Tuning ANN Hyperparameters by CPSOCGSA, MPA, and SMA for Short-Term SPI Drought Forecasting

Abstract

Modelling drought is vital to water resources management, particularly in arid areas, to reduce its effects. Drought severity and frequency are significantly influenced by climate change. In this study, a novel hybrid methodology was built, data preprocessing and artificial neural network (ANN) combined with the constriction coefficient-based particle swarm optimisation and chaotic gravitational search algorithm (CPSOCGSA), to forecast standard precipitation index (SPI) based on climatic factors. Additionally, the marine predators algorithm (MPA) and the slime mould algorithm (SMA) were used to validate the performance of the CPSOCGSA algorithm. Climatic factors data from 1990 to 2020 were employed to create and evaluate the SPI 1, SPI 3, and SPI 6 models for Al-Kut City, Iraq. The results indicated that data preprocessing methods improve data quality and find the best predictors scenario. The performance of CPSOCGSA-ANN is better than MPA-ANN and SMA-ANN algorithms based on various statistical criteria (i.e., R2, MAE, and RMSE). The proposed methodology yield R2 = 0.93, 0.93, and 0.88 for SPI 1, SPI 3, and SPI 6, respectively.

novel approach for predicting the standardised precipitation index considering climatic factors

“…The aim of this approach is to minimise the impact of outliers and make the time series have a normal distribution or close to normal [32]. In this research, the natural logarithm technique was utilised to normalise the time series and make it more static and reduce the predictors' collinearity [23].…”

“…Accordingly, the box and whisker approach was used to clean data from outliers. Also, singular spectrum analysis (SSA) was utilised for denoising time series [32].…”

“…The trial and error method may not necessarily offer the best answer, which is considered time-consuming. As a result, metaheuristic algorithms were integrated with ANN to choose the best learning rate value and the best number of neurons for hidden layers to gain the best input, output mapping and avoid underfitting and overfitting the model [32].…”

“…The data was then categorised into three sets: training (70%, 1990-2010); testing (15%, 2011-2015); and validation (15%, 2016-2020) according to Alawsi, Zubaidi [32] and Soh, Koo [10].…”

Application of Metaheuristic Algorithms and ANN Model for Univariate Water Level Forecasting

“…Additionally, particle swarm optimisation (PSO) has been employed in multiple hydrology areas, for example, WL [38] and streamfow [39]. Moreover, the constriction coefcient-based particle swarm optimisation and chaotic gravitational search algorithm (CPSOCGSA) was proposed by Rather and Bala [40], and it is used in the prediction of drought [41]. Te CPSOCGSA algorithm was proposed under the strategy of hybridising the existing algorithms.…”