传统坝体地震易损性评估所采用的非线性数值模拟手段的计算量很大，为兼顾大坝地震易损性评估的高效 性与准确性，提出基于增量动力分析（incremental?dynamic?analysis,?IDA）与多层感知机（multilayer?perceptron,?MLP） 的混凝土坝地震易损性评估方法。以我国西北地区某混凝土重力坝为例，建立三维坝体-库水-地基有限元模型并 开展多组地震响应计算。利用等步长调幅处理所选地震动记录，并采用三向地震进行幅值输入；选取峰值 [ 地 面 ] 加速度（peak?ground?acceleration,?PGA）为地震动强度指标，坝顶顺河向位移为大坝损伤指标，初步进行混凝土 坝易损性分析。提取各地震动特性参数作为输入，坝顶顺河向位移为输出，训练并测试 MLP 模型；扩充地震动以 获取各地震动特性参数，利用 MLP 模型进行坝顶顺河向位移的快速预测，实现有限元结果的扩充，进行混凝土坝 易损性分析，并绘制易损性曲线。结果表明，将 MLP 模型引入分析可有效扩充数据量，利用 IDA-MLP 耦合方法 建立的大坝地震易损性曲线符合实际规律，验证了采用 MLP 模型预测混凝土坝损伤指标进而扩充数据的可行性， 在保证精度的情况下大幅提高计算效率，为同类型水工建筑物的抗震安全评价和防震减灾提供科学依据。

The?western?region?of?China?is?rich?in?hydropower?resources,?but?strong?earthquakes?occur?frequently. Hydraulic?structures?may?withstand?strong?earthquakes?during?their?life?cycle.?The?damage?caused?by?earthquakes will?affect?the?normal?operation?of?water?conservancy?projects,?which?seriously?threatens?people's?lives?and?social stability.?The?seismic?vulnerability?analysis?provides?the?seismic?basis?for?seismic?design?of?hydraulic?structure?and earthquake ?prevention ?and ?disaster ?reduction, ?which ?is ?very ?important ?to ?ensure ?the ?expected ?performance ?of concrete?gravity?dam?and?safe?operation?under?earthquake. ??????The?seismic?vulnerability?assessment?of?dams?often?requires?the?use?of?nonlinear?numerical?simulation?methods and?a?huge?amount?of?calculation?work.?Considering?the?efficiency?and?accuracy?of?seismic?vulnerability?assessment of ?dams, ?a ?way ?based ?on ?incremental ?dynamic ?analysis ?and ?multilayer ?perceptron ?for ?seismic ?vulnerability assessment?of?concrete?dams?is?proposed.?Taking?a?concrete?gravity?dam?in?northwest?China?as?an?example,?a?three- dimensional ?finite ?element ?model ?of ?dam ?body-reservoir ?water-foundation ?is ?established ?and ?multiple ?seismic response?calculations?are?carried?out.?The?selected?ground?motion?records?are?processed?by?equal?step?amplitude modulation,?and?the?amplitude?input?is?carried?out?by?three-way?seismic.?The?peak?ground?acceleration?is?selected?as the?index?of?ground?motion?intensity,?and?the?displacement?of?dam?top?along?river?is?selected?as?the?index?of?dam damage. ?The ?multilayer ?perceptron ?model ?was ?trained ?and ?tested ?by ?extracting ?the ?vibration ?characteristic parameters?as?input?and?the?displacement?of?dam?top?along?river?as?output.?The?ground?motion?is?extended?to?obtain the?vibration?characteristic?parameters.?The?MLP?model?is?used?to?quickly?predict?the?displacement?of?the?dam?top along?the?river,?and?the?finite?element?results?are?extended.?The?vulnerability?analysis?of?the?concrete?dam?is?carried out,?and?the?vulnerability?curve?is?drawn. ??????The ?results ?are ?as ?follows: ?(1) ?The ?seismic ?vulnerability ?of ?concrete ?gravity ?dam ?is ?analyzed ?based ?on incremental?dynamic?analysis?method.?Incremental?dynamic?analysis's?curve?clusters?of?river-side?displacement, transverse?river?flow?displacement?and?vertical?displacement?of?the?dam?crest?are?plotted,?and?seismic?vulnerability curves?are?plotted.?In?summary,?the?earthquake?damage?grade?of?the?selected?gravity?dam?under?the?designed?ground motion?is?low,?and?about?80%?probability?belongs?to?the?basically?intact?state.?(2)?The?multi-layer?perceptron?model is ?introduced ?into ?the ?incremental ?dynamic ?analysis ?method, ?and ?the ?seismic ?vulnerability ?analysis ?method ?of concrete?gravity?dam?based?on?incremental?dynamic?analysis-multi-layer?perceptron?is?proposed.?After?the?training of?the?multi-layer?perceptron?model,?the?accuracy?of?the?damage?prediction?index?is?higher.?The?coefficient?of determination?is?0.960?2,?the?mean?square?error?is?0.005?6,?and?the?mean?absolute?error?is?0.056?1.?Based?on?this method,?more?analytical?data?can?be?constructed?to?expand?the?finite?element?dynamic?calculation?results.?It?saves the?calculation?work?up?to?75%?and?reduces?the?nonlinear?dynamic?calculation?time.?The?seismic?vulnerability?of dams?can?be?predicted?quickly?and?accurately?as?close?as?possible?to?the?reality,?and?the?vulnerability?curve?can?be drawn?to?improve?the?efficiency?of?seismic?vulnerability?analysis?of?dams.