You are on page 1of 10

Journal of Electrical Engineering & Technology

https://doi.org/10.1007/s42835-021-00973-5

ORIGINAL ARTICLE

Short‑term Price Forecasting Method in Electricity Spot Markets Based


on Attention‑LSTM‑mTCN
Guang Yang1   · Songhuai Du1 · Qingling Duan1 · Juan Su1

Received: 10 March 2021 / Revised: 13 November 2021 / Accepted: 23 November 2021


© The Korean Institute of Electrical Engineers 2021

Abstract
Accurate and rapid price forecasting plays a crucial role in the electricity spot market. Owing to the variability of market
participants’ activities, prices are usually too volatile to forecast accurately. In this study, a short-term price forecasting
model for locational marginal price (LMP) based on a multiple temporal convolutional network (mTCN) and attention-long
short-term memory (ATT-LSTM) is proposed. An attention-LSTM method is adopted to reconstruct the electrical features
for the future. The mTCN is used to extract the hidden information and long-term temporal relationships in the input features
included in electrical features. The effectiveness of the proposed model is demonstrated using datasets from the New England
electricity market (ISO-NE) in the U.S. The proposed model provides accurate load forecasting results in experiments and
comparisons with existing models.

Keywords  Short-term price forecasting · Attention · LSTM · TCN · Power market

1 Introduction integrated moving average (SARIMA), and deep belief net-


work (DBN) was proposed for short-term electricity price
In the electricity market, the benefits of trading are ulti- forecasting [4]. A combined model for short-term price fore-
mately reflected through the price [1], whose fluctuations casting including variational model decomposition, hybrid
have strong economic leverage by affecting the flow and data modeling, feature selection, a generalized regression
allocation of various resources. Electricity price forecast- neural network, and a gravity search algorithm was proposed
ing is of great importance for each participant in the market [5]. Gray correlation analysis was used to select the effec-
[2], and obtaining accurate electricity price forecasts can tive parameters, and datasets from different sources were
enable them to make more profits in electricity transac- processed using deep neural networks. The main features
tions; electricity consumers can make reasonable electric- of the input data were detected, data dimensionality reduc-
ity consumption plans to reduce electricity costs based on tion was performed, and a rough structured artificial neural
the forecast electricity prices. Electricity suppliers can rely network was used to forecast the day-ahead electricity price
on accurate price forecasts to respond to power shortages or [6]. A discrete incremental model with fractional Brownian
surplus events in certain time periods, which is conducive to motion (FBM) was used to predict electricity price series,
improving system load factors, reducing operation costs, and where FBM represents a typical nonsmooth stochastic pro-
ensuring the security and stability of the power system [3]. cess with long-range dependence (LRD) characteristics. The
The complex characteristics of electricity price FBM-driven stochastic differential equation was discretized
series,such as high volatility and non-stationarity, make into a discrete incremental model for electricity price pre-
forecasting difficult. An adaptive hybrid model based on diction [7]. A deep learning architecture with bidirectional
variational pattern decomposition (VMD), adaptive particle long short-term memory (LSTM) units was used to forecast
swarm optimization (SAPSO), the seasonal autoregressive price differences between day-ahead and real-time markets
at the same node [8]. The time-series LSTM electricity price
* Guang Yang model considering the wind-load ratio was used to improve
yangguang2018@cau.edu.cn precision [9]. A proposed multi-branch gated regression
unit (GRU) structure for electricity price forecasting could
1
College of Information and Electrical Engineering, China capture models with both long- and short-term time data
Agricultural University, Beijing, People’s Republic of China

13
Vol.:(0123456789)
Journal of Electrical Engineering & Technology

distributions [10]. A multiple input LSTM method was (3) We use the architecture of a model, including daily
proposed for electricity price and demand forecasting with and weekly forecasts. Multistep forecasting is improved
large data, whose hyperparameters were tuned using the Jaya by paying different attention to each feature and consider-
optimization algorithm to improve forecasting capability ing consequences from previous steps.
[11]. A method was proposed to create new variables based
on temperature to forecast electricity demand and tariffs.
The variables were linearly related to electricity demand,
and the method could avoid clustering of data to different 2 Summary of Method
seasons and accurately determine the temperature [12]. A
day-ahead SMP forecasting model implemented an artificial 2.1 Attentional LSTM
neural network (ANN) algorithm [13]. A statistical data-
filtering method worked at an input data preprocessing stage LSTM is a recurrent neural network with hidden states ht
to evaluate the reliability of input load data by analyzing all and Ct to store short- and long-term sequence feature infor-
possible data confidence levels and filtering out noise/outli- mation, respectively, as shown in Fig. 1. Its three-gate struc-
ers to improve the accuracy of short-term load forecasting ture realizes the long-term memory of the time series. The
models [14]. Input features were separated into historical three gates are the input gate it, forgetting gate ft, and output
and prediction data. Historical data were input to an LSTM ot, and it has a sigmoid activation function,σ.
layer to model its relationships, and its outputs were incor- The parameters of the LSTM model are collected as
porated with those of a fully connected layer [15]. Deep
learning methods have greater adaptability and accuracy in ⎡ ̂it ⎤ ⎡ sigmoid ⎤⎡ Wi Ui bi ⎤
⎡ x ⎤
time series forecasting at different scales because they can ⎢ ô t ⎥ ⎢ sigmoid ⎥⎢ Wo Uo bo ⎥ ⎢ t ⎥
⎢ f̂ ⎥ = ⎢ sigmoid ⎥⎢ W (1)
bf ⎥⎢ t−1 ⎥
h
identify structures and patterns, such as nonlinearity and Uf
⎢ ∼t ⎥ ⎢ ⎥⎢ f ⎥⎣ 1 ⎦
complexity. Bai et al. of Carnegie Mellon University pro- ⎣ Ĉ ⎦ ⎣ tanh ⎦⎣ Wc Ua ba ⎦
posed a temporal convolutional network (TCN) [16], which
including 12 LSTM parameters, such as ­W*, ­U*, and
has performance advantages over classical recurrent neural *
­b . Through the control of the three gates, the relationship
networks for sequential data processing tasks.
between the front and back of the time series is established,
Attention-based neural networks have achieved good
so that the gradient of the model error will not disappear dur-
results in natural language processing, such as machine
ing backpropagation, and the error will remain unchanged.
translation, parsing, and automatic summarization. By
A network of information can be created and trained for
assigning different weights to the hidden layer units of the
long-term dependency [17].
neural network, the attention mechanism enables the hidden
For each moment, we should attach importance to which
layer to pay more attention to the critical information. The
moment in its past is different. If we can set a vector that
TCN model is used in fields such as pattern recognition,
not only considers the weight of moments but changes
anomaly detection, and mental assessment, but its applica-
dynamically according to different moments, we can
tion to feature extraction in load forecasting is relatively lim-
obtainthe attention model. To achieve this, h(t-1) is changed
ited. We propose a method based on the attention-LSTM-
to h(t),considering the weight of time, weight information is
TCN model to forecast the short-term electricity price,
added, and the weight proportion corresponding to past time
aiming to improve prediction accuracy. The main contribu-
t is introduced [18],
tions are summarized as follows:

(1) Attention-LSTM captures the actual variation and


fluctuation trends in features. By considering the loca-
tional marginal price (LMP) and day-ahead cleared
demand in the day-ahead market, the real-time LMP is
forecast based on an improved multiple-TCN model for
the electricity spot market.
(2) The mTCN model, integrating parallel feature pro-
cessing and time-domain modeling, is proposed for
the feature in price forecasting. This model solves two
problems due to its special convolutional structure and
residual block. It can extract deep features and long-term
temporal relationships, and it avoids the vanishing or
exploding gradient during deep network training. Fig. 1  Neuron Structure of LSTM

13
Journal of Electrical Engineering & Technology


⎛ h ⎞ t−1 � � perceptual field, and allowing it to receive information over a
⎜ (t−1)⎟ � h(t−1) longer period of time. With hole convolution, as the number
⎜ ∼ ⎟= 𝜕𝜏(t) (2) of layers increases, the convolution window becomes larger,
⎜ c ⎟ 𝜏=1 c(t−1)
⎝ (t−1)⎠ and the number of holes in the window increases. For a one-
dimensional sequence input x ∈ ­Rn and a filter F: {0, 1, …,
k − 1} → R, the hole convolution operation F on the sequence
𝜕𝜏(t) = soft max(w𝜏(t) ) (3)
element s,
( )
∼ ∑
k−1
w𝜏(t) ∶=g x(t) , h , h F(s) = (x ∗ fd )(s) = f (i)xs−di (6)
(t−1) (𝜏)
( ∼
) (4) i=0
T
=v tanh Wx x(t) + W ∼ h +Wh , h
h (t−1)
where x is the input time series, "*" denotes convolution,
(𝜏)
­fd is the filter corresponding to dilation factor d,xs-di is the
The parameters of the attention-LSTM model are col- input sequence, and k is the filter size. An example of causal
lected as convolution with dilation factors d = 1, 2, 4 and filter size
k = 3 is shown in Fig. 1, ­wherex0, ­x1, …, ­xT are the input
⎡ ̂it ⎤ ⎡ sigmoid ⎤⎡ Wi Ui bi ⎤⎡ xt ⎤ sequences, and ­y0, ­y1, …, ­yT are the output sequences. Obvi-
⎢ ô ⎥ ⎢ ously, the output sensory field can cover all the values in
⎢ t ⎥ = ⎢ sigmoid ⎥⎥⎢⎢ Wo Uo bo ⎥⎢ ∼ ⎥
(5)
bf ⎥⎢ t ⎥
h the input sequence both after and before the time point, as
⎢ f̂ t ⎥ ⎢ sigmoid ⎥⎢ Wf Uf
⎥⎢ ⎥
⎢ C̃̂ ⎥ ⎣ tanh ⎦⎣ W Ua ba ⎦⎣ 1 ⎦ in Fig. 3.
⎣ ⎦ c
TCN uses residual networks to solve the problem of gradi-
The attention mechanism gives more attention to the key ent disappearance or explosion in deep networks. ResNet uses
parts of the input sequence that affect the output results, so a nonlinear change function to describe the input and output
asto better learn its information. This does not increase the of a network, i.e., the input is X and the output is o. F usually
calculation and storage of the model. Therefore, the attention includes operations such as convolution and activation. The
mechanism is introduced in the LSTM model to effectively output is expressed as a linear superposition and nonlinear
improve forecasting. transformation of the input,
o = F(x) + x (7)
2.2 Temporal Convolutional Network
Since deep learning relies on chain backpropagation of
The TCN is essentially a one-dimensional CNN that is opti- errors for parameter updates, once one of the derivatives is
mized and adapted to the time series problem, and improved small, the gradient may become increasingly smaller after mul-
through causal convolution, dilated convolution, and are si tiple concatenations, which is referred to as gradient vanishing.
dual block [16]. For deep networks, the pass to shallow layers is almost gone.
Causal convolution ensures that the prediction results With the use of residuals, a constant term of 1 is added to each
at earlier time steps do not involve future data information, derivative,
which allows the convolutional network to be used in time
series models, as in Fig. 2; hole convolution allows each hid- 𝜕h 𝜕(f + x) 𝜕f
= =1+ (8)
den layer to keep the size of the input sequence of the previ- 𝜕x 𝜕x 𝜕x
ous layer, reducing computational effort while increasing the

Fig. 2  Causal convolution Fig. 3  Dilated convolution

13
Journal of Electrical Engineering & Technology

and the error can still be effectively backpropagated, even


if the original derivative 𝜕f ∕𝜕x is very small.
This approach allows deep networks to directly learn
not only the global transformations but also the underlying
constant relations or residual terms. Increasing the percep-
tual field by causal convolution and building a deep hidden
layer by residual networks enables efficient implementation
of temporal convolutional networks in temporal prediction.
Unlike recurrent neural networks, causal convolution allows
the network to capture longer sequences of information,
and the residual-connected temporal convolutional network
avoids gradient explosion or disappearance, as shown in
Fig. 4.

2.3 Short‑Term Price Forecasting Method Based


on Attention‑LSTM‑TCNs

Based on the TCN model, we analyze the coordination


relationship between the day-ahead and real-time markets
in the electricity spot market; consider the impact of the
day-ahead market LMP, day-ahead cleared demand load,
day-ahead active power portion LMP, day-ahead congested
power portion LMP, and day-ahead lost power portion on the
next day's real-time market; and design an improved multiple
TCN deep learning network model to forecast them. Figure 5 Fig. 5  Multiple TCN-based real-time LMP model processing steps
shows the steps in the model, Fig. 6 shows the improved for electricity spot markets
prediction model, and Table 1 shows the processing of real-
time electricity price data and related data. is {x1, ­x2 … x­ m}, and the time step T and sample labels
In the TCN multiple time series model, the prediction of are added to the original dataset to construct the dataset as
time series is transformed to supervised learning in machine {x1,x2,…,xi,…,xT,yi}, where y­ i denotes the supervised learn-
learning, assuming the number of samples is M, the dataset ing label of the ith sample, the prediction result of using data
of length T for the data at time point T + 24/168.
This paper makes two main improvements in the structure
of the TCN deep learning model.
Layers of attention-LSTM are added between the input
and TCN cell layers. Attention-LSTM layers transform the
input day-ahead market impact factor time series data from
one feature space to another. The real-time market real-time
LMP series as the target space, affected by each dimension
of the input space, the real-time market real-time electric-
ity price vector is a weighted sum of the day-ahead market
impact factor vectors. This layer takes the parse sequence
and inputs it as a newly generated feature to the next TCN
cell layer.
A fully connected layer is added before the result out-
put, acting as a"classifier" in the deep learning network, and
mapping the learned features to the target space.
The input layer is a fully connected network, and a deep
learning model with five TCN hidden layers is built. The
numbers of neurons in the hidden layers are 32, 16, 8, 4, 2;
the length of the convolutional kernel is 3; the batch length is
24; the learning rate is set to 0.001; and the number of model
Fig. 4  Residual block iterations is 50. Min–max normalization is performed on the

13
Journal of Electrical Engineering & Technology

Fig. 6  Multiple TCN-based
real-time LMP forecasting
model for electricity spot
market

Table 1  Real-time electricity Data Input variable t


price data and related data
tables t Real-time LMP t-24/168 Day-ahead Locational Marginal Price
t-24/168 Day-ahead Cleared Demand
t-24/168 Energy Component of Day-ahead LMP
t-24/168 Congestion Component of Day-ahead
LMP
t-24/168 Marginal Loss Component of Day-ahead
LMP
t-24/168 Real-Time LMP

timeseries data. The adaptive moment estimation method for different parameter gradients based on the loss function,
(Adam) is used for gradient optimization. This can dynami- which has the advantages of high computational efficiency,
cally adjust the first- and second-order moment estimation small memory occupation, and easy implementation.

13
Journal of Electrical Engineering & Technology

2.4 Performance Evaluation

We used absolute error (APE), mean absolute percentage


error (MAPE), root mean square error (RMSE), and mean
absolute error (MAE), defined as
| pre |
i ||
|Pi − Preal
APE = | (9)
Preal
i

N | pre real |
100% ∑ ||Pi − Pi || (10)
MAPE =
N 1 Preal
i


�N
� ∑ pre
� (P − Preal )2 Fig. 7  Real-time tariff fluctuations in September 2019
� (11)
� 1 i i
RMSE =
N


N
� pre �
�Pi − Preal
i ��
1 � (12)
MAE =
N
as evaluation metrics of power prediction error, where N is
the number of test data points, Pipre denotes the i-th predicted
power data, and Pireal denotes the i-th measured power data.
Fig. 8  Real-time LMP forecast on July 31, 2019

3 Case Study
3.2 Experiment 1:Real‑time LMP Forecasting in Two
Representative Days
3.1 Experimental Settings
This experiment involved the prediction of 24 h real-time
The experimental environment was implemented with
LMP on July 31, 2019. Data from January 1, 2016, to July
Python 3.6.2, the deep learning development framework was
31, 2019, were used as the training set, data of July 29–30,
TensorFlow 2.0.0a GPU, Intel Core i5-7200U 64-bitCPU at
2019, as the validation set, and data from July 31, 2019, as
2.50 GHz ~ 2.70 GHz, 8 GB RAM, and an Nvidia GeForce
the test set.
940MX graphics card.
We predicted 24 h real-time LMP on December 31, 2019,
The dataset was the annual whole-point data of the New
with data from January 1, 2016, to December 29, 2019,as the
England Electricity Market (ISO-NE) in the United States,
training set, data of December 29–30, 2019, as the validation
selected for the Connecticut (CT) region [19]. The real-time
set, and data of December 31, 2019, as the test set.
electricity price data were collected for 1461 consecutive days,
Figure 8 and Table 2 compare the 24 h real-time LMP
from January 1, 2016 to December 31, 2019, once per hour, for
prediction results of the two models of LSTM, S-TCN,
a total of 35,064 moments, including real-time tariff-related
and M-TCN on July 31, 2019, Fig. 9 and Table 3 present
data, load-related data, and day-ahead tariff-related data.
the same comparisons on December 31, 2019, and Table 4
Fluctuations of the real-time LMP series are shown in
shows the overall evaluation.
Fig. 7, using a time series decomposition of 24 full points per
day in December of the 2019 dataset.
3.3 Experiment 2: Real‑timeLMP intraweek
Figure 7 shows the series of forecast dates, from top to bot-
forecasting
tom, as the original, trend, seasonal, and residual series. It can
be seen that the linear autocorrelation of the real-time market
In this experiment, price data were forecast from Decem-
real-time LMP is not strong and is influenced by random noise.
ber 25 to 31, 2019, a span of 168 h. The training set was
from January 1, 2016, to December10, 2019, the validation

13
Journal of Electrical Engineering & Technology

Table 2  The forecast result Predicted time Actual value ($/ LSTM S-TCN M-TCN
comparisons in July 31, 2019 MWh)
Values APE Values APE Values APE

0:00 32.15 34.38 0.07 28.18 0.12 22.90 0.29


1:00 40.84 39.42 0.03 31.92 0.22 23.30 0.43
2:00 34.46 46.03 0.34 39.54 0.15 21.61 0.37
3:00 30.72 42.97 0.40 33.95 0.11 20.63 0.33
4:00 32.21 41.29 0.28 30.67 0.05 21.27 0.34
5:00 34.82 41.58 0.19 31.97 0.08 22.16 0.36
6:00 29.31 42.78 0.46 34.26 0.17 22.46 0.23
7:00 29.5 39.40 0.34 29.43 0.00 23.98 0.19
8:00 29.75 39.53 0.33 29.59 0.01 28.61 0.04
9:00 26.24 39.40 0.50 29.81 0.14 28.89 0.10
10:00 27.19 37.20 0.37 26.73 0.02 28.97 0.07
11:00 29.62 37.58 0.27 27.57 0.07 33.15 0.12
12:00 35.32 38.91 0.10 29.70 0.16 37.74 0.07
13:00 37.26 42.60 0.14 34.70 0.07 41.40 0.11
14:00 41.83 44.16 0.06 36.40 0.13 41.01 0.02
15:00 38.28 47.29 0.24 40.41 0.06 44.67 0.17
16:00 27.86 45.16 0.62 37.30 0.34 43.71 0.57
17:00 26.9 39.01 0.45 28.16 0.05 44.03 0.64
18:00 26.4 37.73 0.43 27.31 0.03 38.93 0.47
19:00 22.22 36.79 0.66 26.88 0.21 29.97 0.35
20:00 22.74 34.34 0.51 23.21 0.02 29.51 0.30
21:00 22.07 34.59 0.57 23.66 0.07 29.10 0.32
22:00 19.67 34.37 0.75 23.08 0.17 26.58 0.35
23:00 17.61 33.27 0.89 20.97 0.19 23.65 0.34
MAPE 0.8587 0.1093 0.0692
MAX APE 1.2383 0.3388 0.6368

observed that the forecasting profile of the proposed model


is able to match the actual price, and the model captures the
occasional fluctuation. The other models show different devia-
tions in response to sudden variations or spikes. It can be seen
that the real-time LMP prediction accuracy of the proposed
method exceeds that of the other models.

4 Conclusion

We proposed a short-term price forecasting method in elec-


tricity spot markets based on attention-LSTM-mTCN, as
Fig. 9  Real-time LMP forecast on December 31, 2019 follows:

(1) A deep learning network model for short-term price


set from December10–24, 2019, and the testing set from forecasting prediction based on a multiple temporal
December 25 to 31, 2019. convolutional network and attention-long short-term
The statistical metrics of each model on the testing set are memory was designed and improved by considering
shown in Table 5. Figure 10 compares the forecast and actual the locational marginal price and day-ahead cleared
prices of each model from December 25 to 31, 2019. It is demand in the day-ahead market.

13
Journal of Electrical Engineering & Technology

Table 3  The forecast result Predicted time Actual LSTM S-TCN M-TCN


comparisons in Dec 31, 2019 LMP($/
MWh) Values ($/MWh) APE Values ($/MWh) APE Values ($/MWh) APE

0:00 20.5 37.73 0.17 24.66 0.23 16.99 0.17


1:00 15.31 30.22 0.26 22.82 0.44 16.15 0.05
2:00 10.88 22.45 0.35 18.48 0.46 16.04 0.47
3:00 11.24 18.61 0.39 15.06 0.51 14.79 0.32
4:00 15.75 19.24 0.40 15.18 0.53 15.93 0.01
5:00 18.01 22.93 0.34 18.61 0.47 17.79 0.01
6:00 17.35 24.51 0.16 20.33 0.31 21.55 0.24
7:00 18.96 24.14 0.18 19.82 0.33 22.80 0.20
8:00 21.14 25.62 0.14 21.05 0.29 22.83 0.08
9:00 24.54 27.47 0.05 22.70 0.13 22.76 0.07
10:00 27.22 30.55 0.12 25.28 0.07 24.29 0.11
11:00 25.9 33.15 0.12 27.31 0.08 23.10 0.11
12:00 21.78 32.41 0.08 26.31 0.26 21.35 0.02
13:00 22.19 28.89 0.22 23.18 0.38 21.45 0.03
14:00 22.48 29.56 0.29 23.50 0.44 21.59 0.04
15:00 26.59 29.56 0.23 23.72 0.38 23.19 0.13
16:00 24.13 33.24 0.19 26.85 0.04 23.67 0.02
17:00 24.32 30.94 0.15 24.98 0.07 30.48 0.25
18:00 20.84 31.63 0.20 25.12 0.05 25.08 0.20
19:00 19.99 28.18 0.27 22.48 0.01 24.04 0.20
20:00 19.16 27.79 0.22 21.84 0.04 24.38 0.27
21:00 19.43 26.68 0.21 21.20 0.04 22.86 0.18
22:00 20.97 27.00 0.37 21.42 0.09 22.68 0.08
23:00 24.01 28.13 0.60 22.58 0.28 20.51 0.15
MAPE 0.3969 0.1421 0.0615
MAX APE 0.5974 0.5287 0.4742

Table 4  The forecast result Predicted time S-LSTM S-TCN M-TCN


comparisons of 24-h prediction
results between winter and RMSE MAE RMSE MAE RMSE MAE
summer ($/MWh)
2019/7/31 10.80 24.61 4.06 3.24 3.44 2.69
2019/12/31 9.90 7.41 3.17 2.47 1.32 1.27

(2) The model was composed of daily and weekly fore-


Table 5  Load forecasting evaluationon testing set
casts and could perform multistep forecasting to
Time From 25-December -2019 to 31-Decem- improve accuracy by paying different attention to each
ber -2019
feature.
Statistical metrics γmae ($/MWh) γmape (%) γrmse ($/MWh) (3) Forecasting and comparative analysis were performed
using LSTM, gradient boosting decision tree, random
Random forest 5.28 0.46 7.56
forest, and TCN, using the annual ISO-NE series data
GBRT 3.3 0.28 5.16
as the research object. Results showed that the pro-
LSTM 5.64 0.58 7.29
posed forecasting method could significantly reduce
Bi-LSTM 5.82 0.58 7.48
forecasting error and improve accuracy with the
CNN-LSTM 4.86 0.52 6.70
data samples and structural scenarios covered in this
ATT-LSTM 5.28 0.56 7.00
study.
TCN 5.47 0.59 8.15
Proposed model 2.96 0.29 4.45

13
Journal of Electrical Engineering & Technology

Fig. 10  Real-time LMP forecast from 25-December -2019 to 31-December -2019

Acknowledgements  This work was supported by National Key R&D their application(DSA), 978–1–7281–6057–3/19/$31.00, IEEE,
Program of China (No.2016YFB0900100). https://​doi.​org/​10.​1109/​DSA.​2019.​00045
10. Haolin Yang, Kristen R. Schell (2020) HFNet: Forecast-
ing real-time electricity price via novel GRU architec-
tures;978–1–7281–2822–1/20/$31.00, IEEE PMAPS
References 11. Khalid R, Javaid N et al (2020) Electricity load and price forecast-
ing using jaya-long short term memory (JLSTM) in smart grids.
Entropy 22(1):10. https://​doi.​org/​10.​3390/​e2201​0010
1. Su Juan, Du Songhuai, Li Talent (2007) Research on short-term
12. Jasinski To (2020) Use of new variables based on air temperature
spot electricity price forecasting method based on multi-factor
for forecasting day-ahead spot electricity prices using deep neural
wavelet decomposition by neural network, Power Autom Equip-
networks: A new approach. Energy 213:118784
ment, 27(11)
13. Jufri F H, Seongmun O, Jung J (2019) Day-ahead system marginal
2. Songhuai Du, Buying Wen, Chuanwen Jiang, (2004) “Power Mar-
price forecasting using artificial neural network and similar-days
ket,” China Electric Power Press, March, pp 1–9
information. J Electr Eng Technol 14:561–568. https://d​ oi.o​ rg/1​ 0.​
3. Mohammad shahidehpour et al. (2005) Original book, Compiled
1007/​s42835-​018-​00058-w
by Songhuai Du et al., “Market oriented operation of power sys-
14. Kwon B-S, Park R-J, Song K-B (2020) Short-term load fore-
tem,” China Electric Power Press.9, pp 13–14
casting based on deep neural networks using LSTM layer. J
4. Zhang J, Tan Z, Wei Y (2020) An adaptive hybrid model for short
Electr Eng Technol 15:1501–1509. https://​d oi.​o rg/​1 0.​1 007/​
term electricity price forecasting. Appl Energy 258:114087
s42835-​020-​00424-7
5. Heydari A, Nezhad M M, Pirshayan E, Garcia D A, Keynia F, De
15. Bui DM, Le PD, Cao TM et al (2020) A statistical data-filter-
Santoli L (2020) Short-term electricity price and load forecast-
ing method proposed for short-term load forecasting models.
ing in isolated power grids based on composite neural network
J Electr Eng Technol 15:1947–1967. https://​doi.​org/​10.​1007/​
and gravitational search optimization algorithm. ApplEnergy
s42835-​020-​00460-3
277:115503
16. S. Bai, J. Z. Kolter, and V. Koltun, (2018) An empirical evalua-
6. Deng J, Song W, Zio E (2020) A discrete increment model
tion of generic convolutional and recurrent networks for sequence
for electricity price forecasting based on fractional brownian
modelling. In: Proceedings. AAAI conference. artificial. intel-
motion. IEEE Access. 8:130762–130770. https://d​ oi.o​ rg/1​ 0.1​ 109/​
ligence. (AAAI), New Orleans, LA, United states, pp 2159–2166
ACCESS.​2020.​30087​97
17. D. Bahdanau, K. Cho ,Y. Bengio. (2015) Neural machine transla-
7. Jahangir H, Tayarani H, Baghali S, Ahmadian A, Elkamel A
tion by jointly learning to align and translate[J]. ICIR
(2020) A novel electricity price forecasting approach based on
18. J. Cheng, L. Dong, M. Lapata. (2016) Long short-term memory-
dimension reduction strategy and rough artificial neural networks.
networks for machine reading [D].arxiv:​1601.​06733
IEEE Trans Ind Inform 16(4):2369–2361
19. https://​www.​iso-​ne.​com/
8. Ronit Das, Rui Bo, et al (2020) Cross-market price difference
forecast using deep learning for electricity markets. In: IEEE PES
Publisher's Note Springer Nature remains neutral with regard to
innovative smart grid technologies Europe (ISGT-Europe) Virtual,
jurisdictional claims in published maps and institutional affiliations.
26–28
9. Sangli Liu, Liang Zhang, Bin Zou (2019) Study on electricity mar-
ket price forecasting with large-scale wind power based on LSTM,
In: 2019 6th International conference on dependable systems and

13
Journal of Electrical Engineering & Technology

Guang Yang  received his Qingling Duan  received the Ph.D.


M.S.degree in Electronic and- degree in agriculturalelectrifica-
communication engineering from tion and automation from China
XidianUniversity,Xian,China, in AgriculturalUniversity,Beijing,
2009.He is a Senior Engineer of China, in 2011. She is currently
PowerEngineering technology aProfessor of computer science
and a Senior Engineer of Compu- and technology and theDeputy
terTechnology. Currently, he is Director of the Department of
pursuing Ph.D.degree at ChinaA- ComputerEngineering, College
gricultural University,Beijing, of Information and ElectricalEn-
China. His research interestsin- gineering, China Agricultural
clude power system automation, University.She was involvedin
power systeminformatization, the research domains of informa-
forecasting of electricity market tion processing andartificial
andartificial intelligence. intelligence.

Juan Su  received the Ph.D.


Songhuai Du  received the Ph.D. degree in ElectricalEngineering
degree in ElectricalEngineering f ro m C h i n a Ag r i c u l t u r a l
from Shanghai Jiao Tong University,Beijing,China, in
University,Shanghai,China, in 2009. She is currently a associate
2002. He is currently a leading- Professor ofPower Engineering
Professor of PowerEngineering, and the Deputy Director of
College of Information and Elec- theDepartment of Electrical
tricalEngineering, China Agri- Engineering, College ofInforma-
c u l t u r a l Un i ve r s i t y. H i s tion and Electrical Engineering,
researchinterests include power China AgriculturalUniversity.
system automation, relayingpro- Her research interests include
tection and electricity market. power systemautomation, power
system informatization and
electricitymarket.

13

You might also like