Optimization of interpretable hydropower reservoir operation rules by denoising diffusion probabilistic model, parallel chaotic cooperation search algorithm and liquid neural network

The development of hydropower reservoir operation rules is crucial for ensuring their efficient, stable operation and rapid response capabilities. This study proposes an interpretable DDPM-PCCSA-LNN reservoir operation rules framework to address issues such as inadequate scenario representativeness, time-consuming optimization, limited training data, and lack of interpretability in traditional methods. DDPM generates large-scale operation scenarios, PCCSA solves the corresponding optimization problem, and LNN extracts operational rules with SHAP interpreting key factors. The model is applied to the Hongjiadu and Three Gorges hydropower reservoirs and results show that operation rules simulated by proposed model are closer to the actual optimal operation process, and the power generation balances efficiency and sustainability. In the test scenarios of the Hongjiadu hydropower reservoir, compared with traditional model, proposed model achieves 15.3% to 20.3% improvement in SI, while average annual power generation can reach 99.1% to 99.3% of optimal operation model. Based on traditional models, this method adds modules for scenario simulation and interpretability analysis, improves the construction accuracy of operation rules, and provides a valuable technical approach for watershed management.