Search

Machine learning models often converge slowly and are unstable due to the significant variance of random data when using a sample estimate gradient in SGD. To increase the speed of convergence and improve stability, a distributed SGD algorithm based on variance reduction, named DisSAGD, is proposed in this study. DisSAGD corrects the gradient estimate for each iteration by using the gradient variance of historical iterations without full gradient computation or additional storage, i.e., it reduces the mean variance of historical gradients in order to reduce the error in updating parameters. We implemented DisSAGD in distributed clusters in order to train a machine learning model by sharing parameters among nodes using an asynchronous communication protocol. We also propose an adaptive learning rate strategy, as well as a sampling strategy, to address the update lag of the overall parameter distribution, which helps to improve the convergence speed when the parameters deviate from the optimal value—when one working node is faster than another, this node will have more time to compute the local gradient and sample more samples for the next iteration. Our experiments demonstrate that DisSAGD significantly reduces waiting times during loop iterations and improves convergence speed when compared to traditional methods, and that our method can achieve speed increases for distributed clusters. [ABSTRACT FROM AUTHOR]

In recent years, the application of traceability systems in the food and drug industry has developed rapidly, but it is rarely used for wind turbines. From the aspects of low information transparency and information islands in the supply chain process for wind turbines, a reliable traceability system is essential. However, the existing traceability systems are not suitable to be directly applied to wind turbines. Consequently, according to the characteristics of the wind power industry, a semi-centralized traceability architecture based on Internet of Things technology was proposed. Furthermore, a traceability platform was constructed by analyzing the information collected in each stage related to various user needs of wind turbines, and various applications, including manufacturing management and spare parts management, were developed. Compared with the existing systems, the proposed platform was wind-turbine-oriented, effectively improved traceability efficiency and enterprises' information security, and extended the length of the traceability chain by integrating the after-sales information. The traceability of key components of wind turbines during their life cycle provides a useful reference for further improving the parts quality management system of the wind power industry. [ABSTRACT FROM AUTHOR]

Life cycle assessment (LCA) is conducive to the change in the wind power industry management model and is beneficial to the green design of products. Nowadays, none of the LCA systems are for wind turbines and the concept of Internet of Things (IoT) in LCA is quite a new idea. In this paper, a four-layer LCA platform of wind turbines based on IoT architecture is designed and discussed. In the data transmission layer, intelligent sensing of wind turbines can be achieved and their status and location can be monitored. In the data transmission layer, the LCA platform can be effectively integrated with enterprise information systems through the object name service (ONS) and directory service (DS). In the platform layer, a model based on IMPACT 2002+ is developed, and four management modules are designed. In the application layer, different from other systems, energy payback time (EPBT) is selected as an important evaluation index for wind turbines. Compared with the existing LCA systems, the proposed system is specifically for wind turbines and can collect data in real-time, leading to improved accuracy and response time. [ABSTRACT FROM AUTHOR]

Cu-containing CHA type (Cu-CHA) zeolites have been widely investigated owing to their excellent low-temperature activity and high hydrothermal stability in selective catalytic reduction of NOx with NH3 (NH3-SCR). Herein, a series of Cu-SAPO-44 zeolites were prepared by one-pot method with dual-amine templates and the subsequent ion exchange (IE) with NH4NO3. The effect of NH4NO3 treatment on Cu species was investigated by X-ray powder diffraction (XRD), N2 adsorption-desorption isotherm, inductively coupled plasma (ICP); field-emission scanning electron microscope (FE-SEM), high-resolution transmission electron microscope (HRTEM), X-ray absorption fine structure (XAFS), and H2-temperature-programmed reduction (H2-TPR). The results indicated that—besides the main SAPO-44 structure—the CuO phase was detected by XRD in original samples. After IE with NH4NO3, the Cu contents decreased greatly from ICP analysis. The removal of CuO agglomerations and the presence of highly dispersed CuO nanoparticles (~2.36 nm) were confirmed by SEM, TEM and H2-TPR. Furthermore, a significant increase in the proportion of isolated Cu2+ was derived from XAFS. As a result, the activity at higher temperature (≥350 °C) was improved a lot. [ABSTRACT FROM AUTHOR]