The invention discloses a model-self-matching and fusion health forecasting method. The method includes the steps that 1, a forecasting-model total base is constructed according to a plurality of existing mature forecasting models; then the forecasting models in the forecasting-model total base are subjected to forecasting-result error testing based on time series data of the health parameters of a to-be-tested system, multiple forecasting models with forecasting-result errors meeting normal distribution are selected to serve as combinational forecasting models, one matched forecasting model with the forecasting-result errors meeting the tolerance requirement is further determined to be used for forecasting the health parameters, the forecasting results of the matched forecasting model are fused, and finally the forecasting value of the health parameters of the to-be-tested system is obtained. By means of the model-self-matching and fusion health forecasting method, newest observation data can be tracked in real time, automatic selecting and matching of the multiple forecasting models are carried out, the forecasting results of the multiple forecasting models are automatically selected and fused according to the characteristics of the data, and the model-self-matching and fusion health forecasting method is particularly suitable for long-term accurate forecasting of a system with the unobvious change regularity or the complex change regularity.