| High core rockfill dam project is usually characterized by large-scale and long-duration,conducting construction schedule simulation is of vital importance to reasonably predict the dam filling schedule and control the project cost.However,the construction of high core rockfill dam project is an open-air operation,so rainfall can easily cause the moisture content of dam material to exceed the design standard,which in turn affects the compaction quality of dam material.To ensure construction quality,the construction management measures taken by the site operators in the rainy season directly affect the dam filling schedule.Therefore,in order to reasonably carry out the construction schedule planning of layers and the dam project,it is a great challenge for the construction management research of high core rockfill dam to establish the construction schedule simulation model considering the effect of rainfall,and there is an urgent need to study and solve the following key issues:First of all,how to realize the short-term rainfall forecast and daily rainfall forecast which match the time scale of layer construction planning and long-term filling schedule planning.Secondly,how to improve the modeling accuracy of key simulation parameters based on the real-time acquisition of construction monitoring system.Thirdly,how to establish the layer compaction simulation model which is more suitable to the actual construction characteristics of the compaction process,and how to establish the layer compaction simulation model considering the effect of short-term rainfall.Finally,how to establish the construction schedule simulation model of high core rockfill dam considering the influence of daily rainfall.In response to the above problems,this paper mainly carried out the construction schedule simulation study of high core rockfill dam considering rainfall impact,and achieved the following main results:(1)A short-term rainfall prediction method considering unbalanced data characteristics and a non-parametric prediction model of daily rainfall are proposed.Most current studies on short-time rainfall forecasting ignore the influence of unbalanced data characteristics.A few adopt oversampling methods,but the calculation efficiency is low due to the increase of sample size,and the method based on local information interpolation to reduce the unbalanced degree of sample distribution tends to cause model over-fitting.In this paper,a two-stage short-term rainfall forecasting model based on improved Cat Boost and GRU considering unbalanced data features is proposed to improve the prediction accuracy of short-term rainfall series.Firstly,a two-stage modeling approach is proposed,which specifically refers that short-term rainfall forecasting is conducted only when a short-term rainfall event is predicted to occur.The advantage lies in avoiding the problem of increasing sample size caused by the oversampling method and reducing the class imbalance at the same time.Secondly,by calculating the mean,maximum,minimum,standard deviation,skewness,and kurtosis statistical features of wind speed,temperature,and atmospheric humidity respectively,a way to build the classification feature vector for short-term rainfall event prediction is proposed.Thirdly,an improved Cat Boost model based on Multi-Verse Optimizer(MVO)is proposed for short-term rainfall event prediction,which can effectively mitigate prediction bias and improves the prediction accuracy.Finally,when a short-term rainfall event is predicted to occur,short-term rainfall prediction is realized based on the GRU model,which effectively captures the complex nonlinear relationship between multiple time steps and multiple meteorological variables.Based on the observation data from an automatic meteorological station deployed on the filling surface of a high core rockfill dam project in southwest China,application research on short-term rainfall forecasting was carried out.The effectiveness of the proposed model was verified by multiple evaluation indicators and various comparison methods.Current daily rainfall prediction studies suffer from the deficiency of model prediction accuracy relying on empirical specified distribution types.In this paper,a non-parametric prediction model of daily rainfall based on the higher-order Markov chain and Dirichlet process mixture model is proposed to improve the prediction accuracy of daily rainfall series.Firstly,the daily rainfall state simulation model is established based on the higher-order Markov chain.By extending the first-order correlation of the first-order Markov chain to a higher-order correlation,the evolution of the daily rainfall state can be better simulated.Secondly,a non-parametric prediction model of daily precipitation based on the Dirichlet process mixture model is proposed.Defining the prior distribution over an infinite-dimensional parameter space can reduce the dependence on empirically specified distribution types and thus improve the accuracy of daily precipitation prediction.Finally,based on the daily rainfall observation data from a meteorological station near a high core rockfill dam project in southwest China,application research on daily rainfall prediction was carried out.The effectiveness of the proposed model was verified by multiple comparison methods.(2)A fast non-parametric distribution modeling method of compaction construction simulation parameter,and a high-dimensional joint distribution modeling method of the travel time of dam material transportation are presented.At present,the compaction construction simulation parameter modeling study of high core rockfill dam has the problem of requiring the empirical setting of hyper-parameters and low computational efficiency.In response to the above problems,a fast non-parametric distribution modeling method of compaction construction simulation parameter is proposed.Firstly,a distribution modeling method of compaction construction simulation parameter based on the Bayesian field theory is presented.By using scalar fields to define the prior distribution of simulation parameters,the problem of empirically set hyper-parameters can be effectively avoided and the modeling accuracy can be improved.Secondly,the threshold identification of sample outliers based on distribution estimation results is proposed to improve the modeling accuracy of simulation parameters.Thirdly,the rejection sampling method is adopted to generate random samples of compaction construction simulation parameters,which are further used as input of the construction simulation model.Finally,based on the data collected from the intelligent compaction construction monitoring system of a high core rockfill dam project,an application study on distribution modeling of roller speed was carried out.The analysis from several perspectives has verified the high accuracy and computational efficiency of the proposed method.The current study on travel time modeling of dam material transportation for high core rockfill dam project has some shortcomings,including the difficulty in determining the model structure of high-dimensional joint distribution model under the framework of Copula methods when the number of road segments is large,and lack of division method for road segments of dam material transportation.In response to the above problems,firstly,the division method for road segments of dam material transportation is presented based on the Douglas-Peucker algorithm from the perspective of curve vector compression,which can effectively preserve the original geometric characteristics of the road.Secondly,abnormal,duplicate,and isolated records of GNSS trajectory data of trucks are identified to improve the distribution modeling accuracy.Thirdly,a geometry-based map matching algorithm is used to realize the localization and mapping calculation of GNSS trajectory data of trucks to each road segment.Finally,a method for modeling high-dimensional joint probability density distribution of dam material transportation path travel time based on the Neural Spline Flows model is proposed.The advantage lies in effectively avoiding empirical assumptions on the time-space correlation structure of travel speed on adjacent road segments.And the proposed method can effectively overcome the problem of difficulty in determining optimal vine structure and type of Copula function in each layer in the framework of Copula methods.Based on the data collected from the dam material transportation monitoring system of a high core rockfill dam project,an application study on modeling high-dimensional joint probability density distribution of dam material transportation path travel time was carried out.Results of the case study has validated the effectiveness of the proposed method.(3)A layer compaction construction simulation model of the high core rockfill dam that simultaneously considers the effects of short-term rainfall and the uncertainty of compaction construction simulation parameters is established.In the current compaction construction simulation study,the effect of short-term rainfall is ignored,and the influence of compaction simulation parameter uncertainty on the compaction duration is not fully considered.In response to the above problems,firstly,taking the accumulated precipitation and the maximum instantaneous precipitation of short-term rainfall as inputs,the compaction construction suspension model based on the TPOT(Tree-based Pipeline Optimization Tool)algorithm is proposed,which has the advantage of high modeling efficiency.Secondly,in addition to considering the uncertainty of compaction construction simulation parameters such as roller speed,roller deflection angle,and roller offset width,the completed width of the filling layer is proposed as the criterion for judging whether the compaction construction process is completed,to avoid the misjudgment of simulation ending condition caused by the fixed number of compaction bands.On this basis,a refined compaction construction simulation model considering the uncertainty of the number of compaction bands is established.The advantages of the proposed model are that the modeling of construction simulation process is closer to actual construction characteristics and the prediction accuracy of compaction duration is proved.Thirdly,the layer compaction construction simulation method of high core rockfill dam that simultaneously considers the effects of short-term rainfall and the uncertainty of compaction construction simulation parameters is proposed,which can provide a reasonable decision-making basis for the compaction construction planning of high core rockfill dam in the rainy season.Finally,based on a high core rockfill dam project located in southwest China,an application study on compaction construction simulation of core zone was conducted to realize compaction duration prediction under short-term rainfall impact.By comparison with actual monitoring data,the validity of the proposed model was verified.(4)A construction schedule simulation model of high rockfill dam considering the effect of daily rainfall stochastic characteristics is established.Current construction schedule simulation studies of high core rockfill dam usually calculate the suspension period based on multi-year historical daily rainfall data,ignoring the influence of random characteristics of daily rainfall such as the length and distribution of dry or wet spells on the construction schedule.In this paper,a construction schedule simulation model of high core rockfill dam considering the effect of random characteristics of daily rainfall is established.Firstly,based on the aforementioned layer compaction construction simulation model considering the effect of uncertainty of compaction construction simulation parameters and the high-dimensional joint distribution of travel time of dam material transportation,the mathematical model of the construction schedule simulation model of high core rockfill dam considering the effect of random characteristics of daily rainfall is proposed.Secondly,the method for calculating the construction suspension period of high core rockfill dam under daily rainfall conditions is proposed.The advantage of this method is that the influence of stochastic characteristics of daily rainfall on the construction schedule is simultaneously considered,including the number of wet days,length,and distribution of dry or wet spells.Finally,taking a high core rockfill dam project in southwest China as an example,an application study on construction simulation of filling schedule of the core zone was carried out to realize filling schedule prediction under the influence of daily rainfall.The validity of the proposed model was verified by comparing it with the actual monitoring data. |
