基于核函数极限学习机的分布式光伏短期功率预测

发布时间：2018-02-20 17:22

本文关键词： 分布式发电光伏发电预测短期功率用户侧极限学习机光伏覆尘　出处：《农业工程学报》2014年04期 　论文类型：期刊论文

【摘要】：伴随中国农村电网的较快发展,分布式光伏的集成应用是实现新能源就地消纳的重要途径。国家相关政策已对分布式光伏的快速发展进行了相关规划,国家电网公司也出台政策为分布式光伏接入提供便利条件与技术支持,相关的分布式光伏发电功率预测技术需要进行深入研究。针对用户侧分布式光伏发电系统,考虑预测系统的成本约束和运行需求,以及农村电网应用特点,提出一种基于核函数极限学习机的分布式光伏功率预测方法。对于不同容量的分布式光伏发电系统,使用核函数极限学习机构建分布式光伏短期功率预测模型,使用基于权重的训练样本筛选方法提高预测模型计算效率,并通过粒子群算法优化模型参数。预测模型使用低成本的非数值天气预报采样信息,对几十千瓦级的分布式光伏,预测相对误差仅16%～18%,能在低功耗处理器上实现10ms内完成单次发电功率预测,在简化低权重属性后能基本保持原有精度,同时在分布式光伏随机覆尘或逆变器故障条件下预测误差基本不变,具有较高的适应能力。
[Abstract]:With the rapid development of rural power grid in China, the integrated application of distributed photovoltaic (DPV) is an important way to realize the local absorption of new energy. The State Grid Company has also issued policies to provide convenient conditions and technical support for distributed photovoltaic access. The related distributed photovoltaic power prediction technologies need to be deeply studied. Considering the cost constraints and operational requirements of the prediction system and the characteristics of rural power grid applications, a distributed photovoltaic power prediction method based on the kernel function extreme learning machine is proposed. The distributed PV short-term power prediction model is constructed by using kernel function extreme learning machine, and the computational efficiency of prediction model is improved by training sample screening method based on weight. The model parameters are optimized by particle swarm optimization algorithm. The prediction model uses low-cost non-numerical weather prediction sampling information, for tens of kilowatts of distributed photovoltaic, The relative error of prediction is only 16 / 18. It can realize the prediction of single generation power within 10 Ms on a low power processor, and can keep the original precision after simplifying the low weight attribute. At the same time, the prediction error is basically unchanged under the condition of distributed photovoltaic dusting or inverter fault, and it has high adaptability.
【作者单位】：华北电力大学电气与电子工程学院;
【基金】：国家自然科学基金资助项目(51277067)资助
【分类号】：TM615

【参考文献】