Application of a Linear Programming-Driven Multi-Algorithm Integration Framework in Optimizing Crop Planting Strategies

This study examines a rural village in the mountainous region of North China. Based on local crop characteristics, farmland conditions, and 2023 planting data, a linear programming (LP) model was established as the baseline framework. Three distinct single models were developed by integrating reinforcement learning Q-Learning and simulated annealing (SA) algorithms, respectively. Subsequently, hybrid models—LP-Q-Learning, LP-SA, and LP-SA-Q-Learning—were designed for cross-validation. The study aims to maximize net profits from 2024 to 2030, comprehensively considering constraints such as crop growth cycles, market fluctuations, and climate uncertainties. Results indicate: In the short term (2024–2026), the LP-Q-Learning model demonstrates optimal optimization performance by rapidly adapting to market volatility and crop rotation demands through reinforcement learning's dynamic responsiveness. In the long term (2027–2030), the LP-SA-Q-Learning model overcomes local optima traps through the SA algorithm. By integrating Q-Learning's dynamic adjustment capability with LP's deterministic constraints, it demonstrates strong robustness under complex climate and market conditions, significantly outperforming other models in optimization. Overall, the LP-SA-Q-Learning model initially corrects strategy deviations through continuous learning, steadily increasing net returns. It achieves optimal strategy stability and return growth in the mid-to-late stages, validating its core advantages for long-term agricultural planning in mountainous regions.