Time series forecast of foundation pit deformation based on LSSVM-ARMA model
Jin Cao · Rock and Soil Mechanics · 2014
It is difficult to forecast and control the deformation of foundation pit engineering. A time series forecast method of foundation pit deformation based on wavelet transform, least squares support vector machine(LSSVM) optimized by particle swarm optimization(PSO) and autoregressive moving average model(ARMA) is proposed. Firstly, the foundation pit deformation series is decomposed and reconstructed into trend series and random series. Secondly, the trend series future values are forecasted by PSO-LSSVM model; while the random series future values are forecasted by ARMA model. Finally, the sum of trend series and random series future values are used as the final forecast value. This method is used to forecast the deep horizontal displacement of a foundation pit in Kunming. Wherein, the latest measured data of early working stage is used to build forecast model; and the model is used to forecast the later working stage deformation. The result is satisfied.