بهینه سازی شبکه عصبی کانولوشن با الگوریتم حشره آبسوار در پیش بینی وضعیت آب و هوایی

Abstract:

One of the most important and practical issues in today's world is weather forecasting. Forecasting the weather condition reduces the losses and damages of critical weather conditions. Weather forecasting can be effective in applications such as agriculture and air transportation. One of the methods that can be used to learn and predict the weather is deep learning networks, including convolutional neural networks. One of the important challenges of the convolutional neural network is that it performs feature selection unintelligent by using a number of convolution operations. In this article, a collective intelligence method is presented to improve the accuracy of forecasting weather conditions by convolutional neural network. The evaluations using the data sets related to weather conditions show that the proposed method, which has an accuracy and sensitivity of 96.32% and 96.14%, respectively, compared to the CNN deep learning network, has been able to achieve prediction accuracy of about increase by 8.35%.

Keywords: weather forecasting, convolutional neural network, feature selection, water strider algorithm

1 .Introduction

Weather conditions are an interesting and complex subject and very variable. With the change of weather and the occurrence of storms, floods, etc. Currently, there is no effective way to prevent the consequences of damaging weather conditions, so prevention should be prioritized. Today, machine learning and deep learning are considered as two subsets of artificial intelligence systems and are widely used in weather forecasting. One of the practical methods for predicting weather conditions is using the deep learning technique of convolutional neural network. This method also has challenges, one of which is unintelligent feature selection. Another challenge of this method is that the weights used in this network are not necessarily chosen optimally. To solve these challenges and to increase the accuracy of the convolutional neural network in this article, collective intelligence is used to reduce the prediction error.

2 .Research background

In the field of predicting weather variables using neural networks, some researches have been conducted and we will discuss some of them. As an example, in research [1], they presented a statistical and machine learning method for weather forecasting. The results showed that by combining automatic feature engineering with machine learning approaches, the prediction accuracy can be increased. In research [2], they presented a neural network for effective weather forecasting using time series data from the local weather station. The experimental results of the tests show that the proposed model using convolutional neural network provides better prediction compared to LSTM artificial neural network and other classical machine learning methods.





3 .Suggested method

In this section, a proposed method for improving the convolutional neural network using group intelligence of water-borne insects is presented and introduced.

3-1 .Convolution neural network

Convolutional neural networks are a class of deep neural networks. CNNs are very different from other pattern recognition algorithms because CNNs combine feature extraction and classification. Figure (1) shows a simple example of a basic CNN.



3-2- Water strider algorithm

One of the interesting behavior of collective intelligence that has been presented recently is the behavior of water bug algorithm. In this algorithm, behaviors such as foraging and mating are used to find the optimal solution and search the problem space. which is shown in figure (2).



4- Analysis of the proposed method

In the proposed method for analysis, it is evaluated in the MATLAB implementation environment. In the diagram of Figure (4), the value of the fitness function is displayed according to the repetition of the proposed algorithm. The evaluations show that the value of the fitness function is decreasing according to the repetition of the proposed algorithm in forecasting the weather condition.

To evaluate the proposed method based on accuracy and sensitivity indicators in predicting the weather condition from counting correct positive samples (TP), false positive samples (FP), correct negative samples (TN) and samples false negative (FN) is used.



The evaluations show that the proposed method has been able to increase the prediction accuracy by about 8.35% compared to the CNN deep learning network. Here, to complete the experiments, the proposed method is compared with the results of the article [4], which was presented in 2021. In table (1), the MAE and MSE index of the proposed method is compared with other methods.



According to the conducted experiments, the MAE and MSE index in predicting the weather condition, the error of the proposed method in predicting the weather condition is less than other methods.

4- Conclusion

Today, weather forecasting is an important research topic and is used for applications such as agriculture and travel planning. One of the methods of forecasting the weather condition is the use of machine learning and deep learning methods such as convolutional neural network. In this article, at first, the selection of features is done with the water strider algorithm, and then the important features related to the weather conditions are delivered to the convolutional neural network so that the weather conditions can be predicted.

The evaluations show that the proposed method is more accurate in terms of accuracy and sensitivity index for predicting weather conditions than artificial neural network, decision tree, logistic regression and convolutional neural network.

Reference:

1. Ioannou, K., Karampatzakis, D., Amanatidis, P., Aggelopoulos, V., & Karmiris, I. (2021). Low-Cost Automatic Weather Stations in the Internet of Things. Information, 12(4), 146.

2. Wolff, S., O'Donncha, F., & Chen, B. (2020). Statistical and machine learning ensemble modelling to forecast sea surface temperature. Journal of Marine Systems, 208, 103347.

3. Hewage, P., Behera, A., Trovati, M., Pereira, E., Ghahremani, M., Palmieri, F., & Liu, Y. (2020). Temporal convolutional neural (TCN) network for an effective weather forecasting using time-series data from the local weather station. Soft Computing, 24(21), 16453-16482.

4. Bączkiewicz, A., Wątróbski, J., Sałabun, W., & Kołodziejczyk, J. (2021). An ANN Model Trained on Regional Data in the Prediction of Particular Weather Conditions. Applied Sciences, 11(11), 4757.