This subject introduced flow field analysis and statistical learning methods to build wind retrieval and calibration model based on neural networks identification and support vector machine, calculated the wind speed and direction data of undisturbed wind field, reduced the deviation of wind speed and direction between disturbed wind field and wind field that wasn't disturbed. This subject made use of radial basis probabilistic neural networks and support vector machine method to analyze and research, used machine learning methods to get the exclusion model of meteorological data to improve the effectiveness of monitoring data. This subject used smoothing and threshold eliminating techniques, adopted radar reflectivity threshold identification algorithm to realize the identification of dangerous weather, such as thunderstorm. MCT coupler technology and message passing parallel computing was used to achieve regional air-sea mode coupling. Forecasting methods such as dynamic diagnosis, support vector machines, multi-index nesting were adopted to establish a shipborne forecasting model of maritime thunderstorm and clouds.%课题采用流场分析和统计学习的方法，建立基于神经网络系统辨识和支持向量机的风场反演校正模型，计算出未受干扰风场中的风速风向数据，减小干扰风场与未受干扰风场风速风向之间的偏差；利用径向基概率神经网络和支持向量机方法进行分析研究，借助机器学习方法得到气象要素数据奇异值剔除模型，从而提高监测数据的有效性；应用区域平滑滤波和阈值剔除技术，采用基于雷达反射率阈值的识别算法，实现雷暴等危险天气的识别；采用MCT耦合器技术及消息传递的并行计算方式，实现区域海气模式耦合；采用动力诊断、支持向量机、多指标叠套等预报方法，建立海上雷暴、云的船用预报模型。
展开▼
