Застосування LSTM та CNN моделей для прогнозування багатопараметричних показників якості води в реальному часі

Abstract

The paper examines the application of deep neural networks, namely LSTM (Long Short-Term Memory) and CNN (Convolutional Neural Network), for forecasting multiparametric water quality indicators in real time. The architectural features of recurrent networks with memory elements are analyzed, including the mechanism of controlling information flow through forget, input, and output gates. The principles of constructing hybrid CNN-LSTM models are considered, in which convolutional layers extract local spatial features of time series while LSTM layers capture long-term temporal dependencies. The feasibility of using such hybrid models to improve the accuracy of forecasting an integral complex water quality indicator based on multiparametric Big Data models is justified.