Research on Water Resource Carrying Capacity Assessment and Water Quality Forecasting Based on Feature Selection with CNN-BiLSTM-Attention Model of the Min River Basin

Water, Journal Year: 2025, Volume and Issue: 17(6), P. 824 - 824

Published: March 13, 2025

To achieve a more accurate assessment of water resource carrying capacity (WRCC), the indicators resources, social and ecological environment were selected to construct WRCC system on basis combinatorial assignment method with advantages. Moreover, incorporation key quality influences into predictions facilitated performance predictive models. Adaptive Lasso Regression was used select factors affecting quality, whereas CatBoost algorithm ranked importance by in prediction model. The Convolutional Neural Network-Bidirectional Long Short-Term Memory-Attention (CNN-BiLSTM-Attention) model forecast WQI. research results propose new evaluation method. show that average barrier levels for socio-economic development, 34.97%, 34.93%, 30.10%, respectively. Compared other layers WRCC, obstacle degree layer has always been lower. total sewage treatment, greening coverage built-up areas, per capita green space parks main within Min River Basin. Based factor screening, it can be seen dissolved oxygen is positively correlated watershed, while influencing are negatively Total nitrogen had greatest impact conditions regression coefficient −1.7532. From comparison results, known hybrid make MAE value 45% monitoring points reach minimum, RMSE 35% minimum. percentages remaining models reached lowest values 15% 20% 30%, models, MSE relatively small, which conducive predicting

Language: Английский

Study on Motion Response Prediction of Offshore Platform Based on Multi-Sea State Samples and EMD Algorithm

Water, Journal Year: 2024, Volume and Issue: 16(23), P. 3441 - 3441

Published: Nov. 29, 2024

The complexity of offshore operations demands that platforms withstand the variability and uncertainty marine environments. Consequently, analyses platform motion responses must extend beyond single sea state conditions. This study employs Computational Fluid Dynamics (CFDs) software STAR-CCM+ for data acquisition investigates from two perspectives: adaptability analysis to different wave directions varying significant heights. aim is develop a model capable predicting across multiple results demonstrate integrating empirical mode decomposition (EMD) algorithm with residual convolutional neural networks (ResCNNs) Long Short-Term Memory (LSTM) effectively resolves challenge insufficient prediction accuracy under diverse maritime Following EMD incorporation, model’s performance within predictive range was significantly enhanced, coefficient determination (R2) consistently exceeding 0.5, indicating high degree fit data. Concurrently, mean squared error (MSE) Mean Absolute Percentage Error (MAPE) metrics exhibited commendable performance, further substantiating precision reliability. methodology introduces an innovative approach forecasting dynamic structures, providing more rigorous accurate foundation operational decisions. Ultimately, research enhances safety productivity activities.

Language: Английский