Search

Whereas adversarial training is employed as the main defence strategy against specific adversarial samples, it has limited generalization capability and incurs excessive time complexity. In this paper, we propose an attack-agnostic defence framework to enhance the intrinsic robustness of neural networks, without jeopardizing the ability of generalizing clean samples. Our Feature Pyramid Decoder (FPD) framework applies to all block-based convolutional neural networks (CNNs). It implants denoising and image restoration modules into a targeted CNN, and it also constraints the Lipschitz constant of the classification layer. Moreover, we propose a two-phase strategy to train the FPD-enhanced CNN, utilizing $\epsilon$-neighbourhood noisy images with multi-task and self-supervised learning. Evaluated against a variety of white-box and black-box attacks, we demonstrate that FPD-enhanced CNNs gain sufficient robustness against general adversarial samples on MNIST, SVHN and CALTECH. In addition, if we further conduct adversarial training, the FPD-enhanced CNNs perform better than their non-enhanced versions.

Superior iron-based Fischer-Tropsch synthesis (FTS) catalysts (Fe/NMCs) were developed by impregnating high amount of iron (40 wt%) over nitrogen-rich mesoporous carbons (NMCs) with high porosities. The large pore volumes and specific surface areas of NMCs realized the high loadings and proper dispersions of iron while the nitrogen-containing groups enhanced the basicities of the catalysts. Weak metal-support interactions were observed in Fe/NMCs, which improved the reduction, carburization of iron, and further the FTS activities. FTS tests indicated a medium amount of nitrogen (≤8.3 wt%) did not show obvious effect on the activity of the catalyst, but a much higher amount of nitrogen (≥16.5 wt%) could result in a significant decrease of activity. Nitrogen-containing groups could effectively suppress the methane selectivity and improve the lower olefins selectivity in the FTS reaction. The hydrocarbon productivity of the Fe/NMCs was up to 0.62 g HC/(h·g cat.) at 260 °C, 1 MPa, and H2/CO ratio of 1, much higher than that of any other iron catalyst reported under similar conditions. As a result of the superior performance, Fe/NMCs could be one of ideal candidates for iron-based FTS catalysts in the future.

Iron-based Fischer-Tropsch synthesis (FTS) has received renewable interests due to the gradual depletion of crude oil resources and its flexible product adjustment from lower olefins to long-chain hydrocarbons. However, one of the main challenges of iron-based FTS catalysts is the relative lower activity or productivity (per gram catalyst) at low or middle temperature. In this work, ultra-active iron-based FTS catalysts (Fe/SMC) were developed by loading iron over spherical mesoporous carbon (SMC) with huge porosity. The large pore volume (2.22 cm 3 /g) and high specific surface area (767 m 2 /g) of SMC allowed the high iron loading (up to 50 wt%) with proper dispersion. Weak C-Fe interaction was observed in the Fe/SMC catalysts, which was favorable for the reduction of iron and its further carburization to form more active sites of FTS. Activity tests showed that the Fe/SMC catalysts exhibited ultra-high activities in FTS. With the increase of iron loadings (30–50 wt%), the FTS rates (FTY, mmol CO/(h·g cat.)) of Fe/SMC catalysts increased due to the larger iron surfaces exposed at higher iron loadings. Small amount of K promoter significantly enhanced the FTY and apparent turnover frequency (TOF, S −1 ) of the Fe/SMC catalysts. The hydrocarbon productivity of K-promoted catalyst (40FeK/SMC) was up to 0.91 g HC/(h·g cat.) (260 °C, H 2 /CO = 1, 2 MPa), much higher than that of any other supported or unsupported iron-based FTS catalysts reported at similar reaction conditions. The results obtained in this work will open a new avenue for the development of highly active iron-based FTS catalysts.

In this paper, we propose a weighted online sequential extreme learning machine with kernels (WOS-ELMK) for class imbalance learning (CIL). The existing online sequential extreme learning machine (OS-ELM) methods for CIL use random feature mapping. WOS-ELMK is the first OS-ELM method which uses kernel mapping for online class imbalance learning. The kernel mapping avoids the non-optimal hidden node problem associated with weighted OS-ELM (WOS-ELM) and other existing OS-ELM methods for CIL. WOS-ELMK tackles both the binary class and multiclass imbalance problems in one-by-one as well as chunk-by-chunk learning modes. For imbalanced big data streams, a fixed size window scheme is also implemented for WOS-ELMK. We empirically show that WOS-ELMK obtains superior performance in general than some recently proposed CIL approaches on 17 binary class and 8 multiclass imbalanced datasets.

Advantages for biochars used in soil improvement have been proposed to their nutrients release and sorption characteristics which strongly depend on their production conditions. N

Advantages for biochars used in soil improvement have been proposed to their nutrients release and sorption characteristics which strongly depend on their production conditions. N2-flow and air-limited pyrolysis are two different widely-applied oxygen-limited pyrolysis methods for producing biochars, however, their different effects on nutrients release and sorption characteristics of biochars remains unknown. In this study, bamboo derived biochars pyrolyzed in N2-flow (BC-N2) and air-limited environments (BC-Air) at the temperature of 150~750 °C were used to compare the release and sorption of nitrogen and phosphorous nutrients. The results showed that release of nitrogen and phosphorous in BC-Air were always greater than those in BC-N2, the maximum nitrogen and phosphorous release of BC-Air (0.65 mg/g at 750 °C) is about 7.7 times of that of BC-N2 (0.084 mg/g at 450 °C). Both BC-N2 and BC-Air had no/little sorption of phosphate. Meanwhile, the sorption capacity of ammonium nitrogen on BC-Air (1.83 ~ 4.67 mg/g) was always greater than that on BC-N2 (0.23 ~ 1.34 mg/g) at the pyrolysis temperature of 300 ~ 750 °C. Phosphorous-containing minerals in ash was an enhancing factor for the release of phosphorous and sorption of ammonium nitrogen on BC-Air. Furthermore, with increasing pyrolysis temperature, the release amount of phosphorous from BC-Air and the sorption capacity of ammonium nitrogen on BC-Air increased. The results show that high pyrolysis temperature combined with air-limited environment produced biochars are optimal for nutrients enhancement and retention.

A multi-energy system that controls the generation and storage of different types of energy offers great potential for improving the overall system efficiency and operating cost. The economic dispatch problem for multi-energy systems is to find the optimal system configuration that satisfies the demands under a host of constraints. The problem is often formulated as a mixed integer programming problem, which is very hard to solve since the number of variables involved could be in the order of hundreds. The difficulty of the problem is further compounded when model predictive control is implemented because the set points for controlling the multi-energy system at each time step have to be modeled, drastically increasing the number of variables in the optimization problem. In this paper, a large-scale hybrid optimization strategy is proposed to solve the economic dispatch in a multi-energy management system. Mixed-integer linear programming is used to solve a linearly approximated problem to determine the discrete set points followed by nonlinear programming for the continuous variables. We compare different optimization methods for the nonlinear programming problem with respect to the computational time and cost savings. The cost savings is also compared to a rule-based economic dispatch as the baseline. The simulation is performed using the data obtained from an office building in an R modeling and data management environment. We show that the proposed optimization strategy is able to solve the economic dispatch problem of a multi-energy management system with 720 mixed-integer variables within a short timeframe while still accounting for the nonlinearity of the optimization problem.