Effect of flow forecasting quality on benefits of reservoir operation – a case study for the Geheyan reservoir (China)

This paper presents a methodology to determine the effect of flow forecasting quality on the benefits of reservoir operation. The benefits are calculated in terms of the electricity generated, and the quality of the flow forecasting is defined in terms of lead time and accuracy of the forecasts. In order to determine such an effect, an optimiza5 tion model for reservoir operation was developed which consists of two sub-models: a long-term (monthly) and a short-term (daily) optimization sub-model. A methodology was developed to couple these two sub-models, so that both short-term benefits (time span in the order of the flow forecasting lead time) and long-term benefits (one year) were considered and balanced. Both sub-models use Discretized Dynamic Program10 ming (DDP) as their optimization algorithms. The Geheyan reservoir on the Qingjiang River in China was taken as case study. Observed (from the 1997 hydrological year) and forecasted flow series were used to calculate the benefits. Forecasted flow series were created by adding noises to the observed series. Different magnitudes of noise reflected different levels of forecasting accuracies. The results reveal, first of all, 15 a threshold lead time of 33 days, beyond which further extension of the forecasting lead time will not lead to a significant increase in benefits. Secondly, for lead times shorter than 33 days, a longer lead time will generally lead to a higher benefit. Thirdly, a perfect inflow forecasting with a lead time of 4 days will realize 87% of the theoretical maximum electricity generated in one year. Fourthly, for a certain lead time, more 20 accurate forecasting leads to higher benefits. For inflow forecasting with a fixed lead time of 4 days and different forecasting accuracies, the benefits can increase by 5 to 9% compared to the actual operation results. It is concluded that the definition of the appropriate lead time will depend mainly on the physical conditions of the basin and on the characteristics of the reservoir. The derived threshold lead time (33 days) gives a 25 theoretical upper limit for the extension of forecasting lead time. Criteria for the appropriate forecasting accuracy for a specific feasible lead-time should be defined from the benefit-accuracy relationship, starting from setting a preferred benefit level, in terms