基于时间序列的网格化城市管理案件预测模型研究

Urban Grid Management Cases Prediction Model Based on Time Series

针对传统或流行的基于时间序列的预测模型,探索出一种适用于网格化城市管理的成体系的案件预测方法。分别采用博克斯-詹金斯法、Auto-ARIMA以及LSTM模型,对近几年北京市6个城区各站点网格化管理问题案件数量进行预测,通过对比不同模型方法间准确度和实用性,以MAPE 为精度评价指标,分析各个模型应用在城市网格化问题预测方面优势与劣势。研究发现,Auto-ARIMA适合进行对网格化管理问题数量趋势预测,博克斯-詹金斯法在解决滞后性问题中预测准确率很高,但由于预测流程烦琐,因此实用性较差,LSTM预测效果相对准确且平稳,可以在样本输入量、参数以及自身架构上进行进一步优化。

In order to improve the traditional or popular predicting model based on time series, this paper explores the predicting method for systematic case suitable for grid city management. The Box-Jenkins method, Auto- ARIMA and LSTM models have been adopted to predict the number of cases of grid management problems in six urban areas of Beijing in recent years, by comparing the accuracy and practicability between different model methods, and taking MAPE as the accuracy evaluation indicator, and analyzing in application of urban grid management. The experiment results indicate that Auto-ARIMA is suitable for predicting the quantitative trend of grid management problems, and Box-Jenkins method has a high prediction accuracy after solving the lag problem, however, with lower practicability, and LSTM prediction effect is relatively accurate and stable and can be further improved on sample input, parameters, and its framework.