【作者】 周松林；

【导师】 苏建徽； 茆美琴；

【作者基本信息】 合肥工业大学 ， 电力系统自动化， 2013， 博士

【摘要】 风力发电是微电网的重要单元，风速及风电功率的随机性对微电网运行产生了不利影响。研究风速及风电功率预测技术，并在此基础上进一步预测风速的随机性对微电网潮流分布的影响，对微电网能量管理具有重要意义，本文对此课题展开了深入研究。提出了小波-支持向量机以及基于改进型差分进化算法的脊波神经网络等新型预测模型以提高预测精度、建立了概率预测模型对风速及风电功率预测结果的不确定性进行评估、建立了微电网潮流预测模型、分析风速、光照的随机性对微电网潮流的影响。通过上述研究取得了一些创新性研究成果，所提出的方法和结论能够为微电网能量管理提供有价值的参考。本文的主要研究内容及创新点可概括如下：1.在小时风速预测方面，研究基于历史气象数据的支持向量机风速预测方法，采用自适应粒子群算法对最小二乘支持向量机模型参数进行优化；将小波分解技术和支持向量机相结合进行短期风速预测。利用小波分解技术将历史训练集的风速时间序列分解成基础低频分量和高频概貌分量，再使用支持向量机预测各个小波分量，最后将预测结果分别予以重构得到最终风速预测值。同时提出一种风速预测误差修正方法，利用误差的预测值对初步预测风速进行校正从而提高了预测精度。2.在小时风电功率预测方面，采用改进型差分进化算法训练脊波神经网络预测模型，通过直接预测和间接预测两种方法分别预测风电功率。研究表明，所提出的脊波神经网络对高维函数的逼近能力、泛化性能和训练速度都有一定改善，提高了风电功率预测精度。3.在不确定性预测方面，分析了风速及风电功率预测结果的不确定性，建立了概率预测模型对风速及风电功率的预测结果进行不确定性评估。不确定性预测包含两个方面，一个是求出预测风速在某个范围内实际发生的概率，本文采用条件概率计算方法求解这一问题；另一个是在分析风电功率预测误差及功率波动的统计规律基础上，采用非参数区间估计方法求取在一定置信度下的风电功率置信区间。4.在微电网潮流预测方面，研究了风速和光照随机性对微电网潮流的影响。在风速及风电功率预测的基础上，将马尔科夫链模型和蒙特卡洛模拟方法相结合，研究微电网概率潮流预测，预测风速的变化对微电网在并网和孤岛两种运行状态下的潮流所产生的具体影响；建立了风速和光照条件联合概率预测模型，研究同时考虑风、光随机性时，微电网的概率潮流分布。更多还原

【Abstract】 Wind generation system is an important unit of a microgrid. The randomness of wind and windpower has a great adverse influence on operation characteristic of a microgrid. It is of greatsignificance for energy management system of microgrid to predict power flow based on theprediction of wind speed and wind power production. To address these problems, wavelet andsupport vector machine prediction model and the ridgelet neural network trained by improveddifferential evolution algorithm are proposed to enhance prediction precise of wind speed and windpower production. Futher, a probabilistic prediction model is established to assess the uncertaintyof predictive results. Based on these results, the power flow prediction model of a microgrid isestablished to predict and analyze the influence of wind randomness on the microgrid. Theoutcomes of these researches are of great reference value for energy management system of amicrogrid. The main innovative achievements are as follows:1. For the aspect of hourly wind speed prediction, the support vector machine (SVM) predictionmethods by the historical meteorological data are studied. The parameters of least square supportvector machine (LSSVM) are determined by adaptive particle swarm optimization (APSO). A windspeed forecasting model based on wavelet and support vector machine (Wavelet-SVM) isestablished also in which the original wind speed sequences are decomposed into coarse componentsand detail components firstly, and each wavelet component is separately forecasted bycorresponding support vector machine model. Finally, the forecasting results of original wind speedseries are achieved by wavelet reconstruction. In additional, a new forecast method based onpredictive error correction is presented in the paper. The preliminary predicted wind speed iscorrected with predict error to improve the prediction accuracy.2. For the aspect of hourly wind power prediction, the ridgelet neural network based on improveddifferential evolution algorithm is adopted to predict wind power production by the means of directlyprediction and indirectly prediction methods. The simulation results show that the approximationability for high-dimensional function, generalization performance, and train speed of the proposedridgelet neural network are improved. By the prediction model, the wind power productionpredictive precise is enhanced.3. Based on the analysis of uncertainties of wind speed and wind power prediction, theprobabilistic prediction models are established to assess the uncertainty of wind and wind powerpredicted results from two aspects, one is to calculate the probability of actual wind speed within acertain range by the conditional probability calculation method, and the other is to calculate theconfidence intervals of predictive wind power corresponding to a certain confidence level by meansof non-parametric confidence interval estimation on basis of analyzing the statistical regularities of power forecast errors and power fluctuation values.4. For the power flow prediction of microgrid, the influence of the wind randomness on theoperation characteristic of a microgrid is studied firstly, and then Markov process and Monte Carlosimulation methods are combined together to carry out probabilistic forecasting of power flow.Finally, the impact of wind speed on the microgrid power flow is analyzed under the operationmodes of both grid-connection and off-grid. In additional, a conditional joint probability model inview of the randomness of wind and light is established to research the probabilistic power flow ofa microgrid.更多还原