Personalized dialogue generation aims to achieve more human-like conversations by endowing dialogue agents with personas. However, existing dialogue systems still lack the understanding of human communication patterns and tend to generate incoherent or generic responses. One of the major challenges lies in wrong selection of when and which persona information to invoke based on the dialogue flow. Moreover, predefined personas often provide superficial and fragmented descriptions, making it difficult for models to comprehensively understand the character. From the perspective of cognitive science, dialogues inherently organize around concepts that structure the semantic flow. In real-world conversations, humans retrieve relevant information from memory and make associations based on the concepts, then generate contextually appropriate responses. Based on this theory, we decide to model the underlying concept-guided cognitive process in human conversations to address the challenges. In this paper, we propose Concept-guided Persona Perception with Commonsense Reasoning (CPPCR), an effective framework consists of three modules that correspond to different stages of the cognitive process. Specially, the Concept-Guided Persona Selection module utilizes the trained concept extractor, which identifies concepts representing the dialogue semantic flow, to guide the selection of relevant persona sentences. Then, the Commonsense-Augmented Persona Expansion module expands the selected persona sentences through commonsense reasoning. Finally, the Persona-Aware Generation module enhances the model’s perception of connections between persona information and dialogue semantics through prompt tuning. The experimental results demonstrate the improvement of CPPCR in generating responses that are contextually coherent, rich in content, and consistent with the persona.