subsets for corresponding universes of discourse. After trying various types of membership functions (MF), piecewise linear functions are adopted. The membership grad

25、e is given as, ( ) 2 2 1 2 2 1 ( ) ( ) ( ) ( ) a x x or x μ α α α α α α = ? ? ? ?(1) The fuzzy IF-THEN rules are given as, 1 1 1 2 1 11 1: ( ) ( ) ( ): ( ) ( ) ( ) n n n nRule IF X is A AND X is B THEN Y is CRule IF X

26、is A AND X is B THEN Y is CM M Mwhere the Zadeh-Mamdani fuzzy inference is employed for intelligent decision fusion. More precisely, the T-norm fuzzy operator performs AND instruction and is applied to the conjunction

27、 of the input values when the antecedent in a certain rule has more than one part. The T-conorm operator is employed to aggregate the outputs of all rules. Defuzzification is based on the centroid method. 3) Fuzzy Neu

28、ral Network Module For the system to have learning capability, the fuzzy neural network is considered here. The proposed fuzzy neural network is contrived as a five-layered network structure to implement the fuzzif

29、ication, fuzzy inference and defuzzification processes. The classic multilayer perceptrons (MLP), characterized by fuzzy inputs and outputs, and crisp weights, is used as forward propagation network in antecedent

30、rule and consequent layers to discover the association between the fuzzified input and output patterns of the system. Layer 1 is the input layer. The crisp input variables are passed through the fuzzification layer (l

31、ayer 2) where the current inputs are converted into the membership degrees with ij μ , 1,2,..., , j q =where ij μrefers to pre-defined membership function of p×q fuzzy subsets. Membership function ij μ that t

32、ransfers the input variable i x into fuzzified values is given as, ( )minmin min maxmax( )l jj j j j jh jw x xx w x x x x xw x xμ??(2) where ( ), 1,2,... j x j q μ = , max min 0, 1, l h j j j w w w x x = = = ? . Layer

33、 3 is the fuzzy IF-THEN rule layer and each neuron in layer 3 can be viewed as a fuzzy rule, which defined the dependences of fuzzified input and output of system. The outputs of the layer 2, all nodes of the layer 3,

34、 and the layer 4 can be considered as a standard multilayer perceptron (MLP) network and used to perform two-phase propagation-learning cycles. The weights of connections between layer 2 and layer 3 are trained using

35、 a back-propagation algorithm. A sigmoid function ( H β ) is used as the nonlinear transfer function and is given as, ( ) 1(1 ) H e β γ γ ? = +(3) Layer 4 is a consequent layer and it generates the fuzzified output. T

36、his estimated output is compared to the desired output and the mean deviation is input to the back-propagation algorithm, in order to modify the weights in layer 3. Until convergence, the unit implements the fuzzy max-

37、operator in order to aggregate the outputs from fuzzy rules with the same consequent. The output of each neuron can be therefore be given as, 1nk t kt t V H W X β =? ? = ? ? ? ? ∑(4) The cost function p E of back-p

38、ropagation algorithm is defined as ( )2 12d d d k k E T V = ? ∑(5) where d k T is the target output and d k V is the actual output. Layer 5 is the output layer where the outputs are defuzzified into the real world s

39、cale. A weighted sum approach is adopted for defuzzification and is given as, ( ) _ max _ min 1lout k k k k Y R R v=? ? = ? × ? ? ∑(6) where out Yis the numerical output of the proposed FNN, _ min k Rand _ max k

40、 Rare the lower and upper limits for kth range, k v is the fuzzified output of layer 4. C. Communications Home networks connect a wide range of electronic devices within a household for the sake of resource sharing

41、, communication, and remote control. In this system design, wireless (bluetooth) and wired networking (power-line) techniques are employed as the communication solutions between central control PC and smart home appl

42、iances. 1) bluetooth Communication To enable wireless communication, the smart home controller and home appliances are connected through a bluetooth device. Two bluetooth communication devices, bluetooth-enabled USB

43、 dongle and bluetooth serial port adapter (SPA), are employed here [13, 14]. The bluetooth connection interface is developed to establish the bluetooth connection between bluetooth serial port adapter and USB dongle.

44、 The bluetooth serial port adapter acts as a remote slave that powers up separately and is connected to the alarm clock module. The USB dongle detects and configures the parameters and settings of a serial port adapt

45、er through the bluetooth connection. The serial port adapter can then receive data from the USB dongle. 2) X10 Power-line Communication X10 power-line carrier technology is also exploited to transmit and receive sign

46、als between the smart home controller and appliances. It employs signal delivering at the zero crossing point to transmit and receive information [15]. Two X10 devices, a 2-way computer interface and a transceiver mo