عنوان مقاله [English]
Proper temperature forecasting is of significant importance in adapting to climate change at local scales. For this purpose, in this research, feature selection is done using principal component analysis algorithm, then in the post-processing stage of the neural network, the group method of data modeling is improved using the water strider algorithm, so that the temperature prediction can be done optimally. In order to compare, the results show a decrease in mean square error of 0.0469 in the proposed method compared to the feature selection method using mlp.
Keywords: temperature prediction, neural network, data modeling group method, water strider algorithm.
Forecasting temperature fluctuations plays a vital role in studying future climate patterns. Weather forecasting has always been a big challenge for meteorologists to predict the future weather and weather conditions . In their article, Gupta et al. (2022) discussed a model based on multiple linear regression to predict the average temperature of a day. Conclusions are made according to the performance of the evaluated model. This model can successfully predict the average temperature of a day with an error of 2.8 degrees Celsius . Fona et al. (2021) in their paper estimated rainfall based on remote sensing using massive data with the help of integrated routing framework. The results show proper evaluation and error reduction for rainfall forecasting in the proposed BIRSM model and artificial neural network.[3 ].
2- Water strider algorithm
In this water strider algorithm, the food sources required are females. Males are the preferred mating sources. Males send calling signals and females respond to them in the form of attraction or repulsion signals.
Step 1: The first step of early birth
Step 2: Territory Building: Aquatic insects maintain territories to survive, mate, and feed.
The third step of pairing:
Step 3: nutrition:
Step 4: death and success:
3- Analysis of main components
This algorithm is used to reduce the dimension
4- Group method of data modeling
In general, to model complex systems that include a set of data with several inputs and one output, formula (6).
The GMDH neural network is created through a quadratic polynomial of formula (7) and the mean square error of formula (8):
The partial derivative of the matrix equation (Aa=y) is obtained.
5- multilayer perceptron neural network
This network includes an input layer, an intermediate layer and an output layer. In our problem, the best solution is usually obtained with the number of 10 neurons in the hidden layer.
6- Proposal method:
In this research, the database includes the data of the Aqdasiye weather station, it includes 2721 samples with 23 characteristics, which are related to the last 8 years. Using the principal component analysis algorithm, 6 features have been selected from among the 23 available features. The following formulas have been used to evaluate the models.
In formula (14), the mean square error function as the objective function,
Initial preparation stage: In this stage, a population of water strider is created randomly. The position of each aquatic insect includes four components: the maximum number of layers, the maximum number of neurons in each layer, and the pressure parameter selected according to table (2).
Repetition stage: In this stage, the above operations are repeated.
Table (3) shows the position of the male water beetle before and after mating, and the flowchart of the proposed method is shown in Figure (1).
7- Simulation results
7-1- The results of temperature prediction with the proposed method
In the proposed method, the parameters of the population size of water strider are considered equal to 50 and the maximum number of repetitions is equal to 1000, and the results are based on the types of errors according to table (4).
Figure (2) and Figure (3) show error and regression histograms for training and test data, respectively
7-2- The results of temperature prediction with multilayer perceptron neural network trained with water strider algorithm
The number of neurons in the hidden layer is equal to 5, the results are based on the types of errors according to table (5).
Figures (4) and (5) show the histogram of error and regression for training data, test data and total data, respectively.
According to table (6) and figure (6), the neural network of the proposed method has a better performance in temperature prediction in terms of error types for the training, experimental and total data compared to the multilayer perceptron neural network trained with The algorithm has water strider.
This article is based on temperature forecasting using data mining methods. According to tables (4) and (5), the neural network of the improved data modeling group method with the water strider algorithm and the multilayer perceptron neural network trained with the algorithm water strider predicted the temperature with mean square error of 0.0469 and 0.0529, respectively. Therefore, the best performance belongs to the neural network of the proposed method.
. R. Palamuttam, R. Palamuttam, R. M. Mogrovejo, C. Mattmann, B. Wilson, K. Whitehall, R. Verma, L. McGibbney and P. Ramirez, "SciSpark: Applying in-memory distributed computing to weather event detection and tracking," in 2015 IEEE international conference on Big data (Big data), pp. 2020-2026: IEEE, 2015.
. I. Gupta, H. Mittal, D. Rikhari, and A. K. Singh, "MLRM: A Multiple Linear Regression based Model for Average Temperature Prediction of A Day," arXiv preprint arXiv:2203.05835, 2022.
. J. Refonaa and M. Lakshmi, "Remote sensing based rain fall prediction using big data assisted integrated routing framework," Journal of Ambient Intelligence and Humanized Computing, pp. 1-11, 2021.