Combining Prediction Models and Dimensionality Reduction Technology for Water Resources Management Under Incomplete Information and Dynamic Change

The reasonable selection of water resource management alternatives is crucial for decision-makers to develop effective management measures ensuring the sustainable use of water resources. A hybrid decision framework is proposed to address incomplete decision data and dynamic updates of criteria. This framework integrates generalized regression neural networks (GRNN), affinity propagation clustering (AP), probabilistic neural networks (PNN), and a linear weighting method, aiming to achieve long-term socio-economic sustainability. First, based on the PNN model, the missing matrix is filled to obtain a complete evaluation matrix. Then, AP is used to cluster the evaluation criteria to obtain the decision matrix of the cluster center, effectively reducing the dimension of the matrix. For the new evaluation criteria, PNN classifies and updates the cluster center matrix, addressing criteria updates effectively. The alternatives are ranked using the linear weighting method. The framework's effectiveness and reliability are verified through two cases: urban water supply and reservoir optimal operation. Results demonstrate that this hybrid decision-making framework aids managers in selecting optimal schemes.