Electricity Cost Optimization for Large Loads

through Energy Storage and Renewable Energy

Abstract—Demand charge contributes to a significant portion investing in renewable energy infrastructure, supporting re-
of the electric bill for large electric loads. Shaving peak demand search and development, and implementing favorable policies
and adopting optimal demand management strategies can reduce and incentives to encourage the widespread use of renewable
this cost significantly. This paper deals with an optimization
problem to reduce the electricity cost for a bulk load with the energy technologies.
help of energy storage and renewable energy. An optimal demand The most recent works have been done on cost minimization
control strategy is developed based on the formulated problem of residential loads and is landed micro-grids. Large building
that aims to minimize the total electricity cost. The optimization loads have not been analyzed to a great extent because mini-
problem was solved using the MATLAB based modeling system mizing large building loads can be quite strategic and tricky.
CVX. The system considered here is a real system that is situated
in University of California, Riverside. Available renewable energy While analyzing the large building loads, very high peaks are
(solar) prediction data and lithium iron phosphate battery usually observed. Peak shaving has been considered as a very
data has been used for the simulation purpose. Two different efficient and cost-effective approach for large consumer groups
strategies, battery only strategy and battery with solar strategy, as peak demand cost occupies a major portion of the electricity
are tested with the proposed strategy. The results are validated bill for the consumers in commercial and industrial sectors.
with National Renewable Energy Laboratory (NREL) developed
System Advisor Model (SAM). The results show that 14.5-15.4% Renewable energy like solar, wind etc can be applied for peak
peak reduction is possible which leads to a monthly savings of shaving in large building loads for a short period of time.
6, 410to38,820. Peak demand reduction-related research has been done
Index Terms—Demand cost, Optimization, Energy storage, in the area of determining the size and best operation of
Renewable sources, Bulk loads. battery energy storage systems [1]. Large industrial plants are
particularly interested in the peak shaving application because
I. I NTRODUCTION their electricity bill includes both energy cost and demand
The world’s population and economy are experiencing un- cost. The latter one frequently reaches a level of 50% of the
controlled growth, accompanied by rapid urbanization. This entire cost of the electricity bill and corresponds to the highest
trend is expected to lead to a significant surge in energy value of power demand during a specified time period (usually
demand in the coming years. However, conventional energy between 15 minutes and an hour). Battery energy storage
sources are struggling to keep up with this increasing demand, systems (BESS) have previously been used to lower peak
as many of them are becoming scarce and their availability demand and electricity bills utilizing the tactic of discharging
is diminishing. To address this challenge, renewable energy stored energy at peak times for various sorts of tariffs [1], [2]?
has emerged as a popular and viable solution to meet the [17]. Meanwhile, renewable energy can also play a vital role
growing energy demand. Renewable energy sources, such as for supplying the additional energy in cases when the BESS
solar, wind, hydro, and geothermal power, offer sustainable is not enough or gets out of order due to any unavoidable
alternatives that can help satisfy the increasing energy needs occurrence. But the initial installation cost for large solar and
without further depleting finite resources. wind facility is way too high which makes it an economically
Countries around the world have recognized the importance inefficient plan both from the utility as well as customers’
of transitioning to renewable energy sources and have taken perspective [3].Strong public support for renewable energy
initiatives to meet their energy demands through renewable has prompted academics to look at novel ways to improve
means. For example, California has set a goal to rely en- renewable energy penetrations, mostly in the form of solar,
tirely on zero-emission energy sources for electricity by the wind, and wave energy [4].
year 2045, demonstrating a commitment to renewable energy Traditional approaches, such as on-site diesel generators, gen-
??adoption [15]. The shift towards renewable energy brings nu- erate issues such as cost inefficiency, high fuel consumption,
merous benefits. It helps reduce carbon emissions, mitigating carbon dioxide (CO2) emissions, increased transportation and
the impact of climate change and improving environmental maintenance costs, and quicker depreciation of equipment [5]-
sustainability. Additionally, renewable energy sources offer [7]. Multiple algorithms such as genetic algorithm, ant colony
long-term energy security, as they are naturally replenished optimization algorithm have also been applied to reduce the
and not subject to the same constraints as fossil fuels. electricity cost and smart energy management for bulk loads
The adoption of renewable energy also stimulates inno- [6], [8]- [15]. Previous work has been done in the case of a
vation, job creation, and economic growth in the renewable freestanding hybrid solar-wind system [Hong Kong]. Different
energy sector. Governments, businesses, and communities are storage technologies, such as pumped hydro storage and
flywheels, have also been considered in some circumstances A. Control Strategy
[3], [5].The economics of hybrid photovoltaic-diesel-battery The peak power drawn by the load can be shaved by using
power systems for residential loads have been investigated in energy storage and solar energy individually. Extra power is
[3]. However, no study has been done on the techno-economic drawn from the grid during low-demand periods and stored
analysis of a big load grid-connected system with peak shaving in batteries so that it can be used during peak periods later.
for cost optimization. In [13], authors suggest novel net- Batteries will discharge the energy during peak periods which
zero energy management and optimization methodologies for will eventually reduce the power drawn from the grid and
commercial buildings in metropolitan locations powered by will help reduce the electricity bill. Solar energy has been
renewable energy sources. These renewable energy systems used as additional source of renewable energy along with
are linked to energy storage technologies like pumped hydro the energy storage. Simulation has been done assuming two
and hydrogen taxis. The goal of applying these measures is to different combination of the surplus energy such as battery
get buildings to consume as much energy as they create from only and battery and solar. Fig. 1 depicts the system diagram Fig. 1?
renewable sources, resulting in a net-zero energy balance. and power flow from various energy sources to the building
This research paper primarily focuses on addressing the load.
classical cost optimization problem in the context of a large Renewable energy (solar) can be stored in the BESS and
load consisting of multiple university building accounts, while be used in the moment of need, which will eventually reduce
also taking into consideration the constraints imposed by the amount of required energy to be drawn from the grid.
batteries. Previous studies conducted in this field have not Renewable sources can also provide energy to the load directly.
given sufficient attention to cost optimization and validation Any surplus energy source can reduce the cost of electricity
across multiple accounts. In this paper, the integration of by using at a proper time schedule. But the drawback of
grid energy, solar energy, and battery energy storage systems renewable energy sources such as solar and wind generation
(BESS) is explored with the objective of achieving peak is their high intermittency and heavy dependency on weather
shaving and power leveling at any given time. Simulations conditions. Solar generation data and corresponding weather
are performed on two different combinations of these energy data were collected from National Renewable Energy Labora-
sources in order to improve peak shaving capabilities and tory’s (NREL) National Solar Radiation database '.' ??
reduce overall costs. The research problem is formulated as a
convex optimization problem, where the objective function is B. Notation
the total cost that combines energy cost and demand cost. The To solve the cost saving problem analytically, we have used
simulated results demonstrate significant cost savings for bulk some notations for different parameters such as time slot,
loads. The System Advisory Model (SAM) is employed to solar energy, battery storage etc. The notations that have been
validate the modeled system, and the cost reduction obtained used throughout our algorithm to implement the strategy are
from the formulated MATLAB program closely aligns with summarized below.
the analysis conducted using SAM. Additionally, the paper
considers monthly and annual cost savings as crucial aspects TABLE I
SUMMARY OF NOTATIONS
of the techno-economic analysis.
The organization of this paper is as follows: Section II Notation Notation Description
states the control strategy and description of the notations. i index of a slot in the billing period
ti total power drawn from grid in slot i
Problem formulation and constraints have been explained in gj grid power fed into the load in slot i
section III. Section IV describes the optimization algorithm Ci power charged into the battery in slot i
whereas simulation results are shown in section V. A few di power discharged from the battery in slot i
1i power provided to the load in slot i
limitations with future directions have been noted in section
si solar generation in slot i
VI and section VII gives the concluding remarks. Cmax battery’s maximum charging rate
Dmax battery’s maximum discharging rate
Emax battery’s maximum storage capacity
II. SYSTEM DESCRIPTION ηc li-ion battery’s conversion efficiency
ηl li-ion battery’s leakage efficiency
a the energy price measured in ($/KWH)
The system consists of a large load with an average of b the peak price measured in ($/KW)
6682.838 kW daily for any 15-minute period throughout a µ duration of time slot i
year. It is a billing account of multiple buildings situated Eprimary energy left first slot when billing period starts
ei energy left in battery at beginning of slot i
in University of California Riverside campus. Currently the
buildings have no solar, wind and battery storage for providing
energy during on peak hour. A 100-kW solar panel and 5000
kWh LiF eP O4 battery has been assumed to be connected to III. PROBLEM FORMULATION AND CONSTRAINTS
the buildings for the simulation purpose. Solar and wind data The problem to minimize the electricity cost is an optimiza-
of the location has been predicted for respective weather data tion problem itself. It is an optimization problem with battery,
with machine learning approaches. solar, wind and time constraints. wind???
A. Objective Function the remaining battery energy from the previous period minus
The objective of our problem is to minimize the electricity the discharged energy from the previous time slot times the
cost. Electricity cost has two parts. One is energy charge leakage efficiency. We assume that initially the stored energy
(kWh) and another is demand charge (kW). So, in order to in battery is zero. The fourth constraint tells us that at any time
reduce the electricity cost we have to minimize the following slot i the discharged energy from the battery must be less than
equation. or equal to the stored energy in battery. The fifth constraint
tells us that at any time slot i the power discharged from the
n
X battery must be less than or equal to the maximum discharging
M inimize aµ (ti + b · max(ti )) (1) rate which is 5000 kW. The sixth constraint denotes that at
i=1 any time slot i the power charged in to the battery must be
Where, a denotes the energy price measured in ($/kWh) and less than or equal to the maximum charging rate which is
b is the peak price measured in ($/kW) and is the power 5000 kW. The seventh constraint states that at any time slot i
(kW) consumed from the grid in time slot i and µ is the the remaining energy in battery must be less than or equal to
duration of time slot i. µ is 15 minute duration as the peak maximum storage capacity of battery which is 5000 kWh. The
load has been selected based on the peak of any 15 minutes last constraint states that the total power drawn from the grid,
throughout the monthly billing period. For Riverside Public the power supplied directly to the building load, the power
Utility (RPU) rate schedule (designed for large building load discharged from the battery, the power charged into the battery,
and demand is consistently greater than or equal to 150 kW), the stored energy in the battery, solar generation at any time
the energy price is 0.1033 ($/kWh) and peak demand price is instant must be non-negative.
6.88 ($/kW). The rate schedule is not a Time of Use (TOU)
C. Convex Optimization
schedule and does not vary with the different time periods of
the day. So, the first portion of the objective function denotes The formulated problem is a convex optimization problem.
the energy charge or kWh charge and the second portion of Since the energy charge part in the objective function is linear,
the objective function denotes the demand charge or the kW it can be considered as convex. The demand charge part of
charge. Together they comprise the total electricity cost that the objective function is a function of max function which is
we are trying to minimize. a convex function. The energy price measured in ($/kWh), a,
the peak price measured in ($/kW), b the duration of time
B. Constraints slot, µ and the time itself are positive quantities. So, the
We are trying to solve the optimization problem subject to objective function is a convex function as it will minimize the
some constraints. These constraints are mentioned below: nonnegative weighted sum of two convex functions. All the
constraints are linear and therefore, they can be considered as
convex, too. Since the objective function and the constraints
i. ti = gi + ci ; 1≤i≤n are all convex, the stated optimization problem stated in this
ii. gi + di + si = li ; 1 ≤ i ≤ n paper is a convex optimization problem. This problem has been
solved with MATLAB CVX [16] tool to find out the cost and

Eprimay i=1
iii. ei = compare it with the cost in absence of our algorithm.
ηc ci−1 µ + ηl (ei−1 − di µ) 2 ≤ i ≤ n
iv. di µ ≤ ei ; 1≤i≤n IV. OPTIMIZATION ALGORITHM
v. di ≤ Dmax ; 1≤i≤n The algorithm implemented for convex optimization
vi. ci ≤ Cmax ; 1≤i≤n problem solving has been stated below:
vii. ei ≤ Emax ; 1≤i≤n
Algorithm:
viii. ti , gi , ci , di , ei , si ≥ 0; 1≤i≤n
Step 1: Solve the optimization problem using the building
load data along with the battery specifications. each time slot
The explanations of these constraints are described below: The i.
first constraint stands for the fact that at any time slot i, the Step 2: Generate the decision variables ti , gi , ci , di for each
total power drawn from the grid must be equal to the power time slot i.
supplied directly to building load plus the power supplied to Step 3: Take the following actions for each time slot i for the
charge the li-ion battery. The second constraint denotes that at intended day:
any time slot i, the power supplied directly to the load need to • Draw ti kW from grid.
be commensurate with the summation of the solar generation, • Draw si kW from solar panels.
the power discharged from the battery and power drawn from • Supply gi kW directly to building.
the grid. The third constraint focuses on the energy calculation • Use ci kW from grid to charge the battery.
of the battery. At any instant the battery storage must be • Discharge di kW from the battery to meet the building
equal to the energy supplied from the grid to charge it from load demand.
the previous time slot times the conversion efficiency with
 Fig. 1. Load optimization with battery only Fig. 2. Load optimization with solar plus battery

We applied this algorithm continuously for each day of the Table III demonstrates the peak load before and after
month and optimized the monthly cost. optimization has been performed and the total monthly peak
reduction both in kW and as percentage. For battery only
strategy, the usual peak load is 7,910 kW and the optimized
V. SIMULATION RESULTS
peak load is 6,763 kW which stands for 1,147 kW load
The applied algorithm shaves the peak at a certain limit reduction (14.5% of the total load). The combined battery and
each day and hence the maximum demand has been reduced. solar strategy reduces 1,217 kW load after the optimization
Two different strategies have been adopted for the simulation algorithm has been applied on the system which is 15.39% of
purpose with the proper modification to the algorithm. The the total load (7,910kW).
first strategy does optimization with battery only and the
second strategy includes solar generation alongside battery
TABLE II
for optimization purpose. The simulation results produce good COST SAVINGS
outcomes in terms of both peak reduction and cost minimiza-
tion. Both the strategies reduce the peak load by a significant Actual Optimal Monthly
Strategy
Cost cost Savings ($)
amount. Consequently, a considerable amount of money is Battery 502970 496560 6410
saved as demand charge accounts for the most of the billing Battery +
502970 464150 38820
amount. The following figures and tables present the findings Solar
from the simulation results. Figure 3 shows the actual load
without optimization and the load optimized with battery
only strategy. Figure 4 shows the actual unoptimized load
and the optimized load with battery plus solar strategy. The TABLE III
performance of these strategies are compared in table IV and PEAK REDUCTION
table V which show the peak reduction and the cost savings
Peak Peak
respectively for these two strategies. Peak
without with
Strategy Reduction
Optimization Optimization
($)
Table II tabulates the actual cost of electricity and the ($) ($)
Battery 7910 6763 1147
minimized cost after the optimization algorithm and monthly Battery +
savings both in dollar amount and percentage amount for both 7910 6763 1217
Solar
of the systems. Battery only strategy holds an unoptimized
value of $502,970 and optimized cost of $496,560. As a
result of the optimization, $6,410 can be saved on a monthly From the SAM analysis report, the yearly electric bill
basis which is 1.27% of the actual cost. On the other hand, without system is $6,122,238 and yearly electric bill with the
the actual electricity cost of combined battery plus solar optimization algorithm is $5,656,963. Net savings with system
strategy is $502,970 whereas the optimized cost is 464,150. is $465,275 and the payback period is 17.1 years. Therefore,
As a consequence, $38,820 can be saved monthly which from economic point of view, this algorithm can save a huge
represents 7.72% of the actual cost. amount of money for a customer who intend to use the system
for a long period of time.
VI. LIMITATIONS AND FUTURE CONSIDERATION [2] A. Oudalov, D. Chartouni, C. Ohler, G. Linhofer, A. L. Bess, and B. N.
Bess, “Value Analysis of Battery Energy Storage Applications in Power
The algorithm used here does not consider the Time of Systems,” IEEE PES Power Systems Conference and Exposition, pp.
Use (TOU) demand rate which is different based on different 2206–2211, 2006.
[3] K. Sunderland, M. Narayana, G. Putrus and M. Conlon, ”Levelised cost
time periods like on-peak, mid-peak and off-peak. Mid-peak of energy analysis: A comparison of urban (micro) wind turbines and
and off-peak hours usually offer lower rates and incentives solar PV systems,” 51st International Universities Power Engineering
are offered usually for shifting energy use away from more Conference (UPEC), Coimbra, pp. 1-6, 2016.
[4] S. Mishra, P. Palanisamy, “Efficient Power Flow Management and Peak
expensive peak hours. Undoubtedly, this Time of Use policy Shaving in a Microgrid-PV System,” CoRR, abs/1807.07180, 2018.
can help the users save money and reduce the stress on the [5] ABB, (2015) “Microgrids: Integration for power cost and con-
electric grid as well as utility companies. Moreover, change trol” (ABB Article). Cary, NC : ABB Data Center [Online].
https://library.e.abb.com/public/67d130471f5442d882ddf881b9b7d0f8/
of SOC (State of Charge) of batteries has been considered microgrids%20for%20data%20centers.pdf
as linear with time. For non-linear characteristics along with [6] A. Mishra, D. Irwin, P. Shenoy, and T. Zhu, “Scaling Distributed
other constraints of Li-ion batteries, the algorithm may act dif- Energy Storage for Grid Peak Reduction”, Proceedings of the fourth
international conference on Future energy systems (e-Energy ’13), pp.
ferently. So, this algorithm can be applied by considering these 3–14, 2013.
limitations to produce better results in future. Moreover, deep [7] IEC, Electrical Energy Storage (White Paper).
learning models can be employed to learn the relationships [8] M. Alamaniotis, D. Bargiotas, N. G. Bourbakis and L. H. Tsoukalas,
”Genetic Optimal Regression of Relevance Vector Machines for Elec-
between input variables, output variables, and the constraints tricity Pricing Signal Forecasting in Smart Grids,” in IEEE Transactions
that need to be satisfied. The model can then be used to on Smart Grid, vol. 6, no. 6, pp. 2997-3005, Nov. 2015.
optimize the output by adjusting the input variables while [9] C. Li, F. de Bosio, F. Chen, S. K. Chaudhary, J. C. Vasquez and J. M.
Guerrero, ”Economic Dispatch for Operating Cost Minimization Under
ensuring that the constraints are met. Real-Time Pricing in Droop-Controlled DC Microgrid,” in IEEE Journal
of Emerging and Selected Topics in Power Electronics, vol. 5, no. 1,
C ONCLUSION pp. 587-595, March 2017
[10] Z. Iqbal, N. Javaid, M. R. Khan, I. Ahmed, Z. A. Khan and U.
This paper presents an algorithm that can help in peak Qasim, ”Cost and Load Reduction Using Heuristic Algorithms in Smart
Grid,” 2016 30th International Conference on Advanced Information
reduction of large electric loads which results in the reduction Networking and Applications Workshops (WAINA), Crans-Montana,
of cost in electric bill with demand charge. Most of the 2016, pp. 24-30
previous work that dealt with this type of optimization was [11] M. Dabbagh, B. Hamdaoui, A. Rayesz and Mohsen Guizani, “Shaving
Data Center Power Demand Peaks Through Energy Storage and Work-
done only on small to medium residential or commercial loads load Shifting Control”, IEEE Transactions on Cloud Computing, pp. 1-1,
and on simulated systems. The system considered here is a real 2018.
system with a large load comparable to industrial loads, which [12] Khan, FA, Pal, N, Saeed, SH. “Review of solar photovoltaic and wind
hybrid energy system for sizing strategies, optimization techniques
is situated in the University of California, Riverside’s campus. and cost analysis methodologies”, Renewable and Sustainable Energy
The optimization algorithm applied here on this real system Reviews, vol 92, pp. 937–947, 2018.
generates the load profile for optimal control strategy with a [13] Liu, Jia, et al. ”Net-zero energy management and optimization of
commercial building sectors with hybrid renewable energy systems
view to minimizing the total electricity cost. The constraints integrated with energy storage of pumped hydro and hydrogen taxis.”
of the optimization problems used here are formulated with Applied Energy 321 (2022): 119312.
practical assumptions. Two strategies, one with battery only [14] M. S. Islam, M. S. Rahman and M. A. Amin, ”Beat Based Realistic
Dance Video Generation using Deep Learning,” 2019 IEEE Interna-
and the other with solar plus battery are examined here with tional Conference on Robotics, Automation, Artificial-intelligence and
necessary modifications. Both of the strategies are successfully Internet-of-Things (RAAICON), Dhaka, Bangladesh, 2019, pp. 43-47,
able to do peak shaving and cost reduction for the given load doi: 10.1109/RAAICON48939.2019.22.
[15] https://www.npr.org/2018/09/10/646373423/california-sets-goal-of-100-
profile and existing system. As expected, the generated results percent-renewable-electric-power-by-2045
show that the battery plus solar strategy is superior to battery [16] M. Grant and S. Boyd. CVX: Matlab software for disciplined convex
only strategy, providing a monthly savings of almost six times programming (web page and software). http://stanford.edu/boyd/cvx,
June 2009.
than the latter. The results are proven acceptable as they are [17] A. S. M. J. Hasan, L. F. Enriquez-Contreras, J. Yusuf and S.
validated with SAM. With proper long-term planning this Ula, ”A Comprehensive Building Load Optimization Method from
amount of savings can prove to be profitable for the industrial Utility Rate Structure Perspective with Renewables and Energy
Storage,” 2021 International Conference on Smart Energy Sys-
customer with a similar size load. The utility companies can tems and Technologies (SEST), Vaasa, Finland, 2021, pp. 1-6, doi:
also be benefitted from such approach as it the optimized 10.1109/SEST50973.2021.9543188.
load with reduced peak would put less stress on generating
units. The future of energy use will be dominated by renew-
able generation with storage systems and this work can be
made impactful and valuable with proper implementation and
integration

R EFERENCES
[1] A. Oudalov, R. Cherkaoui, and A. Beguin, “Sizing and Optimal Opera-
tion of Battery Energy Storage System for Peak Shaving Application,”
IEEE Lausanne Power Tech, pp. 621–625, 2007.