<p>The multi-type load characteristic sequences under different electricity usage scenarios and geographical climate environments exhibit high complexity and high dimensionality. Existing forecasting methods are plagued by high overlap of feature information and low model predictability, resulting in low economic benefits and poor reliability of regional capacity layout. To address these issues, a regional load demand forecasting and capacity layout optimization method based on multi-type load aggregation interaction characteristics is proposed. The proposed method utilizes a regional load forecasting model based on an improved Informer that considers multi-type load aggregation interaction characteristics, which enhances its analytical capability by incorporating a dimension-segment-wise embedding mechanism and a two-stage attention mechanism. The method also includes a two-stage regional new load capacity layout optimization method that considers both reliability and economic factors. The model’s effectiveness is validated through simulation results. The root mean squared error (RMSE) value of the proposed method is reduced by 72.37%, 65.78%, 51.27%, 45.08%, and 15.05% compared to recurrent neural network (RNN), long short-term memory (LSTM), gated recurrent unit (GRU), Transformer, and traditional Informer models, respectively. The mean absolute error (MAE) value is reduced by 75.53%, 68.02%, 58.36%, 48.17%, and 25.27%, respectively. The R-squared (R²) value is increased by 11.76%, 7.95%, 17.24%, 5.56% and 3.26%, respectively. Additionally, the regional load’s minimum value has increased by 8.95%, and the peak value has decreased by 8.46% compared to the baseline, demonstrating the proposed method’s ability to improve both forecasting accuracy and the optimization of regional power grids.</p>