Вы находитесь на странице: 1из 11

Applied Energy 188 (2017) 497–507

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Cloud computing platform for real-time measurement and verification of


energy performance
Ming-Tsun Ke, Chia-Hung Yeh ⇑, Cheng-Jie Su
Department of Energy and Refrigerating Air-Conditioning Engineering, National Taipei University of Technology, Taipei, Taiwan

h i g h l i g h t s

 Application of PSO algorithm can improve the accuracy of the baseline model.
 M&V cloud platform automatically calculates energy performance.
 M&V cloud platform can be applied in all energy conservation measures.
 Real-time operational performance can be monitored through the proposed platform.
 M&V cloud platform facilitates the development of EE programs and ESCO industries.

a r t i c l e i n f o a b s t r a c t

Article history: Nations worldwide are vigorously promoting policies to improve energy efficiency. The use of measure-
Received 16 October 2016 ment and verification (M&V) procedures to quantify energy performance is an essential topic in this field.
Received in revised form 2 December 2016 Currently, energy performance M&V is accomplished via a combination of short-term on-site measure-
Accepted 8 December 2016
ments and engineering calculations. This requires extensive amounts of time and labor and can result
Available online 20 December 2016
in a discrepancy between actual energy savings and calculated results. In addition, the M&V period typ-
ically lasts for periods as long as several months or up to a year, the failure to immediately detect abnor-
Keywords:
mal energy performance not only decreases energy performance, results in the inability to make timely
Real-time energy performance
Measurement and verification
correction, and misses the best opportunity to adjust or repair equipment and systems.
Cloud computing In this study, a cloud computing platform for the real-time M&V of energy performance is developed.
Particle swarm optimization On this platform, particle swarm optimization and multivariate regression analysis are used to construct
accurate baseline models. Instantaneous and automatic calculations of the energy performance and
access to long-term, cumulative information about the energy performance are provided via a feature
that allows direct uploads of the energy consumption data. Finally, the feasibility of this real-time
M&V cloud platform is tested for a case study involving improvements to a cold storage system in a
hypermarket.
Cloud computing platform for real-time energy performance M&V is applicable to any industry and
energy conservation measure. With the M&V cloud platform, real-time and long-term energy perfor-
mances can be obtained. By tracking fluctuations in energy performance, real-time monitoring or correc-
tion of the operating performance of equipment or system can help to maintain good energy
performance. Thus, real-time energy management can be accomplished based on the above attributes.
In addition, the cloud computing platform developed in this research can improve our national M&V
level. Specifically, it helps government in promoting energy efficiency programs and the development
of energy service industries.
Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction efficiency improvement is considered the fastest and most efficient


method. Under the Tradable White Certificate (TWC), Clean Devel-
Of the numerous energy policies implemented by countries opment Mechanism (CDM), and Demand-Side Management (DSM),
worldwide to effectuate energy-saving carbon reduction, energy Energy Service Companies (ESCO) and Energy-Saving Performance
Contracting (ESPC) energy performance is quantified through mea-
⇑ Corresponding author at: 1, Sec. 3, Chung-hsiao E. Rd., Taipei, Taiwan. surement and verification (M&V) procedures. M&V procedures
E-mail address: albertyeh0131@gmail.com (C.-H. Yeh). have become indispensable key items in the energy efficiency field.

http://dx.doi.org/10.1016/j.apenergy.2016.12.034
0306-2619/Ó 2016 Elsevier Ltd. All rights reserved.
498 M.-T. Ke et al. / Applied Energy 188 (2017) 497–507

Nomenclature

AEPCA Australasian energy performance contracting associa- L1 cognitive learning factor


tion L2 social learning factor
AMI advanced metering infrastructure MCEM Monte Carlo expectation maximization algorithm
ASHRAE American society of heating, refrigerating and air- M&V measurement and verification
conditioning engineers P energy consumption (kW)
CDM clean development mechanism PAdjusted pre-installation energy consumption (kW)
CVRMSE coefficient of variation of the RMSE PMeasured post-installation energy consumption (kW)
DSF double skin façade PSO particle swarm optimization
DSM demand side management pbest personal best position
ECM energy conservation measure Qe refrigeration capacity (kW)
EEDSM energy efficiency and demand-side management R2 coefficient of determination
EPC energy performance contracting r random numbers
EPCPs energy performance contracting projects TOA atmosphere temperature (°C)
ESCO energy service company Teva refrigerant evaporation temperature (°C)
ESPC energy saving performance contracting TWC tradable white certificate
EVO efficiency valuation organization Vmax maximum particle velocity
FEMP federal energy management program Vmin minimum particle velocity
GAE Google app engine w inertia weight
GP Gaussian process Xmax maximum particle position
gbest group’s best position Xmin minimum particle position
HERO home energy rebate offer xi position of particle i
IPMVP international performance measurement and verifica- Z relative error
tion protocol Z_minAvg convergence condition
IoT Internet of Things
k iteration number

Due to the importance of M&V, various M&V guidelines have been of the value of EPCPs. The result showed that third-party M&V
published to illustrate the concepts and methodologies of M&V to mechanism can reduce energy savings risk. International Perfor-
let users quickly understand M&V through simple case mance Measurement and Verification Protocol [9] has been widely
descriptions. adopted internationally. It provides M&V concepts and methodolo-
Related literatures such as the study by Bertoldi and Rezessy [1] gies to determine energy and water savings for various energy con-
described the concept and main elements of TWC scheme in the servation measures (ECMs) in existing and new buildings and
European Union (EU). In TWC, energy saving is viewed as a trad- industrial processes. American Society of Heating, Refrigerating
able commodity. Hence, proper quantification of energy saving and Air-Conditioning Engineers Guideline 14 [10] is used for
through M&V is regarded as an important element of TWC. M&V of energy and demand savings in building energy manage-
Michaelowa and Jotzo [2] used quantitative model to explore the ment projects and in commercial transactions between ESCOs
relations between the markets of ‘‘transaction cost and institu- and end users. Australasian Energy Performance Contracting Asso-
tional rigidities” and CDM. Transaction cost comprises of numerous ciation developed the best practice guide [11]. This guide provides
costs such as baseline development, monitoring costs, and verifica- M&V concepts and methodologies to calculate energy and water
tion and certification. The reduction of transaction costs positively savings. This M&V guide can be applied in residential, commercial
impacts the CDM market. Eskom [3] described that energy effi- and industrial buildings, and industrial processes. Federal Energy
ciency and demand-side management (EEDSM) projects cover a Management Program M&V Guideline [12] contains M&V methods
wide range of electricity-related activities and strategies in South and plans for quantifying the energy and water savings in
Africa. The M&V can be used to provide an impartial quantification performance-based contracts.
and assessment of project impacts and savings that result from However, different energy conservation measures require dif-
EEDSM activities. Vine et al. [4] described standard M&V proce- ferent methods to calculate the baseline model and energy saving
dures and contract terms that can help both end users and the performance. Therefore, many M&V studies have been conducted.
financial companies better understand performance contracting. Lee [13] suggested that long-term monitoring produces accurate
Gan [5] described current situations, barriers, and corresponding energy savings calculations for individual cases because actual
removal measures of ESCO industries in China. One of the stimula- lighting conditions at individual sites may differ from client-
tive measures is to build standard energy performance contracts provided data and this difference may introduce errors in the
and M&V procedures. Bertoldi et al. [6] considered the result of energy savings calculations. Dalgleish and Grobler [14] used linear
reviews and analyses of ESCO industries in the EU and the new regression to construct a baseline model for conveyor belts; their
accession countries to formulate a long-term strategy. One of the model verified that installing a motor sequencing controller on
actions is to standardize contracts and M&V procedures; it will conveyor belts reduced energy use and demand. Dong et al. [15]
help both end users and the financial community to easily under- used regression analysis to construct a baseline model for energy
stand the M&V procedures. Lee et al. [7] used two questionnaire consumption in a building. The parameters for the baseline energy
surveys to identify the risks in energy performance contracting consumption were the outdoor dry-bulb temperature, relative
projects (EPCPs). The results indicated that M&V risks are one of humidity, and global solar radiation. The accuracy of the baseline
the top three key risk factors. Qian and Guo [8] developed the bar- model was verified using statistical indicators such as the coeffi-
gaining model of the shared savings contract and the basic model cient of determination (R2) and the coefficient of variation of the
of the forecast-commitment contract to analyze the uncertainty root-mean-square error (CVRMSE). Reddy and Claridge [16]
M.-T. Ke et al. / Applied Energy 188 (2017) 497–507 499

developed the statistical indicator DEsave/Esave to evaluate the consumption and the model forecast electricity consumption men-
goodness of fit of baseline models. Kissock and Eger [17] used mul- tioned above were explored by recalibrating the model using
tivariable piecewise regression models to construct baseline mod- actual measured data and by verifying model accuracy through
els. In particular, they employed actual case studies of retrofitting statistical indicators. Research results indicated that the main
to show that energy efficiency calculations require baseline adjust- reasons for discrepancies were the usage pattern of the building
ments via variables such as production output and weather condi- and the occupant’s behaviors. The author also indicated that appro-
tions during the post-retrofit period. Kaiser and Pulsipher [18] priate monitoring of energy performance is necessary to ensure the
randomly selected 60 participants in the Home Energy Rebate maintenance of high energy efficiency for buildings. Granderson
Offer Program (HERO) and used the International Performance et al. [29] applied the energy use data of 537 commercial buildings
Measurement and Verification Protocol (IPMVP) Option C to exam- to compare and analyze the prediction accuracy, model complex-
ine the relationship between daily electricity consumption and ity, and computational time of 10 baseline energy models through
parameters such as cooling degree days, heating degree days, testing procedures and performance metrics. Research results indi-
dew-point temperature, and energy price, and a baseline model cated the interval data baseline model showed great promise using
was constructed using regression analysis. Ginestet and Marchio advanced metering infrastructure in collecting building energy
[19] described the IPMVP methodology and its related details data to conduct energy performance test. Burkhart et al. [30]
and used four M&V options, labeled A through D, to calculate combined the Gaussian process (GP) model and Monte Carlo
energy savings gained by improvements to the air handling unit expectation maximization algorithm (MCEM) to develop a new
in a conference room. This case study involved the installation of GP-MCEM approach. Through case verification, this GP-MCEM
variable-speed fans and CO2 sensors in the air handling unit. The approach can significantly increase the robustness and mitigate
rate of outdoor air influx was determined from the indoor CO2 con- volatility of the prediction under the input of data uncertainty. In
centrations. Their improvements effectively reduced the energy addition, this new method can also be applied to the situations
consumed by fans and by preheating outdoor air during the winter. of limited and sparse measurement data for the reduction of
Xia and Zhang [20] used mathematical functions to define M&V M&V data collection and sensor installation costs.
models and plans. They also used case studies to explain how The aforementioned literature review shows that M&V baseline
M&V models can produce optimized M&V plans. Walter et al. models are typically constructed using regression analysis meth-
[21] developed a regression model to assess the baseline energy ods, along with real-time and long-term monitoring to accurately
load model and used cross validation to quantify the uncertainty assess the performance of energy conservation measures. Mean-
in the baseline model predictions. This baseline regression model while, PSO provides more accurate predictions than the least error
and its corresponding uncertainty algorithm were validated using squares techniques. Therefore, in this study, a cloud computing
the energy load data from 17 buildings. Ke et al. [22] used IPMVP platform for measuring and verifying real-time energy perfor-
Option D to analyze energy use in an office building. Actual utility mance is constructed according to the M&V methodology while
bills were compared to simulated results to correct the model and integrating the PSO algorithm, JAVA programming language, and
determine its accuracy. The effects of changes to the energy con- cloud computing techniques. By automatically establishing base-
sumption factors, such as window glass, outer wall material, line models for pre-installation and automatically calculating
worker density, and lighting density, on changes to the office real-time energy performance after post-installation, the proposed
building’s overall energy consumption were also examined. cloud computing platform is expected to reduce the time and cost
Particle swarm optimization (PSO) was developed by Eberhart of M&V, while improving the accuracy of energy performance cal-
and Kennedy [23]. This is a type of optimization algorithm with culation. In addition, this platform enables accessing information
the advantages of simple parameter settings, rapid convergence, regarding fluctuations in energy performance to monitor the
and high accuracy. PSO has been applied to numerous engineering operations of equipment and systems.
and scientific studies. Assareh et al. [24] used PSO and a genetic
algorithm to predict future oil demand based on population, gross
domestic product, and import/export data and found that PSO pro- 2. Introduction to M&V
duced lower average relative errors than did the genetic algorithm.
AlRashidi and El-Naggar [25] used PSO and the least error squares The IPMVP was developed by the non-profit Efficiency Valua-
technique to predict long-term peak demand in electrical systems. tion Organization (EVO) [9] and explains how to establish baseline
Their results were verified with historical data, and PSO was deter- data and calculate energy savings. Baseline data establishment is
mined to be the most accurate prediction method. Satapathy et al. necessary because energy consumption in the baseline period
[26] applied multiple regression linear models to data classifica- (i.e., pre-installation) cannot be measured by instruments after
tion using PSO and a least error squares technique to solve for energy conservation measures (ECM) have been implemented.
the regression coefficient. They found that PSO provided a more Therefore, adjustments to the baseline data are required to deter-
favorable result than did the least error squares technique. mine the energy consumption of the equipment and systems
Andelković et al. [27] used Energy plus software combined with during the baseline period and the energy consumption during
airflow network algorithm to develop double skin façade (DSF) the reporting period (i.e., post-installation) under the same operat-
model for predicting cavity air temperature and velocity to com- ing conditions. The energy performance can then be calculated as
pare simulation results against actual measurement values and shown in Fig. 1 and Eq. (1).
through statistical indicators to validate the accuracy of this model.
Research result showed that the accuracy of this DSF model can be Savings ¼ ðBaseline-Period Use or Demand
used to assess the thermal performance of buildings. Ascione et al.  Reporting-Period Use or DemandÞ
[28] studied the energy use of net zero energy building in the first
 Adjustments ð1Þ
year. The actual value of overall annual electric usage was quite
close to the model predictive value. However, discrepancies were Four options, A through D, are available for M&V. Option A
discovered between the actual electricity consumption and model (Retrofit Isolation: Key Parameter Measurement) and Option B
forecast electricity consumption in the areas of heating and (Retrofit Isolation: All Parameter Measurement) are applicable
domestic hot water, lighting, ventilation, equipment and auxiliary when assessing the performance of a single conservation measure.
items. The reasons for the difference between actual electricity Energy consumption parameters are measured pre-installation and
500 M.-T. Ke et al. / Applied Energy 188 (2017) 497–507

Adjust
Build Correlation
Energy Use

Measured Data
(Old, Baseline)
Measured Data
ECM (Retrofitted)
Installation

Fig. 1. Schematic of energy performance M&V.

post-installation, and the energy performance is calculated using


statistical or engineering calculations. The difference between
Option A and Option B is that all energy consumption parameters
are measured for Option B, whereas only key parameters are mea-
sured for Option A, and the remaining parameters are measured on
the basis of historical data of actual site usage or data reported by
equipment manufacturers. Option C (Whole Facility) and Option D
(Calibrated Simulation) are used to assess energy conservation
measures in entire buildings or factories or in situations where
measurements are difficult. The difference between the two
options is that in Option C, energy savings are calculated using a
combination of energy bill analysis and statistical methods,
whereas in Option D, energy savings are initially calculated by
specialized software that simulate pre-installation and post-
installation energy consumption and then the energy bill data Fig. 2. PSO algorithm flowchart.
are used to correct the simulation models.

The PSO algorithm is illustrated in Fig. 2, and each step is


3. Introduction to PSO described below.

PSO was proposed by Kennedy and Eberhart in 1995 [23]. In (1) For each particle in the initial population, a position and
this algorithm, searches and optimization are based on individual velocity in N-dimensional space are randomly generated.
or group experiences and no mechanism for elimination exists. (2) The fitness value for each particle is determined according to
Optimization is achieved by continually updating searches based the objective function.
on the current optimal solution. In addition, PSO systems converge (3) The fitness value is compared to the particle’s pbest, and the
to optimal solutions faster than genetic algorithms do. particle velocity is adjusted to improve the fitness in the
The basic steps in the PSO algorithm are as follows. First, an ini- next iteration.
tial population and center are randomly generated. Next, an objec- (4) pbest is compared to gbest. If pbest is superior to gbest, then
tive function is used to calculate the fitness value of each particle gbest is adjusted accordingly. gbest is used to adjust all
and the current best value for each particle (pbest). Each particle’s particle velocities for the next iteration.
velocity and position are then adjusted using the current best value (5) Particle positions and velocities are updated using Eqs. (2)
for the group (gbest). The best known solutions of previous itera- and (3).
tions are replaced at each new iteration by the superior solutions. (6) The algorithm is terminated if the termination conditions,
Finally, convergence occurs when the optimal solution is found. which are typically defined as achieving a global optimal
The velocity and position of each particle are defined by Eqs. (2) or reaching the maximum number of iterations, are satisfied.
and (3).
4. Cloud computing platform for measurement and verification
¼ wv ki þ L1 r 1 ðpbest i  xki Þ þ L2 r2 ðgbest i  xki Þ
k k
vkþ1
i ð2Þ of real-time energy performance

xikþ1 ¼ xki þ v kþ1


i ; ð3Þ 4.1. Platform architecture

where vi is the velocity of particle i, xi is the position of particle i, k The architecture of the cloud computing platform for the M&V
is the iteration number, L1 is the cognitive learning factor, L2 is the of real-time energy performance developed in this study is shown
social learning factor, r1 and r2 are random numbers, w is the inertia in Fig. 3. The architecture is divided into three tiers: the on-site
weight, pbest is the personal best position, and gbest is the group’s measurement tier, the cloud computing tier, and the user tier.
best position. These are described below.
M.-T. Ke et al. / Applied Energy 188 (2017) 497–507 501

Fig. 3. System architecture for the M&V cloud computing platform.

(1) On-site measurement tier dynamically created by the server in response to user tier
Measurements of energy consumption and other energy requests.
parameters obtained by instruments (e.g., temperature and (3) User tier
humidity meters, ultrasonic flow meters, and digital amme-  Cloud computing results are transmitted to user-end
ters) are transmitted to data recording devices or computers. computers via the Internet, allowing users to monitor
All collected data are then uploaded to the cloud computing real-time energy consumption and energy performance
platform via the Internet. at any time and from any location.
(2) Cloud computing tier  The interface uses HTML supplemented by JavaScript.
 The cloud computing tier handles creating files, receiving Communication between HTML and JavaScript is handled
data, transmitting data, filtering data, storing data, con- by Ajax (asynchronous JavaScript and XML) technology,
ducting PSO algorithm operations, constructing baseline which greatly reduces the amount of data transferred
equations, and calculating energy savings. between the server and browser and allows for fast
 The Google App Engine (GAE) is adopted as the cloud- server responses.
based server. Computing programs and Web service
interfaces are written in Java and supplemented by tools 4.2. Software architecture
such as the Java development toolkit, Google Web toolkit,
and Java servlets in the Eclipse software. Web pages in Programs developed for this study include PsoRun (server),
hypertext markup language (HTML) or JavaScript are PsoRun (client), and the Data Upload Program. Figs. 4 and 5 show

Fig. 4. Software architecture for the M&V cloud computing platform.


502 M.-T. Ke et al. / Applied Energy 188 (2017) 497–507

Fig. 5. Software development screenshot of the M&V cloud computing platform.

the software architecture and a development screenshot, respec- (2) PsoRun (client): Web page creation
tively. The functions of each program are described as follows.  ProjectList.java is the interface for project lists and can be
divided into:
(1) PsoRun (server): Developed on the GAE platform – ProjectList_SavingView.java, which is the interface for
 ProjectServiceImpl.java is the primary program called in energy savings calculations, and
response to user actions. Its functions involve project – ProjectList_Manage.java, which is the interface for
creation, project storage, project deletion, and PSO managing projects.
operations.
 FileUpload.java is called when users upload pre- (3) Data Upload Program: Program for uploading data
installation, baseline data (e.g., energy consumption and  This program returns data from post-installation real-
other energy parameters) and when cloud computing time readings (i.e., energy consumption and other energy
results are returned. parameters) and uploads this data in a fixed format
 DataUpload.java is called to upload post-installation data to the cloud computing platform for energy savings
(e.g., energy consumption and other energy parameters) calculations.
for energy savings calculations.
 doGet and doPost are hypertext transfer protocol 4.3. Constructing the baseline model using PSO
requests: doGet retrieves and transmits data from the
server in response to user requests and doPost transmits Different energy conservation measures require different types
data from the user to the server. and numbers of energy consumption parameters. Therefore, users
M.-T. Ke et al. / Applied Energy 188 (2017) 497–507 503

can determine the number of independent variables in the baseline weight (w), cognitive learning factor (L1), social learning factor
regression model when constructing a generic multivariate base- (L2), maximum particle velocity (Vmax), minimum particle veloc-
line performance model using the cloud computing platform. ity (Vmin), maximum particle position (Xmax), minimum particle
These models are expanded using simple quadratic equations. position (Xmin), random numbers (r1 and r2), and convergence
For example, Eqs. (4) and (5) show a baseline regression model condition (Z_minAvg).
using three energy consumption parameters (i.e., three indepen- The convergence condition is the relative error of the dependent
dent variables). The pre-installation energy consumption (P) and variable (Z) in the baseline model (Eq. (6)). When the relative error
the measurements of the three independent variables (X, Y, and (Z) is lower than the present value, the PSO program determines
Z) are substituted into the baseline regression model, and the that the convergence conditions have been fulfilled and therefore
PSO algorithm is adopted to determine the coefficient of each inde- terminates the calculations. The interface for inputting the baseline
pendent variable (C0–C9). model parameters is shown in Fig. 6.
n  
depend v ariable  depend v ariablePSO   100%
P ¼ fðX; Y; ZÞ ð4Þ X 
Z% ¼   ð6Þ
P ¼ C 0 þ C 1  X þ C 2  X 2 þ C 3  Y þ C4  Y 2 þ C 5  Z þ C 6  Z2 i
depend v ariable
þ C7  X  Y þ C8  Y  Z þ C9  X  Z ð5Þ
The cloud computing platform first filters the pre-installation 4.4. Energy savings calculations
data by conducting a regression analysis and retaining data that
contain errors of less than 10% for dependent variables. Next, Uploading post-installation data, such as real-time energy con-
PSO regression analysis is used to solve for the independent vari- sumption and other energy parameters, to the cloud computing
able coefficients to complete the construction of the baseline platform for calculations requires inputting a corresponding pre-
model. The filtered data can also be saved as a comma-separated installation project ID and URL, as well as a data file name and file
value file for validation or research. Further, the filtered indepen- path (Fig. 7). Entering the observed post-installation values for
dent variable data can serve as the upper and lower limits to filter energy consumption parameters into the pre-installation baseline
post-installation data returned to the user to ensure that these data model yields the pre-installation energy consumption under
match the ranges of the baseline model and that the calculations of the same operating conditions. Subtracting the post-installation
the energy savings are accurate. value from the pre-installation value yields the energy perfor-
PSO parameters can also be set by the user. Definable parame- mance (Eq. (1)). To ensure the accuracy of the energy savings
ters include the number of particles, number of iterations, inertia calculations, the program for uploading the post-installation data

Fig. 6. Interface for entering the baseline model parameters into the M&V cloud computing platform.
504 M.-T. Ke et al. / Applied Energy 188 (2017) 497–507

Fig. 7. Interface to upload post-installation data to the M&V cloud computing platform.

includes a data filtering function that allows the appropriate data system energy use. According to Eqs. (4) and (5), four parameters,
range to be set. Data records that are not within the preset range refrigeration capacity (Qe), atmosphere temperature (TOA), refriger-
are not uploaded. ant evaporation temperature (Teva), and energy consumption (P),
The cloud computing platform calculates energy performance were selected for the baseline model (Eq. (7)). First, the project
information, including average energy savings (kW), total ID, project name, and PSO parameters (Table 1) were entered
kilowatt-hours savings (kW h), average percentage of energy sequentially into the real-time energy performance M&V cloud
savings (%), and energy cost savings (in New Taiwan Dollars). In computing platform. Next, a data file containing pre-installation
addition, a time interval calculation function was specially energy consumption and other energy parameters was uploaded
designed to allow users to specify a time interval for the energy for the calculations. When the construction of the baseline model
savings calculations, and fluctuations in the pre-installation was complete, the message ‘‘true” was displayed for the field
energy consumption (PAdjusted) and post-installation energy con- ‘‘computations completed” and the C0–C9 coefficients were
sumption (PMeasured) are displayed on a graph. Users can a mouse acquired. The R2 value for the baseline model was also displayed,
over the curve to see the data record associated with a specific time as shown in Fig. 9.
point and to gain a quick view of the energy savings, as shown in
Fig. 8. P ¼ C 0 þ C 1  T OA þ C 2  T 2OA þ C 3  Q e þ C 4  Q 2e þ C 5  T ev a
þ C 6  T 2ev a þ C 7  T OA  Q e þ C 8  Q e  T ev a þ C 9  T OA  T ev a
5. Case study: Energy conservation in a freezer and cold storage
ð7Þ
system
After the post-installation data for energy consumption and
The M&V cloud computing platform was used to examine other energy parameters were uploaded to the cloud computing
energy conservation for the freezer and cold storage system of a platform for the real-time M&V of the energy performance, the
hypermarket. The 150HP air-cooled freezer in this study has been program indicated the energy savings accumulated over 11.5 h.
in use for a long time. Measurement showed that the operating Specifically, the average energy savings was 39.56 kW, the total
efficiency of the freezer is low and the energy consumption of its kilowatt-hours saved was 534.03 kW h, the average percentage of
traditional electric defrost is high. After the assessment, it was energy savings was 44.52%, and energy cost savings was NT
recommended that the high-efficiency 120HP freezer be used as $1869(equivalent to $US 58.4 dollars, $US 1 dollar = $NT 32 dol-
replacement and hot gas bypass defrost be introduced to reduce lars), as shown in Fig. 10.
M.-T. Ke et al. / Applied Energy 188 (2017) 497–507 505

Fig. 8. Energy performance interface for the M&V cloud computing platform.

Table 1 platform for the M&V of the real-time energy performance that
PSO parameter vales. integrates the M&V methodology, PSO, and multivariate regression
Inertia weight (w) 0.4 Cognitive learning factor (L1) 2 analysis modeling was successfully constructed on the GAE cloud
Social learning factor (L2) 2 Maximum particle velocity 5 platform with real-time and automatic calculation of energy per-
(Vmax) formance. The proposed cloud computing platform features a
Minimum particle velocity 5 Maximum particle position 20
user-friendly graphical interface, while combining the PSO algo-
(Vmin) (Xmax)
Minimum particle position 20 Number of particles 300 rithm to effectively enhance the accuracy of the multivariate
(Xmin) regression baseline models.
Maximum number of 10,000 Convergence condition 0.1 The M&V cloud platform can use uploaded real-time energy
iterations (Z_minAvg)
consumption data to automatically create features of pre-
installation energy baseline model and post-installation real-time
energy performance calculation to effectively improve the M&V
6. Conclusions and future work efficiency. In addition, fluctuations in the energy savings can be
observed using the proposed platform, which allows for the mon-
Because nations worldwide have vigorously promoted energy itoring of equipment and system operations and the early detec-
efficiency programs and energy service industries, the accuracy tion of abnormal functioning. In case of any anomaly, equipment
of energy performance quantification using measurement and and system operations can be corrected in a timely manner to
verification (M&V) has become crucial. At present, energy perfor- maintain good energy performance.
mance M&V are mostly conducted by on-site short-term measure- The cloud platform for real-time energy performance M&V
ments combined with engineering calculation. It requires adopts the general form equation to establish baseline regression
substantial input of time and labor, and it may result in discrepan- models. Hence, after the independent variable figures of this base-
cies between the calculated results and actual energy saving line regression model are determined by the users and the values
performance. of all the independent variables are uploaded onto the platform,
Based on the foregoing factors, in this study by utilizing JAVA, a baseline regression model can be established and the energy
HTML, Eclipse, and other software tools, a cloud computing performance can be calculated. Thus, it can provide energy
506 M.-T. Ke et al. / Applied Energy 188 (2017) 497–507

Fig. 9. Screenshot a baseline model being constructed for the pre-installation cold storage system.

Fig. 10. Screenshot of the post-installation energy savings for the cold storage system.
M.-T. Ke et al. / Applied Energy 188 (2017) 497–507 507

performance M&V for any industry and any energy conservation determining energy and water saving, volume I; 2012 <http://www.evo-
world.org>.
measure.
[10] American society of heating, refrigeration and air conditioning engineers.
The limit of this study is that the building of baseline model was Measurement of energy and demand savings, guideline 14-2002. Atlanta:
obtained from the regression of various parameter values. There ASHRAE; 2002 <http://www.ashrae.org>.
are certain limits on data range. When calculating energy perfor- [11] Australian Energy Performance Contracting Association. A best practice guide
to measurement and verification of energy savings. Commonwealth of
mance, the range of values substituted for each parameter need Australia; 2004.
to be in the range of the baseline model values. This research plat- [12] Department of Energy DOE (USA). M&V guidelines: measurement &
form is equipped with data filtering function in which arbitrary verification for performance-based contractors. Version 4.0; 2015 <http://
energy.gov/eere/femp>.
data range can be set by the user to prevent miscalculation of [13] Lee AH. Verification of electrical energy savings for lighting retrofits using
energy performance. In addition, when the range of the parameter short- and long-term monitoring. Energy Convers Manage 2000;41:
values collected in the baseline model is extensive, there is no need 1999–2008.
[14] Dalgleish A, Grobler L. Measurement and verification of a motor sequencing
to consider the limit on this data range. controller on a conveyor belt. Energy 2003;28:913–27.
Combining the superior characteristics of the aforementioned [15] Dong B, Lee SE, Sapar MH. A holistic utility bill analysis method for baselining
cloud platform, this M&V cloud platform developed in this article whole commercial building energy consumption in Singapore. Energy Build
2005;37:167–74.
can improve our national M&V level. Furthermore, it helps govern- [16] Reddy TA, Claridge DE. Uncertainty of ‘‘measured” energy savings from
ment in promoting energy efficiency programs and the develop- statistical baseline models. HVAC&R Res 2000;6:3–20.
ment of energy service industries. [17] Kissock JK, Eger C. Measuring industrial energy savings. Appl Energy
2008;85:347–61.
For future research direction, this M&V Cloud platform can be
[18] Kaiser MJ, Pulsipher AG. Preliminary assessment of the Louisiana home energy
combined with Internet of Things (IoT) to assist users in building rebate offer program using IPMVP guidelines. Appl Energy 2010;87:691–702.
various energy performance indicators to effectively control [19] Ginestet S, Marchio D. Retro and on-going commissioning tool applied to an
energy use and learn about the energy performances of various existing building: operability and results of IPMVP. Energy 2010;35:1717–23.
[20] Xia X, Zhang J. Mathematical description for the measurement and verification
energy conservation measures. In addition, it can be combined of energy efficiency improvement. Appl Energy 2013;111:247–56.
with IoT to collect the energy consumption data of all kinds of [21] Walter T, Price PN, Sohn MD. Uncertainty estimation improves energy
equipment or industries. Through the analysis of macroscopic data, measurement and verification procedures. Appl Energy 2014;130:230–6.
[22] Ke MT, Yeh CH, Jian JT. Analysis of building energy consumption parameters
the energy performance benchmarks of various equipment and and energy savings measurement and verification by applying eQUEST
industries are established. software. Energy Build 2013;61:100–7.
[23] Kennedy J, Eberhart R. Particle swarm optimization. Proc Neural Networks. In:
Proceedings of IEEE International Conference, vol. 1995; 1944. p. 1942–48.
References [24] Assareh E, Behrang MA, Assari MR, Ghanbarzadeh A. Application of PSO
(particle swarm optimization) and GA (genetic algorithm) techniques on
[1] Bertoldi P, Rezessy S. Tradable white certificate schemes: fundamental demand estimation of oil in Iran. Energy 2010;35:5223–9.
concepts. Energy Effi 2008;1:237–55. [25] AlRashidi MR, EL-Naggar KM. Long term electric load forecasting based on
[2] Michaelowa A, Jotzo F. Transaction costs, institutional rigidities and the size of particle swarm optimization. Appl Energy 2010;87:320–6.
the clean development mechanism. Energy Policy 2005;33:511–23. [26] Satapathy SC, Murthy JVR, Prasad Reddy PVGD, Misra BB, Dash PK, Panda G.
[3] Eskom. The measurement and verification guideline for energy efficiency and Particle swarm optimized multiple regression linear model for data
demand-side management (EEDSM) projects and programmes. Version 10; classification. Appl Soft Comput 2009;9:470–6.
2010 <http://www.eskom.co.za>. [27] Andelković AS, Mujan I, Dakić S. Experimental validation of a EnergyPlus
[4] Vine EL, Murakoshi C, Nakagami H. International ESCO business opportunities model: application of a multi-storey naturally ventilated double skin façade.
and challenges: a Japanese case study. Energy 1998;23:439–47. Energy Build 2016;118:27–36.
[5] Gan DL. Energy service companies to improve energy efficiency in China: [28] Ascione F, Bianco N, Böttcher O, Kaltenbrunner R, Vanoli GP. Net zero-energy
barriers and removal measures. Procedia Earth Planet Sci 2009;1:1695–704. buildings in Germany: design, model calibration and lessons learned from a
[6] Bertoldi P, Rezessy S, Vine E. Energy service companies in European countries: case-study in Berlin. Energy Build 2016;133:688–710.
current status and a strategy to foster their development. Energy Policy [29] Granderson J, Touzani S, Custodio C, Sohn MD, Jump D, Fernandes S. Accuracy
2006;34:1818–32. of automated measurement and verification (M&V) techniques for energy
[7] Lee P, Lam PTI, Lee WL. Risks in Energy Performance Contracting (EPC) projects. savings in commercial buildings. Appl Energy 2016;173:296–308.
Energy Build 2015;92:116–27. [30] Burkhart MC, Heo Y, Zavala VM. Measurement and verification of building
[8] Qian D, Guo J. Research on the energy-saving and revenue sharing strategy of systems under uncertain data: a Gaussian process modeling approach. Energy
ESCOs under the uncertainty of the value of Energy Performance Contracting Build 2014;75:189–98.
Projects. Energy Policy 2014;73:710–21.
[9] Efficiency Valuation Organization (EVO). International performance
measurement and verification protocol: concepts and options for

Вам также может понравиться