一种基于数据平滑技术的回归预测算法

在基于固定窗口宽度滑动窗口模型的基础上。提出了一种基于回归参数存储的预测模型，该模型设置了计算区、数据区和参数区。计算区用于获得最近一个滑动窗口中的数据。数据区用于接收新数据，参数区存储最近若干组滑动窗口数据所计算得到的线性回归参数值，作为计算预测结果的原始数据集。按照这种模型的处理思路。提出了一种基于数据平滑技术的回归预测算法，随着窗口的滑动。对数据区中的数据进行回归分析，获得前面若干组滑动窗口数据的回归函数并存入参数区中，检验当前窗口中数据获得的回归函数预测效果。实验分析表明。通过修正当前回归函数的参数。可以使预测函数的预测精度得到很大程度的提高。

Data Stream Prediction Model Based on Parameters Smoothing Methods

Fixed window width model means the size of the sliding window width is a fixed number. Based on this model,this paper proposed a data stream prediction model based on regression parameters smoothing method. Compared with the sliding window model, a new buffer zone is set for the storage of the parameters which calculated by recent data. Then a new data stream prediction algorithm is proposed. By using the data in current data buffer to get the prediction function ,we can get the prediction value in cur- rent window. Given a fixed deviation value,if the difference between the prediction results and observation data not less than the deviation,parameters correction step need to be done, in which the prediction function parameters need to be changed. Experiments show us that this algorithm can reduce the space and time complexity, and also improves the prediction accuracy.