Integrating green technologies into traditional industries requires precise matching between technological supply and situational demand, yet many existing knowledge graphs model only categorical entity relationships and lack the constraint structure needed to determine whether a technology is truly fit for purpose, including legal validity, geographic jurisdiction, organizational ownership, and IP transaction history. Building upon Data Pragmatics theory, we propose the \textbf{Green Pragmatics Graph (GPG)}, a constraint-oriented schema design approach that models five pragmatic dimensions as first-class graph elements. Taking hydrogen energy as a case study, we construct a GPG from over 100,000 patents (10 node types, 19 relationship types) and implement a \textbf{Text-to-Cypher} interface with a two-stage fine-tuning strategy combining supervised fine-tuning (SFT) and GRPO reinforcement learning with online execution-based rewards. Starting from a 7B-parameter base model with 24.6% execution accuracy (EX), SFT raises EX to 56.4%, and subsequent reinforcement learning further boosts it to 68.1%. On our execution-validated, schema-specific Text-to-Cypher benchmark, this schema-trained 7B model exceeds the best \emph{zero-shot} commercial LLM baseline (61.4% EX), highlighting the benefit of targeted training for complex constraint schemas rather than a general capability advantage. These findings suggest that constraint-oriented schema design, combined with targeted domain fine-tuning, can substantially improve situational retrieval in complex green technology ecosystems.