适用于风速预测的微分神经网络方法及其在风能利用中的应用研究

In the process of wind energy utilization, to improve the forecasting skill of local wind speed is the urgent and key problem. Numerical Weather Prediction (NWP) is currently the popular method to forecast wind speed for wind industry, but the NWP models always have large uncertainties in the case of point forecasts near the ground. And there is no widely applicable methods of uncertainty modeling and optimization. Under this circumstance, this application explores the intersection of NWP simulation and artificial intelligence in the field of wind forecasting, which is to research the differential neural network for wind speed forecasts and its application wind energy utilization. This application will be based on the NWP simulation and the utilization of multiple meteorological data, to develop the theory and method of NWP wind speed forecasts combining with the principle of atmospheric dynamics and artificial intelligence, and to solve the problem of differential function within the forecasting process. The completion of this process will develop new methods for the solution of different equation and new ideas of wind speed forecasts, and will provide theories and research basis for the related uncertainty optimization problems, and will provide practical techniques for wind speed forecasts of wind farms. Therefore, this application has its important theoretical and practical significance.

在风电开发利用过程中，提高风速预测水平是当前亟待解决的热点问题和关键技术问题。使用数值天气模式进行风速预测是目前风电行业的普遍趋势，但数值模拟在近地层点预测中通常存在较大的不确定性，且尚未提出具有广泛适用性的不确定性建模和优化方法。鉴于此，本项目探索气象数值模式与人工智能方法在风电预测领域的交叉，研究适用于风速预测的微分神经网络方法及其在风能利用中的应用，以气象数值模式为基础，综合利用多元气象数据，发展大气动力学原理与人工智能相结合的风速数值预测的理论和方法，解决并优化其中的微分函数求解问题。本项目的完成，将发展微分函数求解的新方法和风速预测的新思路，为相关的不确定性优化方法提供理论与研究基础，提供适用于风电场风速预测的实用技术，具有重要的理论和现实意义。

随着风力发电的快速发展和风电在电网中比例的不断增大，提高风速预测水平是进一步提高风电穿透比、维护电网安全稳定运行的重要手段，也是当前亟待解决的关键技术问题。针对当前风速预测中的数值预报不确定性大、优化建模脱离大气运动的物理规律等实际问题，本项目的研究以气象数值模拟为基础，综合利用气象观测、数值模式输出等资料，在研究数值模拟与人工智能交叉的基础上，发展了适用于风速数值模拟的特征数据挖掘方法，构造并改进了多元自组织神经网络模型及其微分形式网络，实现了风速数值预测复杂目标函数的优化求解，进一步发展了基于集合数值模拟的风速预测的不确定性优化方法，并探讨了其中的模型复杂度、集合成员优化、样本分类策略等问题。项目开展了我国山东省、宁夏回族自治区、等不同地区的风电场风速预测的模拟试验，结果显示，本项目发展的预测模型能够在实际风力发电中提高风电场风速预测的准确性、可靠性和稳定性。