Hydraulic engineering transportation management cooperative control system based on digital twinning

Abstract

The invention discloses a hydraulic engineering operation and management cooperative control system based on digital twinning, and the system comprises the following steps: a data assimilation module which collects observation station, remote sensing and meteorological data and assimilates the data to obtain an assimilation estimator; the dictionary and lifting module is used for generating a lifting variable sequence according to hydrodynamic force prior; the structure identification and uncertainty module is used for identifying a lifting linear model and recursively predicting uncertainty under the constraints of conservation, monotonicity, dissipation and spectral radius; the terminal security module constructs a positive invariant terminal security set according to a support function; the opportunity constraint module is used for setting a total out-of-limit probability budget and generating an opportunity constraint substitution set; the Koopman rolling optimization module is used for solving the control sequence and executing the first control quantity; and the online updating and correcting and deformation and verification module is used for executing residual error triggering low-rank correction, projection return and support surface parameter online deformation. According to the invention, risk-controllable and stable and efficient collaborative scheduling is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent control of water conservancy projects, and in particular to a water conservancy project operation and management collaborative control system based on digital twinning. BACKGROUND

[0002] The existing water conservancy project operation and management relies on mechanism hydrodynamic models and experience scheduling rules, and is assisted by PID, LQR or conventional MPC to drive the linkage of gates and pump stations, but the uncertainty of incoming water and exogenous inflow, the boundary and change rate constraints of the execution mechanism, the conservative setting of model linearization and fixed safety margin, and the spatiotemporal alignment and quality control problems caused by heterogeneous data sources make the state estimation and prediction error accumulate, the out-of-limit alarm and the end infeasibility appear alternately, and the boundary tightness and risk level are difficult to be self-consistent and unified by the rolling optimization.

[0003] The existing integration of digital twinning and MPC mostly adopts low-dimensional linear prediction models and empirical features, lacks an observable dictionary and systematic variable construction based on hydrodynamic prior, and is difficult to meet the structural constraints of conservation, monotonicity, dissipation and spectral radius, the opportunity constraint often allocates a static or offline budget that cannot reflect the spatiotemporal correlation uncertainty and covariance structure, and the terminal constraint is mostly approximated by a spherical domain or a box domain, which is difficult to describe the geometric shape of water level, flow and control boundary and lacks positive invariance, thereby causing end infeasibility and optimization shock in the long-term rolling process.

[0004] The offset caused by model mismatch and disturbance is often handled by full-state re-identification, which has high computational overhead and lacks projection back to positive guarantee after updating, and the existing scheme generally lacks a low-rank incremental correction mechanism for improved linear models and a joint feasible region projection process, and also lacks a closed-loop strategy that unifies the shrinkage of support function as the core with the shrinkage coefficient and links with the spatiotemporal budget allocation variable, resulting in risk budget depletion or excessive conservatism.

[0005] Therefore, how to provide a water conservancy project operation and management collaborative control system based on digital twinning is a problem that those skilled in the art need to solve. SUMMARY

[0006] One object of the present application is to provide a water conservancy project operation and management collaborative control system based on digital twinning, which integrates digital twinning assimilation, observable dictionary and improved variable modeling, structural constraint identification that meets conservation / monotonicity / dissipation / spectral radius, opportunity constraint risk budget allocation, support function positive invariance terminal safety set, residual trigger low-rank incremental correction and projection back to positive, and completes the rolling optimization collaborative scheduling of water level, flow and execution mechanism, and has the advantages of controllable out-of-limit probability, full-time domain feasibility, fast response, self-adaptation to disturbance, high resource utilization efficiency and scalable deployment.

[0007] According to the water conservancy project operation and management collaborative control system based on digital twinning, the water conservancy project operation and management collaborative control system based on digital twinning comprises: a data assimilation module, configured to converge station data, remote sensing data and meteorological data according to a unified timestamp, assimilate the converged data to generate assimilated estimates, and establish a time index and a space index; a dictionary and lifting module, configured to construct an observable dictionary according to a hydrodynamic prior, map the assimilated estimates, control variables and exogenous inflows into a lifting variable sequence, and a structure identification and uncertainty module, configured to identify a lifting linear model under the constraints of conservation, monotonicity, dissipation and spectral radius, and recursively predict uncertainty; a terminal safety module, configured to map water level, flow and actuator boundaries to a lifting space, define a positive invariant terminal safety set according to a support function, and record support surface parameters; a chance constraint module, configured to set a total exceedance probability budget, configure spatiotemporal budget allocation variables, and generate a chance constraint surrogate set in combination with the predicted uncertainty; a Koopman rolling optimization module, configured to establish a Koopman model predictive control based on the lifting linear model, the chance constraint surrogate set and the terminal safety set, solve a control sequence and execute a first control variable; an online updating and correction module, configured to update the lifting variables, the spatiotemporal budget allocation variables and the predicted uncertainty by assimilating observations, generate a residual error trigger flag, trigger a low-rank incremental correction when the time comes, and project back to a joint feasible region; a deformation and verification module, configured to deform the support surface parameters according to the assimilated estimates, uniformly shrink and adjust the spatiotemporal budget allocation variables according to a shrinkage coefficient, complete positive invariance verification, and take effect in the next control cycle.

[0008] Optionally, the modules are connected through the following methods: The station data, remote sensing data and meteorological data are assimilated by digital twinning to obtain assimilated estimates, an observable dictionary is constructed according to a hydrodynamic prior, and lifting variables are generated; The lifting linear model is identified under the constraints of conservation, monotonicity, dissipation and spectral radius, and the predicted uncertainty is formed; The water level, flow and actuator boundary constraints are mapped to the lifting space, a positive invariant terminal safety set is constructed according to a support function, and support surface parameters are recorded; The total exceedance probability budget is set, the chance constraint surrogate set is generated in combination with the predicted uncertainty, and the spatiotemporal budget allocation variables are configured; The Koopman model predictive control is established based on the lifting linear model, the chance constraint surrogate set and the terminal safety set, and the first control variable is executed; The observation data after execution are obtained, the lifting variables, the spatiotemporal budget allocation variables and the predicted uncertainty are updated, and residual error statistics are calculated; Performing low-rank incremental correction on the boosted linear model when the residual trigger condition is met and projecting back to the joint feasible region of conservation, monotonicity, dissipation and spectral radius constraint; According to the assimilation estimator, the parameters of the support surface are linearly changed, and the positive invariance is verified. If the verification fails, the shrinkage coefficient is uniformly shrunk, and the spatio-temporal budget allocation is adjusted. The next control cycle takes effect.

[0009] Optionally, the assimilation estimator generation and the boosted variable generation specifically include: A unified timestamp is used to collect the water level and flow of the station, and remote sensing water surface shape and rainfall products are obtained. Synchronous meteorological elements and exogenous inflow prediction are performed to establish time index and spatial index; Uniform sampling interval and spatial segmentation are used for missing data correction, anomaly processing, denoising and unit consistency to form a quality-controlled observation sequence; Based on the hydrodynamic prior, the assimilation state is generated and paired with the observation sequence under the same index, and the digital twin data assimilation is performed to obtain the assimilation estimator; The observable dictionary is set according to the assimilation estimator and the hydrodynamic prior, covering basic hydraulic state quantities, combined quantities, spatial gradient quantities, cumulative quantities along the river section, control inputs and exogenous inflows, and the structure mask is applied to limit physically consistent items; The mean value of the dictionary items is removed and the scale is unified to obtain the normalized feature set; The assimilation estimator, control input and exogenous inflow are mapped to the normalized dictionary vector according to the time index and spatial index to form a continuous boosted variable sequence.

[0010] Optionally, the boosted linear model identification and uncertainty prediction generation specifically includes: The model structure and function are determined, the boosted linear model is used to represent the time advancement and observation relationship, the state transition matrix is used to obtain the next time boosted variable from the current boosted variable, the input matrix is used to introduce the control quantity influence, the exogenous input matrix is used to introduce the exogenous inflow influence, the bias term is used to describe the fixed offset, and the reconstruction matrix is used to restore the physical state from the boosted variable; The identification data is organized, the adjacent time boosted variables are paired according to the time index, and the control quantity and exogenous inflow at the same time index are aligned to form a time-ordered sample set, the identification criteria and convergence threshold are set, and the decision variables are the state transition matrix, the input matrix, the exogenous input matrix, the bias term and the reconstruction matrix; Setting constraints and performing identification, the conservation constraint defines that the linear aggregation related to the conservation of mass in the time advance is only contributed by the control quantity and the contribution of the exogenous inflow, the monotonicity constraint defines that the non-diagonal elements of the state transition matrix are not less than zero and the elements of the input matrix and the exogenous input matrix are not less than zero, the dissipation constraint defines that the energy after the first update is not higher than the energy before the update minus the dissipation amount by the energy measurement matrix and the dissipation weight matrix, and the spectral radius constraint sets the shrinkage margin and checks the upper bound of the spectral radius; Constructing prediction uncertainty, first, estimating the model parameter covariance according to the identification residual statistics, and giving the process noise covariance, then recursively lifting the variable uncertainty according to the time index, and then mapping it to the physical state uncertainty through the reconstruction matrix, forming the prediction uncertainty description containing two types of uncertainty.

[0011] Optionally, the terminal safety set construction and support surface parameter management specifically includes: Establishing a list of boundary elements and binding indexes, listing the upper limit of water level, the lower limit of ecological flow, the upper and lower limits of gate opening, the opening rate limit, the upper and lower limits of pump station power, the power change rate limit, the exogenous inflow range, the working condition interval, the corresponding cross section index, the actuator index, the equipment index and the time index; Expressing each boundary element as a lifting space inequality using the reconstruction matrix, extracting the direction coefficient and threshold value for each inequality and establishing the number, forming a list of support surface parameters; Organizing all inequalities into a half-space set, defining the terminal safety set with the half-space set, recording the mapping relationship of the number of half-spaces, the direction coefficient and the threshold value list; Performing single-step time advance based on the state transition matrix, the input matrix, the exogenous input matrix and the bias term, and using the upper and lower limits and the change rate limit for boundary value, calculating the advance evaluation quantity on each number and preserving the calculation index; Comparing the advance evaluation quantity with the corresponding threshold value one by one, recording the number that meets the condition in the terminal safety set, and marking the number that does not meet the condition as a face to be contracted and recording the deviation amplitude; Generating a data description of the terminal safety set, including the direction coefficient, the threshold value, the number, the advance evaluation quantity index and the corresponding relationship of the index to the cross section and the equipment, and storing it with the time index and the space index.

[0012] Optionally, the total exceedance probability budget configuration and opportunity constraint replacement set generation specifically includes: Defining a unified risk benchmark, setting a total exceedance probability budget, giving the effective time index range and space index range, and recording the version identifier for consistent use; Reading the prediction uncertainty, calculating the fluctuation amplitude and the correlation description for each boundary constraint item on each time index and space index, and generating a corresponding confidence measure table; A spatiotemporal budget allocation variable is set for each boundary constraint term and each time index, with an interval of zero to one, a summation constraint is imposed that the total does not exceed the total over-limit probability budget, and an allocation variable index table is recorded; A safety margin is synthesized from the confidence measure and the spatiotemporal budget allocation variable, each boundary constraint term is rewritten as a deterministic inequality, each is numbered, a set of opportunity constraint alternatives is collected, and a parameter list is formed; A one-to-one correspondence between the set of opportunity constraint alternatives and the time index and the space index is established, the number, the parameter, and the index mapping are saved, and are stored in a structured list to ensure consistent reference.

[0013] Optionally, the Koopman model predictive control rolling optimization specifically includes: A time configuration is set, giving the sampling step, the prediction time domain length, and the control time domain length, so that the reference sequence corresponds one-to-one to the time index; A decision structure is defined, the control sequence corresponds to the control value of each time index, and the lifting variable sequence corresponds to the lifting variable value of each time index; An equation relationship of the Koopman lifting linear model is assembled, the next time lifting variable is obtained using the state transition relationship, the input mapping, the exogenous inflow mapping, and the bias term, and the lifting variable is converted into the predicted water level and the predicted flow using the reconstruction matrix; The set of opportunity constraint alternatives is applied, the safety margin is synthesized from the spatiotemporal budget allocation variable and the confidence measure, two types of checks are performed at each time index and space index, the first type is that the predicted water level plus the safety margin does not exceed the upper limit of the water level, and the second type is that the predicted flow minus the safety margin does not fall below the lower limit of the flow, and the number registration is completed; The terminal safety set is applied, the terminal lifting variable is limited according to the support surface parameter at the end of the prediction time domain, and the number corresponding relationship is recorded; Supplementary boundary constraints are written, the input upper and lower bounds and the input change rate bounds are written, the exogenous inflow range set is limited, and is stored consistently with the time index and the space index; The objective function is described in detail, which is composed of four parts of linear summation and is set with weights respectively: The state deviation term, at each time index and space index, the square sum of the difference between the predicted water level and the reference water level is calculated, and the square sum of the difference between the predicted flow and the reference flow is calculated, and then the water level deviation weight and the flow deviation weight are multiplied and summed; The control amplitude term, at each time index, the square sum of the control value is calculated and multiplied by the control amplitude weight; The control increment term, at adjacent time indices, the square sum of the control difference is calculated and multiplied by the control increment weight, and the adjacent time indices are determined by the fixed sampling step; An energy consumption term, which is calculated by multiplying the instantaneous energy consumption at each time index by an energy consumption weight, is accumulated according to the pump station power and efficiency relationship, and the efficiency relationship is derived from the equipment parameter table or the calibration result; The four parts are linearly added according to weight coefficients to form a target function, and the weight is non-negative and takes a value after dimension normalization; A Koopman model predictive control rolling optimization is performed to obtain a control sequence at a starting time index and execute a first control amount.

[0014] Optionally, the online assimilation, lifting variable updating, residual error statistics, and space-time budget allocation variable adjustment specifically include: After the first control amount is executed, water levels, flow rates, gate openings, pump station powers, and external inflows are collected according to time indexes, spatial index alignment and unit consistency are completed, and an observation sequence is formed; Digital twin data assimilation is performed according to the hydrodynamic prior and the observation sequence to generate an assimilated estimate and record the time index; The assimilated estimate, the control amount, and the external inflow are mapped into a dictionary vector according to an observable dictionary, and the lifting variable is updated at the current time index; The current prediction uncertainty is calculated from the prediction uncertainty and the process noise of the last time index according to the time advancement rule, and the confidence measure is calculated together with the assimilated estimate; Residual error statistics, including absolute error, mean square error, and cumulative sum statistics, are calculated, and residual error trigger flags are generated according to threshold values; The space-time budget allocation variable is re-assigned, and the confidence measure and the used budget are set to a value between zero and one at each time index and spatial index according to the confidence measure and the used budget, and the total exceeds the probability budget; The lifting variable, the space-time budget allocation variable, the prediction uncertainty, the confidence measure, and the residual error trigger flag are stored according to the time index and the spatial index.

[0015] Optionally, the low-rank incremental correction and joint feasible region projection specifically include: A trigger condition is determined, the residual error trigger flag is set as a starting condition, the update window length, the low-rank order, the shrinkage margin, the upper limit of the projection number, and the check threshold are given, and the time index range is recorded; An incremental correction data set is organized, and the lifting variable sequence, the control amount sequence, the external inflow sequence, the predicted water level sequence, and the predicted flow rate sequence are arranged according to the time index and the spatial index within the update window to complete dimension normalization and missing data processing; An incremental structure is set, a state transition matrix, an input matrix, and an external input matrix are added with low-rank increments, a bias term is added with a vector increment, and a reconstructed matrix maintains the established mapping; The low-rank increment parameters are obtained by minimizing the prediction error with the increment correction data set, and the state transition matrix, the input matrix, the exogenous input matrix and the bias term are updated, and the position of the time index is recorded; The projection back is performed, first satisfying the conservation constraint, then the monotonicity constraint, then the dissipation constraint, and finally the spectral radius constraint, and the projection number and the convergence state are saved; The stability and feasibility check is carried out, including the spectral radius check, the energy inequality check, the conservation relationship check and the monotonicity check, and when all the checks are satisfied, the current model parameters are replaced, and when the checks are not satisfied, the last version of the parameters and the control strategy are used and the time index is recorded; The prediction uncertainty is recursively predicted according to the time index, the model parameter covariance and the process noise covariance are updated, the lifting variable covariance is calculated and is stored synchronously with the space index, and the input is provided for the rolling optimization.

[0016] Optionally, the terminal security set online linear variation, positive invariance verification and space-time budget allocation variable linkage adjustment specifically comprises: The input is determined, the support surface parameter list is determined using the assimilation estimate, the time index and the space index, including the direction coefficient and the threshold, and the numbering order is fixed; The single-step time advance is performed, the state transition matrix, the input matrix, the exogenous input matrix and the bias term are called and the boundary value is used, the advance evaluation quantity is calculated in sequence, the advance evaluation quantity is compared with the corresponding threshold one by one, and the passed number and the failed number are marked; The failed number is uniformly scaled, the threshold is reduced by the same proportion according to the scaling coefficient, the direction coefficient remains unchanged, and the updated support surface parameter list is formed; The advance evaluation quantity is recalculated under the same time index and space index, and the threshold is compared again to generate a new positive invariance determination result; The stop rule is set, and any of the following situations occurs to stop: the positive invariance is established, the scaling number reaches the upper limit; the threshold reaches the lower limit of scaling, and the total of the space-time budget allocation variables reaches the total exceeding probability budget; The terminal security set is updated to the new version, the space-time budget allocation variable is adjusted, the value is limited to between zero and one, the total does not exceed the total exceeding probability budget, the security margin used by the opportunity constraint replacement set is updated accordingly, and the time index effective in the next control period is recorded.

[0017] The beneficial effects of the present application are: The application improves variables through digital twin assimilation and observable dictionary construction, completes structure retention identification under the constraints of conservation, monotony, dissipation and spectral radius, and forms prediction uncertainty, combines with opportunity constraint space-time budget allocation and support function terminal safety set, uniformly processes water level, flow and actuator boundary in Koopman model prediction control, makes rolling optimization feasible in long time domain and limits the overrun probability within the preset budget, and realizes the trade-off between water resource allocation and energy consumption indicators under the given reference.

[0018] In the execution phase, the low-rank incremental correction triggered by the residual and the joint feasible region projection are introduced, and the online update driven by the assimilation and the space-time budget linkage are matched, so that the improved linear model and the safety set boundary can be quickly corrected under the conditions of disturbance and model slow change, thereby reducing the frequency of terminal infeasibility and optimization shock, reducing the full-quantity re-identification overhead, improving the response speed and robustness to flood peak inflow and working condition switching, and obtaining tighter constraint utilization rate and higher scheduling accuracy under the same risk level. BRIEF DESCRIPTION OF DRAWINGS

[0019] The accompanying drawings are included to provide a further understanding of the application, and constitute a part of the specification, together with the embodiments of the application, to explain the application, and do not constitute a limitation on the application. In the drawings:

[0020] Fig. 1 A flowchart of a water conservancy engineering operation management collaborative control system based on digital twin is proposed for the application; Fig. 2 A Koopman model prediction control structure diagram of a water conservancy engineering operation management collaborative control system based on digital twin is proposed for the application. DETAILED DESCRIPTION

[0021] The application will now be further described in detail in conjunction with the drawings. These drawings are all simplified schematic diagrams, and only schematically illustrate the basic structure of the application, and therefore only show the components related to the application.

[0022] REFERENCE Figs. 1-2 A water conservancy engineering operation management collaborative control system based on digital twin, comprising the following steps: A data assimilation module is used to converge the data of stations, remote sensing and weather according to a unified time stamp and assimilate to generate assimilated estimates, establish time index and space index; A dictionary and improvement module is used to construct an observable dictionary according to hydrodynamic prior, and map the assimilated estimates, control variables and exogenous inflow into an improved variable sequence; A structure identification and uncertainty module is used to identify an improved linear model and recursively predict uncertainty under the constraints of conservation, monotony, dissipation and spectral radius; Terminal safety module, for mapping water level, flow and actuator boundary to lifting space, defining positive invariant terminal safety set by support function and recording support surface parameters; Chance constraint module, for setting total exceedance probability budget, configuring space-time budget allocation variable, combining prediction uncertainty to generate chance constraint surrogate set; Koopman rolling optimization module, for establishing Koopman model predictive control based on lifting linear model, chance constraint surrogate set and terminal safety set, solving control sequence and executing first control quantity; Online updating and correction module, for assimilating observation to update lifting variable, prediction uncertainty and space-time budget allocation variable, generating residual error trigger flag, triggering low-rank incremental correction and projecting back to joint feasible region; Deformation and verification module, for deforming support surface parameters according to assimilated estimate, uniformly shrinking and adjusting space-time budget allocation variable according to shrinkage coefficient, completing positive invariance verification and taking effect in next control period.

[0023] The present application establishes a collaborative control system composed of data assimilation, dictionary and lifting, structure identification, terminal safety, chance constraint, Koopman rolling optimization, online correction and deformation verification, realizes dynamic consistency of digital twin model and physical system, unifies water level, flow and actuator constraints in prediction and control process, builds lifting variable driven by assimilation and combines with Koopman model predictive control to form self-adaptive water conservancy dispatching strategy, realizes accurate prediction and safe control of water power system under multi-time and space scale, and maintains system stability, risk control and energy consumption optimization under extreme inflow, complex boundary and equipment nonlinear conditions.

[0024] In the present embodiment, the modules are realized by the following methods: The assimilated estimate is obtained by digital twin assimilation of station, remote sensing and meteorological data, the observable dictionary is constructed according to water power prior and the lifting variable is generated; The lifting linear model identification is completed under the constraints of conservation, monotony, dissipation and spectral radius, and the prediction uncertainty is formed; The water level, flow and actuator boundary constraints are mapped to lifting space, the positive invariant terminal safety set is constructed by support function and the support surface parameters are recorded; The total exceedance probability budget is set, the chance constraint surrogate set is generated combined with prediction uncertainty and the space-time budget allocation variable is configured; The Koopman model predictive control rolling optimization is established based on lifting linear model, chance constraint surrogate set and terminal safety set, and the first control quantity is executed; Obtain post-execution observation data, update and improve variables, spatiotemporal budget allocation variables and prediction uncertainty, and calculate residual statistics; When the residual triggering condition is met, a low-rank incremental correction is performed on the lifting linear model and the model is projected back to the joint feasible region of conservation, monotonicity, dissipation and spectral radius constraints. Based on the assimilation estimate, the support surface parameters are deformed online and their positive invariance is verified. If the condition is not met, the parameters are uniformly contracted according to the contraction coefficient and the spatiotemporal budget allocation is adjusted accordingly, taking effect in the next control cycle.

[0025] This invention constructs a closed-loop rolling optimization process from data assimilation to Koopman model predictive control, linking observation, modeling, constraints, and control within a unified framework. This allows the results of digital twin assimilation to directly drive improved spatial modeling, with spatiotemporal feasibility ensured by chance constraints and terminal safety sets. By recursively predicting uncertainty and adaptively adjusting the risk level through budget allocation, a dynamic balance is achieved between maintaining model structure, satisfying constraints, and ensuring control stability. This enables high-precision water level and flow control and multi-objective collaborative optimization under complex inflow and nonlinear boundary conditions.

[0026] In this embodiment, the generation of assimilation estimators and the generation of boosting variables specifically include: A unified timestamp was used to collect station water level and flow rate data and obtain remote sensing water surface morphology and rainfall products. Meteorological elements and exogenous inflow predictions were synchronized, and time and spatial indexes were established. Unify the sampling interval and spatial segmentation, perform missing measurement correction, anomaly handling, noise reduction and unit consistency to form a quality-controlled observation sequence; Based on hydrodynamic priors, pre-assimilation states are generated and paired with observation sequences under the same index. Digital twin data assimilation is then performed to obtain assimilation estimates. An observable dictionary is set up based on assimilation estimates and hydrodynamic priors, covering basic hydraulic state quantities, combined quantities, spatial gradient quantities, cumulative quantities along the river section, control inputs and exogenous inflows, and a structural mask is applied to limit physically consistent entries. The dictionary entries are subjected to mean removal and scaling to obtain a normalized feature set; The assimilation estimator, control input, and exogenous inflow are mapped to normalized dictionary vectors by time and spatial indices, forming a continuous sequence of boosting variables.

[0027] The application completes multi-source alignment of stations, remote sensing, weather and exogenous inflow by unifying timestamps and indexes, and forms high-quality observation sequences through missing data correction, anomaly processing, denoising and unit consistency, combined with hydrodynamic priors to carry out assimilation to suppress bias and improve spatiotemporal consistency, then screens physical consistent features by observable dictionary and structure mask and implements mean removal and scale unification, maps assimilation estimates, control inputs and exogenous inflow into continuously promoted variable sequences, so that the nonlinear process obtains stable linear representation in the promotion space, thereby significantly enhancing the prediction distinguishability and control availability.

[0028] In the embodiment, the promotion linear model identification and prediction uncertainty generation specifically includes: The model structure and function are defined, the promotion linear model is used to represent the time advancement and observation relationship, the state transition matrix is used to obtain the next time promotion variable from the current promotion variable, the input matrix is used to introduce the control quantity influence, the exogenous input matrix is used to introduce the exogenous inflow influence, the bias term is used to describe the fixed offset, and the reconstruction matrix is used to restore the physical state from the promotion variable; The identification data is organized, the promotion variables of adjacent time are paired according to the time index, and are aligned with the control quantity and the exogenous inflow at the same time index, forming a sample set sorted by time, setting identification criteria and convergence threshold, and defining the decision variable as the state transition matrix, the input matrix, the exogenous input matrix, the bias term and the reconstruction matrix; The constraints are set and the identification is performed, the conservation constraint limits that the linear aggregation related to mass conservation only changes in time advancement by the contribution of the control quantity and the exogenous inflow, the monotonicity constraint limits that the non-diagonal elements of the state transition matrix are not less than zero and limits the elements of the input matrix and the exogenous input matrix are not less than zero, the dissipation constraint limits the energy after one update by the energy measurement matrix and the dissipation weight matrix, and the spectral radius constraint sets the contraction margin and checks the upper bound of the spectral radius; The prediction uncertainty is constructed, first, the model parameter covariance is estimated according to the identification residual quantity, and the process noise covariance is given, then the promotion variable uncertainty is recursively propagated according to the time index, and the physical state uncertainty is mapped through the reconstruction matrix, forming the prediction uncertainty description containing two types of uncertainty.

[0029] The application defines the time advancement and observation structure of the promotion linear model, and organizes the identification samples paired by time index under the constraints of conservation, monotonicity, dissipation and spectral radius to complete the structure preserving identification of the state transition, input and exogenous influence matrices, then constructs the parameter and process covariances from residual statistics and recursively propagates to the physical state layer to form the prediction uncertainty, so that the model provides a stable, usable and verifiable prediction base for chance constrained and Koopman rolling optimization under the conditions of physical consistency, distinguishability and measurable risk.

[0030] In the embodiment, the terminal security set construction and support surface parameter management specifically include: A boundary element list is established and indexed, and upper and lower limits of water level, ecological flow, gate opening, opening rate, pump station power, power rate, external inflow range, and working condition interval are listed, corresponding to section index, actuator index, device index, and time index; Each boundary element is expressed as a lifting space inequality using a reconstruction matrix, a direction coefficient and a threshold are extracted for each inequality, and a number is established, forming a support surface parameter list; All inequalities are organized into a half-space set, and the terminal security set is defined by the half-space set, and the mapping relationship between the number of half-spaces, the number, and the direction coefficient and threshold list is recorded; Based on the state transition matrix, the input matrix, the external input matrix and the bias term, a single-step time advance is performed, the boundary value is respectively taken as the upper and lower limits and the rate limit, the advance evaluation quantity is calculated on each number and the calculation index is reserved; The advance evaluation quantity and the corresponding threshold are compared one by one, the number that meets the condition is recorded in the terminal security set, and the number that does not meet the condition is marked as a surface to be contracted and the deviation amplitude is recorded; A data description of the terminal security set is generated, including the direction coefficient, the threshold, the number, the advance evaluation quantity index, and the corresponding relationship between the index and the section and the device, and is stored in the same time index and space index.

[0031] The present application converts water level, flow and actuator boundary into a series of indexes by means of reconstruction matrix, extracts direction coefficient and threshold to form support surface parameters, and defines positive invariant terminal security set by half-space set, calculates advance evaluation quantity on each number by single-step time advance, compares with threshold, marks the one that does not meet the condition as a surface to be contracted and records the deviation, and finally forms a traceable data description and index mapping, so that the end feasibility is structured, verifiable and maintainable in rolling optimization.

[0032] In the embodiment, the total overrun probability budget configuration and opportunity constraint replacement set generation specifically include: A unified risk benchmark is defined, a total overrun probability budget is set, a valid time index range and a space index range are given, and a version identifier is recorded for consistent use; The prediction uncertainty is read, the fluctuation amplitude and the correlation description are calculated for each boundary constraint item on each time index and space index, and the corresponding confidence measure table is generated; A space-time budget allocation variable is set for each boundary constraint item and each time index, the interval is limited to zero to one, the total sum constraint is applied, and the total overrun probability budget is not more than the total sum, and the allocation variable index table is recorded; The confidence measure and the space-time budget allocation variable are synthesized into a safety margin, each boundary constraint term is rewritten as a deterministic inequality, is numbered, is collected into a set of chance constraint alternatives, and a parameter list is formed; A one-to-one correspondence between the set of chance constraint alternatives and the time index and the space index is established, the number, the parameter and the index mapping are saved, and the structured list is stored to ensure consistent reference.

[0033] The present application sets the total overrun probability budget uniformly and binds the time and space indexes, generates the confidence measure in the form of fluctuation amplitude and correlation after reading the prediction uncertainty, configures the space-time budget allocation variable with a value between zero and one and subject to the total constraint for each boundary and each time, synthesizes the confidence and the budget into a safety margin to rewrite the random constraint into a deterministic inequality and number the chance constraint alternative set, and establishes a one-to-one mapping with the index and a parameter list, so that the risk allocation can be quantified and tracked, and consistent with the rolling optimization.

[0034] In the embodiment, the Koopman model predictive control rolling optimization specifically includes: Set the time configuration to give the sampling step, the prediction time domain length and the control time domain length, so that the reference sequence corresponds to the time index one by one; Explicit decision structure, control sequence corresponding to the control amount value of each time index, improve the variable sequence corresponding to the improvement variable value of each time index; Assemble the Koopman promotion linear model equation relationship, use the state transition relationship, input mapping, exogenous inflow mapping and bias term to get the next time promotion variable, and use the reconstruction matrix to convert the promotion variable into predicted water level and predicted flow; Apply the set of chance constraint alternatives, use the space-time budget allocation variable and the confidence measure to synthesize the safety margin, perform two types of checks at each time index and space index, the first type is that the predicted water level plus the safety margin does not exceed the water level upper limit, and the second type is that the predicted flow minus the safety margin does not exceed the flow lower limit, and complete the numbering registration; Apply the terminal safety set, limit the terminal promotion variable according to the support surface parameter at the end of the prediction time domain, and record the number corresponding relationship; Supplement the boundary type constraint, write in the input upper and lower limit and the input change rate limit, limit the exogenous inflow range set, and store it with the time index and the space index; Detailed description of the objective function, which is composed of four parts of linear addition and is set with weights respectively: State deviation term, at each time index and space index, square and accumulate the difference between the predicted water level and the reference water level, square and accumulate the difference between the predicted flow and the reference flow, multiply by the water level deviation weight and the flow deviation weight respectively, and sum up; a control amplitude term, in which the numerical value of the control quantity at each time index is squared and accumulated and multiplied by a control amplitude weight; a control increment term, in which the difference value of the control quantity at adjacent time indexes determined by a fixed sampling step is squared and accumulated and multiplied by a control increment weight; an energy consumption term, in which the instantaneous energy consumption is calculated according to the relationship between the pump station power and the efficiency at each time index, and is accumulated and multiplied by an energy consumption weight, and the efficiency relationship is derived from the device parameter table or the calibration result; the four parts are linearly added according to the weight coefficients to form the objective function, and the weight is non-negative and is taken after dimension normalization; Koopman model predictive control rolling optimization is performed, and the control sequence is obtained at the starting time index and the first control quantity is executed.

[0035] The present application aligns the reference sequence with the time index under the unified time configuration, obtains the water level and flow prediction by the Koopman promotion linear model in series with the state transition and reconstruction, synthesizes the safety margin by the space-time budget allocation variable and the confidence measure, rewrites the upper limit of the water level and the lower limit of the flow as an opportunity constraint, constrains the terminal safety set to predict the terminal, and at the same time applies the upper and lower bounds of the input and the change rate bound and the exogenous inflow range, drives the rolling optimization by the weighted objective function composed of the state deviation, the control amplitude, the control increment and the energy consumption to output the first control quantity, so that the risk is controlled, the terminal is feasible and the energy consumption is balanced in the same optimization framework.

[0036] In the embodiment, the online assimilation, promotion variable updating, residual error statistics and space-time budget allocation variable adjustment specifically include: After the first control quantity is executed, the water level, flow, gate opening, pump station power and exogenous inflow are collected according to the time index, the spatial index alignment and unit consistency are completed, and the observation sequence is formed; Digital twin data assimilation is performed according to the hydrodynamic prior and the observation sequence, and the assimilated estimate is generated and the time index is recorded; The assimilated estimate, the control quantity and the exogenous inflow are mapped into a dictionary vector according to the observable dictionary, and the promotion variable is updated at the current time index; The current prediction uncertainty is calculated from the prediction uncertainty and the process noise of the last time index according to the time advancing rule, and the confidence measure is calculated together with the assimilated estimate; Residual error statistics are calculated, including absolute error, mean square error and cumulative sum statistics, and residual error trigger marks are generated according to the threshold; The space-time budget allocation variable is re-assigned, and the value is set to be between zero and one at each time index and spatial index according to the confidence measure and the used budget, and the total exceeds the probability budget; The promotion variable, the space-time budget allocation variable, the prediction uncertainty, the confidence measure and the residual trigger flag are stored according to the time index and the space index.

[0037] The application updates the promotion variable according to the observation dictionary, recursively predicts the uncertainty and the confidence measure by assimilating the water level, the flow, the gate and the pumping station power collected after the first control variable is executed, triggers the adaptive adjustment of the space-time budget allocation variable based on the residual statistics, and keeps the total probability of exceeding the limit constraint, thereby continuously calibrating the model and the risk characterization, reducing the exceeding limit and the energy consumption, and stably maintaining the end feasibility and the control smoothness in the rolling optimization.

[0038] In the embodiment, the low-rank incremental correction and the joint feasible region projection specifically include: The trigger condition is determined, the residual trigger flag is set as the starting condition, the update window length, the low-rank order, the shrinkage margin, the upper limit of the projection number and the checking threshold are given, and the time index range is recorded; The incremental correction data set is organized, the promotion variable sequence, the control variable sequence, the exogenous inflow sequence, the predicted water level sequence and the predicted flow sequence are arranged according to the time index and the space index in the update window, and the dimension unification and the missing data processing are completed; The incremental structure is set, the state transition matrix, the input matrix and the exogenous input matrix are added with low-rank increments, the bias term is added with a vector increment, and the reconstruction matrix keeps the given mapping; The low-rank incremental parameters are obtained by minimizing the prediction error based on the incremental correction data set, the state transition matrix, the input matrix, the exogenous input matrix and the bias term are updated, and the starting position of the time index is recorded; The projection back is performed, the conservation constraint is satisfied first, then the monotonicity constraint, the dissipation constraint and finally the spectral radius constraint, the projection number and the convergence state are saved; The stability and the feasibility are checked, including the spectral radius check, the energy inequality check, the conservation relationship check and the monotonicity check, when all the checks are satisfied, the current model parameters are replaced, when the checks are not satisfied, the last version parameters and the control strategy are used and the time index is recorded; The prediction uncertainty is recursively updated according to the time index, the model parameter covariance and the process noise covariance are updated, the promotion variable covariance is calculated and stored synchronously according to the space index, and the input for the rolling optimization is provided.

[0039] The present application takes residual triggered low rank incremental correction as a starting point, first organizes promotion variables, control variables, exogenous inflow and predicted water level and predicted flow in a given window to form correction data and minimize prediction error to obtain incremental parameters, and then implements low rank update on state transition, input and exogenous input matrix and projects back to positive according to the order of conservation, monotony, dissipation and spectral radius, and replaces the model after passing the stability and feasibility check and recursively updates the parameter covariance and process noise covariance to generate new prediction uncertainty input, thereby suppressing model drift and maintaining the feasibility and robustness of rolling optimization without full-amount re-identification.

[0040] In the embodiment, the terminal security set is adjusted in linkage with linear variation, positive invariance verification and space-time budget allocation variable, which specifically includes: Explicit input, using assimilation estimates, time index and space index to determine the support surface parameter list, including direction coefficient and threshold, and fixed number order; Perform single-step time advance, call state transition matrix, input matrix, exogenous input matrix and bias term and use boundary value, calculate the promotion evaluation quantity by number, compare the promotion evaluation quantity with the corresponding threshold one by one, and mark the passed number and the failed number; Uniformly shrink the failed number, the threshold is reduced by the same proportion according to the shrinkage coefficient, and the direction coefficient remains unchanged, forming the updated support surface parameter list; Recalculate the promotion evaluation quantity under the same time index and space index and compare the threshold again to generate a new positive invariance judgment result; Set the stop rule, stop when any of the following conditions occurs: positive invariance is established, shrinkage times reaches the upper limit; the threshold reaches the shrinkage lower limit, and the total of space-time budget allocation variables reaches the total exceedance probability budget; Update the terminal security set to the new version, adjust the space-time budget allocation variable at the same time, limit the value to between zero and one, so that the total does not exceed the total exceedance probability budget, and update the security margin used by the opportunity constraint alternative set accordingly, and record the time index effective in the next control period.

[0041] The present application determines the support surface parameter list by assimilation estimates, time index and space index and fixes the number, calculates the promotion evaluation quantity by number under single-step time advance and compares it with the threshold, uniformly shrinks the failed number according to the shrinkage coefficient and keeps the direction coefficient unchanged, and repeatedly judges until the positive invariance is satisfied or the stop condition is reached, and adjusts the space-time budget allocation variable and the opportunity constraint security margin in linkage, so that the terminal security set and the risk budget are self-consistent and stable in the next control period, thereby avoiding the end infeasibility and strategy shock, reducing the conservatism and improving the utilization rate of constraint tightness and the proportion of feasible solutions of rolling optimization.

[0042] Embodiment 1: To verify the feasibility of the application in implementation, the application is applied to a typical cascade reservoir-river-pump station joint dispatching scene, the basin contains three upstream reservoir areas, two gate control sections and a group of pump stations, the goal is to meet the reservoir safety water level, downstream ecological flow and urban drainage demand under heavy rainfall, the traditional method relies on experience curve and fixed margin, in the face of uncertain inflow and execution mechanism change rate limit, there are often problems such as insufficient pre-discharge, end infeasibility and high energy consumption, therefore, the system converges the station water level, remote sensing rainfall and weather field with a unified timestamp, and forms a spatial resolution consistent assimilated estimate through digital twin assimilation, according to the hydrodynamic prior, an observable dictionary is constructed, the assimilated estimate, control and external inflow are mapped to the promotion variable sequence, under the constraints of conservation, monotonicity, dissipation and spectral radius, the promotion linear model is identified, and the uncertainty is recursively predicted to provide input for risk quantification.

[0043] In the application process, first, set the total overrun probability budget and the spatiotemporal budget allocation variable, jointly predict the uncertainty to generate a set of chance constraint alternatives, and then map the water level upper limit, ecological flow lower limit and execution mechanism boundary to the promotion space to construct a positive invariant terminal safety set in the form of support function, then based on the promotion linear model, the chance constraint alternative set and the terminal safety set, the Koopman model predictive control is established, the target function including water level deviation, flow deviation, control increment and energy consumption is set, the control sequence is output by rolling optimization and the first control quantity is implemented, after execution, new period observations are collected and assimilation update is performed, residual statistics are calculated and trigger flags are generated, when the trigger is established, the model is corrected by low rank increment and projected back to the joint feasible region, the structural constraints and stability are maintained, at the same time, the support surface parameters of the terminal safety set are implemented online deformation according to the assimilated estimate, the number of unsatisfied positive invariance is uniformly contracted by contraction coefficient, and the spatiotemporal budget allocation variable is adjusted, so that the risk and feasibility are consistent in the next control period.

[0044] To quantify the effect, a continuous rainstorm situation is constructed for comparison with traditional experience scheduling, the inflow peak intensity is in the high quantile interval, the pump station rated power and gate opening rate are limited according to the equipment manual, the statistics cover multiple control periods, the key indicators include maximum water level overrun, overrun probability, ecological flow undersupply rate, pump station total energy consumption, outflow fluctuation coefficient and optimization solving time, the results are shown in Table 1, it can be seen that under the same inflow and constraint conditions, the system significantly reduces the overrun risk and reduces the energy consumption, while keeping the solution real-time and end feasibility consistent: Table 1 Performance comparison table

[0045] It can be seen from the comparison that the digital twin assimilation improves the state consistency and the inflow description accuracy, the structure retention identification and the opportunity constraint budget enable the rolling optimization to maintain feasibility in a long time domain, the positive invariant terminal safety set guarantees the terminal constraint, the low rank correction and the projection back to the positive inhibit model drift and oscillation, and the overall implementation realizes the comprehensive goals of controllable overrun probability, energy consumption reduction and response speedup.

[0046] The above merely describes preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art, according to the technical solution and the inventive concept of the present application, can make equivalent replacements or changes within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A collaborative control system for the operation and management of water conservancy projects based on digital twins, characterized in that, include: The data assimilation module is used to aggregate station, remote sensing, and meteorological data according to a unified timestamp, assimilate and generate assimilation estimates, and establish time and spatial indexes. The dictionary and boosting module is used to construct an observable dictionary based on hydrodynamic priors, mapping assimilation estimates, control variables, and exogenous inflows into a sequence of boosting variables. The structure identification and uncertainty module is used to identify and recursively predict uncertainties in improved linear models under constraints of conservation, monotonicity, dissipation, and spectral radius. The terminal safety module is used to map water level, flow rate and actuator boundary to the lifting space, define a positive invariant terminal safety set according to the support function and record the support surface parameters; The opportunity constraint module is used to set the total probability budget of exceeding the limit, configure the spatiotemporal budget allocation variables, and generate an opportunity constraint alternative set in combination with the prediction uncertainty. The Koopman rolling optimization module is used to establish Koopman model predictive control based on the lifting linear model, the opportunity constraint substitution set and the terminal safety set, solve the control sequence and execute the first control variable; The online update and correction module is used to assimilate observation update enhancement variables, prediction uncertainty and spatiotemporal budget allocation variables, generate residual trigger flags, and when triggered, perform low-rank incremental correction and project back to the joint feasible region. The deformation and verification module is used to uniformly shrink and adjust the spatiotemporal budget allocation variables according to the shrinkage coefficient based on the deformation support surface parameters of the assimilation estimate, complete the positive invariance verification, and take effect in the next control cycle.

2. The collaborative control system for water conservancy project operation and management based on digital twins according to claim 1, characterized in that, The modules are connected in the following way: Assimilation estimates are obtained by assimilating station, remote sensing and meteorological data through digital twins. An observability dictionary is constructed based on hydrodynamic priors and improvement variables are generated. Under the constraints of conservation, monotonicity, dissipation, and spectral radius, the identification of lifting linear models and the formation of prediction uncertainty are completed; Map water level, flow rate and actuator boundary constraints to the lifting space, construct a positive invariant terminal safety set according to the support function and record the support surface parameters; Set a total over-limit probability budget, jointly predict uncertainties to generate an opportunity constraint alternative set, and configure spatiotemporal budget allocation variables; Based on the lifting linear model, the opportunity constraint substitution set, and the terminal safety set, a Koopman model is established for predictive control rolling optimization, and the first control variable is executed. Obtain post-execution observation data, update and improve variables, spatiotemporal budget allocation variables and prediction uncertainty, and calculate residual statistics; When the residual triggering condition is met, a low-rank incremental correction is performed on the lifting linear model and the model is projected back to the joint feasible region of conservation, monotonicity, dissipation and spectral radius constraints. Based on the assimilation estimate, the support surface parameters are deformed online and their positive invariance is verified. If the condition is not met, the parameters are uniformly contracted according to the contraction coefficient and the spatiotemporal budget allocation is adjusted accordingly, taking effect in the next control cycle.

3. The water conservancy project operation and management collaborative control system based on digital twin according to claim 2, characterized in that, The generation of the assimilation estimator and the generation of the boosting variables specifically include: Data from monitoring stations, remote sensing data, and meteorological data are aggregated according to a unified timestamp, and time and spatial indexes are established. By unifying the sampling interval and spatial segmentation, missing measurement correction, noise reduction and unit consistency are completed to form an observation sequence; Based on hydrodynamic priors, digital twin data assimilation is performed to generate assimilation estimates; Based on the assimilation estimator and the hydrodynamic prior verification observable dictionary, the entries include basic hydraulic state quantities, combined quantities based on state quantities, spatial gradient quantities, cumulative quantities along the river section, control inputs and exogenous inflows; The assimilation estimator, control input, and exogenous inflow are mapped to observable dictionary vectors, generating a sequence of boosting variables that correspond one-to-one with the time and spatial indices.

4. The water conservancy project operation and management collaborative control system based on digital twin according to claim 2, characterized in that, The improvement of uncertainty generation in linear model identification and prediction specifically includes: A lifting linear model structure is adopted, which includes a state transition matrix, an input matrix, an exogenous input matrix, a bias term, and a reconstruction matrix. Pair boosting variables, control quantities, and exogenous inflows according to time index to generate a set of identification samples and set identification criteria; Identify the state transition matrix, input matrix, exogenous input matrix, bias term, and reconstruction matrix under conservation constraints, monotonicity constraints, dissipation constraints, and spectral radius constraints; Based on the identification of residuals and data statistics, the model parameter covariance and process noise covariance are estimated. The uncertainty of variables is increased by recursion according to the time index. The reconstructed matrix is ​​mapped to the physical state uncertainty, forming a description of the prediction uncertainty.

5. A collaborative control system for water conservancy project operation and management based on digital twins according to claim 2, characterized in that, The construction and support surface parameter management of the terminal security set specifically includes: Establish a set of boundary features and bind it with time and spatial indices; By reconstructing the matrix, the set of boundary features is transcribed into lifting space inequalities, and the directional coefficients and thresholds are extracted and numbered to form support surface parameters. A half-space set is constructed using support surface parameters, and a terminal security set is defined, recording the mapping between quantity and number; Under single-step time advancement, the state transition matrix, input matrix, exogenous input matrix and bias term are applied to calculate the advancement evaluation quantity for each half of the space based on the boundary element values; The evaluation quantity will be compared with the corresponding threshold one by one. If all of them do not exceed the threshold, they will be determined as the positive and invariant terminal safety set. If there is an out-of-limit number, it will be marked as a surface to be contracted. Store the data description of the terminal security set, including direction coefficients, thresholds, numbers, and propulsion evaluation indexes, and keep them consistent with the time and spatial indexes.

6. A collaborative control system for water conservancy project operation and management based on digital twins according to claim 2, characterized in that, The total excess probability budget allocation and opportunity constraint substitution set generation specifically includes: Set the total out-of-limit probability budget and lock the time and spatial index ranges; Based on the uncertainty of prediction, the fluctuation amplitude and correlation description of each boundary constraint term at each time index and spatial index are calculated to form a confidence measure; Establish a spatiotemporal budget allocation variable for each boundary constraint term and each time index, limit the value to between zero and one, and set the sum to not exceed the total over-limit probability budget; Based on the confidence level measure and the spatiotemporal budget allocation variables, the safety margin is synthesized, the boundary constraint terms are rewritten as deterministic inequalities, and the opportunity constraint substitution set is formed and numbered. Establish a one-to-one correspondence between the opportunity constraint substitution set and the time index and spatial index, and save the number and parameter list.

7. A collaborative control system for water conservancy project operation and management based on digital twins according to claim 2, characterized in that, The Koopman model predictive control rolling optimization specifically includes: Fixed-time index, setting sampling step size, prediction time domain length, control time domain length and reference sequence; Based on the lifting linear model, lifting variables, control quantities, exogenous inflows and bias terms are joined by adjacent time indices, and the reconstruction matrix is ​​used to generate predicted water level and predicted flow. Load the opportunity constraint substitution set, calculate the safety margin based on the spatiotemporal budget allocation variable and confidence measure, and require that the predicted water level plus the safety margin does not exceed the upper limit of the water level and the predicted flow minus the safety margin is not lower than the lower limit of the flow in each time index and spatial index. Apply a terminal safety set at the end of the prediction time domain and constrain the terminal boosting variables according to the support surface parameters; Write the upper and lower bounds of the input, the bound of the rate of change of the input, and the range of the exogenous inflow, and bind them with the time index and the spatial index; The objective function is defined, consisting of state deviation, control amplitude, control increment and energy consumption, and weights are assigned to each. The state deviation is calculated based on the reconstructed predicted water level and reference water level, as well as the reconstructed predicted flow rate and reference flow rate. The Koopman model predictive control rolling optimization is solved at the starting time index to generate a control sequence and execute the first control variable.

8. A collaborative control system for water conservancy project operation and management based on digital twins according to claim 2, characterized in that, The online assimilation, improved variable update, residual statistics, and spatiotemporal budget allocation variable adjustment specifically include: After executing the first control variable, water level, flow rate, gate opening, pump station power and exogenous inflow are collected according to time index to complete spatial index alignment and unit consistency, forming an observation sequence; Digital twin data assimilation is performed based on hydrodynamic priors and observation sequences to generate assimilation estimates and record time indices; The assimilation estimator, control variable, and exogenous inflow are mapped to dictionary vectors according to the observable dictionary, and the boosting variable is indexed and updated at the current time. Based on the time-progression rule, the current prediction uncertainty is calculated from the prediction uncertainty of the previous time index and the process noise, and the confidence metric is calculated in combination with the assimilation estimator. Calculate residual statistics, including absolute error, mean square error, and cumulative sum statistics, and generate residual trigger flags based on thresholds; Reassign the spatiotemporal budget allocation variable, setting its value between zero and one in each time and spatial index based on the confidence metric and the used budget, and ensuring that the sum does not exceed the total over-limit probability budget. Store boosting variables, spatiotemporal budget allocation variables, forecast uncertainty, confidence metrics, and residual trigger flags using time and spatial indexes.

9. A collaborative control system for water conservancy project operation and management based on digital twins according to claim 2, characterized in that, The low-rank incremental correction and joint feasible region projection specifically include: Using the residual trigger flag as the starting condition, set the update window length, low-rank order, shrinkage margin, and upper limit of projection times; Within the update window, improve variables, control quantities, exogenous inflows, predicted water levels, and predicted flow rates are organized by time and spatial indexes to form an incremental correction data set. Low-rank incremental correction is applied to the lifting linear model, which is applied to the state transition matrix, input matrix and exogenous input matrix respectively. The bias term adopts vector increment, and the reconstruction matrix remains unchanged. The incremental parameters are obtained and the matrix group and bias terms are updated based on the incremental correction dataset. The constraints are projected back to the joint feasible region in the following order: conservation constraints, monotonic constraints, dissipative constraints, and spectral radius constraints. Perform stability and feasibility checks. If all conditions are met, replace the current model parameters. If not, use the parameters and control strategies from the previous version. Based on the uncertainty of the recursive prediction of the replaced model, update the model parameter covariance, process noise covariance and lifting variable covariance, and store them in a consistent manner with the time index and spatial index.

10. A collaborative control system for water conservancy project operation and management based on digital twins according to claim 2, characterized in that, The online deformation and positive invariance verification of the terminal security set, along with the coordinated adjustment of spatiotemporal budget allocation variables, specifically includes: The support surface parameter numbers of the terminal security set are located using assimilation estimators, temporal indices, and spatial indices. The advancement evaluation quantity for each number is calculated based on the linear model and boundary values ​​and compared with the threshold to form a positive invariance determination. For those that fail to pass the numbering test, a uniform shrinkage threshold is applied based on the shrinkage coefficient, while the directional coefficient remains unchanged. The propulsion evaluation quantity is recalculated using the same time index and spatial index, and the positive invariance determination is performed again. Update the terminal security set according to the judgment result, and record the mapping between the direction coefficient, threshold and number; Adjust the spatiotemporal budget allocation variables to ensure that the sum does not exceed the total over-limit probability budget, synchronously update the safety margin of the opportunity constraint substitution set, and set it to take effect in the next control cycle.

Images