Technical Measures to Mitigate Load Fluctuation for Large-Scale Customers to Improve Power System Energy E ﬃ ciency

: Industrial equipment such as electric arc furnaces and steel mills are often associated with rapid and high load disturbances, so their power systems require additional control equipment to limit the frequency. However, proper ancillary service fees are not paid in these cases with extreme and variable load demands. The frequency regulation reserve equipment adds to power generation costs. Therefore, variable power generation loads lead to increases in the cost of energy production. We propose a load frequency control method that is applied on the customer end instead of the power supply end to reduce the operating reserve required to improve the energy e ﬃ ciency of the power system. We analyzed the load ﬂuctuation of steel mill customers using real data sampled at two-second intervals from the energy management system in Korea. We developed an automatic generation control program to simulate the power system’s frequency characteristics. We also propose compensation techniques for mitigation of the system’s frequency deviation at the customer end based on an energy storage system, pump storage hydro generator, customer generator, and plant process adjustment. To recover the frequency deviation, we calculated the compensation facility capacity and analyzed static characteristics, and we proved the feasibility via simulations.


Introduction
Power system operator adjusts the schedule for power feeding to meet both the power demand and supply demand with the aim of managing power systems stably [1][2][3]. Generally, the demand forecasting for dispatch is based on one-hour estimations to facilitate planning of the generation schedule for the next day's market. However, the power demand during the power system operation is fluctuated greatly by the short-period load variation within 1 h [4][5][6]. These intermittent load fluctuations may cause critical unbalance problems in operating power systems [7][8][9]. Therefore, the power system operator purchases ancillary service at a high price and provides such services for frequency following for the stabilization of power system [10,11].
The ancillary service is an indispensable function in terms of maintaining the quality of electricity. In addition, if this value added is not provided, it is impossible to maintain reliability by stabilizing the power system [12][13][14][15]. The ancillary service is the concept of power supply and demand regulation service, and there are services such as load following and energy imbalance as operation control to secure a supply and demand balance [14][15][16][17][18][19][20][21].
The supplemental reserve consists of operational and standby reserves, and the operational condition corresponds to power being generated that is already synchronized with the ISO controlled grid. The reserve is 1000 [MW] under normal operation and 1500 [MW] in an emergency. The allocated reserve for standby conditions is 1500 [MW] under normal operation and 1000 [MW] in winter and summer.
The spinning reserve [32] corresponds to the sum of the regulating reserve and supplemental reserve. During operation, this reserve is 2500 [MW]. The non-spinning reserve is 1500 [MW]. The peak demand in 2013 was approximately 76.5 [GW], and the spinning reserve accounted for 3.1% of the peak load.

Load Fluctuation of Steel Mills in Korea
The daily load curves of the total power system and the steel manufacturing industries in winter, spring and summer are shown in Figure 1a,b, respectively. This load data is recorded every 2 s from the EMS. The load curve of the total power system shows an accumulated load characteristic due to the overlap of the various loads.
The variable loads have similar patterns for hourly, daily, weekly, monthly, and yearly periods, but patterns may become more random due to the fluctuating demand loads of industry and increasing environmental effects and human behavior. The smaller number of demand loads has a noisy cyclic complex curve while the larger number of loads has a fairly smooth curve. However, the fluctuation of the total demand load in the power system depends on that of the combination of individual loads.
Due to increased heater and air conditioning loads in winter and summer, the load demand is higher than that in the spring. If the individual load is separated from the total overlap load, the short-term load fluctuation appears to be serious. In particular, the load fluctuation for the steel manufacturing industry appears to be large, as shown in Figure 1b. This is caused by the instantaneous operation of many different facilities used for arc furnace, rolling, etc. This load curve shows a similar pattern in winter and summer. On the other hand, during the daytime in spring, the load curve shows a high power consumption pattern. Figure 1b shows higher power consumption at daytime in spring than winter and summer, and at night time than during daytime in winter and summer, because winter and summer.
The spinning reserve [32] corresponds to the sum of the regulating reserve and supplemental reserve. During operation, this reserve is 2500 [MW]. The non-spinning reserve is 1500 [MW]. The peak demand in 2013 was approximately 76.5 [GW], and the spinning reserve accounted for 3.1% of the peak load.

Load Fluctuation of Steel Mills in Korea
The daily load curves of the total power system and the steel manufacturing industries in winter, spring and summer are shown in Figure 1a,b, respectively. This load data is recorded every 2 s from the EMS. The load curve of the total power system shows an accumulated load characteristic due to the overlap of the various loads.  If the daily load data in Figure 1 is analyzed by time period, the results show that various cycle components are combined, as shown in Figure 2. To analyze such cycle components, a frequency band analysis method based on weighted moving average (W-MA) and a method for calculating the spectrum density by way of fast Fourier transformation (FFT) coding of the auto-correlation function are used [33][34][35][36][37]. The results of these two different methods are known to be almost the same.
Energies 2020, 13, x FOR PEER REVIEW 4 of 28 The variable loads have similar patterns for hourly, daily, weekly, monthly, and yearly periods, but patterns may become more random due to the fluctuating demand loads of industry and increasing environmental effects and human behavior. The smaller number of demand loads has a noisy cyclic complex curve while the larger number of loads has a fairly smooth curve. However, the fluctuation of the total demand load in the power system depends on that of the combination of individual loads.
Due to increased heater and air conditioning loads in winter and summer, the load demand is higher than that in the spring. If the individual load is separated from the total overlap load, the short-term load fluctuation appears to be serious. In particular, the load fluctuation for the steel manufacturing industry appears to be large, as shown in Figure 1b. This is caused by the instantaneous operation of many different facilities used for arc furnace, rolling, etc. This load curve shows a similar pattern in winter and summer. On the other hand, during the daytime in spring, the load curve shows a high power consumption pattern. Figure 1b shows higher power consumption at daytime in spring than winter and summer, and at night time than during daytime in winter and summer, because demand side management (DSM) is carried out by time-of-use (TOU) pricing. The TOU standard of Korea is classified according to the season and time. Typically, the load levels are divided into three levels: light load level (23:00-09:00), heavy load level (09:00-10:00, 12:00-13:00, 17:00-23:00), and peak load level (10:00-12:00, 13:00-17:00).
If the daily load data in Figure 1 is analyzed by time period, the results show that various cycle components are combined, as shown in Figure 2. To analyze such cycle components, a frequency band analysis method based on weighted moving average (W-MA) and a method for calculating the spectrum density by way of fast Fourier transformation (FFT) coding of the auto-correlation function are used [33][34][35][36][37]. The results of these two different methods are known to be almost the same. The correlation between the spectrum density (SL(ω)) and angular frequency (ω) is shown as Equation (1). Disturbance has a different cycle depending on time. The results show a 15 min cycle standard deviation of disturbance, which is almost proportional to the square root of the system capacity, as shown in Equation (2).   The correlation between the spectrum density (S L (ω)) and angular frequency (ω) is shown as Equation (1). Disturbance has a different cycle depending on time. The results show a 15 min cycle standard deviation of disturbance, which is almost proportional to the square root of the system capacity, as shown in Equation (2). To analyze the load variability, the authors analyzed the frequency spectrum of the steel manufacturing load using FFT. If the daily load data is analyzed by spectrum analysis using FFT, the results show the characteristics of the frequency band, as shown in Figure 3. To analyze the load variability, the authors analyzed the frequency spectrum of the steel manufacturing load using FFT. If the daily load data is analyzed by spectrum analysis using FFT, the results show the characteristics of the frequency band, as shown in Figure 3.  Table 2 shows the correlation between the frequency band and the load fluctuation time. If we assume that the disturbance analyzed has a normal distribution, it can be presented as shown in Figure 4.    Table 2 shows the correlation between the frequency band and the load fluctuation time. If we assume that the disturbance analyzed has a normal distribution, it can be presented as shown in Figure 4.
X: Random variables; µ: Average data; σ 2 : Fluctuation data.    The distribution of the load fluctuation components is shown as the range of size of fluctuation components, and this can be shown by the range of standard deviation. Regarding the standard deviation adopted in this study, the variability of 2σ with a probability of 95.44% was used as the acceptable load fluctuation, so as to ensure compatibility. Table 3 shows the data distribution percentage. σ shows the probability of exceeding the data range, and the 3σ regulation implies that the range of σ standard deviations from the mean includes almost every value (99.7%). Standard deviation analyzes the difference between the size of each load fluctuation and the mean fluctuation, while the statistics are reported as means, thus providing the distribution of load fluctuation. In this paper, we calculated the load increase rate considering the load forecasting component, and this is used in the moving average (MA) method. The load fluctuation data with respect to time period is obtained by comparing the MA data with the increasing rate data.
The statistical equations to analyze the size of average load and load fluctuation are shown as Equations (4)- (7). The mean load capacity is calculated to analyze the range of load fluctuation, and the equation for calculating the mean is shown in Equation (4). Equation (4) is used to calculate the growth rate of the load capacity and the size of the load fluctuation, as shown in Equation (5). With respect to the temporal means, the size of load fluctuation is calculated using the difference between average load and load growth rate, and the load fluctuation for average load time can be calculated using Equation (6).
The distribution curve of load fluctuation was calculated using the equation for calculating standard deviation, Equation (7), where N is load average time and I is the size of real-time load fluctuation.
Energies 2020, 13, 4812 7 of 27 where 'N' is average time of load, and 'i' is the real-time load fluctuation value. Table 4 shows the calculation criteria for load fluctuations using the moving average method presented in this paper. We calculated the MA for the time period, and calculated the MA increase rate for 1 h, 10 min, and 30 s. The calculation of load fluctuation magnitude with an average time of 15 min was based on the average increase rate of 1 h. The calculation for fluctuation magnitude for average times of 10, 5 and 1 min were based on an average increase rate of 10 min, and the rest were based on the 30 s increase rate.
By using the load variability analysis method proposed, the size of load fluctuation in the daily load data can be determined for the steel manufacturing industry, major steel manufacturers, and representative steel manufactures. The magnitudes of load fluctuation for the total steel industry according to the load levels and the seasons are shown in Table 5.

AGC Program
PSS/E can simulate a power system's transient effects for only a few seconds, so we developed an AGC program called KEPCO Automatic Generation Control (K-AGC) to simulate the power system load fluctuation for longer than 10 min [38,39]. This program is composed of a text viewer, load-flow calculation, and AGC simulation and can simulate both the transient and steady-state behavior of a power system. Figure 5 shows a block diagram of the K-AGC program. It can accommodate several areas to configure a power system. Figure 6 shows the AGC configuration. One area is composed of SYS, LOAD, economic dispatch control (EDC), and ACE. SYS is a damping constant that represents the frequency characteristics for a difference between the generation and zone load, and LOAD is the load fluctuation for the frequency deviation. The AGC is executed for each zone, and every zone is connected to other zones by tie lines.

AGC Program
PSS/E can simulate a power system's transient effects for only a few seconds, so we developed an AGC program called KEPCO Automatic Generation Control (K-AGC) to simulate the power system load fluctuation for longer than 10 min [38,39]. This program is composed of a text viewer, load-flow calculation, and AGC simulation and can simulate both the transient and steady-state behavior of a power system. Figure 5 shows a block diagram of the K-AGC program. It can accommodate several areas to configure a power system. Figure 6 shows the AGC configuration. One area is composed of SYS, LOAD, economic dispatch control (EDC), and ACE. SYS is a damping constant that represents the frequency characteristics for a difference between the generation and zone load, and LOAD is the load fluctuation for the frequency deviation. The AGC is executed for each zone, and every zone is connected to other zones by tie lines.   Figure 7a shows a flow chart of the steady-state analysis. The program can calculate the frequency deviation for a load change and the load deviation for a frequency change. There are two modes in the steady-state analysis: governor-lock (GL) and GF modes. The GL mode considers only the load coefficient (KL), but the GF mode considers the generator coefficient (KG) and KL. Figure 7b shows a flow chart for the transient-state analysis, which has three modes: GL, GF, and AGC. The procedure for LFC is summarized as follows:

AGC Program
PSS/E can simulate a power system's transient effects for only a few seconds, so we developed an AGC program called KEPCO Automatic Generation Control (K-AGC) to simulate the power system load fluctuation for longer than 10 min [38,39]. This program is composed of a text viewer, load-flow calculation, and AGC simulation and can simulate both the transient and steady-state behavior of a power system. Figure 5 shows a block diagram of the K-AGC program. It can accommodate several areas to configure a power system. Figure 6 shows the AGC configuration. One area is composed of SYS, LOAD, economic dispatch control (EDC), and ACE. SYS is a damping constant that represents the frequency characteristics for a difference between the generation and zone load, and LOAD is the load fluctuation for the frequency deviation. The AGC is executed for each zone, and every zone is connected to other zones by tie lines.   Figure 7a shows a flow chart of the steady-state analysis. The program can calculate the frequency deviation for a load change and the load deviation for a frequency change. There are two modes in the steady-state analysis: governor-lock (GL) and GF modes. The GL mode considers only the load coefficient (KL), but the GF mode considers the generator coefficient (KG) and KL. Figure 7b shows a flow chart for the transient-state analysis, which has three modes: GL, GF, and AGC. The procedure for LFC is summarized as follows:  Figure 7a shows a flow chart of the steady-state analysis. The program can calculate the frequency deviation for a load change and the load deviation for a frequency change. There are two modes in the steady-state analysis: governor-lock (GL) and GF modes. The GL mode considers only the load coefficient (K L ), but the GF mode considers the generator coefficient (K G ) and K L . Figure 7b shows a flow chart for the transient-state analysis, which has three modes: GL, GF, and AGC. The procedure for LFC is summarized as follows: Figure 8 shows the developed AGC program. The text-based input window of Figure 8a is for the input of system model and initial data, and the execution window of Figure 8b is for the calculation of the size of load fluctuation and the results of frequency deviation. Step 1: Input power system data; Step 2: Set the reference frequency deviation; Step 3: Calculate initial load deviation; Step 4: Select simulation mode from among GL, GF, and AGC modes; Step 5: Calculate frequency deviation and determine whether frequency error is within the tolerance. If the error is less than the tolerance, go to step 6. Else increase/decrease load deviation by dP and go to step 4; Step 6: determine load deviation and calculate cost for ancillary service.    Step 1: Input power system data; Step 2: Set the reference frequency deviation; Step 3: Calculate initial load deviation; Step 4: Select simulation mode from among GL, GF, and AGC modes; Step 5: Calculate frequency deviation and determine whether frequency error is within the tolerance. If the error is less than the tolerance, go to step 6. Else increase/decrease load deviation by dP and go to step 4; Step 6: determine load deviation and calculate cost for ancillary service. We analyzed the results using the 1-bus stand-alone model and the data in Table 6 to confirm the reliability of the developed AGC program. As shown in Figure 9, the frequency deviation characteristics for the 500 [MW] load fluctuation at 10 s were the same as the PSS/E (version 33.4.0 SIEMENS PTI) results. The AGC program's accuracy and reliability were verified by comparing the calculation results with those of PSS/E [40].   Step 1: Input power system data; Step 2: Set the reference frequency deviation; Step 3: Calculate initial load deviation; Step 4: Select simulation mode from among GL, GF, and AGC modes; Step 5: Calculate frequency deviation and determine whether frequency error is within the tolerance. If the error is less than the tolerance, go to step 6. Else increase/decrease load deviation by dP and go to step 4; Step 6: determine load deviation and calculate cost for ancillary service. Energies 2020, 13, x FOR PEER REVIEW 10 of 28 We analyzed the results using the 1-bus stand-alone model and the data in Table 6 to confirm the reliability of the developed AGC program. As shown in Figure 9, the frequency deviation characteristics for the 500 [MW] load fluctuation at 10 s were the same as the PSS/E (version 33.4.0 SIEMENS PTI) results. The AGC program's accuracy and reliability were verified by comparing the calculation results with those of PSS/E [40]. (a) (b)

An Analysis of Frequency Characteristics
In this paper, we conducted a simulation by dividing the model into a single model and a network model according to the connections between generators and loads, as shown in Figure 10.
In addition, this model is divided into a single capacity system for individual companies, a business type capacity system, and a total system capacity system. The single model considers the generators owned by individual companies and the characteristics of the steel manufacturing load. The generators owned by the companies in Zone 2 of the single model are based on a locked output. With respect to the network model, Load_b in Figure 10b shows the steel manufacturing load, and Load_a shows the total load of the power system excluding the steel manufacturing load. Zone 1 comprises generation data and general load data and Zone 2 comprises steel load data, fluctuation data and the data of the generators owned by companies. (a)

An Analysis of Frequency Characteristics
In this paper, we conducted a simulation by dividing the model into a single model and a network model according to the connections between generators and loads, as shown in Figure 10. In addition, this model is divided into a single capacity system for individual companies, a business type capacity system, and a total system capacity system. The single model considers the generators owned by individual companies and the characteristics of the steel manufacturing load. The generators owned by the companies in Zone 2 of the single model are based on a locked output. With respect to the network model, Load_b in Figure 10b shows the steel manufacturing load, and Load_a shows the total load of the power system excluding the steel manufacturing load. Zone 1 comprises generation data and general load data and Zone 2 comprises steel load data, fluctuation data and the data of the generators owned by companies.
In a simulation model, the susceptance of the tie line, the maximum power demand, the load level, the reserve, and the self-generation capacity data are needed.
The inertia coefficient of the generators was calculated as 4.85 using Equation (8).
M i : Inertia coefficient of the generator; P i : Capacity of the generator; P total : Total capacity of the generator. owned by individual companies and the characteristics of the steel manufacturing load. The generators owned by the companies in Zone 2 of the single model are based on a locked output. With respect to the network model, Load_b in Figure 10b shows the steel manufacturing load, and Load_a shows the total load of the power system excluding the steel manufacturing load. Zone 1 comprises generation data and general load data and Zone 2 comprises steel load data, fluctuation data and the data of the generators owned by companies. In a simulation model, the susceptance of the tie line, the maximum power demand, the load level, the reserve, and the self-generation capacity data are needed.
The inertia coefficient of the generators was calculated as 4.85 using Equation (8).

= × (8)
Mi: Inertia coefficient of the generator; Pi: Capacity of the generator; Ptotal: Total capacity of the generator.
Regarding the susceptance of the tie line, the fault capacity was calculated using load flow data, and Thevenin's equivalent impedance for the respective bus was used. The susceptance of the tie line for steel company P was 0.00096 + j0.00578 [Ω], and the susceptance of the tie line for the load of all the steel companies was 00020 + j0.00159 [Ω].
The This implies that the reserve model is calculated by the proportion to the maximum demand, and the frequency regulating reserve ( ) has a value of 2.03%, and the stand-by/replacement reserve ( ) has a value of 3.38% (spring and fall season) and 4.05% (winter, summer season). To analyze the frequency characteristics of load fluctuation, the load fluctuation was divided into season, load level and load capacity.
In general, with respect to response speed, GF operation is fastest, followed by AGC operation. Thus, to analyze the frequency characteristics, integrated times of 10 and 30 s were applied to GF operation, and integrated times of 60, 300, 600 and 900 s were applied to AGC operation.
In this paper, to show the frequency deviation distribution, the frequency deviation of a steadystate power system within 0.

An Analysis of the Single Capacity Model
To analyze the single capacity model, we used the load fluctuation data in Table 5, the single model in Figure 10a, and the simulated reference data in Table 7.
The characteristics of dynamic frequency of representative steel manufacturer (P company) is shown in Figure 11. In a few typical cases, Figure 11 shows frequency characteristics for the magnitude of load fluctuation within the reserve and outside the reserve. As shown in Figure 11a, the magnitude of load fluctuation is within the frequency regulating reserve, so the frequency deviation is stable within −0.0542 [Hz]. However, the magnitude of load fluctuation is outside the This implies that the reserve model is calculated by the proportion to the maximum demand, and the frequency regulating reserve (P 1st reserve ) has a value of 2.03%, and the stand-by/replacement reserve (P 2nd reserve ) has a value of 3.38% (spring and fall season) and 4.05% (winter, summer season). To analyze the frequency characteristics of load fluctuation, the load fluctuation was divided into season, load level and load capacity.
In general, with respect to response speed, GF operation is fastest, followed by AGC operation. Thus, to analyze the frequency characteristics, integrated times of 10 and 30 s were applied to GF operation, and integrated times of 60, 300, 600 and 900 s were applied to AGC operation.
In this paper, to show the frequency deviation distribution, the frequency deviation of a steady-state power system within 0.

An Analysis of the Single Capacity Model
To analyze the single capacity model, we used the load fluctuation data in Table 5, the single model in Figure 10a, and the simulated reference data in Table 7.
The characteristics of dynamic frequency of representative steel manufacturer (P company) is shown in Figure 11. In a few typical cases, Figure 11 shows frequency characteristics for the magnitude of load fluctuation within the reserve and outside the reserve. As shown in Figure 11a, the magnitude of load fluctuation is within the frequency regulating reserve, so the frequency deviation is stable within −0.0542 [Hz]. However, the magnitude of load fluctuation is outside the reserve in Figure 11b, so the frequency regulating reserve is insufficient; thus, the frequency characteristics of the generator were not reflected. As a result, the frequency deviation was as significant as −2.975 [Hz]. In Figure 11c, the magnitude of load fluctuation is within the stand-by/replacement reserve, and the frequency deviation is recovered to 60.0 [Hz]. However in Figure 11d, the magnitude of the frequency deviation is outside the stand-by/replacement reserve, so the frequency is not recovered. Outside the limits of the reserve, the frequency characteristics of the power system do not reflect the characteristics of generation power-frequency. In other words, the results depend on the frequency characteristics of the load. The frequency deviation distribution of a steel manufacturing company P is shown in Figure 12.  11d, the magnitude of the frequency deviation is outside the stand-by/replacement reserve, so the frequency is not recovered. Outside the limits of the reserve, the frequency characteristics of the power system do not reflect the characteristics of generation power-frequency. In other words, the results depend on the frequency characteristics of the load. The frequency deviation distribution of a steel manufacturing company P is shown in Figure 12.

An Analysis of the Business Type Capacity Model
We grouped Korea's leading steel manufacturing companies, P, K and H, together and analyzed their frequency characteristics in conjunction with load fluctuation. To this end, we used the load fluctuation data in Table 5, the single model in Figure 10a, and the simulated reference data in Table  8. The dynamic frequency characteristics for business type capacity are shown in Figure 13. The magnitude of load fluctuation exceeds the frequency regulating reserve and stand-by/replacement reserve in Figure 13b,d, showing significant frequency deviation. The distribution of the frequency deviation for the load fluctuation of major business types is shown in Figure 14. Since the business type model has a greater reserve and a higher load level than the single capacity model, the ratio of [State I] is higher and the frequency deviation is, in general, lower.

An Analysis of the Business Type Capacity Model
We grouped Korea's leading steel manufacturing companies, P, K and H, together and analyzed their frequency characteristics in conjunction with load fluctuation. To this end, we used the load fluctuation data in Table 5, the single model in Figure 10a, and the simulated reference data in Table 8. The dynamic frequency characteristics for business type capacity are shown in Figure 13. The magnitude of load fluctuation exceeds the frequency regulating reserve and stand-by/replacement reserve in Figure 13b,d, showing significant frequency deviation. The distribution of the frequency deviation for the load fluctuation of major business types is shown in Figure 14. Since the business type model has a greater reserve and a higher load level than the single capacity model, the ratio of [State I] is higher and the frequency deviation is, in general, lower.

An Analysis of the Network Connected Model
We analyzed the frequency characteristics of the power systems for the load fluctuation of whole domestic steel manufacturing companies. To analyze the network model, we used the load fluctuation data in Table 5, the network model in Figure 10b, and the simulated reference data in Table 9.

An Analysis of the Network Connected Model
We analyzed the frequency characteristics of the power systems for the load fluctuation of whole domestic steel manufacturing companies. To analyze the network model, we used the load fluctuation data in Table 5, the network model in Figure 10b, and the simulated reference data in Table 9.

An Analysis of the Network Connected Model
We analyzed the frequency characteristics of the power systems for the load fluctuation of whole domestic steel manufacturing companies. To analyze the network model, we used the load fluctuation data in Table 5, the network model in Figure 10b, and the simulated reference data in Table 9.
The dynamic frequency characteristics of the network model are shown in Figure 15. Since the magnitude of load fluctuation in whole steel manufacturing companies is within the range of the stand-by/replacement reserve, the frequency deviation was low, showing a stable state. In addition, the frequency deviation distribution in Figure 16 shows a [State I] ratio of 100%. This implies that,   The dynamic frequency characteristics of the network model are shown in Figure 15. Since the magnitude of load fluctuation in whole steel manufacturing companies is within the range of the stand-by/replacement reserve, the frequency deviation was low, showing a stable state. In addition, the frequency deviation distribution in Figure 16 shows a [State I] ratio of 100%. This implies that, as there is sufficient reserve that is capable of supporting the variability in the loads of whole steel manufacturing companies, the frequency deviation is within 0.     The dynamic frequency characteristics of the network model are shown in Figure 15. Since the magnitude of load fluctuation in whole steel manufacturing companies is within the range of the stand-by/replacement reserve, the frequency deviation was low, showing a stable state. In addition, the frequency deviation distribution in Figure 16 shows a [State I] ratio of 100%. This implies that, as there is sufficient reserve that is capable of supporting the variability in the loads of whole steel manufacturing companies, the frequency deviation is within 0.

The Results of Frequency Deviation Analysis
The frequency deviation distributions of the single capacity model, the business type model, and the network model analyzed in this paper are shown in Figure 17. If the frequency deviation characteristics of companies P, K and H are compared with the those of the business type model and the network model, the frequency deviations gradually decrease. This is because the business type model and the network model have greater characteristics of generation power-frequency and load-frequency than the single capacity model. In other words, as the load level and reserve increase, the frequency deviation for load fluctuation decreases.

The Results of Frequency Deviation Analysis
The frequency deviation distributions of the single capacity model, the business type model, and the network model analyzed in this paper are shown in Figure 17. If the frequency deviation characteristics of companies P, K and H are compared with the those of the business type model and the network model, the frequency deviations gradually decrease. This is because the business type model and the network model have greater characteristics of generation power-frequency and loadfrequency than the single capacity model. In other words, as the load level and reserve increase, the frequency deviation for load fluctuation decreases.

Frequency Deviation Analysis of a Power System with Compensator
The power system control block diagram consists of an inertia generator, damping coefficient, and governor characteristics [1]. We analyzed the frequency deviation characteristics with load fluctuation by adding a model of a compensation facility, as shown in Figure 18 [38,39,41]. To determine the frequency deviation, the compensation facility capacity was calculated, and the static characteristics were analyzed. The transfer function between the load fluctuation and frequency deviation of a power system without a compensator is represented by Equation (9), where M is a momentum constant, D is the load damping, and ( ) is the governor characteristic:

Frequency Deviation Analysis of a Power System with Compensator
The power system control block diagram consists of an inertia generator, damping coefficient, and governor characteristics [1]. We analyzed the frequency deviation characteristics with load fluctuation by adding a model of a compensation facility, as shown in Figure 18 [38,39,41]. To determine the frequency deviation, the compensation facility capacity was calculated, and the static characteristics were analyzed. The transfer function between the load fluctuation and frequency deviation of a power system without a compensator is represented by Equation (9), where M is a momentum constant, D is the load damping, and µ(s) is the governor characteristic: Energies 2020, 13, x FOR PEER REVIEW 16 of 28

The Results of Frequency Deviation Analysis
The frequency deviation distributions of the single capacity model, the business type model, and the network model analyzed in this paper are shown in Figure 17. If the frequency deviation characteristics of companies P, K and H are compared with the those of the business type model and the network model, the frequency deviations gradually decrease. This is because the business type model and the network model have greater characteristics of generation power-frequency and loadfrequency than the single capacity model. In other words, as the load level and reserve increase, the frequency deviation for load fluctuation decreases.

Frequency Deviation Analysis of a Power System with Compensator
The power system control block diagram consists of an inertia generator, damping coefficient, and governor characteristics [1]. We analyzed the frequency deviation characteristics with load fluctuation by adding a model of a compensation facility, as shown in Figure 18 [38,39,41]. To determine the frequency deviation, the compensation facility capacity was calculated, and the static characteristics were analyzed. The transfer function between the load fluctuation and frequency deviation of a power system without a compensator is represented by Equation (9), where M is a momentum constant, D is the load damping, and ( ) is the governor characteristic: LFC with a compensation facility with characteristic constant K comp is represented Equation (10), where ∆ f comp is the improved compensator frequency deviation: To be considered a steady-state frequency (∆ f std ) in Korea, the improved frequency deviation should be within ±0.2 [Hz]. Thus, the minimum compensation facility's capacity for the improved frequency to become a steady-state frequency is: where the characteristic constant of the compensation facility is: Equation (12) represents a case where load fluctuation occurs within the spinning reserve range considering speed droop and load damping.
When the load fluctuation exceeds the governor's operating range, the frequency deviation characteristics are proportional to the compensation facility capacity, as shown in Equations (13) and (14): where ∆P GF_lim is the spinning reserve capacity of the governor. By substituting the compensation model into Equation (14), the compensated characteristic equation for the control system is obtained: The compensation capacity equation for steady-state frequency (∆ f std ) operation is: where (∆P L = 1/R · ∆ f GF_lim ) is the frequency regulation reserve divided by each generator due to the limitations of the governor response. Figure 19 shows the compensation capacity and slope for steady-state frequency deviation caused by load fluctuation.
Energies 2020, 13, x FOR PEER REVIEW 17 of 28 To be considered a steady-state frequency ( ) in Korea, the improved frequency deviation should be within ±0.2 [Hz]. Thus, the minimum compensation facility's capacity for the improved frequency to become a steady-state frequency is: where the characteristic constant of the compensation facility is: Equation (12) represents a case where load fluctuation occurs within the spinning reserve range considering speed droop and load damping.
When the load fluctuation exceeds the governor's operating range, the frequency deviation characteristics are proportional to the compensation facility capacity, as shown in Equations (13) and (14): where _ is the spinning reserve capacity of the governor. By substituting the compensation model into Equation (14), the compensated characteristic equation for the control system is obtained: The compensation capacity equation for steady-state frequency ( ) operation is: where ( = 1/ ⋅ _ ) is the frequency regulation reserve divided by each generator due to the limitations of the governor response. Figure 19 shows the compensation capacity and slope for steady-state frequency deviation caused by load fluctuation.

Compensator Characteristics
In this paper, we selected the following compensator models for load fluctuation mitigation: an ESS, a peak reduction (PR) generator, and a customer-owned (CO) generator. The compensator model is based on the IEEE model, the ESS is a battery-based BESS1 model, the PR generator is the GOVHYD model, and the CO generator is the IEEEG1 model. Equations (17)- (19) represent the mathematical models of each compensator. P BESS = ((1/R)/(0.3s + 1))∆ f (17) As the speed droop of each compensator, 1% for BESS, 3% for GOVHYD, and 5% for IEEEG1 are applied, and then summarized as an equation for the frequency deviation, as shown in Equations (20)- (22).
An independent ACE model was constructed as shown in Figure 20 to analyze the PFR characteristics of each model of the compensator. The system inertia of the power system was '10' and a load damping of '3' was applied. The governor and turbine models compared and analyzed the characteristics of the 10 [MW] load fluctuation at 5 s under non load-following conditions.
= (1/ )(5 + 1)( + 1) (0.2 + 1)(38 + 1)(0.1 + 1) An independent ACE model was constructed as shown in Figure 20 to analyze the PFR characteristics of each model of the compensator. The system inertia of the power system was '10' and a load damping of '3' was applied. The governor and turbine models compared and analyzed the characteristics of the 10 [MW] load fluctuation at 5 s under non load-following conditions. Figure 21 shows the frequency characteristics of each compensator, and the response characteristics of ESS are the fastest, so it is very effective in PFR, and its value as an ancillary service is very high. However, due to its shorter duration, lower capacity and higher cost compared to other compensators, appropriate choices are needed.   Figure 21 shows the frequency characteristics of each compensator, and the response characteristics of ESS are the fastest, so it is very effective in PFR, and its value as an ancillary service is very high. However, due to its shorter duration, lower capacity and higher cost compared to other compensators, appropriate choices are needed.

Improved Compensator Control Method
The control block diagram for the PFR improvement method is shown in Figure 22, and the compensation coefficient setting by frequency deviation is applied to each compensator. The characteristics are analyzed as the self-tuning value with respect to frequency deviation proportionality. For the characteristics analysis of the improved control method, the characteristics of the compensators were analyzed by applying the frequency error gain value under the same conditions as the configuration diagram in Figure 22. We applied gains values of '1', '2.4' and '3.2' for BESS, '1', '18', '24', for GOVHYD, and '1', '15', and '48' for IEEEG1 as the error correction factors for each compensator, and the frequency characteristics are shown in Figure 23.
When a frequency deviation occurs due to load fluctuations, since the output of the governor is controlled by the difference from the reference frequency, the improved control method with an error correction factor has a lower frequency deviation of the PFR term and a higher recovery speed than the conventional control method. The larger the frequency error correction factor, the faster the recovery speed of the PFR, but as the self-excited vibration (oscillation) occurs after non-ideal control and frequency recovery exceeding the system maximum response speed of each compensator, a correction factor must be set for the system response characteristics.

Improved Compensator Control Method
The control block diagram for the PFR improvement method is shown in Figure 22, and the compensation coefficient setting by frequency deviation is applied to each compensator. The characteristics are analyzed as the self-tuning value with respect to frequency deviation proportionality.

Improved Compensator Control Method
The control block diagram for the PFR improvement method is shown in Figure 22, and the compensation coefficient setting by frequency deviation is applied to each compensator. The characteristics are analyzed as the self-tuning value with respect to frequency deviation proportionality. For the characteristics analysis of the improved control method, the characteristics of the compensators were analyzed by applying the frequency error gain value under the same conditions as the configuration diagram in Figure 22. We applied gains values of '1', '2.4' and '3.2' for BESS, '1', '18', '24', for GOVHYD, and '1', '15', and '48' for IEEEG1 as the error correction factors for each compensator, and the frequency characteristics are shown in Figure 23.
When a frequency deviation occurs due to load fluctuations, since the output of the governor is controlled by the difference from the reference frequency, the improved control method with an error correction factor has a lower frequency deviation of the PFR term and a higher recovery speed than the conventional control method. The larger the frequency error correction factor, the faster the recovery speed of the PFR, but as the self-excited vibration (oscillation) occurs after non-ideal control and frequency recovery exceeding the system maximum response speed of each compensator, a correction factor must be set for the system response characteristics. For the characteristics analysis of the improved control method, the characteristics of the compensators were analyzed by applying the frequency error gain value under the same conditions as the configuration diagram in Figure 22. We applied gains values of '1', '2.4' and '3.2' for BESS, '1', '18', '24', for GOVHYD, and '1', '15', and '48' for IEEEG1 as the error correction factors for each compensator, and the frequency characteristics are shown in Figure 23.

Improved Compensator Control Method
The control block diagram for the PFR improvement method is shown in Figure 22, and the compensation coefficient setting by frequency deviation is applied to each compensator. The characteristics are analyzed as the self-tuning value with respect to frequency deviation proportionality. For the characteristics analysis of the improved control method, the characteristics of the compensators were analyzed by applying the frequency error gain value under the same conditions as the configuration diagram in Figure 22. We applied gains values of '1', '2.4' and '3.2' for BESS, '1', '18', '24', for GOVHYD, and '1', '15', and '48' for IEEEG1 as the error correction factors for each compensator, and the frequency characteristics are shown in Figure 23.
When a frequency deviation occurs due to load fluctuations, since the output of the governor is controlled by the difference from the reference frequency, the improved control method with an error correction factor has a lower frequency deviation of the PFR term and a higher recovery speed than the conventional control method. The larger the frequency error correction factor, the faster the recovery speed of the PFR, but as the self-excited vibration (oscillation) occurs after non-ideal control and frequency recovery exceeding the system maximum response speed of each compensator, a correction factor must be set for the system response characteristics.

Case Study
We propose four ways in which steel mills can execute the LFC: an ESS, a PR generator, GF operation of a CO generator, and changing industry processes to mitigate load fluctuation. Additionally, we investigated the frequency impact of a load change using a single bus model for four cases. We used an independent power system model to simulate the four proposed measures, as shown in Figure 24. We assumed that the power system generator is in a governor lock condition (load limit), and that the demand end is connected to a compensation facility that is capable of GF and AGC. In this paper, the generator capacity of the system is the same as the contracted power of the load, and the governor operating range corresponding to the frequency adjustment reserve is allocated to the generator. The load is selected as the most frequent value of the average power in units of 5 min for each time period of the intermittent load, and the frequency deviation characteristic is simulated by inputting the load fluctuation magnitude. The simulation data are shown in Table 10. %R is 5% MW/Hz, and %D is 3% MW/Hz. The 1st operating reserve is 14.01 [MW], and the 2nd operating reserve is 27.95 [MW], which are proportionally allocated according to each generator's capacity and correspond to GF and AGC operation, respectively. When a frequency deviation occurs due to load fluctuations, since the output of the governor is controlled by the difference from the reference frequency, the improved control method with an error correction factor has a lower frequency deviation of the PFR term and a higher recovery speed than the conventional control method. The larger the frequency error correction factor, the faster the recovery speed of the PFR, but as the self-excited vibration (oscillation) occurs after non-ideal control and frequency recovery exceeding the system maximum response speed of each compensator, a correction factor must be set for the system response characteristics.

Case Study
We propose four ways in which steel mills can execute the LFC: an ESS, a PR generator, GF operation of a CO generator, and changing industry processes to mitigate load fluctuation. Additionally, we investigated the frequency impact of a load change using a single bus model for four cases. We used an independent power system model to simulate the four proposed measures, as shown in Figure 24. We assumed that the power system generator is in a governor lock condition (load limit), and that the demand end is connected to a compensation facility that is capable of GF and AGC. In this paper, the generator capacity of the system is the same as the contracted power of the load, and the governor operating range corresponding to the frequency adjustment reserve is allocated to the generator. The load is selected as the most frequent value of the average power in units of 5 min for each time period of the intermittent load, and the frequency deviation characteristic is simulated by inputting the load fluctuation magnitude.
(load limit), and that the demand end is connected to a compensation facility that is capable of GF and AGC. In this paper, the generator capacity of the system is the same as the contracted power of the load, and the governor operating range corresponding to the frequency adjustment reserve is allocated to the generator. The load is selected as the most frequent value of the average power in units of 5 min for each time period of the intermittent load, and the frequency deviation characteristic is simulated by inputting the load fluctuation magnitude. The simulation data are shown in Table 10. %R is 5% MW/Hz, and %D is 3% MW/Hz. The 1st operating reserve is 14.01 [MW], and the 2nd operating reserve is 27.95 [MW], which are proportionally allocated according to each generator's capacity and correspond to GF and AGC operation, respectively. The simulation data are shown in Table 10. %R is 5% MW/Hz, and %D is 3% MW/Hz. The 1st operating reserve is 14.01 [MW], and the 2nd operating reserve is 27.95 [MW], which are proportionally allocated according to each generator's capacity and correspond to GF and AGC operation, respectively.

Addition of ESS
An ESS is used for various purposes, such as load shifting, peak shaving, smoothing of photovoltaic and wind turbine output, and frequency regulation. The duration of the ESS reserve operation is shorter than that of the generator, but the load following speed is faster, so we can use the ESS as a reserve resource. We adopted an ESS for frequency regulation to control the frequency deviation caused by mill-load fluctuation. Figure 25 shows the power system simulation model with an ESS for facility compensation.

Addition of ESS
An ESS is used for various purposes, such as load shifting, peak shaving, smoothing of photovoltaic and wind turbine output, and frequency regulation. The duration of the ESS reserve operation is shorter than that of the generator, but the load following speed is faster, so we can use the ESS as a reserve resource. We adopted an ESS for frequency regulation to control the frequency deviation caused by mill-load fluctuation. Figure 25 shows the power system simulation model with an ESS for facility compensation. We used a 30-s 3-sigma average load variation from steel mill customer B, which is a 147 [MW] increase. We assumed that the 1st reserve was 14.01 [MW] and that the output power of the CO generator was fixed at 900 [MW]. The speed droop in this case was 1%. In this simulation, we analyzed the effect of the ESS during 3 min after the load fluctuation before the AGC operation. According to   We used a 30-s 3-sigma average load variation from steel mill customer B, which is a 147 [MW] increase. We assumed that the 1st reserve was 14.01 [MW] and that the output power of the CO generator was fixed at 900 [MW]. The speed droop in this case was 1%. In this simulation, we analyzed the effect of the ESS during 3 min after the load fluctuation before the AGC operation. According to Table 10

GF Operation of a Customer-Owned Generator
In the next case, we assumed that a customer-owned generator was used for the LFC, as shown in Figure 27. We simulated two cases: one for GF operation and another for AGC operation, including the GF mode. The speed droop in this case was 5%.The given data are as follows: a load fluctuation of 219.

GF Operation of a Customer-Owned Generator
In the next case, we assumed that a customer-owned generator was used for the LFC, as shown in Figure 27. We simulated two cases: one for GF operation and another for AGC operation, including the GF mode. The speed droop in this case was 5%.The given data are as follows: a load fluctuation of 219.

GF Operation of a Customer-Owned Generator
In the next case, we assumed that a customer-owned generator was used for the LFC, as shown in Figure 27. We simulated two cases: one for GF operation and another for AGC operation, including the GF mode. The speed droop in this case was 5%.The given data are as follows: a load fluctuation of 219.

GF Operation of a Customer-Owned Generator
In the next case, we assumed that a customer-owned generator was used for the LFC, as shown in Figure 27. We simulated two cases: one for GF operation and another for AGC operation, including the GF mode. The speed droop in this case was 5%.The given data are as follows: a load fluctuation of 219.

Addition of Peak Reduction Generator
A Diesel generator and hydroelectric generator can be used for peak reduction. We considered a hydroelectric generator as a compensation facility to mitigate the load fluctuation. For peak reduction in this case, we used two hydroelectric generators with 100 [MW]. The generator is denoted by G p in Figure 29, which has a 3% droop rate. The assumed load variation was a 155.5 [MW] increase, which was a 5-min 3-sigma fluctuation during light-load operation of the power system. The 2nd reserve was 27.95 [MW], and the customer-owned generator power output was fixed at 900 [MW]. We analyzed the frequency response for the three cases during 10 min of operation: (i) without G p , (ii) GF operation with G p only, and (iii) GF + AGC operation with G p .

Addition of Peak Reduction Generator
A Diesel generator and hydroelectric generator can be used for peak reduction. We considered a hydroelectric generator as a compensation facility to mitigate the load fluctuation. For peak reduction in this case, we used two hydroelectric generators with 100 [MW]. The generator is denoted by Gp in Figure 29, which has a 3% droop rate. The assumed load variation was a 155.5 [MW] increase, which was a 5-min 3-sigma fluctuation during light-load operation of the power system. The 2nd reserve was 27.95 [MW], and the customer-owned generator power output was fixed at 900 [MW]. We analyzed the frequency response for the three cases during 10 min of operation: (i) without Gp, (ii) GF operation with Gp only, and (iii) GF + AGC operation with Gp. Figure 30 illustrates the simulation output for the PR generator case. Without PR generators, the results show a frequency deviation of 2.99 [Hz] due to the lack of power supply. The GF-limited operation using two hydroelectric generators has a frequency deviation of 0.83 [Hz], and the output contribution of the PR generators is 92. 16 [MW]. For case (c), the output increases to 126.97 [MW], and the frequency stabilizes without deviation. The results show that using a PR generator is an effective way for LFC to stabilize the frequency within 10 min.

Change of Industry Processes
The load fluctuation caused by a steel mill can be improved by adjusting the manufacturing process without decreasing productivity through load re-distribution and re-allocation. Load redistribution is the equalization of load changes by adjusting all other mill processes except the essential processes. In other words, it is the same as the method of leveling the load through peakcut. In the case of a process with large load fluctuations, it is a method of improving the process by reducing the use of other loads and making the magnitude of the load fluctuation constant. In addition, load re-allocation is the alteration of the process sequence according to the power system demand or factory load level. In this case, the process is changed according to the daily load level. However, although theoretical proposals are possible, there are difficulties in practical application. We simulated three cases of process changes, as shown in Table 11. Case 1 is for the original process, case 2 is for load re-distribution, and case 3 is for load re-allocation.  Figure 30 illustrates the simulation output for the PR generator case. Without PR generators, the results show a frequency deviation of 2.99 [Hz] due to the lack of power supply. The GF-limited operation using two hydroelectric generators has a frequency deviation of 0.83 [Hz], and the output contribution of the PR generators is 92. 16 [MW]. For case (c), the output increases to 126.97 [MW], and the frequency stabilizes without deviation. The results show that using a PR generator is an effective way for LFC to stabilize the frequency within 10 min.

Addition of Peak Reduction Generator
A Diesel generator and hydroelectric generator can be used for peak reduction. We considered a hydroelectric generator as a compensation facility to mitigate the load fluctuation. For peak reduction in this case, we used two hydroelectric generators with 100 [MW]. The generator is denoted by Gp in Figure 29, which has a 3% droop rate. The assumed load variation was a 155.5 [MW] increase, which was a 5-min 3-sigma fluctuation during light-load operation of the power system. The 2nd reserve was 27.95 [MW], and the customer-owned generator power output was fixed at 900 [MW]. We analyzed the frequency response for the three cases during 10 min of operation: (i) without Gp, (ii) GF operation with Gp only, and (iii) GF + AGC operation with Gp. Figure 30 illustrates the simulation output for the PR generator case. Without PR generators, the results show a frequency deviation of 2.99 [Hz] due to the lack of power supply. The GF-limited operation using two hydroelectric generators has a frequency deviation of 0.83 [Hz], and the output contribution of the PR generators is 92. 16 [MW]. For case (c), the output increases to 126.97 [MW], and the frequency stabilizes without deviation. The results show that using a PR generator is an effective way for LFC to stabilize the frequency within 10 min.

Change of Industry Processes
The load fluctuation caused by a steel mill can be improved by adjusting the manufacturing process without decreasing productivity through load re-distribution and re-allocation. Load redistribution is the equalization of load changes by adjusting all other mill processes except the essential processes. In other words, it is the same as the method of leveling the load through peakcut. In the case of a process with large load fluctuations, it is a method of improving the process by reducing the use of other loads and making the magnitude of the load fluctuation constant. In addition, load re-allocation is the alteration of the process sequence according to the power system demand or factory load level. In this case, the process is changed according to the daily load level. However, although theoretical proposals are possible, there are difficulties in practical application. We simulated three cases of process changes, as shown in Table 11. Case 1 is for the original process, case 2 is for load re-distribution, and case 3 is for load re-allocation.

Change of Industry Processes
The load fluctuation caused by a steel mill can be improved by adjusting the manufacturing process without decreasing productivity through load re-distribution and re-allocation. Load re-distribution is the equalization of load changes by adjusting all other mill processes except the essential processes. In other words, it is the same as the method of leveling the load through peak-cut. In the case of a process with large load fluctuations, it is a method of improving the process by reducing the use of other loads and making the magnitude of the load fluctuation constant. In addition, load re-allocation is the alteration of the process sequence according to the power system demand or factory load level. In this case, the process is changed according to the daily load level. However, although theoretical proposals are possible, there are difficulties in practical application. We simulated three cases of process changes, as shown in Table 11. Case 1 is for the original process, case 2 is for load re-distribution, and case 3 is for load re-allocation. For each case, three kinds of load variances were added for 10 s while linearly increasing the load by 100 [MW] for 30 s. Figure 31 shows the simulation input profiles for these 3 cases. In Figure 31a  For each case, three kinds of load variances were added for 10 s while linearly increasing the load by 100 [MW] for 30 s. Figure 31 shows the simulation input profiles for these 3 cases. In Figure  31a, there is a step change of 21 [MW] at time zero, 75 [MW] at 10 s, and 131. 5 [MW] at 20 s, which lasted for 10 s each. A similar method was used for cases 2 and 3.   Figure 32 shows the transient simulation results for the process change. Figure 32a shows that the frequency deviations at steady state are 2.76, 2.24, and 1.73 [Hz] for cases 1, 2, and 3, respectively. Rearranging the process to decrease the load fluctuation magnitude over time resulted in a better frequency response with respect to the operational reserve.  Figure 33 summarizes the mitigation effects of the simulation results. We can improve the frequency deviation response by 30% just by changing the process sequence without any extra investment. The frequency deviation is recovered within one minute when using the ESS, 5 min when using the pump storage hydro-generator, and 10 min when using the customer's generator. Figure 32 shows the transient simulation results for the process change. Figure 32a shows that the frequency deviations at steady state are 2.76, 2.24, and 1.73 [Hz] for cases 1, 2, and 3, respectively. Rearranging the process to decrease the load fluctuation magnitude over time resulted in a better frequency response with respect to the operational reserve.  Figure 33 summarizes the mitigation effects of the simulation results. We can improve the frequency deviation response by 30% just by changing the process sequence without any extra investment. The frequency deviation is recovered within one minute when using the ESS, 5 min when using the pump storage hydro-generator, and 10 min when using the customer's generator.

Conclusions
In this study, we analyzed the effect of the variability of large-scale industrial load on the frequency of power systems and the load fluctuation frequency spectrum of the steel manufacturing industry, showing substantial load fluctuation, time zones where load fluctuation occurs, and the magnitude of load fluctuation. In addition, we analyzed the frequency characteristics using the EMS (Energy Management System) real-time data when powering systems including steel mills, resulting in customers with three kinds of capacity: individual capacity, group capacity, and total capacity. In addition, we developed an AGC program to simulate the effect of the magnitude of load fluctuation on the frequency of the power system. To this end, load fluctuation data was categorized by season and load level and then the frequency deviation characteristics of the single capacity model, the business type model, and the network model were analyzed. We also proposed ways to mitigate the system's frequency deviation at the customer end using an ESS, pump storage hydro-generator, customer generators, and the adjustment of plant processes. To recover from frequency deviations, the capacity of the compensation facilities was calculated, and the static characteristics were analyzed. We proved the feasibility of the measures through the simulation results.  Figure 33 summarizes the mitigation effects of the simulation results. We can improve the frequency deviation response by 30% just by changing the process sequence without any extra investment. The frequency deviation is recovered within one minute when using the ESS, 5 min when using the pump storage hydro-generator, and 10 min when using the customer's generator.

Conclusions
In this study, we analyzed the effect of the variability of large-scale industrial load on the frequency of power systems and the load fluctuation frequency spectrum of the steel manufacturing industry, showing substantial load fluctuation, time zones where load fluctuation occurs, and the magnitude of load fluctuation. In addition, we analyzed the frequency characteristics using the EMS (Energy Management System) real-time data when powering systems including steel mills, resulting in customers with three kinds of capacity: individual capacity, group capacity, and total capacity. In addition, we developed an AGC program to simulate the effect of the magnitude of load fluctuation on the frequency of the power system. To this end, load fluctuation data was categorized by season and load level and then the frequency deviation characteristics of the single capacity model, the business type model, and the network model were analyzed. We also proposed ways to mitigate the system's frequency deviation at the customer end using an ESS, pump storage hydro-generator, customer generators, and the adjustment of plant processes. To recover from frequency deviations, the capacity of the compensation facilities was calculated, and the static characteristics were analyzed. We proved the feasibility of the measures through the simulation results.

Conclusions
In this study, we analyzed the effect of the variability of large-scale industrial load on the frequency of power systems and the load fluctuation frequency spectrum of the steel manufacturing industry, showing substantial load fluctuation, time zones where load fluctuation occurs, and the magnitude of load fluctuation. In addition, we analyzed the frequency characteristics using the EMS (Energy Management System) real-time data when powering systems including steel mills, resulting in customers with three kinds of capacity: individual capacity, group capacity, and total capacity. In addition, we developed an AGC program to simulate the effect of the magnitude of load fluctuation on the frequency of the power system. To this end, load fluctuation data was categorized by season and load level and then the frequency deviation characteristics of the single capacity model, the business type model, and the network model were analyzed. We also proposed ways to mitigate the system's frequency deviation at the customer end using an ESS, pump storage hydro-generator, customer generators, and the adjustment of plant processes. To recover from frequency deviations, the capacity of the compensation facilities was calculated, and the static characteristics were analyzed. We proved the feasibility of the measures through the simulation results.
The analysis methods used in this paper and the results of this paper are expected to be used as ways of fairly sharing the cost of regulating frequency variations with those who provide the causes of such variation, thus establishing a foundation of justice for the electric power industry.
Load fluctuations will grow with the increased demand for power in the future. Moreover, the effect of intermittent power generation on power systems due to the increase in renewable