Conference PaperPDF Available

Classification and Regression Based Methods for Short Term Load and Price Forecasting: A Survey

Authors:

Abstract

Due to increase in electronic appliances, electricity is becoming basic necessity of life. Consumption of electricity depends on various factors like temperature , wind, humidity, weekend, working days and season. In electricity load forecasting, many researchers perform data analysis on electricity data provided by utilities to extract meaningful information. Smart Grid (SG) is power supply network which allows consumers to monitor their energy usage. It integrates different components of electricity like variety of operations, smart appliances, data collected from smart meters and efficient energy sources. To reduce the consumption of electricity, accurate prediction is compulsory. A good forecasting model makes an acceptable use of all characteristics of electric loads based data and also reduces dimensionality of that data. Various machine learning techniques are proposed for load and price forecasting in literature. In this research, we present a survey based on different short term electricity forecasting techniques for price and load. We broadly categorize different types of techniques into traditional machine learning and deep learning techniques.
Classification and Regression Based Methods for
Short Term Load and Price Forecasting: A
Survey
Hira Gul1, Arooj Arif1, Sahiba Fareed1, Mubbashra Anwar1, Afrah Naeem1,
Nadeem Javaid1,*
Abstract Due to increase in electronic appliances, electricity is becoming basic
necessity of life. Consumption of electricity depends on various factors like tem-
perature, wind, humidity, weekend, working days and season. In electricity load
forecasting, many researchers perform data analysis on electricity data provided by
utilities to extract meaningful information. Smart Grid (SG) is power supply net-
work which allows consumers to monitor their energy usage. It integrates differ-
ent components of electricity like variety of operations, smart appliances, data col-
lected from smart meters and efficient energy sources. To reduce the consumption
of electricity, accurate prediction is compulsory. A good forecasting model makes
an acceptable use of all characteristics of electric loads based data and also reduces
dimensionality of that data. Various machine learning techniques are proposed for
load and price forecasting in literature. In this research, we present a survey based on
different short term electricity forecasting techniques for price and load. We broadly
categorize different types of techniques into traditional machine learning and deep
learning techniques.
Key words: Load Forecasting, Support Vector Machine, Smart Grid, Price Fore-
casting, Deep Learning, Neural Network
1 Introduction
In this era of technology, energy consumption is expanding exponentially. Demand
of electricity is increasing day by day as number of electronic appliances are in-
Hira Gul
COMSATS University Islamabad, e-mail: hiragul001@gmail.com
Corresponding author
Nadeem Javaid, COMSATS University Islamabad e-mail: nadeemjavaidqau@gmail.com
1
2 Hira et al.
DATA PREPROCESSING
FEATURE SELECTION
FEATURE EXTRACTION
PREDICTION MODEL
05
04
03
02
06
HISTORICAL DATA
EXPERIMENT RESULTS
Fig. 1: General Forecasting Model
creasing. Consumption of electricity is divided into six areas: residential, industrial,
commercial, agricultural, transportation and other government. Residential area is
consuming 65% of total electricity [1]. Traditional grid is used as electromechani-
cal technology, which faces several difficulties like one-way communication, central
distribution, manual monitoring, few sensors and manual restoration. Smart Grid
(SG) is established to resolve the above-mentioned issues. It plays a leading role
in balancing the consumption, generation and transmission of electricity [2]. It is
power grid which monitors generation, transmission and consumption of electric-
ity [3]. It helps producers to produce required electricity according to consumers
need. So, forecasting of electricity is necessary for utilities to balance between de-
mand and supply of electric load [4].
Demand of electricity is dependent on numerous factors like humidity, wind,
temperature, season, weekend and weekdays. It also depends on number of house-
holders and their daily routine. Data analytics is procedure of assessing data by us-
ing statistical software to obtain useful information. In electricity load forecasting,
many researchers do data analysis on electricity data to extract meaningful infor-
mation. Many machine learning techniques are proposed for load forecasting. Large
amount of data is required for prediction. This data contains sensitive information
with thousands of entries. Accurate load forecasting reduces the electricity price as
it minimizes the consumption in peak hours [5].
Based on time interval, electricity load forecasting is split into three parts. First
is Short Term Load Forecasting (STLF), it forecasts the electricity from a time pe-
riod of 24 hours to one week. Second is Medium Term Load Forecasting (MTLF),
it forecasts load from a period of one week to one year. Third is Long Term Load
Forecasting (LTLF), it forecasts the load from a time period of one year to more than
two years [6]. Many load forecasting approaches developed in literature like Support
Classification and regression based methods for STLPF 3
Vector Machine (SVM), linear regression and deep neural network. In deep learn-
ing, Convolutional Neural Network (CNN) is most commonly used method for load
forecasting. CNN consists of input layer, multiple hidden layers and output layer.
Hidden layer contains multiple layers like convolutional layers and pooling layers.
Any pooling layer can be used among max, average or sum pooling. [7]. Forecast-
ing models generally consist of six steps including historical data, preprocessing,
feature selection, feature extraction, prediction model and experimental results as
shown in figure 1.
2 Related Work
This section describes the related work of electricity forecasting for short term load
forecasting.
2.1 Problem Addressed
In this section, we describe the problem statement of related papers.
2.1.1 Machine Learning versus Deep Learning
Machine learning techniques are used for electric load forecasting. These techniques
are Random Forest (RF) and Gradient Boosting (GB). Although, these techniques
have some drawbacks. Computational complexity of RF is very high, as it takes
very long time to build decision tree. Error rates are very high in RF [2]. In [3], au-
thors discussed the problems of data redundancy in feature selection and extraction.
Existing models used only univariate data for price forecasting that is not sufficient
for accurate forecasting. There is need of big data for electricity price forecasting
yet computational complexity of such data is very high. Over fitting problem occurs
in decision tree which means that a model performs well on training data and its
performance degrades on testing data.
In existing studies, features directly fed into model without any preprocessing.
According to authors, parameter tunning is considered to be important for accu-
rate forecasting. So, there is need of feature selection and extraction technique for
load forecasting [4]. In [5], previous Recurrent Neural Network (RNN) models do
not use future hidden state vector and available past information. If a state vector
is generated incorrectly at any specific time then cannot be corrected at that spe-
cific time, as it is very important to enhanced forecasting at specific time. In [6],
many authors used optimization techniques still there is need to optimize the hybrid
models. Authors consider the problem of back propagation in hybrid approach com-
posed of DWT, EMD and RVFL. Back propagation is an algorithm which consists
4 Hira et al.
of two main problems: slow convergence and trapped in local minimum. In [8],
authors discussed drawbacks of ANN in DCANN. ANN has some disadvantages
like slow convergence, less generalizing ability, trapped in local minimum and poor
initialization of parameters.
2.1.2 Traditional Forecasting Methods
In [9], WD is only used for pre-processing and reconstruction results based on in-
dependent prediction. Such combination is not sufficient for Support Vector Re-
gression (SVR). Moreover, the purpose of Multi-resolution Wavelet Decomposi-
tion (MWD) is need to be considered with SVR. In existing models, traditional ap-
proaches were used for load forecasting. These techniques are: time series and linear
regression. They mainly focus on aggregated electric load demand pattern. Now au-
thors are combining machine learning techniques with deep learning approaches.
So, there is need to consider the integrated deep learning techniques with combina-
tion of machine learning [10]. Short term day ahead load forecasting is very chal-
lenging job, because it depends on external environment factors like temperature,
wind and humidity. Existing day ahead load forecasting models improve the accu-
racy by compensating the computational cost. In short computational complexity of
hybrid model is a big challenge [11]. In [12], authors addressed the issue of data
redundancy among features in minimal Redundancy Maximal Relevance (mRMR).
They also discussed that influential factors like weather changes, day type and eco-
nomical condition affects the load forecasting. Authors ensemble different models to
address above mention problems. In [13], Linear regression and ARIMA gives best
results with linear problems while these models perform unsatisfactory with non-
linear time series data. Therefore, authors proposed a hybrid model which works
better for non-linear time series data. Time-Varying parameter Regression (TVR)
has a problem of price instability, fluctuating input variables, managing availability
of data and complex data inputs [14]. In [15], authors discussed the limitations of
ANN like over-fitting problem, trapped in local minimum and poor initialization of
parameters. To solve these issues, authors used combination of different models for
short term load forecasting. In [16], authors discussed that load forecasting is af-
fected by unstable factors like temperature, price, policy management and holidays
data. Due to fluctuation in these factors, noisy data is generated which gives inaccu-
rate results. Authors used EMD to decompose original data into IMFs and forecast
the electric load. In [17], authors discuss the issue of cyber security in load forecast-
ing. Hackers injected false information in original dataset. Data integrity problems
are relatively new in this domain.
2.2 Solution Proposed
This section describes the proposed solution of related papers.
Classification and regression based methods for STLPF 5
2.2.1 Machine Learning versus Deep Learning
In [2], authors proposed Grey Correlation Analysis (GCA) for feature selection
to remove the repetition of features. They combined Kernel Function with Prin-
ciple Component Analysis (KPCA). They used Differential Evolution (DE) which
is based on Support Vector Machine (SVM) classifier for the classification of price
forecasting. In [3], authors proposed probability density forecasting model for short
term power load forecasting. Authors proposed Multi-Layer Perceptron (MLP)
function with deep neural network. Kernel density estimation method consist of
three sub modules deep learning, loss function and quantile regression.
In [4], authors proposed a deep learning model called Stack Denoising Auto en-
coders (SDAs) for feature extraction. This model train SVR to forecast the electric
load of day ahead. By using heterogeneous deep model, accuracy of forecast is
better with less errors. In [5], authors proposed RICNN which is combination of
RNN and 1- Dimensional Convolutional Neural Network (1-D CNN). 1-D CNN
inception module is used to adjust the prediction time and hidden state vector val-
ues. They calculate the hidden state vector from adjacent time steps. As a conse-
quence, RNN generates optimized and robust network through prediction time and
hidden state vector. In [6], authors proposed hybrid incremental learning method.
In this method three techniques are combined: Discrete Wavelet Transform (DWT),
Empirical Mode Decomposition (EMD) and Random Vector Functional Link net-
work (RVFL). RVFL is very useful as it generates weights randomly among input
layer and hidden layer. It also provides nearest possible solution for parameters cal-
culation. In incremental learning, when they add DWT and EMD it impressively
increases accuracy and effectiveness of short term load forecasting. In [8], authors
proposed a model named Dynamic choice artificial neural network (DCANN) which
selects input dynamically for ANN. DCANN is hybrid model which consists of su-
pervised and unsupervised learning problems. This approach selects unsupervised
learning technique to choose input variable for individual output. Authors train cor-
respondence inputs and outputs through supervised learning.
2.2.2 Traditional Forecasting Methods
In [9], authors proposed hybrid incremental learning method. In this method, three
techniques are combined: DWT, EMD and Random Vector Functional Link net-
work (RVFL). RVFL is very useful as it generate weights randomly among input
layer and hidden layer. It also provide nearest possible solution for parameters cal-
culation. In incremental learning, when authors add DWT and EMD it impressively
increases accuracy and effectiveness of short term load forecasting. In [10], uthors
proposed an algorithm which is based on deep neural network for STLF. In this
algorithm there are input layer and output layer. Input layers denote the past data
while output layer denote the future energy load. There is a deep energy which has
two main processes: feature extraction and forecasting. In feature extraction, there
are three more layers of convolutional layers and three pooling layers. In [11], au-
6 Hira et al.
thors proposed a hybrid ANN day ahead based load forecasting model for smart
grid. Proposed model consists of three components: pre-processing module, fore-
cast module and optimization module. In preprocessing module, irrelevant variable
and data redundancy are removed from input sample. In forecast module, ANN is
used with sigmoid function and multivariate auto regression algorithm.
In optimization module, heuristic problem solving approach is used to reduce
error. For this purpose they used enhanced differential evolution algorithm. In [12],
authors proposed hybrid model EMD-mRMR-FOA-GRNN which is combination of
Empirical Mode Decomposition (EMD), minimal Redundancy Maximal Relevance
(mRMR), fruit Fly Optimization Algorithm (FOA) and General Regression Neural
Network (GRNN). Firstly, they divide load series data into Intrinsic Mode Func-
tions (IMFs), secondly, mRMR is applied to select the best features. Finally, FOA
is used to enhance the factors in GRNN. In [13], authors proposed AS-GCLSSVM
hybrid model for short term load forecasting. AS-GCLSSVM stands for Autocorre-
lation feature selection and Least Squares Support Vector Machine wolf algorithm
and cross validation. In [14], authors proposed Hybrid Iterative Reactive Adaptive
(HIRA) model. This model consists of two steps. In first step, they identified only
those parameters which affect the electricity prices. In second step, selected vari-
ables are used in price forecasting by applying hybrid approach. This HIRA model
is combination of statistical models and neural network tools. In [15], authors pro-
posed Wavenet ensemble model for STLF. In this model, authors combined different
components like input parameters, mean, median , mode, cross-validation, selection
and algorithms. Authors predicted hourly load forecasting through one step ahead
strategy. In [16], authors proposed a short term forecasting model named as EMD
Mixed ELM. In this model, authors combined EMD and Extreme Learning Machine
(ELM). They used EMD for denoising and normalization of complicated features
from a large dataset. The performance of ELM depends upon type of kernel they
used. Authors used mixed kernel for ELM. Mixed kernel is combination of UKF
kernel and RBF kernel. In [17], authors proposed framework for load forecasting
named as systematic data integrity simulation.
2.3 Future Challenges/ Open Research Issues
This section discusses the limitation and future challenges of literature.
2.3.1 Machine Learning versus Deep Learning
In [2], authors used RF for decision tree. Computational complexity of RF is very
high, as it takes very long time to build decision tree. DE is used for tunning pa-
rameters, however it is prone to slow convergence in local optima. In future, authors
will apply real time requirements in the proposed model. In [3], authors used RNN
model which do not use future hidden state vector and available past information. If
Classification and regression based methods for STLPF 7
a state vector is generated incorrectly at any specific time then cannot be corrected,
as it is very important for enhanced forecasting in prediction time. In [4], authors
did not specify the condition on which basis they extract the features. They did not
consider many important parameters like temperature, wind and humidity. In [5],
authors used EMD and it has mode mixing problem. ELM is feedforward neural
network which is most widely used, although drawback of ELM is it does not up-
date weights and parameters . In future, authors will combine proposed model with
ANN and SVR and applied on real time applications to evaluate the performance [6].
In [8], main limitation of the paper is computational power which is very high. The
classifier consumes more resources than existing models. Authors did not validate
that generated inputs are corrected or not.
2.3.2 Traditional Forecasting Methods
In [9], there is need to tune the parameters for better results. Also, different selec-
tion and extraction techniques can be used for different buildings in order to get
more better results. For seasonal data prediction, there is need to provide more sam-
ple data that is required to increase the consistency of training data. In future, this
model can be applied on different regions to attain accurate electricity forecasting of
that regions. In [10], authors concluded that due to complex neuron structure in neu-
ral network the computational power is very high as compared to existing models.
Three layers of pool makes the model more complex. Over fitting problem arises
which affects the training data. In [11],an enhanced signal processing technique for
features selection and extraction and some optimization technique needs to be con-
sidered for scheduling based application. In [12], prediction performance reduced
due to poor generalization capability of GRNN. Computational complexity of pro-
posed model is also high. In [13], authors consider AutoCorrelation Function (ACF)
relationship between two parameter. However, additional external parameters like
temperature, weather, holidays and festival related parameters need to used. In [14],
computational time is high of proposed model. It is because four different models
are combined together. This model performs well on big data but worst on small
datasets. In [15], authors compared ensemble learning algorithms with base learner
classes like deep neural network or ELM. However, they did not validate the af-
fect of feature selection method. ELM as being a feedforward neural network have
prominent role in SG operations besides this, It needs to be intensified [16].
8 Hira et al.
Table 1: Related work.
Problem Identified Proposed Solu-
tion
Results Limitations
In [2], computational
complexity of RF is very
high, as it takes very
long time to build deci-
sion tree
GCA for feature
selection is proposed
to remove the rep-
etition of features
and they combined
KPCA.
Proposed model gives
98% accuracy and shows
more robustness as com-
pared to NB and DT
DE is used for
tunning parameters,
however, it shows
convergence at local
optima
In [3], over fitting prob-
lem occurs in decision
tree, which means deci-
sion tree performs good
in training but not in pre-
diction
Probability density
forecasting model is
proposed
The error rate of proposed model is less
than RF and GB
RNN models do
not used future
hidden state vector
and available past
information.
In [4], features directly
move into model with-
out any preprocessing,
for accurate results tun-
ning parameter is neces-
sary
SDAs model is pro-
posed by authors
Error rate of proposed
model is less than plain
SVR and ANN
Authors did not ex-
plain the rules on
which basis they per-
form features selec-
tion and extraction
In [5], previous Re-
current Neural Network
(RNN) models do not
used future hidden state
vector and available past
information
Authors proposed
RICNN
Value of MAPE is 4.779
while the values of
MAPE in benchmark
techniques are 8.084,
7.371 and 5.634. These
values show that pro-
posed model is more
accurate than existing
models
RNN consumes
more time for
training the dataset
and deduction of the
results as compared
to MLP and CNN
In [6], the problem
of back propagation in
hybrid approach com-
posed of DWT, EMD and
RVFL
Authors proposed
hybrid incremental
learning method.
In this method
three techniques are
combined: DWT,
EMD and RVFL
Value of RMSE is
218.329 while the values
of MAPE in benchmark
techniques are 355.503
for GLMLF, 307.892
for SHLFN, 278.511
for RF and 244.820
for EMD-RVFL. These
values show that pro-
posed model is more
accurate and efficient
than existing models
This model will de-
composed with vari-
ous other model like
deep learning, sup-
port vector regres-
sion and kernel ridge
regression
In [8], drawbacks of
ANN in DCANN. ANN
has some disadvantages
like slow convergence,
Less generalizing perfor-
mance, trapped at local
minimum and poor ini-
tialization of parameters
DCANN is proposed
which selects input
dynamically for
ANN
Results of proposed
model with dynamic
selection are 10.71%
and 8.39%. These values
show that proposed
model is more accurate
and than existing models
Main limitation of
the paper is compu-
tational power which
is so high
Classification and regression based methods for STLPF 9
In [9], WD only used
for pre-processing and
reconstruction results
based on independent
prediction. Such combi-
nation are not sufficient
for SVR
Authors proposed
hybrid incremental
learning method.
In this method,
three techniques are
combined: DWT,
EMD and RVFL
Authors compared pure
SVR and hybrid SVR
and experiment shows
that addition of MWD in-
creases the accuracy of
forecasting
For seasonal data
prediction, there
should be more
sample data required
to increase the con-
sistency of training
data.
In [10], need to con-
sider the integrated deep
learning techniques with
combination of machine
learning
A algorithm which is
based on deep eural
network is proposed
for STLF
The results of previous
models are 9% and
11% while, Results of
proposed model are
9.77% for MAPE and
11.66% for RMSE. Au-
thors claimed that they
achieved high accuracy
Due to too much
neural network
the computational
power is very high as
compared to existing
models
In [11], existing day
ahead load forecasting
model improves the ac-
curacy by compensating
the computational cost.
Computational complex-
ity of hybrid model is a
big problem
Authors proposed
a hybrid ANN day
ahead based load
forecasting model
for smart grid
Value of MAPE is
1.23 while values of
existing models are
3.18 and 2.31
FS technique is not
performs satisfac-
tory, it needs more
refinement
In [12], authors ad-
dressed the issue of data
redundancy among fea-
tures in mRMR
Authors proposed
hybrid model EMD-
mRMRFOA-GRNN
Performance of EMD-
mRMR-FOA-GRNN
model is better than
existing models
Prediction perfor-
mance reduced due
to poor generaliza-
tion capability of
GRNN
In [13], Linear Regres-
sion and ARIMA gives
best results with lin-
ear problems while these
models perform unsatis-
factory with non-linear
time series data
Authors proposed
AS-GCLSSVM
hybrid model
Values of MAPE, MAE
and B2are 0.5596,
32.2088 and 0.9952.
Based on results authors
concluded that proposed
model is better than
existing model
Computational com-
plexity of proposed
model is high be-
cause of GWO algo-
rithms
In [14], over-fitting
problem, trapped in
local minimum and
poor initialization of
parameters
Wavenet ensemble
model for STLF
Performance of WNN
ensemble model is better
than existing models
Authors will com-
pare ensemble
learning algorithms
with base learner
class like deep
neural network or
ELM
In [15], the limitation
of ANN like overfitting
problem, trapped in local
minimum and poor ini-
tialization of parameters
Proposed Wavenet
ensemble model for
STLF
Performance of WNN
ensemble model is better
than existing models
Authors will com-
pare ensemble
learning algorithms
with base learner
class like deep
neural network or
ELM
10 Hira et al.
In [16], EMD is used
to denoise original sig-
nals. However, it has
mode mixing problem
and ELM cannot update
the weights and biases
Authors combined
EMD and ELM
Values of MAE, RMSE,
MAPE and TIC are
7.3550, 9.5823, 08093
and 0.0052 perspec-
tively, Performance of
EMD-mRMR-FOA-
GRNN model is better
than existing models
Prediction perfor-
mance reduced due
to poor generaliza-
tion capability of
GRNN
In [17], Linear Regres-
sion and ARIMA gives
best results with lin-
ear problems while these
models perform unsatis-
factory with non-linear
time series data
Authors proposed
AS-GCLSSVM
hybrid model for
short term load
forecasting
Values of MAPE, MAE
and B2are 0.5596,
32.2088 and 0.9952,
Based on results authors
concluded that proposed
model is better than
existing model
Computational com-
plexity of proposed
model is high be-
cause of GWO algo-
rithms
3 Conclusion
Demand of electricity is increasing exponentially as number of electronic appli-
ances are increasing day by day. Different forecasting models were proposed in last
decade. In this paper, we conduct a survey based on load and price forecasting. We
discussed traditional machine learning approaches and deep learning approaches.
In this research, we compare the performance of different load and price models
to observe the best results. The research has been moving towards new and more
efficient techniques and replacing old approaches. There is a clear move towards
hybrid techniques techniques. Hybrid techniques are efficient, better computational
complexity and more flexible. This research brings open challenges for future.
Classification and regression based methods for STLPF 11
References
1. Masip-Bruin, X. Marin-Tordera, E. Jukan, A.& Ren, G.J. (2018). “Managing resources conti-
nuity from the edge to the cloud: Architecture and performance.” Future Gener. Comput. Syst.
, 79, 777785.
2. Guo, Z, Zhou, K, Zhang, X, & Yang, S. (2018). “A deep learning model for short-term power
load and probability density forecasting.” Energy, 160, 1186-1200.
3. Wang, K, Xu, C, Zhang, Y, Guo, S, & Zomaya, A. Y. (2017). “Robust big data analytics for
electricity price forecasting in the smart grid.” IEEE Transactions on Big Data, 5(1), 34-45.
4. Tong, C, Li, J, Lang, C, Kong, F, Niu, J, & Rodrigues, J. J. (2018). “An efficient deep model
for day-ahead electricity load forecasting with stacked denoising auto-encoders.” Journal of
Parallel and Distributed Computing, 117, 267-273.
5. Kim, J, Moon, J, Hwang, E, & Kang, P. (2019). “Recurrent inception convolution neural net-
work for multi short-term load forecasting.” Energy and Buildings, 194, 328-341.
6. Qiu, X, Suganthan, P. N, & Amaratunga, G. A. (2018). “Ensemble incremental learning ran-
dom vector functional link network for short-term electric load forecasting.” Knowledge-Based
Systems, 145, 182-196.
7. Wang, H. Z., Li, G. Q., Wang, G. B., Peng, J. C., Jiang, H., & Liu, Y. T. (2017). “Deep learning
based ensemble approach for probabilistic wind power forecasting.” Applied energy, 188, 56-
70.
8. Wang, J, Liu, F, Song, Y., & Zhao, J. (2016).“ A novel model: Dynamic choice artificial neural
network (DCANN) for an electricity price forecasting system.” Applied Soft Computing, 48,
281-297.
9. Chen, Y., & Tan, H. (2017). “Short-term prediction of electric demand in building sector via
hybrid support vector regression.” Applied energy, 204, 1363-1374.
10. Kuo, P. H., & Huang, C. J. (2018). “A high precision artificial neural networks model for
short-term energy load forecasting. ”Energies, 11(1), 213.
11. Ahmad, A., Javaid, N., Mateen, A., Awais, M., & Khan, Z. (2019). Short-term load forecasting
in smart grids: an intelligent modular approach. Energies, 12(1), 164.
12. Liang, Y., Niu, D., & Hong, W. C. (2019). Short term load forecasting based on feature ex-
traction and improved general regression neural network model. Energy, 166, 653-663.
13. Yang, A., Li, W., & Yang, X. (2019). Short-term electricity load forecasting based on feature
selection and Least Squares Support Vector Machines. Knowledge-Based Systems, 163, 159-
173.
14. Cerjan, M., Petricic, A., & Delimar, M. (2019). HIRA Model for Short-Term Electricity Price
Forecasting. Energies, 12(3), 568. Ribeiro, G. T., Mariani, V. C., & dos Santos Coelho, L.
(2019).
15. Enhanced ensemble structures using wavelet neural networks applied to short-term load fore-
casting. Engineering Applications of Artificial Intelligence, 82, 272-281.
16. Chen, Y, Kloft, M, Yang, Y, Li, C, & Li, L. (2018).“ Mixed kernel based extreme learning
machine for electric load forecasting.” Neurocomputing, 312, 90-106.
17. Moon, J, Kim, K H, Kim, Y, & Hwang, E. (2018, January).“ A short-term electric load fore-
casting scheme using 2-stage predictive analytics.” In 2018 IEEE International Conference on
Big Data and Smart Computing (BigComp) (pp. 219-226). IEEE.
Article
Full-text available
With emergence of automated environments, energy demand increased with unexpected ratio, especially total electricity consumed in the residential sector. This unexpected increase in demand in energy brings a challenging task of maintaining the balance between supply and demand. In this work, a robust artificial ecosystem-inspired optimizer based on demand-side management is proposed to provide the optimal scheduling pattern of smart homes. More precisely, the main objectives of the developed framework are: i) Shifting load from on-peak hours to off-peak hours while fulfilling the consumer intends to reduce electricity-bills. ii) Protect users comfort by improving the appliances waiting time. Artificial ecosystem optimizer (AEO) algorithm is a novel optimization technique inspired by the energy flocking between all living organisms in the ecosystem on earth. Demand side management (DSM) program is modeled as an optimization problem with constraints of starting and ending of appliances. The proposed optimization technique based DSM program is evaluated on two different pricing schemes with considering two operational time intervals (OTI). Extensive simulation cases are carried out to validate the effectiveness of the proposed optimizer based energy management scheme. AEO minimizes total electricity-bills while keeping the user comfort by producing optimum appliances scheduling pattern. Simulation results revealed that the proposed AEO achieved a minimization electricity-bill up to 10.95, 10.2% for RTP and 37.05% for CPP for the 12 and 60 min operational time interval (OTI), respectively, in comparison to other results achieved by other optimizers. On the other hand peak to average ratio (PAR) is reduced to 32.9% using RTP and 31.25% using CPP tariff.
Thesis
Electricity theft (ET) is a major problem in developing countries. It a�ects the economy that causes revenue loss. It also decreases the reliability and stability of electricity utilities. Due to these losses, the quality of supply e�ects and tari � imposed on legitimate consumers. ET is an essential part of Non-technical loss (NTL) and it is challenging for electricity utilities to �nd the responsible people. Several methodologies have developed to identify ET behaviors automatically. However, these approaches mainly assess records of consumers' electricity usage, may prove inadequate in detecting ET due to a variety of theft attacks and irregularity of consumers' behavior. Moreover, some important challenges are needed to be addressed. (i) The number of normal consumers has been wrongly identi�ed as fraudulent. This leads to high False-positive rate (FPR). After the detection of theft, on-site inspection is needed to validate the detected person, either is it fraudulent or not and it is costly. (ii) The imbalanced nature of datasets which negatively a�ect on the model's performance. (iii) The problem of over�tting and generalization error is often faced in deep learning models, predicts unseen data inaccurately. So, the motivation for this work to detect illegal consumers accurately. We have proposed four Arti�cial intelligence (AI) models in this thesis. In system model 1, we have proposed Enhanced arti�cial neural network blocks with skip connections (EANNBS). It makes training easier, reduces over�tting, FPR and generalization error, as well as execution time. Temporal convolutional network with enhanced multi-layer perceptron (TCN-EMLP) is proposed in system model 2. It analyzes the sequential data based on daily electricity-usage records, obtained from smart meters. At the same time, EMLP integrates the non-sequential auxiliary data, such as data related to electrical connection type, property area, electrical appliances usage, etc. System model 3 based on Residual network (RN) that is used to automate feature extraction while three tree-based classi�ers such as Decision tree (DT), Random forest (RF) and Adaptive boosting (AdaBoost) are trained on the obtained features for classi�cation. Hyperparameter tuning toolkit is presented in this system model, named as Hyperactive optimization toolkit. Bayesian is used as an optimizer in this toolkit that aims to simplify the tuning process of DT, RF and AdaBoost. In system model 4, input is forwarded to three di�erent and well-known Machine learning (ML) techniques, i.e., Support vector machine (SVM), as an input. At this stage, a meta-heuristic algorithm named Simulated annealing (SA) is employed to acquire optimal values for ML models' hyperparameters. Finally, ML models' outputs are used as features for meta-classi�ers to achieve �nal classi�cation with Light Gradient boosting machine (LGBM) and Multi-layer perceptron (MLP). Furthermore, Pakistan residential electricity consumption dataset (PRECON1), State grid corporation of china (SGCC2) and Commission for energy regulation (CER3) datasets is used in this thesis. SGCC dataset contains 9% fraudulent consumers, which are extremely less than non-fraudulent consumers, due to the imbalance nature of data. Furthermore, many classi�cation techniques have poor predictive class accuracy for the positive class. These techniques mainly focus on minimizing the error rate while ignoring the minority class. Many re-sampling techniques are used in literature to adjust the class ratio; however, sometimes, these techniques remove the important information that is necessary to learn the model and cause over�tting. By using six previous theft attacks, we generate theft cases to mimic the real world theft attacks in original data. We have proposed the combination of oversampling and under-sampling techniques that is Near miss borderline synthetic minority oversampling technique (NMB-SMOTE), Tomek link borderline synthetic minority oversampling technique with support vector machine (TBSSVM) and Synthetic minority oversampling technique with near miss (SMOTE-NM) to handle imbalanced classi�cation problem. We have conducted a comprehensive experiment using SGCC, CER and PRECON datasets. The performance of suggested model is validated using di�erent performance metrics that are derived from Confusion matrix (CM).
Article
Full-text available
In competitive power markets, electric utilities, power producers, and traders are exposed to increased risks caused by electricity price volatility. The growing reliance on renewable sources and their dependence on weather, nuclear uncertainty, market coupling, and global financial instability are contributing to the importance of accurate electricity price forecasting. Since power markets are not all equally developed, different price forecasting methods have been introduced for individual markets. The aim of this research is to introduce a short-term electricity price forecasting method that addresses the problems of price volatility, a varying number of input parameters, varying data availability, and a large number of parameters and input data. Furthermore, the proposed model can be used on any market as it targets the characteristics and specifics of each market. The proposed Hybrid Iterative Reactive Adaptive (HIRA) method consists of two phases. In analysis phase, fundamental parameters which affect the electricity price are identified depending on market development. Obtained parameters are used as data inputs for price forecasting using a hybrid method. The HIRA model combines a statistical approach for large data set analysis and a similar day method with neural network tools. Similar days are examined using a statistical method which combines correlation significance, price volatility, and forecasting accuracy of the historical data. Data are collected based on their availability and electricity prices are forecasted in several iterations. All relevant data for price forecasting are collected, categorized, and arranged using simple indicators which makes the HIRA model adaptive and reactive to new market circumstances. The proposed model is validated using the Hungarian Power Exchange (HUPX) electricity price data records. The results show that with HIRA model forecasting, the error is stable and does not depend on price volatility. The HIRA method has proven to be applicable for forecasting electricity prices in real-time market conditions and enables effective hedging of price risk in the production or market portfolio.
Article
Full-text available
Daily operations and planning in a smart grid require a day-ahead load forecasting of its customers. The accuracy of day-ahead load-forecasting models has a significant impact on many decisions such as scheduling of fuel purchases, system security assessment, economic scheduling of generating capacity, and planning for energy transactions. However, day-ahead load forecasting is a challenging task due to its dependence on external factors such as meteorological and exogenous variables. Furthermore, the existing day-ahead load-forecasting models enhance forecast accuracy by paying the cost of increased execution time. Aiming at improving the forecast accuracy while not paying the increased executions time cost, a hybrid artificial neural network-based day-ahead load-forecasting model for smart grids is proposed in this paper. The proposed forecasting model comprises three modules: (i) a pre-processing module; (ii) a forecast module; and (iii) an optimization module. In the first module, correlated lagged load data along with influential meteorological and exogenous variables are fed as inputs to a feature selection technique which removes irrelevant and/or redundant samples from the inputs. In the second module, a sigmoid function (activation) and a multivariate auto regressive algorithm (training) in the artificial neural network are used. The third module uses a heuristics-based optimization technique to minimize the forecast error. In the third module, our modified version of an enhanced differential evolution algorithm is used. The proposed method is validated via simulations where it is tested on the datasets of DAYTOWN (Ohio, USA) and EKPC (Kentucky, USA). In comparison to two existing day-ahead load-forecasting models, results show improved performance of the proposed model in terms of accuracy, execution time, and scalability.
Article
Full-text available
One of the most important research topics in smart grid technology is load forecasting, because accuracy of load forecasting highly influences reliability of the smart grid systems. In the past, load forecasting was obtained by traditional analysis techniques such as time series analysis and linear regression. Since the load forecast focuses on aggregated electricity consumption patterns, researchers have recently integrated deep learning approaches with machine learning techniques. In this study, an accurate deep neural network algorithm for short-term load forecasting (STLF) is introduced. The forecasting performance of proposed algorithm is compared with performances of five artificial intelligence algorithms that are commonly used in load forecasting. The Mean Absolute Percentage Error (MAPE) and Cumulative Variation of Root Mean Square Error (CV-RMSE) are used as accuracy evaluation indexes. The experiment results show that MAPE and CV-RMSE of proposed algorithm are 9.77% and 11.66%, respectively, displaying very high forecasting accuracy.
Article
Smart grid and microgrid technology based on energy storage systems (ESS) and renewable energy are attracting significant attention in addressing the challenges associated with climate change and energy crises. In particular, building an accurate short-term load forecasting (STLF) model for energy management systems (EMS) is a key factor in the successful formulation of an appropriate energy management strategy. Recent recurrent neural network (RNN)-based models have demonstrated favorable performance in electric load forecasting. However, when forecasting electric load at a specific time, existing RNN-based forecasting models neither use a predicted future hidden state vector nor the fully available past information. Therefore, once a hidden state vector has been incorrectly generated at a specific prediction time, it cannot be corrected for enhanced forecasting of the following prediction times. To address these problems, we propose a recurrent inception convolution neural network (RICNN) that combines RNN and 1-dimensional CNN (1-D CNN). We use the 1-D convolution inception module to calibrate the prediction time and the hidden state vector values calculated from nearby time steps. By doing so, the inception module generates an optimized network via the prediction time generated in the RNN and the nearby hidden state vectors. The proposed RICNN model has been verified in terms of the power usage data of three large distribution complexes in South Korea. Experimental results demonstrate that the RICNN model outperforms the benchmarked multi-layer perception, RNN, and 1-D CNN in daily electric load forecasting (48-time steps with an interval of 30 min).
Article
Along with the deregulation of electric power market as well as aggregation of renewable resources, short term load forecasting (STLF) has become more and more momentous. However, it is a hard task due to various influential factors that leads to volatility and instability of the series. Therefore, this paper proposes a hybrid model which combines empirical mode decomposition (EMD), minimal redundancy maximal relevance (mRMR), general regression neural network (GRNN) with fruit fly optimization algorithm (FOA), namely EMD-mRMR-FOA-GRNN. The original load series is firstly decomposed into a quantity of intrinsic mode functions (IMFs) and a residue with different frequency so as to weaken the volatility of the series influenced by complicated factors. Then, mRMR is employed to obtain the best feature set through the correlation analysis between each IMF and the features including day types, temperature, meteorology conditions and so on. Finally, FOA is utilized to optimize the smoothing factor in GRNN. The ultimate forecasted load can be derived from the summation of the predicted results for all IMFs. To validate the proposed technique, load data in Langfang, China are provided. The results demonstrate that EMD-mRMR-FOA-GRNN is a promising approach in terms of STLF.
Article
Short-Term Electricity Load Forecasting (STLF) has become one of the hot topics of energy research as it plays a crucial role in electricity markets and power systems. Few researches aim at selecting optimal input features (Feature Selection, FS) when forecasting model is established, although more and more intelligent hybrid models are developed to forecast real-time electricity load. In fact, a good FS is a key factor that influence prediction accuracy. Based on the idea of selecting optimal input features, a hybrid model, AS-GCLSSVM, is developed to forecast electricity load in this research, which combines ACF (AutoCorrelation Function) and LSSVM (Least Squares Support Vector Machines). ACF is applied to select the informative input variables, and LSSVM is for prediction. The parameters in LSSVM are optimized by GWO (Grey Wolf Optimization Algorithm) and CV (Cross Validation). The proposed model is to forecast the half-hour electricity load of the following week. Experimental results show that it is an effective approach that can improve the forecasting accuracy remarkably, compared with the benchmark models.
Article
Accurate load forecasting is critical for power system planning and operational decision making. In this study, we are the first to utilize a deep feedforward network for short-term electricity load forecasting. Our results are compared to those of popular machine learning models such as random forest and gradient boosting machine models. Then, electricity consumption patterns are explored based on monthly, weekly and temperature-based patterns in terms of feature importance. Also, a probability density forecasting method based on deep learning, quantile regression and kernel density estimation is proposed. To verify the efficiency of the proposed methods, three case studies based on daily electricity consumption data for three Chinese cities for 2014 are conducted. The empirical results demonstrate that (1) the proposed deep learning-based approach exhibits better forecasting accuracy in terms of measuring electricity consumption relative to the random forest and gradient boosting model; (2) monthly, weekly and weather-related variables are key factors that have a great influence on household electricity consumption; and (3) the proposed probability density forecasting method is capable of forecasting high-quality prediction intervals via probability density forecasting.
Article
Short term electric load forecasting, as an important tool in the electricity market, plays a critical role in the management of electric systems. Proposing an accuracy and optimization method is not only a challenging task but also an indispensable part of the energy system. More and more accurate forecasting methods are needed by different people in different areas. This paper proposes a novel short-term electric load forecasting method EMD-Mixed-ELM which based on empirical mode decomposition (EMD) and extreme learning machine (ELM). EMD-Mixed-ELM first uses the empirical mode decomposition to decompose the load series for capturing the complicated features of the electric load and de-noising the data. Considering that the performance of extreme learning machine (ELM) is greatly influenced by the choice of kernel, the mixed kernel method is proposed for ELM. The mixed kernel combines the RBF kernel and the UKF kernel. The forecasting results of the EMD-Mixed-ELM are proved to be better than all the other three methods (RBF-ELM, UKF-ELM and Mixed-ELM) and other existing methods (MFES, ESPLSSVM and Combined method). To verify the forecasting ability of the EMD-Mixed-ELM, half-hourly electric load data from the state of New South Wales, Victoria and Queensland in Australia are used in this paper as a case study. The experimental results clearly indicate that for this three datasets, the forecasting accuracy of the proposed method is superior to other methods.
Article
Short-term electric load forecasting plays an important role in the management of modern power systems. Improving the accuracy and efficiency of electric load forecasting can help power utilities design reasonable operational planning which will lead to the improvement of economic and social benefits of the systems. A hybrid incremental learning approach composed of Discrete Wavelet Transform (DWT), Empirical Mode Decomposition (EMD) and Random Vector Functional Link network (RVFL) is presented in this work. RVFL network is a universal approximator with good efficiency because of the randomly generated weights between input and hidden layers and the close form solution for parameter computation. By introducing incremental learning, along with ensemble approach via DWT and EMD into RVFL network, the forecasting performance can be significantly improved with respect to both efficiency and accuracy. The electric load datasets from Australian Energy Market Operator (AEMO) were used to evaluate the effectiveness of the proposed incremental DWT-EMD based RVFL network. Moreover, the attractiveness of the proposed method can be demonstrated by the comparison with eight benchmark forecasting methods.