Flood routing with nature-inspired optimization techniques and artificial neural networks

Within the scope of this thesis, flood prediction is performed by optimizing the weight and bias parameters of the artificial neural network with nature-inspired optimization algorithms such as artificial bee colony algorithm, particle swarm optimization algorithm, firefly algorithm, frog leap algorithm and genetic algorithm. The flood forecasting performance of meta-heuristic optimization techniques was evaluated by comparing the forecasting results with the single artificial neural network algorithm. The forecasts were made using the flood data of Ordu-Turna water in 2009 and 2013. During modeling, 30-minute flows at the upstream station were used as input and 30-minute flows downstream were used as target variables. In the model setup, 70% of the data is divided into training and 30% of the data is divided into testing. The performance of the models was tested according to statistical metrics such as root mean square error (RMSE), mean absolute error (MAE), Akaike Information Criterion (AIC), Nash-Sutcliffe Efficiency (NSE), Kling-Gupta Efficiency (KGE), coefficient of determination (R2), mean bias error (MBE), bias factor (BF) and percent bias (Pbias). As a result of the analysis, the ABCANN hybrid model predicted the 2009 floods with RMSE:1.84, AIC:71.63, NSE:0.94, R2:0.97, and Pbias:1.07, and the 2013 floods with RMSE:1.23, AIC:19.99, NSE:0.95, R2:0.95, and Pbias:0.63. In addition, according to the percentage changes of RMSE values, meta-heuristic optimization algorithms were found to reduce the error level of the single artificial neural network model in flood forecasting in the range of 45% to 93%. In addition, the model performances were visualized with heat maps, cumulative scatter plots, Taylor diagrams, violin plots, box plots, scatter plots, bar charts and line plots to emphasize the superiority of the ABCANN hybrid approach. stuck in the local optimum by mimicking the interaction of swarm members with each other and helps to determine the most appropriate model parameters, produced the most accurate prediction results. The results of the study help decision makers and policy makers to anticipate potential flood risks, take proactive measures and develop appropriate response strategies. Keywords: hyper parameter optimization, swarm intelligence, flood routing, artificial intelligence