【作者】 张晋华；

【导师】 田德； 刘永前；

【作者基本信息】 华北电力大学 ， 可再生能源与清洁能源， 2014， 博士

【摘要】 随着电力系统中风电并网比例的增加,风能的随机波动性对传统电力系统经济调度和安全运行带来挑战。研究在风电功率预测与电力系统的负荷约束条件下,风电场内机组优化调度问题,不仅能减少风力发电机组的冗余运行和磨损浪费,避免机组的频繁启停,还可以降低运行成本,提高风电场输出功率的电能质量,有效减轻风电波动性对电网的影响,从而在保证电力系统安全性的前提下,提高电力系统的消纳风电能力和经济效益。以风电场功率预测数据为基础,重点研究了以降低风力发电机组疲劳载荷损伤相对量和降低集电系统损耗为目标的风电场内机组优化调度的算法,完成了以下研究工作。(1)提出了不同运行工况下风力发电机组关键部件相对疲劳损伤量的计算方法。根据华北某风电场的风资源数据,利用瑞利分布的风速累积分布函数,基于GH-Bladed模拟了1.5MW风力发电机组的疲劳载荷,利用雨流循环计数法,得到风力发电机组各个部件的疲劳载荷谱,然后根据仿真计算和Miner法则得到的风力发电机组关键部件的相对疲劳损伤量,可为风电场内机组优化运行提供评价准则。(2)基于相空间重构的神经网络风电功率预测算法的应用。风电场内机组优化调度是以风力发电机组的短期和超短期功率预测值为研究基础,根据混沌-相空间重构的原理可知风力发电机组的风速和风电功率时间序列数据具有混沌的属性的基础上,将相空间重构与神经网络相结合,建立混沌-Elman、混沌-BP和混沌-Volterra级数的风电功率预测模型,经实例验证,分析比较得出混沌-Elman模型的预测效果相对较好,能够提高预测的精度和稳定性。(3)建立以风电场集电系统网损最小为目标的机组优化调度模型。以风电场内集电系统网损最小为目标函数,电网调度要求、风力发电机组有功输出的功率上下限、风力发电机组无功输出的功率上下限、风力发电机组端电压上下限、变压器变比上下限等为约束条件,建立机组优化调度的数学模型,分别采用粒子群优化算法和粒子群-遗传优化算法进行寻优。结果表明,粒子群-遗传算法在优化效果和运算效率方面均优于单一粒子群算法。(4)建立了以风电场内机械损伤量最小为目标的机组组合优化模型。基于前述的相对疲劳损伤量模型,建立机组组合模型,合理配置机组启停方案,以期在调度期内风电场整体机械损伤最小,延长机组运行效率和使用寿命。然后利用改进二进制粒子群优化算法(BPSO)、遗传优化算法(GA)、粒子群-遗传混合优化算法(BPSO-GA),进行优化求解。结果表明,BPSO-GA比单一GA和BPSO提高了优化性能,运行期间总疲劳损伤量最小；引入粒子群优化参数的BPSO-GA算法的计算时长相对BPSO算法略长,但比GA算法计算时长要短；三种模型的计算时长从大到小依次为：GA, BPSO-GA, BPSO。(5)建立了基于机组优先级分类的风电场内功率分配模型。以风力发电机组发电功率、风速平均值和均方根差值作为特征值,分析机组发电性能,并分别采用SOFM神经网络算法与基于模拟退火遗传算法的模糊C均值聚类算法建立机组优先级分类模型。将发电性能较好的一类作为优先执行发电计划的机组,计及线路损耗后的发电计划,对风电场内其余机组进行两层优化,外层是以风力发电机组相对疲劳损伤量最小为目标的机组优化出力,内层是确定机组间负荷满足电网调度要求的最优功率分配。通过两层分配,得到既满足电网调度需求,又降低风电场运行损耗的功率分配结果。结果表明,基于遗传模拟退火算法的模糊聚类算法分类方法的疲劳损伤量比自组织特征映射神经网络分类方法的疲劳损伤量较小,SAGA-FCM分类方法下停机的机组台数较多。风力发电机组分类后优化调度,能够使风电场机组运行优化,提高风电场输出电能质量。更多还原

【Abstract】 With the increase of wind power integration in power system, the stochastic variability of wind energy has been posing great challenges on traditional economic dispatch and operational security. Research on optimal dispatch in a wind farm with constrains on power system load and wind power forecasting could not only reduce operational redundancy and mechanical abrasion of wind turbines, avoid frequent on-off operation, but also decrease operation cost, improve power quality. All these above would eventually mitigate the negative impacts from wind power on power system, thus to maximize wind power penetration and economic benefits on the condition of secure system operation.Based on the wind power forecasting data, this dissertation mainly discussed the unit optimal dispatch algorithms aiming to reduce the wind turbine relative fatigue loading damage and the line loss of collection system in the wind farm. The main contribution of the dissertation can be summarized as follows:(1) A relative fatigue loss method for wind turbine key components in different operation conditions is established. Based on the wind resource data of a wind farm in Northern China, with the Rayleigh distribution of the wind speed cumulative distribution function, this dissertation simulated the fatigue load of1.5MW wind turbine using GH-Bladed and obtained the fatigue load spectrum of wind turbine components using the rain flow cycle count method. Then, according to the Miner laws and the simulation calculation, the relative fatigue damage of the wind turbine key components was calculated, providing evaluation criterion for unit optimal dispatch in the wind farm.(2) A wind power forecasting model is built based on neural network with the phase space reconstruction. Unit optimal dispatch in the wind farm is based on the wind turbine short-term and ultra short-term power forecasting. According to the principle of chaos-phase space reconstruction, it can be seen that the wind speed and the wind power time series data of wind turbines have the property of chaotic. On this basis, the phase space reconstruction was combined with neural network to establish chaos-BP and chaos-Elman, chaos-Volterra series wind power forecasting models. Tests showed that the forecasts of Elman model are more accurate than that of the others, improving accuracy and stability of the forecasting.(3) An unit optimal dispatch model is established minimizing the losses of collection system in a wind farm. With constraint conditions including wind farm output’s satisfaction of grid dispatch, upper and lower limit of the wind generator active power output, upper and lower limit of the wind generator reactive power output power, upper and lower limit of the wind generator voltage and upper and lower limit of the transformer ratio, with the minimum collecting system network loss in the wind farm being the objective function, the mathematical model of optimal dispatch was established by using particle swarm optimization algorithm and genetic-particle swarm optimization algorithm respectively. The results showed that genetic-particle swarm optimization algorithm was better than particle swarm optimization algorithm in both optimization performance and computing efficiency.(4) A unit commitment model is established with an objective function of minimization of mechanical losses in a wind farm. Based on relative fatigue damage model built before, unit commitment model was established for the rational allocation of unit start-up in wind farm, which could obtain the minimum mechanical damage in the scheduling period, prolong the operation efficiency and life. And then the improved binary particle swarm optimization algorithm, genetic optimization algorithm, genetic-particle swarm optimization algorithm were utilized to obtain an optimized solution. The results showed that BPSO-GA improved the optimal performance better than GA and BPSO, and minimized the amount of fatigue damage in runtime. The calculation time of the BPSO-GA algorithm which introduced the particle swarm optimization parameters was slightly longer than that of the BPSO algorithm, but shorter than that of the GA algorithm. The calculation time of three models in decreasing order is:GA, BPSO-GA, BPSO.(5) A model for optimally allocating wind turbine output is presented based on units classification. After the analysis of large amount of historical data of unit in the wind farm, the average and root mean square difference of each wind turbine power and the wind speed were extracted as the characteristic value to analyze the unit performance, the units were classified with the SOFM neural network algorithm and the fuzzy c-means clustering algorithm based on simulated annealing and genetic algorithm. The units with better electrical performance in the classification were given priority in the execution of the power generation plan. Considering the line loss of the power generation plan, the rest of the units in the wind farm were optimized in two layers, the outer layer being the unit commitment optimization with the target of minimum relative fatigue damage of wind turbine, the inner layer being the optimal power dispatch of the units satisfying the grid requirement. Finally the power allocation model with the output of a wind farm in compliance with the dispatch requirements of the grid was established. Both the two kinds of classification algorithm can get unit commitments meeting the requirements of power grid dispatch. After comparison, however, it is found that the fatigue damage value of fuzzy clustering algorithm based on genetic simulated annealing algorithm is smaller than that of the SOFM neural network algorithm. Optimal dispatch based on wind turbines classification could optimize units operation in wind farm and improve the power quality of wind farm.更多还原