The fracture pressure of tight conglomerate reservoirs in the Mahu Block varies significantly, which easily leads to unbalanced staged fracturing stimulation of horizontal wells and poor production performance. There is an urgent need to accurately predict fracturing sweet spots to improve the stimulation effect of horizontal wells. To this end, based on RF (random forest) and cross-validation, the primary controlling factors of fracture pressure in tight conglomerate reservoirs were screened. The hyperparameters of BiLSTM (bidirectional long short-term memory) network were optimized using AE (alpha evolution) algorithm, and a hybrid prediction model for fracture pressure was established. Fracture pressure was introduced into fracturing sweet spot evaluation to calculate fracturing sweetness, and an innovative fracturing sweet spot prediction model for tight conglomerate reservoirs was developed. Finally, the model was verified using field data from the Mahu Block. The results show that the screening of primary controlling factors and hyperparameter optimization have effectively improved the accuracy of the hybrid fracture pressure prediction model, with a prediction RMSE(root mean square error) of 0.75, MAPE (mean absolute percentage error) of 0.63%, and Root mean square error,R² (coefficient of determination)of 0.92. The R² of the regression between fracturing sweetness and FSI(flow scanning imaging) data is 0.87. When the fracturing sweet spot prediction model is applied to different blocks, the R² still exceeds 0.70. The study concludes that fracturing sweetness has a strong correlation with oil production, and the established fracturing sweet spot prediction model has good applicability. It can effectively guide the fracturing design and staged-clustered optimization of horizontal wells in tight conglomerate reservoirs, providing technical support for achieving balanced reservoir modification and enhanced oil recovery.