动态指数平滑优化模型及其应用

Optimized dynamic exponential smoothing model and its applications

针对常规二次指数平滑模型在实际应用中的不足,引入动态平滑参数的概念,建立了一类不需选取平滑初值、平滑权重能自行调整的指数平滑模型;以预测误差平方和SSE最小为目标,构造了优选并自动生成最佳平滑参数使预测模型得以优化的最速下降算法.通过严谨的数学论证和算法分析较完整地解决了指数平滑预测中,平滑参数靠经验确定并且因静态、平滑初值难以确定而易导致预测偏差等问题.最后用所建模型对中国消费品零售总额进行了预测,结果明显优于传统模型.

From analyzing the deficiency of the traditional model, we put forward the concept of dynamic smoothing parameter and a new class of quadratic exponential smoothing model, the main features of our model are that it's unnecessary to put any initial value in smoothing time series and the weight number in the formula is always adjusting with the parameter self_actingly. Aiming the square sum of error (SSE), we also construct the steepest_descent algorithm to iterate and select an optimal parameter for optimizing the new model. By deducting mathematically,some questions, i.e., the smoothing parameter was static and determined only by one's experiences, and the smoothing initial value is difficulty to determine and leads to a deviation easily, are completely resolved. At last, the total sales of consumer goods in China is forecasted by using the new model, and the result apparently is superior to the traditional model's one.