2. Literature Review on Enhancing Hydroelectric Power Plant Efficiency
With the advancement of technology, machine learning perceives and analyzes data to solve real-world problems. Artificial neural networks, on the other hand, can learn from data to recognize patterns and partition the data into layers of abstraction [
4].
Hydrological issues can be addressed through mathematical models that are computed with the assistance of computers. Artificial neural networks are employed to generate solutions for intricate problems related to hydrological phenomena [
5]. Optimizing energy production and planning processes in a hydropower plant is a significant and complex challenge, considering technological, economic, physical, and environmental factors. In this study, it is shown how to produce energy most efficiently in a HEPP.
The two most frequently mentioned approaches for enhancing power efficiency are managing reservoir operations and regulating water flow. This process involves various uncertainties, including volatile demand, water availability, weather conditions, and financial considerations [
6].
The amount of energy produced is closely linked to water pressure and the power generated at the turbine shaft. This mechanical power can be calculated using the
P Equation (1) [
6].
where ηt = hydraulic efficiency of the turbine, ρw = water intensity, g = acceleration due to gravity, Q = flow rate of water acting on the turbine, h = height of water acting on the turbine.
Equation (1) shows that the power produced at the generator shaft is directly reliant on the flow of water. A hydropower plant experiences variability in water flow throughout the year. Effectively utilizing this incoming water is crucial for optimal operation. Kayhomayoon et al. employed machine learning models to predict energy production at a hydroelectric power plant, achieving the highest accuracy with the adaptive neuro-fuzzy inference system (ANFIS) method [
7]. Coulibaly, Anctil, and Bobée used an artificial neural network model to predict energy inputs in hydroelectric power plants [
8]. Hussin et al. used an artificial neural network to estimate electricity production and water consumption in a hydroelectric power plant [
9].
Barzola-Monteses et al. utilized artificial neural networks (ANNs) to forecast hydropower generation, with model parameters optimized through a grid search algorithm [
10]. The resulting model proved to be a dependable tool for energy management.
Gao et al. introduced an ANN model for forecasting power one day in advance [
11]. Rahman et al. developed LSTM, convolutional neural network (CNNE), and recurrent neural network models (RNN) to predict power energy [
12].
Liu et al. presented a study at the Manwan Hydroelectric Power Plant in Yunnan Province, China, where the application of a genetic algorithm integrated with dynamic programming increased flexibility in output allocation, reducing leaks by 79% and expanding high-efficiency zones by 43% [
13]. Xiao et al. developed the butterfly optimization algorithm (BOA), a newly developed meta-heuristic method that is widely used in solving various optimization problems to realize the optimal operation of the reservoir [
14]. Grini et al. proposed a new approach called the state-of-the-art algorithm stochastic dual dynamic programming STRO to solve the mid-term medium-term reservoir management problem [
15]. Chen et al. established the coordination model of electricity generation and ecological flow using the multi-objective genetic algorithms (MOGAs) method for the operation of the cascade hydropower system [
16]. Ehtearm et al. presented a hybrid deep learning model to predict hydropower production [
17]. He et al. proposed a multi-objective optimization operation model to optimize the seasonal flood-limited water levels (FLWLs) of the reservoir in order to maximize hydropower benefits and reduce flood risk [
18]. In order to benefit from the advantages of hydropower, Shen et al. developed a two-layer planning optimization model for the cascading hydro-PV complementary system, considering the energy market [
19].
In energy markets, hourly energy prices are set according to the supply-demand balance. There is a direct correlation between demand and energy price. Estimating the market clearing price (MCP) is a very important factor for all energy producers in terms of increasing the profitability of businesses. Estimating the price will allow businesses to plan their production strategies. With this feature, the study provides a great advantage to the power plant by providing forecasting foresight.
Li et al. presented a multi-stage, risk-neutral preventive maintenance planning model for a hydropower producer priced in a deregulated market by applying the Stochastic Dual Dynamic Integer Programming (SDDIP) algorithm [
20]. Li et al. propose a stochastic hydropower unit commitment (SHUC) model to maximize the hydropower producer’s total revenue, including current revenue, future revenue (i.e., opportunity cost), and startup and shutdown costs [
21]. Considering multiple price determinants, time matching and hydropower system characteristics with significant difficulties, Liu et al. proposed a new methodology for the long-term auction strategy of cascade hydropower stations (CHSs), a price determinant based on supply function equilibrium (SFE) [
22]. A literature comparison is given in
Table 1.
Power generation plants have start-stop costs. This price varies according to the type of facility. In a study on the start-stop cost of hydrogen generators in HEPPs, Osburn et al. found that the units in the power plant would cost between USD 274 and USD 411 per start-stop [
31]. At this point, the study increases the operational life of the mechanical systems and increases the economic benefits by preventing unnecessary start-stops of the power plant by making the right planning.
Bicil conducted a doctoral thesis on Pricing in the Electricity Market, specifically Price Forecasting in the Turkish Electricity Market [
32]. The study evaluated both the effect of the estimation method on estimation performance and the effect of data properties on estimation performance. Demirezen made an estimation of electricity prices for the day-ahead market in Turkey [
33].
Başoğlu and Bulut developed a hybrid system by supporting ANNs with expert systems [
34]. This system was trained using data from the last 10 years. Fan et al. examined an integrated machine learning model for predicting day-ahead electricity prices [
35]. Rodriguez et al. conducted Forecasting Energy prices using neural networks, fuzzy logic, and their combination [
36]. In their study, they used a four-layer perceptron neural network consisting of an input layer, two hidden layers, and an output layer. Li et al. estimated electricity price study in the grid environment [
37]. Cui and Song conducted research on electricity price estimation based on chaos theory [
38]. Vahidinasab et al. conducted a day-ahead price estimation study in reconstructed power systems using ANN [
39]. In their study, they used a modified Levenberg–Marquardt (LM) learning algorithm to estimate prices in the Pennsylvania-New Jersey-Maryland (PJM) market. Zhang and Cheng estimated day-ahead electricity prices by using artificial intelligence [
40]. Tang and Gu estimated day-ahead electricity prices using ANNs [
41]. Yan and Chowdhury estimated electricity market clearing prices in an unregulated electricity market, aiming to estimate the medium-term market clearing price and decide on, program, and plan a bidding strategy [
42]. Singhal and Swarup stated that estimating electricity prices was a difficult task for online trade and e-trade and used a neural network approach to estimate marketing behavior based on historical prices, amounts, and other information in order to estimate future prices and units [
43]. Özgüner conducted a study to estimate short-term electricity prices in the Turkish electricity market [
44]. Anbazhagan and Kumarappan estimated the day-ahead market clearing price by using the discrete cosine transform (DCT) as input to the neural network [
45]. Geidel and Zareipour estimated the Spanish day-ahead electricity market price by analyzing the long-term effect of wind power on the electricity market [
46]. Kölmek and Navruz estimated the day-ahead price in the electricity balancing and settlement market in Turkey using ANNs [
47]. Sahay and Tripathi estimated the short-term price of the electricity market analysis by using ANNs [
48]. Kotur and Zarkovic estimated short-term and long-term electricity prices and loads using neural network models [
49]. Keles et al. estimated day-ahead electricity spot prices by applying ANNs [
50].
A neural network model was developed in the present study for energy planning and estimating the MCP of a HEPP.
This study proposes a method for operating a specific hydroelectric power plant with maximum economic efficiency using real data. The proposed method is based on the principle of employing two separate artificial neural network models in tandem. The first model utilizes collected environmental data (flow rate time series, reservoir active volume) to predict the water accumulation in the reservoir of the dam for the next day. The second model uses the output of the first model and energy market data as inputs to optimize the sale of energy generated by the dam for maximum profit throughout the day. It is observed that with the proposed method, the profitability of the dam increases by 37.21%.
4. Conclusions
This study investigates the use of artificial intelligence to control the operation of the dam reservoir of HEPPs to operate at maximum efficiency. No similar studies have been found; this study will be useful in resolving uncertainties regarding resources and production in a HEPP. Two ANN models were used in the study, and the contribution and results of the ANN models are evaluated. Additionally, the study was tested by applying it to different ANN models, and the best result was achieved with the MLP Levenberg–Marquardt (LM) algorithm, which is the ANN training function.
The first phase of the study used an ANN to estimate the water coming into the dam reservoir of a HEPP, which predicted incoming water with 96.5% accuracy. Estimations on successive days showed that the error rate would increase for the first few hours following a sudden change of incoming water, and then the error rate would return to the expected value. It is expected that if the data set used in ANN training is enriched, the predictive ability would increase and the error rate improve, and long-term data collection would make prediction of incoming water more reliable and accurate.
The second phase of the study estimated the MCP to determine the production capacity that provided the most efficient outcome in the energy market. The short-term data were evaluated hourly; the average hourly estimate was 90% of MCP. When evaluated on a daily basis, the estimate was 95% of MCP. When evaluating the benefits provided by the study, for example, on December 6, 2021, it was observed that the predictions made by the YSA model enabled the operation to achieve 37.21% higher production revenue. This study provided the Darıca-2 HEPP operation directorate with the opportunity to make energy production plans and energy sales with maximum profitability.
It was concluded that when the estimate of water and MCP were made together as a hybrid study of a HEPP, production planning was made with an accuracy of 95%. The study has determined that utilizing the maximum amount of data possible for water forecasting and training the artificial neural network (ANN) for short-term market clearing price (MCP) prediction with the most up-to-date data leads to much more accurate results.
