Machine Learning for Hydropower Scheduling: State of the Art and Future Research Directions

Bordin, Chiara; Skjelbred, Hans Ivar; Kong, Jiehong; Yang, Zhirong

doi:10.1016/j.procs.2020.09.190

Cited by 36 publications

(28 citation statements)

References 37 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Given that such choices can be based on the reservoirs' spatial coupling and, therefore, depend on the problem, it is possible to find several possibilities in the literature. For example, methods such as the principal component analysis [33], global sensitivity analysis [34], clustering criteria [17], or techniques based on machine learning [35] can be used to address these choices.…”

Section: A Inflow Aggregationmentioning

confidence: 99%

“…In ( 35) zβ,NS-1 is the critical value β of the t-student distribution with NS -1 degree of freedom, and σNS is the standard deviation related to fcNS [41]. The equation (35) means that, although it is impossible to know the exact value fc, it can be assumed that with 100 (1-β) % confidence, its value is not greater than CI.…”

Section: Analysis Frameworkmentioning

confidence: 99%

“…The first CI is calculated for the most detailed model, that is, 1A, by using NS trees, each with NK inflow scenarios. Then, the average value of the NS expected costs and the respective standard deviation are computed to obtain (35) via the forward simulations. The one-node LP employed in these simulations, for the instance 1A, can be rewritten as follows: where: JS1 number of inequality constraints used to represent the FCF of month t.…”

Section: Analysis Frameworkmentioning

confidence: 99%

See 2 more Smart Citations

Inflow Aggregation and Run-of-the-River Inflow Energy for Reducing Dimensionality in the Long-Term Generation Scheduling Problem

Fredo

Finardi

Larroyd³

et al. 2021

IEEE Access

View full text Add to dashboard Cite

The long-term generation scheduling (LTGS) problem aims to build an operating policy over a multi-year planning horizon, correlating thermal generation and deficit costs with water storage. The LTGS is modeled as a linear multistage stochastic program, whose state-of-art solution is the stochastic dual dynamic programming (SDDP). One critical challenge is to represent the time-dependency of river inflows accurately. Despite recent advances, modeling simplifications are needed to allow the LTGS computational tractability via SDDP since it is necessary to increase the state space to include the time-dependent variables properly. Thus, most works simplify the hydro production function (HPF) to represent the inflows in detail to overcome this negative aspect. However, the literature lacks this trade-off, i.e., finding a balance between stochastic inflow modeling and HPF representation. Thus, this paper proposes an alternative approach to analyzing this trade-off, using the inflow aggregation and run-of-the-river energy inflow instead of individual inflow in SDDP. The proposed approach drastically reduces the number of state variables in SDDP, allowing a detailed HPF representation in the LTGS. We employ a one-sided confidence interval for expected cost estimation to show that our approach provides better performance than the individualized inflows. The analysis is performed in several large-scale computational instances using a power system with 53 geographically widespread hydro plants. The numerical results demonstrate that inflow aggregation provides, on average, a 4 % reduction on the operating cost, whereas for the run-of-the-river inflow energy, the reduction is, on average, 2 %. INDEX TERMS Inflow modeling, hydro production function, the long-term generation scheduling problem, stochastic dual dynamic programming, run-of-the-river inflow energy.

show abstract

Section: A Inflow Aggregationmentioning

confidence: 99%

Section: Analysis Frameworkmentioning

confidence: 99%

Section: Analysis Frameworkmentioning

confidence: 99%

See 1 more Smart Citation

Inflow Aggregation and Run-of-the-River Inflow Energy for Reducing Dimensionality in the Long-Term Generation Scheduling Problem

Fredo

Finardi

Larroyd³

et al. 2021

IEEE Access

View full text Add to dashboard Cite

show abstract

“…The energy generation process is divided into three stages: preoperation, real-time, and post-operation. The pre-operation steps can be separated into long-term, medium-term, short-term, and real-time scheduling [4].…”

Section: Introductionmentioning

confidence: 99%

Hydropower Operation Optimization Using Machine Learning: A Systematic Review

Bernardes

Santos

Abreu

et al. 2022

View full text Add to dashboard Cite

The optimal dispatch of hydropower plants consists of the challenge of taking advantage of both available head and river flows. Despite the objective of delivering the maximum power to the grid, some variables are uncertain, dynamic, non-linear, and non-parametric. Nevertheless, some models may help hydropower generating players with computer science evolution, thus maximizing the hydropower plants’ power production. Over the years, several studies have explored Machine Learning (ML) techniques to optimize hydropower plants’ dispatch, being applied in the pre-operation, real-time and post-operation phases. Hence, this work consists of a systematic review to analyze how ML models are being used to optimize energy production from hydropower plants. The analysis focused on criteria that interfere with energy generation forecasts, operating policies, and performance evaluation. Our discussions aimed at ML techniques, schedule forecasts, river systems, and ML applications for hydropower optimization. The results showed that ML techniques have been more applied for river flow forecast and reservoir operation optimization. The long-term scheduling horizon is the most common application in the analyzed studies. Therefore, supervised learning was more applied as ML technique segment. Despite being a widely explored theme, new areas present opportunities for disruptive research, such as real-time schedule forecast, run-of-river system optimization and low-head hydropower plant operation.

show abstract

“…The successful ML's applications in the hydropower sector are aimed at the forecast of the inflow, streamflow, consumption, generation projections, cavitation problems [13,14], and do not concern the lifetime prediction or crack propagation. Nowadays, predicting cracks in runners and lifetime issues are only solved by remote diagnostic centers or experts.…”

Section: Introductionmentioning

confidence: 99%

Limitations of Modern Diagnostic and Prognostic Systems for a Hydraulic Unit’s Health

Georgievskaia

2021

Eng

View full text Add to dashboard Cite

Modern diagnostic systems for the hydraulic unit’s health play an important role in ensuring the reliability and safety of the hydroelectric power plant (HPP). However, they cannot provide timely detection of such dangerous operational defects as fatigue cracks. This article reflects two main reasons for this problem. The first one is a high level of the individuality of hydraulic units, which does not allow the effective use of statistical methods of information processing, including BIG DATA and MACHINE LEARNING technologies. The second is the fundamental impossibility to identify cracks in some key components of hydraulic units only on the basis of data analysis from a standard diagnostic system usually used at the HPP. Developed computational studies on the example of Francis turbines confirmed this. It is proposed to supplement the functionality of standard diagnostic systems with a prognostic block for an individual analytical forecast of the unit’s residual lifetime based on the calculated assessment of fatigue strength. This article presents the developed conceptual diagram and the demonstration version of the proposed analytical predictive system. The comparison of the standard vibration diagnostic system and the proposed solution as a tool for the early detection of cracks in a Francis turbine runner shows some advantages of the proposed approach.

show abstract

Machine Learning for Hydropower Scheduling: State of the Art and Future Research Directions

Cited by 36 publications

References 37 publications

Inflow Aggregation and Run-of-the-River Inflow Energy for Reducing Dimensionality in the Long-Term Generation Scheduling Problem

Inflow Aggregation and Run-of-the-River Inflow Energy for Reducing Dimensionality in the Long-Term Generation Scheduling Problem

Hydropower Operation Optimization Using Machine Learning: A Systematic Review

Limitations of Modern Diagnostic and Prognostic Systems for a Hydraulic Unit’s Health

Contact Info

Product

Resources

About