PDF(2506 KB)
Ultra-Short-Term Forecasting of Upstream Water Level at the Three Gorges Reservoir Using Deep Learning
XIE Shuai, CAO Hui, WANG Dong, ZHANG Zheng, ZHOU Tao
Journal of Changjiang River Scientific Research Institute ›› 2026, Vol. 43 ›› Issue (4) : 45-51.
PDF(2506 KB)
PDF(2506 KB)
Ultra-Short-Term Forecasting of Upstream Water Level at the Three Gorges Reservoir Using Deep Learning
[Objective] Although artificial intelligence-based water level forecasting methods have achieved promising results in predicting upstream water levels at various power stations, including the Three Gorges Reservoir, there remains room for improvement. To obtain more accurate water level predictions for the Three Gorges Reservoir, this study develops an ultra-short-term forecasting model with a 15-minute time scale based on deep learning techniques, providing enhanced technical support for real-time reservoir operation. [Methods] The dataset comprises four categories: (1) water level data from the Three Gorges Reservoir and downstream areas; (2) inflow and spillage flow rates of the Three Gorges; (3) total power output of the Three Gorges power plant; and (4) precipitation between Cuntan and the Three Gorges area. Four water level forecasting models, including a baseline model and three comparative models, were developed to predict water level changes over the next 24 hours. Each model was constructed using both LSTM and RNN neural networks. The primary distinctions among these models lie in the processing of input and output data as well as the temporal scales of the data. By comparing the performance of different models under varying conditions, we analyze how model configurations impact prediction accuracy. [Results and Conclusion] (1) Regardless of input conditions, water level forecasting models built with LSTM outperform those using RNN, achieving Mean Absolute Errors (MAE) of 3.58 to 4.40 cm and maximum absolute errors of 56.99 to 110.03 cm. (2) All four water level forecasting models constructed using LSTM exhibit good performance, with the best-performing model incorporating dynamic reservoir capacity and interval rainfall impacts, achieving an MAE of 3.58 cm and a maximum absolute error of 56.99 cm. (3) Differences in model input variables are the dominant factor affecting forecast accuracy across various conditions. Incorporating reservoir water level information allows the model to better account for dynamic reservoir capacity effects, while adding interval rainfall data provides more precise inflow estimates, significantly enhancing prediction accuracy. This approach reduces the MAE by 18.64% compared to the baseline model. This study demonstrates that integrating relevant hydrological and meteorological factors into LSTM-based models can substantially improve the precision of short-term water level forecasts, thereby supporting effective reservoir management.
ultra-short-term water level forecasting / interval rainfall / dynamic reservoir storage / deep learning / Three Gorges Reservoir / upstream water level
| [1] |
|
| [2] |
|
| [3] |
|
| [4] |
|
| [5] |
The forecasting accuracy of downstream water levels greatly impacts the economic operation of reregulating hydropower stations. If present, large errors in water level forecasting may require the hydropower station to discard water, which decreases the revenue from hydropower generation. Especially when the reregulating hydropower station undertakes the task of peak load regulation of the power grid, its power output may change drastically over a short period of time. Such large changes in the discharged flow of the hydropower station may inevitably cause an unsteady flow, which can lead to large fluctuations in the water levels of downstream rivers. In this case, the current forecasting methods, such as standard rating curves and empirical formulae, may inevitably result in large deviations when forecasting the change in the downstream water level. In order to reduce the volume of water discarded as a result of forecasting errors, a back propagation neural network-based forecasting model for downstream water levels of a reregulating hydropower station was constructed. The model enabled the direct, accurate, real-time forecasting of changes in downstream water levels by using measurable operation data from a hydropower station and process data on downstream water level changes. This method was applied to the actual hydropower generation dispatch of the Gezhouba Hydropower Plant in China. The results showed that as the peak load regulation volume increased, the forecasting errors of two conventional methods (standard rating curves and empirical formulae) became continuously superimposed. However, the water level forecasting error of the neural network-based method was small and could fully meet the real-time dispatching requirements of the hydropower station. In particular, under large peak load regulations, the maximum absolute values of the forecasting errors of the two conventional methods were close to 1 m, while that of the neural network-based method could be controlled within 0.3 m.
|
| [6] |
徐杨, 刘亚新, 汪涛, 等. 基于LSTM的三峡水库短期上游水位预测方法研究[J]. 水利水电快报, 2022, 43(10): 13-18.
(
|
| [7] |
郭爽. 梯级水电站短期水位预测与调控策略研究[D]. 邯郸: 河北工程大学, 2021.
(
|
| [8] |
李力. 复杂水力联系下梯级水库短期联合优化调度研究[D]. 武汉: 华中科技大学, 2021.
(
|
| [9] |
张森, 谢帅, 王永强, 等. 基于机器学习的电站耗水率计算及其对水位预测影响研究[J]. 水力发电, 2023, 49(12): 94-100.
(
|
| [10] |
蹇德平, 丁义. 区间降雨径流预报方案及其梯级水库运行实用分析[J]. 四川水力发电, 2016, 35(增刊2): 126-128.
(
|
| [11] |
张余龙. 长江上游水库群一维水动力模型应用及其与水文、调度模型耦合研究[D]. 武汉: 华中科技大学, 2021.
(
|
| [12] |
徐鼎甲. 进行日调节时水电站下游水位的计算[J]. 水利水电技术, 1995, 26(4): 2-4.
(
|
| [13] |
郭爽, 龙岩, 王孝群, 等. 沙坪二级水电站短期水位预测与实时调控策略[J]. 水电能源科学, 2022, 40(8):83-87.
(
|
| [14] |
刘亚新, 樊启祥, 尚毅梓, 等. 基于LSTM神经网络的水电站短期水位预测方法[J]. 水利水电科技进展, 2019, 39(2): 56-60, 78.
(
|
| [15] |
刘晓阳, 姚华明, 张海荣, 等. 基于机器学习的三峡水库h尺度坝前水位预测[J]. 人民长江, 2023, 54(2): 147-151.
(
|
| [16] |
刘晓阳, 郭乐. 基于机器学习方法的葛洲坝水电站下游水位预测研究[J]. 水利水电快报, 2022, 43(10):19-22,36.
(
|
| [17] |
王永强, 张森, 谢帅, 等. 基于深度学习的三峡电站未来坝前最大最小水位预测[J]. 长江科学院院报, 2024, 41(12): 9-14.
最大最小水位是计算梯级水库调度问题、水电站经济运行问题等时要考虑的重要约束条件,其精确预测能为水电站经济运行提供支持。常用的迭代计算容易误差积累,导致多时段预测结果可信度降低。选取对时间序列问题有良好处理效果的长短时记忆网络模型(LSTM),将其应用于三峡电站未来4 d最大最小坝前水位预测中,依据水量平衡预测框架构建传统预测模型;基于LSTM使用不同特征变量构建2种深度学习模型,并比较其预测效果。计算结果表明,考虑三峡库区水面线传播规律后的深度学习模型预测具有精确稳定的预测效果,99%预测绝对误差<40 cm。
(
|
| [18] |
Several variants of the long short-term memory (LSTM) architecture for recurrent neural networks have been proposed since its inception in 1995. In recent years, these networks have become the state-of-the-art models for a variety of machine learning problems. This has led to a renewed interest in understanding the role and utility of various computational components of typical LSTM variants. In this paper, we present the first large-scale analysis of eight LSTM variants on three representative tasks: speech recognition, handwriting recognition, and polyphonic music modeling. The hyperparameters of all LSTM variants for each task were optimized separately using random search, and their importance was assessed using the powerful functional ANalysis Of VAriance framework. In total, we summarize the results of 5400 experimental runs ( ≈ 15 years of CPU time), which makes our study the largest of its kind on LSTM networks. Our results show that none of the variants can improve upon the standard LSTM architecture significantly, and demonstrate the forget gate and the output activation function to be its most critical components. We further observe that the studied hyperparameters are virtually independent and derive guidelines for their efficient adjustment.
|
| [19] |
Flood forecasting is an essential requirement in integrated water resource management. This paper suggests a Long Short-Term Memory (LSTM) neural network model for flood forecasting, where the daily discharge and rainfall were used as input data. Moreover, characteristics of the data sets which may influence the model performance were also of interest. As a result, the Da River basin in Vietnam was chosen and two different combinations of input data sets from before 1985 (when the Hoa Binh dam was built) were used for one-day, two-day, and three-day flowrate forecasting ahead at Hoa Binh Station. The predictive ability of the model is quite impressive: The Nash–Sutcliffe efficiency (NSE) reached 99%, 95%, and 87% corresponding to three forecasting cases, respectively. The findings of this study suggest a viable option for flood forecasting on the Da River in Vietnam, where the river basin stretches between many countries and downstream flows (Vietnam) may fluctuate suddenly due to flood discharge from upstream hydroelectric reservoirs.
|
| [20] |
. Rainfall–runoff modelling is one of the key\nchallenges in the field of hydrology. Various approaches exist, ranging from\nphysically based over conceptual to fully data-driven models. In this paper,\nwe propose a novel data-driven approach, using the Long Short-Term Memory\n(LSTM) network, a special type of recurrent neural network. The advantage of\nthe LSTM is its ability to learn long-term dependencies between the provided\ninput and output of the network, which are essential for modelling storage\neffects in e.g. catchments with snow influence. We use 241 catchments of the\nfreely available CAMELS data set to test our approach and also compare the\nresults to the well-known Sacramento Soil Moisture Accounting Model (SAC-SMA)\ncoupled with the Snow-17 snow routine. We also show the potential of the LSTM\nas a regional hydrological model in which one model predicts the discharge\nfor a variety of catchments. In our last experiment, we show the possibility\nto transfer process understanding, learned at regional scale, to individual\ncatchments and thereby increasing model performance when compared to a LSTM\ntrained only on the data of single catchments. Using this approach, we were\nable to achieve better model performance as the SAC-SMA + Snow-17, which\nunderlines the potential of the LSTM for hydrological modelling applications.
|
| [21] |
Accurate forecasting of the tail water level (TWL) is of great importance for the safe and economic operation of hydropower stations. The prediction accuracy is significantly influenced by the backwater effect of downstream tributaries and the operation of adjacent hydropower stations, but the explicit quantification method of the backwater effect is lacking. In this study, a deep-learning-model-based forecasting method for TWL predictions under the backwater effect is developed and applied in the Xiangjiaba (XJB) hydropower station, which is influenced by the backwater effect of downstream tributaries, including the Hengjiang River (HJR) and the Minjiang River (MJR). Firstly, the random forest algorithm was used to analyze the influence of HJR and MJR flows with different lag times on the TWL prediction error of the XJB hydropower station. The results show that the time lags of the backwater effect of HJR and MJR run offs on the TWL of the XJB are 5~7 h and 1~2 h, respectively. Then, the run off thresholds of the HJR and MJR for impacting the TWL of the XJB station are obtained through scenario comparison, and the results show that the run off thresholds of the HJR and the MJR are 700 m3/s and 7000 m3/s, respectively. Finally, based on the analysis of the time lag and the threshold of the backwater effect, a deep learning model (LSTM)-based TWL forecasting method is established and applied to predict the TWL of the XJB station. The results show that the forecasting model has a good predictive performance, with 98.22% of absolute errors less than 20 cm. The mean absolute error over the validation dataset is 5.27 cm, and the maximum absolute error is 63.35 cm. Compared with the LSTM-based prediction model without considering the backwater effect, the mean absolute error decreased by 31%, and the maximum absolute error decreased by 71%.
|
/
| 〈 |
|
〉 |
