| 引用本文: | 黄南天,唐立鹏,戴千斌.计及NWP风速横纵向误差的循环确定性策略梯度风速修正模型[J].电力系统保护与控制,2024,52(23):10-21.[点击复制] |
| HUANG Nantian,TANG Lipeng,DAI Qianbin.Wind speed correction method for numerical weather prediction based on a recurrent deterministic policy gradient algorithm considering lateral and longitudinal error[J].Power System Protection and Control,2024,52(23):10-21[点击复制] |
|
| 摘要: |
| 目前数值天气预报(numerical weather prediction, NWP)风速修正方法大多基于纵向误差的幅值特性评估建模，忽视了NWP风速序列与历史风速序列间超前或延迟效应导致的横向误差对NWP风速修正的潜在影响。因此，提出计及NWP风速横纵向误差的循环确定性策略梯度(recurrent deterministic policy gradient, RDPG)风速修正模型。首先，基于加权欧式距离相似度确定各个NWP风速待修正点超前/滞后时序区间(lead/lag timing interval, LLTI)边界，构成强相关时序区间。然后，采用LLTI内统计特征代替该区间NWP原始气象特征，并基于沙普利可加性解释法(shapley additive explanation, SHAP)归因理论，根据每个特征的边际贡献评估不同特征对模型输出的影响程度。最后，建立基于RDPG的NWP风速修正模型。通过算例分析，验证了所提方法在超短期、短期预测的不同时间尺度下修正NWP风速的有效性和可行性。 |
| 关键词: 数值天气预报 横向误差 纵向误差 超前/滞后时序区间 循环确定性策略梯度 |
| DOI：10.19783/j.cnki.pspc.240541 |
| 投稿时间：2024-05-06修订日期：2024-08-20 |
| 基金项目:吉林省科技发展计划项目资助(20210201126GX) |
|
| Wind speed correction method for numerical weather prediction based on a recurrent deterministic policy gradient algorithm considering lateral and longitudinal error |
| HUANG Nantian1,TANG Lipeng1,DAI Qianbin2 |
| (1. Key Laboratory of Modern Power System Simulation and Control & Renewable Energy Technology, Ministry of Education
(Northeast Electric Power University), Jilin 132012, China; 2. Liaoning Pushihe Pumped Storage Co., Ltd., Dandong 118216, China) |
| Abstract: |
| The current numerical weather prediction (NWP) wind speed correction methods are based on the amplitude characteristic evaluation modeling of the longitudinal error. They ignore the potential impact of the lateral error caused by the lead or delay effect between the NWP wind speed series and historical wind speed series on NWP wind speed correction. Therefore, an NWP wind speed reinforcement correction method based on the recurrent deterministic policy gradient (RDPG) algorithm and considering lateral and longitudinal errors is proposed. First, based on weighted Euclidean distance similarity, the lead/lag timing interval (LLTI) boundary of each NWP wind speed to be corrected is determined to form a strongly correlated timing interval. Subsequently, the statistical characteristics of the LLTI are used to replace the original meteorological characteristics of the NWP. Based on the Shapley additive explanations (SHAP) attribution theory, the influence of different features on the model output is evaluated using the marginal contribution of each feature. Finally, an NWP wind speed correction model based on the RDPG is established. Through an example analysis, the effectiveness and feasibility of the proposed method for modifying the NWP wind speed on different time scales of ultra-short-term and short-term predictions are verified. |
| Key words: numerical weather prediction longitudinal error lateral error lead/lag timing interval recurrent deterministic policy gradient |
