Under the global low-carbon transition, integrated energy systems (IES) have emerged as a critical solution but face challenges from renewable integration volatility. To address this issue, this study proposes a bi-level optimisation framework combining Long Short-Term Memory (LSTM)-based renewable energy forecasting with adaptive operations. The upper level optimizes system configuration to minimize curtailment and investment costs, while the lower level dynamically adjusts operations to reduce costs and flexibility shortages. An iterative feedback loop replaces theoretical curtailment estimates with operational data, ensuring consistency between system design assumptions and actual performance. Compared to conventional methods, the framework achieves 16.5% lower emissions, 5.5% operating costs reduction, 17.8% less curtailment, and zero flexibility shortages. The study offers a novel paradigm for integrated planning-operation optimisation in high-renewable systems.