Abstract: The risk of power load becomes the hot spot in the electric power industry; however, due to the single factor evaluation, the traditional power load forecasting model is not adequately comprehensive and systematic. Hence, it cannot accurately predict the risk and may cause hidden danger of power failures. To address this issue, the risk of power load is analyzed and forecast by collecting data from multiple sources:customer service center, machine, and historical weather records and so on. First by cleaning and sorting the data and then by the K-Mean clustering, variables are chosen which have strong correlation with risk degree of transformer to construct the Bayesian discriminant models. The experimental results show that this model can accurately predict the risk of transformer at a certain probability of 99.53%. In the practical aspect, this model can provide prevention scheme and control decisions to power supply security and contribute to reduce customer's electricity failure and improve customer satisfaction.