Abstract: Prophet is a time series prediction model open sourced by Facebook. It is good at processing daily periodic data with large outliers and trend changes. Aiming at the problem that Prophet time series model has low prediction accuracy on short-term data, this paper proposes an improved Prophet_SVR model based on Prophet to predict dissolved oxygen parameters in the next 2 hours, and uses a comparative model to perform comparative experiments on the same data. The experimental results were compared by root mean square error (ERMSE) and average absolute percentage error (EMAPE). The results show that the compared with Prophet time series model, in Prophet_SVR model, ERMSE is reduced by 0.197 1 and EMAPE is decreased by 3.890 4%. Through experimental comparison, it can be seen that the Prophet_SVR prediction model is more effective in reducing the overall prediction error and improving the accuracy of a single numerical prediction, and the method can train the model in a shorter time and with higher efficiency, which provides a reference for short-term water quality parameter prediction.

Key words: time series, dissolved oxygen prediction, Prophet time series prediction model, SVR

摘要： Prophet是Facebook开源的一种时间序列预测模型，擅长处理具有大异常值和趋势变化的日常周期数据。针对Prophet时序模型在短时间数据上预测精度较低的问题，提出了基于Prophet改进的Prophet_SVR模型对未来2 h内溶氧参数进行预测，并利用对比模型在相同数据上进行对比试验。试验结果通过均方根误差(ERMSE)和平均绝对百分比误差(EMAPE)进行对比。结果显示：Prophet_SVR模型的试验结果相对于Prophet时序模型ERMSE下降0.197 1，EMAPE下降3.890 4%。试验对比可知，Prophet_SVR预测模型在降低预测整体误差和提升单个数值预测精度上效果更优。该方法训练模型的时间更短、效率更高，为短期水质参数预测提供了参考。

关键词: 时间序列, 溶氧预测, Prophet时序预测模型, 支持向量回归