1、4759 Control Optimization for Chilled Water Thermal Storage System Under Complicated Time-of-Use Electricity Rate Schedule Jijun Zhou Member ASHRAE Song Deng, PE Guanghua Wei, PE Associate Member ASHRAE David E. Claridge, PhD, PE Member ASHRAE Member ASHRAE ABSTRACT The existence of a 1.4-million-ga
2、llon chilled water ther- mal storage tankgreatly increases the operationalflexibility of a campus-wide chilled water system under a four-price time- of- use electricity rate structure. While signiJicant operational savings can be expected, the complication in the rate structure also requires more so
3、phisticated control over the charging and discharging processes of the thermal storage tank. A chiller start-stop optimization program was developed and implemented into the energy management und control system (EMCS) to determine the number of chillers that need to be brought on line and the start
4、und stop times for euch chiller every day, based on the prediction of the campus cool- ing load within the next 24 hours. With timely and accurate weather forecasting, the actual tankcharging und discharging process closely matches the simulated process. The chiller plants operating schedules are dy
5、namically optimized to deliver required cooling capacity at lowest possible operating costs. INTRODUCTION Thermal storage is economically attractive under a time- of-use electricity rate schedule. A typical use of the chilled water thermal storage tank is to charge the tank during off- peak hours wh
6、en utility costs are relatively low and discharge the tank during on-peak hours to reduce operating costs. Equally significant savings may be achieved if a facility is subject to demand charges, in which case the thermal storage tank is used to “level off the peak demand in order to avoid hefty dema
7、nd charges. Another less obvious benefit ofthermal storage is the decoupling of the thermal load profile from the operation of the equipment, adding an element of flexibility and reliability to the system (ASHRAE 2003). W. Dan Turner, PhD, PE Oscar Contreras While it is conceptually simple to take a
8、dvantage of the thermal storage tank, the actual control strategy for a chilled water system with a thermal storage tank may turn out to be rather complicated, especially if multiple chillers are involved and the facility is under a complicated utility rate schedule. Even when the capacity of the th
9、ermal storage tank is big enough to serve the cooling load for the entire on-peak period, needs for operational optimization still exist and the tank should be used only to the extent “necessary” since unneces- sary production of the storage almost always introduces extra heat losses (Tamblyn 1985).
10、 Many papers have been published in the past two decades on the subject of thermal storage, including several that emphasize the operation and control issues (Shavit 1985; Fiorino 1991). Few sources, however, discuss in detail the optimization process for facilities under a complicated time-of-use u
11、tility rate structure. Unsuccessful thermal storage projects are seldom documented, but they do exist. Liu (1999) described how rehabilitation was able to improve the performance of an unsuccessful storage system and reduce the operating costs. Many other case studies can be found in the literature,
12、 some of which detail step-by-step control optimization procedures for a specific electricity rate schedule (Wei 2002). A complete control strategy for a ther- mal storage system describes the sequences of operation under all possible operating modes, such as charging storage, charg- ing storage whi
13、le meeting load, meeting load from discharg- ing, meeting load from discharging and direct equipment operation, demand-limiting control, etc. (ASHRAE 2003). The most economical operating schedule has to be determined with many factors in mind, such as load prediction, utility rates, demand charges,
14、and capacities of the chillers and the thermal storage tank. J. Zhou is a research associate, G. Wei and S. Deng are assistant directors of the Energy Systems Laboratory, and W.D. Turner and D.E. Car- idge are professors in the Department of Mechanical Engineering, all at Texas A however, it is not
15、always desirable to 1 1 1 1 1 1 2 2 1 1 1 1 1 1 2 2 have a full tank if the campus load is relatively low. The number of chillers that should be brought online during the charging cycle and the exact times to stadstop a specific chiller should be determined accurately to take full advantage of the f
16、our-tiered rate structure. PROCEDURES TO OPTIMIZE CHILLER STARTS AND STOPS The exact chiller stadstop schedules will vary from day to day. It is obvious that, for the months from May to Septem- ber, the thermal storage tank should be charged during night- time when the electricity price is lower and
17、 be discharged in the afternoon when the electricity price is higher. Assume the tank is almost depleted at 10 p.m., and one or more chillers would be brought online in sequence. If the chiller(s) produce more cooling than what is demanded by the campus at the time, the tank would be gradually charg
18、ed. The tank would 186 ASHRAE Transactions: Research then be fully charged sometime in the early afternoon before the higher-priced period starts. Then all the chillers would be shut down and the tank start to discharge. At around 10 p.m., the thermal storage tank would be almost depleted again, and
19、 a new tank charging cycle would start. This tank charging/discharging sequence is conceptually straightforward. However, the exact times for starting and stopping the chillers that will take full advantage of the ther- mal storage tank and the four-tiered rate schedule can only be determined from m
20、ore detailed analysis. The optimal chiller startstop sequence varies from day to day depending on the actual cooling load of the campus. A chiller startistop optimi- zation (CSSO) program was developed and implemented in the energy management and control system (EMCS) to auto- mate the decision-maki
21、ng process to determine the most economical operating schedule ofthe central chiller plant each day, based on many factors and restraints. Unless otherwise stated, the CSSO procedures discussed in this paper address the warmer months when at least three different electricity prices are involved in a
22、 single day. The procedures for the chiller starthtop optimization include the following steps. First, the cooling load of the campus is estimated for the next 24 hours with a load predic- tion model based on the weather forecast. Next, the best period of time for the tank to discharge is selected,
23、including as many higher-priced hours as possible. The time period to charge the tank is automatically determined as well. Then the average chiller production rate required during the charging period is calculated from the total estimated campus load and the avail- able charging hours. After that, t
24、he number of chillers to be brought online and the runtimes for each chiller are deter- mined based on the predicted campus load. Finally, the start/ stop times for each chiller are scheduled precisely, after taking many restrictions and specific requirements into consider- ation. 2,500 - 2,000 . 9.
25、 .- -1,000 -1,500 . 40 50 60 70 80 90 100 Ambient Temperature (deg F) Figure 2 Thermal storage tank discharge rate (weekdays). Campus Load Prediction Model The total cooling load of all the buildings on the campus chilled water loop is regarded as a simple function of the outside air temperature. At
26、 any time, this total cooling load can be estimated from the instantaneous chilled water consump- tion rate at the central chiller plant, which can be calculated from the flow rate and the supply and the return temperatures of the chilled water that are continuously monitored and trended by the EMCS
27、. Unfortunately, the chilled water flow meter was not working properly at the time and therefore there was no direct means to measure the chilled water consumption of the campus. The chilled water supplied to the campus comes from two sources: the chillers andor the thermal storage tank. When the ch
28、illers produce more than the campus needs, the excess chilled water flows through the thermal storage tank and charges it. When the chillers produce less than needed, the tank discharges to help meet the demand. When all chillers are shut down, the tank provides all the needed chilled water to the c
29、ampus. Under the last scenario, the campus cooling load can be evaluated from the tanks discharging rate. Fortunately, the thermal storage tank itself is metered adequately, with a total of 30 temperature sensors evenly distributed along its vertical dimension. From the overall temperature changes w
30、ithin the tank during certain periods of time, the rate at which the tank was discharged (or charged) can be calculated. Figure 2 and Figure 3 show the thermal storage tanks discharging (or charging) rate, calculated from the trended data, for weekdays and weekends, respectively. A positive value in
31、dicates the tank was discharging and a negative value indicates the tank was being charged. On both figures, clear separations among data are shown. The top cloud of data on each figure shows the operation scenario when all the chillers were shut down and the campus chilled water was solely supplied
32、 by the tank itself. Therefore, the discharging rates are all positive. These data points actually outline the campus 2,500 2,000 1 e 1,500 1 % 1,000 E 500 K IF 2 00 .2 a : -500 x I- -1,000 -1.500 . I .= . 40 50 60 70 80 90 1 O0 Ambient Temperature (deg F) Figure 3 Thermal storage tank discharge rat
33、e (weekends). ASHRAE Transactions: Research 187 2,000 , I - “. . . . . - I -ibientTernparature Model 40 50 60 70 80 90 1 O0 mbient Temperature (deg F) Figure 4 Campus cooling load prediction model. cooling load profiles for weekdays and weekends. The middle cloud of data shows the operation scenario
34、 when one chiller was running and the tank may discharge or be charged depend- ing on the campus load. Similarly, the bottom cloud of data was for the operation scenario when two chillers were running and the tank would always be charged since the campus load never exceeded the capacity of two fully
35、 loaded chillers. For simplicity, the campus cooling load profiles for weekdays and weekends are modeled with simple three-point-change-point (3PCP) linear models, as shown in Figure 4. The ambient temperature profile for the next 24 hours is generated from the weather channels daily weather forecas
36、t. The temperature profile is generated with a sine function, which peaks at 3 p.m. and bottoms at 3 a.m. The predicted ambient temperature at any hour is calculated as where Thi = daily high temperature from weather forecast, T, = overnight low temperatures from weather forecast, n = the hour of th
37、e day (from O to 23). Figure 5 shows a comparison between the generated ambient temperature profile and the actual hour-by-hour weather data from the weather channel. The close match between the two indicates the ambient temperature prediction model is sufficiently accurate. P E 75.0 E 70.0 6/11/02
38、6/11/02 6/11/02 6/12/02 6/12/02 6/12/02 6/12/02 6/13/02 6/13/02 6/13/02 6/13/02 6/14/02 600 1200 18:OO 001) 6.01) 1200 10.01) 0:OO 6IW 1200 18:OO Q00 Time Figure 5 Modeled ambient temperature profile vs. actual hourly weather data. Determination of Daily Tank ChargelDischarge Periods With the campus
39、 coling load predicted for the next 24 hours, the total required chilled water production could be determined. The next question to address is when and how long to run the chillers or, in other words, to determine the tank charging/discharging periods. One also has to remember that the tank may be d
40、ischarging even when one chiller is running. However, the tank discharging period is defined as the period when all the chillers are ofnine and the campus cooling is solely supplied by the tank itself. The rest of the time when at least one chiller is running is defined as the tank-charging period.
41、Since it is advantageous to use (discharge) the thermal storage tank during higher-price periods, the tank discharge period is selected hour by hour starting from the most expen- sive hours to the least expensive hours. For the months from June through August, when all four electricity prices come i
42、nto play, the hours to use the tank (and therefore avoid running the chillers) should be selected in the sequence shown below. Accumulate the predicted campus cooling loads hour by hour, in the sequence below, until the maximum tank capacity is reached. In rare cases, the accumulated cooling load fo
43、r the most expensive 12 hours (i.e., from 10 a.m. to 10 p.m.) may still be less than the tank capacity. What to do next depends on the month. For May and September, since all the rest of the hours are charged for the cheapest rate (price-1), the tank only needs to store enough chilled water for thos
44、e 12 hours. The I Price 4 I Price 3 I Price 2 I Price 1 I 188 ASHRAE Transactions: Research tank should start discharging at 10 a.m. and start the charging process at 10 p.m. For June, July, and August, however, the tank should be exploited further. There are another four hours (8 a.m. to 10 a.m. an
45、d 10 p.m. to 12 a.m.) in the price-2 period that can take advantage ofthe thermal storage tank. Keep accu- mulating the hourly cooling loads for the price-2 hours, until the tank capacity is reached. The discharging process starts at the earliest selected hour and ends at the latest selected hour. T
46、he weekends are different from the weekdays due to the difference in rate schedules. For May and September, only the eight hours fi-om 2 p.m. to 10 p.m. are charged at price-2 and all other hours are charged at price- 1. The tank should only be charged to the amount required by these eight hours. Th
47、e weekends of June, July, and August are charged at three prices daily. Therefore, methods similar to those described for the weekdays should be used to determine the time period for the tank discharging process. The tank charging process starts as soon as the tank discharging process is over, when
48、the tank charge is expected to be depleted (or almost depleted). Because the physical plant requires the chillers to be started under operators observation, the actual tank charging process should be started before the operator on the last shift leaves duty, which is around 10:30 p.m. For that reaso
49、n, the latest time to stop the tank discharging cycle (and to start the tank charging cycle) was set at 10 p.m. The tank charging cycle ends at the start of the tank discharg- ing cycle. Total Predicted Chiller Load and Chiller Run Time The total predicted chiller load (Lp) is defined as the load seen by all the chillers for the next charging cycle, which is essentially the total estimated campus load (L,) for the next 24 hours, unless the tank is not depleted at the start ofthe charging cycle. In that case, the total predicted chiller load is the esti- mated total campus load le
