Smart Grid and Grid-Connected Systems

Part1: The Basic of Smart Grid and Grid-Connected

Systems
Chapter 6: Uncertainty in Power System

prepared by Dr. Adib Allahham , Newcastle University, Newcastle, UK, January 2020
 Smart Grid and Grid-Connected Systems
Part1: The Basic of Smart Grid and Grid-Connected
Systems

Chapter 6: Uncertainty in Power System

prepared by Dr. Adib Allahham , Newcastle University, Newcastle, UK, January 2020

Please cite as:

Adib Allahham (2019): Smart Grid and Grid-Connected Systems. Teaching materials as part of the EU-funded project “Advanced
Teaching and training on Smart grid & Grid Integration of Renewable Energy Systems (AT-SGIRES)”, Newcastle University, Newcastle,
UK.

"The European Commission support for the production of this publication does not constitute an endorsement of the contents which reflects the views only of the
authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein."
 Outline
 Sources of Uncertainty
 Load and generation forecasting methods
 Planning and operation under uncertainty
 Case study
 Sources of Uncertainty
In a power system, there are two types of uncertainty [1]:
1) Uncertainty related to modelling
The mathematical model is not a perfect representation of the
system

Difference between the mathematical model of the system and its

real behaviour

The values calculated by the model are different from the real values

This type of uncertainty includes also calculation and measurement errors

 Sources of Uncertainty
2) The uncertain behaviour of some power system components, such as:
 Output energy from generation units (particularly renewable
generators);
 Energy demand;
 Outage of network components;
 Prices of fuel and energy; and
 Weather conditions.
 Sources of Uncertainty
Summary of some aspects of uncertainties [2]
Some aspects of uncertainties
Uncertainty location Aspects
Generation Intermittent output of renewable energy sources (RESs)
Failure or outage of distributed generators (DGs)
Operation strategy of energy storage systems (ESSs)

Network Failure or outage of feeders

Active network reconfiguration
Load demand Growth rate of load demand
Demand response and management
Charging/discharging strategy of electric vehiclues (EVs)

Wholescale market Electicity price

Production costs of components/assets
 Sources of Uncertainty
 Dealing with uncertainty requires the load demand and generation
forecasting which usually uses historical data to produce a prediction of
future load/generation.

 Load forecasting is used to schedule generation units; consequently,

power system operation depends on accurate load forecasting.

 Generation forecasting, especially for renewable sources, is used to

predict the future contribution of these sources in supplying the load.
 Load and generation forecasting
 Load forecasting is the prediction of the electric load over a given future
time horizon.
 Examples:
 In the residential energy sector, the energy consumption of a home
depends on the presence of the resident of this home, and the time at
which he uses specific appliances, and the duration for which he uses
them.
 This example demonstrates that load is variable and its value is uncertain
and depends on many other variables/parameters
 Load and generation forecasting
Load forecasting is classified according to:
 the time horizon; and
 forecasting techniques.

According to the time horizon, the load forecasting can be divided into three
main categories:
 Short-term forecast: The time horizon varies from one hour to one
week;
 Medium-term forecast: The time horizon varies from one week to one
year;
 Long-term forecast: The time horizon is higher than one year.
 Load and generation forecasting
 Load forecasting techniques are classified as:
 Qualitative; and
 Quantitative.

 Qualitative methods do not use any mathematical models and they are
based on judgments, opinions, and personal experience.
 Qualitative methods include the Delphi method and curve fitting.

 Quantitative methods are based on mathematical models and include

decomposition methods, regression analysis, and exponential smoothing.
 Quantitative methods can be sub-divided into two further categories:
 Time-series; and
 Associative models.
 Load and generation forecasting
 Time-series models analyse the historical patterns within load data and
predict the future load based on these patterns.

 The associate or causal models suppose that the load is driven by other
external variables and forecast the load based on the modelled
relationships between load and these variables [3].
Load and generation forecasting
 Load and generation forecasting
1. Associate quantitative methods

A. Linear regression methods

 Linear regressions methods are used to represent the relationship
between the load and other parameters which could have a causal
relationship for example, time of day, type of consumer, and weather
conditions.

 The methods suppose that the load is composed of two parts:

 a standard load trend; and
 a component linearly dependent on one or more parameters affecting
the load.
 Load and generation forecasting
The general mathematical representation of regression methods is [4]:
𝑛
𝐿 ( 𝑡 )=𝐿𝑛 ( 𝑡 ) + ∑ 𝑎𝑖 𝑥 𝑖 ( 𝑡 ) +𝑒 (𝑡 )
𝑖=1
where:
: the standard load at time t;
: a coefficient;
: the parameters affecting the load;
: the error term, : the number of observations.
 Load and generation forecasting
 The accuracy of the regression method depends on the relationship
between the potential future operating conditions and historical data.

 For annual load forecasting, the following characteristics of the historical

load data must be considered:
 The load growth during a year;
 the load change between the different seasons; and
 The load changes occurring every day.
 Load and generation forecasting
B. Multiple regression methods

 These methods can be applied for load forecasting considering a set of

parameters such as weather conditions, per capita growth rate, and
electricity prices.
 One of the techniques used for multiple regression methods is the least-
square estimation technique [4], given by the following equation:
𝑦 𝑡 =𝑉 𝑡 𝑎𝑡 +𝑒 𝑡
where:
is the forecasted load at the sampling time t;
is the vector of affecting parameters to consider;
is the vector of regression coefficients;
is the error at sampling time t.
 Load and generation forecasting
2. Time-series quantitative methods

 The time-series is a sequence of time-ordered measurements/data

collected at a given time resolution (half-hour, an hour, etc.) over a given
duration (day, week, month, year, etc.).

 These forecasting methods assume that the future has the same
characteristics and qualities as the past.
Load and generation forecasting

Main characteristics of time-series

 Load and generation forecasting
 The trend indicates the long-term development/change in the data.
 This trend can be rising or descending.
 The trend might face irregular variations which do not reflect a typical
behaviour and have been caused by unusual conditions.
 The cycle indicates the wavelike change remaining for a period more than
one year.
 The seasonality indicates regular change that recurs every calendar year
(and/or every week or day).

 All these characteristics including trend, cycles, and seasonality might face
random variations that might have been caused by chance [5].
 Load and generation forecasting
A. Naïve method

 The naïve method assumes the forecast is the actual/real value of the last
period.
 This method is simple to use with no computational cost.
 It does not use any data analysis; hence its accuracy is low [5].
 The naïve method can be expressed as follows:
𝐹 ( 𝑡 )= 𝐴 ( 𝑡 −1 )

Where:
is the forecasted value at the time period and
is the observed value at the time period .
 Load and generation forecasting
B. Averaging methods

 The averaging method uses the average of past data as a forecast.

 This averaging category includes:

 Moving average;
 Weighted moving average; and
 Exponential smoothing.

 The moving average methods use the average of a number of the most
recent values of the observation/measurements.
 Load and generation forecasting
 This can be expressed in the following equation [5]:
𝑛
𝐹 ( 𝑡 )=∑ ¿ ¿ ¿
𝑖 =1

where:
is the forecasted value at the time period
the index corresponding to the period,
: the number of periods/data point during the considered
period, and
is the observed value at the time period .
 In weighted moving average methods, more weights are assigned to the
recent data points. The weights must sum to one.
 Load and generation forecasting
C. Exponential smoothing

 The forecast value during the next time step will be calculated as follows
[5]:

𝐹 𝑡 =𝐹 𝑡 −1 +𝛼 ( 𝐴𝑡 −1 − 𝐹 𝑡 −1 )

Where:
is the forecasted value at the time step ;
is the actual load at the time step ;
is the load forecast at the time step ;
: is the smoothing coefficient where .
 The amount of represents the error.
 Load and generation forecasting
3. Techniques for trends

 The trend can be linear or nonlinear.

 The techniques based on linear trends try to draw a straight line as close
as possible to the data points.

 This can be done by drawing a straight line with minimizing the squared
values of vertical deviations between the line and the data points.
 Load and generation forecasting
 The straight line equation can be given as follows [5]:.

𝐹 ( 𝑡 )=𝑎+𝑏𝑡
𝑛 ∑ ( 𝑡𝑦 ) − ∑ 𝑡 ∑ 𝑦
𝑏=
𝑛 ∑ 𝑡 2−( ∑ 𝑡 )
2

𝑎=
∑ 𝑦 −𝑏 ∑ 𝑡
𝑛

Where:

is the forecasted value at the period time

is the number of time periods after value of at ,

is the slope of the line, and is the value of time series.

 Load and generation forecasting
3. Techniques based on Seasonal indexes

 Seasonal variations are cyclic occurrences that are not driven by the
underlying trend or recent historical observations;

separate techniques are required to integrate this behavior into a forecast

model which causes the forecast to deviate from the trend or average
methods [5]

 There are mainly two techniques:

 Additive; and
 multiplicative.
 Load and generation forecasting
 In the additive technique, the seasonality index is expressed as a quantity,
then this quantity is added or subtracted from the average or the trend of
the series.

 In the multiplicative technique, the seasonality index is expressed as a

percentage of the average (or trend) amount. Then, this index is
multiplied by the overall average over a season to generate the forecast
for this season.
 Load and generation forecasting
4. Forecast error
 Forecast errors calculation helps to determine if the errors are within
reasonable bounds. If the error value is high, it will be necessary to take
corrective action.
 Forecast errors enable the decision-markers to act on full information
around both a prediction and its accuracy.
 The forecast error is the difference between the actual value and the
value that was predicated for a given period [5, and 6]:
𝑒𝑡 = 𝐴𝑡 − 𝐹 𝑡
Where:
is the forecast error at a given time
and are respectively the observed and forecasted values at the given time
 Load and generation forecasting
 Forecast errors can be used to evaluate the success or failure of the used
forecasting technique, and to choose between the different forecasting
techniques.
 The forecast error can be estimated using the following measures:

1) Mean Absolute Deviation (MAD)

∑ | 𝐴𝑡 − 𝐹 𝑡|
𝑀𝐴𝐷=
𝑛
2) Mean Squared Error (MSE)
𝑀𝑆𝐸=
∑ ( 𝐴𝑡 − 𝐹 𝑡 )2
𝑛− 1

3) Mean Absolute Percent Error (MAPE)

| 𝐴𝑡 − 𝐹 𝑡|
∑ 𝐴𝑡
𝑀𝐴𝑃𝐸= ∗100
𝑛
 Load and generation forecasting
5. Forecasting of renewable generation
 The output of renewable energy resources depends on the weather
conditions.
 Wind turbines, for example, have an intermittent output which is subject
to wind speed and direction, air temperature and humidity.
 The power output of solar photovoltaic (PV) is subject to the solar
radiation.
 Accurate forecasting of wind and solar resources usually depends on the
accuracy with which the driving weather conditions can be forecast.
 In [7], an exhaustive literature review is presented to show the application
of the aforementioned time-series techniques for short-term forecasting
of renewable resources.
 Planning and operation under uncertainty
 The integration of renewable energy sources (RES) such as PV and wind
turbines, which have intermittent and uncertain output, into the
distribution networks leads to challenges in planning and operation [8]

 The traditional deterministic planning methods, such as “fit and forget”,

ignore the different operating conditions of the network and the
uncertainties of the RESs

 These deterministic methods have drawbacks, including unnecessary

upgrade of the distribution network, and the increases in network losses

New planning methods are required to enhance the integration of RESs into
the distribution network
 Planning and operation under uncertainty
Key characteristics of the required planning methods:

 Optimal placement of RESs:

 The optimal allocation of RES is an important issue in the planning of
distribution networks.

 Optimal allocation of RES leads to many benefits for the network such
as the deferral of reinforcement, reduction of network losses,
increasing system reliability, and minimizing curtailment of the RES [9].

 The RES are placed in a way which satisfies the security constraints of
the network and meets the optimal planning of the networks [10].
 Planning and operation under uncertainty
 Combining planning and operation:
 Combination of planning and operation of distribution networks
enables the simultaneous optimization of both.
 Operational models can be integrated into the planning models to
avoid/defer network reinforcement.
 Technical operational challenges in the distribution networks arising
from the integration of RESs/DERs [11]:
 Voltage rise/drop
 Hosting capacity for dynamic active load demand
 Fault level
 Reactive power support
 Accommodation of RESs/DERs
 Planning and operation under uncertainty
 Considering uncertainty:
 The uncertainty aspects affect the planning models and their impacts
increase when multiple aspects are combined.

 Sizing and placement of energy storage systems (ESS):

 ESSs provide many services to the distribution network, including peak
shaving, deferring network reinforcement, improving supply reliability,
and reducing the fluctuation of RES.

 Placement and sizing of ESS has an important impact on the planning

of distribution networks.
 Planning model of distribution networks
 Typical mathematical formulation of planning model of distribution
network
𝑀𝑖𝑛 𝐹 ( 𝑥 𝑠𝑡 , 𝑦 𝑠𝑡 ) =[ 𝑂𝐹 1 , 𝑂𝐹 2 , …, 𝑂𝐹 𝑛 ]
𝐺 ( 𝑥 𝑠𝑡 , 𝑦 𝑠𝑡 ) =0
𝐻 ( 𝑥 𝑠𝑡 , 𝑦 𝑠𝑡 ) ≤ 0

 In this formulation, the are the decision variables

 Decision variables are related to the network and to the DERs
 Planning model of distribution networks
 Decision variables can be:
 placement (locations and sizes) of new substations,
 locations and sizes of new feeders,
 locations of reserve feeders,
 locations and sizes of DGs/RESs,
 locations and sizes of new energy storage systems (EESs), and
 locations and sizes of voltage control devices such as the capacitors
and static VAR compensator (SVC).
 Planning model of distribution networks
 denotes the set of objective functions
 Objective functions can be classified as technical, economic, and
environmental.
 Economic objectives can be
 Minimization of investment and operation costs,
 Minimization of energy losses, and
 maximization of net profit value.
 Planning model of distribution networks
 Technical objectives include
 Maximization of system reliability, and
 Improvement of voltage profile.

 Environmental objectives could be

 Minimization of carbon emissions, and
 Maximization of renewable energy yield.
 Planning model of distribution networks
 and are respectively the equality and inequality constraints

 Constraints can be technical and economic.

 Economic constraints introduce the financial limits for the network

operator to build and install new substations, feeders, and DGs.
 Planning model of distribution networks
 There are many technical constraints considered in the literature:
• Size limits of network assets such as substations, and feeders;
• Connectivity of the electric network;
• Power flow equations;
• Nodal real and reactive power balance equations;
• Limit constraints of the energy storage devices (ESS), distributed
generators (DG), ramp of DG, RES, and on-load tap changers (ONTC);
• Bus voltage limits;
• Constraints of demand-side response (DSR).
 Case study
 A new short-term electricity demand forecasting model, which has been
developed and implemented as part of the Smarter Network Storage
(SNS) project by Newcastle University, will be described [12, 13, 14].

 The model is based on the same linear- and auto-regressive methods

where the explanatory variables have been selected to cover a long
historical period.

 The resulting model has higher accuracy and deteriorates less significant
as the prediction window increases.
 Case study
 The model is based on the same linear- and auto-regressive methods

 The explanatory variables have been selected to cover a long historical

period.

 The resulting model has higher accuracy and deteriorates less significant
as the prediction window increases.
 Case study
 The model combines an autoregressive element with generic
explanatory variables, such as
 time of day
 day of week, month, and
 daylight
 The autoregressive terms were selected by establishing which lagged
values had a correlation coefficient of 0.75 or higher – this value was
selected to ensure a compromise between accuracy and computational
efficiency.
 Case study
 The forecasting method is based on multiple linear regression (MLR).
 An MLR model calculates the desired value based on a weighted sum of
explanatory variables:

𝑌 = 𝛽0 + 𝛽1 𝑋 1+ 𝛽 2 𝑋 2 +𝛽 3 𝑋 3 +…+ 𝛽𝑛 𝑋 𝑛 +𝑒

Where:
 Y is the desired value,
 represents the value of an explanatory variable,
 is the weighting given to the term within the MLR model
 e is an error term, representing the uncertainty in the model.
 Case study
 In short term demand forecasting, additional data are available, which
can be used to explain the variation in electricity demand.

 The additional data available are the demand in the last few hours, and
weather variables which will be available to the demand forecast
through weather forecasts.
 Case study

Lagged demand Values

 It is likely that the demand value at a time, t, will be related to the
demand values over the last few hours.
 In order to assess which values are likely to be significant, the
autocorrelation of the demand was calculated.
 Autocorrelation involves calculating the correlation of the demand data
with itself at various time lags.
 Case study
Autocorrelation of Electricity Demand

1.0
0.8
0.6
Correlation

0.4
0.2
0.0

0 10 20 30 40 50 60 70

Lag (Hours)

Autocorrelation of electricity demand for lags up to two weeks [14]

 Case study
 The autocorrelation of demand falls as the lag increases but then
increases again as it moves toward the same time in the previous day.
This is because demand is dominated by a diurnal cycle, meaning the
correlation varies with the same 24-hour periodicity.

 While correlation does not imply causation, it is likely that the demand
over the last few hours, along with the demand at the same time on the
previous day, will be useful explanatory variables for demand prediction.
 Case study

Ambient temperature
 There is evidence within power system analysis that electricity demand
is influenced by the weather, and particularly by ambient temperature
[14].

 Analysis was carried out to identify the relationship between the

demand and ambient temperature at Leighton Buzzard.
 Case study

Demand against temperature, on a month by month basis. January is represented by the bottom left
plot, followed by February-April along the bottom row, and December represented by the top right plot
 Case study
 The relationship between temperature and demand for each month of
the year shows that there is minimal correlation, even during the winter
months when, intuitively, low temperatures should correspond to high
demand.

 The ambient temperature data was available on an hourly basis, but the
SNS forecasting requires data at a 30-minutes resolution.

 Linear interpolation was used to fill in the half-hourly values.

 Case study

Dummy variables
 Dummy variables can be used to explain much of the variation. This is
because the variation is associated with concepts such as time of day,
day of week, month and whether there is daylight.

 Additionally, there is an interaction between at least some of these

variables.
 Case study

Daily load profile for each season

The peak demand being more pronounced in autumn and winter

References

[1] Ivey, M., Akhil, A., Robinson, D., Stamber, K. and Stamp, J., 1999. Accommodating uncertainty in planning and
operations. Transmission Reliability Program, Office of Power Technologies, US Department of Energy.

[2] Li, R., Wang, W., Chen, Z., Jiang, J. and Zhang, W., 2017. A review of optimal planning active distribution
system: Models, methods, and future researches. Energies, 10(11), p.1715.

[3] Bunn, D.W. and Farmer, E.D., 1985. Review of short-term forecasting methods in the electric power industry.
Comparative models for electrical load forecasting, pp.13-30.

[4] Singh, A.K., Khatoon, S., Muazzam, M. and Chaturvedi, D.K., 2012, December. Load forecasting techniques
and methodologies: A review. In 2012 2nd International Conference on Power, Control and Embedded Systems
(pp. 1-10). IEEE.

[5] Stevenson, W.J., Hojati, M. and Cao, J., 2007. Operations management (Vol. 8). Boston: McGraw-Hill/Irwin.
References

[6] Adhikari, N.C.D., Domakonda, N., Chandan, C., Gupta, G., Garg, R., Teja, S., Das, L. and Misra, A., 2019. An
Intelligent Approach to Demand Forecasting. In International Conference on Computer Networks and
Communication Technologies (pp. 167-183). Springer, Singapore

[7] Ghofrani, M. and Alolayan, M., 2017. Time Series and Renewable Energy Forecasting. In Time Series Analysis
and Applications. IntechOpen.

[8] Omu, A., Choudhary, R. and Boies, A., 2013. Distributed energy resource system optimisation using mixed
integer linear programming. Energy Policy, 61, pp.249-266.

[9] Aghaei, J., Muttaqi, K.M., Azizivahed, A. and Gitizadeh, M., 2014. Distribution expansion planning considering
reliability and security of energy using modified PSO (Particle Swarm Optimization) algorithm. Energy, 65,
pp.398-411.

[10] Abapour, S., Zare, K. and Mohammadi-Ivatloo, B., 2015. Dynamic planning of distributed generation units in
active distribution network. IET Generation, Transmission & Distribution, 9(12), pp.1455-1463.
References

[11] Li, R., Wang, W., Chen, Z., Jiang, J. and Zhang, W., 2017. A review of optimal planning active distribution
system: Models, methods, and future researches. Energies, 10(11), p.1715.

[12] Greenwood, D., Long Term Electricity Demand Forecast for Energy Storage Scheduling – Smarter Network
Storage. 2014, Newcastle University.

[13] Hong, T., Gui, M., Baran, M.E. and Willis, H.L., 2010, July. Modeling and forecasting hourly electric load by
multiple linear regression with interactions. In IEEE PES General Meeting (pp. 1-8). IEEE.

[14] Greenwood, D., Short-Term Electricity Demand Forecasting – Smarter Network Storage. 2014, Newcastle
University
 Thank you very much
for your attention!

Contribution by Newcastle University to the EU-funded project “Advanced Teaching and training on Smart
grid & Grid Integration of Renewable Energy Systems (AT-SGIRES)”

prepared by Dr. Adib Allahham , Newcastle University, Newcastle, UK, January 2020

Please cite as:

Adib Allahham (2019): Smart Grid and Grid-Connected Systems. Teaching materials as part of the EU-funded project “Advanced
Teaching and training on Smart grid & Grid Integration of Renewable Energy Systems (AT-SGIRES)”, Newcastle University, Newcastle,
UK

"The European Commission support for the production of this publication does not constitute an endorsement of the contents which reflects the views only of the
authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein."

57
 AT-SGIRES
Contact Information, Website and Social
Media Sites

http://meu.edu.jo/sites/atsgires [email protected]

[email protected] facebook.com/ATSGIRES

[email protected] [email protected]

AT-SGIRES Project Group @ATSGIRES

/company/atsgires