Risk dispatching model of cascade hydropower station reservoirs considering power generation and water abandonment risks

To quantify the risk situation during the operation of cascade hydropower station reservoirs, a risk operation model driven by forecast information was proposed. Based on long-short term memory (LSTM) neural network, a risk situation prediction model was established. By constructing a mapping relationship between predicted inflow runoff, reservoir operating conditions, scheduling decisions, output damage degree, and abandoned water volume, the risk of reservoir scheduling in cascade hydropower stations under complex operating conditions was dynamically deduced. A risksituation based scheduling model for cascade hydropower station reservoirs was established, and risk pre control models for insufficient power generation and water abandonment were proposed. A case study was conducted on a cascade hydropower station consisting of the Jinping Ⅰ Reservoir and the Ertan Reservoir in the lower reaches of the Yalong River Basin. The results show that compared to the conventional maximum power generation model, the risk scheduling model compensates for future output shortage periods by reducing a certain amount of output in advance, reducing the depth of output shortage of the Jinping Ⅰ Reservoir and Ertan Reservoir by 46% and 63%, respectively. Additionally, it increases power generation discharge in advance to reserve storage capacity, thus alleviating the water surplus in the subsequent periods in the wet season, leading to a reduction of total water surplus of the Jinping Ⅰ Reservoir and Ertan Reservoir by 4.8% and 5.4%, respectively. The risk situation prediction model can incorporate the uncertainty of inflow into the risk scheduling model, enabling dynamic response to the operational state of the hydropower station reservoir and changes in inflow situation through risk control decisions.