Time series forecasting using ensemble learning methods for emergency prevention in hydroelectric power plants with dam
Introduction
Based on the historical water affluence forecast for a period, the daily plant operation program (DPOP) is created, which provides the necessary requirements to meet the contractual and control needs determined by the National Electric System Operator to meet the demand required for the period [1]. The problem of forecasting affluence in relation to the generation of energy is complex [2], given that its poor planning can cause high costs for the future electricity generation or in the spillage of water. This stochastic problem is evaluated through optimizers that view energy savings over time [3].
In an emergency in which the safety limit in relation to the reservoir capacity can be exceeded, the plant controls the reservoir using the plant’s operation manual. Considering that the control of the plant is directly linked to information from external bodies [4], [5]. If there is a delay in passing on this information in an emergency, the decision-making time will become short, which may expose the plant to a safety risk. The use of information recorded based on the region’s water affluence curve in relation to the volume of rain can be mathematically modeled to predict risk situations and thus alert operators so that they are prepared for decision-making in advance.
Some researchers conducted promising studies focused on time series forecasting to assess the load on the electrical system. Kaboli et al. [6] showed that load forecasting can improve the operation of the electrical system, considering that the electrical generation must follow electrical load demand. According to Kazemzadeh et al. [7] load forecasting is one of the main studies necessary for the planning and expansion of the electrical power system, in this approach, techniques combined with optimization algorithms are a promising alternative for the evaluation in question. The uncertainty associated with the need for generation leads to problems in the management of smart active distribution networks, so an accurate forecasting model can improve planning for the operation of power systems [8].
Regarding time series forecasting, fuzzy systems are being used by some authors [9]. In [10] a fuzzy interval model based on multiple time-frequency spaces and the induced ordered weighted average aggregation operation is used, the signal is decomposed into empirical modes of ensemble and is reconstructed as several time-frequency spaces for visibility graphs. The results presented using the proposed model prove to be efficient for solving the problem, being more promising than well-consolidated models. Tanuwijaya et al. [11] used a first-order single-valued neutrosophic hesitant fuzzy time series for stock index forecasting in Indonesia and Argentina. The presented results show that the proposed method outperforms the classic fuzzy time series.
The use of techniques based on deep learning for time series forecasting has been growing considerably recently [12]. Kasburg and Stefenon [13] used the long short-term memory (LSTM) to predict the generation of photovoltaic energy, with the objective of improving the capture of the sun and thus the efficiency in photovoltaic panels. In [14] a hybrid model with stacked residual LSTMs and adaptive grouping of sub-series is applied to predict multivariate time series. The results of applying the model are superior in the effectiveness in public data sets when compared to other algorithms.
With the need to predict failures in the electrical power system to keep the system working, the ensemble learning model has promising results for assessing chaotic time series [15]. The ensemble learning model is based on the divide-to-conquer paradigm, devoted to improving the models’ accuracy in classification [16], clustering, or regression tasks [17], [18]. Several weak or base models individually perform a specific task, and when its results are combined, an efficient model in terms of accuracy is obtained [19].
Specifically for flood forecasting in hydroelectric power plants, Davydov et al. [20] assessed extreme water discharges from flood control facilities. Based on this assessment, it is possible to define modes of operation for a hydroelectric plant and flood control facilities. Hydroelectric plants have influence on flood control, as depending on their operation they can quickly increase the level downstream [21], [22].
Regarding the risk related to the use of dams with large flooded areas, Sidek et al. [23] conducts a study on the impacts of a possible rupture of a dam. From this study, the flow to the downstream area is evaluated to obtain an estimate of the flooded area. According to Assahira et al. [24] and Resende et al. [25], the construction of hydroelectric plants using a reservoir has major impacts on social, ecological, and environmental as it substantially changes the hydrological regime of a region.
Davydov et al. [26] present a work to flood control based on the calculation of the maximum flow rates for specific rivers. Using this approach it is possible to manage floods from the water reservoirs of the hydroelectric power station. Chusov et al. [27] developed a study about land protection from floods based on the use of reservoirs distributed as part of a hydrosystem with a hydroelectric power plant. This research shows the importance of the management of the water flow in hydroelectric reservoirs.
Applying the LSTM model that will be evaluated comparatively in this paper, Le et al. [28] perform flood forecasting according to daily discharge and rainfall. Accuracy can reach 99 % depending on the case evaluated, this shows that there is an influence on the accuracy in the prediction depending on the data set used since the same model only reaches 87 % for an evaluation of another data set. According to Razali et al. [29] to have efficient flood control it is necessary to study the inflow design to determine the dam structure.
The ensemble learning model is used for forecasting in several areas of knowledge [30], standing out for fault diagnostics and prognostics [31] as well as component failure [32]. This model can be used to assess the risks associated with establishing relationships with corporate partners to business failure prediction models [33]. Ensemble learning methods can be applied to assess the carrying capacity of reinforced concrete columns, which are a concern in structural design and also in performance evaluation procedures [34], it may also have a promising application for predicting the useful life of industrial components [35], which is an evaluation similar to the one being proposed in this paper.
Based on the characteristics of the ensemble models of having less computational effort and greater accuracy in certain applications, the bagging, boosting, random subspace, bagging plus random subspace, and stacked generalization ensemble models are evaluated in this paper. The general purpose of this application is to predict the emergency situations in hydroelectric power plants with dam, and thus, advance the necessary decision-making in a situation of risk to the dam. The forecast analysis was carried out based on the variation of 1-hour ahead, enough time to carry out preventive actions in a hydroelectric power plant.
The contributions of this paper for hydroelectric power plants are summarized in the following:
- •
The first contribution is related to the technical feasibility assessment of flood forecasting to the improvement of the safety of the hydroelectric plant through a model that can anticipate the decision-making of the plant management. Subsequent action results in a change in the emergency procedure, and this was not done in this research;
- •
The second contribution is related to the evaluation of the use of an ensemble learning model to forecast the inflows of floods. In this respect, the well-known bagging, boosting, bagging plus random subspace, random subspace, and stacking ensembles are combined with support vector regression to forecast the inflows of floods with high accuracy. Comparative applications of different ensemble models are rare, considering that specific models for specific solutions are typically explored;
- •
The third contribution is about the model’s ability to process raw data, which has nonlinearities. Even with raw data the model has higher accuracy, and less error that the ensemble models can have compared to well-established LSTM models. This shows the deep learning techniques that are being used widely today may have inferior results to the ensemble models depending on the application.
Table 1 defines the symbol used in this paper.
The remainder of this paper is organized as follows: In Section 2, the operation of the hydroelectric plant is detailed and the issue of water inflow is discussed. In Section 3, the proposed method is presented. In Section 4, the results obtained are presented and discussed. Finally, the conclusions are discussed in Section 5, about the applicability of the model.
Section snippets
Operation of a hydroelectric power plant
The operation of a hydroelectric plant is usually performed according to the DPOP. The DPOP is a daily electro-energetic program issued by the operation programming area of the National Electric System Operator. DPOP has the inclusions or changes in the operational restrictions of the national energy generation and transmission facilities, the forecast of daily load, the programming of automatic generation control, in addition to the forecast of the conditions for the operation of the
Ensemble learning models
An ensemble learning model is an approach based on the divide-to-conquer paradigm devoted to improving the models’ accuracy in classification, clustering, or regression tasks. Several weak or base models individually perform a specific task, and when its results are combined, an efficient model in terms of accuracy is obtained [19]. The success of ensemble learning methods is because each model learns different features of the data, and then when the results are combined, the whole pattern of
Analysis of results
In this section, the analysis will be carried out for each model based on the evaluation of the alteration of several parameters (according to Section 3.5). After the models’ training, the results of the best assessment for each model will be presented and discussed. For a better presentation, the best results are highlighted in bold. The first model to be evaluated is the bagging approach, which is presented in Table 3.
Considering the MSE and RMSE the results for the best configuration were
Conclusion
In this paper, we applied several ensemble learning models, and it was clear that these methods have promising results for forecasting time series. It is possible to foresee an emergency, based on the variation of the level of the dam, carrying out a preventive analysis in order to guarantee the safety of the installations of a hydroelectric plant. Based on an accurate forecast, it is possible to alert specific teams so that they are prepared for an adverse situation.
Among the models used, the
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
The authors would like to thank the National Council of Scientific and Technologic Development of Brazil – CNPq (Grants number: 307958/2019-1-PQ, 307966/2019-4-PQ, 404659/2016-0-Univ), PRONEX ‘Fundação Araucária’ 042/2018. The authors are thankful to the Coordination for the Improvement of Higher Education Personnel (CAPES), awarding a doctoral scholarship to one of the authors. The authors would like to thank the Canadian Bureau for International Education (CBIE), Government of Canada, who
References (79)
- et al.
Assessing solution quality and computational performance in the long-term generation scheduling problem considering different hydro production function approaches
Renew. Energy
(2019) - et al.
Long-term electrical energy consumption formulating and forecasting via optimized gene expression programming
Energy
(2017) - et al.
A hybrid data mining driven algorithm for long term electric peak load and energy demand forecasting
Energy
(2020) - et al.
A novel electrical net-load forecasting model based on deep neural networks and wavelet transform integration
Energy
(2020) - et al.
Ensemble approach based on bagging, boosting and stacking for short-term prediction in agribusiness time series
Appl. Soft Comput.
(2020) - et al.
Tree mortality of a flood-adapted species in response of hydrographic changes caused by an amazonian river dam
Forest Ecol. Manage.
(2017) - et al.
Massive tree mortality from flood pulse disturbances in amazonian floodplain forests: the collateral effects of hydropower production
Sci. Total Envir.
(2019) - et al.
Short-term forecasting COVID-19 cumulative confirmed cases: perspectives for Brazil
Chaos Solitons Fractals
(2020) - et al.
Failure mode classification and bearing capacity prediction for reinforced concrete columns based on ensemble machine learning algorithm
Adv. Eng. Inf.
(2020) - et al.
Ensemble of optimized echo state networks for remaining useful life prediction
Neurocomputing
(2018)
Spatiotemporal analysis of heavy rain-induced flood occurrences in Germany using a novel event database approach
J. Hydrol.
Multi-step ahead meningitis case forecasting based on decomposition and multi-objective optimization methods
J. Biomed. Inf.
Wavelet group method of data handling for fault prediction in electrical power insulators
Int. J. Electr. Power Energy Syst.
Forecasting Brazilian and American COVID-19 cases based on artificial intelligence coupled with climatic exogenous variables
Chaos Solitons Fractals
Intelligent impulse finder: a boosting multi-kernel learning network using raw data for mechanical fault identification in big data era
ISA Trans.
Stacked generalization
Neural Netw.
Predicting the number of hospital admissions due to mental disorders from air pollutants and weather condition descriptors using stacked ensemble of deep convolutional models and LSTM models (SEDCMLM)
J. Clean. Prod.
A novel support vector machine ensemble model for estimation of free lime content in cement clinkers
ISA Trans.
Echo state network applied for classification of medium voltage insulators
Int. J. Electr. Power Energy Syst.
A practical tutorial on the use of nonparametric statistical tests as a methodology for comparing evolutionary and swarm intelligence algorithms
Swarm Evol. Comput.
Two-stage and multi-stage decompositions for the medium-term hydrothermal scheduling problem: a computational comparison of solution techniques
Int. J. Electr. Power Energy Syst.
Comparing spatial and scenario decomposition for stochastic hydrothermal unit commitment problems
IEEE Trans. Sustain. Energy
Mixed-integer nonseparable piecewise linear models for the hydropower production function in the unit commitment problem
Electr. Power Syst. Res.
Solving stochastic hydrothermal unit commitment with a new primal recovery technique based on lagrangian solutions
Int. J. Electr. Power Energy Syst.
Probabilistic forecasting with fuzzy time series
IEEE Trans. Fuzzy Syst.
A fuzzy interval time-series energy and financial forecasting model using network-based multiple time-frequency spaces and the induced-ordered weighted averaging aggregation operation
IEEE Trans. Fuzzy Syst.
A novel single valued neutrosophic hesitant fuzzy time series model: applications in Indonesian and Argentinian stock index forecasting
IEEE Access
Deep neural networks for multivariate prediction of photovoltaic power time series
IEEE Access
Deep learning for photovoltaic generation forecast in active solar trackers
IEEE Latin Am. Trans.
A hybrid method with adaptive sub-series clustering and attention-based stacked residual LSTMs for multivariate time series forecasting
IEEE Access
A novel doublet extreme learning machines for delta 3D printer fault diagnosis using attitude sensor
ISA Trans.
A novel decomposition-ensemble learning framework for multi-step ahead wind energy forecasting
Energy
Ensemble deep learning with multi-objective optimization for prognosis of rotating machinery
ISA Trans.
Ensemble approaches for regression: a survey
ACM Comput. Surv.
The simulation model for a flood management by flood control facilities
MATEC Web of Conferences
Influence of hydroelectric power station on flood situation and modeling of hydrological characteristics
E3S Web of Conferences
Mathematical models of operating regimes of flood control facility’s system
Journal of Physics: Conference Series
Dam break modelling and flood hazard mapping for cascading dams in Sungai Perak hydroelectric scheme
International Conference on Dam Safety Management and Engineering
Mathematical simulation of flood management by hydro systems with temporarily flooded reservoirs
Energy Management of Municipal Transportation Facilities and Transport
Cited by (60)
Centralized decomposition approach in LSTM for Bitcoin price prediction
2024, Expert Systems with ApplicationsA solar radiation intelligent forecasting framework based on feature selection and multivariable fuzzy time series
2023, Engineering Applications of Artificial Intelligence