基于模态分解和云粒子网络的大坝基岩地震动输入研究

Study of seismic wave input of dam bedrock based on modal decomposition and cloud particle network

在大坝抗震安全评价中基岩地震动输入多采用实测数据或人工生成等方式,而当坝址仪器损坏或历史震害资料不足时,确定基岩地震动就变得尤为困难。本文提出对大坝基岩地震动进行反演的研究思路,并开发了基于经验模态分解和云粒子网络的分解—训练—反演混合模型,在不依赖场地历史震害资料的情况下,仅用少量周边测站数据即可确定大坝的基岩地震动。首先,选取坝址周边地表及基岩的地震动实测记录,采用经验模态分解法将地震加速度序列分解;其次,通过粒子群算法建立与神经网络连接权值的映射,采用云理论优化粒子群算法的全局寻优能力,建立反演模型,将分解后的加速度序列作为训练集进行反演训练;然后,选取与大坝处于相似地质情况的地表实测地震动信息,结合反演模型对大坝基岩输入地震动进行反演;最后,以紫坪铺大坝为研究实例,通过对比传统输入方法,验证该模型的适用性。结果表明:本文所提的混合模型综合性能稳定,能较好地反演地震加速度序列,模型决定系数均大于0.9,平均绝对百分比误差均在11%左右;采用本文反演得到的基岩地震动进行计算,较已有研究成果计算误差降低0.79%~17.28%,与工程实际动力响应更为吻合。本文方法可为解决大坝基岩输入地震动的获取提供一条新途径。

The input of bedrock seismic wave in the existing seismic evaluation of dams mostly uses actual data or manual generation,and it becomes particularly difficult to determine bedrock seismic wave when the dam site instrumentation is damaged or historical information is inadequate.Therefore,the research idea of inversion of seismic wave input to bedrock of earth and rock dams is proposed,and a hybrid decomposition-training-inversion model based on empirical modal decomposition and cloud particle network was developed,determination of seismic wave of dam bedrock by using only a small number of peripheral measurement stations without relying on historical seismic data of the site.Firstly,the measured seismic wave records of surface and bedrock were selected,and the acceleration sequence was decomposed by the empirical modal decomposition method.Secondly,the particle swarm algorithm was used to establish the mapping with the neural network connection weights,optimizing the global search capability of particle swarm algorithms using cloud theory,establish the inversion model,and use the decomposed acceleration sequence as the training set for inversion training.Then,the seismic wave information measured at the surface,which was in a similar geological situation as the dam,was selected and combined with the inversion model to invert the seismic wave input to the bedrock of the dam.Finally,the Zipingpu dam is used as a research example to verify the applicability of the model by comparing the traditional input methods.The results show that the hybrid model proposed in this paper has a comprehensive and stable performance and can invert the seismic acceleration sequences better,with the model coefficient of determination are greater than 0.9,mean absolute percentage error are about 11%.At the same time,the calculation of bedrock seismic wave obtained from the inversion of this paper reduces the calculation error by 0.79%~17.28%compared with the existing research results,which is more consistent with the actual dynamic response of the project,thus providing a new way to solve the problem of acquiring seismic wave from the bedrock input of the dam.