|
Comparison of MLP and RBF Neural Networks in Estimating the Average Relative Humidity Case Study: South Khorasan
|
Mehdi Asadi *   |
| Farhangian University |
|
|
Abstract: (21 Views) |
Abstract
Relative humidity is one of the environmental factors affecting the growth and spread of plant diseases and is of particular importance due to its direct relationship with rainfall. Therefore, this research aim is to estimate the average relative humidity in South Khorasan province as a dry region using MLP and RBF neural networks. The study utilized RStudio software to analyze the following variables over 23 years (2000-2023): average relative humidity, maximum relative humidity, minimum relative humidity, average air temperature, maximum air temperature, minimum air temperature, average wind speed, and total hours of sunshine, each in six different scenarios. The results showed that the correlation value (R) between the observed and predicted data in both methods is equal to 0.98; the value R2 in the MLP method is equal to 0.953, and in the RBF, method is equal to 0.965. The RMSE for the MLP method is equal to 0.845, and the RBF method is equal to 0.740. The MSE for the MLP method is equal to 0.714, and for the RBF method, it is equal to 0.548. In total, the MLP and RBF networks have relatively good accuracy in estimating the average annual relative humidity. Still, the error of the RBF network (0.561) is less than that of the MLP network (0.707) and performs better than the MLP in estimating relative humidity.
Introduction
Recently, many scientists and researchers have attempted to prove that fluctuations and changes in the climate have occurred by examining some climate parameters. Accordingly, one of the challenges of the 21st century is the issue of fluctuations and climate change. The main reasons for these fluctuations are the increase in incoming energy from the sun and global warming due to the intensification of greenhouse effects. These changes, in turn, cause hydrological anomalies, such as droughts and floods, and affect the environment and human life. Among the meteorological variables, compared to temperature and precipitation, little research has been done in the field of investigating the temporal changes of relative humidity, while it has a direct effect on the amount of visibility, cloud formation, fog, and smog. In addition, relative humidity is an environmental factor affecting plant growth and the spread of plant diseases, and it is of particular importance due to its direct relationship with rainfall.
Study area
South Khorasan province, with an area of 82,864 square kilometers, is in the east of Iran and is located between the geographical coordinates of 57 degrees and 46 minutes to 60 degrees and 57 minutes of east longitude and 30 degrees and 35 minutes to 34 degrees and 14 minutes of north latitude. The mentioned province shares borders with Afghanistan to the east, Yazd, Isfahan, and Semnan provinces to the west, Khorasan Razavi province to the north, and Sistan-Baluchistan and Kerman provinces to the south. Due to its low latitude and closeness to the orbit of the Tropic of Cancer, and due to the atmosphere's general circulation during the warm period, it is located under the influence of a tropical high-pressure center and has a winter rainfall regime. The average rainfall of the region is about 180 mm per year, and the annual average relative humidity is 36%.
Materials and Methods
This research utilized meteorological statistics (average, maximum, and minimum relative humidity; average, maximum, and minimum air temperature; average wind speed; total sunny hours) from synoptic stations in South Khorasan province (Birjand, Qain, Beshroieh, Ferdous, Nehbandan, Khor Birjand) as annual averages over 23 years from 2000 to 2023. This research utilized MLP and RBF neural networks, leveraging RStudio's Neuralnet and Kernlab libraries to model the annual average relative humidity in South Khorasan province. The model divides the data into three parts: training, validation, and test sets, using six stations from 2000 to 2023: Birjand, Qain, Beshroieh, Ferdous, Nehbandan, and Khor Birjand. From 138 data points, 104 (75%) annual data points were collected for network training, 21 (15%) data points yearly for validation, and 13 (10%) yearly data points were used as test data.
Results and discussion
Therefore, in the MLP method, the highest degree of correlation between the observed and predicted data in the network training stage is associated with variance (0.0364), standard deviation (0.0069), and RMSE (0.2803) related to the first scenario with a value of 967.0, and the lowest correlation with variance (0.0514), standard deviation (-0.0009), and RMSE (0.8456) corresponding to the sixth scenario with a value of 0.872. Also, the best structure of the designed networks is related to the first scenario in a network with a layer of 14 hidden neurons and the lowest error rate (0.0007) in the training phase, and a layer of 9 hidden neurons and the lowest error rate (0.0011) in the test phase.
Conclusion
In this study, RBF and MLP neural networks predict the annual relative humidity of stations in South Khorasan province, including Birjand, Qain, Beshroieh, Ferdous, Nehbandan, and Khor Birjand. Considering the relative humidity values obtained from the Meteorological Department of South Khorasan Province as the target values in the training of neural networks, the performance of six scenarios involving different networks was investigated. The investigations determined that the models mentioned can reliably estimate relative humidity. Based on this, the highest measured and estimated relative humidity is 40.97% for the Meteorological Organization, 41.53% for the MLP method, and 40.88% for the RBF method, which is the highest value measured in 2023, and the highest estimated value occurred in 2000. The lowest measured and estimated relative humidity values are as follows: 28.25% for the Meteorological Organization, 29.22% for the MLP method, and 28.57% for the RBF method. These values represent the lowest measured value in 2015 and the highest estimated values for the MLP method in 2011 and for the RBF method in 2015.
|
Article number: 1 |
| Keywords: Neural Network, MLP, RBF, Relative humidity, South Khorasan. |
|
|
Full-Text [PDF 870 kb]
(14 Downloads)
|
Type of Study: Research |
Subject:
Special
Received: 2024/09/2
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
|
|
