Forecasting water usage based on the CaffeNet model combined with the developed student psychology-based optimizer

This research paper presents an advanced water demand forecasting model through CaffeNet deep-learning architecture as well as a developed student psychology-based optimizer (DSPBO), aiming to improve the predictability of water consumption for the domestic, industrial, and agricultural sectors. The combined CaffeNet-DSPBO model has performed well in the performance evaluation to capture the complex nonlinear relationships caused by weather conditions, seasonality, and sector-specific patterns, and is trained using real data from the Yangtze River Delta of China. The main findings show a model with low RMSE values of 0.25 (domestic), 0.40 (industrial), and 0.58 (agricultural) and high correlation coefficients of 0.87, 0.75, and 0.62, respectively. This indicates that the domestic consumption sector, in particular, can be considered a reliable and accurate forecasting model. Also, the model demonstrated superior performance compared to other meta-heuristic algorithms in terms of convergence stability and solution accuracy. Another performance advantage is the training time of less than an hour and the inference latency of less than 10 ms. The results show how important this can be in combining deep-learning and better optimization techniques for predicting multi-sector water needs, paving the way for sustainable yet efficient management of this precious resource.