With the increasing scale of distributed photovoltaic connected to the distribution network, conducting research on distributed photovoltaic siting and sizing is of great significance for improving the economy and security of the distribution network. Currently, most research on photovoltaic siting and sizing planning adopts conventional single period programming method for problem modeling, ignoring the medium- and long-term loads growth property of the distribution network, which will have a negative impact on the actual application effect of the model. Therefore, a bi-level multi-stage programming model for multi-objective photovoltaic siting and sizing planning that considers load classification growth is proposed. Firstly, the dynamic time warping （DTW） and LightGBM algorithm are combined to achieve classification and prediction of medium- and long-term loads. Then, a two-stage dynamic planning model is constructed for multi-objective photovoltaic siting and sizing planning considering load classification growth, and the multi-objective bi-level crisscross optimization algorithm （MOBL-CSO） is introduced for solving. The experimental results in the modified IEEE 33-bus system indicate that the proposed scheme can comprehensively consider the load changes throughout the entire planning period for optimization, and can obtain a photovoltaic siting and sizing planning scheme that is more long-term safe and economic compared to traditional methods.