Consumption mode stability analysis(CMSA)is the precondition of load forecasting

the essence of which is to judge the similarity of a customer's consumption features in different historical time periods. Extracting consumption features from data of a long time period reduces the feasibility and accuracy of CMSA

using the short-time data might also be of low accuracy because daily consumption data are influenced by random factors. A method is therefore proposed to extract the customers' consumption modes by principal component analysis and taking daily consumption coefficients and daily consumption volatility as original consumption features. The stability of consumption mode is further judged by the Euclidean distance between the factor loading vectors of the principal components in two historical periods. A numerical example is provided by a residential community. Results show that the monthly consumption forecast accuracy is apparently different from the consumers in stable and unstable consumption modes judged by the proposed method and there is significant positive correlation between the similarity distance and the forecast error. It's also concluded that proper data period is of utmost importance to the feasibility and accuracy of CMSA. It seems 16 weeks will be appropriate for the CMSA problem.