This paper introduces an algorithm which uses multi-objective fuzzy optimization within a hybrid prediction model to improve the forecasting accuracy of slope deformation. Single forecasting models used in hybrid prediction are optimized by an empirical threshold law. The technique described in this paper uses mean square error to determine the fitting error of the prediction model and a gray relational grade to describe the developing relevance between monitored deformation curve and the predictive model. The multi-objective optimization model is established by introducing a fuzzy satisfaction function aimed at simultaneously decreasing the fitting error while increasing the developing relevance. Particle swarm optimization (PSO) is employed to solve the optimal weight of the hybrid prediction model, and the eventual multi-objective refined forecasts. The model has been used to conduct predictive analysis on an unstable slope (Laowubao) located in the 25th tender of YiBa highway, Hubei, China. The forecast results indicate that the new model obtains a smaller fitting error and can be considered more relevant than a traditional single prediction model. This technique can help effectively improve the precision of the prediction model and shows great application value.
Monitoring of displacement is a fundamental function of slope monitoring leading to the prediction of further slope displacement. This is one of the most important technical problems in slope engineering. Existing mechanical theory cannot solve the problem of slope displacement prediction completely. Mathematical methods have been applied to predict slope deformation, and have achieved reasonable results (Jin and Xu. 2008). These methods predict the trend of slope deformation by studying variation of the time series of monitoring data. Examples include the Saito method, grey theory model, filtering model, exponential smoothing model, regression model, neural network method, catastrophe theory model, time series analysis method, prediction model considering mechanical mechanism, etc (Zhao et al., 2007; Li et al., 2011). The various prediction methods tend to produce different results, with each having its own advantages and disadvantages.
