AMBIENT AIR QUALITY MODELLING OF TROPOSPHERIC OZONE USING ARTIFICIAL NEURAL NETWORK (ANN) WITH FEED FORWARD BACK PROPAGATION ALGORITHM

• Main Author: AYESHA THOBARONY , ZENETH
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/31900
• Institution: Institut Teknologi Bandung

<p align="justify">Abstract: Predicting tropospheric ozone concentration is one of the important factors in air quality management plans. To provide predicted concentration, an Artificial Neural Network (ANN) model is developed in this research. Used in ANN study using Backpropagation Algorithm (BP) are widely used in <br /> <br /> environmental problem analysis which involves large data set, for example air pollution. The purpose of this research is to predict ozone concentration using ANN. The major capability of ANN is its ability to represent relationships both in linear and non linear fashion, and its characteristic to learn the relationship amongst variables directly on modeled data. BP algorithm in ANN can identify linier pattern from input-output variable with steps: 1) designing ANN model structure; 2) learning and training time data set; and 3) verification. Half hourly data from Air Quality Monitoring Station (AQMS) DKI 1 Thamrin which represents business district and DKI 3 Jagakarsa which represents residential area of 2-year period (2011-2012) were utilized. Input configuration consisted of ozone precursor concentration (NO, NO₂, CH₄, NMHC), meteorology (temperature, solar radiation, humidity, wind direction and wind speed) and both. Historical ozone concentration is utilized as target. Time lapse method was applied for precursor data. To accommodate the time required for photochemical reaction to take place for DKI 1 station, data set is divided into 3 parts of training, testing, and validation with ratio of 58%, 21%, and 21%, respectively. For DKI 3 station, data set is divided with ratio 59%, 20.5%, and 20.5%. Statistical parameters to measure the accuracy are regression, RMSE (Regression Mean Square Error) and MBE (Mean Bias Error). To obtain as indicator of ANN parameter input in yielding an optimal model. The most optimal ANN model for DKI 1 station is the combination input of precursor and meteorological parameter with R = 0.73, while most optimal ANN model for DKI 3 station is used meteorological input with R = 0.81. The study shows that ANN can predict average peak diurnal O₃ <br /> <br /> concentration, indicating by its abilty to produce similar diurnal pattern with monitoring data..<p align="justify">