A varying comprehensive hydropower coefficient for medium/long-term operation of a single reservoir

The following symbols are used in this paper:

A = total output numbers for the entire plant;

A_i = output percent in interval i (%);

a_i = output numbers in interval i;

B = total output numbers of the plant in the interval i;

B_j = output percent in the j category (%);

b_j = total output numbers of the j category;

E₀ = actual power generation (billion kWh);

E_k = benchmark power generation (billion kWh);

H = water head (m);

i = index of the output intervals;

I_m = inflow in the m period (m³/s);

j = index of the unit categories;

k = comprehensive hydropower coefficient for the plant;

k_ij = varying values of the j type unit in the i interval;

M = output levels;

m = numbers of periods;

N = output (kW);

P = unit categories;

P_m = output in the m period (kW);

PU_m = upper limits of the output (kW);

PL_m = lower limits of the output (kW);

Q_m = outflow in the m period (m³/s);

QU_m = upper limits of the outflow (m³/s);

QL_m = lower limits of the outflow (m³/s);

q = power discharge (m³/s);

V_m = storage capacity in the m period (m³);

VU_m = upper limits of the storage capacity (m³);

VL_m = lower limits of the storage capacity (m³);

Due to the limitations of traditional fossil energy (Ming et al. 2017), renewable energy is playing an increasingly important role in energy supply. Water power is a typical example of renewable energy, especially in China. There are many studies on improving the efficiency of hydroelectric power production to increase the availability of energy (Inglesi-Lotz & Blignaut 2014), and reservoir operation is a key issue for improving hydropower efficiency. A large number of these studies have focused on the optimization method for reservoir operation (Liao et al. 2017; Feng et al. 2018), while the hydropower efficiency evaluation is seldom studied (Chang et al. 2017).

In medium/long-term reservoir operation (Uen et al. 2018), hydropower output N is calculated by k × q × h, where q is the power discharge, h is the water head, and k is the comprehensive hydropower coefficient, which indicates the conversion efficiency from water power to electricity (Hidalgo et al. 2009). The coefficient k equals the gravitational constant multiplied by the efficiency of the turbine-generator set (Hidalgo et al. 2014), and is a key parameter in the simulation of hydropower generation. The k value is usually taken as a fixed constant for the entire plant to simplify calculations (Hidalgo et al. 2012). However, the k value varies among different units, and with the water head and hydropower discharge (Liu et al. 2012).

Much research has been done on identifying the factors that determine k. For example, Liu et al. (2012) identified influencing factors and Xu et al. (2017) did a sensitivity analysis on units that affect the k value. Diniz et al. (2007) and Cordova et al. (2014) studied the hydraulic efficiency of turbines and established a relationship between efficiency and h and q (Finardi & da Silva 2006; Diniz et al. 2007; Cordova et al. 2014). In these methods, the relationship between k and h, and k and q were based on direct estimation from observed data. However, an average value of h does not meaningfully describe the variation of hydropower plants. Therefore, these methods are restricted in applicability to short-term operations, and are difficult to be used for medium/long-term operations.

Other studies have focused on adjusting the efficiency curves or exploring the performances of the units. For example, Barros & Peypoch (2007) adopted a random cost frontier method to analyze technical efficiency. Barros (2008) sought to identify the best practices for improving the performance of hydroelectric generating plants. Hidalgo et al. (2010) used a simulator to improve the efficiency and reliability of plant data. Cordova et al. (2014) described a system to calculate the efficiency of each individual unit, while Li et al. (2014) described a three-dimensional interpolation technique to represent the generation function of each individual unit. Hidalgo et al. (2014) used an iterative calculation to obtain the efficiency functions. However, the improvement in accuracy of the efficiency curves was used for a single unit type, and could not be applied to the entire hydropower plant operation.

Advances in technology and communication have led to big data research. Big data form a good basis for overcoming the limitations of existing methods, and for investigating the variation rules for k. Therefore,the purpose of this study is to develop a novel method for obtaining the variations in k using observed big data. Our intention is that the proposed method should reduce the generation simulation and efficiency evaluation errors for medium/long-term and short-term operations. Our specific aims are to: (1) determine a comprehensive varying k using a temporal-spatial aggregation estimation method, (2) test the derived k using the observed big data, and (3) re-evaluate reservoir operation benefits using the derived k. The rest of the paper is organized as follows. The ‘Methodology’ presents two estimation methods to generate the k–h relationship for an entire hydropower plant. ‘Case study’ describes a case study using the Three Gorges Project (TGP), China. In ‘Results and discussion’, the varying k value is tested for different time periods, and its impacts on reservoir operation efficiency are evaluated. ‘Conclusions’ summarizes the conclusions of this study.

Two estimation methods for determining a varying k (the direct method and the temporal-spatial aggregation method) are presented for comparative purposes (Figure 1). The direct method is carried out using observed output, water head, and discharge data (section: ‘The direct estimated method’), while the temporal-spatial aggregation method built the k–h relationship using the efficiency curve (it is also called hill chart) and the observed big data (section: ‘The temporal-spatial aggregation estimated method’). These two methods are then tested for hydropower simulation using an error analysis.

However, the water head varies with time, and in the medium to long term an average value is unable to meaningfully describe the processes of the hydropower plant. Thus, the k value derived using Equation (1) is not always suitable since the value of h is difficult to be determined. Therefore, different k values should be derived at different time scales, respectively.

The efficiency curves of plant units are used to improve the estimation accuracy of k. The k value is obtained by interpolation from efficiency curves when h and q are known.

An estimation is implemented to test the derived k–h relationship. The estimation is obtained for the direct and for the temporal-spatial aggregation methods. The accuracy of the varying coefficient is evaluated by comparing the estimated results with the observed power output. The power generation, the error percentage of the power generation, the mean error percentage of 10 days, and the root mean square error (RMSE) are chosen as the evaluation criteria.

The TGP is, currently, the largest multipurpose hydro-development project ever built. The TGP is crucial for the water resources development of China's largest river, the Yangtze River (Figure 2). The benefits provided by the TGP include flood control, power generation, and navigation improvement. The TGP receives inflow from a 4.5 × 10³km long channel with a contributing drainage area of 1 × 10⁶km². The mean annual runoff at the dam site is 4.51 × 10¹¹ m³. The flood storage capacity of the TGP is very large (2.215 × 10¹⁰ m³), and plays a very important role in the flood control of the Yangtze River. The TGP has 32 sets of 70 × 10⁴ kW hydraulic turbo generators, which equates to 22.50 × 10⁴ kW in total installed capacity. The TGP produces an annual electricity output of 8.468 × 10¹⁰kWh, and a large proportion of this is used to supply eastern and central China. The TGP also improves the navigation conditions downstream in the dry season, as well as in the 660 km-long waterway upstream, and improves downstream water quality during the dry season. In addition, the TGP enhances fish habitats in the reservoir, as well as enhancing tourism and recreational activities (Liu et al. 2011). Table 1 lists the TGP parameters (Li et al. 2010).

Although the hydropower units have the same rated power, they are produced by four different manufacturers. Based on the manufacturer and the time of installation (Table 2), the units were classified into eight categories, each with a slight difference of performance.

Observed data from 2015 to 2018 were used to illustrate the method. The observed data include: water levels in the upstream and downstream of the reservoir at 5 minute intervals, hydropower for each unit at 5 minute intervals, and the power flow of the entire plant every 2 hours. 1,280,000 observed data were used to derive the varying k value. This is a typical example of combining big data with a case study, and thus obtaining more reliable results.

The data during 2015–2016 were used to estimate the varying k, while the data during 2017–2018 were used to test the estimation accuracy by comparing the estimated and observed hydropower generation.

The efficiency curves of the units were used to estimate k (Figure 3).

In addition, the maximum output of the entire plant, the relationship between water levels and reservoir capacity, and the relationship between release and downstream water levels were used in the reservoir simulation and optimization.

The varying k was obtained by applying the direct method to the 2015–2016 data. Due to the characteristic of medium/long term operation, 10 days is selected as the time scale for the calculation, the average value of N, q, and h within 10 days is solved, and the corresponding k value is obtained according to the formula; the 72 points are shown in Figure 4. It is clear from the calculated results that there is a significant relationship between k and h. As a result, a function curve was used to fit the relationship between k and h as shown in Figure 4. According to the water head of the power station, the corresponding k value can be obtained by the fitting curve, and then the power generations are estimated.

However, the correlation is not perfect, and this fitting may decrease the estimation accuracy. Furthermore, the average water heads were used instead of the actual water heads, and this led to errors in the k value.

Each unit's output ranges from 0 to 70 × 10⁴ kW. However, in actual operations the output was seldom less than 50 × 10⁴kW. Therefore, output ranges from 50 × 10⁴ kW to 70 × 10⁴kW were used.

There were a total of five levels, and these formed four intervals. Using the observed data from 2015 to 2016, the actual output operation times for the different intervals were counted and the frequencies were used as weights (Table 3). Then the weights of the eight categories are counted according to the observed data, as shown in Table 4.

For the i interval, the value of k_ij was calculated using the efficiency curve, as shown in Figure 5. The k_i value was calculated using k_ij and b_j for interval i, as shown in Figure 6. The varying k value was then obtained using k_i and a_i for the entire plant.

Finally, the relationship between k and h was derived (Figure 7). According to the observation data, the water head at the plant was 76 m–110 m. Thus, the k–h relationship was estimated only for this range. It is observed that a water head of 94 m is a critical point. The value of k increases with increasing water head before this point, and then decreases after this point.

The k value can be determined based on the observed water head. Combining with observed power discharge, the estimated outputs can be obtained. Ten days (a traditional Chinese time period) was chosen as the time interval in the computation.

The estimated results for 2017–2018 were compared with the observed power generation. Differences between estimated and observed values are shown in Table 5. An estimation using a constant k value has been included to demonstrate the inherent characteristic of k, and to highlight the necessity of using a varying k value.

It is clear that the accuracy of the temporal-spatial aggregated k is higher than that of the fixed k and the direct k. As shown in Table 5, the annual power generation errors for the temporal-spatial aggregated k are only slightly greater than 0.7%, but the generation errors for the fixed k exceed 1.9%. In addition, the RMSE of the temporal-spatial aggregated k is smaller than that of the fixed k, indicating a good estimation.

The accuracy of the direct estimation method is the lowest of the three methods, and the error percentages for the entire year and for the mean of each 10 days were both almost 2.5%. The RMSE for the direct estimation method is much higher than for the other estimation methods. Thus, it can be concluded that the directly determined k is not the most suitable for power generation simulation, and therefore temporal-spatial aggregation was chosen as the estimation method for the varying k.

As shown in Figures 8 and 9, the relative error has huge differences in the flood period, thus the estimated accuracy of the k value is lower than in the non-flood period. On the contrary, the direct estimation of the k value in the flood season is much better than the other two estimation methods. It can be considered to combine the temporal-spatial aggregated method and the direct-determined method during the year to further improve the estimation accuracy of the k value.

The k varies constantly according to what has been described above. Therefore, it is necessary to investigate the relationship between the benefits' evaluation and the time interval, and to determine whether a shorter interval results in a better estimation. It is also necessary to investigate the stability of the proposed method under different conditions. Repeated estimations were carried out for three time intervals: 1 day, 10 days, and 1 month.

Comparing with the observed data, the estimated annual power generation was almost the same for all three time intervals (Table 6). The mean error percentage for daily intervals was the largest. These results ran contrary to expectations. The estimations for monthly intervals yielded a reasonable mean error percentage and a variance that was the highest of the three intervals.

It is concluded that 10-day intervals are optimal for the benefits' evaluation process. Both the estimated errors and the details are reasonable. Monthly intervals are useful when the operating period is long. Daily intervals are efficient when a high evaluation accuracy is required. Different time intervals could be chosen based on the concrete demand.

As shown in Figure 10, the conventional operation of the Three Gorges Power Station is implemented using reservoir operating rule curves (Feng et al. 2017).

Table 7 lists a comparison of two k value approaches for hydropower generation based on the observed data. When the k value is a fixed constant of 9.0, the calculated energy produced in 2017 and 2018 is 92.31 and 93.26 billion kWh, respectively. When a varying k value is used, the calculated energy produced is 91.81 and 93.49 billion kWh, respectively. The observed power generation in 2017 and 2018 was 97.61 billion kWh and 101.62 billion kWh, respectively. It is indicated that the actual operation is superior to the conventional rules by 6%–8%. However, the RHUR for the varying k is different to the fixed k. It is indicated that the improvement in benefit should be different to that obtained in the fixed evaluation. The varying k produces a more realistic hydropower generation by improving the estimation accuracy.

The optimal reservoir operation can be used as a benchmark to evaluate the potential benefit of hydropower generation. The optimal reservoir operation model is built with the single objective function of maximizing hydropower generation. The dynamic programing algorithm was used to solve the optimization model based on the historical inflows and the results were compared with the observed data. Moreover, the following constraints were considered in optimal operation:

1. Water balance constraint:

   

   (6)

   where m is the number of periods, V_m is the storage capacity in the m period, and I_m and Q_m are the inflow and outflow of the hydropower station, respectively.
2. Storage capacity constraints:

   

   (7)

   where VL_m and VU_m are the lower and upper limits of the storage capacity of the hydropower station, respectively.
3. Release constraints:

   

   (8)

   where QL_m and QU_m are the lower and upper limits of the outflow of the hydropower station, respectively.
4. Output constraints:

   

   (9)

   where PL_m and PU_m are the lower and upper limits of the output of the hydropower station, respectively.

5. Non-negative constraints and other constraints.

Since the actual water level is higher than the flood-limited water level (the conventional upper water level boundary) during flood seasons (Liu et al. 2015), two water level boundaries were available as constraints for evaluating potential hydropower benefits. According to the design requirements of the power station, in the non-flood season, the upper and lower water levels are 175 m and 145 m, respectively. In the flood season, the feasible interval ranges from 140 m to 145 m. The optimal operation results for 2017 and 2018 were obtained using deterministic dynamic programming.

The results of the optimal model with different upper water level limits are shown in Table 8. It can be found that the corresponding RHUR values are negative based on the observed data, indicating that the actual operation does not produce maximum power, and that there is considerable potential for increasing the hydropower generated. The results of the varying k are about 1.7% higher than the results of the fixed one, which indicates that the varying k can accurately evaluate the optimal space of operation benefits and the optimal operation is sensitive to the k value.

Table 8 also lists the optimal results within actual operation boundary of water levels. Due to the increased water levels, the power generation potential both noticeably increased in new intervals for 2017 and 2018 than the designed one. In 2017 and 2018, the benefits' rate with the actual operation water level was 0.72% and 0.65% higher than for the designed operation water level for the varying simulation. This clearly illustrates that potential hydropower generation is higher with different operation water level boundaries.

This study proposed the temporal-spatial aggregation method for estimating varying k to minimize the errors in hydropower generation simulation for medium/long-term operations. The k–h relationship was obtained using unit efficiency curves and observed big data. The TGP was used as a study case. It was concluded that:

1. The method proposed was successfully used to determine the varying k for the entire power plant for medium/long-term reservoir operations. The temporal-spatial aggregation method produced better results than the direct estimation of k method, and is suitable for calculations in any time scales.

2. Using the varying k value improved the simulation accuracy. Using a time interval of 10 days resulted in the relative highest accuracy in hydropower generation simulation.

3. The derived k plays an important role in benefit evaluation of reservoir operation. Conventional power generation operations can be improved by using the varying k. The potential for improvement is large.

The method proposed can be applied to other hydropower plants to improve hydropower simulation and to provide a more balanced benefit evaluation. However, further research is needed to assess measurement errors.

All data used during the study are proprietary and may only be provided with restrictions (e.g., anonymized data). Including the observed big data, the operation parameters, the maximum output of the entire plant, the relationship between water levels and reservoir capacity, and the relationship between outflow and downstream water levels were used in the reservoir simulation and optimization. All of the data are illustrated in the section ‘Data description’.

This study was supported by the National Key Research and Development Program of China (2016YFC0402202), the National Natural Science Foundation of China (U1865201) and the Innovative Research Groups of the Natural Science Foundation of Hubei, China (2017CFA015). The authors thank the editor and anonymous reviewers for their comments that improved the paper. The authors declare that they have no conflict of interest.