IRJET - Demand Response for Energy Management using Machine Learning and LSTM Neural Network

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 07 Issue: 03 | Mar 2020 www.irjet.net p-ISSN: 2395-0072
© 2020, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 3062
Demand Response for Energy Management Using Machine Learning
and LSTM Neural Network
Harshiyanas M
Department of Computer Science and Engineering &APJ Abdul Kalam Technological University-India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Prediction of power demand plays a necessary
role within the electrical trade, becauseitprovidesthepremise
for creating selections in installation designingand operation.
In a dynamic atmospherestandardpredictiontechniquearen't
enough, and a lot of refined ways area unit required. This
paper proposes an hour-ahead demand response prediction
for energy management systems. Advanced Reinforcement
learning and Artificial Neural Network for modelling and
understanding very complex systems will allow companies to
represent the complexity of electricity demand in a way
previously unachievable. With the help ofhour-ahead demand
response algorithm for energy management systems, a steady
price prediction model based on LSTM neural network is
presented. More specifically, the objective is to achieve more
accurate forecasts for the electricity demand. Aproperscheme
of energy management can be achieved for multiple units
minimize user energy bills and help the user to significantly
reduce its electricity cost comparedwithabenchmarkwithout
demand response.
Key Words: Machine learning, LSTM neural network,
demand response, energy management
1. INTRODUCTION
Changes in electric variables such as electricityprice, energy
consumption and its operations requires energy
management systems to take accurate real time actions
correctly and instantly[1].Demand response become
significant role in terms of promoting efficiency and
reliability of energy systems with the help of modern
advances in informationand communicationsystems(ICT's)
and smart metering infrastructure, as it affords the capacity
to balance electricity supply and demand incongruity by
regulating loads[2]. A well-designed DR scheme in an EMS
can have significant positive effects on society, like
improving human comfort levels, facilitating the
accommodation of renewable resources, reducing
worldwide energyexpenditure,and reducingrelianceonfuel
resources related to high carbon emissions [3],[4]. Several
energy management system (EMS) structures under an
hourly pricing DR were developed to determine the optimal
appliance scheduling based on a mixed-integer linear
programming (MILP) approach, with the aim of reducing
user costs and enhancing energy efficiency [5]-[7].
2. RELATED WORKS
For past few years, with the rapid evolution of artificial
intelligence (AI), muchattention has been devoted to the use
of AI for optimal decision-making. Through the studies
developed,considersreinforcementlearning(RL)foroptimal
decision making [8]. Some research has also been done on
adopting RL forsolving decision-making problems in energy
management. For example, in [9], [10], the authors applied
RL-based algorithms to energy trading games among
different strategic players with incomplete information,
enabling each player used the learning scheme to choose a
strategy to trade energy in an independent market, so as to
maximize the average revenue.
Recently, the authorsof[11],[12]presentedaprice-basedDR
scheme linking the service provider to its customers via RL
methodology, where the scheme was modelled as a Markov
decision process (MDP), then Q-learning was used to make
optimal decisions. However, despite these efforts, there still
exist two significant limitations. The studies in [13], [14]
proposes batch RL algorithmscheduleloadsunderdayahead
pricingschemes.In[15]-[17],RLalgorithmsusedtoobtainan
energy consumption plan for electronic vehicles (EV). First,
most studies focused on only one kind of appliance such as
thermostatically controlloador EV thatdidnotconsiderhow
the proposed learning algorithms would enable decision-
making when dealing with multiple types of appliances.
Second, all studies considered day-ahead energy
management, but the hour-ahead DR exhibits greater
potential for balancing power systems due to the dynamic
constraints associated with energy generation and the
uncertainty in prediction [18].
Going through the above-mentioned issues, this paper
proposes an hour ahead demand response algorithm for
multiple units of load in a grid. To solve issues regarding
uncertainty in futureprice, astablepriceforecastingmodelis
presented using Long Short-Term Memory (LSTM)Neural
Network. Over a past few years, price forecasting methods
become an important topic in electrical engineering and
several implementation methods attempted. And by using
LSTM the power consumption can be monitored and
advanced measures can be taken by the service providers
based on the predicted output. The LSTM approach is
comparativelyeasytoimplementandhavegoodperformance
than other techniques such as ARIMA model [19].
3. SYSTEM DESIGN
First of all wemust understandaboutanenergymanagement
system and its architecture. Here a block diagram shows a
structure of energy management systems across a grid. This

area contains several energy consumption units under
service provider.
Fig -1: Energy Management System Architecture
3.1 Vector Conversion
Vectors are a foundational element of linear algebra. Vectors
are used throughout the field of machine learning in the
description of algorithms and processes such as the target
variable (y) when training an algorithm. A vector is a tuple of
one or more values called scalars. Vectors are built from
components, which are ordinary numbers. The LSTM input
layer is specified by the “input shape” argument on the first
hidden layer of the network. The three dimensions of this
input are:
a) Samples. One sequence is one sample. A batch is
comprised of one or more samples.
b) Time Steps. One time step is one point of observation in
the sample.
c) Features. One feature is one observation at a time step.
This means that the input layer expects a 3D array of data
when fitting the model and when making predictions, even if
specific dimensions of the array contain a single value, e.g.
one sample or one feature. When defining the input layer of
your LSTM network, the network assumes you have 1 or
more samples and requires that you specify the number of
time steps and the number of features. You can do this by
specifying a tuple to the “input shape” argument..
3.2 Understanding LSTM Neural Network Vector
Conversion
Recurrent neural networks re unrolled programmatically
during training and prediction. An LSTM network is a
recurrent neural network that has LSTM cell blocks in place
of our standard neural network layers. These cells have
various components called the input gate, the forgetgateand
the output gate.
Fig -2: LSTM cell diagram
Notice first, on the left hand side, we have our new
word/sequence value xt being concatenated to the previous
output from the cell ht−1. The first step for this combined
input is forit to be squashedvia a tanh layer.Thesecondstep
is that this input is passed through an input gate. An input
gate is a layer of sigmoid activated nodes whose output is
multiplied by the squashed input. These input gate sigmoids
canact to “kill off” any elements oftheinputvectorthataren’t
required. A sigmoid functionoutputsvaluesbetween0and1,
so the weights connecting the input to these nodes can be
trained to output values close to zero to “switch off” certain
input values (or, conversely, outputs close to 1 to “pass
through” other values).
The next step in the flow of data through this cell is the
internal state or forget gateloop. LSTM cells have an internal
state variable st. This variable, lagged onetime step
i.e. st−1 is added to the input data to create an effective layer
of recurrence. This addition operation, instead of a
multiplication operation,helpstoreducetheriskofvanishing
gradients. However, this recurrence loop is controlled by a
forget gate – this works the same as the input gate, but
instead helps the network learn which statevariablesshould
be “remembered” or “forgotten”. Finally, we have an output
layer tanh squashing function, the output of which is
controlled by an output gate. This gate determines which
values are actually allowed as an output from the cell ht. The
mathematics of the LSTM cell looks like this: Input First, the
input is squashed between -1 and 1 using a tanh activation
function.
3.3 System Structure
The structure shows the load or energy consumption is
monitored by theserviceprovider.Theserviceprovidergives
the hour ahead electricity price to the energy management
system and the system can give the load control command
across the units. In designing, the model is started with a
dataset that used as input and by training and testing this
dataset, and an output is plotted as graph. Using this graph,
the predicted values can be analysed. The following figure
shows the bock diagram of the system.

Fig -3: System Structure
3.4 Predicted Result
By performing the complete machine learning purpose
through LSTM neural network, the output is the forecasted
price on the hour ahead manner. After training the data , we
test the prediction result in order to make sure about the
efficiency of the system. For which the data has been splitted
as 90% for training and 10% for testing. The result can be
plotted on a graph such that electricity load in y axis and
hours in x axis. This can be figured out as shown below.
Fig -4: Plotted Graph
4. PERFORMANCE EVALUATION
Performance of the price forecasting model is evaluated on
the basis of some data collected during the previous years.
Figure 6 shows a comparison of forecasted prices for last 7
days of January 2020, where the blue line represents the
actual price and red line denotes the forecasted price. From
the figure we can see that the trends in actual price is quite
same as forecasted price. Compared to the price forecasting
results in [20], [21], indicating that the LSTM model in this
paper make accurate and reasonable price forecasting
method.
Fig -5: Result analysis
5. CONCLUSION
In this paper we proposed a price-based demand response
for energy consumption in a day ahead energy system using
LSTM neural network.Theperformanceevaluationdone and
shows that actual and forecastedpricesaretendstobesame.
REFERENCES
[1] M. Muratori and G. Rizzoni, “Residential demand
response: Dynamic energy management and time-
varying electricity pricing,” IEEE TransactionsonPower
systems, vol. 31, no. 2, pp. 1108–1117, 2016.
[2] M. Yu, R. Lu, and S. H. Hong, “A real-time decision model
for industrial load managementina smartgrid,”Applied
energy, vol. 183, pp. 1488–1497, 2016.
[3] Y. M. Ding, S. H. Hong, and X. H. Li, “A demand response
energy management scheme for industrial facilities in
smart grid,”IEEETransactionsonIndustrial Informatics,
vol. 10, no. 4, pp. 2257–2269, 2014.
[4] Y.-C. Li and S. H. Hong, “Real-time demand bidding for
energy management in discrete manufacturing
facilities,” IEEE Transactions on Industrial Electronics,
vol. 64, no. 1, pp. 739–749, 2017.
[5] F. De Angelis, M. Boaro, D. Fuselli, S. Squartini, F. Piazza,
and Q. Wei, “Optimal home energy management under
dynamic electrical and thermal constraints,” IEEE
Transactions on Industrial Informatics, vol. 9, no. 3, pp.
1518–1527, 2013.
[6] N. G. Paterakis, O. Erdinc, A. G. akirtzis,andJ.P.Catalao,
“Optimal household appliances scheduling under day-
ahead pricing and load shaping demand response
strategies,” IEEE TransactionsonIndustrial Informatics,
vol. 11, no. 6, pp. 1509–1519, 2015.
[7] S. Pal and R. Kumar, “Electric vehicleschedulingstrategy
in residential demandresponseprograms withneighbor
connection,” IEEE Transactions on Industrial
Informatics, vol. 14, no. 3, pp. 980–988, 2018.
[8] M. Yu and S. H. Hong, “A real-time demand-response
algorithm for smart grids: A stackelberg game
approach,” IEEE Transactions on Smart Grid, vol. 7, no.
2, pp. 879–888, 2016.
[9] H. Wang, T. Huang, X. Liao, H. Abu-Rub, and G. Chen,
“Reinforcement learning in energy trading gameamong
smart microgrids,” IEEE Transactions on Industrial
Electronics, vol. 63, no. 8, pp. 5109–5119, 2016.
[10] H. Wang, T. Huang, X. Liao, H. Abu-Rub, and G. Chen,
“Reinforcement learning for constrained energytrading
games with incomplete information,” IEEE transactions
on cybernetics, vol. 47, no. 10, pp. 3404–3416,2017.
[11] R. Lu, S. H. Hong, X. Zhang, X. Ye, and W. S. Song, “A
perspective on reinforcement learning in price-based
demand response for smart grid,” in 2017 International
Conference on Computational Science and
Computational Intelligence(CSCI),pp.1822–1823,IEEE,
2017.
[12] R. Lu, S. H. Hong, and X. Zhang, “A dynamic pricing
demand response algorithm for smart grid:
Reinforcement learning approach,” Applied Energy,vol.
220, pp. 220–230, 2018.
[13] F. Ruelens,B. J.Claessens, S.Quaiyum, B. De Schutter, R.
Babuska, and R. Belmans, Reinforcement learning
applied to an electric water heater: From theory to
practice,” IEEE Transactions onSmartGrid,no.99,pp.1–
9, 2016.
[14] F. Ruelens, B. J. Claessens, S. Vandael, B. De Schutter, R.
Babuska, ˇ and R. elmans, “Residential demand

response of thermostatically controlled loads using
batch reinforcement learning,” IEEE Transactions on
Smart Grid, vol. 8, no. 5, pp. 2149–2159, 2017.
[15] S. Vandael, B. Claessens, D. Ernst, T. Holvoet, and G.
Deconinck, “Reinforcement learning of heuristicevfleet
charging in a day-ahead electricity market,” IEEE
Transactions on Smart Grid, vol. 6, no.4,pp.1795–1805,
2015.
[16] T. Liu, Y. Zou, D. Liu, and F. Sun, “Reinforcement
learning of adaptiveenergy managementwithtransition
probability for a hybrid electric tracked vehicle,” IEEE
Transactions on Industrial Electronics, vol. 62, no. 12,
pp. 7837–7846, 2015.
[17] A. Chis , J. Lunden, and . Koivunen, “Reinforcement
learning-based plug-in electric vehicle charging with
forecasted price,” IEEE Transactions on ehicular
Technology, vol. 66, no. 5, pp. 3674–3684, 2017.
[18] X. Huang, S. H. Hong, and Y. Li, “Hour-ahead price based
energy management scheme for industrial facilities,”
IEEE Transactions on Industrial Informatics, vol. 13, no.
6, pp. 2886–2898, 2017.
[19] N. Chaabane, “A hybrid arfima and neural network
model for electricity ˆprice prediction,” International
Journal of Electrical Power & Energy Systems, vol. 55,
pp. 187–194, 2014.
[20] I. P. Panapakidis and A. S. Dagoumas, “Day-ahead
electricity price forecasting via the application of
artificial neural network basedmodels,”AppliedEnergy,
vol. 172, pp. 132–151, 2016.
BIOGRAPHIES
Harshiyanas M
Department of Computer Science
and Engineering
APJ Abdul Kalam Technological
University-India

IRJET - Demand Response for Energy Management using Machine Learning and LSTM Neural Network

More Related Content

What's hot (20)

Similar to IRJET - Demand Response for Energy Management using Machine Learning and LSTM Neural Network (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET - Demand Response for Energy Management using Machine Learning and LSTM Neural Network