分布式光伏接入的用户侧微电网功率预测方法

发布时间：2019-04-12 11:42

【摘要】：伴随我国新型城镇化建设,在增加用电需求的同时为分布式能源接入创造了环境。通过用户侧微电网进行分布式光伏集成应用,是实现新能源就地消纳,降低碳排放与环境污染的重要途径。国家相关政策已对分布式光伏未来一段时间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相关技术支持。分布式光伏接入用户侧微电网后,为保证微电网与配电网的稳定与经济运行,相关微电网负荷预测技术与分布式光伏功率预测技术需要结合用户侧微电网的应用特点进行深入研究。提出采用核函数极限学习机构建功率预测模型,采用粒子群算法离线优化预测模型相关参数,对于在线功率预测系统,在保证预测精度的同时重点降低预测模型复杂度,从而构建离线参数寻优与在线功率预测相结合的预测方法。阐述国内外功率预测技术的发展现状,同时简要论述核函数极限学习机和粒子群算法的相关理论基础。 (1)考虑到微电网负荷波动较大,使用分时训练样本进行参数寻优,获得一天各待预测时刻的最优参数。为提高负荷预测系统运行效率,仅选择同类型日期的高相关时段历史数据进行模型训练。对于平均负荷140千瓦至1300千瓦的四个不同类型的微电网分别进行1个月的负荷预测,周预测误差通常小于10%,最大不超过15%。由于微电网负荷可能在较短时间内出现较快增长,研究中对预测模型参数采用周期更新的方式,且在更新后能保持原有负荷预测精度。 (2)对于分布式光伏功率预测,使用基于属性权重的训练样本筛选机制来降低预测模型构建复杂度。预测方法基于低成本的气象信息记录值而非数值天气预报,针对几十千瓦级的分布式光伏系统进行1个月的功率预测,预测误差约16%至18%。同时可根据属性权重值简化预测模型,在预测精度基本不变的条件下进一步降低计算时间。此外在分布式光伏随机覆尘或逆变器部分故障等条件下,预测模型无需人为干预或更新参数,即可保持原有预测精度与效率。
[Abstract]:Along with the construction of new urbanization in China, the environment for distributed energy access is created while increasing the demand for electricity. Distributed photovoltaic integrated application through user-side microgrid is an important way to realize on-site consumption of new energy sources and reduce carbon emissions and environmental pollution. The national relevant policies have made relevant plans for the rapid development of distributed photovoltaic in the future, and the State Grid Corporation has also issued policies to provide convenience and related technical support for distributed photovoltaic access. When distributed photovoltaic is connected to the user-side microgrid, in order to ensure the stability and economic operation of the microgrid and distribution network, The related microgrid load forecasting technology and the distributed photovoltaic power forecasting technology need to combine the user side microgrid application characteristics to carry on the in-depth research. In this paper, the kernel function limit learning mechanism is used to build the power prediction model, and the particle swarm optimization algorithm is used to optimize the parameters of the prediction model offline. For the on-line power prediction system, the prediction accuracy is guaranteed and the complexity of the prediction model is reduced. Thus, a prediction method combining off-line parameter optimization with on-line power prediction is constructed. The development status of power prediction technology at home and abroad is described, and the theoretical basis of kernel function limit learning machine and particle swarm optimization is briefly discussed. The main results are as follows: (1) considering the large load fluctuation of microgrid, the optimal parameters of each time to be predicted are obtained by using time-sharing training samples to optimize the parameters. In order to improve the efficiency of load forecasting system, only the historical data of high correlation period with the same type of date are selected for the training of the model. For four different types of microgrids with an average load of 140kW to 1300 kW for one month, the weekly forecast error is usually less than 10 percent and the maximum is no more than 15 percent. Because the load of microgrid may increase rapidly in a short time, the forecasting model parameters are updated periodically in the study, and the original load forecasting accuracy can be maintained after updating. (2) for the distributed PV power prediction, the training sample selection mechanism based on attribute weight is used to reduce the complexity of the prediction model. The forecasting method is based on low-cost meteorological information records rather than numerical weather forecasts. The prediction error is about 16% to 18% for a distributed photovoltaic system with several tens of kilowatts. At the same time, the prediction model can be simplified according to the attribute weight value, and the calculation time can be further reduced under the condition that the prediction precision is basically constant. In addition, under the conditions of distributed photovoltaic random dust cover or partial inverter failure, the prediction model can maintain the accuracy and efficiency of the original prediction without human intervention or updating of parameters.
【学位授予单位】：华北电力大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TM615

【参考文献】