This study presents two machine learning models, namely, the light gradient boosting machine (LightGBM) and categorical boosting (CatBoost), for the first time for predicting flash flood susceptibility (FFS) in the Wadi System (Hurghada, Egypt). A flood inventory map with 445 flash flood sites was produced and randomly divided into two groups for training (70%) and testing (30%). Fourteen flood controlling factors were selected and evaluated for their relative importance in flood occurrence prediction. The performance of the two models was assessed using various indexes in comparison to the common random forest (RF) method. The results show areas under the receiver operating characteristic curves (AUROC) of above 97% for all models and that LightGBM outperforms other models in terms of classification metrics and processing time. The developed FFS maps demonstrate that highly populated areas are the most susceptible to flash floods. The present study proves that the employed algorithms (LightGBM and CatBoost) can be efficiently used for FFS mapping.
The integration of solar energy in smart grids and other utilities is continuously increasing due to its economic and environmental benefits. However, the uncertainty of available solar energy creates challenges regarding the stability of the generated power the supply-demand balance’s consistency. An accurate global solar radiation (GSR) prediction model can ensure overall system reliability and power generation scheduling. This article describes a nonlinear hybrid model based on Long Short-Term Memory (LSTM) models and the Genetic Programming technique for short-term prediction of global solar radiation. The LSTMs are Recurrent Neural Network (RNN) models that are successfully used to predict time-series data. We use these models as base predictors of GSR using weather and solar radiation (SR) data. Genetic programming (GP) is an evolutionary heuristic computing technique that enables automatic search for complex solution formulas. We use the GP in a post-processing stage to combine the LSTM models’ outputs to find the best prediction of the GSR. We have examined two versions of the GP in the proposed model: a standard version and a boosted version that incorporates a local search technique. We have shown an improvement in terms of performance provided by the proposed hybrid model. We have compared its performance to stacking techniques based on machine learning for combination. The results show that the suggested method provides significant improvement in terms of performance and consistency.
1 February 2019 in Journal of Computational Science
In wireless sensor/ad hoc networks, all wireless nodes frequently flood the network channel by transmitting control messages causing “broadcast storm problem”. Thus, inspired by the physical backbone in wired networks, a Virtual Backbone (VB) in wireless sensor/ad hoc networks can help achieve efficient broadcasting. A well-known and well-researched approach for constructing virtual backbone is solving the Connected Dominating Set (CDS) problem. Furthermore, minimizing the size of the CDS is a significant research issue. We propose a new parallel scatter search algorithm with elite and featured cores for constructing a wireless sensor/ad hoc network virtual backbones based on finding minimum connected dominating sets of wireless nodes. Also, we addressed the problem of VB node/nodes failure by either deploying a previously computed VBs provided by the main pSSEF algorithm that does not contain the failed node/nodes, or by using our proposed FT-pSSEF algorithm repairing the broken VBs. Finally, as nodes in a VB incur extra load of communication and computation, this leads to faster power consumption compared to other nodes in the network. Consequently, we propose the virtual backbone scheduling algorithm SC-pSSEF which aims to find multiple VBs using the VBs provided by the pSSEF algorithm and switch between them periodically to prolong the network life time.