Smart Control of Electric Lamp using Artificial Intelligence based Controller Anupam Purwar

Divya Joshi

Mani Sankar Dasgupta

Visiting Scientist Department of Aerospace Engineering, Indian Institute of Science, Bangalore, India [email protected]

Student Member IEEE Department of Electrical & Electronics Engineering Birla Institute of Technology and Science, Pilani, India [email protected]

Professor Department of Mechanical Engineering Birla Institute of Technology and Science, Pilani, India [email protected]

Abstract— There is need and scope of controlling the power consumed by conventional electric lamps in presence of some natural light. An artificial intelligence based control system has been developed to control a lamp dimmer circuit with bidirectional triode thyristor. The light present in the room is sensed and voltage supplied to the lamp is controlled by varying the time constant of the circuit through change of resistance of a multi-turn potentiometer with a stepper motor. The resistance set in the lamp dimmer circuit is in accordance with signals from an adaptive Neural Network running in MATLAB®. The artificial neural network (ANN) runs in real time in MATLAB® environment to control the time constant of the lamp dimmer circuit for controlling of power consumption. Under test conditions, energy savings up to 35% is achieved. Keywords—adaptive control; energy saving; lamp dimmer; target illuminance; artificial neural network; artificial intelligence.

I.

INTRODUCTION

According to International Energy Agency, 19% of all electricity produced is consumed in lighting [1]. The ever growing lighting demand necessitates development of power efficient devices. Some researchers attempted intelligent switching of street lighting system based on localized conditions [2, 3], recently work on dimmable outdoor lighting system is also reported [4, 5]. Some researchers focused on control of lamps for energy efficiency, for street lighting the focus mainly were on ballast based control of sodium vapor lamps [6] or Metal-halide lamps [7]. Comparative study on LED ballasts for energy efficiency is also reported [8]. For intelligent stage lighting, stepper motor controlled lighting system with ARM based processor is reported [9] but this system does not attempt energy saving. Wojnicki et. al. [10] have proposed a conceptual frame work of Artificial Intelligence (AI) based control system. But development of an AI-based control system to reduce power consumption in lamps and quantification of the same using real time demonstration has not been reported. This paper presents the design, development and demonstration of a novel AI-based control system for reducing power consumption in electric lamps. Artificial intelligence systems have demonstrated to have high level decision making

and learning capabilities similar to human beings, which cannot be programmed into conventional systems. This provides higher flexibility and adaptability in AI based control systems [11]. Hence, an AI based control system has been designed to control the power consumed by lamp. The power required by a lamp for a target illuminance level at a location changes as naturally available light changes over time. Thus, there is a scope to reduce electricity consumed by a lamp. Natural lighting already available in a location of a room is used as reference to achieve target illuminance. The AI based control system has been developed and tested with a setup consisting of a lux meter (input unit), microcontroller (interfacing unit), lamp dimmer alternating current circuit, stepper motor (illuminance control element) and trained artificial neural network in MATLAB® environment (processing unit) has been devised. To evaluate the effectiveness of proposed system, comparison of power consumption of the lamp operated with the proposed control system is made with an ordinary lamp. II.

DESIGN OF CONTROL SYSTEM

Nonlinear processes are difficult to control because of multiple variations of the nonlinear behaviour. The issue becomes more complicated if a nonlinear characteristic, here the light availability and corresponding artificial illumination required in room changes with time during a day. So there is a need for an adaptive control of power consumption in lamps. Adaptive systems commonly employ methods of computational intelligence such as artificial neural networks (ANN) [12][13].

Fig.1. Lamp dimmer circuit

978-1-4673-6540-6/15/$31.00 ©2015 IEEE

Fig.3. Relation between Illuminance and Resistance

calculates the final output value. The number of neurons in hidden layer of network has been optimized to keep the neural network simple and computationally fast [15][16]. A. ANN 1 The ANN-1 is designed and trained using data obtained experimentally as shown in Fig.8. In ANN-1 the input layer is taking the supply voltage reading, and output layer gives lamp dimmer circuit resistance R as output. The architecture of this ANN is shown in Fig. 2. This ANN does not form part of the actual control system, but only relates voltage V, across lamp with the resistance, R of lamp dimmer circuit.

Fig.2. Architecture of ANN1

The design goals of this ANN based control system are accurate modelling of the non-linear process and the constraint is ease of implementation on a basic micro controller to minimize hardware costs. Another objective is testing of this control system with a lamp dimmer circuit. A lamp dimmer circuit [14] consisting of an incandescent bulb has been selected as shown in Fig.1. For this circuit, the voltage supplied or power fed to the lamp is controlled by varying the Resistance R, which in turn changes the triac T firing angle and changes the voltage supplied to lamp. Hence, the control system has to control the value of resistance R based on the light requirement. In first step, rigorous experiments have been carried out to get data of natural light availability at different times during a day and corresponding Resistance R required, then this data has been used to train the ANN. Different number of hidden layers and neurons have been used to accurately model this nonlinear process, then complexity of ANN has been minimized for easy implementation on a low cost microcontroller. After rigorous optimization, a control system with feed-forward ANN consisting of sigmoid hidden neurons and linear output neurons has been selected. The presented neural networks ANN-1 and ANN-2 consists of three layers: Input Layer, Hidden layer and Output layer. In both the neural networks input is passed to four neurons present in hidden layer, the hidden layer uses tan sigmoid as the activation function and passes the values to output layer where the outputs from four neurons are aggregated and fed to a pure linear activation function, which

B. ANN2 The data required for training this artificial neural network has been generated in three steps. First, relation between illuminance output of lamp and voltage supplied is established experimentally as shown in Fig.7. Then, resistance, R of the circuit for a required voltage, V supplied to lamp has been found from ANN1. This provides the data relating deficient illuminance L (Lux) and R (kΩ) which is used to train ANN-2 as shown in Fig.3. In ANN 2, the input layer is taking lux reading of required light intensity and gives lamp dimmer circuit resistance as output. The architecture of ANN-2 is shown in Fig. 4. Hence, this neural network calculates lamp dimmer circuit resistance for a required light intensity level in real time.

Fig.4. Architecture of ANN2

220 Volt supply to the lamp. A lamp dimmer circuit is constructed as shown in Fig. 1. Supply is 220 Volt 50 Hertz Alternating Current, L is 60 watt incandescent lamp, R is 0 to 35 kilo ohm multi turn round potentiometer, C is 1 micro F capacitor, D is 30V 1-2 Ampere Diac and T is 30V 5 Ampere Triac. The voltage V across the lamp and corresponding luminance reading in Lux at the target location has been recorded as shown in Fig.7. Similarly by varying resistance R, the corresponding output Voltage, V across the lamp is recorded as shown in Fig. 8. IV.

Fig.5. Curve Fit :Training performance of ANN1

IMPLEMENTATION

The developed control system has been implemented on Advance RISC Machine (ARM) based microcontroller, for interfacing the final control element, stepper motor and input light sensor with a computer. The arrangement is shown in block diagram in Fig 9. Light intensity L measured by lux meter is input to the microcontroller, which is passed to control system running in MATLAB[18] environment on computer. The output from control system is sent to stepper motor for setting particular resistance R in the multi-turn potentiometer of the lamp dimmer circuit. Thus, the ANN2 running in real time reads the deficient light intensity L and gives the corresponding value of R to be set for a target level of illumination. V.

Fig.6. Curve Fit :Training performance of ANN2

DISCUSSION

A modular design approach has been used to develop this AI based control step, the three steps in the process of construction of the ANN-2 essentially provide this modularity. The Illuminance(Lux) vs. Voltage plot (Fig. 7) is dependent upon place, light source and source to target distance; which are local issues. The Voltage vs. Resistance plot (Fig. 8) is dependent upon the lamp dimmer circuit design. ANN-1 provides the necessary method to find data relating

C. Training of Artificial Neural Network The ANN-1 (Fig. 2) and ANN-2 (Fig. 4) have been trained using Levenberg Marquardt Back Propagation Method (LMBPM) owing to its higher efficiency as compared to other training algorithms[17]. The training is carried out multiple times till a satisfactory fit with mean square error of the order of 10-2 is achieved. The accuracy of the fit obtained using ANN-1 and ANN-2 is shown in Fig. 5 and Fig. 6 respectively. III.

EXPERIMENTATION

A. Experimental Setup Variation of illuminance at a location is modeled by varying the voltage supply to a 60 watt incandescent lamp using a variac in a dark room, and the light intensity is recorded with a digital lux meter kept at a fixed distance from the bulb. The distance between lamp and lux meter is adjusted in a way that the sensitivity range of the lux meter is well utilized and accordingly luminance measured by lux meter is 250 lx for

Fig.7. Illuminance vs. Voltage

Fig.8. Voltage vs. Resistance in lamp dimmer circuit

LIGHT SENSOR (INPUT DEVICE) 0-5 V ANALOG SIGNAL

ARM 32 BIT MICRO CONTROLLER NXP LPC 1768

PC WITH ANN CONTROLLER RUNNING IN MATLAB

TIME CONSTANT OF LAMP DIMMER CIRCUIT

STEPPER MOTOR DC 12V, 0.1 Amp

DC 12 V SUPPLY

Fig.9. Block diagram of control system implementation

Deficient illuminance in room (Lux)

is made about the power saving with the proposed adaptive controller for lamp against a normal lamp without controller. Using another 60 Watt lamp fitted in series with a variac as an external source, the light intensity at the measuring point in the dark room is varied from 0 to 250 lux in 25 steps, maintaining each light intensity level for a constant interval of time. During the same, the 60 Watt lamp with adaptive controller is kept active and the power consumption is recorded. This roughly simulates a day long situation with variation in available natural lighting. Overall for the full duration of the experiment 34.55% power saving is recorded as shown in Fig. 10. Power consumed by stepper motor being less than 1.2 Watt is neglected. Thus, this work demonstrates valuable saving of electricity with the developed control system. VI.

CONCLUSION

In this work an AI-based control system has been designed and developed to reduce electricity consumption in electric lamp. The same has been implemented using a microcontroller and ANN running in MATLAB. Then, an experiment is carried out to quantify the effectiveness of this ANN- based controller in reducing electricity consumption. In this experiment, natural light availability is varied artificially and electricity saving of about 35 percent has been observed. Thus, establishing the capability of AIbased control system. The ANN can be retrained with different set of data for different target lighting conditions and hence this provides a promising solution to manage power consumed in lamps installed in places like airport, factory, long corridors etc. This smart control scheme can also be used for dimming CFLs and LED lamps as well by modifying the ANN according to their dimmer circuit and thus aid in reducing electricity consumption in all types of lamps. With similar logic, the work can also be extended to other Alternating current devices like air conditioners, heating devices etc . REFERENCES [1]

Fig.10. Graph showing power consumed by lamp (in W) vs. light intensity required (in Lux) in the room [2]

resistance, R for a required voltage. But the control system consists of only ANN-2, which runs in real time to control lamp dimmer circuit resistance, R according to ambient light measured by light sensor. Dimmers have been implemented in various illumination control applications but they don't have fine control over dimming. The motivation behind employing artificial neural network is to obtain smart, smooth and adaptive control. The entire control system is also affordable, whose major cost consists of microcontroller available within INR 600. The cost of entire system does not exceed INR 950, which is cheaper than a large wattage dimmer. A comparative study

[3]

[4]

[5]

World on the edge – Energy Data Efficiency 2010. Available at www.earth-policy.org/datacenter/xls/book_pb4_ch4-5_4.xls (accessed on June 12, 2015) Tsang, P.W.M.; Wang, R.W.C.; Development of a distributive lighting control system using Local Operating Network, Consumer Electronics, IEEE Transactions on , vol.40, no.4, pp.879-889, Nov 1994 De Dominicis, C.M.; Flammini, A.; Sisinni, E.; Fasanotti, L.; Floreani, F.; On the development of a wireless self localizing streetlight monitoring system. Sensors Applications Symposium (SAS), 2011 IEEE, pp.233-238, 22-24 Feb. 2011 Zotos, N.; Stergiopoulos, C.; Anastasopoulos, K.; Bogdos, G.; Pallis, E.; Skianis, C.; Case study of a dimmable outdoor lighting system with intelligent management and remote control, Telecommunications and Multimedia (TEMU), 2012 International Conference on , vol., no., pp.43-48, July 30 2012-Aug. 1 2012 Mircea P. and Costin C., Energy Consumption Saving Solutions Based On Intelligent Street Lighting Control System, U.P.B. Sci. Bull., Series C, Vol. 73, Iss. 4, PP 297-308, 2011

[6]

[7]

[8]

[9]

[10] [11]

[12]

[13]

[14] [15] [16] [17] [18]

Van Tichelen, P.; Weyen, D.; Geens, R.; Lodeweyckx, J.; Heremans, G.; ,"A novel dimmable electronic ballast for street lighting with HPS lamps," industry Applications Conference, 2000. Conference Record of the 2000 IEEE , vol.5, no., pp.3419-3422 vol.5, 2000 Stockwald, K.; Kaestle, H.; Weiss, H.; Significant efficacy enhancement of low wattage metal halide hid lamp systems by acoustically induced convection configuration, IEEE 35th International Conference on Plasma Science, 2008. ICOPS 2008. Galkin, I.; Milashevski, I.; Teteryonok, O.; Comparative Estimation of Efficiency of LED Dimmers at Different Modulation Techniques, Power Electronics and Applications (EPE 2011), proceedings of the 2011-14th European Conference Aug. 30 2011 Hui Ren; Chaohui Lu; Zhibin Su; Jingjing Zhang; , "Stepper Motor Drive of Stage Intelligent Light Based on Embedded," Measuring Technology and Mechatronics Automation (ICMTMA), 2010 International Conference on , vol.3, no., pp.489-492, 13-14 March 2010 Wojnicki et. al, Towards AI-based Lighting Control Systems Passino and Antsaklis ,A System and Control Theoretic perspective on Artificial Intelligence Planning Systems, Source: http://www3.nd.edu/~pantsakl/Publications/57-AAI89.pdf (accessed on June 12, 2015) User-Adaptive and Other Smart Adaptive Systems:Possible Synergies, http://dfki.de/~jameson/pdf/eunite01.jameson.pdf, (accessed on April 18, 2014) Wilamowski B. M., Neural Networks and Fuzzy Systems for Nonlinear Applications, Proceedings of 11th International Conference on Intelligent Engineering Systems, (2007) 13-15 Lamp dimmer circuit, Available at http://www.circuitstoday.com/diacapplications (accessed on August 18, 2014) SN Sivanadam N. S., Sumathi S., Deepa S. N., Introduction to Neural Networks using MATLAB 6.0, Tata McGraw Hill (2006) Rajasekaran S., Pai G. A. Vijaylakshmi, Neural Networks, Fuzzy Logic and Genetic Algorithms, PHI Learning Pvt. Ltd, (2004) Hagan and Menhaj, Training Feedforward Networks with Marquardt Algorithm, IEEE Transactions on Neural networks, 1994 Math Works Inc, Matlab Documentation and Help, Available at http://www.mathworks.com (accessed on October 18, 2011)