皖南农村地区需水量特征分析及预测模型研究

Analysis and predictive model research of water demand features in southern Anhui rural areas

需水量的准确预测对于提高农村地区水资源利用率、优化农饮水厂日常调度有着重要作用,然而现有预测算法面临农村部分地区可用历史数据少、质量差等问题,同时部分预测算法的计算负荷较高,无法有效适应农饮水厂运维环境。选取2021-2022年安徽省南部不同地区3处农饮水厂日供水量数据作为研究对象。经过特征分析发现,皖南农村地区日需水量存在季节性变化趋势明显而短期内波动较大的特点。针对该数据特点及实际计算需求,选取了使用网格搜索法优化后的ARIMA模型进行预测,并与SVR、LSTM、RF等模型进行对比分析。研究结果表明,使用网格搜索法的ARIMA模型可以更好地适应农饮水厂运维环境,利用其对于周期性、季节性数据较强的学习能力,对农饮水厂供水量变化趋势和规律进行准确预测,具有一定的普适性,且计算负荷明显优于其他方法。模型对比分析显示,ARIMA模型的预测精度最高,在3处农饮水厂测试集数据上的平均绝对百分比误差(MAPE)分别为1.827、1.454、2.714。

Accurate prediction of water demand plays a crucial role in enhancing the utilization efficiency of water resources in rural areas and optimizing the daily scheduling of rural drinking water plants.However,existing forecasting algorithms face challenges such as limited availability and poor quality of historical data in rural areas.Additionally,the computational load of some algorithms is high,making them less adaptable to the operational environment of rural drinking water facilities.In this study,daily water supply data from three different rural drinking water plants in southern Anhui Province from 2021 to 2022 were selected as the research subjects.Through feature analysis,it was observed that daily water demand in rural areas of southern Anhui exhibits significant seasonal trends and considerable short-term fluctuations.Addressing the data characteristics and computational requirements,this study selected the ARIMA forecasting model optimized using the grid search method for prediction,and conducted comparative analyses with other models such as SVR,LSTM and RF.The results indicate that the ARIMA model,when enhanced with grid search,is better suited for the operational environment of rural drinking water facilities.Leveraging its strong learning capability for periodic and seasonal data,the model accurately predicts the trends and patterns in water supply at these facilities.Furthermore,it demonstrates a certain degree of universality and has a significantly lower computational burden compared to other methods.Model comparison analysis demonstrates that the ARIMA model exhibits the highest prediction accuracy.The Mean Absolute Percentage Error(MAPE)for the test dataset of three agricultural drinking water plants are 1.827,1.454 and 2.714,respectively.