Research On Application Of Parallel Computing In Nowcasting System

Nowcasting is a technology of weather forecast for the next 0 to 6 hours weather conditions and possible disasters in advance. nowcasting system can provide meteorogical data for the government and the disaster relief related department to reference and weather forecast,can make may face a disaster area personnel evacuation in advance, reduce unnecessary people’s life and property losses. Therefore, studies of approaching nowcasting system has important practical significance.In nowcasting system, already has various weather forecast algorithm, such as the cross correlation method and optical flow method,there are two main techniques of nowcasting system, they have been widely used in meteorological field. Relative to the cross correlation method, optical flow method calculate more complex, but for the strong convective weather, optical flow method can obtain better prediction effect. Optical flow method from the point of view of the partial differential equation to get optical flow field, as introduction of strict constraints in the process of calculation, obtain results need layers of recursion, so the optical flow method, so the calculation time is longer, if use the single-core processor to calculate the optical flow field,that need quite a long time, it affect the popularization and application of optical flow method in forecast technology.Due to the weather forecast system for real time demand is higher, in the face of disaster, forecast must be every minute counts, so the prediction accuracy at the same time, must be the key to solve problems of various kinds of algorithm running time length.In recent years, with the development of parallel computing, the CPU- GPU collaborative parallel computing plays a very good speedup in optical flow method, therefore, apply parallel computing to meteorological field, can accelerate the speed of all kinds of forecast algorithm, can improve the accuracy of weather forecasts, yesterday,it cannot meet the needs of the business, but the precision is very high algorithm,now it can apply to the real business becomes possible, so the parallel computing technology used in nowcasting system is a kind of inevitable technology trends.This thesis is based on guangdong province meteorological observatory short-term nowcasting system, it is an actual research topic,it analyzed and research of various weather forecast algorithm. put forward the thought of use parallel processing to the nowcasting system, and implement the algorithm. good speed ratio and the actual results have been achieved.The main work of this thesis can be summarized as the following aspects:(1) analyzed and research of various weather forecast algorithm. put forward the thought of use parallel processing to the nowcasting system, and implement the algorithm.(2) build a multi-core parallel processing calculation and development platform, and respectively implement the multi-core cross correlation method, multi-core optical flow method and multi-core variational method and multi-core Z-I relation recognition method on the parallel computing platform, also has test and analysis to performance speedup, accelerate eventually get close to 4.(3) use Gtx660 as hardware development platform, built a GPU parallel computing platform base on CUDA, has realized the multi-threaded optical flow method and multi-thread variational method, accelerating around 100 times.