基于局部敏感哈希的天气形势场图像相似预报
Similarity Prediction of Weather Situation Field Image Based on Locality Sensitive Hashing

气象相似预报是指从海量高维历史天气数据样本中查找与某时刻天气数据样本最相似的个例，并以此为依据制作天气预报。由于搜索空间巨大，现有的相似预报方法存在检索速度慢的问题。借鉴图像检索的思想，本文提出了一种基于局部敏感哈希的天气形势场相似预报方法，利用局部敏感哈希函数将大量高维的历史形势场图像进行哈希散列，建立历史形势场的索引结构，并以相似离度作为相似性准则衡量样本之间的相异程度，得到相似个例排序，从而减少计算量。实验结果表明，该方法能够对高维海量的历史形势场进行有效搜索，在保证相似预报准确率的基础上，降低了相似预报的时间复杂度，提高了相似预报的检索效率。
Meteorological similarity prediction refers to finding the individual cases that are most similar to the weather data samples at a certain time from a large number of high dimensional historical weather data samples and making weather forecasts based on them. Because of the huge search space, the current similarity prediction method has the problem of slow retrieval speed. Using the idea of image retrieval as a reference, this paper proposes a similarity prediction method for the weather situation field based on locality sensitive hashing. The method hashes a large number of high-dimensional historical situation field images with locality sensitive hashing function to establish the index structure of the historical situation field, and measures the degree of dissimilarity between samples by using similarity divergence as a similarity criterion to get the order of similar cases, thus reducing the amount of calculation. The experimental results show that the proposed method can effectively search the high-dimensional massive historical situation field, reduce the time complexity of similarity prediction while ensuring the accuracy of similarity prediction, and improve the retrieval efficiency of similarity prediction.