含分布式储能系统的主动配电网负荷预测模型构建OA
Construction of an active distribution network load forecasting model with distributed energy storage system
针对含分布式储能系统的主动配电网短期负荷预测中储能运行扰动显著、节点电压波动增强、负荷演化非平稳性突出的特点,本研究围绕储能接入条件下负荷预测精度与物理一致性难以兼顾的问题,构建了一种融合电压波动感知与拓扑约束的短期负荷预测模型.建模过程中,首先面向工程场景中储能充放电数据缺失、记录不连续等问题,结合节点功率平衡关系与时间连续性约束,对缺失时段储能功率进行重构,以提高关键输入变量的完整性与可用性;随后,针对储能调节引起的局部电压变化对负荷响应的影响,构建节点电压波动指标,并对电压特征进行修正,将历史负荷、重构后的储能功率、节点电压及其波动特征进行统一时间对齐和多变量融合;在此基础上,建立LSTM-PatchTST联合预测框架,由LSTM提取局部时序动态特征,PatchTST刻画长时段依赖关系,并通过残差修正与结果融合增强模型在复杂工况下的稳定性;同时,将储能接入线路状态和线路载流限制以拓扑约束损失项的形式嵌入训练过程,以提升预测结果对实际配电网运行边界的适配能力.以某含分布式储能系统的主动配电网半年运行数据为对象开展仿真分析,在15 min的时间尺度下构建训练集、验证集和测试集.结果表明,所建模型能够较准确跟踪储能调节活跃时段的负荷变化趋势,在负荷突升和回落区间均表现出较好的响应能力,综合平均绝对百分比误差为2.73%,较两种对比方法分别下降23.53%和32.43%.实验结果表明,通过协同引入储能功率重构、电压波动感知和拓扑约束训练,可有效增强模型对源网荷储耦合运行特征的表征能力,提高主动配电网短期负荷预测的精度、稳定性与工程适用性,并可为储能调度优化和配电网智能运行提供更可靠的数据支撑.
Short-term load forecasting(STLF)in active distribution networks(ADNs)with distributed energy storage systems(DESS)is complicated by operational disturbances,pronounced voltage fluctuations,and highly non-stationary load patterns.To improve forecasting accuracy while ensuring physical consistency under DESS integration,a short-term load forecasting model integrating voltage fluctuation awareness and topological constraints is proposed.First,to handle missing or discontinuous DESS data in practical engineering scenarios,a power reconstruction method is developed based on nodal power balance and temporal continuity constraints,thereby improving the completeness of key input variables.Second,to capture the influence of DESS-induced local voltage variations on load response,a nodal voltage fluctuation index is constructed to refine voltage features,followed by the fusion of historical load,reconstructed DESS power,and time-aligned voltage characteristics.A hybrid LSTM-PatchTST framework is then established,in which LSTM captures local temporal dynamics and PatchTST characterizes long-term dependencies,while residual correction and ensemble learning further improve model stability.In addition,DESS capacity boundaries and line flow limits are incorporated into the training process as topological constraint loss terms to enhance model adaptability to the physical operating limits of the grid.Simulation results based on half-year operational data from a real ADN demonstrate that the proposed model accurately tracks load trends during active DESS regulation.At a 15-minute resolution,the model achieves a mean absolute percentage error(MAPE)of 2.73%,corresponding to a reduction of 23.53%and 32.43%compared with two benchmark methods.The results indicate that the synergistic integration of power reconstruction,voltage awareness,and topological constraints significantly enhances the ability of the model to represent source-network-load-storage coupling and provides reliable data support for DESS dispatch and smart grid operation.
杨韵;程永峰;谢祥中;涂炼;廖晔
广东电网有限责任公司电力调度控制中心,广东 广州 510220广东电网有限责任公司电力调度控制中心,广东 广州 510220广东电网有限责任公司电力调度控制中心,广东 广州 510220广东电网有限责任公司电力调度控制中心,广东 广州 510220北京清能互联科技有限公司,北京 100084
信息技术与安全科学
分布式储能系统主动配电网负荷预测电压波动感知储能-负荷耦合拓扑约束建模
distributed energy storage systemactive distribution networkload forecastingvoltage fluctuation awarenessstorage-load couplingtopology-constrained modeling
《储能科学与技术》 2026 (4)
1302-1311,10
中国南方电网有限责任公司科技项目(036000KK52210065(GDKJXM20210096)).
评论