Scholarlink

<p>Agricultural heritage tourism plays an important role in the conservation of traditional agro-landscapes and the development of rural areas. Current route optimisation algorithms, mainly used for urban tourism, are unsuitable for the multi-objective constraints of agricultural heritage environments, such as seasonal farming activities, ecological sensitivity, carrying capacity constraints and socio-cultural heritage preservation. This paper introduces a new framework called Dueling Double Deep Q-Network (D3QN) for intelligent route optimization in Globally Important Agricultural Heritage Systems (GIAHS). The route planning problem is defined as a dynamic Markov Decision Process which integrates tourist preferences, real-time environmental restrictions and sustainability measures. The multi-objective reward function combines four dimensions — tourist satisfaction, route efficiency, heritage sustainability, and personalization which are weighted based on the grid-search optimization and in keeping with the conservation principles of GIAHS. The framework includes carrying capacity and activity calendars for agriculture in the state representation. Importantly, no real-world field deployment has been done; all experiments are done in a simulated environment. The experiments show that D3QN can achieve 18.5% higher tourist satisfaction (<i>p</i> &lt; 0.001, d = 1.42) and 23.3% higher route efficiency (<i>p</i> &lt; 0.001, d = 1.67) than the baseline methods (genetic algorithms) and can also improve the compliance of the route to the sustainability indicators by 31.6% (<i>p</i> &lt; 0.001, d = 1.89), while allowing inference of the route in real time, with a latency of about 150ms per route, using the Honghe Hani Rice Terraces (UNESCO World Heritage Site) as a reference scenario. The sensitivity of the parameters to changes in the tourists’ preference distribution supports the generalizability of the findings in the simulated setting. The findings indicate that deep reinforcement learning offers a promising paradigm for managing sustainable tourism in agricultural heritage sites, with pilot implementation in real-world cases pending.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Intelligent route optimization for agricultural heritage tourism using deep reinforcement learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月3日

查看全文
  • Jiya Sun

摘要

Agricultural heritage tourism plays an important role in the conservation of traditional agro-landscapes and the development of rural areas. Current route optimisation algorithms, mainly used for urban tourism, are unsuitable for the multi-objective constraints of agricultural heritage environments, such as seasonal farming activities, ecological sensitivity, carrying capacity constraints and socio-cultural heritage preservation. This paper introduces a new framework called Dueling Double Deep Q-Network (D3QN) for intelligent route optimization in Globally Important Agricultural Heritage Systems (GIAHS). The route planning problem is defined as a dynamic Markov Decision Process which integrates tourist preferences, real-time environmental restrictions and sustainability measures. The multi-objective reward function combines four dimensions — tourist satisfaction, route efficiency, heritage sustainability, and personalization which are weighted based on the grid-search optimization and in keeping with the conservation principles of GIAHS. The framework includes carrying capacity and activity calendars for agriculture in the state representation. Importantly, no real-world field deployment has been done; all experiments are done in a simulated environment. The experiments show that D3QN can achieve 18.5% higher tourist satisfaction (p < 0.001, d = 1.42) and 23.3% higher route efficiency (p < 0.001, d = 1.67) than the baseline methods (genetic algorithms) and can also improve the compliance of the route to the sustainability indicators by 31.6% (p < 0.001, d = 1.89), while allowing inference of the route in real time, with a latency of about 150ms per route, using the Honghe Hani Rice Terraces (UNESCO World Heritage Site) as a reference scenario. The sensitivity of the parameters to changes in the tourists’ preference distribution supports the generalizability of the findings in the simulated setting. The findings indicate that deep reinforcement learning offers a promising paradigm for managing sustainable tourism in agricultural heritage sites, with pilot implementation in real-world cases pending.