Search

A selective ensemble hybrid modeling prediction method based on wavelet transformation is proposed to improve the fitting and generalization capability of the existing prediction models of the coal face gas concentration, which has a strong stochastic volatility. Mallat algorithm was employed for the multi-scale decomposition and single-scale reconstruction of the gas concentration time series. Then, it predicted every subsequence by sparsely weighted multi unstable ELM(extreme learning machine) predictor within method SERELM(sparse ensemble regressors of ELM). At last, it superimposed the predicted values of these models to obtain the predicted values of the original sequence. The proposed method takes advantage of characteristics of multi scale analysis of wavelet transformation, accuracy and fast characteristics of ELM prediction and the generalization ability of L1 regularized selective ensemble learning method. The results show that the forecast accuracy has large increase by using the proposed method. The average relative error is 0.65%, the maximum relative error is 4.16% and the probability of relative error less than 1% reaches 0.785.

Literature on the design of efficient nonfouling membranes by in-situ modification is poor, which can be explained by the difficulty to control membrane formation mechanisms when a third material is added to the casting solution, or by the lack of stability of matrix polymers with surface-modifiers. We present polyvinylidene fluoride membranes formed by vapor-induced phase separation and modified with a tri-block copolymer of poly(styrene) and poly(ethylene glycol) methacrylate moieties (PEGMA 124 - b -PS 54 - b -PEGMA 124 ). After characterizing the copolymer, we move onto membrane formation mechanisms. Membrane formation is well controlled and leads to structure close to bi-continuous. Considering the formulation chosen, PVDF/PEGMA 124 - b -PS 54 - b -PEGMA 124 solutions are less viscous and more hydrophilic than virgin PVDF solutions. Both effects promote non-solvent transfer, thus decreasing the chances for crystallization. Hydrophilic capability of membranes is increased from about 59 mg/cm 3 to 650 mg/cm 3 , leading to a severe drop of non-specific protein adsorption, up to 85–90%, also depending on its nature. Biofouling at the micro-scale by modified Escherichia coli and Streptococcus mutans is almost totally inhibited. Finally, biofouling is importantly reduced in dynamic conditions, as measured from the water flux recovery ratio of 69.4%, after 3 water-BSA filtration cycles, much higher than with a commercial hydrophilic PVDF membrane (47.3%). These membranes hold promise as novel materials for water-treatment or blood filtration. [ABSTRACT FROM AUTHOR]