ABSTRACT

In this paper, we discuss an approach that predicts the quantitative value of rainfall. The proposed algorithm uses a combination of data mining and neuro-fuzzy inference system for prediction. The model is demonstrated on north interior Karnataka (a state in India) rainfall data as a case study. This model is applicable to any geographical area provided apt predictors are included. For north interior Karnataka rainfall prediction predictors are derived from local and global climate conditions. The local condition variables are derived from the mean sea level pressure, temperature, and wind speed in south India. The global variables affecting the north interior Karnataka rainfall include, Darwin sea level pressure, the ENSO indices and southern oscillation. The data mining technique, association rule mining, is used to study the correlation among the predictors; clustering is used for predictor selection as well as membership function creation for fuzzyfication. Neuro-fuzzy inference system is further used for fine tuning the “If-then” rules and crisp value prediction of the rainfall. The prediction accuracy is observed to be good considering Tropical Meteorological Department data.

摘要

在本文中，我们讨论了一种预测降雨量数值的方法。该算法结合使用数据挖掘和神经模糊推理系统进行预测。该模型以卡纳塔克邦北部内陆地区（印度的一个邦）降雨数据作为案例研究进行了演示。该模型适用于任何包含适当预测变量的地理区域。对于卡纳塔克邦北部内陆地区，降雨量预测是根据当地和全球气候条件得出的。当地条件变量源自印度南部的平均海平面压力、温度和风速。影响卡纳塔克邦北部内陆降雨的全球变量包括达尔文海平面气压、ENSO 指数和南方涛动。数据挖掘技术、关联规则挖掘、用于研究预测变量之间的相关性；聚类用于预测器选择以及模糊化的隶属函数创建。神经模糊推理系统进一步用于微调“If-then”规则和降雨量的清晰值预测。考虑到热带气象部门的数据，预测精度良好。