基于机理模型与LSTM混合预测的风电富集区用电需求评估与消纳分析——以内蒙古为例

Demand Assessment and Accommodation Analysis of Electricity Consumption in Wind Power-Enriched Areas Based on a Hybrid Mechanistic Model and LSTM Forecasting—A Case Study of Inner Mongolia

摘要: 针对风电富集区弃风限电问题中风电出力与用电负荷时序错配的现实困境，本文以内蒙古自治区为研究对象，依托中国矿业大学(北京)理学院与美林数据技术股份有限公司共建的“科技矿场”校企合作平台，基于大学生创新训练项目实践，构建了一套面向需求侧的风电消纳分析与评估方法。研究首先基于社会用电量、最大负荷及负荷率等指标，对区域用电需求规模与时序结构进行刻画，并建立综合需求评估框架；随后以乌兰察布典型风电富集区为例，将用电需求分解为长期趋势与短期波动，分别采用机理模型与LSTM网络进行建模，形成混合预测方法，并通过多种误差指标对预测性能进行检验；在此基础上，将预测负荷引入风电消纳约束关系，设置不同需求侧调节强度情景，对可调节负荷提升对风电消纳能力的影响进行量化评估。结果表明：趋势项与实际用电需求变化方向高度一致(相关系数为0.921)，混合预测模型整体误差较小(RRMSE为7.41%，且相较单一LSTM模型显著降低了预测误差)；在消纳情景分析中，基准情景下风电消纳率约为0.85，而在中度与强化需求侧调节情景下分别提升至0.90和0.95。该实践表明，“科技矿场”平台提供的真实工程场景与数据支持，有效促进了学生将数学方法(机理模型、LSTM等)应用于复杂能源问题的能力。研究表明，仅依赖用电需求总量的自然增长难以有效缓解风电消纳受限问题，而通过需求侧调节优化负荷时序结构，在不额外增加电源侧装机规模的前提下即可显著提升风电消纳水平，从而为风电富集区从需求侧挖掘消纳潜力提供了可行的分析思路与定量依据。

Abstract: To address the mismatch between wind power generation and electricity demand in wind power-rich regions, which leads to wind curtailment, this study takes Inner Mongolia Autonomous Region as the research object. Relying on the university-enterprise collaborative platform “Technology Mining Field” jointly established by the School of Science, China University of Mining and Technology (Beijing) and Meilin Data Technology Co., Ltd., and based on the practice of an undergraduate innovation training project, a demand-side-oriented framework for wind power consumption analysis and evaluation is developed. First, regional electricity demand is characterized in terms of scale and temporal structure using indicators such as total electricity consumption, peak load, and load factor, and a comprehensive demand assessment framework is established. Then, taking Ulanqab, a typical wind power-rich region, as a case study, electricity demand is decomposed into long-term trends and short-term fluctuations. A hybrid forecasting method combining a mechanism model and an LSTM network is proposed, and its performance is evaluated using multiple error metrics. On this basis, the predicted load is incorporated into wind power consumption constraints, and different demand-side regulation scenarios are designed to quantitatively assess the impact of flexible load enhancement on wind power integration capacity. The results show that the trend component is highly consistent with the actual electricity demand (correlation coefficient of 0.921), and the hybrid forecasting model achieves relatively low overall error (RRMSE of 7.41%), significantly outperforming the single LSTM model. In the consumption scenario analysis, the wind power utilization rate is approximately 0.85 under the baseline scenario, and increases to 0.90 and 0.95 under moderate and enhanced demand-side regulation scenarios, respectively. The study demonstrates that the “Technology Mining Field” platform, by providing real engineering scenarios and data support, effectively enhances students’ ability to apply mathematical methods (such as mechanism models and LSTM) to complex energy problems. Furthermore, it indicates that relying solely on the natural growth of electricity demand is insufficient to alleviate wind power curtailment, whereas optimizing the temporal structure of load through demand-side regulation can significantly improve wind power integration without increasing generation capacity, thereby offering a feasible analytical approach and quantitative basis for exploiting demand-side potential in wind power-rich regions.