Water level forecasting using feed forward neural networks optimized by African Buffalo Algorithm (ABO)

• 第一著者: Ahmed, Ehab Ali
• フォーマット: 学位論文
• 言語: English
• 出版事項: 2019
• 主題: QA75 Electronic computers. Computer science
• オンライン･アクセス: http://umpir.ump.edu.my/id/eprint/29019/1/Water%20level%20forecasting%20using%20feed%20forward%20neural%20networks%20optimized%20by%20african%20buffalo%20algorithm.pdf; http://umpir.ump.edu.my/id/eprint/29019/

Water is an essential requirement for human life and activities associated with industries and agriculture. An accurate forecasting model would be helpful in providing a warning of impending flood during the flooding time and assist in regulating reservoir outflows during the low flows. This reason motivated the researchers to exploit the evolution of machine learning to develop water level forecasting systems that were characterized by accuracy, simplicity and low cost. This development goal is to reduce the impact of water variation in river water levels. The machine learning applications, especially Feed forward neural network (FFNN) which inspired from the human biological nervous system have been successful in solving several complex problems. The FFNN training process which is an optimization task to find the optimal controlling parameters (weights and biases) is considered as the main issues in any model performance. Due to that, many algorithms employ different training algorithms to guide the network for providing an accurate result with less training and testing error. These algorithms have succeeded with different accuracy levels, but it is still suffering from some weaknesses. Weakness such as trapped in local minima, slow convergence and finding a good rate between exploitation and exploration of the search space. This research proposed a swarm intelligence training algorithm, Improved African Buffalo Optimization algorithm (IABO) based on the Metaheuristic method called the African Buffalo Optimization algorithm (ABO). ABO has been successful in solving many improvement problems. These successes motivate the development and investigation of its efficiency in training Feed Forward Neural Networks (FFNNs), for solving training process issues. Additionally, the study investigated the effect of neurons number in the hidden layer, the number of population swarm, and the stopping criteria (iterations) on the model’s performance. Water level data set was chosen to test the proposed IABO-trained algorithm. The results were verified by benchmarking with the performance of the Particle Swarm Optimization (PSO) and Backpropagation (BP) algorithms. The results demonstrated the superiority of the IABO-trained algorithm in avoiding local minima, convergence speed, and accuracy compared to the benchmarking (BP and PSO) algorithms in water level forecasting tasks.