Agricultural diesel engines can be operated continuously under complex and variable conditions for extended periods. The higher requirements are posed on the Diesel Particulate Filter (DPF) regeneration and reliability. This study aims to enhance the safe and efficient performance of DPF regeneration, emissions, and fuel economy of agricultural diesel engines. An intelligent multi-objective optimization was also proposed to enhance multiple cyclic training. Taking the agricultural engines as the research object, the sample data of the bench test was obtained after experimental design. A prediction model was then constructed in the conditions of DPF regeneration using a Back Propagation neural network (BPNN). According to the Seagull Optimization Algorithm (SOA), the Adaptive Memory Seagull Optimization (AMSO) was proposed to optimize the structure parameters of the BPNN model, in order to meet the requirements of precision. The targets of optimization included T4, T5, O2 concentration, Brake Specific Fuel Consumption (BSFC), NOx, and smoke opacity. In the specific conditions of agricultural diesel engines, multi-objective optimization of injection and intake control parameters was realized under regeneration mode using the Non-dominated Sorting Genetic Algorithm (NSGA)-III. The AMSO-BP prediction model was used to evaluate the fitness of the Pareto optimal solutions, in order to validate the NSGA-III optimal dataset. The optimized MAP values were written into the Engine Control Unit (ECU) after bench tests. Steady-state and World Harmonized Transient Cycle (WHTC) experiments were carried out to verify the multi-objective optimization of the model. The results indicate that the AMSO significantly outperformed the SOA in the optimization of the BPNN prediction model. The AMSO-BPNN prediction model more accurately utilized experimental data to predict the T4, T5, O2 concentration, BSFC, NOx, and smoke opacity, with the R² values of 0.97, 0.99, 0.95, 0.99, 0.98, and 0.95, respectively, on the validation set. Furthermore, the Mean Absolute Percentage Errors (MAPE) on the validation set were reduced by 0.1%, 0.33%, 0.8%, 0.04%, 0.08%, and 4.2%, respectively, compared with the SOA-BP model. The steady-state tests before and after optimization showed that T4 and T5 increased by an average of 3.14%, where the concentration of exhaust oxygen increased by 2.07% and 10.79%, respectively. Simultaneously, the average reductions in the NOx, smoke opacity, and BSFC were 8.68%, 12.07%, and 1.03%, respectively. Efficient and safe DPF regeneration was achieved to significantly reduce the emissions of diesel engines. There was particularly notable optimization under the operation of agricultural diesel engines, which were often working at low speeds and high loads, fully meeting the usage requirements of these engines. The WHTC transient test also showed that the effectiveness of the optimization was achieved in the T4, T5, and O2 concentrations, which increased by 31%, 2.6%, and 0.5%, respectively. Additionally, NOx and soot emissions decreased by 10.4% and 0.8%, and BSFC was reduced by 3.5%. These findings demonstrate that better performance was obtained under the complex and variable conditions of agricultural diesel engines. In conclusion, the safe and efficient DPF regeneration can be expected to reduce the emissions and operational costs for agricultural diesel engines. The specific requirements of agricultural diesel engines can also be fully met to enhance the system reliability in the engineering application. This research can offer practical guidance to optimize the control parameters under the regeneration mode of diesel engines, particularly for more sustainable and efficient agricultural machinery. [ABSTRACT FROM AUTHOR]