The invention discloses a particle filter load prediction analysis and transformer selection method and device based on clustering analysis. By preprocessing variable historical load data, processed data is subjected to clustering analysis, and several types of loads with different characteristics are extracted. On the basis, the several type of loads with the different characteristics are subjected to load prediction through a particle filter method separately, the predicted result times different weights, the final load prediction result is obtained, and the method and device can be appliedto distribution transformer selection evaluation. Accordingly, the prediction accuracy is increased, model change is flexible, and calculation is simplified.