1、4683 Impact of Modeling Accuracy on Predictive Optimal Control of Active and Passive Building Thermal Storage Inventory Simeng Liu Student Member ASHRAE ABSTRACT This paper evaluates the impact of modeling accuracy on the model-based closed-loop predictive optimal control of both passive building th
2、ermal capacitance and active thermal energy storage (TES) systems to minimize utility cost. The following guidelines have been derived: for an internal heat gain-dominated commercial building, the deviation of build- ing geometry and zoning from the reference building only marginally afects the opti
3、mal control strategy; reasonable simplijcations are acceptable without loss of cost savings potential. Building construction characteristics determine the building passive thermal storage capacity. Zone temperature setpoints are affected more than TES operation by. this construction mismatch, and a
4、loss of cost savings potential is found in some cases. It is advisable to make sure the construc- tion material is well modeled. Zone temperature setpoint projles and TES performance are strongly afected by mismatches in internal heat gains, especially when they are underestimated. Since they are a
5、key factor determining the building cooling load, efforts should be invested to keep the internalgain mismatch as small aspossible. Eficiencies of the building energy systems have no direct impact on building cooling load, but they affect both zone temperature setpoints and active TES operation beca
6、use of the coupling of the base chiller and the TESchillel: Relative eficiencies of the base and TESchillers will determine the balance of operation of the two chillers. Mismatch in this category may besignijcant. To avoid critical modeling mismatch, system ident$cation techniques may be useful in i
7、mproving the modelingprocess. Gregor P. Henze, Ph.D., P.E. Member ASHRAE INTRODUCTION Unlike energy conversion equipment, thermal energy distribution, storage, and control systems have not optimized their energy efficiency and peak load reduction potential. Advancements are needed to improve thermal
8、 storage and control systems and systems integration from a whole building perspective, while meeting occupant comfort and perfor- mance requirements (NETLDOE 2000). This paper evaluates the impact of modeling accuracy on the predictive optimal control of both passive building thermal capacitance an
9、d active thermal energy storage (TES) systems to minimize an objective function of choice, including total energy consumption, energy cost, occupant discomfort, or a combination thereof. In this paper, “active” denotes that ther- mal storage systems, such as ice storage, require an additional fluid
10、loop to charge and discharge the storage tank or to deliver cooling to the existing chilled water loop. Building thermal capacitance is “passive” because it requires no additional heat exchange fluid in addition to the conditioned airstream. To investigate the predictive optimal control of active an
11、d passive building thermal storage inventory, a simulation envi- ronment has been developed based on a state-of-the-art dynamic building simulation program currently under devel- opment by the U.S. Department of Energy (DOE) (2003). This simulation environment is designed to carry out a closed-loop
12、optimization of the hourly zone temperature setpoints and thermal energy storage system charge/discharge rates to reduce building peak cooling demand and associated operat- ing costs. Past research shows that the optimum strategy is affected by many factors, such as utility rate structures and plant
13、 energy efficiency. The accuracy of the building model used for the model-based optimal control relative to the actual building behavior is, thus, of great importance to the quality of the optimal strategy. Simeng Liu is a graduate student and Gregor P. Henze is an assistant professor of architectur
14、al engineering at the University of Nebraska- Lincoln. 02004 ASHRAE. 151 The modeling process entails the truthful representation of the actual characteristics of a specific building. Modeling accuracy may be increased by either improving the simulation program itself or by accurately collecting dat
15、a and information on the building to be modeled. However, it is impossible and impractical to collect complete, accurate information for modeling purposes. Some degree of mismatch is unavoidable with respect to building geometry, consmiction material prop- erties, internal heat gain, and performance
16、 characteristics of the building energy systems. Therefore, it is important to quantify the impact of various modeling mismatches on predictive optimal control. Investigations have been carried out in five different categories of modeling mismatch that are likely to occur in the modeling process. Th
17、is paper summa- rizes and analyzes the results, and provides a comprehensive assessment and guidelines for modeling. LITERATURE REVIEW Prior studies on building thermal mass utilization to reduce peak cooling load and associated electrical demand show that cost savings vary widely among the publishe
18、d case studies, and that the optimal operating strategy is sensitive to many factors (Rab1 and Norford 199 1 ; Conniff 199 1 ; Morris et al. 1994; Keeney and Braun 1996, 1997). In a simulation study presented by Braun (1 990), cost savings for a design day varied from 0% to 35%, depending on system
19、type and utility rate. Anderson and Brandemuehl (1 992) demonstrated energy and cost savings potential by precooling the building structure, calling attention to the importance of the mass of furnishing, which significantly affects the precooling strategy. Braun et al. (2001) developed a tool to eva
20、luate different precooling strat- egies by comparing the HVAC utility costs in each application. Simulation studies were carried out for selected locations, climates, and utility rate structures. A comparison showed cost savings varying from 40% at best to zero, or even excess costs for some less fa
21、vorable cases. In a review article on load control using building thermal mass, Braun (2003) concluded that the savings potential is very sensitive to the utility rates, building and plant characteristics, and weather conditions and occupancy schedule. The greatest cost savings were realized for the
22、 case of heavy construction, good part-load character- istics, and low ambient temperature, which enabled free cool- ing during night ventilation. Henze et al. (1 997) developed a simulation environment to evaluate various control strategies for active ice-based building thermal storage systems. Com
23、pared to the three Location Weather conventional control methods (chiller-priority, storage-prior- ity, and constant-proportion), the optimal control strategy, determined by dynamic programming, demonstrated signifi- cant cost saving potential in most simulation cases. Research also shows that cost
24、savings can be affected by many factors. In a parametric analysis by Henze (2003), the optimization of TES with different ice storage systems, chiller types, rate structures, and building types was investigated. Results showed that the most promising cost savings were associated with favorable utili
25、ty rate structures and good plant energy efficiency. In the current study, we developed an optimal controller that combines the merits of both passive building thermal mass load control and optimal operation of active thermal energy storage systems. This task was formulated as a nonlinear multivaria
26、ble closed-loop predictive optimal control problem. ASSUMPTION AND RESTRICTIONS There are two categories of factors affecting the model- based optimal control of the active and passive building ther- mal storage inventory. The building independent category includes climate, location, and utility rat
27、e structure. These are factors that cannot be changed or manipulated. They cannot be mismatched in the optimization simulation. The building dependent category includes all aspects related to building modeling. In the simulation environment, the building model is generated by the user in the form of
28、 an input file where mismatches occur. It is important to know what happens if a mismatched model is employed in the context of real-time control and its “optimal” result is applied to the real building. In this paper, model-related factors are summarized into the following five categories of parame
29、ters: 1. Geometry 2. Zoning 3. 4. 5. Construction materials including external and internal construction Internal heat gains, including light, equipment, and occu- pancy Characteristics of the plant, including the base chiller and TES system In order to focus on these factors only, all building inde
30、- pendent factors (e.g., climate, location, and utility rate) are kept the same in all simulation cases. Table 1 lists the building independent factors. Phoenix, AZ Phoenix, AZ, TMY2 file used in the simulation Utility Rate Structure On-peak period: 9 a.m. - 6 p.m. On-peak energy rate: 0.20 $/kWh Of
31、f-peak energy rate: 0.05 $/kWh On-peak demand rate: 10.00 $/kW Off-peak demand rate: 0.00 $/kW 152 ASH RAE Transactions: Research TO examine the effects of model variation on optimiza- tion, the weather data file has been modified. Ten identical days are generated by repeating the weather patterns o
32、f July 2 1 ten times from July 21 to 30, thus eliminating the effects of weather variation on the optimization. Another consideration is the thermal history of the building. Zhou et al. (2003) pointed out that different assumptions of the thermal history could generate widely different results durin
33、g the early part of the simulation. To eliminate the start-up effects on the optimi- zation mentioned above, all simulations in this analysis are conducted from July 21 to 30, but results from the beginning will not be considered and presented in the following discus- sion. Case Daily Energy Consump
34、tion (kl) (average) Daily Energy Cost ($) (average) DEFINITION OF TERMS Nighttime setback control Optimization of perfect match PM 5,048,878 5,154,819 23 1.62 183.80 Before implementing the predictive optimal controller in a field or lab application, simulation studies were carried out to analyze th
35、e impact of the five categories of modeling vari- ations listed above. Two simulation environments were set up: the first one was used to carry out the optimization for all simu- lation cases using the mismatched models. After the optimal solution was generated by the first simulation environment, t
36、he optimal strategy in the form of hourly zone temperature setpoints (passive storage) and TES charge/discharge rates (active storage) was applied in the second simulation environ- ment. Here, a reference building model carried out the simu- lation (without optimization) and the associated results,
37、including cost and energy consumption, were calculated. This second environment was meant to represent the application of the controller in a real building. CSR I Plannina Model I Execution Model 100% I 1W% OW Model Accuracy Figure 1 Overview of the execution model and planning models. Execution Mod
38、el The execution model (EM) represents the real building and is intended to execute the optimal strategy previously found by the optimizer to calculate the energy cost and other related simulation results. In this analysis, the EM represents a one-story office building with five thermal zones. The b
39、uilding is occupied from 8:OO a.m. to 5:OO p.m. with 0.1 persodm? and 45 W/m2 of internal heat gain. Planning Model The planning model (PM) refers to the mismatched build- ing model used in the optimization simulation environment. Figure 1 gives an overview of the planning models for the five catego
40、ries of mismatch. For a specific building, model accuracy decreases with the number of parameter categories that vary from the execu- tion model. In the perfect match case, the planning model is identical to the execution model. Although optimization results calculated using this model are considere
41、d as a theo- retical benchmark, the existence of local minima means the results are not necessarily the best of all the planning models. All other optimization results of mismatch planning models will be compared with the perfect match case, and the devia- tions will be identified. Cost Savings Rati
42、o Since the analysis in this paper is based on the comparison of daily optimal results of different planning models, daily energy cost is selected as the objective function in the optimi- zations; demand cost is not included in this analysis. The saving potential of each planning model is discussed
43、in terms of a cost savings ratio (CSR). The optimal control strategy generated by each planning model will be applied to the execu- tion model; cost savings are calculated by comparing the cost with those of conventional nighttime setback control. Cost savings achieved by optimization of the perfect
44、 match model are considered as the benchmark value. The cost savings ratio (CSR) of each planning model is defined as the ratio of cost savings achieved by the planning model compared to the benchmark value. Before investigating the mismatched planning models, simulations of conventional nighttime s
45、etback control and optimization of the perfect match planning model were carried out. Table 2 gives the summary of energy consumption and cost savings of these two cases. Table 2. Summary of Cost and Energy Consumption of Night Setback Control and Optimization of Perfect Match Case I Cost Saving I I
46、 20.6% I Figure 2 Geometries of the original building and the building model with simplified geometv. ANALYSIS AND DISCUSSION Geometry One of the biggest improvements in building simulation software is the capability of visualizing the building model. Users can now generate a building model that clo
47、sely mirrors the actual building. However, it is still impossible to accurately model every aspect of the buildings behavior. In addition, there is always a trade-off between the increased realism and cost of modeling efforts. As far as the optimal controller is concerned, we need to identify how mu
48、ch “realism” is required when setting up a building model and applying it to our optimization environment and what kind of consequences result from a particular mismatch. In the analysis of the geometry category, the execution model (EM) is set up as a cross-shaped building with five zones. Mismatch
49、ed planning models (PMs) vary in shape and area of fenestration. Figure 2 offers a schematic of the geom- etry of both EM and PM. From Figure 2, we can see that in the mismatched PM, the building shape is modeled as a simple box, which is the simplest representation of a building. Two simplification approaches were investigated. One approach was to keep the total external surface area the same as the original (referred to as PM-A in Figures 3 and 4), and the other approach was to keep the total volume constant (referred to as PM-V in Figures 3 and 4). Figures 3 and 4 compare the optimizat
