Introduction: Today, dust phenomena are among the most important environmental hazards and pose a serious threat to human health and the environment. Dust in barley as one of the pollutants has various adverse effects and negative consequences, among which can be reduced growth and yield of agricultural products, intensification of damage caused by pests and plant diseases, increased road accidents due to reduced visibility, The cancellation of flights and the resulting financial losses, increased treatment costs, closure of industrial units, pollution of water resources, increased erosion of buildings, decreased efficiency of solar photovoltaic systems due to turbidity.
Objective: Therefore, due to the importance of dust and in order to predict how dust is spread, the artificial neural network model was used. This model can be useful and cost-effective information for future implementation of air pollution control strategies and cost reduction.
Materials and Methods: To model the dust distribution using artificial neural network model, statistics and meteorological information of Kashan synoptic station, which were recorded daily by the Environment Department in 1996, were used. The proposed neural network model has four input layers that include humidity, temperature, wind speed, wind direction and an output layer, the daily concentration of suspended particles is 2.5 micrometers per cubic meter. The model training process was performed using multilayer perceptron neural network and post-diffusion rule and using sigmoid membership function in Matleb software environment. In the neural network model, the number of neurons in the hidden layer and the appropriate number of rounds or IPAC to achieve the best neural network structure, with the least error for each model, were determined using trial and error. The number of neurons and apex for the model in 2017 is 15 and 37,000, respectively.
Results and discussion:
The correlation coefficient of the model for predicting PM2.5 concentration is equal to 0.80 which is obtained by comparing real data with simulated data. The validation results of the model with real data are close to 80%, so the neural network model can be used to predict PM2.5 concentration. According to the average regression diagram, the predicted values obtained from the model are closer to the diagonal axis and have no dispersion. Also, based on the results of the step-by-step regression method, it was determined that among the four variables used for relative humidity modeling, it has the most impact and importance in dust emission modeling.
Conclusion: According to the accuracy and the results, this method can be used to predict the air pollution of Kashan caused by suspended particles. Due to the high capability of the perceptron neural network in predicting the concentration and distribution of dust, the application of this model can be a suitable and fast solution for predicting the amount and spread of dust.