The selection of relevant socio-economic indicators for power load mid-long term forecasting lacks theoretical foundation in recent research papers. Data mining techniques are introduced into association analysis of power consumption growth. Twenty-five relevant indicators are selected from statistics of China's thirty provinces， autonomous regions and municipalities. Three membership functions are adopted to assign values and Apriori algorithm is then used to calculate fuzzy confidence value between power consumption and different influencing factors. Several dominant factors including gross domestic product（GDP）， industrial output value， total value of imports and exports， total investment in fixed assets and per capita disposable income， are identified in power load growth. And general growth rule of China's power consumption is finally obtained based on self-organizing map neural network. The research procedure provides a new strategy to choose correlative factors in yearly prediction of power load.