Establishment And Empirical Analysis Of A Contaminant Concentration Forecasting Model

In recent years,environmental pollution has caused tremendous problems to the national economy and people’s lives,and it has also become one of the global issues.Along with the continuous advancement of industrialization and urbanization,these problems are even more serious for developing countries.In order to mitigate the impact of air pollution,improve air quality,and achieve sustainable development of the economy and humankind,countries around the world have paid extensive attention to and studied the pollutants forecasting models.However,in actual operation,the main factors that cause air pollution are difficult to determine,and various factors interact with each other.The inherent volatility and intermittent nature of pollutant concentrations even greatly increase the difficulty of prediction.Therefore,it is pretty critical to determine the main factors that cause air pollution and accurately and effectively forecast the concentration of corresponding pollutants.Moreover,it is of great significance to establish a scientific and stable forecasting models and adopt corresponding air quality protection measures.Domestic and foreign scholars have proposed a large number of pollutants forecasting related research in current years and therefore pollutants forecasting accuracy has been greatly improved to some extent.Most studies used single prediction models,which are simple and easy to implement.However,they cannot achieve desireable performance and meet the needs of actual production and people’s lives.On the other hand,hybrid forecasting models based on optimization algorithms and data pre-processing methods greatly improve the pollutants prediction accuracy.However,existing models rarely focus on uncertainty forecasting,and fewer are involved in the screening of pollutant indicators.Hence,these models are liable to reduce the scientificity andaccuracy of the forecasting results,which poses a great challenge to the safety and stability of the pollutants forecasting models.Under such backgrounds,a novel hybrid pollutants forecasting model is proposed in this paper.It is based on pollutants selection,de-nosing technique,imperialist competitive algorithm and extreme learning machine,which mainly includes four parts:pollutants attribute selection,data pre-processing and reconstruction,deterministic forecasting and andinterval forecasting.The main contents can be listed as follows:Firstly,a fuzzy rough set theory was used to determine major pollutants for each city.Then,an ICEEMDAN decomposition method is applied to decompose the original pollutants time series,eliminate redundant noise and extract the primary characteristics in the data preprocessing.Moreover,ICA is used to optimize the weights and thresholds ofextreme learning machine.Finally,in the deterministic and interval forecasting part,a novel hybrid forecasting model ICEEMDAN-ICA-ELM is developed in this paper.Owing to measure the effectiveness and generalization ability of the proposed model,an evaluation system includes a hypothesis test,nineevaluation criteria,five experiments,and six different study cities are introduced to perform comprehensive evaluation.The simulation results show that the proposed prediction model can not only possess strong generalization ability,but also greatly improve the prediction performance with better and more stable forecasting results,in comparison with the other six forecasting models(i.e.,ARIMA,BP,GRNN,PSO-ELM,ICA-ELM,EMD-ICA-ELM).The proposed hybrid pollutants forecasting model in this paper can not only effectively reduce the forecasting error,largely improve the accuracy,and enhance the stability of prediction results,but also benefit the improvement and development of air pollution forecasting and monitoring,which provides guidance for relevant decision makers and people’s daily lives.The main innovations of this study can be illustrated as follows:First,the application of fuzzy rough set theory in polluting indicators selection is successfully developed in this paper;then the ICEEMDAN decomposition method decomposes,de-noises,and reconstructs pollutants time series to fully exploit the inherent information characteristics.Furthermore,the ICA provides a novel viable option for solving optimization problems.The proposed hybrid forecasting model ICEEMDAN-ICA-ELM is used for deterministic and uncertain forecasting,which provides comprehensive forecasting information.Finally,the experimental results indicate that the developed forecasting model in this paper is superior to the other six forecasting models used for comparison.