Development of real-time damage estimation system of lifeline facilities using neural networks

主要研究成果如下：1.对1991年9月27-28日9119号台风和1993年1月秋木大地震这两次自然灾害进行了现场调查。调查了9119号台风停电对生命线系统和城市生活的影响。为了评估生命线系统的可靠性和适用性，1)模拟了地震地面运动的时空变化；2)从理论上研究了生命线系统的物理损伤程度与功能服务率之间的关系；3)开发了综合生命线分析系统的计算程序ILAS。为了探测和定位关东地区及周边地区发生的地震，开发了一个实时强震数据处理系统。并以一辆四轮S…的实时到达时间数据为例进行了验证地震台阵显示，自动系统计算的震源位置和发震时刻比较可靠。利用地震监测数据和土壤区划数据，模糊推理使我们能够进行燃气管网的破坏评估。决策分析也采用了模糊集理论。通过比较实际损伤数据和模糊推理结果，对提出的隶属度函数进行了检验。并对整个系统进行了样本计算。提出了一种基于神经网络的生命线系统震害评估方法。该网络使用地震烈度作为输入来估计结构的损伤。通过监督学习训练的网络在回忆测试中很好地估计了观察到的损伤程度。提出了一种利用神经网络优化电话业务震后恢复过程的有效方法。结果表明，神经网络可以有效地确定受损设备的维修顺序和修复路径，是确定修复过程的一种非常有用的工具。较少

The main results of this may be summarized as follows :1. Site investigations were conducted for the recent two natural disasters, the typhoon 9119 on September 27-28, 1991 and the Kushiro-oki earthquake on January , 1993. Effects of power Outage to lifeline systems and urban life were investigated for the typhoon 9119. Soil effect to structural damage was confirmed for the Kushiro-oki earthquake.In order to evaluate the reliability and serviceability of the lifeline system, 1) the space-time variation of earthquake ground motion was simulated ; 2) the relationship between the degree of physical damage to the lifeline system and the functional service ratio was theoretically investigated; and 3) an outline of the computer program ILAS (Integrated Lifeline Analysis System) was developed.3. A real-time strong-motion data processing system was developed to detect and locate earthquake occurring in and around the Kanto area. It has been demonstrated with real arrival time data for a four-s … More tation seismic array that hypocenter location and origin time calculated by the automatic system are quite reliable.4. Using earthquake monitoring data and soil zoning data, fuzzy reasoning enables us to perform damage estimation of gas networks. Decision analysis also employs fuzzy set theory. Proposed membership functions were examined by comparing actual damage data and the results of fuzzy reasoning . A sample calculation was also conducted as a total system.5. A use of neural networks was proposed for earthquake damage estimation of lifeline systems. The proposed network estimates the structural damage using seismic intensities as inputs. The network trained by supervised learning well estimated the observed damage extent in a recall test.6. An effective procedure was presented for optimization of post-earthquake restoration process of telephone service using neural networks. Results show that the repair sequencing of damaged equipment and the restoration route can be effectively determined and that the neural system may be a very useful tool for determining restoration process. Less