Personalized Emotional Support Conversation (PESC) systems are crucial for providing effective emotional solace tailored to users’ unique needs. This paper details our first-place system in the NLPCC 2025 Shared Task 8 on PESC. To address the challenge of enhancing Large Language Models’ (LLMs’) ability to generate personalized and highly supportive responses in this task, we proposed and integrated a series of innovative optimization strategies. Key methods include: utilizing Forward-looking Information by integrating users’ future turn responses as guiding signals into the system prompt to enhance coherence and goal-orientation; employing an optimized two-turn Sliding Dialogue History Window mechanism for efficient context management; and significantly improving generated text diversity through Output Decoupling of the model’s own outputs in the dialogue history. At the model training level, we adopted LoRA for parameter-efficient fine-tuning and applied NEFTune, adding noise to the embedding layer to enhance model robustness. Experimental results show that our approach achieved a top composite score of 42.69.