Source: Tensoren Technology

Our client is a medium-sized hydropower station in southern China, and its power generation mainly depends on the inflow of water from a certain local river basin. Our analysis shows that the inflow water volume is highly correlated with the monthly rainfall of the hydropower station. Therefore, we decided to start with rainfall, predict the amount of incoming water, and make corresponding optimization analysis of water storage and release.

　　Expected precipitation and expected inflow water

We found out that the cumulative precipitation in the basin where the client is located has a decreasing trend, which may pose a challenge to achieve the annual power generation plan. Therefore, we recommend that the client should consider this factor when making future power generation plans. At the same time, we found out that the frequency and intensity of heavy rains in the basin are on the rise, and thus the difficulty of flood management has increased. It is necessary to improve flood management.

　　Climate data shows a downward trend in precipitation

The client’s historical data shows that in some years, the amount of abandoned water exceeded 100 million cubic meters. Based on the local electricity price, the loss of power generation revenue was nearly RMB 4 million. Given that the marginal cost of hydropower is almost zero, this part of the loss could have been the profit of the power station! The client told us that the loss of abandoned water was mainly due to the rainstorm process and the forecast of the time and size of the flood peak. Therefore, we further used machine learning to analyze and simulate the rainfall process and flood peak formation characteristics of the basin to improve the forecast for peak flood peak and adjust the water storage and the discharge plans accordingly.