基于LSSVM-ARMA模型的高层建筑地基变形时间序列预测

Time series forecast for foundation deformation of tall building based on LSSVM-ARMA model

由于建筑物沉降受多种因素的影响和制约,其变化规律很难用一个显式的数学公式予以正确表达。本文基于时间序列预测法,结合小波变换、粒子群优化的最小二乘支持向量机和自回归移动平均模型建立了联合的预测方法和模型。将沉降变形时间序列通过小波分解和重构为趋势时间序列、随机时间序列。分别对趋势时间序列和随机时间序列采取滚动预测,最后将两个序列预测结果叠加即为最终预测结果。通过算例分析表明,该方法用于建筑物沉降与倾斜预测是可行的。

Due to the settlement of buildings influenced by many factors and constraints, it is difficult to use an explicit mathematical formula to express the change. Based on time series forecasting method, the wavelet transform, least squares support vector machine （LSSVM） optimized by panicle swarm optimization （PSO） and autoregressive moving average model （ARMA） are combined together and a united forecast method and model is proposed in this paper. The time series of settlement deformation is decomposed and reconstructed by wavelet transform, which is a trend series, and random series. The trend series and random series use different models to forecast, finally, the sum of trend series and random series are used as the final forecast value. Take advantage of an exampIe, the method is feasible for the prediction of building settlement and building inclination.