Gaussian mutation-alpine skiing optimization algorithm-recurrent attention unit-gated recurrent unit-extreme learning machine model: an advanced predictive model for predicting evaporation

Evaporation prediction is essential for efficient water resource management and environmental monitoring. Our study introduces a new model named recurrent attention unit (RAU)–gated recurrent unit (GRU)–extreme learning machine model (ELM) to extract temporal features and predict evaporation data. While the ELM model may not extract temporal features from data, the GRU model can extract temporal features from time series data. The GRU–ELM model may not pay attention to important features of time series data. Our study introduces a RAU to address this limitation of the GRU–ELM model. RAU can extract and store important features. Our study also introduces a new optimization algorithm called Gaussian mutation (GM)–Alpine skiing optimization algorithm (ASOA) to improve the exploitation ability of the ASOA for feature selection. The novelties of the current research include the development of the RAU–GRU–ELM (RGE) model for predicting evaporation data and the introduction of a new optimizer for selecting inputs. The GM–ASOA model chose the maximum temperature, minimum temperature, and number of sunny hours as the optimal input scenario. Determining the best input scenario was difficult and complex as there were 26-1 input combinations. GM–ASOA automatically selects the most appropriate input scenario for predicting evaporation. While ELM was unable to extract spatiotemporal patterns, GM–ASOA–RGE was able to extract spatiotemporal patterns and make accurate evaporation predictions. The study results indicated that the GM–ASOA RAU–GRU–ELM (GM–ASOA–RGE) had the best accuracy among other models. The GM–ASOA–RAU–GRU–ELM (GM–ASOA–RGE) decreased training MAE of the ASOA–RGE, RGE, GRU–ELM, RAU–GRU, GRU, and ELM by 4.6