The　accurate　load　forecasting　is　of　great　significance　to　power　companies.　In　this　paper,　we　proposed　a　probabilistic　long-term　load　forecasting　model　based　on　stacked　Long　Short-Term　Memory　(LSTM).The　data　set　is　the　load　data　of　a　power　plant　from　2005　to　2013.Firstly,　data　preprocessing　is　aimed　to　eliminate　outliers　and　missing　values,　it　can　improve　the　accuracy　of　prediction.　Secondly,　we　obtained　the　point　prediction　value　with　stacked　LSTM,　and　the　result　shows　that　the　proposed　model　performs　better　on　prediction　accuracy　than　other　models,　such　as　Support　Vector　Regression　and　Artificial　neural　network　(BP).　Finally,　we　proposed　a　probability　density　prediction　method　based　on　error　statistics,　comparing　with　point　prediction　method,　it　can　provide　more　uncertain　information　for　long-term　load　forecasting.　©　2019　Association　for　Computing　Machinery.