深度强化学习优化的学习向量量化聚类算法

基于学习向量量化(Learning Vector Quantization,LVQ)单次迭代聚类效果不稳定和随着数据维度增大,聚类效果下降的缺陷,采用了一种深度强化学习优化的LVQ聚类算法。将LVQ算法的每一次迭代看做深度强化学习的一个状态,LVQ算法初始化一组原型向量后,用原型向量与数据集中每一个数据点做“拉近”或“远离”运算来完成一次迭代。优化算法挑选一部分数据点,并与原型向量做运算,将这一过程作为一个动作,选取的数据子集不同,产生的动作也不同,把这些动作组成动作集,选定动作后,根据奖赏函数找到最佳动作,进入下一状态。通过对UCI公共数据集和碳碳沉积数据集试验,得出优化后的算法FMI提升3%到10%,Dunn指数提升2%到9%,准确率提高3%到6%,用于公共数据集及碳碳沉积材料的生产过程数据分析性能较优。

The Burial of Biogenic Silica, Organic Carbon and Organic Nitrogen in the Sediments of the East China Sea

We sampled the sediments of the East China Sea during 2005 and 2006, and analysed the contents of the biogenic matters: biogenic silica, organic carbon, and organic nitrogen. From the surface distribution we found the contents of these substances to be in the ranges of 0.72%-1.64%, 0.043%-0.82%, and 0.006%-0.11%, respectively. Their distributions were similar to each other, being high inside the Hangzhou Bay and low outside the bay. The vertical variations of the contents were also similar. In order to discuss the relation between them we analysed the variations of content with depth. They increased in the first 7 cm and then decreased with depth. The peaks were found at depths between 20 to 25 cm. The distribution of carbonate showed an opposite trend to that of biogenic matters. The content of total carbon was relatively stable with respect to depth, and the ratio of high organic carbon to carbonate showed a low burial efficiency of carbonate, which means that the main burial of carbon is organic carbon. In order to discuss the source of organic matters, the ratio of organic carbon to organic nitrogen was calculated, which was 8.01 to 9.65, indicating that the organic matter in the sediments was derived mainly from phytoplankton in the seawater.

Combined influence of sedimentation and vegetation on the soil carbon stocks of a coastal wetland in the Changjiang estuary

Coastal wetlands play an important role in the global carbon cycle. Large quantities of sediment deposited in the Changjiang （Yangtze） estuary by the Changjiang River promote the propagation of coastal wetlands, the expansion of saltmarsh vegetation, and carbon sequestration. In this study, using the Chongming Dongtan Wetland in the Changjiang estuary as the study area, the spatial and temporal distribution of soil organic carbon （SOC） stocks and the influences of sedimentation and vegetation on the SOC stocks of the coastal wetland were examined in 2013. There was sediment accretion in the northern and middle areas of the wetland and in the Phragmites australis marsh in the southern area, and sediment erosion in the Scirpus mariqueter marsh and the bare mudflat in the southem area. More SOC accumulated in sediments of the vegetated marsh than in the bare mudflat. The total organic carbon （TOC） stocks increased in the above-ground biomass from spring to autumn and decreased in winter; in the below-ground biomass, they gradually increased from spring to winter. The TOC stocks were higher in the below-ground biomass than in the above-ground biomass in the P. australis and Spartina alterniflora marshes, but were lower in the below-ground biomass in S. mariqueter marsh. Stocks of SOC showed temporal variation and increased gradually in all transects from spring to winter. The SOC stocks tended to decrease from the high marsh down to the bare mudflat along the three transects in the order： P. australis marsh 〉 S. alterniflora marsh 〉 S. mariqueter marsh 〉 bare mudflat. The SOC stocks of the same vegetation type were higher in the northern and middle transects than in the southern transect. These results suggest that interactions between sedimentation and vegetation regulate the SOC stocks in the coastal wetland in the Changjiang estuary.