| 引用本文: | 苏 楠,章少辉,白美健,等.灌区用水调度的知识图谱模型构建——以淠史杭灌区瓦西干渠灌域为例[J].灌溉排水学报,2023,42(11):112-120. |
| SU Nan,ZHANG Shaohui,BAI Meijian,et al.灌区用水调度的知识图谱模型构建——以淠史杭灌区瓦西干渠灌域为例[J].灌溉排水学报,2023,42(11):112-120. |
|
| 摘要: |
| 【目的】人工智能技术以全新的自适应学习视角为灌区智慧化建设带来了异于传统的研究范式，迄今已有众多智能算法与技术应用于灌区用水调度决策研究中，但已有成果缺乏便利性和易操作性，无法真正落地。【方法】为此，本文通过对比分析人工智能技术中的随机森林和BP神经网络模型，发现融合SHAP（Shapley additive explanation，SHAP）方法的随机森林模型具有更优的灌区用水调度预测效果。在此基础上，基于历史实测数据将灌区用水调度变量划分为不同梯度，表征出9.82×1011组用水调度场景样本，并采用预先训练的随机森林模型获得了各样本对应的调度流量预测值，由此形成了灌区用水调度场景与调度流量值之间映射关系的基础数据库，即在现有调度历史数据的基础上利用机器学习模型丰富和加密了调度场景，得到了能实现基本覆盖现实调度场景的调度场景库。基于该场景数据库，利用Neo4j图形库构建出淠史杭灌区用水调度流量预测值知识图谱模型。【结果】利用该图谱模型，灌区用水调度管理人员仅需确认目标调度场景中各调度变量在知识图谱模型中的近似梯度值，即可检索获得调度流量预测值。【结论】经应用验证表明，由该知识图谱模型获得的调度流量预测值误差在淠史杭灌区用水管理人员的经验认知范围内，且可实现调度流量值的实时检索。 |
| 关键词: 淠史杭灌区；用水调度；机器学习；SHAP方法；知识图谱 |
| DOI：10.13522/j.cnki.ggps.2023047 |
| 分类号: |
| 基金项目: |
|
| Construction of Knowledge Graph Model for Irrigation Water Scheduling |
|
SU Nan, ZHANG Shaohui, BAI Meijian, ZHANG Baozhong
|
|
1. China Institute of Water Resources and Hydropower Research, Beijing 100038, China;
2. State Key Laboratory of Simulation and Regulation of Water Cycle in River Basin, Beijing 100038, China
|
| Abstract: |
| 【Background and Objective】The adoption of artificial intelligence technology, particularly through adaptive learning, is reshaping the landscape of intelligent construction in the field of irrigation. This innovative perspective is divergent from traditional research paradigms, as numerous intelligent algorithms and technologies have been employed in the realm of irrigation water dispatching decisions. However, existing results often lack user-friendliness and ease of operation, preventing them from practical implementation. In light of this, this paper undertakes a comparative analysis of the Random Forest model and the BP Neural Network model in artificial intelligence technology.【Method】Building on this insight, the study divides historical data, based on measured values, into various gradients. These gradients encapsulate water irrigation scheduling variables, resulting in 9.82×1011 water dispatching scenarios. A pre-trained Random Forest model is employed to predict the dispatching flow corresponding to each scenario, ultimately constructing a fundamental database mapping the relationship between irrigation area water dispatching scenarios and dispatching flow values. This process enriches and encodes the scheduling scenario database through the utilization of machine learning models and existing scheduling history data. Leveraging this scenario database, and Neo4j technology, a knowledge map-based water dispatching flow prediction model is established for the Pishihang irrigation district.【Result and Conclusion】The key advantage of this model is its user-friendly approach. Irrigation area water scheduling managers can effortlessly determine the approximate gradient value of each scheduling variable within the knowledge graph model for a specific scheduling scenario. By doing so, they can readily access the corresponding scheduling flow prediction value through a simple query. Application results demonstrate that the knowledge map-based scheduling flow prediction model boasts a remarkable accuracy within the cognitive range of Pishihang irrigation district water management, facilitating real-time scheduling flow inquiries. This research represents a significant step toward more user-friendly and effective irrigation water management. |
| Key words: Pishihang Irrigation District; water dispatching; machine learning; SHAP method; knowledge graph |
