Search

CN-121997844-A - Hydraulic engineering intelligent design method and system based on multi-source heterogeneous data fusion

CN121997844ACN 121997844 ACN121997844 ACN 121997844ACN-121997844-A

Abstract

The invention relates to the technical field of hydraulic engineering, and particularly discloses a hydraulic engineering intelligent design method and a hydraulic engineering intelligent design system based on multi-source heterogeneous data fusion, which are used for acquiring weather radar rainfall forecast data and river network hydrodynamic monitoring data, and carrying out space-time alignment and cleaning to obtain a multi-source data set; the method comprises the steps of constructing a continuous rainfall space-time function by taking discrete rainfall data as supervision through implicit neural representation, constructing a physical enhanced neural differential equation, embedding the continuous rainfall space-time function as an external driving item, ensuring consistency of dynamic mapping and a physical rule through residual constraint of a hydraulic control equation, constructing physical constraint by utilizing hydrodynamic monitoring data and hydraulic residual errors at random space-time points, carrying out iterative optimization on dynamic mapping parameters, taking the optimized dynamic mapping as an environment simulator, combining probability distribution of random interference events, and solving gate opening and closing time sequences meeting safety constraint through distributed robust optimization.

Inventors

  • ZHANG LISUN
  • ZHANG GUOWEN
  • LIU YANG
  • ZHONG ZHIJIAN
  • Xiong Suqing
  • HU YAN
  • ZHOU XIAOHUI
  • WANG XIAOBAO
  • WANG LICHAO
  • HUA JIE
  • HU BO
  • HUANG WEI
  • ZHANG NA
  • WAN GUOYONG
  • Huang Lanbo
  • ZHANG XIUFENG

Assignees

  • 中铁水利信息科技有限公司
  • 河北省水利规划设计研究院有限公司

Dates

Publication Date
20260508
Application Date
20260410

Claims (10)

  1. 1. The hydraulic engineering intelligent design method based on multi-source heterogeneous data fusion is characterized by comprising the following steps of: S1, acquiring rainfall forecast data of a weather radar and hydrodynamic monitoring data of a river network, and performing cleaning treatment on an empty coordinate system and an abnormal value to obtain an aligned multisource data set; S2, adopting implicit neural representation to take discrete rainfall data in a multisource data set as supervision, and constructing a continuous rainfall space-time function capable of outputting rainfall intensity at any space-time position; s3, constructing a physical enhanced nerve differential equation, embedding a continuous rainfall space-time function into the nerve differential equation as an external driving term to describe dynamic mapping of the river network hydraulic state evolution along with time, wherein the internal structure of the nerve differential equation ensures consistency with a physical rule by introducing residual constraint of a hydraulic control equation; S4, utilizing hydrodynamic monitoring data in the multisource data set as a fitting target, constructing physical constraint based on hydrodynamic residual errors of dynamic mapping at random space-time points, and carrying out iterative optimization on parameters of the dynamic mapping to obtain optimized dynamic mapping; and S5, taking the optimized dynamic mapping as an environment simulator, combining probability distribution of random interference events in upstream incoming flow, and solving a gate opening and closing time sequence meeting preset safety constraint through a robust optimization algorithm to be used as a design scheme of pump gate group joint scheduling.
  2. 2. The intelligent hydraulic engineering design method based on multi-source heterogeneous data fusion according to claim 1, wherein the step S2 specifically comprises: Constructing a learnable continuous mapping function taking space-time coordinates as input and rainfall intensity as output, embedding characteristics of the input coordinates by the mapping function through multi-resolution hash coding, processing embedded characteristics through a plurality of cascaded micro-transformable layers, wherein each transformation layer comprises linear transformation and sine period activation functions; Taking the discrete rainfall data in the multi-source data set as supervision, constructing a loss function to measure the deviation between the mapping function output and the discrete observed value; Based on the loss function, the parameters of the mapping function are adjusted by using a gradient descent method until convergence, so that a continuous rainfall space-time function capable of outputting rainfall intensity at any space-time position is obtained.
  3. 3. The intelligent hydraulic engineering design method based on multi-source heterogeneous data fusion according to claim 2, wherein the parameters of the mapping function are adjusted by using a gradient descent method based on a loss function, and the method specifically comprises the following steps: Calculating the parameter gradient of a learnable mapping function in the continuous rainfall spatiotemporal function based on the deviation value output by the loss function; According to the space-time distribution characteristics of the parameter gradients, self-adaptively encrypting sampling points in local space-time areas with loss values exceeding a threshold value, and recalculating the parameter gradient contributions of the corresponding areas; and updating parameters by adopting an adaptive learning rate associated with the current iteration step number in combination with the frequency response characteristic of the sine period activation function, judging the descending amplitude of the loss function after the parameters are updated, and terminating iteration if the descending amplitude of a plurality of continuous iteration steps is smaller than a preset value.
  4. 4. The intelligent hydraulic engineering design method based on multi-source heterogeneous data fusion according to claim 1, wherein the step S3 specifically comprises: Constructing differential relation expression taking a river network hydraulic state as a dependent variable and a time coordinate as an independent variable, wherein the differential relation expression comprises a state change rate, a current state and a learning mapping structure between external driving items; taking the output of the continuous rainfall space-time function as an external driving item, and embedding the continuous rainfall space-time function into the source item position in the differential relation expression according to the physical conservation principle; In the updating process of differential relation expression, a residual calculation layer of a hydraulic control equation is introduced, the residual calculation layer takes the current state as input, outputs a state quantity to be substituted into a deviation value after the control equation, and the corresponding deviation value is used as a correction term to be superimposed on the state change rate, so that the evolution process is ensured to meet the physical conservation law.
  5. 5. The intelligent design method for hydraulic engineering based on multi-source heterogeneous data fusion according to claim 4, wherein the method is characterized in that the output of continuous rainfall space-time function is used as an external driving item, and the output is embedded into the source item position in differential relation expression according to the principle of physical conservation, and specifically comprises the following steps: the rainfall intensity output value of the target river reach position at the current moment is read from the continuous rainfall space-time function, and the net rain flux entering the river at the corresponding moment is calculated by combining the water collecting area and the yield coefficient corresponding to the corresponding river reach; writing the net rain flux into the right side of the differential relation expression in a mode of lateral inflow of unit river length according to the source item form of the continuous equation in the Saint Violet equation set; And (3) carrying out physical conservation constraint on the net rain flux written into the source item according to the water surface width value corresponding to the current water level of the river reach, so as to ensure that the water level lifting rate caused by lateral inflow is matched with the river channel energy regulating capacity.
  6. 6. The intelligent hydraulic engineering design method based on multi-source heterogeneous data fusion according to claim 1, wherein the step S4 specifically comprises: Calculating the deviation between the output value and the measured value of the current dynamic mapping monitoring point by taking hydrodynamic monitoring data in the multisource data set as a fitting reference; Randomly sampling physical constraint points in a global space-time range, substituting hydraulic state values at the sampling points into a hydraulic control equation, and calculating continuity deviation and momentum deviation to form a physical constraint residual error; And carrying out weighted combination on the deviation amount of the monitoring point and the physical constraint residual error to construct comprehensive loss, carrying out reverse correction on the dynamically mapped parameters according to the loss value, and recalculating the physical constraint residual error after each correction until the comprehensive loss meets the convergence condition.
  7. 7. The intelligent hydraulic engineering design method based on multi-source heterogeneous data fusion according to claim 6, wherein the forming process of the physical constraint residual is as follows: A self-adaptive sampling strategy based on gradients is adopted in a space-time calculation domain, and the sampling density of physical constraint points is increased in a region with the change rate exceeding a threshold value according to the change rate distribution of the dynamic mapping output hydraulic state; the space-time coordinates at the sampling points are input into dynamic mapping to obtain corresponding water level values and flow values, and partial derivatives of the water level and the flow on time and space are calculated by utilizing an automatic differentiation technology; Substituting the water level value, the flow value and partial derivatives thereof into a continuous equation and a momentum equation in the san View equation set, calculating the difference between the left end and the right end of the equation, and taking the corresponding difference as a physical constraint residual.
  8. 8. The intelligent hydraulic engineering design method based on multi-source heterogeneous data fusion according to claim 1, wherein the step S5 specifically comprises: extracting the types and intensity distribution of random disturbance events of weed masses and floaters in upstream inflow based on historical hydrologic data, and constructing a probability density function of the disturbance events; Performing discrete sampling on the probability density function to generate a plurality of random interference scenes, inputting each scene as a boundary condition into the optimized dynamic mapping, and deducing evolution tracks of river network hydraulic states under different interferences; The gate opening and closing time sequence is used as a decision variable, the highest water level of the river channel in all random interference scenes does not exceed the elevation of the embankment as a constraint condition, the total energy consumption of a pump station is minimum as an objective function, and a distributed robust optimization method is adopted to solve the gate opening and closing time sequence meeting the constraint.
  9. 9. The intelligent hydraulic engineering design method based on multi-source heterogeneous data fusion according to claim 8, wherein the method for solving gate opening and closing time sequences meeting constraints by adopting a distributed robust optimization method comprises the following steps: Based on the hydraulic state evolution tracks deduced under a plurality of random interference scenes, extracting the highest water level sequence of the river channel corresponding to each random interference scene, and constructing an empirical distribution function taking the gate opening and closing time sequence as an independent variable; Based on an empirical distribution function, describing an offset range between the actual distribution and the empirical distribution of the random interference scene by adopting a moment uncertainty set, and converting the original constraint condition into a distribution robust constraint which is required to be satisfied for all possible distributions; and (3) equivalently converting the distributed robust constraint into a split planning form containing risk measurement parameters, iteratively solving the corresponding split planning through a binary search method, updating the gate opening and closing time sequence in each iteration, and verifying the feasibility of the gate opening and closing time sequence on all the distributions in the moment uncertainty set until the gate opening and closing time sequence which meets the preset safety constraint and minimizes the total energy consumption of the pump station is obtained.
  10. 10. The hydraulic engineering intelligent design system based on multi-source heterogeneous data fusion, which is characterized by being used for executing the hydraulic engineering intelligent design method based on multi-source heterogeneous data fusion as claimed in any one of claims 1-9, and comprising the following steps: the data acquisition and preprocessing module is used for acquiring weather radar rainfall forecast data and river network hydrodynamic monitoring data, and performing space coordinate system and outlier cleaning processing to obtain an aligned multisource data set; the continuous rainfall space-time function construction module is used for constructing a continuous rainfall space-time function capable of outputting rainfall intensity at any space-time position by taking discrete rainfall data in the multi-source data set as supervision by adopting implicit neural representation; the physical enhancement nerve differential equation construction module is used for constructing a physical enhancement nerve differential equation, embedding a continuous rainfall space-time function into the nerve differential equation as an external driving term to describe dynamic mapping of the river network hydraulic state evolution along with time, and ensuring consistency with a physical rule by introducing residual constraint of a hydraulic control equation into an internal structure of the nerve differential equation; The dynamic mapping optimization module is used for constructing physical constraints based on hydraulic residuals of dynamic mapping at random space-time points by taking hydrodynamic monitoring data in a multi-source data set as a fitting target, and carrying out iterative optimization on parameters of the dynamic mapping to obtain optimized dynamic mapping; And the pump gate group joint scheduling scheme solving module is used for solving a gate opening and closing time sequence meeting preset safety constraint through a robust optimization algorithm by taking the optimized dynamic mapping as an environment simulator and combining probability distribution of random interference events in upstream incoming flow to be used as a pump gate group joint scheduling scheme.

Description

Hydraulic engineering intelligent design method and system based on multi-source heterogeneous data fusion Technical Field The invention relates to the technical field of hydraulic engineering, in particular to a hydraulic engineering intelligent design method and system based on multi-source heterogeneous data fusion. Background Along with the rapid development of weather radar observation technology, hydrological automatic measurement and report technology and numerical simulation methods, more and more multi-source heterogeneous data are accumulated in the field of hydraulic engineering planning and design. How to effectively fuse the data with different sources, different formats and different sampling frequencies and apply the data to the intelligent design of a pump gate group joint scheduling scheme has become a research hotspot in the current hydraulic engineering field. When the local geological exploration data is lost or random interference such as weed groups in upstream inflow cannot be covered by sensing equipment, the neural network can forcedly smooth the abrupt change areas which are not covered by data in the training process in order to strictly meet the mathematical continuity requirement of the san-Violet range group, so that the generated scheduling scheme perfectly accords with the physical conservation law in mathematics, but exactly avoids the truly existing embankment leakage weak zone or sundry blockage section in engineering practice, so that the physical equation is changed from the original perfect security guardrail to the disguise for covering the real engineering risk, and finally, scheduling instructions fail under extreme working conditions and cause the consequences of embankment or waterlogging. Disclosure of Invention The invention aims to provide a hydraulic engineering intelligent design method and system based on multi-source heterogeneous data fusion, so as to solve the problems in the background. The aim of the invention can be achieved by the following technical scheme: a hydraulic engineering intelligent design method based on multi-source heterogeneous data fusion comprises the following steps: S1, acquiring rainfall forecast data of a weather radar and hydrodynamic monitoring data of a river network, and performing cleaning treatment on an empty coordinate system and an abnormal value to obtain an aligned multisource data set; S2, adopting implicit neural representation to take discrete rainfall data in a multisource data set as supervision, and constructing a continuous rainfall space-time function capable of outputting rainfall intensity at any space-time position; s3, constructing a physical enhanced nerve differential equation, embedding a continuous rainfall space-time function into the nerve differential equation as an external driving term to describe dynamic mapping of the river network hydraulic state evolution along with time, wherein the internal structure of the nerve differential equation ensures consistency with a physical rule by introducing residual constraint of a hydraulic control equation; S4, utilizing hydrodynamic monitoring data in the multisource data set as a fitting target, constructing physical constraint based on hydrodynamic residual errors of dynamic mapping at random space-time points, and carrying out iterative optimization on parameters of the dynamic mapping to obtain optimized dynamic mapping; and S5, taking the optimized dynamic mapping as an environment simulator, combining probability distribution of random interference events in upstream incoming flow, and solving a gate opening and closing time sequence meeting preset safety constraint through a robust optimization algorithm to be used as a design scheme of pump gate group joint scheduling. The invention further provides a further scheme that the S2 specifically comprises the following steps: Constructing a learnable continuous mapping function taking space-time coordinates as input and rainfall intensity as output, embedding characteristics of the input coordinates by the mapping function through multi-resolution hash coding, processing embedded characteristics through a plurality of cascaded micro-transformable layers, wherein each transformation layer comprises linear transformation and sine period activation functions; Taking the discrete rainfall data in the multi-source data set as supervision, constructing a loss function to measure the deviation between the mapping function output and the discrete observed value; Based on the loss function, the parameters of the mapping function are adjusted by using a gradient descent method until convergence, so that a continuous rainfall space-time function capable of outputting rainfall intensity at any space-time position is obtained. The invention further provides a method for adjusting parameters of a mapping function by using a gradient descent method based on the loss function, which specifically comprises the following st