Вы находитесь на странице: 1из 13

Implementation of Interruptible Load Program for

Peak Shaving Using Binary Particle Swarm


Optimization

Reland Marc B. Tinambacan John Neil Bibera


EECE department, College of Engineering EECE department, College of Engineering
Mindanao State University – Iligan Institute of Technology Mindanao State University – Iligan Institute of Technology
Iligan City, Philippines Iligan City, Philippines
trelandmarc@gmail.com

Abstract—The Interruptible Loads (ILs) which participate in solutions within its disposal to address the power situation of
the contingency reserve market can mitigate the reserve shortage Mindanao. The recent installations of Coal-fired power plants
during emergency situations. In this paper, a model on the day- have helped shift the supply deficiency to generation adequacy
ahead scheduling for the ILs participants in the contingency
reserve spot market is introduced. The Distribution Utility [2]. As reported in the 2017 power situation report in
implements this model to determine the interruption schedule of Mindanao, the Mindanao grid has been stable for the first half
the ILs considering system operating reserves. It is shown that of the year compared to previous years [3]. However, the
the interruptible load program helps to reduce the system peak occurrence of shortages is still to be expected and therefore,
demand and raise the level of system reserve during the peak must not be taken lightly.
hours and in cases of tight power situations. This paper aims to
assess the implementation of the Interruptible Load Program
(ILP) in addressing the Peak demand. An Optimal day-ahead In previous years, when generation fails to meet the
interruption scheduling for the participating Interruptible Load increasing demand, the rise of peak demand is expected. Peak
customers is obtained. The interruption scheduling is a demand can occur in a short span of time within a day.
noncontiguous and non-convex optimization problem for which Contingency reserves are expected to be utilized in order to
simulation-based algorithm like Binary Particle Swarm mitigate this sudden spike in load demand. In Mindanao, the
Optimization is used. The simulation is based on a 9-bus IEEE
test system. MATLAB is used for the simulation. The ILP has
DOE has employed diesel peaking plants like Mapalad Diesel
been shown to be an effective contingency reserve by Distribution Power Plant, EEI Diesel Peaking Power Plant, and Iligan
Utility in dealing with the reduction of the system peak demand. Diesel Power Plant [4]. Likewise, DOE also initiated the
The simulated Interruptible Load Program was able to reduce program called Interruptible Load Program (ILP).
the system peak demand by 160.48 MWh. It has also reduced the
Total Generation Cost by PHP 1,448,284.03 and average The ILP is a demand-side management solution
Electricity Rates by 0.4%. The results obtained satisfy the goals
of ILP. Furthermore, the ILP modeling within the electricity
established by the Department of Energy (DOE) and the
market was achieved. Energy Regulatory Commission (ERC) to help mitigate the
energy supply deficiency in the power grid until new
Keywords—component; formatting; style; styling; insert (key capacities become available [5]. Any consumer connected to
words) the grid who also has a back-up generation unit is a potential
ILP participant [6]. ILP participants will be compensated,
I. INTRODUCTION should they participate during periods of energy supply
deficiency. ILPs are effective in mitigating situations of tight
Due to the increasing demand for electrical energy, energy supply.
Philippines has been facing shortages in its supply. Prior to
2017, notable cases of these shortages have occurred. The As of 2017, Wholesale Electricity Spot Market (WESM)
power situation outlook from the National Grid Corporation of has been integrated in Mindanao. The competition of
the Philippines (NGCP) showed a 49-MW reserve deficit in generation in the electricity market is to be expected. As
Mindanao based on computed data of 10:00 AM, April 14, Distribution Utility (DU) entered bilateral contracts, electricity
2016 [1]. The Mindanao grid is expected to have 1,420 MW of rates increased with the utilization of peaking plants.
available supply against a 1,469- MW system peak demand. Addressing peak demands with higher electricity rates to
The occurrence of El Niño affected the operation of Hydro customers is not economically efficient. Moreover, resorting
Power Plants in Mindanao which caused a supply deficiency to Load Management such as rotational blackouts may not be
in both the Agus and Pulangi Hydro Power Plants [2]. a good practice.
In light of the series of energy shortages of energy in
Mindanao, Department of Energy (DOE) has come up with

1
This paper aims to assess the implementation of the ILP in evaluates its individual fitness. A moving particle cannot be
addressing the Peak demand. An Optimal interruption understood as moving without speed, so in order for a particle
scheduling for the participating Interruptible Load (IL) to find the best position to achieve, it needs its velocity to do
customers is obtained. The interruption scheduling is a so. If the best position achieved by each particle can be termed
𝑝
noncontiguous and non-convex optimization problem for as particle best (𝑥𝑏 ) and the best position of the best
which simulation-based algorithms like Binary Particle Swarm 𝑔
performing particle can be termed as global best (𝑥𝑏 ), then
Optimization is used. The simulation is based on a 9-bus IEEE the position and velocity of each particle can be updated by
test system. MATLAB is used for the simulation. every iterative learning approach which can be achieved by
these equations:
II. MODELING AND IMPLEMENTATION OF ILP
𝑝 𝑔
𝑣𝑖𝑘+1 = 𝑤𝑣𝑖𝑘 + 𝑐1 𝑟𝑎𝑛𝑑(𝑥𝑏 − 𝑥𝑖𝑘 ) + 𝑐2 𝑟𝑎𝑛𝑑(𝑥𝑏 − 𝑥𝑖𝑘 )
The primary purpose of the study is to optimize a daily
schedule of possible ventures of DU to the ILP to reduce the
peak power level. Specifically, the study aims the following: 𝑥𝑖𝑘+1 = 𝑥𝑖𝑘 + 𝑣𝑖𝑘

a. Characterize the ILs to be included in the ILP. Where 𝑣𝑖𝑘 – Velocity of particle 𝑖 at iteration 𝑘, 𝑤 is the
b. Develop a load demand profile based on WESM data. Inertia coefficient, 𝑐1 , 𝑐2 are the acceleration coefficient for
c. Design an overall system model considering the ILP particle and global best respectively. While rand is any
model, test system constraints, compensation cost, random number generated between 0 and 1. The particles will
day-ahead scheduling, and perform a system continue to update their position until a maximum number of
simulation through Optimal Power Flow (OPF) using iterations are achieved or a stopping criterion is satisfied. The
MATPOWER. final solution for a problem is the global best or the global best
d. Develop a MATLAB program for the optimization once the iterative process converges.
problem towards Binary Particle Swarm
implementation given participating Interruptible PSO uses continuous values for every particle. For load
Load data, DU contracted capacity, and power scheduling problems, solutions for each schedule works as
system constraints. on/off or binaries, which, by means, can be understood as a
e. Evaluate the results from ILP implementation. discreet solution. Hence, a discreet solution space is needed
for load scheduling problems. A discrete binary version of
Several published studies present the ILP model, and in PSO is devised to solve for such problems with discreet
which, certain aspects were taken. Those related works solutions space [13].
provide relevant overview on the different literature and
modeling techniques that both useful and significant. The For Binary Particle Swarm Optimization (BPSO), same
works of Yu et al., [7] presented the modeling and assessment velocity function is used. However, the velocity is only limited
of interruptible-load programs in the electricity markets. to a certain range, +𝑉𝑚𝑎𝑥 and -𝑉𝑚𝑖𝑛. Finally, the velocity
Another work of Qi et. al [8] studies the incentive interruptible can then be mapped to [0, 1] using the sigmoid function
load contract with risk preference of power companies. defined as,
Electricity market risk management using forward contracts
with bilateral options, conducted by Chung et al. [9]. 𝟏
∅𝒊 (𝒕 + 𝟏) =
Fahrioglu et al.,[10] provide a model of interruptible load 𝟏+ 𝒆−𝒗𝒊(𝒏,𝒕+𝟏)
contract for minimum compensation cost. Wang et al., also
conducted a research that provides the optimal purchase model The position can then be updated by comparing sigmoid to
of an interruptible load contract [11]. any random number. 𝑥𝑖 = 1 If 𝑟𝑎𝑛𝑑 < ∅𝒊 (𝒕), 𝑥𝑖 = 0
otherwise.
III. PARTICLE SWARM OPTIMIZATION
IV. OPTIMAL POWER FLOW
Particle Swarm Optimization (PSO) is an optimization
method based on population that aims to solve continuous In this paper, the objective of the OPF problem is to
non-linear function [12]. The method was designed from the identify the minimum generation cost of generator units that
behaviors of a group of birds that seems to flock towards one meet equality and inequality constraints. The equality
common destination. The solutions to the problem can be constraints represent conventional power flow equations and
understood as a swarm of particles. For a given problem, the the inequality constraints represent the system operating and
particles are initially spread to a certain position throughout a control limits.
solution space in a random manner. As the optimization
progresses to find a solution for a problem, the particles move Mathematically, the OPF problem is formulated as a
around the solution space while being simultaneously nonlinear optimization problem with equality and inequality
influenced by the best position achieved by each particle and constraints, as shown below:
the best position of the best performing particle. A fitness
function determines the performance of a particle and 𝑂𝑏𝑗𝑒𝑐𝑡𝑖𝑣𝑒 𝐹𝑢𝑛𝑐𝑡𝑖𝑜𝑛 ∶ 𝑓(𝑥, 𝑢)

2
𝐸𝑞𝑢𝑎𝑙𝑖𝑡𝑦 𝐶𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡𝑠 ∶ 𝑔(𝑥, 𝑢) = 0 Across the sea, ILP had been widely accepted as an
𝐼𝑛𝑒𝑞𝑢𝑎𝑙𝑖𝑡𝑦 𝐶𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡𝑠 ∶ ℎ(𝑥, 𝑢) ≤ 0 ancillary service in services that have been referred to as such.
According to the North American Electric Reliability
A. Objective Function Corporation (NERC) in (2007), interruptible load
management (ILM) is recognized as one of the contingency
Input-Output load characteristics of generator units exhibit reserve services. Similarly, the Australian electricity market
a nonlinear, convex structure. The objective function of OPF recognizes "load shedding", both as a frequency control
is presented in equation. service and a network loading control ancillary service [15].
𝑁𝐺
𝐹= ∑ 2
(𝑎𝑖 𝑃𝐺𝑖 + 𝑏𝑖 𝑃𝐺𝑖 + 𝑐𝑖 ) According to the amended resolution, the de-loading
𝑖=1 compensation shall be computed by multiplying the "de-
loaded kWh" less registered kWh in the meter. If any, during
B. Limit Conditions periods when the DU requests a participating customer (PC) to
Load Flow Equations de-load, multiplied by the incremental de-loading cost per
kWh.
𝑁
𝑃𝐺𝑖 − 𝑃𝐿𝑖 − 𝑉𝑖 ∑ 𝑉𝑗 (𝐺𝑖𝑗 𝑐𝑜𝑠𝜃𝑖𝐽 + 𝐵𝑖𝑗 𝑠𝑖𝑛𝜃𝑖𝑗 ) = 0 𝐷𝑙𝐶 = 𝑀𝐶 + (𝑖𝐷𝑅 𝑥 𝐶𝐿)
𝑗=1

𝑁
𝑖𝐷𝑅 = (𝐺𝐶𝑜𝐹 ∗ 𝐹𝐶𝑅) − 𝑃𝐶𝐴𝑅
𝑄𝐺𝑖 − 𝑄𝐿𝑖 − 𝑉𝑖 ∑ 𝑉𝑗 (𝐺𝑖𝑗 𝑐𝑜𝑠𝜃𝑖𝐽 + 𝐵𝑖𝑗 𝑠𝑖𝑛𝜃𝑖𝑗 ) = 0
𝑗=1 Where:

𝑖 = 1, … , 𝑁 𝐷𝑙𝐶: De-loading Compensation


𝑀𝐶: Maintenance Cost
Generator Limits 𝑖𝐷𝑅: Incremental De-loading Rate
𝑚𝑖𝑛 𝑚𝑎𝑥
𝑃𝐺𝑖 ≤ 𝑃𝐺𝑖 ≤ 𝑃𝐺𝑖 𝐶𝐿: Compensable Load in kWh
𝑚𝑖𝑛
𝑄𝐺𝑖 ≤ 𝑄𝐺𝑖 ≤ 𝑄𝐺𝑖 𝑚𝑎𝑥 𝐺𝐶𝑜𝐹: Generation Cost of Fuel
𝑚𝑖𝑛
𝑉𝐺𝑖 ≤ 𝑉𝐺𝑖 ≤ 𝑉𝐺𝑖 𝑚𝑎𝑥 𝐹𝐶𝑅: Fuel Consumption Rate
𝑃𝐶𝐴𝑅: Participating Customer Average Rate
𝑖 = 1, … , 𝑁𝐺
Where Generation Cost of Fuel is the average price of
Where 𝑃𝑚𝑖𝑛 𝑚𝑎𝑥
𝐺𝑖 𝑎𝑛𝑑 𝑃𝐺𝑖 is the minimum and maximum diesel fuel in major oil companies as of the end of the previous
active power of the generator 𝑖𝑡ℎ respectively. While calendar month in the City or Municipality where the
𝑄𝑚𝑖𝑛
𝐺𝑖
𝑎𝑛𝑑 𝑄𝑚𝑎𝑥
𝐺𝑖
is the minimum and maximum reactive power participating customer is resides, Fuel Consumption Rate is
0.34 liter/kWh as fixed by the commission and Participating
of the generator 𝑖𝑡ℎ respectively and 𝑉𝑚𝑖𝑛 𝑚𝑎𝑥
𝐺𝑖 𝑎𝑛𝑑 𝑉𝐺𝑖 minimum
𝑡ℎ Customer Average Rate is the participating customer’s
voltage value of the generator 𝑖 respectively.
average rate for the current billing period, and for simplicity,
this rate is the average rate in the system. Compensable kWh
Transformer Limits is the actual de-loaded kWh for the current billing period and
Maintenance Cost is Php 0.32/kWh multiplied by
𝑚𝑖𝑛 𝑚𝑎𝑥
𝑇𝑇𝑖 ≤ 𝑇𝑇𝑖 ≤ 𝑇𝑇𝑖 Compensable kWh or Php 23,548.00/month, whichever is
𝑖 = 1, … , 𝑁𝐺 lower.

𝑇𝑚𝑖𝑛
𝑇𝑖 𝑎𝑛𝑑 𝑇𝑇𝑖
𝑚𝑎𝑥
is the minimum and maximum level
control of the transformer 𝑖𝑡ℎ respectively.

V. BILLING AND COMPENSATION

The Energy Regulatory Commission (ERC) promulgated


resolution No. 08, Series of 2013 in which ILP was amended.
Effective as of May 2013, this resolution provides rules to
govern the ILP to the distribution utilities (DU). ILP is an
agreement between the DU and the participating customer to
compensate the incremental cost incurred due to the full or
partial de-loading of the customer when it de-loads from the
DU as requested from ERC. The Interim Mindanao Electricity
Market (IMEM) and the ERC basically function as the Fig. 1 Compensation Rate
mediators between the DU and the power consumers who
wish to enter an agreement or contract for ILP.

3
Fig. 1 shows the compensation price rate for the over a minimum duration before another curtailment. For
participating customer. Most participating ILPs are within the simplicity, there is a requirement for maximum curtailment
1MW mark and above which gives them about Php 6.3/kWh period (MAX OFF TIME) and minimum duration between
of rate price. curtailments (MIN ON TIME).

ILP is also integrated in a test-bus system where the ILs’


are connected in the load buses. Each ILs’ are designated on
respective load buses based on the hypothetical data from
previous studies on ILP. In the case of ILP activation, the ILs’
are curtailed within the load points.

B. Load Forecast Model

A load forecast is simulated based on the load forecast of


Wholesale Electricity Spot Market (WESM) as observed on
the website www.wesm.ph. The WESM day-ahead projection
for the system demand of the Luzon grid has a percentage
Fig. 2 Price behaviors of interruptible load participants error of at most 3%. The forecast load profile is designed like
an actual load demand of a DU. As observed in Fig. 3in page
Fig. 2 shows the price exponential behaviors based on 20, demand spikes during peak hours and lesser demands
different reserved levels. When the quantity reserve is below during off-peak hours are included in the load profile.
100%, as required by the commission the price increases, thus
attracting more participating customers. A negative reserve The referenced load profile from WESM has a demand
means that not enough generating plants are available in the peak of 11 GW. In this study, the referenced load profile is
next 24 hours that can supply the required reserve power in the simulated on a per-unit conversion with a base of 10 MW. As
entire system, which normally leads to a rolling blackout. The a result, the demand peak would be equivalent to 1.1 p.u.
bidding scheme insures that the grid stays operational by
ensuring a load shedding effect rather than rolling blackouts.
12000
VI. PROBLEM FORMULATION 10000
Load (MW)

8000
As Interruptible Load Program (ILP) is utilized during
6000
Peak hours or as a contingency reserve in emergency
situations, Distribution Utility (DU) executes a day ahead 4000
scheduling optimization among the participating customers. In 2000
a day ahead scheduling, hourly reserves are determined based
on the Yellow alert level designed by ERC. Interruptible 0
Loads' (IL) initial incentives were formulated by the 1 3 5 7 9 11 13 15 17 19 21 23
compensation cost designed by ERC. A new incentive Time (hr.)
function introduced in this research modifies the compensation
cost based on available reserves.
Fig. 3 WESM Load Forecast
A. Interruptible Load Program Characterization
C. System Model
The ILP is incorporated in a day-ahead electricity market.
This enables the ILP participants to plan for their individual
curtailment capacity as well as the individual compensations. 1. Interruptible Load Program Model
Initial compensations are obtained based on the compensation
cost designed by ERC.

The interruption scheduling of IL deals with the same


complexity as the scheduling of generators. Aside from the
individual curtailment capacity and initial compensations of
ILs, the ability to provide curtailments and curtailment
requirements of ILs is necessary. There are ILP participants
that prefer continuous interruptions while others require
frequent interruptions. There are ILP participants that allow
interruptions over a certain amount of time while there are
also ILP participants that require to be connected to the grid Fig. 4 ILP Model

4
Fig. 4 depicts the interruptible load participants entering generation factors. With the application of Optimal Power
the market and how it aids the network system to mitigate Flow (OPF) in this study, the optimal generation cost is
peak load through the interruption of its own load and to obtained. Generation Cost is then used in the latter part of this
deviate from blackouts. study, particularly in the Optimization. The case scenario for
the distribution system is manifested by the 9-bus IEEE test
2. Interruptible Load Program Activation system.

During the duration of the simulation, it is important to


determine as to when the ILP is activated. Based on the
MERALCO ILP, this is condense in Fig. 5, the National Grid
Corporation of the Philippines (NGCP) will notify the DU
when an anticipated power supply deficiency is detected
within the system [5]. Levels of the available reserve are
categorized in order to classify the type of notification the
designated operator shall give. These classifications are
known as the alert levels. Among these levels are the normal
state, yellow alert, and red alert. The ILP is activated when red
alert level is notified, but with the occurrence of yellow alert
level, ILP is prepared for possible activation.

In this study, yellow alert level is the basis for the


activation of the ILP. The total power of the system at a
specific hour should reach yellow alert line in the activation of
ILP.

Fig. 6 IEEE 9-bus Test System

In Fig. 6, an illustration of the 9-bus IEEE test system is


displayed. The test system includes 3-load bus and 3-generator
bus.

4. Compensation Model

Since the IL market is proposed to function as a day-


ahead market, the IL market participants would have
information on the next 24 hours ahead forecast; transmission
capacity limits across the system and outage conditions on the
system. It can therefore be assumed that the price is unstable
in the next day. It can be expected that the competition in the
spot market will remain healthier the next day because of the
scheduled interruption which drives the price lower. However,
determining the exact function of how the system reacts to this
behavior would be very difficult to assume. For the sake of
simplicity, the price is modeled in such a way that it is not
affected by the previous hours and assumes that the system
does not have an unpredictable behavior.

While this computation is considered by the commission


as standard, due to its lack of appeal or any show of interest
among the participating customers, the compensation cost is
Fig. 5 MERALCO Interruptible Load Program
actually very low. This is why the researchers propose to
implement a factor that will encourage participating customers
to join. Using the calculation of the commission, the fixed
3. Test System Model.
bidding rate will be equivalent to the De-loading
Compensation over Compensable kWh as presented below.
In order to incorporate this study in a real-life scenario, the
ILP is integrated with a test-bus system. To model this study
De − loading Compensation
within a distribution system is vital because network 𝛽0𝑖 = ( )
constraints and power losses are considered as well as Compensable kWh 𝑖

5
The participants made their bidding interruptible load in Table I: Sample Interruption Schedule
kWh in the spot market, the final bidding price is sensitive to
the available reserved in the system, and the bidding function Time (hr.)
is defined as follows:
PC 1 2 3 4 5 6 7 h
𝑅
(1− ) 1 0 0 0 1 0 0 0 0
𝛽𝑖 = 𝛽0𝑖 𝑒 𝑅𝐸𝑆𝑄
2 0 0 0 0 1 0 1 0
Where 𝛽𝑖 is the compensation price of customer i. 𝛽0𝑖 is l 0 0 1 0 0 0 1 0
the initial price of interruption of customer i, R is the available
reserve in the entire system and R is equal to RESQ when R >
RESQ and 𝑅𝐸𝑆𝑄 is the minimum reserve required for the 6. System Simulation
entire system.
In the simulation, a day-a-head is executed and the load is
forecasted using the 9-bus test system and load flow curve
Start
together with the estimated generation charge using OPF on an
hourly basis. Each hour of OPF in the test system is executed
using MATLAB optimum power flow solver
MATPOWER6.0.
Gather Data
After the initial simulation of the forecast load, the
simulation proceeds to the ILP and simulate each hour with
Binary Particle Swarm Optimization to schedule the ILP’s
Compute initial price until termination criteria is achieved. Multiple runs were
simulated in the second phase during which each best results
were recorded for further analysis. In these runs, we iterated
50th time and multiple run optimizations of 250.

Compute hourly price D. Optimization

The pseudo-code of the proposed minimization of


electricity cost using Binary Particle Swarm Optimization
(BPSO) is shown in Fig. 8.
End
In the initialization phase of PSO algorithm, the dimension
of the swarm particles was based on the number of
Fig. 7 Price Calculation participating customers and number of hours in a day, each
particle represents a schedule of interruption. The initial
position of particles and velocity of particles where randomly
initialized. While initial fitness, particle best and global best
5. Day-Ahead schedule Model where set to infinity.

Based on a forecast load profile, ILP is executed as a Day 1. Scheduling of BPSO


Ahead Scheduling. Prior to the day of ILP execution, each IL
characteristics are gathered. It also includes the calculated The curtailment schedule is created using BPSO. Each
compensation price set by Energy Regulatory Board (ERC) particle is an 𝑛 by 𝑡 matrix where 𝑛 is the number of
and the number of affiliated IL. The day ahead scheduling is participants and 𝑡 is the time slot. 𝑥(𝑛, 𝑡) is equal to 1 is the
designated in an hourly basis with 24 hours in a day. Hence, 𝑛𝑡ℎ IL is curtailed during the 𝑡𝑡ℎ time interval; otherwise, it is
the schedule consists of 24 hour time slots. For simplicity, the
equal to 0. The speed of each particle 𝑖 in 𝑘𝑡ℎ iteration as
day-ahead schedule can be mapped in a ℎ by 𝑙 matrix. The ℎ
stands for the number of hour time slots while the 𝑙 stands for follows:
the number of Interruptible Loads Participants. A sample of
schedule is shown below from 1 to h hour and participating 𝒗𝒊 (𝒏, 𝒕)𝒌+𝟏 = 𝒗𝒊 (𝒏, 𝒕)𝒌 + 𝒄𝟏 𝒓𝟏 (𝒙(𝒑𝒃𝒆𝒔𝒕𝒊 )𝒌 − 𝒙𝒊 (𝒏, 𝒕)𝒌 )
customers of 1 to l, where one (1) means that it is schedule for
an interruption and a zero (0) means it is not schedule for +𝒄𝟐 𝒓𝟐 (𝒙(𝒈𝒃𝒆𝒔𝒕)𝒌 − 𝒙𝒊 (𝒏, 𝒕)𝒌 )
interruption.
It is then mapped to a probability by the sigmoid function
and discretized by comparing the certain number to a random

6
number between 0 to 1. The particles are then placed Generator Cost Function:
randomly in a solution space at the start of the simulation.
This is achieved by a random number generator with a 2
𝐶𝐺𝑡 = ∑ 𝐶𝑖 (𝑃𝑔𝑖 ) = ∑ 𝑎𝑖 𝑃𝑔𝑖 + 𝑏𝑖 𝑃𝑔𝑖 + 𝑐𝑖
uniform distribution probability.
𝑖 𝑖

Where:
Start
𝐶𝐺𝑡 : Generation Cost of time t
𝑖: Generator Unit
Initialize 𝐶𝑖 : Rate of Generation of Generator Unit i
𝑃𝑔𝑖 : Real Power Demand for Generator Unit i
𝑎𝑖 : Coefficients of an 1st order polynomial cost function
Generate first 𝑏𝑖 Coefficients of an 2nd order polynomial cost function
Swarm 𝑐𝑖 Coefficients of a last order polynomial cost function

Evaluate the Interruptible Load Cost Function:


fitness of all
ILCt = ∑ βPC (CIL ∗ xPC(t) )
PC
Record Update the
𝐼𝐿𝐶𝑡 : Interruptible Load Compensation
personal best position of 𝑃𝐶: Participating Customer
fitness of all particles 𝛽𝑃𝐶 : Compensation Cost of PC
𝐶𝐼𝐿 : Compensable kWh
Find the global 𝑥𝑃𝐶(𝑡) : Interruption Schedule at time t of PC
Update the
best particle velocity of 4. Penalty Function
particles
The penalty function will impose the overall penalty of
each particle schedule output which includes the peak,
underload and average rate.
Swarm met the
termination 𝑓2 (𝑥𝑖 ) = ∑(𝑏𝑝𝑖(𝑡) + 𝑙𝑝𝑖(𝑡) + 𝑟𝑖(𝑡) )
𝑡∈𝜀
𝜀 = { 𝑃𝑒𝑎𝑘 𝐻𝑜𝑢𝑟𝑠 }

𝐶𝑜 , 𝑙𝑜𝑎𝑑𝑖(𝑡) > 𝑏𝑎𝑠𝑒𝑝𝑒𝑎𝑘 ∗ 𝑐𝑚


End 𝑏𝑝𝑖(𝑡) = {
0, 𝑙𝑜𝑎𝑑𝑖(𝑡) ≤ 𝑏𝑎𝑠𝑒𝑝𝑒𝑎𝑘 ∗ 𝑐𝑚

𝐶𝑜 , 𝑙𝑜𝑎𝑑𝑖(𝑡) < 𝑙𝑜𝑤𝑝𝑒𝑎𝑘 ∗ 𝑐𝑚


Fig. 8 Particle Swarm Optimization Flow Chart 𝑙𝑝𝑖(𝑡) = {
0, 𝑙𝑜𝑎𝑑𝑖(𝑡) ≥ 𝑙𝑜𝑤𝑝𝑒𝑎𝑘 ∗ 𝑐𝑚
2. Repair Algorithm
𝐶𝑜 , 𝑟𝑎𝑡𝑒𝑖(𝑡) > 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡 𝑟𝑎𝑡𝑒𝑡
𝑟𝑖(𝑡) = {
To reduce the runtime of the program the optimization 0, 𝑟𝑎𝑡𝑒𝑖(𝑡) ≤ 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡 𝑟𝑎𝑡𝑒𝑡
will only run when it is a peak hour. To establish the timetable
for peak hours, the first run of the program is to locate the 𝐺𝐶𝑖(𝑡) + 𝐼𝐿𝐶𝑖(𝑡)
𝑟𝑎𝑡𝑒𝑖(𝑡) =
peak hours by setting the schedule interruption, 𝑥𝑖 (𝑛, 𝑡), to all 𝑙𝑜𝑎𝑑𝑖(𝑡)
zeros, which indicates that no interruption is scheduled. This
will make the program run faster, which called the Repair Where 𝑐𝑚 is the allowable error range, 𝐶𝑜 is the
Algorithm.
optimization penalty cost 𝑥𝑖 is the schedule matrix of particle
3. Cost Function i, 𝑏𝑝𝑖 is the high peak penalty 𝑙𝑝𝑖 is the under load level
penalty, 𝑟𝑖 is the rate penalty, 𝑟𝑎𝑡𝑒𝑡 forecasted rate at time t
Cost Function is the computation of Generation Charge and 𝑙𝑜𝑎𝑑𝑖(𝑡) is the total load at time t, in particle schedule i.
and IL Compensation for a time period.

7
5. Objective Function 7. Total Load

The Objective function is what the researchers intend to To calculate the load profile after the interruption schedule
minimize to obtain the optimized results. OPF is used, and for simplicity only real power will serve as
the area of concern.

𝑙𝑜𝑎𝑑𝑡 = ∑ 𝑙𝑜𝑎𝑑𝑘(𝑡) + (𝐶𝐼𝐿 ∗ 𝑥𝑃𝐶(𝑡) )𝑘


Start 𝑘

𝑃𝐿 = 𝑙𝑜𝑎𝑑𝑡 + 𝑙𝑜𝑠𝑠𝑡

𝑃𝐺 − 𝑃𝐿 = 0
Gather Data
where:

𝑙𝑜𝑎𝑑𝑡 : Total Load Demand at time t


𝑙𝑜𝑎𝑑𝑘(𝑡) : Load Demand at node k
t = 1:24
𝑙𝑜𝑎𝑑𝑡 : Total Real Power Demand
𝑙𝑜𝑠𝑠𝑡 : Total Real Power Loss
t=t+1
𝐶𝐼𝐿 : Compensable kWh
𝑥𝑃𝐶(𝑡) : Interruption Schedule at time t of PC
𝑃𝐿 : Total Real Power Load Demand
Peak NO 𝑃𝐺 : Total Real Power Consumption in the
Hour? Generators

YES 8. Total Operating Cost

𝑓1 (𝑥𝑖 ) = ∑(𝑙𝑜𝑎𝑑(𝑥𝑖 )𝑡 − 𝐵𝑎𝑠𝑒𝐿𝑜𝑎𝑑(𝑥𝑖 )𝑡 ) The total operating cost is the cost within a 24 hour
𝑡∈𝜀 interval of all the generating plant and interruptible load
participating customers’ compensation cost.
𝑓2 (𝑥𝑖 ) = ∑(𝑏𝑝𝑖(𝑡) + 𝑙𝑝𝑖(𝑡) + 𝑟𝑖(𝑡) )
𝑡∈𝜀
𝑂𝑝𝑒𝑟𝑎𝑡𝑖𝑛𝑔 𝐶𝑜𝑠𝑡 = ∑[𝐶𝐺𝑡 + 𝐼𝐿𝐶𝑡 ]
𝑓(𝑥𝑖 ) = 𝑓1 (𝑥𝑖 ) + 𝑓2 (𝑥𝑖 ) 𝑡

9. Average Rate
Return fitness 𝑓(𝑥𝑖 )
This is the sum of the total cost of generator procurement
and total ILP compensation per hour over the total load
connected in the system at the hour t as shown below.
End
𝐶𝐺𝑡 + 𝐼𝐿𝐶𝑡
𝑟𝑎𝑡𝑒(𝑡) =
Fig. 9 Fitness Function 𝑙𝑜𝑎𝑑𝑡

E. Assessment

1. Validity
6. Delta Function
The validity of the study is assessed through the
comparison between case scenarios. The case scenarios, -
The Delta function calculates the difference between the
namely the case where ILP is activated during peak hours and
base load profile and the ILP load profile on peak hours.
the case when purchase of peaking plants in the spot market is
employed during peak hours. The assessment includes the
𝑓1 (𝑥𝑖 ) = ∑(𝑙𝑜𝑎𝑑𝑡 − 𝐵𝑎𝑠𝑒𝐿𝑜𝑎𝑑𝑡 ) comparison between the reductions of the electricity rates and
𝑡∈𝜀 generation costs as well the peak reduction. Moreover, the
𝜀 = { 𝑃𝑒𝑎𝑘 𝐻𝑜𝑢𝑟𝑠 } validity is also assessed based on the yellow alert level of the

8
total reserves. Once the total generation approaches the yellow Table III: Interruptible Load Characteristics
state, the Interruptible Load curtailments become desirable.

2. Case Study

Case 1: Peaking Plant Procurement

The first scenario is a system simulation with the


procurement of Peaking Plants in mitigating peak demand
during peak hours.

Case 2: Implementation of ILP

The second case scenario is a system simulation with the


implementation of ILP in mitigating the Peak Demand. This
includes the optimal day-ahead interruption scheduling of
Interruptible Loads as well as the calculation of the
curtailment compensations.

VII. RESULTS AND DISCUSSIONS

The research has been simulated using MATLAB on a core


i7 machine with 16GB of RAM. During the duration of the
simulation, multiple runs of the program have been executed
in order to obtain the absolute best solution. In the latter part
of this chapter, the worst and best solution, as well as the
mean and standard deviation are tabulated. As mentioned in
the previous chapter, two case scenarios were focused on in
this research. Moreover, those case scenarios will determine
the validity of this research by comparison between the two.

Table II: Binary Particle Swarm Optimization Parameters


B. Test Case System With Interruptible Load Program

BPSO Parameters Value


Number of Iteration 50
Number of Swarm 25
Minimum Velocity 0
Maximum Velocity 8
Minimum Inertia 0.4
Maximum Inertia 0.9
Acceleration Coefficient 1 2.05
Acceleration Coefficient 2 2.05
A. Interruptible Load Characteristics
This study involved a total of 30 Interruptible Loads (IL).
These IL have characteristics that are based on hypothetical
data [16]. Among these characteristics includes curtailment
capacity, maximum curtailment period (Max OFF time),
minimum duration between curtailments (Min ON time), and
the bus number connections of each IL. While the hourly
initial compensation price is computed using the pricing
function in the previous chapter. The characteristics of ILs are Fig. 10 IEEE 9-bus Test System with Interruptible Loads
shown in Table III.

9
The simulation of ILP was executed through the 9-bus E. Optimization
IEEE test system. The goal is to incorporate this study in a real
life scenario where ILs’ are connected in the load buses. Table VI: Optimization Runs
During peak hours, ILP is activated instead of peaking plant
procurements. Therefore, ILs’ are disconnected from its load
bus. As a result, the demand during peak hours is reduced. Total Generation Cost (Mil
250 Runs
OPF is then applied to obtain the generation cost which is used Php)
for the cost function in the optimization. Fig. 10 displayed the Average 82.661
9-bus IEEE test system where IL is connected in its designated Standard Deviation 110.447
load bus. Best 82.411
Worst 82.968
C. ILP Activation Levels Simulation time (sec) 82,769

The simulation of ILP is executed in multiple runs. In


Table IV: ILP Activation levels Table VI, the average, standard deviation, best, and worst
values for the total generation was presented. The average and
standard deviation of total generation cost has closer values.
Status Range There is an 82.661 Million PHP Average total generation cost
Normal State P < 555 MW with a standard deviation of 110.447 Million PHP wherein
Yellow Alert P < 580 MW < P fluctuations occur in the final solution quality of total
generation cost, and this can be observed in Fig. 11. The worst
Red Alert P > 605 MW
and best total generation cost is within the range of the
standard deviation. This means that the optimization using
Binary Particle Swarm Optimization was able to achieve
D. ILP Interrption Schedule consistent results and thus, serves as a good solution for the
total generation cost across multiple runs.

Table V: Day-Ahead Interruption Schedules

Fig. 11 Total Generation Cost in Multiple Runs

F. Evaluation

a. Total Generation

The simulation of ILP aims to reduce the Peak Demand. In


Fig. 12, the results for the Total Generation in a 24-hour time
slot are displayed. It can be observed in the figure that the
Yellow Alert Level is around 580 MW while the Red Alert
The simulation of ILP aims to obtain the optimal day- Level is roughly 605 MW. The Red line in the figure
ahead interruption schedule for the IL. Through the process of represents the Red Alert level, it means the available reserves
optimization, the goal was to obtain acceptable interruptions. are zero. Touching or exceeding above the Red line indicates a
The optimal day-ahead interruption schedule mapped in a 24- deficiency in generation. As standard for several distribution
hour by 30-IL can be found in Table V, where 1’s indicates utility, Yellow Alert Level should be the basis for ILP
interruption and 0’s for non-interruption. activation. It can be observed in the figure that the
interruptible load curtailment (the case 2) approaches the
Yellow Line and avoids the Red Line. Meanwhile, case 1
touches and exceeds the Red line on specific hours. This
indicates the use of Peaking Plants from the spot market.

The total forecasted generation in case 1 is 12,967.83


MWh in which 155.37MWh is the procurement of peaking

10
plant assuming the 580 MW yellow alert level as indication b. Average Rate of Electricity
for peaking plant procurement.
The simulation of ILP aims to reduce the average
It can be observed that the forecasted generation in case 1 electricity rates. In page 58 Fig. 13, the results for the average
and case 2 were 12,967.83 and 12807.35 MWh respectively electricity rates in a within 24-hour time slot are displayed. It
which is a difference of 160.48 MWh where case 2 is less than is notable that the implementation of ILP yields reduction in
case 1, hence indicating a reduced generation procurement. average rate of electricity from PHP 6.45 in case 1 to PHP
The average and maximum generation procurement is also 6.43 in case 2. The numerical data for the average rate of
reduced in case 2 which indicates a reduced peak demand in electricity for case 1 and case 2 can be found in Table VIII.
the system.
Table VIII: Power Generation Rate
The numerical data for the total generation in case 1 and
case 2 can be found in Table VII.

Table VII: Hourly Power Generation Procurement

7.4
Case 1
7.2 Case 2
7
Rate (PHP/kWh)

6.8
6.6
6.4
6.2
6
5.8
5.6
5.4
1 3 5 7 9 11 13 15 17 19 21 23
Time (hr.)

Fig. 12 Hourly Total Power Generation for Case 1 and


Fig. 13 Hourly Average Rate of Electricity for Case 1 and
Case 2
Case 2

11
c. Hourly Generation Cost

The simulation of ILP aims reduce to hourly generation


costs. In Fig. 13, the results for the hourly generation costs in a
24-hour time slot are displayed where the blue bar indicating
case 1 and red bar as case 2. It can be observed in the Figure
stated that with the implementation of ILP, the hourly
generation costs have been reduced. The Yellow Bar
represents the total hourly interruptible load compensation
cost. The Green bar represents the reduced generation costs
due to ILP.

The data indicates the total operation costs of case 1 and


case 2 were PHP 84,109,287.59 and PHP 82,661,003.56,
respectively. Hence, there is PHP 1,448,284.03 savings in
operation cost when ILP is activated as an alternative of
peaking plant procurements which is a 1.72% decrease in the
overall operation cost. Fig. 13 Hourly Operation Cost with Generation and
Compensation Cost
Meanwhile, the generation cost in case 1 and case 2 were
PHP 84,109,287.59 and PHP 81,604,865.60 respectively.
Hence, there is PHP 2,504,421.99 savings in generation cost
which is 2.98% decrease in procurement cost. VIII. CONCLUSION

The numerical data for the generation cost for case 1 and The Interruptible Load Program (ILP) was shown to be
case 2 as well as the total compensation cost for ILP can be effective contingency reserve by Distribution Utility in dealing
found in Table IX. with the reduction of system peak demand. The results
obtained satisfy the goals of ILP. The ILP modeling within the
electricity market was achieved. It was shown that the ILP
helps to reduce the system peak demand and raise the level of
system reserve during peak hours and in cases of tight power
Table IX: Total Cost of the Program situations.

The case scenarios in this study were the following: 1.)


Employment of Peaking Plants from the spot market to reduce
peak load, and 2.) Peak load reduction by ILP. Based on the
discussions in the preceding chapter, the second case scenario
proved to be more effective than the first case.

Moreover, the simulated ILP was able to reduce the system


peak demand by 160.48 MWh. It also has reduced the total
generation cost by PHP 1,448,284.03 and average electricity
rates by 0.4%. The interruptible schedule was also able to
satisfy the constraints. Finally, the study was able to impose
the new pricing function with more enticing incentives than
the proposed ERC compensation.

The researchers have found limitations within the program


which they recommend to be improved. One of these is the
structure of the scheduling. A day ahead scheduling is not
flexible enough especially when load forecasting is used as the
basis for curtailments. An hour ahead scheduling correction to
the day ahead scheduling is recommended.

The researchers also recommend presenting this study to


potential interruptible load customers to elicit feedback and
reactions. With communication among potential Interruptible
Load customers, the planning of curtailments would be more
flexible by knowing their possible load shifting.

12
The researchers also recommend the have a continuation Electricity Market,” no. April, pp. 0–8, 2015.
of the study with data based on the Mindanao Gird System and [16] K.-Y. Huang, H.-C. Chin, and Y.-C. Huang, “A Model
Economy. Reference Adaptive Control Strategy for Interruptible
Load Management,” IEEE Trans. Power Syst., vol. 19,
IX. REFERENCES no. 1, pp. 683–689, 2004.
[1] Virgil Lopez, “Mindanao on red alert due to deficient
power reserves | Money | GMA News Online,” GMA
News, 2016. [Online]. Available:
http://www.gmanetwork.com/news/money/companies/
562691/mindanao-on-red-alert-due-to-deficient-
power-reserves/story/. [Accessed: 05-Dec-2017].
[2] DOE, “2016 Philippine Electricity Demand-Supply
Snapshot,” pp. 1–14, 2016.
[3] DOE, “Power supply and demand highlights (january-
june 2017),” no. June, 2017.
[4] DOE, “DOE Continues To Provide Solutions on
Mindanao Power Situation | DOE | Department of
Energy Portal,” DOE, 2014. [Online]. Available:
https://www.doe.gov.ph/doe-continues-provide-
solutions-mindanao-power-situation. [Accessed: 05-
Dec-2017].
[5] Meralco, “Interruptible load program,” 2015.
[6] NGCP, “NGCP urges its customers to volunteer for
ILP,” NGCP, 2016. [Online]. Available:
https://www.ngcp.ph/article-
view.asp?ContentID=9404. [Accessed: 06-Dec-2017].
[7] T. S. C. and K. P. W. C.W. Yu, S. Zhang, “Modelling
and evaluation of interruptible-load programmes in
electricity markets,” IEE Proceedings-Generation,
Transm. …, vol. 152, no. 5, pp. 201–212, 2005.
[8] Q. Qi, X. Li, and Y. Li, “Study on incentive
interruptible load contract with risk preference of
power companies,” 3rd Int. Conf. Deregul. Restruct.
Power Technol. DRPT 2008, no. April, pp. 587–591,
2008.
[9] C. W. Y. and K. P. W. T.S. Chung, S.H. Zhang and
Abstract:, “Electricity market risk management using
forward contracts with bilateral options,” IEE
Proceedings-Generation, Transm. …, vol. 150, no. 5,
pp. 201–212, 2003.
[10] M. Fahrioglu and F. L. Alvarado, “Designing
incentive compatible contracts for effective demand
management,” IEEE Trans. Power Syst., vol. 15, no.
4, pp. 1255–1260, 2000.
[11] J. Wang, X. Wang, and X. Ding, “The forward
contract model of interruptible load in power market,”
Proc. IEEE Power Eng. Soc. Transm. Distrib. Conf.,
vol. 2005, pp. 1–5, 2005.
[12] J. Kennedy and R. Eberhart, “Particle Swarm
Optimization,” pp. 1942–1948, 1995.
[13] J. Kennedy and R. C. Eberhart, “Dc 20212,” pp. 4–8,
1997.
[14] North American Electric Reliability Corporation
(NERC), “Data Collection for Demand-Side
Management for Quantifying its Influence on
Reliability,” no. December, pp. 1–57, 2007.
[15] AEMO, “Guide to Ancillary Services in the National

13

Вам также может понравиться