没有合适的资源?快使用搜索试试~ 我知道了~
一种创新的基于随机森林的非线性集成范式,用于碳价格预测的改进特征提取和深度学习1
需积分: 0 3 下载量 180 浏览量
2022-08-03
18:56:49
上传
评论 1
收藏 3.73MB PDF 举报
温馨提示
试读
13页
一种创新的基于随机森林的非线性集成范式,用于碳价格预测的改进特征提取和深度学习1
资源详情
资源评论
资源推荐
An innovative random forest-based nonlinear ensemble paradigm of
improved feature extraction and deep learning for carbon
price forecasting
Jujie Wang
a,b,
⁎
,XinSun
a
,QianCheng
a
,QuanCui
a
a
School of Management Science and Engineering, Nanjing University of Information Science and Technology, Nanjing 210044, China
b
Collaborative Innovation Center on Forecast and Evaluation of Meteorological Disasters, Nanjing University of Information Science and Technology, Nanjing 210044, China
HIGHLIGHTS
• Propose a new model based on nonsta-
tionary and nonlinear data: CEEMDAN-
SE-LATM-RF
• The efficacy was tested in different car-
bon trading markets in China.
• Proposed model outperformed the
other 4 benchmark methods.
• Improved feature extraction heightens
the forecast accuracy.
• Nonlinear ensemble model repre-
sents better generalization ability for
robustness.
GRAPHICAL ABSTRACT
abstractarticle info
Article history:
Received 19 July 2020
Received in revised form 4 October 2020
Accepted 11 October 2020
Available online 16 October 2020
Editor: Kuishuang Feng
Keywords:
Carbon price prediction
Hybrid model
Improved feature extraction
Long short-term memory network
Nonlinear ensemble algorithm
Random forest
Carbon price is the basis of developing a low carbon economy. The accurate carbon price forecast can not only
stimulate the actions of enterprises and families, but also encourage the study and development of low carbon
technology. However, as the original carbon price series is non-stationary and nonlinear, traditional methods
are less robust to predict it. In this study, an innovative nonlinear ensemble paradigm of improved feature extrac-
tion and deep learning algorithm is proposed for carbon price forecasting, which includes complete ensemble
empirical mode decomposition (CEEMDAN), sample entropy (SE), long short-term memory (LSTM) and random
forest (RF). As the core of the proposed model, LSTM enhanced from the recurrent neural network is utilized to
establish appropriate prediction models by extracting memory features of the long and short term. Improved fea-
ture extraction, as assistant data preprocessing, represents its unique advantage for improving calculating effi-
ciency and accuracy. Removing irrelevant features from original time series through CEEMDAN lets learning
easier and it's even better for using SE to recombine similar-complexity modes. Furthermore, compared with
simple linear ensemble learning, RF increases the generalization ability for robustness to achieve the final nonlin-
ear output results. Two markets' real data of carbon trading in china are as the experiment cases to test the effec-
tiveness of the above model. The final simulation results indicate that the proposed model performs better than
Science of the Total Environment 762 (2021) 143099
Abbreviations: ADF, Augmented Dickey-Fuller; AE, Approximate entropy; ANN, Artificial neural network; ARIMA, Autoregressive integrated moving average; BDS, Brock-Decher-
Scheikman; BPNN, Back propagation neural network; CEEMD, Complete ensemble empirical mode decomposition; CEEMDAN, Complete ensemble empirical mode decomposition
with adaptive noise; EMD, Empirical mode decomposition; EEMD, Ensemble empirical mode decomposition; EWT, Empirical wavelet transform; GARCH, Generalized autoregressive con-
ditional heteroscedasticity; GPR, Gaussian process regression; GRU, Gated recurrent unit; IMF, Intrinsic mode function; KNN, K-nearest neighbor; LSSVM, Least squares support vector ma-
chine; LSTM, Long short-term memory; MAE, Mean absolute error; MAPE, Mean absolute percentage error; RF, Random forest; RMSE, Root mean square error; RNN, Recurrent neural
network; SE, Sample entropy; SVM, Support vector machine; VAR, Vector auto regression; VARIMA, Vector autoregressive integrated with moving average; WT, Wavelet transform;
XGboost, Extreme gradient boosting.
⁎ Corresponding author at: School of Management Science and Engineering, Nanjing University of Information Science and Technology, Nanjing 210044, China.
E-mail address: jujiewang@126.com (J. Wang).
https://doi.org/10.1016/j.scitotenv.2020.143099
0048-9697/© 2020 Elsevier B.V. All rights reserved.
Contents lists available at ScienceDirect
Science of the Total Environment
journal homepage: www.elsevier.com/locate/scitotenv
the other four benchmark methods reflected by the smaller statistical errors. Overall, the developed approach
provides an effective method for predicting carbon price.
©2020ElsevierB.V.Allrightsreserved.
1. Introduction
From a global perspective, climate changes and natural disasters
caused by greenhouse gas emissions are more and more serious in re-
cent years (Huang and He, 2020). This phenomenon has much to do
with the continuous economic activities of humans, but it is inevitable.
The best approach at present is the market-based policy formul ation
of pricing for carbon emissions, which can calculate the cost of emis-
sions by directly considering the economic efficiency (Sun and Wang,
2020; Song et al., 2019). Therefore, recent years have seen the trading
system of carbon emissions gaining global popularity (Wu and Liu,
2020). Looking at the international market, it admits of no doubt that
the European Union Emissions Trading System (EU-ETS) is the largest,
most liquid and most influential greenhou se gas emission reduction
mechanism in the world, which has provided an effective operating par-
adigm for carbon emissions trading around the world and accumulated
a wealth of data and experience (Zhang and Wei, 2010). Afterwards, as
one of the major contributors in reducing global carbon emissions,
China has successively established eight carbon markets in the pilot cit-
ies including Beijing (2013 ), Shan ghai (2013), Chongqing (2014),
Guangdong (2013), Tianjin (2013), Shenzhen (2013), Hubei (2014)
and Fujian (Song et al., 2018). Moreover, in 2017, the national develop-
ment and reform commission offi cially announced that China will
launch carbon trading market pilots and put this project on a vital posi-
tion in the 13th five-year plan, which indicates our firm confidence in
developing a green economy. In light of this, enhancing carbon price
forecasting precision can not only assist in policy implementation for
manufacturing processes and investment decisions, but also create an
effective and stable carbon pricing mechanism (Zhu et al., 2019). How-
ever, as a burgeoning policy-setting market created by artificial regula-
tion, the price series of carbon have the characteristics of nonstationary
and nonlinear, which are caused by internal market mechanisms as well
as external environmental uncertainties (Yang and Liang, 2017; Yang
et al., 2020). Therefore, it is a great challenge to predict carbon prices
and more researchers get involved in this research issue.
Numerous studies show that the carbon price is, to some extent, pre-
dictable only if use some certain methods, which are essentially the time-
series prediction technology. This technology is broadly divided into two
classes: statistical and economics models and artificial intelligence
models. Among them, statistical and economics models are relatively tra-
ditional, including autoregressive moving integrated moving average
(ARIMA), vector autoregression (VAR), generalized autoregressive condi-
tional heteroscedasticity (GARCH) and the like. Byun and Cho (2013) pre-
dicted the volatility of carbon price through GARCH-type models and K-
nearest neighbor model (KNN). It can be found that an improved
GARCH model obtained the most accurate result in all of the selected
models. García-Martos et al. (2013) verified that the approach of vector
autoregressive integrated with moving average (VARIMA) gave better re-
sults for very short forecasting horizons (one-day-ahead forecasts) in the
case of nonstationary and nonlinear price series. Benz and Truck (2009)
predicted the carbon spot price and volatility in the different phases of
returns with the Markov-switching method and a standard AR-GARCH
model. Although these statistical and economics methods are able to
achieve effective prediction results and possess good theoretical descrip-
tion based on statistical logic, their implementations are rooted on the lin-
ear hypothesis and mainly for very short forecasting span, which don't
have adequate ability to catch the nonlinear characteristics hidden in
the time-series. For the sake of solving this problem, artificial intelligence
models sprung up and were applied to predict time-series, which are the
data-driving method (Sun et al., 2020). Artificial intelligence models
mainly cover artificial neural network (ANN) and support vector machine
(SVM). Fan et al. (2015) applied ANN with a multi-layer perceptron
model for carbon price forecasting, which proved prediction accuracy
and fitting capacity are higher than many single and variant models.
Zhu et al. (2016) developed an ensemble paradigm of forecasting energy
price with the feature of nonstationary and nonlinear through the least
squares SVM (LSSVM) model and the final research result also showed
the superiority over the models such as ARIMA and ANN. Patel et al.
(2015) combined ANN with SVM as the hybrid model, which resulted
in the best overall prediction performance. Also, the hybrid model incor-
porating the artifici al intelligen ce model and statistical model can repre-
sent relatively better performance. Zhu and Wei (2013) then made a
combination between LSSVM and ARIMA, which proved better carbon
price prediction results compared by simple statistical models. Further-
more, for the reason that time series prediction is relative not only to
the current data but also to the data at a passed time, whose information
will be lost if only applying the latest data, the recurrent neural network
(RNN) connecting with hidden units was established for enabling the net-
work to keep the memory of recent information, which is unlike tradi-
tional ANNs (Lin et al., 1998). It has been widely used in many fields,
especially for nonstationary and nonlinear series. Rather et al. (2015)
found that according to the prediction error and correlation between
the original data and output data, RNN generates much better prediction
results. Chen et al. (2013) developed an enhanced RNN for multi-step-
ahead flood forecasting with the feature of nonstationary and nonlinear,
representing that RNN could deal with the association between the past
and future data by the memory features. Meanwhile, in the time series
forecast field, the long short-term memory (LSTM) as an enhanced special
RNN has a wide use. The “gate” mechanism is the core of the algorithm,
which can selectively filter the input information and more important his-
torical data information will be extracted for the following prediction (Li,
2020). Huang et al. (2019a) established the LSTM prediction model for
carbon emissions in China and the emulation result showed that the pre-
diction accuracy is higher compared with back propagation neural net-
work (BPNN) and Gaussian process regression (GPR). Krishan et al.
(2019) used LSTM on the field of air quality, which referred to the carbon
emission and results showed that LSTM did have a good capacity to cap-
ture complicated features in the nonstationary and nonlinear series. Liu
and Shen (2019) also established an improved LSTM model for obtaining
higher carbon price forecasting accuracy to some extent. These
extinguished results encouraged us to explore the possibilities of
predicting carbon prices in China via LSTM. Nonetheless, given a large
fluctuationtowardstheoriginalcarbonpriceseries,theeffectthrougha
single AI technique is not very outstanding.
And except for single AI technique, data preprocessing plays an im-
portant role in the time-series prediction, among which feature extrac-
tion is exactly vital. On the one hand, feature extraction can reduce the
feature numbers and dimensions to prevent the emergence of an
overfitting problem. On the other hand, it can remove irrelevant fea-
tures and let learning easier. Generally, the methods of feature extrac-
tion are broadly divided into wrapper methods and filter methods.
Wrapper methods usually cost highly and have a slow learning rate be-
cause of many times' learner training, whereas filter methods are relied
on original data and have a faster calculating rate than wrapper
methods (Wang a nd Li, 2018). Hence, considering the convenience
and speed case, researchers may prefer the later. Zhu et al. (2013) man-
ifested that the exploitation of various internal logic and necessary fea-
tures implied at different frequencies can be made possible through
decomposing methods. Wavelet transform (WT) is one of the decompo-
sition method s widely utilized in the area of prediction. Chevallier
J. Wang, X. Sun, Q. Cheng et al. Science of the Total Environment 762 (2021) 143099
2
(2011) applied the wavelet packet transforms into carbon price fore-
casting, which represented a good performance. Sun et al. (2018) also
used WT as the basis decomposing model in the research of China
Emiss ions-Trading Scheme. But these representations of effectivity
mainly depend on the wavelet basis function selected by researchers'
subjectivity without a specific theory foundation. Empirical mode de-
composition (EMD) is an adaptive method overcoming the drawback
of reliance on the subjective experience of setting a basis function previ-
ously. Zhu et al. (2017) proved that using EMD can well capture several
components with different features. Gao and Jian (2014) proposed a hy-
brid model comprising particle swarm optimization (PSO), SVM and
EMD. In the EMD part, several stationary intrinsic mode functions
(IMFs) and a residual series will be put into a neural network for train-
ing. For the sake of innovation, Gilles (2013) built a new self-adaptive
signal decomposition method named empirica l wavelet transform
(EWT) by combing EMD with WT, whose final result showed a better
performance. However, the process of decomposition through EMD is
easy to emerge modal mixing problem and its physical meaning is lack-
ing (Tian and Hao, 2020). To tackle the problem, Wu and Huang (2009)
carried out a study and improved EMD, which was named as ensemble
empirical mode decomposition (EEMD). Qin et al. (2015) utilized EEMD
as a data preprocessing method for improving the prediction effect of
the carbon price. Wu et al. (2019) also combined EEMD with LSTM to
predict the spot price of west texas intermediate crude oil. It can be
found that EEMD is an enhanced EMD, which can improve the phenom-
enon of modal mixing effectively by offsetting and restraining the ef-
fects of noises in man y times' experiences. Despite robustness and
effectiveness of forecasting based on EEMD, there is still a drawback to-
wards it. Increasing the times of integration can reduce the error of re-
construction, whereas it expands the scale of calculation and remains
residual noises to a certain amplitude. Besides, Wu and Huang (2009)
said that the problem of modal splitting may occur. To overcome this
defect, complete ensemble empirical mode decomposition (CEEMD)
as an improved method of EEMD is applied. Zhang et al. (2018) proved
that complete ensemble empirical mode decomposition with adaptive
noise (CEEMDAN) as a signal processing technology can not only solve
the modal aliasing problem, but also lessen the white noise interference
and save the computing time. What's more, Cao et al. (2019) combined
CEEMDAN with LSTM, indicating that CEEMDAN can exploit more hid-
den information than EMD and the hybrid model surpasses the single
one. Therefore, CEEMDAN can be seen as a relatively progressive de-
composition method at present and utilized by this paper for the reason
that its error of reconstruction is nearly zero by adding adaptive white
noise into each phase.
Extant studies have shown that the AI prediction models with
the feature extraction part can not only achieve the effects of data
preprocessing and improv e the calcul ating efficiency, but also es-
tablish an appropriate prediction model for the time series. But
several major drawbacks still remain. First of all, after
decomposing the carbon price series, each sub-sequence has been
put into a prediction model for the output results, which didn't
consider the similar complexity and correlation among them so
as to lower efficiency and accuracy. Secondly, the predicti on
model for each sub-sequence is the same without the realization
that each mode is different for its unique feature and frequency,
so the respectiv e establishment of mo dels with more prope r pa-
rameters is of vital importa nce (Che, 2015). Thirdly, after achieving
the prediction results of each sub-sequence, existing final ensem-
ble models mainly limit to the linear form such as obtaining the
final forecast result t hrough combining the prediction values of
all the decomposed modes (Zhu et al., 2018). For the reason that
it is not usually applicabl e for all the cases, a line ar ensemble ap-
proach may affect the accuracy of predicting (Liao and Tsao,
2006
). There are two main ty pes of nonlinear integration methods.
One o f them is serialization methods with strong dependencies
among individual learners and the o ther is paralleliz ation met hods
generated simultaneously without strong dependencies among in-
dividual learners. Representative of the former i s boost ing and the
latter is bagging, which develop the extre me gradient boosting
(XGboost) and the random forest (RF) respectively. By comparing
these two methods, we can find that the XGb oost is more se nsitive
to overfitting if the data is noisy and it is often takes longer for
being built in sequence (Fan et al., 2020). What's more, RF is
more adjustable.
In order to solve these existing problems towards carbon price
forecast, a novel hybr id model incorporat ing CEEMDAN, Sample
entropy (SE), LSTM and Random forest (RF) is put forward. From
the perspectiv e of meth odology, it develops an innov ative r andom
forest-based nonlinear ensemble paradigm of improved feature
extraction and deep learning algorithm for higher accuracy in the
case of nonst ationary and nonl inear carbon pr ice forecast. Firstly,
the original carbon price series is decomposed into several simple
stationary modes with the application of CEEMDAN algorithm.
Then, the obtained simple modes with similar co mplexity are
recombined according to the SE algorithm, so as to boost calculat-
ing efficiency and accuracy. Considering that different modes
have their own frequency and characteristic, LSTM can then be ap-
plied t o set an appropriate prediction model for each reconstructed
component because of i ts strong long and shor t term memory. At
last, after forecast results of reconstructed components have been
achieved through the deep learning algorithm, RF as a nonlinear
ensemble bagging learning model is utilized to aggregate the final
carbon price forecast result for the further improved predictio n
accuracy.
From the above, the main innovations and contributions of this re-
search compared to the findings in the literature are shown in the fol-
lowing four points:
a. Considering the neglect of similar complexity and correlation among
decomposed modes, an improved feature extraction incorporating
CEEMDAN and SE is adopted for screening different features effec-
tively from the original carbon price series so as to the higher effi-
ciency and accuracy.
b. With the realization that respective establishment of models is of
vital importance and in order to capture more complicated features,
LSTM replaces RNN as the crucial prediction model.
c. For the reason that nonlinear ensemble learning can get smaller er-
rors and more stability than a linear approach, this research applies
RF as integrated algorithm to improve the forecast accuracy.
d. The novel hybrid model for carbon price forecast setting as an adap-
tive nonlinear ensemble learning paradigm is firstly proposed,
which excels single model and represents its unique robustness.
The structure of the rest of this paper is as follows: the methodologies
and brief proposed model structure are outlined in Section 2. The case
study with data collecting, preprocessing and relative measurement indi-
ces are elaborated in Section 3. Section 4 describes the forecast results as
well as discussions in more detail. At last, Section 5 draw a conclusion.
2. Methodology
2.1. Complete ensemble empirical mode decomposition (CEEMDAN)
EMD proposed b y Huang et al. (1998) has been widely utilized in
many fields, which is an adaptive si gnal decomposition met hod
without any assumptions about data. However, the problem of
modal aliasing causes the decomposed intrinsic functions affecting
each other, which deprives the physical meaning of t he IMF. To
solve this probl em, Wu and Huang (2009) proposed EEMD, which
can offset the effects of noise during the procession of decomposition
by making several times' experiments. U nfortunately, there is resid-
ual noise in the components, which lowers efficiency. Ove rall,
J. Wang, X. Sun, Q. Cheng et al. Science of the Total Environment 762 (2021) 143099
3
剩余12页未读,继续阅读
创业青年骁哥
- 粉丝: 18
- 资源: 341
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功
评论0