莫高窟洞窟微环境场模型与传感器布局优化技术研究

发布时间：2018-01-22 18:35

本文关键词： 采样优化 kriging插值模拟退火无线传感器网络敦煌莫高窟　出处：《浙江大学》2017年硕士论文　论文类型：学位论文

【摘要】：敦煌莫高窟中有壁画4.5万平方米,泥质彩塑2415尊,是世界上现存规模最大,内容最丰富的佛教艺术地,目前世界各国对文化遗产的保护也逐渐重视。由于微气象环境对壁画的保存有着很大的影响,尤其是过高的湿度,二氧化碳浓度,以及不适的温度,都会加速壁画等文物的损坏。为了更好的控制这些微气象环境,首先要对这些微气象环境做出准确合理的监测。在微环境的监测中,无线传感器的使用已经较为普遍。但是敦煌莫高窟中由于洞窟大小以及形状的复杂性,即使在同一洞窟中,不同位置的温湿度也有着不小的差异。使用大量传感器必然可以提高微气象环境监测的准确度,但同时也导致了较高的传感器部署成本,以及大量的能源消耗,此外对历史文物的展示带来了很大的不便。因此本文的出发点是如何使用较少量的传感器,对洞窟中的微气象环境做出合理准确的监测。因此本文以敦煌465窟为例,通过在该洞窟中部署的传感器,获取大量的温湿度数据。然后在该场景下首次提出插值算法和模拟退火算法相结合的方法,来对传感器的数量和布局进行优化。这里主要从两点出发,首先寻找适合洞窟微气象环境插值的空间插值算法,其次寻找合理传感器布局,对洞窟中的微气象环境进行准确高效的监测。首先本文随机挑选出20组传感器布局,每组布局中的传感器个数为10个,然后从温度均方根误差以及湿度均方根误差的角度对比反距离加权法,以及不同模型的普通kriging法,包括线性模型的kriging,指数模型的kriging,幂模型的kriging,球状模型的kriging。最终挑选出适合洞窟微气象环境插值的插值算法。然后以上一步中挑选出来的最适合的空间插值方法为基础,并结合先验知识对传感器进行分组并插值验证,根据不同组传感器的插值误差来对每组传感器的权重进行设置,然后根据这些传感器的权重优化传统模拟退火算法中的状态转移过程,从而加速最优状态的发现,减少计算的迭代次数。然后选择合理的目标函数,通过上述改进的模拟退火算法,在全局范围内进行智能化的搜索,该方法可以避免陷入局部最优解,从而找出全局较优解。在模拟退火算法中,设定初始解的时候需要确定传感器的个数,出于传感器监测精度以及传感器部署成本的考虑,实验中分别设计了 3个,4个,5个,6个,7个,8个,9个,10个,11个,12个样点。通过对每一组样点进行插值,并求出目标函数值,通过分析目标函数值变化趋势,以及结合温湿度传感器的部署成本,以及能耗等方面,最终确定传感器个数,以及最终布局。最后通过测试数据集来对最终的传感器布局进行验证。
[Abstract]:Dunhuang Mogao Grottoes, with 45,000 square meters of murals and 2,415 clay sculptures, is the largest and most abundant Buddhist art site in the world. At present, countries all over the world pay more and more attention to the protection of cultural heritage. Due to the micrometeorological environment has a great impact on the preservation of murals, especially high humidity, carbon dioxide concentration, and uncomfortable temperature. In order to better control these micro-meteorological environment, first of all to make accurate and reasonable monitoring of these micro-meteorological environment. In the monitoring of micro-environment. The use of wireless sensors has become more common, but in Dunhuang the Mogao Grottoes, because of the complexity of the size and shape of the grottoes, even in the same cave. The use of a large number of sensors will inevitably improve the accuracy of micro-meteorological environment monitoring, but also lead to higher sensor deployment costs and a large amount of energy consumption. In addition, the display of historical relics has brought great inconvenience. Therefore, the starting point of this paper is how to use a small number of sensors. The micrometeorological environment in the cave is monitored reasonably and accurately. Therefore, taking Dunhuang Grottoes 465 as an example, the sensors deployed in the cave are used. Get a large amount of temperature and humidity data. Then in this scenario, the first proposed interpolation algorithm and simulated annealing algorithm to optimize the number and layout of the sensor. Firstly, the spatial interpolation algorithm suitable for the interpolation of the cave micrometeorological environment is found, and then the reasonable sensor layout is found. The micro-meteorological environment in the cave is monitored accurately and efficiently. Firstly, 20 groups of sensors are randomly selected in this paper, and the number of sensors in each group is 10. Then the inverse distance weighting method is compared from the root mean square error of temperature and the root mean square error of humidity, as well as the ordinary kriging method of different models, including the kriging of linear model. Krigingof exponential model, kriging of power model. Finally, the interpolation algorithm suitable for the interpolation of the cave micrometeorological environment is selected. Then the most suitable spatial interpolation method selected from the above step is used as the basis. Combining with prior knowledge, the sensors are grouped and verified by interpolation, and the weights of each group of sensors are set according to the interpolation errors of different groups of sensors. Then according to the weights of these sensors, the state transfer process in the traditional simulated annealing algorithm is optimized, which accelerates the discovery of the optimal state, reduces the number of iterations, and then selects the reasonable objective function. Through the above improved simulated annealing algorithm, intelligent search in the global scope, this method can avoid falling into the local optimal solution, and find out the global optimal solution, in the simulated annealing algorithm. When setting the initial solution, we need to determine the number of sensors. Considering the sensor monitoring accuracy and sensor deployment cost, we designed three, four, five, six, seven, eight, respectively. 9, 10, 11 and 12 samples are interpolated to each set of samples, and the value of objective function is obtained. The change trend of objective function value is analyzed, and the deployment cost of temperature and humidity sensor is analyzed. Finally, the number of sensors and the final layout are determined. Finally, the final sensor layout is verified by the test data set.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：K879.21;TP212.9

【参考文献】