基于物理约束长短时记忆网络的能源转型背景下梯级水电站长期运行管理

Abstract On the path to energy transition, substantial increases in wind and solar power are expected to heighten the complexity of ensuring the continuous load-generation balance for grid stability. Hydropower could be the low-carbon source of flexibility to integrate wind and solar power, but seasonal fluctuations and multiple coupling uncertainties would shape traditional hydropower operations. Here we propose a new simulation–optimization–learning approach that addresses uncertainties and nonlinear dynamic hydropower operation characteristics to extract long-term operational rules for cascade hydropower plants under energy transition. The approach consists of three key steps: Simulation, using Kirsch–Nowak Streamflow Generator and ARIMA to track hydrological and meteorological uncertainties; Optimization, developing the objective-driven optimal model that consider nonlinear dynamic hydropower operation characteristics to obtain optimal schemes before and after energy transition; and Learning, constructing physics-constrained LSTM networks (PCLSTM) that incorporate physical reservoir operation constraints to learn operational rules from the optimal schemes. Case studies are conducted for a hydro-wind-solar hybrid system in Southwest China’s Wujiang River Basin. Results show that: (1) The effective operational rules adapted to energy transition can be extracted; (2) Compared to long-short term memory networks, the operational rules extracted by PCLSTM can enhance simulation accuracy and reduce the degree of reservoir level penalty, effectively guiding hydropower operations both before and after energy transition. The average degree of reservoir level penalty in all hydropower plants could decrease from 22 % to 5 % before the energy transition and from 10 % to 4 % after energy transition; (3) Compared to stochastic dual dynamic programming (SDDP), the real simulation results from the proposed approach are close to those of the state-of-the-art existing approaches and can address problems that the SDDP cannot solve. (4) Hydropower could perform intra-annual hydraulic-electricity spatial-temporal exchange to accommodate wind and solar power, at the expense of sacrificing a portion of hydropower generation.