Along with the increasing development of the suburban society and economy, accurate load forecasting has a significant meaning to planning and operation of suburban power systems. This paper proposes a novel interval forecast model for daily peak load. Firstly, a time series of daily peak load is decomposed into the mean trend and the high frequency component. For the mean trend, a number of factors that may have influence to the load are analyzed and combined by principle component analysis (PCA), and the principle components are used to train the least squares-support vector machine (LS-SVM) so as to forecast the mean trend as a point value. On the other hand, the high frequency component is predicted to fall into an interval using the Markov chain model. Finally, the forecast result, which is also an interval, is obtained by summing the two forecast results together. To further improve the quality, an error correction part is proposed to correct the initial forecast interval. Simulation studies are conducted for two suburban areas in southern China, and the results have demonstrated that compared with traditional methods, the proposed model provides accurate and stable forecast results.