南水北调东线工程通过江都水利枢纽抽引长江水，准确及时的引水口水位预报对提高水利工程调度和水资源配置的科学性至关重要。选择引水口附近的三江营潮位站为代表，由于三江营地处潮区界以内，所以在预报其潮位时需要综合考虑潮汐引力、上游来水量和区间降水量等影响因素。首先采用潮汐调和分析法计算得到天文潮位，再采用二次校正法、多元线性回归法、支持向量机（support vector machine,SVM）法和随机森林法，对三江营的日均潮位和逐时潮位进行预报，并比较4种方法的预报精度。结果表明，4种方法得到的年日均潮位预报结果均能达到乙级以上精度，预报效果较好，其中：二次校正法的预报精度最高且最稳定，日均潮位和逐时潮位预报均能达到甲级精度，适合用于水利工程精细化调度；SVM法的预报精度次之，多元线性回归法和随机森林法相对较差，但后3种方法的预见期长于二次校正法，适合用于对预见期要求较高的潮位预警。
The East Route of South-to-North Water Transfer Project diverts water from the Yangtze River through Jiangdu Water Conservancy Project.Accurate and timely water-level forecast of the diversion project plays a key role in the scientific operation of water conservancy project and reasonable allocation of water resources.The Sanjiangying tide station near the diversion project was selected as a representative.Due to the Sanjiangying tide station locating in the tidal limit,the water level was affected by a combination of astronomical tide and upstream inflow.The interval precipitation should be taken into consideration in the tidal level forecast since the precipitation was unevenly distributed throughout the year because of the unique subtropical monsoon climate.Due to the numerous influencing factors,the tidal level forecast of Sanjiangying can be more complicated. Tidal harmonic analysis method was used to forecast astronomical tide.The average daily tidal level and the hourly tidal level can be obtained by the tidal harmonic analysis method.Four methods,including secondary correction method,multiple linear regression method,support vector machine method and random forest method were applied to forecast the average daily tidal level of Sanjiangying station.The hourly tidal level forecast was based on the forecast of the average daily tidal level of the above four methods and the tidal harmonic analysis method.The precision index,including qualified rate,absolute error and root mean square error were used to compare the accuracy of four methods for average daily tidal level and hourly tidal level.Moreover,a linearly dependent coefficient was used to compare the degree of fitting between simulated tidal level and measured tidal level.Finally,the optimum forecasting scheme was recognized. The qualified rate of a simulated tidal level using tidal harmonic analysis method was too low to meet the actual engineering requirement and therefore other methods are needed to improve the forecast precision.The results of simulated average daily tidal level and simulated hourly tidal level showed that the accuracy of the simulated average daily tidal level determined the accuracy of simulated hourly tidal level.Better average daily tidal level forecast was followed by a better hourly tidal level forecast.In addition,the accuracy of the model training period was higher than the verification period,which was consistent with the assumption of the general case.The prediction accuracy of the average daily tidal level by four methods can reach class B or above in both the training period and verification period.Moreover,the precision index showed that secondary correction method owed the highest accuracy and stability with prediction accuracy reaching class A,followed by the support vector machine method,and the multiple linear regression method and random forest method had the relatively worst performance.The order was kept when forecasting the hourly tidal level,with secondary correction method also reaching class A.The result was also verified by linearly dependent coefficient;secondary correction method had the highest linearly dependent coefficient in the forecast of average daily tidal level and hourly tidal level,which means the simulated tidal level forecasted by secondary correction method was closer to the measured tidal level.Thus,it can be seen that secondary correction method was the best method to forecast tidal level in the four methods on account of its highest prediction accuracy and stability. Out of comprehensive consideration,secondary correction method is suitable for applying in the fine scheduling of the water conservancy project in contrast to support vector machine,multiple linear regression and random forests,which are more appropriately used for tidal level warning owing to longer prediction period.