Raw Water Supply System Operation And Risk Research With Considering Forecast Uncertainty Under Low Water Scenario

Raw water is the main source to ensure urban life,production and ecological environment,and its water safety directly affects the sustainable development and social stability of the city.In recent years,the acceleration of urbanization and the increase of urban population have led to the continuous expansion of the water supply scale in the central urban area and the increase in the total urban water consumption yearly,which have made the contradiction between the supply and demand of urban water resources become more and more severe.The objective existence of the uneven temporal and spatial distribution of resources and the imbalance of water demand in the society,makes the construction of water transfer projects and the joint operation of multiple water sources become important ways to alleviate the water shortage in urban areas.Medium and longterm runoff forecasting can provide conditions for formulating annual water supply plans,and its accuracy is directly related to the decision-making quality of the reservoir operation.However,current runoff forecasting models tend to focus on a single forecasting model,and there is room for improvement.In addition,the impact of runoff forecast results and their uncertainty on the water supply strategy is not fully reflected in the actual dispatch process.Affected by global climate change,the assumption of consistency in runoff sequence no longer exists,hydrometeorological drought events occur frequently,and existing scheduling schemes are difficult to cope with the potential challenges brought by low water scenario to water resources system,so it is urgent to carry out intensive research.Taking the raw water reservoirs in the Ningbo as the research object,this paper focuses on the scientific problems of raw water supply system with considering prediction uncertainty under low water scenario,and carries out the following research work:(1)In view of the fact that the multi-dimensional Copula function is mostly limited to Archimedean-type Copula functions when analyzing the encounter characteristics of runoff,and that the maximum likelihood estimation(ELM)is easy to obtain unreliable estimation when the runoff sample size is small,this research selects four large reservoirs as the research object,and the probability of synchronous asynchronous encounter for runoff sequences during multiple research periods is calculated.Based on analyzing of the optimal marginal distribution of a single runoff sequence,a variety of Copula functions are used to construct the combined distribution of pairwise runoff variables,the interior-point optimization algorithm based on the gradient(IPOAG)is used to estimate the parameters of Copula functions and determine the optimal Copula function,and a universal hierarchical nested Copula function is constructed with reference to the step by step recursion.Studies have demonstrated that,compared with the maximum likelihood estimation method,the Copula function parameters estimated by the IPOA-G have a better fitting effect when the structural variables are jointly distributed.The annual runoff of the four reservoirs is combined to construct a nested function with Gaussian copula in the upper layer and Gaussian copula and Clayton copula in the lower layer.(2)In view of the problem that the factor screening effect of the medium and long-term runoff forecasting model is general and the forecast accuracy still needs to be improved,according to the adaptive framework of ‘signal decomposition-component prediction-modal reconstruction’,a mixed runoff prediction model driven by multi-influencing factors is constructed in the Baixi Reservoir.Based on the mutation year of runoff sequences,the study period is divided into the calibration period and the verification period.Then,discrete wavelet transform(DWT),variational mode decomposition(VMD)and complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)are used to decompose monthly runoff series into multiple modal components,the Pearson correlation coefficient and random forest method are used to screen the teleconnected and conventional factors,and long-short-term memory artificial neural network(LSTM)is used to forecast runoff modal components and reconstruct monthly runoff series.Research has shown that the prediction effect of the mixed model constructed with the decomposition model of CEEMDAN or VMD is better than that of the mixed model constructed with the decomposition model of DWT.Multi-group mixed models have good forecasting performance in both the calibration period and the verification period.The NSE in the calibration period is greater than 0.84,while that in the verification period is greater than 0.78.The best hybrid model for monthly runoff prediction in the Baixi Reservoir is the prediction model composed of CEEMDAN,Pearson correlation coefficient and RF,LSTM.(3)In view of the fact that the past topology traversal method can not meet the modeling requirements of multi-source to multi-waterworks in a complex water supply system,a generalized method of topological relationship with traversing the river entity as the core is proposed,and the optimal dispatching model concerning the multi-source joint water supply is constructed.Condiering the problem that traditional runoff generation methods in the dry years are mostly limited to selecting the measured data in typical years,a multiple random highdimensional sequence generation method is proposed by combining the hierarchical nested Copula function coupled with the Latin hypercube sampling(LHS)and multi-factor nearest neighbor sampling regression model(MNNBR).By setting the joint distribution probability to generate multiple runoff schemes,and input them into the optimal scheduling model,so as construct the mapping relationship between different low-water scenarios and water shortage risks in the water-receiving area,and propose a water supply risk strategy to deal with the low-water scenarios.Studies have shown that under the low water level conditions with a design guarantee rate of 90%,the water shortage risk in the Jiangdong Water Plant is 0.222;under the low water level conditions with a design guarantee rate of 95%,the water shortage risk in the Beilun Water Plant is 0.249.(4)Aiming at the problem that the existing methods are difficult to describe the variation of multi-stage runoff prediction error of multiple hydrological stations with time,this paper analyzes the distribution characteristics of single station prediction error and multi station prediction error by using the characteristics of similar climate conditions in the same hydrological area and spatial correlation of multi station prediction error generated by the same prediction model.And then,a stochastic optimal operation model of reservoir group water supply coupled with Monte Carlo scenario tree and vine copula function is constructed.Studies have shown that,compared with the deterministic forecast of the inflow,the inflow considering the forecast error can reduce the water supply damage depth by 32.0%,36.5%,and 31.7%,respectively,under three inflow conditions.