Load forecasting of large scale distribution network needs to focus on the accuracy and calculation efficiency. Because of the large scale and complex correlation factors, the traditional centralized or simple distributed platform environment is difficult to meet the application requirements. A solution based on the big data platform is proposed. Combined with the characteristics of large-scale distribution network load forecasting application scenario, the technical route of key links such as data storage, data preprocessing, characteristic analysis, forecasting algorithm, etc. is proposed in the scheme. The classified storage of multiple types of data is realized by the application of diversified distributed storage mode；data preprocessing, load characteristic analysis, etc. is realized by the application of machine learning technology；rolling shortterm forecasting based on sliding window operation and medium and long-term forecasting based on stateless operation is realized by the application of spark stream computing technology. The practical application results in a province prove that the solution is effective in improving the forecasting accuracy and calculation efficiency.