| 66 | 0 | 78 |
| 下载次数 | 被引频次 | 阅读次数 |
【目的】针对水闸水平位移序列受工况变化、异常跳点和外部驱动滞后影响，传统阈值清洗与单一模型难以兼顾异常识别和预测适用性的问题，提出分工况ARIMAX-MAD稳健清洗与短期预测方法。【方法】以某水闸逐日监测数据为对象，依据水位、降雨和运行扰动划分汛期高扰动与非汛期平稳工况。以水位差、气温和降雨量为外生变量，经滞后相关分析建立ARIMAX模型，利用模型残差表征当前工况下难以解释的异常偏离，并结合MAD稳健尺度识别异常；对清洗后训练序列实施固定参数单步滚动预测，与HST模型比较。【结果】ARIMAX-MAD在保持序列均值、中位数和主体波动特征的同时削弱异常跳点影响，残差标准差由0.554 mm降至0.402 mm，峰度由4.02降至3.67。汛期三类外生变量的滞后约为2 d、2 d和5 d；ARIMAX(0，1，3)的RMSE、MAE和MaxAE分别为0.162 mm、0.140 mm和0.282 mm，较HST降低80.8%、76.7%和84.2%，残差自相关控制更优。非汛期ARIMAX(2，0，0)在MAE和残差相关控制方面较优，HST在RMSE、MaxAE和R2/NSE方面较优。【结论】分工况建模提高了清洗和预测评价的针对性。ARIMAX-MAD主要用于异常识别和训练序列稳健化；汛期宜采用滞后ARIMAX刻画短时响应，非汛期可结合HST控制整体水平与极端偏差，可为水闸位移监测数据处理和短期预测提供参考。
Abstract：Objective: Sluice displacement monitoring data are often affected by changing operating conditions， local outliers and delayed responses to external drivers， which limits the reliability of conventional amplitude-threshold cleaning and single-model forecasting. This study proposes an operating-condition-based ARIMAX-MAD framework for robust data cleaning and short-term displacement forecasting. Methods: Daily horizontal displacement data from the ZSZEX04 monitoring point of a sluice were divided into a high-disturbance flood-season condition and a relatively stable non-flood-season condition according to water-level， rainfall and operation-related disturbance characteristics. Water-level difference， air temperature and rainfall were used as exogenous variables. Lagged correlation analysis was first conducted to determine the delayed effects of these variables， and condition-specific ARIMAX models were then established. ARIMAX residuals were used to represent deviations that could not be explained by the current operating condition， and a median absolute deviation (MAD)-based robust threshold was applied to identify outliers. The cleaned series was evaluated using fixed-parameter one-step-ahead rolling forecasts and compared with a conventional hydrostatic-seasonal-time (HST) model. Results: The ARIMAX-MAD method reduced local abnormal jumps while largely preserving the mean， median and main fluctuation characteristics of the displacement series. After cleaning， the residual standard deviation decreased from 0.554 mm to 0.402 mm， and kurtosis decreased from 4.02 to 3.67. Under flood-season conditions， displacement responses showed lagged relationships with water-level difference， air temperature and rainfall of approximately 2 d， 2 d and 5 d， respectively. The lagged-variable ARIMAX(0，1，3) model achieved an RMSE， MAE and MaxAE of 0.162 mm， 0.140 mm and 0.282 mm， respectively， reducing these errors by approximately 80.8%， 76.7% and 84.2% compared with the HST model， with better residual autocorrelation control. Under non-flood-season conditions， ARIMAX(2，0，0) and HST showed different advantages: HST performed better in RMSE， MaxAE and R/NSE， whereas ARIMAX showed lower MAE and better control of residual temporal correlation. Conclusion: Operating-condition-based modeling improves the specificity of cleaning and forecasting for sluice displacement monitoring data. ARIMAX-MAD mainly contributes to anomaly identification and robust training-series preparation. A lagged-exogenous-variable ARIMAX model is more suitable for short-term dynamic response forecasting under high-disturbance flood-season conditions， whereas HST can be combined under stable non-flood-season conditions to control overall displacement level and extreme deviations. The proposed framework provides a practical reference for sluice displacement data processing， short-term forecasting and safety warning.
[1] 曹海， 李海枫， 徐涛， 等. 近坝库岸边坡安全监控及风险预警研究[J]. 水利水电技术(中英文)， 2025， 56(S2): 561-567.
[2] 何宁， 李登华， 陈鸽， 等. 土石坝安全监测数据处置与评价技术研究进展[J]. 水利水运工程学报， 2025(5): 88-100.
[3] 于国卿， 汪自力， 任红旭. 新陈代谢GM(1， 1)模型预测水闸垂直位移[J]. 人民黄河， 2008， 30(12): 116-117.
[4] 李学德， 周明明， 赵津磊， 等. 基于改进GM(1， 1)模型的水闸沉降预测研究[J]. 人民珠江， 2020， 41(11): 72-75.
[5] 宋福生. 改进的变权组合模型在水闸沉降预测中的应用[J]. 水利技术监督， 2023(11): 243-246.
[6] 钟成. 基于ARIMA的水库大坝安全预测方法研究[J]. 水利科学与寒区工程， 2025， 8(4): 111-114.
[7] 汪远征， 徐雅静. 多元平稳时间序列ARIMAX模型的应用[J]. 统计与决策， 2007， 23(18): 132-135.
[8] 张利. 基于时间序列ARIMA模型的分析预测算法研究及系统实现[D]. 镇江: 江苏大学， 2008.
[9] 艾玲. 时间序列短期预测的方法和技术[D]. 上海: 华东师范大学， 2010.
[10]向华琦. 基于时间序列分析法的大坝变形监测数据分析研究[D]. 杨凌: 西北农林科技大学， 2012.
[11] 王潇骁， 顾嘉辉， 李悦溪. 基于SARIMAX模型的城市轨道交通客流预测[J]. 高科技与产业化， 2024， 30(8): 40-42.
[12] MULLA S， PANDE C B， SINGH S K. Times series forecasting of monthly rainfall using seasonal auto regressive integrated moving average with EXogenous variables (SARIMAX) model[J]. Water Resources Management， 2024， 38(6): 1825-1846.
[13] SHAH V， PATEL N， SHAH D， et al. Forecasting maximum temperature trends with SARIMAX: A case study from ahmedabad， India[J]. Sustainability， 2024， 16(16): 7183.
[14] LONG C Y， MARSANI M F， MA H， et al. Predicting daily water level with exogenous variables ARIMAX-MS-GARCH model[J]. Journal of Water and Climate Change， 2026， 17(2): 74-93.
[15] 王成， 胡添翼， 顾艳玲， 等. 大坝安全监测的ARIMA-BP组合预测模型[J]. 三峡大学学报(自然科学版)， 2018， 40(1): 20-24.
[16] 杨晨蕾， 包腾飞， 胡安玉， 等. 考虑残差的小波G-Verhulst-ARIMA大坝变形组合预测模型及应用[J]. 水电能源科学， 2020， 38(12): 94-97.
[17] 胡安玉， 包腾飞， 杨晨蕾， 等. 基于LSTM-Arima的大坝变形组合预测模型及其应用[J]. 长江科学院院报， 2020， 37(10): 64-68.
[18] 杨恒， 岳建平， 周钦坤. 利用SVM与ARIMA组合模型进行大坝变形预测[J]. 测绘通报， 2021(4): 74-78.
[19] MA N， NIU X B， CHEN X H， et al. Analysis of concrete dam deformation prediction based on the ResNet-GRU-SGWO model[J]. Advances in Civil Engineering， 2024， 2024: 4791788.
[20] TIAN K， YANG J， CHENG L. Deep learning model for the deformation prediction of concrete dams under multistep and multifeature inputs based on an improved autoformer[J]. Engineering Applications of Artificial Intelligence， 2024， 137: 109109.
[21] LI Y T， BAO T F， GAO Z X， et al. A new dam structural response estimation paradigm powered by deep learning and transfer learning techniques[J]. Structural Health Monitoring， 2022， 21(3): 770-787.
[22] 刘桓辰， 朱静， 郭梦京. 融合特征因子筛选的拱坝变形深度学习预测模型[J]. 水利水电技术(中英文)， 2025， 56(3): 123-134.
[23] 丁勇康， 远近， 毛延翩， 等. 基于IHHO-LSTM-KAN的大坝变形预测模型[J]. 水利水电技术(中英文)， 2025， 56(5): 170-182.
[24] 段尧彬， 刘邓， 满翰林， 等. 基于LSTM-Informer模型的乌东德水库水位多步长预测研究[J]. 水利水电技术(中英文)， 2025， 56(S1): 1-5.
[25] 赵二峰， 李章寅， 袁冬阳. 基于双阶段注意力机制的大坝变形深度学习预测模型[J]. 河海大学学报(自然科学版)， 2023， 51(6): 44-52.
[26] 周兰庭， 邓思源， 柳志坤， 等. 基于SSA-LSTM-GF的混凝土坝变形预测方法[J]. 河海大学学报(自然科学版)， 2023， 51(2): 73-80.
[27] 严英杰， 盛戈皞， 陈玉峰， 等. 基于时间序列分析的输变电设备状态大数据清洗方法[J]. 电力系统自动化， 2015， 39(7): 138-144.
[28] PEARSON R K. Data cleaning for dynamic modeling and control[C]//1999 European Control Conference (ECC). Karlsruhe， Germany: IEEE， 1999: 2584-2589.
[29] LEYS C， LEY C， KLEIN O， et al. Detecting outliers: Do not use standard deviation around the mean， use absolute deviation around the median[J]. Journal of Experimental Social Psychology， 2013， 49(4): 764-766.
[30] ROUSSEEUW P J， CROUX C. Alternatives to the median absolute deviation[J]. Journal of the American Statistical Association， 1993， 88(424): 1273-1283.
[31] LEYS C， DELACRE M， MORA Y L， et al. How to classify， detect， and manage univariate and multivariate outliers， with emphasis on pre-registration[J]. International Review of Social Psychology， 2019， 32(1): 5.
[32] YANG Z G， WANG J， HUANG Q J， et al. Anomaly detection algorithm of single variable time-series data based on dynamic parametrization for subsurface fluid data anomaly detection[J]. Geophysical Journal International， 2025， 243(1): 1-12..
[33] ZHU Y T， NIU X Q， YAN T Y， et al. A robust displacement monitoring model for high-arch dams integrating signal dimensionality reduction and deep learning-based residual correction[J]. Structural Control and Health Monitoring， 2024， 2024: 3330769.
[34] MEJRI N， LOPEZ-FUENTES L， ROY K， et al. Unsupervised anomaly detection in time-series: an extensive evaluation and analysis of state-of-the-art methods[J]. Expert Systems with Applications， 2024， 256: 124922.
[35] TASHMAN L J. Out-of-sample tests of forecasting accuracy: an analysis and review[J]. International Journal of Forecasting， 2000， 16(4): 437-450.
[36] BERGMEIR C， BENÍTEZ J M. On the use of cross-validation for time series predictor evaluation[J]. Information Sciences， 2012， 191: 192-213.
[37] HEWAMALAGE H， ACKERMANN K， BERGMEIR C. Forecast evaluation for data scientists: common pitfalls and best practices[J]. Data Mining and Knowledge Discovery， 2023， 37(2): 788-832.
[38] SCHWARZ G. Estimating the dimension of a model[J]. The Annals of Statistics， 1978， 6(2): 461-464.
[39] AKAIKE H. A new look at the statistical model identification[J]. IEEE Transactions on Automatic Control， 1974， 19(6): 716-723.
[40] CAO X， SHENG J B， JIANG C， et al. Concrete dam deformation prediction model considering the time delay of monitoring variables[J]. Scientific Reports， 2025， 15(1): 8458.
[41] BERGER A， KIEFER M. Comparison of different response time outlier exclusion methods: A simulation study[J]. Frontiers in Psychology， 2021， 12: 675558.
[42] IGLEWICZ B， HOAGLIN D C. How to detect and handle outliers[M]. Milwaukee: ASQC Quality Press， 1993.
[43] ROMO-CHAVERO M A， CANTORAL-CEBALLOS J A， PÉREZ-DÍAZ J A， et al. Median absolute deviation for BGP anomaly detection[J]. Future Internet， 2024， 16(5): 146.
基本信息:
中图分类号:TV698.11
引用信息:
[1]杨登翔,马福恒,俞扬峰.基于ARIMAX-MAD残差分析的水闸位移监测数据稳健清洗与分工况预测[J].水利水电技术(中英文)().
基金信息:
黄河水科学研究联合基金重点支持项目（U2243244）
2026-08-31
2026-08-31
2026-08-31