preprocessing

Simulating the flow in order for managing the water allocation in drought and wet periods is of great importance. According to the researches conducted during several decades in this regard, computational intelligence methods combined with wavelet are known to be effective. In this paper, Wavelet-Principal Component Analysis-Random Forest (WPCARF) hybrid approach is suggested to model daily flow of Polroud river. In the suggested model, first, hydrometric data is processed by wavelet transform and applied to the PCA algorithm along with meteorological data. Afterward, their output vectors were entered into the random forest network. The results have shown that the PCA algorithm can improve the performance accuracy and speed of the model, despite reducing the input vectors and simplifying them. Also, it can integrate a model with increased simulation time and input vectors uncertainty having a lower impact on model capability leading to a more uniform decreasing trend. Furthermore, preprocessing the data accompanied by PCA could enhance the agreement index by 5 and 8 percent during one and three days of the simulation and increase the model ability for more accurate simulation of river flow. On the other hand, results for the best proposed hybrid model during the one-day simulation time were R=0.911 and RMSE=7.095m3/S, while these values were R=0.817 and RMSE=8.681 m3/S in the best hybrid model for three-day simulation time. This indicates the adequate capacity of the suggested combined model for long-term simulation times.

Waste generation in today industries is a serious problem. Waste generation from the production stage to the final disposal is an inevitable issue. Development of the cities and the industrialization causes the everyday increasing in solid waste generation. Therefore, knowing the waste values is an essential tool for solid waste management systems. In this research, artificial neural network is used as a financial tool for modeling solid waste generation in Mashhad. For this purpose, first, some pre-processing on the dependent and independent variables are done and the effect of this procedure on the accuracy of the model is investigated. Research findings clearly indicate that by using some preprocessing on the input data accurate results can be obtained. Three different conditions have been evaluated and the best one is selected which contains logarithm, trend removing and standardizing. The selected network has two hidden layers with five neurons in each one. Network performance parameters are MAPE, MSE and R2 that equals to 0.06, 0.46 and 0.86 respectively.