This learning rule is based on gradient descent. The connection weights between different layers are initialized with pseudo-random values and are changed in a direction that will reduce the error. When a learning pattern is clamped, the activation values are propagated to the output units, and the actual network output is compared with the desired output values.
Published in Chapter:
Artificial Neural Network Modelling of Sequencing Batch Reactor Performance
Eldon R. Rene (University of La Coruña, Spain), Sung Joo Kim (University of Ulsan, South Korea), Dae Hee Lee (University of Ulsan, South Korea), Woo Bong Je (University of Ulsan, South Korea), Mirian Estefanía López (University of La Coruña, Spain), and Hung Suck Park (University of Ulsan, South Korea)
Copyright: © 2012
|Pages: 24
DOI: 10.4018/978-1-61350-116-0.ch019
Abstract
Sequencing batch reactor (SBR) is a versatile, eco-friendly, and cost-saving process for the biological treatment of nutrient-rich wastewater, at varying loading rates. The performance of a laboratory-scale SBR was monitored to ascertain the chemical oxygen demand (COD) and total nitrogen (T-N) removals under four different operating conditions, by varying the operating time for the nitrification/denitrification steps, i.e., the cycle times. A multi-layered neural network was developed using COD, T-N, carbon to nitrogen ratio (C/N), aeration time, and mixed liquor suspended solids concentration (MLSS) data. This chapter compares the neural simulation results to the experimental results and extracts information on the significant factors affecting SBR performance. The application of artificial neural networks to biological processes such as SBR is a relatively new technique in wastewater and water quality management, and the results presented herein indicate the promising start of the adoption of computational science in this domain of research.