A hydropower station water regime analysis system based on big data

Abstract

The application discloses a hydropower station water regime analysis system based on big data, relates to the technical field of water conservancy informatization and hydrology and water regime, and comprises a data bus module, a unified risk scale module, an editable structure memory module, a phase change structure signature module, an anti-fact bifurcation deduction module and a constraint feasible region synthesis module and a closed loop backfilling module.

Description

Technical Field

[0001] This invention relates to the field of water conservancy and hydropower informatization and hydrological information analysis technology, specifically a hydropower station hydrological information analysis system based on big data. Background Technology

[0002] During hydropower station operation, it is necessary to consider the interconnected impacts of rainfall forecasts, upstream inflow changes, and station-wide scheduling conditions. This requires continuous analysis and risk assessment of reservoir water levels, outflow rates, downstream control section water level change rates, gate opening and closing processes, and unit start-up and shutdown status to support safe scheduling and emergency response under multiple constraints. With the increase in the number of monitoring points, sampling frequency, and external forecast data sources, hydrological observation data and scheduling data exhibit characteristics such as high dimensionality, strong time-varying nature, frequent data gaps, and distribution drift. This necessitates that the hydrological analysis system not only provide prediction results but also provide executable scheduling action sequences under constraints and be able to self-calibrate based on actual responses after execution.

[0003] Existing technologies typically achieve risk control by statistically analyzing hydrological observation data or using model-based hydrological forecasting and scheduling simulations, combined with empirical thresholds or fixed safety margins, and by screening candidate scheduling schemes under multiple scenarios. However, certain limitations exist, such as: the uncertainty caused by inconsistent timestamps between different data sources and the lack of data completion is difficult to characterize uniformly, leading to inconsistent risk scales in the analysis chain; phased hydrological changes often rely on fixed thresholds for discrimination, making it difficult to adaptively identify abrupt changes and structural drift; prediction models and scheduling feasibility domain screening are often constructed separately, making it difficult to ensure scale consistency and closed-loop self-calibration between prediction results, constraint boundary contraction, and strategy output; and the actual response after execution is difficult to effectively feed back into the analysis and prediction stages, leading to error accumulation during long-term system operation and reducing the stability and interpretability of strategy output.

[0004] Therefore, there is an urgent need for a hydropower station hydrological analysis system based on big data, which can achieve unified alignment and handling of missing measurement uncertainties of multi-source hydrological observation data and scheduling condition data within the data stream of the operation window, forming a verifiable and consistent risk scale. Under the consistent risk scale, it can identify phased hydrological changes, predict future hydrological evolution, and output executable scheduling action sequences under multiple constraints. At the same time, after execution, it can use the real response to perform online calibration of the risk scale and state representation, thereby meeting the needs of continuous analysis and risk control under complex hydrological conditions and multi-constraint scheduling. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and propose a hydropower station hydrological analysis system based on big data to solve the above-mentioned problems.

[0006] The objective of this invention is achieved through the following technical solution: a hydropower station hydrological analysis system based on big data, comprising a data bus module, a unified risk scale module, an editable structure memory module, a phase change structure signature module, a counterfactual bifurcation deduction module, a constrained feasible region synthesis module, and a closed-loop reinjection module. The data bus module collects and aligns hydrological observation data and scheduling condition data to form an operational window data stream, and outputs rainfall forecast data and upstream inflow scenario data. The closed-loop reinjection module forms historical operational window samples and writes them into the calibration sample library of the unified risk scale module. The unified risk scale module calculates the current time-based hydrological data based on the calibration sample library. A risk scale parameter is used to determine the phase transition trigger threshold and the unified scale parameter for contraction at the same scale as the constraint boundary. An editable structural memory module performs channel attenuation, directional erasure, and structural writing on the structural memory state based on the unified risk scale parameter, and outputs hydrological latent variables. A phase transition structural signature module constructs a structural signature based on the hydrological latent variables, calculates the intensity of structural changes, and determines the phase transition trigger threshold according to the unified risk scale parameter to generate phase transition trigger results and output phase transition migration instructions. A counterfactual bifurcation inference module uses the current structural memory state as the initial bifurcation state, receives rainfall forecast data and upstream inflow scenario data, and executes candidate scheduling action sequences. The structure memory recursion module, consistent with the editable structure memory module, generates future hydrological trajectories. It determines the phase transition trigger threshold within the trajectory based on unified risk scale parameters and outputs a phase transition migration command when the threshold is met. This command simultaneously rewrites the directional erase intensity and structure writing intensity in the structure memory recursion to complete the stage migration within the bifurcation. The constraint feasible region synthesis module calculates the constraint default degree for the future hydrological trajectory and determines it as the predicted default degree. It then performs same-scale contraction on the constraint boundary based on unified risk scale parameters to generate a policy family that satisfies the contraction boundary. When the policy family is not empty, it outputs the policy family's internal structure. The selected scheduling action sequence is used to output the set of unsatisfiable constraints and the triggering reasons associated with the unified risk scale parameters and phase transition triggering results when the policy family is empty. The closed-loop feedback module is used to collect the real response after executing the scheduling action sequence output by the constraint feasible domain synthesis module, calculate the real default degree peak, and write the real default degree peak into the calibration sample library of the unified risk scale module to update the unified risk scale parameters at subsequent time points. Based on the deviation direction between the predicted default degree and the real default degree peak, the module generates a correction instruction and feeds it back to the editable structure memory module to adjust the directional erasure intensity, thereby forming a closed loop of phase transition self-transfer and policy family generation driven by the unified risk scale.

[0007] The data bus module includes a timing alignment unit, a missing data processing unit, and a feature loading unit. The timing alignment unit is used to align the hydrological observation data and the scheduling operation data with timestamps according to a unified sampling interval to form a running window data stream. The missing data processing unit is used to perform consistency completion on the running window data stream and output a missing data identifier. The feature loading unit is used to write the missing data identifier and the running window data stream together into the input interface of the editable structure memory module, so that the editable structure memory module can explicitly absorb the uncertainty caused by missing data when updating the structure memory state.

[0008] The unified risk scale module includes a calibration sample library, a risk upper bound estimation unit, and a margin solving unit. The calibration sample library stores historical operating window samples, and each historical operating window sample contains the true default rate peak. The risk upper bound estimation unit sorts the true default rate peaks in the calibration sample library and aggregates them with random weights to generate a risk upper bound estimation distribution. The margin solving unit solves for the unified risk scale parameters under the constraints of the target risk threshold and confidence level.

[0009] The calibration sample library stores preset operating window samples when the system is first run, and updates historical operating window samples using a sliding window method during system operation to ensure that uniform risk scale parameters can be calculated and are adaptively updated according to seasonal changes and scheduling conditions.

[0010] The editable structure memory module includes a structure memory matrix group, an attenuation gating generation unit, an erase gating generation unit, a write gating generation unit, and a readout unit. The structure memory matrix group is used to store a preset number of structure memory states, and each structure memory state is in matrix form. The attenuation gating generation unit is used to generate channel attenuation gating, the erase gating generation unit is used to generate directional erase gating, the write gating generation unit is used to generate structure write gating, and the readout unit is used to output latent water conditions variables based on the structure memory matrix group.

[0011] The erase gate generation unit generates an erase key based on the erase key generation path and generates a directional erase gate accordingly. The write gate generation unit generates a write key based on the write key generation path and generates a structure write gate accordingly. The erase key generation path and the write key generation path are independent of each other so that the directional erase gate and the structure write gate can be independently adjusted under the same input conditions.

[0012] The phase change structure signature module includes a delayed embedding unit, a structure signature construction unit, and a trigger determination unit. The delayed embedding unit is used to construct a time embedding sequence corresponding to a preset scale set based on the latent hydrological variables. The structure signature construction unit is used to calculate the structure signature on the time embedding sequence and obtain the structural change intensity. The trigger determination unit is used to determine the phase change trigger threshold based on the unified risk scale parameter, and compare the structural change intensity with the phase change trigger threshold to generate a phase change trigger result and output a phase change migration command.

[0013] The counterfactual bifurcation deduction module includes a bifurcation generation unit, an isomorphic recursion unit, and an intra-trajectory migration unit. The bifurcation generation unit generates candidate scheduling action sequences and copies the current structural memory state as the initial bifurcation state of each candidate scheduling action sequence. The isomorphic recursion unit, based on the initial bifurcation state and receiving rainfall forecast data and upstream water inflow scenario data, performs structural memory recursion, consistent with the editable structural memory module, on each candidate scheduling action sequence to generate a future hydrological trajectory. The intra-trajectory migration unit determines the intra-trajectory phase transition trigger threshold based on a unified risk scale parameter on the future hydrological trajectory, and outputs an intra-trajectory phase transition migration command when the intra-trajectory phase transition trigger threshold is met. The intra-trajectory phase transition migration command simultaneously rewrites the directional erasure intensity and structural writing intensity of the isomorphic recursion unit.

[0014] The constraint feasible region synthesis module includes a default degree calculation unit, a boundary contraction unit, a strategy family generation unit, a strategy selection unit, and a failure interpretation unit. The default degree calculation unit calculates the default degree of water level constraint, ecological discharge constraint, downstream water level change rate constraint, gate opening and closing rate constraint, and unit start-up and shutdown constraint, and outputs the maximum default degree as the constraint default degree. The boundary contraction unit performs same-scale contraction on the boundaries of the above five types of constraints according to the unified risk scale parameter to form a contraction boundary. The strategy family generation unit selects candidate scheduling action sequences that meet the contraction boundary to form a strategy family. The strategy selection unit selects the scheduling action sequence with the minimum constraint default degree within the strategy family as the scheduling action sequence. The failure interpretation unit outputs the set of unsatisfiable constraints and the triggering reasons associated with the unified risk scale parameter and phase change triggering results when the strategy family is empty.

[0015] The control unit of the hydropower station hydrological analysis system based on big data is configured to execute the hydrological analysis method, which includes steps S1 to S6. Step S1 involves the data bus module collecting and aligning hydrological observation data and scheduling condition data to form an operating window data stream, and outputting rainfall forecast data and upstream inflow scenario data. Step S2 involves the closed-loop reinjection module providing historical operating window samples, and the unified risk scale module calculating unified risk scale parameters. Step S3 involves the editable structure memory module performing channel attenuation, directional erasure, and structure writing on the structure memory state based on the unified risk scale parameters, and outputting hydrological latent variables. Step S4 involves the phase change structure signature module constructing a structure signature based on the hydrological latent variables, and determining the phase change trigger threshold according to the unified risk scale parameters to generate a phase change trigger result and output the phase change migration. The instructions are as follows: Step S5 involves the counterfactual bifurcation inference module copying the current structural memory state, receiving rainfall forecast data and upstream water inflow scenario data, and performing structural memory recursion on candidate scheduling action sequences to generate future hydrological trajectories. Within the trajectory, a phase transition trigger threshold is determined based on unified risk scale parameters, and a phase transition migration instruction is output when the phase transition trigger threshold is met, thereby simultaneously rewriting the directional erasure intensity and structural writing intensity in the structural memory recursion. Step S6 involves the constraint feasible region synthesis module performing same-scale contraction on the constraint boundary based on unified risk scale parameters to generate a policy family and output a scheduling action sequence. After executing the scheduling action sequence, the closed-loop feedback module feeds back the actual default rate peak value to update the unified risk scale parameters at subsequent times and feeds back correction instructions to the editable structural memory module to complete the closed-loop optimization.

[0016] The beneficial effects of this invention are: The unified risk scaling module generates unified risk scaling parameters using the peak value of the actual default rate in the calibration sample library as the sole benchmark. This parameter is also used as a unified scale parameter for determining the phase transition trigger threshold and the contraction of the constraint boundary at the same scale. This ensures that the phase transition structure signature module's trigger determination of the structural change intensity and the constraint feasible region synthesis module's contraction of the constraint boundary are coordinated and consistent on the same risk scale. This avoids the scale separation problem caused by setting phase transition determination and constraint contraction on different scales, which results in phase transition being triggered but the feasible region is still too wide, or the feasible region being contracted but the phase transition not being identified. As a result, the strategy family generation and failure interpretation output have stable scale meaning and auditable consistency.

[0017] The editable structure memory module performs channel attenuation, directional erasure, and structure writing on the structure memory state under the unified risk scale parameters. It also explicitly suppresses the writing channel through missing measurement markers, so that the uncertainty of missing measurement in the running window data stream is structurally absorbed and reflected in the latent hydrological variables. This makes the structure signature constructed by the phase change structure signature module based on the latent hydrological variables more representative of the phase changes of the real hydrological state rather than the pseudo changes caused by missing measurement points. This improves the stability of the phase change triggering results, reduces the probability of false triggering caused by missing measurement completion errors, and keeps the input state of subsequent counterfactual bifurcation inference and constraint feasible domain synthesis reliable.

[0018] The phase change structural signature module constructs a time-embedded sequence on a preset scale set and forms a structural signature. It uses the intensity of structural change and the phase change trigger threshold determined by a unified risk scale parameter to generate phase change triggering results and phase change migration instructions. This makes the phase switching not dependent on fixed water level thresholds or fixed empirical segments, but directly driven by the structural changes of latent hydrological variables. Furthermore, the phase change migration instructions further act on the trajectory migration of the counterfactual bifurcation inference module, prompting the predicted trajectory to synchronously update the directional erasure intensity and structural writing intensity during phase switching. This ensures that the future hydrological trajectory maintains the consistency of structural memory recursion during cross-phase inference, reducing the accumulation of prediction bias caused by phase drift.

[0019] The counterfactual bifurcation inference module uses the current structural memory state as the initial bifurcation state. It performs structural memory recursion on candidate scheduling action sequences in the same way as the editable structural memory module to generate future hydrological trajectories. Within the trajectory, it triggers phase transition migration and gating strength rewriting based on unified risk scale parameters. This makes the comparison of candidate scheduling action sequences no longer an offline simulation of a static model, but an isomorphic inference of the dynamic coupling process of scheduling action, hydrological state, and stage migration under the same structural memory mechanism. As a result, the predicted default degree obtained by the constraint feasible region synthesis module can simultaneously reflect the impact of scheduling action sequences on water level, discharge and start / stop constraints, as well as the impact on stage migration position, thereby improving the targeting of strategy family screening and scheduling action sequence selection.

[0020] The constraint feasible domain synthesis module performs same-scale contraction on the boundaries of water level constraints, ecological discharge constraints, downstream water level change rate constraints, gate opening and closing rate constraints, and unit start-up and shutdown constraints under the action of unified risk scale parameters. It calculates the constraint default degree based on future hydrological trajectories and determines it as the predicted default degree, thereby forming a strategy family that satisfies the contraction boundary and outputting the selected scheduling action sequence within the strategy family. This ensures that the output scheduling action sequence has a controllable default margin under risk scale constraints. Furthermore, when the strategy family is empty, it outputs the set of unsatisfiable constraints and the triggering reasons associated with the unified risk scale parameters and phase transition triggering results. This allows infeasible states to be located to specific constraint categories and risk scale levels, realizing a closed-loop operation mode of outputting actions when feasible and reasons when infeasible, reducing repeated trial and error by dispatchers under abnormal hydrological conditions.

[0021] After the execution of the scheduling action sequence, the closed-loop feedback module collects the actual response and calculates the peak value of the actual default degree. It writes the peak value of the actual default degree into the calibration sample library to update the unified risk scale parameters for subsequent time periods. It also generates correction instructions based on the deviation direction between the predicted default degree and the actual default degree peak value and feeds them back to the editable structure memory module to adjust the directional erasure intensity. This enables the two key links of risk scale and structure memory to obtain online calibration capabilities simultaneously. The former ensures that the scale parameters are updated adaptively with seasonal and operating condition changes, while the latter ensures that the suppression of misjudged structure residence by structure memory gradually increases in long-term operation. This allows phase transition triggering, trajectory inference and strategy output to maintain stable convergence in continuous operation. Attached Figure Description

[0022] Figure 1 This is a flowchart of the present invention; Figure 2 The system architecture of this invention Figure 1 ; Figure 3 The system architecture of this invention Figure 2 . Detailed Implementation

[0023] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1 like Figure 1As shown, in this embodiment, the hydropower station hydrological analysis system based on big data is deployed on the station-side computing platform or the centralized control center computing platform. The computing platform is connected to the station's measurement and monitoring system. The system consists of a data bus module, a unified risk scale module, an editable structure memory module, a phase change structure signature module, a counterfactual bifurcation inference module, a constraint feasible domain synthesis module, and a closed-loop reinjection module. Each module is connected in series through data frames indexed at the same time. The data frames at any given time contain hydrological observation data, scheduling condition data, rainfall forecast data, and upstream water inflow scenario data, thereby ensuring that the data processing process is continuous and traceable in the time dimension.

[0025] In terms of data sources, hydrological observation data is output from water level measurement devices, flow measurement devices, rainfall measurement devices, and downstream control section water level measurement devices. Scheduling operation data is output from gate opening measurement devices, gate opening and closing status quantities, unit output measurement devices, unit start and stop status quantities, reservoir operating water level target values, and ecological discharge target values. Rainfall forecast data is provided by the meteorological forecast interface and covers the prediction time domain with preset forecast lead time. Upstream inflow scenario data is formed by fusing measured inflow from upstream hydrological stations with the basin forecast model and forming scenario sets with scenario numbers. The scenario sets provide inflow sequence under different scenarios within the same prediction time domain, enabling the counterfactual bifurcation inference module to provide consistent exogenous driving input to candidate scheduling action sequences under the same time index.

[0026] When the system is running, the control unit organizes the data flow and computation flow according to the following steps, without any breaks in the chain between the steps.

[0027] S1, performed by the data bus module, performs timing alignment, missing data processing, and feature loading to form a running window data stream. Specifically, it aligns the hydrological observation data and scheduling data with timestamps according to a unified sampling interval to obtain the time index. Aligned data frames ;right Perform consistency completion on the missing test fields and generate missing test flags. Missing test mark and It has a one-to-one field structure, and the missing data flag field is used to indicate whether the corresponding observation field is missing; , Rainfall forecast data and upstream water inflow scenario data are loaded together as input tensors. and within a continuous time period Defined as a running window data stream, where The window length is preset in the system parameters.

[0028] S2 is provided with historical operating window samples by the closed-loop feedback module and written into the calibration sample library of the unified risk scaling module. Subsequently, the unified risk scaling module calculates the unified risk scaling parameters for the current moment. Specifically, after each execution of the scheduling action sequence, the closed-loop feedback module extracts the peak value of the actual constraint default from the actual response and forms historical operating window samples. The sample is written to the calibration sample library; the unified risk calibration module reads the sample set from the calibration sample library. And extract the corresponding set of peak values ​​of actual constraint default. The peak value of the actual constraint default degree is calculated using the following formula. In the formula, For the first The peak value of the true constraint default rate for a sample of historical operating windows; , , , , These represent the degree of violation of water level constraints, ecological discharge constraints, downstream water level change rate constraints, gate opening and closing rate constraints, and unit start-up and shutdown constraints, respectively. To ensure the auditable and recalculationable nature of the calibration sample library, the five types of actual constraint violation degrees are shown in the sample. The calculation method used within the calculation is consistent with the actual default rate and maintains the same convention as the definition of predicted default rate in step S6, that is, using a non-negative quantity to represent the default magnitude and satisfying zero within the boundary and positive when exceeding the limit: Among them, the water level constraint default rate is obtained by comparing the actual reservoir water level with the allowable upper limit boundary; the ecological discharge constraint default rate is obtained by comparing the actual discharge flow with the ecological discharge lower limit boundary; the downstream water level change rate constraint default rate is obtained by comparing the actual downstream control section water level change rate with the allowable upper limit boundary; the gate opening and closing rate constraint default rate is obtained by comparing the actual gate opening rate change with the allowable upper limit boundary; the unit start-up and shutdown constraint default rate is obtained by comparing the actual unit start-up and shutdown frequency or start-up and shutdown interval with the allowable conditions; and the peak value is taken within the time range of the historical operation window as... , , , , .Will Sorted from smallest to largest At the confidence level parameter Determine the serial number in, For serial number, For the sample size, This represents the confidence level parameter. Based on this, a unified risk scale parameter is obtained. ;in, To standardize risk scaling parameters. Among them, As a unified scale parameter, it is used simultaneously for determining the phase transition trigger threshold and for the same-scale shrinkage of the constraint boundary. The unified risk scale parameter changes dynamically as the sample library is updated.

[0029] S3 uses an editable structure memory module to perform channel attenuation, directional erasure, and structure writing on the structure memory state based on unified risk scale parameters, and outputs the latent hydrological variables. Specifically, the editable structure memory module maintains the structure memory state. The structural memory state is used to carry the coupled structure of hydrological observation data and scheduling condition data within the operating window; at time... Upon arrival, the input tensor will be... Input into the editable structure memory module, based on unified risk scale parameters Generate attenuation coefficient Erasure intensity With write strength ,in, The attenuation coefficient is... For the erasure intensity, For write strength, the three satisfy the following condition: The mapping relationship changes monotonically and is written into the system parameter configuration; to facilitate implementation and ensure parameter reproducibility, the mapping relationship can be configured as a linear saturation form, for example... , , ,in , , As the baseline parameter, , , The slope coefficient, , , This is the upper limit parameter; when performing a structural memory state update, first... Channel attenuation is performed to obtain Then based on the missing test mark Apply a suppression factor to the write channel corresponding to the missing test field to prevent the padded value from being written to the structure memory state with high confidence, and then... Perform targeted erasure and perform structure writing to obtain and from Reading Hidden Variables of Water Conditions As a latent variable in the water condition output, among which, It is a vector of latent water conditions, used to uniformly represent the dynamic characteristics, lag effects, and scheduling functions under the current operational phase.

[0030] S4, through the phase change structure signature module, constructs a structure signature based on latent hydrological variables and calculates the intensity of structural changes. It then determines the phase change trigger threshold based on unified risk scale parameters to generate a phase change trigger result and output a phase change migration command. Specifically, the phase change structure signature module processes the sequence of latent hydrological variables within the running window. Perform delayed embedding to construct structural signatures ,in, The embedding window length is preset in the system parameters; the intensity of structural variation is defined by the adjacent differences in the structural signature. in, For structural change intensity, for The structural signature of time, It is the Euclidean norm. The phase transition trigger threshold is determined based on a unified risk scaling parameter. ; in, For low threshold system parameters, The system parameter is the slope coefficient. To standardize risk scale parameters.

[0031] S5 uses the counterfactual bifurcation deduction module to copy the current structural memory state as the initial bifurcation state and receives rainfall forecast data and upstream water inflow scenario data. It performs structural memory recursion on candidate scheduling action sequences to generate future hydrological trajectories. Within the trajectory, it determines the phase transition trigger threshold based on unified risk scale parameters and outputs a phase transition migration command when the threshold is met, simultaneously rewriting the directional erase intensity and structural write intensity in the recursion. Specifically: the counterfactual bifurcation deduction module at time... Generate a set of candidate scheduling action sequences Each candidate scheduling action sequence covers the prediction time domain and consists of the gate opening trajectory, unit output trajectory, and unit start-up and shutdown trajectory; for each candidate scheduling action sequence Copy the current structure memory state As the initial state of the bifurcation ,in, For candidate scheduling action sequence index, For the first The initial state of the bifurcation of the candidate scheduling action sequence is determined, and under the same input of rainfall forecast data and the same upstream inflow scenario data, structural memory recursion consistent with the editable structural memory module is executed to obtain the future hydrological trajectory. ,in, To predict the length of the time domain, For the first Candidate fork in the prediction step The hydrological trajectory at the location includes at least the reservoir water level, outflow, downstream control section water level, and the response of gate opening and generating unit output; during the recursive process, the structural signature within the trajectory is continuously calculated. Intensity of structural changes within the trajectory And based on unified risk scale parameters, a phase transition trigger threshold within the trajectory is formed. ,when Output phase transition command within the trajectory ,in, This refers to the phase transition migration command within the trajectory, and the directional erasure intensity of the subsequent recursive steps is determined by the phase transition migration command within the trajectory. With structural writing intensity Simultaneous rewriting during execution ensures that the structural memory recursion after the stage migration within the fork is consistent with the coupling law of the new stage, thereby avoiding trajectory drift caused by continuously writing the old stage structure.

[0032] S6, through the constrained feasible region synthesis module, performs same-scale contraction on the constraint boundary based on unified risk scale parameters to generate a family of strategies and outputs a sequence of scheduling actions. The closed-loop feedback module then feeds back the actual default rate peak after executing the scheduling action sequence to update the unified risk scale parameters for subsequent time periods and feeds back correction instructions to the editable structured memory module to complete the closed-loop optimization. Specifically, for each future hydrological trajectory... Five types of constraint violation degrees are calculated, and the maximum value is taken as the constraint violation degree. These constraint violation degrees are also determined as predicted violation degrees. Specifically, the water level constraint violation degree is defined as the predicted extent to which the reservoir water level exceeds the upper limit of the allowable range; the ecological discharge constraint violation degree is defined as the predicted extent to which the discharge flow is lower than the lower limit of the ecological discharge range; the downstream water level change rate constraint violation degree is defined as the predicted extent to which the downstream control section water level change rate exceeds the upper limit of the allowable range; the gate opening and closing rate constraint violation degree is defined as the predicted extent to which the gate opening rate change exceeds the upper limit of the allowable range; and the unit start-up and shutdown constraint violation degree is defined as the predicted extent to which the unit start-up and shutdown frequency or start-up and shutdown interval violates the allowable conditions. The original boundaries of the five types of constraints are then standardized using a unified risk scale parameter. Performing same-scale contraction yields the contraction boundary, and based on this, candidate scheduling action sequences that satisfy the contraction boundary are selected to form a policy family. ,in For a set of strategy families, when When not empty, the scheduling action sequence with the minimum predicted default degree is selected within the policy family as the output scheduling action sequence and sent to the execution system. When empty, the set of unsatisfiable constraints is output along with the triggering reason associated with the unified risk scale parameter and phase transition triggering result for dispatcher review; the closed-loop backfeed module collects the actual response and calculates the peak value of the actual default degree after the execution of the scheduling action sequence. ,Will The calibration sample library written into the unified risk scale module is used for updating the unified risk scale parameters at subsequent time points, and is also based on the predicted default rate. Peak of actual default rate The deviation direction forms a correction signal in, As a correction signal, This represents the peak value of the actual default rate. To predict the degree of default, the correction signal is mapped as a correction instruction and fed back to the editable structure memory module. This is used to adjust the value of the directional erasure intensity in the next running window, so that the suppression of misjudged structures is enhanced in subsequent windows. This achieves self-calibration and self-correction of the phase transition self-migration and strategy family generation closed loop driven by a unified risk scale during continuous operation.

[0033] Example 2 like Figure 1 and Figure 2 As shown, in this embodiment, based on the system architecture and data processing link of Embodiment 1, the unified risk scale module is refined and enhanced, so that the unified risk scale parameters have a calculation link that can be cold-started, rolled updated, and audited and recalculated, and form a stable data closed loop with the closed-loop feedback module. The data input and output relationships, data flow organization methods of the running window, and generation methods of hydrological latent variables, structural signatures, future hydrological trajectories, strategy families, and scheduling action sequences of the other modules are kept consistent. Only the internal structure of the unified risk scale module, the sample organization of the calibration sample library, the risk upper bound estimation method, and the margin solution method are supplemented and explained to ensure that those skilled in the art can directly implement it.

[0034] The unified risk scale module in this embodiment includes a calibration sample library, a risk upper bound estimation unit, and a margin solution unit. The calibration sample library stores historical operating window samples and records the actual default rate peak and the corresponding window start and end timestamps. The historical operating window samples are generated by the closed-loop recharge module after each scheduling action sequence is executed and written into the calibration sample library. The actual default rate peak is calculated using the same caliber to ensure auditability and recalculation. This same caliber includes consistent calculation methods for water level constraint default, ecological release constraint default, downstream water level change rate constraint default, gate opening / closing rate constraint default, and unit start / stop constraint default, with the maximum value taken as the actual default rate peak. Specifically, within a historical operating window, the five types of constraint default rates are calculated hourly for the actual response sequence, and the peak value within the window is taken to obtain the actual default rate peak. The actual default rate peak is denoted as... ,in For sample index, and satisfying ; in, For the first The peak value of the true default rate for a sample of historical operating windows; For the sample The corresponding set of timestamps; Indicates the timestamp The degree of violation of water level constraints is calculated from the actual reservoir water level and the allowable boundary. This represents the degree of violation of ecological discharge constraints calculated from the actual discharge flow and the ecological discharge boundary. This represents the degree of violation of the downstream water level change rate constraint calculated from the actual downstream control section water level change rate and the allowable boundary. This represents the failure rate of the gate opening and closing rate constraint calculated from the actual gate opening rate change rate and the allowable boundary. This represents the default rate of unit start-up and shutdown constraints calculated from the actual unit start-up and shutdown status and allowable conditions. All five types of default rates use non-negative quantities to represent the default magnitude, satisfying the unified definition of zero default and positive exceedance, thus ensuring comparability between different samples.

[0035] To achieve a cold start for the system's initial operation, the calibration sample library stores preset operating window samples during the system's first run. These preset operating window samples are derived from historical monitoring archives and historical dispatch condition archives of the hydropower station. The process of forming the preset operating window samples is as follows: a time period covering the dry season, normal water season, and wet season is selected from the historical archives; the actual hydrological observation data and actual dispatch condition data within the time period are aligned with a unified sampling interval to form an operating window data stream; the actual default value peak of each window is calculated based on the above five types of constraint default degree calibrators and written into the calibration sample library. This ensures that the calibration sample library has an initial sample set with no less than the preset number of samples when the system goes online, avoiding the problem of phase change trigger threshold and constraint boundary contraction lacking a unified scale due to the inability to calculate unified risk scale parameters.

[0036] To achieve adaptive updates based on seasonal changes and scheduling conditions, the calibration sample library uses a sliding window approach to update historical samples during system operation. Specifically, the calibration sample library maintains a maximum capacity (N) and a sample retention strategy. When the closed-loop backfeed module writes new samples... If the number of samples has not reached the maximum capacity, they are directly added and written. If the number of samples has reached the maximum capacity, the sample with the earliest window end timestamp is removed first and a new sample is written to ensure that the sample set always reflects the risk distribution characteristics of the most recent operating period. The maximum capacity (N) is preset in the system parameters and can be set according to the seasonal cycle length of the hydropower station operation to cover the number of samples covering multiple complete scheduling cycles.

[0037] The risk upper bound estimation unit is used to sort the peak values ​​of the true default rates in the calibration sample library and aggregate them with random weights to generate a risk upper bound estimation distribution. Specifically, the risk upper bound estimation unit reads the current valid sample set from the calibration sample library. And sort in ascending order to obtain At the same time, a random weight vector is generated. And satisfy and ,in, For a random weight vector, For the first Each weight; to ensure the random weight aggregation process is auditable and recalculated, the random weight vector is generated by a pseudo-random number generator and adopts a recalculated random weight generation rule, which includes at least a random seed and a normalization method: in the first... A random seed is determined during the generation of the next candidate. The random seed Consists of the calibration sample library version identifier, refresh time index, and candidate generation sequence number. The random seed is jointly determined and recorded in the audit log; it is generated based on the random seed. nonnegative intermediate quantities and according to Normalization yields the random weight vector, thus enabling the repeated generation of a consistent weight vector given a sample set and a random seed. The random weight vector is used to randomize and weight the contributions of different samples to form a discrete representation of the risk upper bound estimation distribution. The risk upper bound estimation unit constructs a weighted quantile candidate set based on the ranking results and random weights. The upper bound of each candidate risk is determined by the position where it crosses the threshold according to the weighted cumulative weight, and the index of the upper bound of the candidate risk satisfies ; in, The index for the upper bound of the candidate risk; Indicates and The corresponding weights; This is the confidence level parameter. This leads to the candidate values ​​for the upper bound of risk. By applying random weight vectors corresponding to different random seeds... The second candidate generation yields a set of candidate values. This results in discrete samples of the risk upper bound estimation distribution. The candidate value set is used to support the audit recalculation of the unified risk scale parameters, that is, given the calibration sample library samples, confidence level parameters and random weight generation rules, a consistent risk upper bound estimation process can be obtained repeatedly.

[0038] The margin solving unit is used to solve for a unified risk scale parameter under the constraints of the target risk threshold and confidence level. It then outputs this unified risk scale parameter as a unified scale parameter for determining the phase transition trigger threshold and shrinking the constraint boundary at the same scale. Specifically, the margin solving unit receives discrete samples of the risk upper bound estimation distribution. A conservative upper bound strategy is adopted to select a unified risk scale parameter, wherein the maximum value of the candidate value set is taken as the unified risk scale parameter. ; in, To standardize risk scale parameters, This serves as a set of candidate values. This ensures that the unified risk scale parameter has a conservative coverage capability for sample fluctuations at the same confidence level. The unified risk scale parameter can be recalculated at every moment or at every operating window boundary and output to the phase transition structure signature module and the constraint feasible region synthesis module, enabling the phase transition trigger threshold and constraint boundary contraction to be linked at the same scale. The output process is as follows: the phase transition structure signature module receives the unified risk scale parameter and determines the phase transition trigger threshold accordingly; the constraint feasible region synthesis module receives the unified risk scale parameter and performs constraint boundary contraction at the same scale accordingly.

[0039] In this embodiment, the output of the unified risk scale parameter is used not only for determining the phase transition trigger threshold and shrinking the constraint boundary on the same scale, but also for unifying the deviation correction judgment scale of the closed-loop refeeding module. Specifically, the closed-loop refeeding module obtains the true default rate peak after executing the scheduling action sequence output by the constraint feasible region synthesis module. And obtain the predicted default degree from the constrained feasible region synthesis module. The predicted default degree is determined by the constrained default degree calculated from the future hydrological trajectory, and the closed-loop recharge module calculates the deviation direction. in, The direction of deviation This represents the peak value of the actual default rate. To predict the degree of default, the deviation direction is mapped to a correction command and fed back into the editable structure memory module to adjust the directional erasure intensity of the next running window. The generation and feeding back process of the correction command has been given in Example 1 and will not be repeated in this example. It is only emphasized that since the unified risk scale parameter is formed by the true default value peak of the calibration sample library and can be updated on a rolling basis, the deviation direction maintains a consistent scale under different seasons and different operating conditions. This makes the adjustment of the directional erasure intensity by the correction command have a stable amplitude meaning and avoids over-correction or under-correction when the sample distribution drifts.

[0040] Through the above structure and process, the calibration sample library realizes the cold start of the pre-set running window samples when the system goes online and the sliding window update during the operation. The risk upper bound estimation unit realizes the sorting of the peak value of the true default degree and the random weight aggregation to form the risk upper bound estimation distribution. The margin solution unit solves the unified risk scale parameter under the constraints of the target risk threshold and confidence level and outputs the unified scale parameter as the determination of the phase transition trigger threshold and the contraction of the constraint boundary at the same scale. Thus, the system can complete phase transition detection, bifurcation inference, feasible region contraction, strategy output and closed-loop backfeeding update at a unified scale during continuous operation. The data processing link remains consistent and uninterrupted from the formation of historical samples to the output of unified risk scale parameters and the use of module linkage.

[0041] Example 3 like Figures 1 to 3 As shown, in this embodiment, based on the closed-loop hydrological analysis system of Embodiment 1 and the unified risk scale module calculation link of Embodiment 2, the editable structured memory module, phase change structured signature module, counterfactual bifurcation deduction module, and constraint feasible domain synthesis module are further structurally enhanced. This makes phase change migration, memory deletion and writing, and bifurcation deduction form a strongly coupled closed loop in the same information flow. Moreover, the screening and selection of candidate scheduling action sequences are directly dependent on the stage migration results of the structured memory recursion, thereby ensuring that the system can still provide consistent and auditable scheduling action sequence output or failure explanation output under high uncertainty hydrological conditions.

[0042] The data source in this embodiment is the same as in embodiment 1. Hydrological observation data, scheduling data, rainfall forecast data, and upstream water inflow scenario data are fed into the editable structure memory module via the data bus module to form a running window data stream. At the same time, the closed-loop recharge module writes the actual default rate peak into the calibration sample library of the unified risk scale module to update the unified risk scale parameters. After the unified risk scale parameters are updated at each running window boundary or within a preset refresh cycle, they are output to the phase transition structure signature module, the counterfactual bifurcation inference module, and the constraint feasible region synthesis module, respectively, as unified scale parameters for determining the phase transition trigger threshold and the contraction of the constraint boundary at the same scale. This embodiment focuses on describing the implementation details of gating generation, structure signature construction, isomorphic recursion and trajectory migration, strategy family selection, and failure interpretation that were not elaborated in embodiments 1 and 2.

[0043] The editable structure memory module includes a structure memory matrix group, an attenuation gating generation unit, an erase gating generation unit, a write gating generation unit, and a readout unit. The structure memory matrix group is used to store a preset number of structure memory states, and each structure memory state is in matrix form. For ease of implementation, the preset number corresponds to the number of multi-head structure memory state heads, denoted as . , No. Head structure memory state is recorded as ,in For multi-head structure memory state head number, For the header index, For the state dimension, For memory slot dimensions, For the first Head structure memory state, at any time The structural memory state can be represented as The readout unit is used to map the structured memory matrix group to obtain the latent variables of the water situation. ,in The low-dimensional representation vector is shared by the phase change structure signature module and the counterfactual bifurcation inference module; the attenuation gating generation unit generates channel attenuation gating to control the time decay of the structure memory state; the erase gating generation unit generates directional erase gating to erase the old stage structure when the stage changes; and the write gating generation unit generates structure write gating to write the coupled structure of hydrological observation data and scheduling data in the current window.

[0044] To ensure that gating for erasing and writing can simultaneously achieve both the "strong erase old stage structure" and the "strong write new stage structure" during the phase transition phase, this embodiment sets the key generation paths for erase gating and write gating to be independent of each other. Specifically, the erase gating generation unit generates the erase key according to the erase key generation path. The write gate generation unit generates a write key based on the write key generation path. ,in, For the erase key, For the write key, the erase key generation path and the write key generation path use different parameter sets, and both are input with the same risk scale parameter. The missing measurement marker and the latent water condition variable are designed to ensure that they can be adjusted independently under the same input conditions. For ease of implementation, the erase key and write key can be generated through different linear mappings and nonlinear activations, and normalization makes them usable for memory slot addressing. After the erase key generation path outputs the erase key, a directional erase gating is obtained. The write key generates a path and outputs the write key to obtain the structure write gate. ,in, For targeted erasure gating, Write gates to the structure. For the state dimension, For the memory slot dimension, attenuation gating is applied to the output channel of the generation unit. ,in, As a channel attenuation gating, the three factors work together on the structure memory matrix group to ensure that the structure memory state update satisfies the order of attenuation, then erasure, and then writing. Moreover, the erasure intensity and writing intensity are adjusted at different stages by a unified risk scale parameter and a phase transition migration command.

[0045] At any moment Upon arrival, the editable structure memory module receives the input tensor output by the data bus module. The input tensor includes the aligned data frame of the running window data stream at the current time step, as well as the missing test flag and exogenous driving summary features. The readout unit reads the structural memory state from the previous time step. Obtain the hidden variables of the water situation at the previous moment and Together, the gating input vector required for the current update is generated. To explicitly absorb the uncertainty of missing tests, the missing test flag is used to apply a suppression coefficient to the write gating during the gating generation process, so that the write gating corresponding to the missing test field remains low in activation at this moment. The structure memory state update adopts the following core update form, and the following formula is used as the core formula and is used to disclose the implementation path: in, For the first The head's structural memory state The number of heads in the multi-head structure memory state; Element-wise multiplication; For the reason and The mapping obtained by the first Header candidate write content matrix; channel attenuation gating Controlling historical structure decay and targeted erasure gating Control the erasure of the old stage structure and the gating of structure writing. Control the writing of new structures; readout units The execution reads out the hidden variables of the water situation. The water condition hidden variables serve as the input to the subsequent phase transition structure signature module and the shared state description recursively derived by the counterfactual bifurcation deduction module.

[0046] The phase transition structure signature module includes a delayed embedding unit, a structure signature construction unit, and a trigger determination unit. The delayed embedding unit constructs a time embedding sequence corresponding to a preset scale set based on latent hydrological variables. The preset scale set is denoted as... ,in For each scale, the embedding scale length is specified. Delayed embedding unit from Constructing the temporal embedding matrix The structural signature construction unit is for each Calculate the structural signature vector And the overall structural signature is obtained by splicing them together. ,in, For a pre-defined scale set, For the first Embedding scale length, For the first A time embedding matrix, For the first A structural signature vector, For the overall structural signature, to ensure simplicity and stability, the structural signature can be a combination of second-order statistics and related statistics of the embedding matrix; subsequently, the structural signature construction unit calculates the structural change intensity. As a measure of the difference between structural signatures at adjacent time points, the core calculation formula is: in, For structural change intensity, for The overall structure signature at any given moment for The overall structure signature at any given moment It is a Euclidean norm. The triggering decision unit is based on a unified risk scale parameter. Determine the phase transition trigger threshold The threshold is determined using a linear threshold mapping that changes in the same direction as the uniform scale, and its core expression is: ;in, For low threshold system parameters, The system parameter is the slope coefficient. To standardize risk scaling parameters. When The phase transition trigger result is generated and the phase transition migration command is output. The phase transition migration instruction includes a migration activation identifier and a migration intensity factor. The migration intensity factor is used to rewrite the directional erase intensity and the structure write intensity in the migration unit within the trajectory of the counterfactual bifurcation deduction module, so that the bifurcation recursion can maintain self-consistency across stage switching.

[0047] The counterfactual bifurcation deduction module includes a bifurcation generation unit, an isomorphic recursive unit, and a trajectory intra-transition unit. The bifurcation generation unit at time... Generate candidate scheduling action sequences and copy the current structure memory state as the initial bifurcation state of each candidate scheduling action sequence. The copying method is to copy the structure memory matrix group head-to-head one by one. ,in The candidate scheduling action sequence is indexed; the isomorphic recursive unit performs structural memory recursion on each candidate scheduling action sequence under the same rainfall forecast data and the same upstream inflow scenario data to generate future hydrological trajectories. The structural memory recursion is consistent with the structural memory state update form of the editable structural memory module, the difference being that the input tensor is replaced by the candidate scheduling action sequence to replace the currently executed action and uses predicted exogenous driving data; specifically, in the prediction step... The above recursion yields Corresponding latent water conditions And output future water situation trajectory .

[0048] The in-trajectory migration unit is used to determine the in-trajectory phase transition trigger threshold based on a unified risk scale parameter on the future hydrological trajectory. When the in-trajectory phase transition trigger threshold is met, it outputs an in-trajectory phase transition migration command and simultaneously rewrites the directional erase intensity and structural write intensity of the isomorphic recursive unit using the in-trajectory phase transition migration command. Specifically, for each candidate bifurcation... In each prediction step Up by Constructing structural signatures within the trajectory And calculate the intensity of structural changes within the trajectory. ; in, For the first Candidate fork in the prediction step The intensity of structural changes within the trajectory at that point. Sign the structure within the trajectory. Determine the phase transition trigger threshold within the trajectory using a unified risk scale parameter. And it remains consistent with the determined online threshold, when Time-based generation of phase transition migration instructions within the trajectory This is mapped to a gating strength rewriting factor, so that the structural memory recursion from this prediction step continues to be recursively recursively according to the gating strength after the stage transition. The rewriting method is to simultaneously rewrite the strength parameters used to generate directional erase gating and structural write gating inside the isomorphic recursion unit and keep the erase key generation path and write key generation path independent of each other, thereby ensuring that the migration within the trajectory does not destroy the gating independence and avoids structural memory write lag after the stage switch.

[0049] The constraint feasible region synthesis module includes a default degree calculation unit, a boundary contraction unit, a strategy family generation unit, a strategy selection unit, and a failure interpretation unit. The default degree calculation unit calculates the default degree of water level constraints, ecological release constraints, downstream water level change rate constraints, gate opening / closing rate constraints, and unit start-up / shutdown constraints for each future hydrological trajectory and outputs the maximum default degree as the constraint default degree. The boundary contraction unit performs same-scale contraction on the boundaries of the above five types of constraints according to a unified risk scale parameter to form a contraction boundary. The strategy family generation unit selects candidate scheduling action sequences that satisfy the contraction boundary to form a strategy family. The strategy selection unit selects the scheduling action sequence with the minimum constraint default degree within the strategy family as the output scheduling action sequence. The failure interpretation unit outputs the set of unsatisfiable constraints and the triggering reasons associated with the unified risk scale parameter and phase transition triggering results when the strategy family is empty. To ensure that strategy selection is reproducible and outputs a single value, the strategy selection unit adopts a minimization criterion, the core expression of which is: ; in, Index the selected candidate scheduling action sequence. For the set of policy family indexes, For the first The predicted default rate corresponding to each candidate scheduling action sequence is given by the output of the scheduling action sequence. .when When empty, the failure explanation unit outputs the unsatisfiable constraint set by checking the minimum reachable default degree under the contraction boundary for each of the five types of constraints and adding the unsatisfiable constraints to the set. At the same time, it outputs the trigger reason field, which includes the unified risk scale parameter level and the phase transition trigger result status, so that the scheduler can distinguish between the different trigger types of "risk scale contraction causing the feasible domain to be empty" and "intra-trajectory stage migration causing the predicted trajectory to cross the constraint boundary".

[0050] In this embodiment, in addition to generating historical operation window samples and updating unified risk scale parameters, the closed-loop refeedback module also performs a closed-loop function for error correction under gate independence. Specifically, it collects the actual response and calculates the peak value of the actual default degree after the execution of the scheduling action sequence. And obtain the predicted default degree from the constrained feasible region synthesis module. The closed-loop recharge module calculates the deviation direction. ; in, The direction of deviation This represents the peak value of the actual default rate. To predict the degree of default, the deviation direction is mapped as a correction command and fed back to the editable structure memory module. This allows the erase gating generation unit to adjust the strength parameters of the erase key generation path within the subsequent operation window without changing the strength parameters of the write key generation path. This enhances the suppression of misjudged structure residency while maintaining the independence of erase gating and write gating, ensuring the stable convergence of the closed-loop link of "phase change trigger threshold - gating deletion / writing - isomorphic bifurcation - shrinking feasible region - strategy selection - feedback correction" during long-term operation.

[0051] Through the above implementation, this embodiment, based on the complete closed-loop process of embodiment 1 and the calculable and roll-upable unified risk scale parameters of embodiment 2, discloses the following implementation methods: the structure memory matrix group and gating generation method of the editable structure memory module; the independent setting method of the erase key generation path and the write key generation path; the multi-scale delay embedding and structure change intensity calculation method of the phase change structure signature module; the isomorphic recursion and trajectory migration gating intensity rewriting method of the counterfactual bifurcation inference module; the strategy family generation and minimization selection method of the constraint feasible domain synthesis module; and the failure explanation output method. This enables those skilled in the art to directly build and run the system described in this embodiment, and the data processing process, from the input of the data bus module to the output of the scheduling action sequence and then to the closed-loop backfeeding update, remains continuous and uninterrupted.

[0052] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A hydropower station hydrological analysis system based on big data, characterized in that, The system includes a data bus module, a unified risk scale module, an editable structure memory module, a phase transition structure signature module, a counterfactual bifurcation deduction module, a constraint feasible region synthesis module, and a closed-loop recharge module. The data bus module collects and aligns hydrological observation data and scheduling data to form an operational window data stream, and outputs rainfall forecast data and upstream inflow scenario data. The closed-loop recharge module generates historical operational window samples and writes them into the calibration sample library of the unified risk scale module. The unified risk scale module calculates the unified risk scale parameters for the current moment based on the calibration sample library and outputs unified scale parameters for determining the phase transition trigger threshold and the contraction of constraint boundaries at the same scale. The editable structural memory module is used to perform channel attenuation, directional erasure, and structural writing on the structural memory state based on the unified risk scale parameters, and outputs hydrological latent variables. The phase change structural signature module is used to construct a structural signature based on the hydrological latent variables, calculate the structural change intensity, and determine the phase change trigger threshold according to the unified risk scale parameters to generate a phase change trigger result and output a phase change migration instruction. The counterfactual bifurcation deduction module is used to take the current structural memory state as the initial bifurcation state, and receive the rainfall forecast data and the upstream inflow scenario data, and perform structural memory recursion consistent with the editable structural memory module on the candidate scheduling action sequence to generate a future bifurcation result. The system calculates the incoming water condition trajectory and determines the phase transition trigger threshold within the trajectory based on the unified risk scale parameters. When the phase transition trigger threshold is met, a phase transition migration command within the trajectory is output. This command simultaneously rewrites the directional erase intensity and structural write intensity in the structural memory recursion to complete the stage migration within the bifurcation. The constraint feasible region synthesis module calculates the constraint default degree for the future water condition trajectory and determines the constraint default degree as the predicted default degree. Based on the unified risk scale parameters, it performs same-scale contraction on the constraint boundary to generate a policy family that satisfies the contraction boundary. When the policy family is not empty, it outputs the selected scheduling action sequence within the policy family. When the policy family is empty, the output is the set of unsatisfiable constraints and the triggering reasons associated with the unified risk scale parameters and the phase transition triggering results. The closed-loop feedback module is used to collect the real response after executing the scheduling action sequence output by the constraint feasible domain synthesis module, calculate the real default degree peak, and write the real default degree peak into the calibration sample library of the unified risk scale module to update the unified risk scale parameters at subsequent times. Based on the deviation direction between the predicted default degree and the real default degree peak, a correction instruction is generated and fed back to the editable structure memory module to adjust the directional erasure intensity, thereby forming a closed loop of phase transition self-transfer and policy family generation driven by the unified risk scale.

2. The hydropower station hydrological analysis system based on big data as described in claim 1, characterized in that, The data bus module includes a timing alignment unit, a missing data processing unit, and a feature loading unit. The timing alignment unit is used to align the hydrological observation data and the scheduling condition data with timestamps at a uniform sampling interval to form the running window data stream. The missing data processing unit is used to perform consistency completion on the running window data stream and output a missing data identifier. The feature loading unit is used to write the missing data identifier and the running window data stream together into the input interface of the editable structure memory module, so that the editable structure memory module can explicitly absorb the uncertainty caused by missing data when updating the structure memory state.

3. The hydropower station hydrological analysis system based on big data as described in claim 1, characterized in that, The unified risk scale module includes a calibration sample library, a risk upper bound estimation unit, and a margin solving unit. The calibration sample library is used to store the historical operating window samples, and each historical operating window sample contains the true default rate peak. The risk upper bound estimation unit is used to sort the true default rate peaks in the calibration sample library and aggregate them with random weights to generate a risk upper bound estimation distribution. The margin solving unit is used to solve for the unified risk scale parameters under the constraints of the target risk threshold and confidence level.

4. The hydropower station hydrological analysis system based on big data as described in claim 3, characterized in that, The calibration sample library stores preset operating window samples during the first system run and updates the historical operating window samples using a sliding window method during system operation to ensure that the unified risk scale parameter can be calculated and is adaptively updated with seasonal changes and scheduling conditions.

5. The hydropower station hydrological analysis system based on big data as described in claim 1, characterized in that, The editable structure memory module includes a structure memory matrix group, an attenuation gating generation unit, an erase gating generation unit, a write gating generation unit, and a readout unit. The structure memory matrix group is used to store a preset number of structure memory states, and each structure memory state is in matrix form. The attenuation gating generation unit is used to generate channel attenuation gating. The erase gating generation unit is used to generate directional erase gating. The write gating generation unit is used to generate structure write gating. The readout unit is used to output the latent hydrological variables based on the structure memory matrix group.

6. The hydropower station hydrological analysis system based on big data as described in claim 5, characterized in that, The erase gating generation unit generates an erase key based on the erase key generation path, and generates the directional erase gating accordingly. The write gating generation unit generates a write key based on the write key generation path, and generates the structure write gating accordingly. The erase key generation path and the write key generation path are independent of each other, so that the directional erase gating and the structure write gating can be independently adjusted under the same input conditions.

7. The hydropower station hydrological analysis system based on big data as described in claim 1, characterized in that, The phase transition structure signature module includes a delayed embedding unit, a structure signature construction unit, and a trigger determination unit. The delayed embedding unit is used to construct a time embedding sequence corresponding to a preset scale set based on the latent hydrological variables. The structure signature construction unit is used to calculate the structure signature on the time embedding sequence and obtain the structure change intensity. The trigger determination unit is used to determine the phase transition trigger threshold based on the unified risk scale parameter, and compare the structure change intensity with the phase transition trigger threshold to generate the phase transition trigger result and output the phase transition migration instruction.

8. The hydropower station hydrological analysis system based on big data as described in claim 1, characterized in that, The counterfactual bifurcation deduction module includes a bifurcation generation unit, an isomorphic recursion unit, and an intra-trajectory migration unit. The bifurcation generation unit generates the candidate scheduling action sequence and copies the current structural memory state as the initial bifurcation state of each candidate scheduling action sequence. The isomorphic recursion unit, based on the initial bifurcation state and receiving the rainfall forecast data and the upstream water inflow scenario data, performs structural memory recursion, consistent with the editable structural memory module, on each candidate scheduling action sequence to generate the future hydrological trajectory. The intra-trajectory migration unit determines the intra-trajectory phase transition trigger threshold based on the unified risk scale parameter on the future hydrological trajectory, and outputs the intra-trajectory phase transition migration command when the intra-trajectory phase transition trigger threshold is met. The intra-trajectory phase transition migration command is used to simultaneously rewrite the directional erasure intensity and structural writing intensity of the isomorphic recursion unit.

9. The hydropower station hydrological analysis system based on big data as described in claim 1, characterized in that, The constraint feasible domain synthesis module includes a default degree calculation unit, a boundary contraction unit, a strategy family generation unit, a strategy selection unit, and a failure interpretation unit. The default degree calculation unit is used to calculate the default degree of water level constraint, ecological discharge constraint, downstream water level change rate constraint, gate opening and closing rate constraint, and unit start-up and shutdown constraint, and outputs the maximum default degree as the constraint default degree. The boundary contraction unit is used to perform same-scale contraction on the boundaries of the above five types of constraints according to the unified risk scale parameter to form a contraction boundary. The strategy family generation unit is used to screen candidate scheduling action sequences that satisfy the contraction boundary to form the strategy family. The strategy selection unit is used to select the scheduling action sequence with the minimum constraint default degree in the strategy family as the scheduling action sequence. The failure interpretation unit is used to output the set of unsatisfiable constraints and the triggering reasons associated with the unified risk scale parameter and the phase transition triggering result when the strategy family is empty.

10. The hydropower station hydrological analysis system based on big data as described in claim 1, characterized in that, The control unit of the big data-based hydropower station hydrological analysis system is configured to execute a hydrological analysis method, which includes steps S1 to S6. Step S1 involves the data bus module collecting and aligning the hydrological observation data and the scheduling condition data to form the operating window data stream, and outputting the rainfall forecast data and the upstream inflow scenario data. Step S2 involves the closed-loop reinjection module providing the historical operating window samples, and the unified risk scale module calculating the unified risk scale parameters. Step S3 involves the editable structure memory module performing channel attenuation, directional erasure, and structure writing on the structure memory state based on the unified risk scale parameters, and outputting the hydrological latent variables. Step S4 involves the phase change structure signature module constructing the structure signature based on the hydrological latent variables, and determining the phase change trigger threshold according to the unified risk scale parameters to generate the phase change trigger result and output the phase change. The migration instruction, step S5, involves the counterfactual bifurcation inference module copying the current structural memory state, receiving the rainfall forecast data and the upstream water inflow scenario data, and performing structural memory recursion on the candidate scheduling action sequence to generate the future hydrological trajectory. Within the trajectory, the phase transition trigger threshold is determined based on the unified risk scale parameter, and when the phase transition trigger threshold is met, a phase transition migration instruction is output to simultaneously rewrite the directional erasure intensity and the structural writing intensity in the structural memory recursion. Step S6 involves the constraint feasible region synthesis module performing same-scale contraction on the constraint boundary based on the unified risk scale parameter to generate the policy family and outputting the scheduling action sequence. After executing the scheduling action sequence, the closed-loop feedback module feeds back the actual default rate peak value to update the unified risk scale parameter at subsequent times and feeds back the correction instruction to the editable structural memory module to complete the closed-loop optimization.

Images