Real-Time Evolving Deep Learning Models for Predicting Hydropower Generation

Deep learning models have shown great promise for predicting hydropower generation. Previous research has focused on energy output prediction or predictive maintenance using traditional artificial neural networks (ANNs). However, these models lack sustainability in the face of changing environmental conditions. The need for dynamic, real-time modeling becomes apparent in rapidly changing environments, where speed and accuracy of execution are critical. In this paper, we present a framework for real-time evolving deep learning (RT-EDL) models designed to accurately predict hydropower generation on a daily, weekly, and monthly basis. Our evolving model employs backpropagation techniques and a stochastic gradient descent optimizer to continuously fine-tune the model using newly acquired data points in real time. To validate our approach, we conduct a case study using the RT-EDL model and show how the hyperparameters in the evolving model can be adjusted to achieve optimal operation. Our experimental results not only demonstrate the feasibility and effectiveness of our real-time evolving model, but also highlight its superiority over traditional deep learning methods.