利用多變量類神經網路執行智慧建築物之照明控制

Abstract

由於近年來油電雙漲，喚起國人們的節能環保意識，舉例來說，許多家庭已經開始使用LED燈和購買節能產品。本研究利用室外光所提供之照度當作控制系統之干擾，在滿足量測點最小之照度需求下進而利用控制器關閉不必要之燈具，利用照度模擬軟體Tracepro模擬工作區域及整個房間之照度分佈，並使用最佳選擇控制與倒傳遞學習類神經方法，利用電腦之Matlab軟體及Simulink軟體將兩種控制方法實現，且有效的控制在實驗室裡的九個燈具開關，倒傳遞學習類神經方法並可以進一步的擴展到其他房間之不同燈具數量與不同量測點之燈具控制。 本研究建立的多變量控制系統包含十個輸入及九個輸出，這兩種控制方法能有效的讓我們節省能源，達成控制目標，在具有一個量測點下，最佳選擇控制方法可以節省約21.4 %的日常用電量，倒傳遞學習類神經方法則可以節省約42.0%的日常用電量；在具有兩個量測點下，倒傳遞學習類神經方法可節省約26.6%得日常用電量。把實驗室的傳統T5螢光燈管換成LED燈管時，並在具有兩個量測點下則可節省約14.0%的日常用電量，此外傳統T5螢光燈管在未控制前耗電量為每天5.712度，控制後為4.192度;LED燈管在未控制前耗電量為每天4.160度，控制後為3.593度。

Since the price of gas and utilities has increased in recent years, people have become more environmentally conscious. For example, many families have started to use LED lights and buy energy efficient products. This study provides two methods for setting up these control systems for intelligent buildings. In addition, intelligent buildings have also become more advanced to help save energy. This study analyzes the distribution of illumination in both an individual working place and a whole room with Tracepro software. Illumination provided by the outdoor light is a disturbance; as a result, the controller would adjust by turning off some unnecessary lamps. The two methods used are Best Choice control and back propagation learning neural network control methods. Each is used to control the nine lamps in our experiment room; moreover, the back propagation learning neural network control method can be also applied to other rooms with different number of lamps and different positions of measurement points. Controllers in this study are all constructed by a Matlab software and a Simulink software in computers and the multivariable control system contains ten inputs and nine outputs. These two control methods let us save energy efficiently. With one measurement point, the Best Choice control method can save about 21.4% of daily lighting energy use. In the same situation, the back propagation learning neural network control method can save about 42.0% of daily lighting energy use. Moreover, with two measurement points, the back propagation learning neural network control method can save about 26.6% of daily lighting energy use. After changing the traditional T5 fluorescent luminaires to LED luminaires in the experiment room, the back propagation learning neural network control method can save about 14.0% of daily lighting energy use at two measurement points. Furthermore, without control, T5 fluorescent luminaires consume about 5.712 Kw-h per day; LED luminaires consume about 4.160 Kw-h per day. However, with control, T5 fluorescent luminaires consume about 4.192 Kw-h per day; LED luminaires consume about 3.593 Kw-h per day.