【作者】 孙川永；

【导师】 王式功； 罗勇； 陶树旺；

【作者基本信息】 兰州大学 ， 气象学， 2009， 博士

【摘要】 风电产业的大规模发展使风电对电网的冲击问题越来越凸显，许多地方出现了拉闸限电的情形，随着百万千瓦级风电基地、千万千瓦级风电基地的规划及建设，急需一套行之有效的风电场风电功率预报系统来满足风电上网调度的实际需求，为国家大力发展清洁能源的政策提供技术保障。国外在此方面已有十余年的发展历史，而国内目前的风电功率预报方法主要集中于一些统计方法，并不能满足风电上网调度的精度和时效要求。采用数值预报模式和风电功率统计预报模型相结合的集成系统进行预报，是解决风电场风电功率短期预报的有效方法，基于此本论文以RAMS区域大气模式为工具对高分辨的风电场风速预报以及风电功率预报问题进行了初步探讨。风电场风电功率预报的前提就是风机叶片扫风范围内准确的风速预报，而对于利用数值模式进行风速预报而言，由于模式稳定性及网格结构以及风机机型的差异，数值模式很难在涡轮及叶片扫风高度范围内做出详细的计算。通过统计方法找出各高度间风速的分布规律是解决这一问题的最直接的方法。通过分析我国内陆河北省张北县地区和吉林地区风电场内的风速变化及风速廓线变化特性发现，虽然两地的风速分布形态不一样，但各高度间风速的差异分布大体相同：即由夜间到白天逐渐缩小，在中午达到最小，由白天到夜间再逐渐增大，并且在各个阶段又相对稳定。即在日出后由地面向上的热量输送逐渐增强，湍流加强，各层间的风速差异减少，并迅速趋于稳定；日落湍流减弱，各层间的风速差异迅速增大，并趋于夜间时段的相对稳定。这一规律的发现对解释涡轮高度不同时间相同风速条件下风机出力不同及风电功率建模有重要意义。通过对海陵岛的60米高测风塔不同高度的NRG测风资料分析发现：当沿海地区有海陆风发生时，海风和陆风阶段风速廓线存在较大差异，海风阶段风速的垂直切变明显小于陆风阶段。并发现即使没有海陆风发生，当风向为海洋吹向陆地时，风速随高度的垂直切变同样小于陆地吹向海洋的时段。这对模式预报风速的订正及沿海地区风电功率预报建模有很高的实用价值。将SRTM3 90米分辨率地形资料引入到模式当中，有效分辨率可达500米，实现了模式的高分辨率预报。通过对张北地区空间分辨率为1km，时间分辨率为1小时的风速预报结果分析，发现RAMS模式风速预报结果基本可以满足风电场风速预报的要求，与国际上的风速预报水平相当，1-84小时的平均均方根误差为2-3m/s。为了对数值模式预报结果进行改进，利用人工神经网络方法进行了订正试验，通过预报因子的选择试验，本文初步确定了利用预报风速、风向、气压场的主成分作为预报因子进行订正的方案。通过神经网络方法，将风速廓线随时间的变化引入神经网络方法中，确定了选择不同高度的风速作为风电功率预报因子的预报方案，对1-72个预报时刻分别进行风电功率预报建模。并比较了“单机法”与“整体法”两种预报方式下风电场风电功率预报的误差，发现“单机法”效果好于“整体法”。通过风电功率预报实验分析发现，2008年3月份风电场风电功率预报前42小时的平均均方根误差占装机容量的百分比为17．5%，72小时总的平均均方根误差占装机容量的百分比为21%，2008年4月份72小时总的平均均方根误差占风电场装机容量的百分比为24%，与国际上的平均功率预报误差基本持平，但还有待改进。更多还原

【Abstract】 As the wind power industries are developing cosmically,The impulse of windenergy to the grid network is protruding. In many places, the situation of wind farm isout of the grid network exits. As the million kilowatt and multimillion kilowatt windpark base being layout and constructed, a system of wind power forecasting is neededimminently to solve the problem of the compatible between grid network and windenergy, which can provide technique for the policy of develop clean energy large-scaleForeign countries have been developing short-term wind power forecasting techniquesfor more than 10 years, but in China the method of wind power forecasting focus onthe statistical method, which can’t satisfy the request of precision and time in gridnetwork operation. An integrated system, which is a combination of both numericalprediction models and statistical power capacity models, is used to make forecasting.This is an effective approach for wind power forecasting for wind farms. In this article,we use the regional climate model RAMS to give the forecasting of wind speed inwind farm with a high resolution, and we establish a system of wind power forecasting.The precondition of wind power forecasting is the precision of wind speedforecasting between the distance of turbine blade. But for wind speed forecasting usingnumerical model, Because the different type of wind turbine and the model’scharacteristic the numerical model hardly can compute the wind speed at hub heightand wind speed at different height in the distance of wind turbine blade in detail. Theimmediate method to solve this problem is to find the statistical relation between windspeed at different height.By analyzing the wind speed variety and wind difference between different heightof wind farm at Zhangbei and Jilin district, we found that although the wind speed isdifferent for these two places, but the wind difference is very similar between differentheight: wind difference of different height is becoming smaller from night to day andreach the smallest value at noon, wind difference of different height is becoming largerfrom day to night, and show a stabilization condition at each stage. After the sunrise,the heat transmission increases, the air turbulence increases, the wind difference between different height is becoming smaller and keep steady. After sunset the airturbulence decrease the wind difference between different height also goes down, andkeep steady. This rule is a good explanation for the different output of turbine with thesame wind speed at different time during a day, and it’s very important for wind powerprediction.By analyzing the observation wind speed at different height from a wind tower of60m, we find that The wind profile have great difference when land-sea breezehappens. Wind shear is smaller when blowing form sea than blowing from land. eventhere is no land-see breeze when wind blowing from sea the wind profile shear is alsosmaller than blowing from land. This rule is very important for the improvement ofwind speed prediction.And it has great value for the wind power forecasting along thecoast.By using the SRTM3 topographty data which with a resolution of 90m, we cancarry out the forecasing of 500m resolution in operation. By analyzing the wind speedforecasting result with a resolution of 1 km and with time interval 1 hour, we found thatRAMS model can give the proper wind speed forecasting which can reach the demandof wind speed forecasting in wind farm, and the forecasting error of wind speed isclose to the error of wind speed abroad, the average root-mean-square error is 2-3m/sfrom1 to 84 hours. To improve the forecasting result of numerical model, we use theartificial neural network method to make the test. By choosing different forecastingfactors we confirm the method using wind speed, wind direction and the principalcomponent of air pressure to improve the result.Finally, we introduce the rule of wind profile changing as time past by intoArtificial Neural Network. we use the wind speed at different height to forecasting thewind power of wind farm, and set up 72 wind power forecasting model fore every hour.By compare the average root-mean-square error between "single turbine method" and"the whole wind farm method", we find "single turbine method" is better than"thewhole wind farm method". Through the test of wind power forecasting in March, 2008,we find the average root-mean-square wind power forecasting error of the first 42hours is 17.5% of the nameplate capacity of the wind farm, The averageroot-mean-square wind power forecasting error of the 72 hours is 21% of the nameplate capacity of the wind farm. For the test of April,2008 The averageroot-mean-square wind power forecasting error of the 72 hours is 24% of thenameplate capacity of the wind farm. The average root-mean-square wind powerforecasting error is close to the results abroad, but it’s needed improved further.更多还原