Application of Machine Learning Method in Jinan Temperature
Prediction
Mingzhao Liu
and Zhongmu Li
*
Institute of Astronomy, Dali University, Dali 671003, China
Keywords: Machine Learning, LSTM, Temperature Prediction.
Abstract: Machine learning algorithms based on big data have been widely used in many fields, such as healthcare,
education, manufacturing, and finance. It can extract features from existing data, learn the changing laws of
existing data, and then judge and predict new data. It is practical to apply machine learning to temperature
prediction. This paper uses a machine learning algorithm and Jinan temperature data to build a prediction
model, and compares the predicted with the actual value. The prediction accuracy of the model is 90.1%. The
results show that the model can predict the daily maximum temperature in Jinan, and the machine learning
method has good applicability in the field of temperature prediction.
1 INTRODUCTION
With the increasing severity of global warming,
people gradually realize the importance of
temperature in daily production and life. Extreme
high or low temperatures will affect the safety of
people's lives and property. In particular, persistent
heat has melted Arctic glaciers and raised sea levels,
causing irreversible damage to coastal cities. In recent
years, the frequency of extreme temperature
phenomenon in Jinan is getting higher and higher. As
the provincial capital, Jinan has a large population
density, and the damage caused by extreme high or
low temperature is particularly obvious. To better
cope with temperature changes, it is necessary to
predict the future temperature. Since ancient times,
people have been looking for ways to predict
temperature, hoping to predict temperature to a
certain extent through efforts. Temperature prediction
belongs to a climate factor prediction in climate
prediction. At present, there are two main methods for
climate prediction, namely statistical and dynamic
methods. Statistical method refers to the use of
statistical methods to analyze the linear relationship
between various prediction factors. Common
statistical methods include multiple linear regression
(Cannon and McKendry, 2002; Mekanik et al., 2013),
singular spectrum analysis (Chau and Wu, 2010),
singular value decomposition (Yun et al., 2002;
Fattorini and Brandini, 2002) and grey prediction
model (Kung et al., 2003), etc. Dynamic method
refers to a numerical model that adds initial
conditions to partial differential equations based on
physical laws, and solves to obtain future climate
change. Although, statistical methods and dynamic
methods are the two main methods of climate
prediction, the shortcomings of these two methods are
very obvious. The statistical method lacks the
research on the physical mechanism, and the
prediction ability of extreme values is insufficient.
The dynamic method is sensitive to the initial value
and cannot make full use of historical data.
In recent years, artificial intelligence has developed
rapidly. Machine learning (Sammut and Webb, 2010)
and deep learning (Lu et al., 2014) technology are the
core content of artificial intelligence (Jordan and
Mitchell, 2015), and they are widely used in various
fields. The development of artificial intelligence
promotes the generation of prediction methods based
on data-driven models (Lu et al., 2014). Machine
learning can automatically learn the patterns of
changes in existing data and apply these patterns to
new data. The training of machine learning model can
not be separated from a large number of data. The
accuracy of the model prediction depends on the
amount of data. The four basic characteristics of big
data are the amount of data, the speed of data update,
the diversity of data, and the accuracy of data. The
meteorological data has already these four
characteristics (Markus et al., 2019). Therefore,
machine learning is very suitable for climate
prediction. Common machine learning algorithms
252
Liu, M. and Li, Z.
Application of Machine Learning Method in Jinan Temperature Prediction.
DOI: 10.5220/0011920100003612
In Proceedings of the 3rd International Symposium on Automation, Information and Computing (ISAIC 2022), pages 252-255
ISBN: 978-989-758-622-4; ISSN: 2975-9463
Copyright
c
2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)
include random forest algorithm (Simone et al.,
2011), artificial neural network (Agatonovic-Kustrin
et al., 2000), support vector machine (Tripathi et al.,
2006), logistic regression (Menard, 2004) and long
short-term memory network, etc. In this paper, we use
a representative method in machine learning, long
short-term memory network. The network and Jinan
climate data are used to build the model. The model
is then used to predict the daily maximum
temperature in Jinan. We calculate the error between
the predicted and the real value to evaluate the model
performance.
2 DATA AND MODELS
2.1 Data Sources
The data used in this article comes from the daily
observation data of China's surface meteorological
stations of the "National Meteorological Science Data
Center". We selected the daily maximum temperature
data of Jinan station from year 1951 to 2019 for
research. The temperature data characteristics are
shown in Figure 1. Temperature data is a typical time
series data with obvious periodic attribute. We input
it to the prediction model, and the model
automatically extracts data features and makes
predictions.
Figure 1: Temperature change graph
2.2 Data Processing
We use the info function in Python to view the
relevant information of the sample data. Then we call
the MinMaxScaler function in the
sklearn.preprocessing package to normalize the
sample data and transform the sample data to be
between (0, 1). The data are selected from year 1951
to 2012 as the training set, with a total of 22,639
pieces of data; the data from year 2013 to 2019 are
selected as the test set, with a total of 2,556 pieces of
data. Missing data is not much in the selected data,
which has little impact on model training. Therefore,
we will directly delete the missing data. After the data
are divided, the training data are directly passed to the
model, and the model is predicted after training. The
predicted value is compared with the test data, and the
Explained Variance Score (EVS), Mean Absolute
Error (MAE) and Mean Squared Error (MSE) are
calculated.
2.3 Long Short-Term Memory
Principle
The long short-term memory network model is a
special form of the recurrent neural network.
Recurrent neural network is a kind of neural network
with memory ability, and its main purpose is to
process sequence data. The innovation of recurrent
neural network is to store the past information in the
memory unit and produce output after the interaction
with the current input. However, when the amount of
data increases and the data has long-term dependence,
the recurrent neural network will have the problem of
gradient explosion and gradient disappearance. To
solve this problem, long short-term memory network
was created. The long short-term memory network
effectively makes up for the shortcomings of the
recurrent neural network by introducing a special
gating mechanism. It was first developed and applied
by two scientists, Schmidhuber and Hochreiter, in
1997, and has received much attention in recent years
and has been widely used in other works. The storage
of the long short-term memory network is called the
gating unit. The function of the gating unit is like a
"gate", controlling the retention or discarding of data
and allowing information to pass through selectively.
The long short-term memory network has three gated
units, i.e., forget gate, input gate and output gate. The
three gates respectively control the output of various
information. The update process is from Formula (1)
to Formula (6).
G
sigmoi
d
w
H
,X
(1)
G
sigmoi
d
w
H
,X
(2)
S
tanhw
H
,X
(3)
C
G
C
G
S
(4)
G
sigmoi
d
w
H
,X
(5)
H
G
∗tanhC
)
(6)
Application of Machine Learning Method in Jinan Temperature Prediction
253
3 MODEL PREDICTION AND
RESULT ANALY SIS
3.1 Experimental Platform
Construction
This research uses Anaconda to build a Python3.8
environment, and uses the Tensorflow framework
developed by Google to complete the construction of
the entire model. The training process of machine
learning model is completed on the Jupyter notebook
provided by Anaconda. Anaconda is a software for
installing and managing Python-related packages. It
can be used to install different versions of packages
and their dependencies on the same machine and
switch between different environments. Tensorflow
is a deep learning framework released by Google in
2015, and released version 2.0 in 2019 to make up for
the defects.
3.2 Construction of Prediction Model
We use the long short-term memory network and the
highest temperature data of Jinan from January 1,
1951 to December 31, 2019 to model. The network
model constructed in this experiment consists of three
LSTM layers, three Dropout layers and a Dense layer,
where the nodes of the LSTM layer are 64, 50 and 50
respectively. The batch_size and the number of
training epochs are set to 32 and 30. The optimization
algorithm of the model uses Adam. We use the
trained model to predict the daily maximum
temperature in Jinan, and analyze the EVS, MAE and
MSE of the predicted value and the real value. The
results show that the model can predict the daily
maximum temperature in Jinan with high accuracy.
3.3 Evaluation Standard
The purpose of prediction is to predict the future time.
To analyze whether the prediction results are correct,
the model needs to be evaluated after training. Model
evaluation is an integral part of the model
development process. It helps to discover the best
model for expressing data and how well the selected
model will work in the future. We need to develop
evaluation criteria to evaluate the prediction effect of
the model. These criteria can explain the overall
prediction performance of the prediction model and
compare the advantages and disadvantages of the
prediction methods. In this paper, the training set data
is used for model construction and training, The
training set data is not involved in model evaluation.
The test set data is not involved in model building, it
can be used to evaluate the accuracy of model
predictions. This study needs to evaluate the accuracy
of the temperature prediction results of the machine
learning method, the selected evaluation criteria are
Explained Variance Score (EVS), Mean Absolute
Error (MAE) and Mean Squared Error (MSE). The
calculation formulas of these evaluation criteria are
shown in Formula (8) to Formula (10).
3.4 Result Analysis
The model was trained by training set data to predict
the daily maximum temperature in Jinan City. We
compare the prediction results with the test set data.
The fitting results are shown in Figure 2.
Figure 2: Comparison of predicted and actual results
It can be seen from the Figure 3 and Figure 4 that
the fitting of the model is good. MSE and MAE
decreased with the increase of training times.
Figure 3: Mean Squared Error versus number of training
sessions
EVS 1
var
y
y
^
var
y
(8)
MAE
1
n
|
y
y
^|
(9)
MSE
1
n
y
y
^
(10)
ISAIC 2022 - International Symposium on Automation, Information and Computing
254
Figure 4: Mean Absolute Error versus number of training
sessions
4 CONCLUSION
This paper uses a machine learning method to predict
the daily maximum temperature in Jinan City. Via the
comparison between the predicted and the actual
value, it is found that the long short-term memory
network has a good performance in fitting the
prediction of the daily maximum temperature of Jinan
City. The accuracy rate reaches 90%, and the values
of MSE and MAE are 0.0049 and 0.0565. Therefore,
long short-term memory has good prospects in
temperature prediction and climate prediction, and it
is a trend to apply machine learning methods to
climate prediction. Temperature is characterized by
randomness and uncertainty, and many climatic
factors affect temperature. Therefore, the follow-up
research can consider the impact of more climate
factors on the temperature, and then better predict the
temperature.
ACKNOWLEDGEMENTS
This work is supported by the Yunnan Academician
Workstation of Wang Jingxiu (202005AF150025)，
National Natural Science Foundation of China (No.
11863002), and Sino-German Cooperation Project
(No. GZ 1284).
REFERENCES
Agatonovic-Kustrin, S., and Beresford, R. (2000). Basic
concepts of artificial neural network (ann) modeling
and its application in pharmaceutical research. Journal
of Pharmaceutical & Biomedical Analysis, 22(5), 717-
727.
Chau, K., Wu, C. (2010). A hybrid model coupled with
singular spectrum analysis for daily rainfall prediction.
Journal of Hydroinformatics, 12(4), 458-473.
Cannon, A. J., and Mckendry, I. G. (2010). A graphical
sensitivity analysis for statistical climate models:
application to indian monsoon rainfall prediction by
artificial neural networks and multiple linear regression
models. International Journal of Climatology, 22.
Fattorini, M., and Brandini, C. (2020). Observation
Strategies Based on Singular Value Decomposition for
Ocean Analysis and Forecast. Water, 12(12): 3445.
Jordan, M. I., and Mitchell, T. M. (2015). machine learning:
trends, perspectives, and prospects. Science,
349(6245):255-260.
Kung, C. Y., Kung, C. J., and Tsai, S. Y. (2003). Study of
computer game forecasting in Taiwan market
application of grey prediction model.
Lu, C., Lei, Y., Singh, V., et al. (2014). Determination of
input for artificial neural networks for flood forecasting
using the copula entropy method. Journal of
Hydrologic Engineering, 19(11), 217-226.
Lu, C., Lei, Y., Singh, V., et al. (2014). Determination of
input for artificial neural networks for flood forecasting
using the copula entropy method. Journal of
Hydrologic Engineering, 19(11), 217-226.
Markus, et al. (2019). Deep learning and process
understanding for data-driven earth system science.
Nature.
Menard, S. (2004). Logistic regression. American
Statistician, 58(4), 364.
Mekanik, F., Imteaz, M. A., Gato-Trinidad, S., and
Elmahdi, A. (2013). Multiple regression and artificial
neural network for long-term rainfall forecasting using
large scale climate modes. Journal of Hydrology,
503(503), 11-21.
Sammut, C., and Webb, G. I. (2010). Machine learning.
Kluwer Academic Publishers, 10.1007/978-0-387-
30164-8(Chapter 89), 139-139.
Simone, V., Matteo, Z., et al. (2011). Application of a
random forest algorithm to predict spatial distribution
of the potential yield of ruditapes philippinarum in the
venice lagoon, italy. Ecological Modelling.
Tripathi, S., Srinivas, V. V., and Nanjundiah, R.S. (2006).
Downscaling of precipitation for climate change
scenarios: a support vector machine approach. Journal
of Hydrology, 330(3-4), 621-640.
Yun, W. T., Stefanova, L., and Krishnamurti, T. N. (2003).
Improvement of the multimodel superensemble
technique for seasonal forecasts. Journal of Climate,
16(22): 3834-3840.
Application of Machine Learning Method in Jinan Temperature Prediction
255
