深度類神經網路技術於辦公室照明系統之模擬與分析

Application of Deep Artificial Neural Networks to the Simulation and Analysis of Office Lighting System

黃亭雅(Ting-Ya Huang)

指導教授 ： 陳柏翰

國立臺灣大學/工學院/土木工程學研究所/碩士(2018年)

https://doi.org/10.6342/NTU201802024

摘要

本論文主要探討深度類神經網路（DNNs），也就是大家耳熟能詳的深度學習於辦公室照明設計之模擬與分析應用研究，在符合國內外辦公室照明法規的前提下，並同時考慮成本，提出一種合理的照明設計建議機制，以解決照明設計師、建築師、室內設計師等使用者在進行照明設計時，因燈具種類過多而造成效率不佳之問題。

以辦公室為主要研究案例，提供使用者只需輸入參數：空間參數、照度需求，就能應用本研究所提出之人工智慧預測模型得到：「燈具種類」、「燈具配置方式」、「燈具配置數量」及「燈具成本」，並同時預測照明四項重要指標：平均照度、眩光度、照度均勻度及用電密度，進而告之光環境之品質狀況。而建議會根據「照明品質最佳解」及「配置成本最佳解」分別提供兩種燈具種類之建議。

藉由Python並結合Keras進行程式撰寫，經由研究結果顯示，應用深度類神經網路之倒傳遞類神經網路，學習並整理於辦公室照明設計之輸入參數與目標之間的內在映射規則，以達成預測燈具配置方式及照明品質評估指標，並提出設計建議之目標具可行性。且此模型具延展性，未來可藉由匯入更多燈具資料，來擴充此模型之應用性。

關鍵字

辦公室照明設計 ； 深度類神經網路 ； 照明設計建議模型 ； 照明品質指標 ； 人工智慧

並列摘要

This thesis focuses on the deep artificial neural networks, which is familiar with deep learning in the simulation and analysis of office lighting design applications. Domestic and foreign office lighting regulations is under the premise while taking into account the cost, propose a reasonable lighting design recommend. It is the proposed mechanism to solve the problem of inefficiency caused by many types of lights when users of lighting designers, architects, interior designers, etc., perform lighting design.

Taking the office is as the main research case, even if users with only the input parameters: spatial parameters, illumination requirements, the mechanism can apply the artificial intelligence model proposed by this research to obtain: "light type", "light configuration method", "light configuration quantity”, and "cost of lighting", in the same time to predict four important indicators of lighting: average illumination, glare, uniformity and power density, and then report the quality of the light environment. The proposal will provide the recommendation about two types of lights that is based on the "Best Solutions for Lighting Quality" or "Best Solutions for Configuration Costs".

Using Python with Keras for program writing, the results show that deep neural network-based back-propagation neural network is applied to learn and organized the internal mapping rules between the input parameters and goals of office lighting design to predict the light configuration method and the lighting quality evaluation index and also proposing the feasibility of the design proposal. The application of the derivative model will be expanded by importing more fixture data in the future.

並列關鍵字

office lighting ； deep neural networks ； lighting design ； lighting quality index ； Artificial intelligence ； DIALux

參考文獻

外文文獻：

[1] Aste, N., Manfren, M., & Marenzi, G., “Building Automation and Control Systems and performance optimization: A framework for analysis. Renewable and Sustainable Energy Reviews”, 75, 313-330（2017）.

[2] Ahmad, A.S., Hassan, M.Y., Abdullah, M.P., Rahman, H.A., Hussin, F., Abdullah, H., Saidur, R., “A review on applications of ANN and SVM for building electrical energy consumption forecasting”, Renewable and Sustainable Energy Reviews, 33, 102–109（2014）.

[3] “ASHRAE 90.1”, Energy Standard for Buildings Except Low-Rise Residential Buildings（2017）.

[4] Balvís, E., Sampedro, Ó., Zaragoza, S., Paredes, A., & Michinel, H., “A simple model for automatic analysis and diagnosis of environmental thermal comfort in energy efficient buildings”, Applied Energy, 177, 60-70（2016）.