An adaptive flood forecasting method, a terminal and a storage medium

Abstract

The application discloses a self-adaptive flood forecasting method, a terminal and a storage medium. The method comprises the following steps: generating a high signal-to-noise ratio local hydrological real-time data stream through adaptive sampling and a double-channel differential quantization technology; receiving an upstream input data stream, performing flood evolution calculation based on a viscoelastic dynamics model and fusing a dynamically identified input delay to obtain an original prediction value; for model parameters, a confidence ellipse is constructed based on ridge regression, the residual of real-time data and the prediction value is compared through statistical projection, the measurement noise and the river physical drift are intelligently distinguished, and the parameters are updated smoothly; finally, the original prediction value is corrected by solving a convex optimization problem subject to safety constraints, and a physically credible final prediction value is output. The corresponding terminal comprises corresponding modules for realizing the above steps. The application realizes high-precision, self-adaptive and intrinsically safe flood forecasting.

Description

Technical Field

[0001] This application relates to the fields of hydrological monitoring, water conservancy engineering and disaster prevention and mitigation, specifically to an adaptive flood forecasting method, terminal and storage medium. Background Technology

[0002] The watershed flood forecasting terminal is a core device for flood control decision-making. Based on real-time water level and rainfall data, it uses built-in hydrological or hydrodynamic models to predict future flood events. Existing terminals are usually pre-installed with fixed models (such as one-dimensional Saint-Venant equation solvers), and their operation is highly dependent on a set of pre-calibrated physical parameters, such as river cross-sectional geometry and riverbed roughness.

[0003] However, the physical environment of river channels exhibits significant time-varying characteristics. Through flood erosion or siltation, the morphology and surface roughness of the riverbed undergo slow but irreversible changes (physical drift). A key problem with existing technologies is that when the terminal detects a deviation between real-time data and model predictions, automated algorithms cannot effectively distinguish whether this deviation is caused by instantaneous random noise from sensor measurements (non-physical fluctuations) or by substantial changes in the physical properties of the riverbed (such as water level rise due to siltation).

[0004] If the system corrects for all deviations, the model parameters will oscillate wildly in a short period due to tracking measurement noise, resulting in "overfitting" and causing prediction failure. If the system does not react to deviations, the model will gradually lose accuracy due to parameter aging. Currently, the industry relies on professional teams to conduct "re-measurement and calibration" on-site periodically, which greatly increases maintenance costs and cannot meet real-time requirements.

[0005] Therefore, there is an urgent need for an adaptive flood forecasting technology and terminal that can automatically and accurately distinguish between noise and drift, and maintain high accuracy and stability during long-term unattended operation. Summary of the Invention

[0006] This application aims to overcome the shortcomings of the prior art and provide an adaptive flood forecasting method, terminal and storage medium to solve the technical problem that existing flood forecasting terminals cannot automatically identify whether the model prediction deviation is caused by instantaneous measurement noise or drift of river physical parameters during long-term operation, thus making it impossible to perform accurate and stable adaptive calibration without relying on manual on-site calibration.

[0007] In a first aspect, this application provides an adaptive flood forecasting method, applied to a flood forecasting terminal, the method comprising the following steps: S1. Through adaptive sampling and dual-channel differential quantization technology, analog signals from at least one sensor are continuously processed to generate a local hydrological real-time data stream with a high signal-to-noise ratio. S2. Receive upstream input data stream, construct a river channel evolution model based on viscoelastic dynamics to simulate the flow lag effect, and dynamically identify the arrival delay of the upstream input data stream; use the delay-calibrated upstream input data stream and the local hydrological real-time data stream to perform flood evolution calculations and obtain the original predicted value. S3. For the parameters of the river evolution model, a statistical confidence ellipse is constructed based on ridge regression to define a reasonable range of values; by statistically projecting and comparing the residuals of the local hydrological real-time data stream and the original predicted values, measurement noise and river physical drift are distinguished, and the parameters are smoothly updated after the drift is confirmed. S4. The original predicted value is corrected by solving a convex optimization problem with safety constraints to output the final predicted value.

[0008] Step S1 includes: performing dual-channel processing on the analog signal; the first channel performs calculations using a preset voltage threshold as the modulus and performs high-resolution analog-to-digital conversion on the remainder to obtain a remainder value; the second channel counts the number of integer cycles in which the signal exceeds the voltage threshold to obtain an index value; simultaneously, dynamically determining the sampling time based on the changing trend of the hydrological physical quantity; performing an arithmetic combination of the obtained remainder value and the index value to reconstruct the digital value of the hydrological physical quantity; outputting the continuously reconstructed digital value as a local real-time hydrological data stream, and performing real-time verification and filtering based on sensor range and physical limit rules.

[0009] Step S2 includes: establishing a deterministic state-space equation with water level and flow velocity as state variables, and introducing a generalized rheological unit model to dynamically calculate channel friction resistance and reproduce the water level-flow hysteresis relationship; based on pre-stored historical upstream input data sequences and local historical hydrological response data sequences, applying Krylov subspace projection technology to dynamically identify the effective delay time of upstream water reaching the current section; using the effective delay time to time-align the current upstream input data stream, and inputting the aligned data stream and the local real-time hydrological data stream into the state-space equation to solve for the original predicted value.

[0010] Step S3 includes: processing local historical hydrological data sequences using the ridge regression algorithm, calculating baseline estimates of model parameters, and generating a parameter confidence ellipse covering a preset probability based on sensor noise distribution; calculating the residual between the current value of the local real-time hydrological data stream and the original predicted value, mapping the residual to the model parameter space, and calculating its Mahalanobis distance from the baseline estimate; if the distance falls outside the confidence ellipse for multiple consecutive periods and is in the same direction, it is determined that physical drift of the river channel has occurred; otherwise, it is determined to be random noise; when it is determined to be physical drift, the ridge regression calculation is re-executed based on the new data stream within the sliding time window, and a smoothing constraint term for the parameter mutation amplitude is introduced to obtain an updated parameter set, and then the confidence ellipse is regenerated.

[0011] Step S4 includes: setting a discrete-time high-order safety constraint function including the warning water level, the historical highest water level, and the maximum allowable fluctuation rate; at each calculation step, using a first-order Taylor expansion to linearize the nonlinear hydrodynamic equation at the current trajectory point, and calculating the tangent plane of the safety constraint function, thereby transforming the nonlinear constraint set into a linear half-space constraint set, forming a convex feasible region; constructing a quadratic objective function that simultaneously minimizes the deviation of the output trajectory from the original predicted value and the degree of abrupt change in the trajectory itself; subsequently, within the convex feasible region, solving the optimization problem using a sequential quadratic programming algorithm, and obtaining the optimal solution sequence is the final predicted value after safety correction.

[0012] Secondly, this application provides an adaptive flood forecasting terminal for implementing the method described in the first aspect, the terminal comprising: The acquisition module is configured to generate local real-time hydrological data streams through adaptive sampling and dual-channel differential quantization processing; The dynamic evolution module is configured to perform evolution calculations on the upstream input data stream and the local hydrological real-time data stream based on a viscoelastic dynamic model and integrate dynamically identified input delays, and output the original predicted values. The parameter calibration module is configured to construct a confidence ellipse for the model parameters based on ridge regression, and to perform locking or smoothing updates on the parameters of the evolution module according to the position of the residual between the real-time data and the predicted value in the ellipse. The safety control module is configured to optimize and correct the original predicted value under safety constraints using a tangent projection and sequential quadratic programming algorithm, and output the final predicted value. Furthermore, the acquisition module includes a dual-channel parallel analog-to-digital conversion circuit, configured as a high-resolution residual quantization channel and a low-resolution period counting channel, respectively; the dynamic evolution module includes an embedded computing unit integrating a Krylov subspace projection coprocessor; the parameter calibration module includes a hardware acceleration unit for performing ridge regression and Mahalanobis distance calculations; and the safety control module includes a dedicated sequential quadratic programming solver.

[0013] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the adaptive flood forecasting method described in the first aspect.

[0014] The beneficial effects of this application, compared with the prior art, are that it provides an adaptive flood forecasting method, terminal, and storage medium. The method includes: generating a high signal-to-noise ratio local hydrological real-time data stream using adaptive sampling and dual-channel differential quantization techniques; receiving upstream input data streams, performing flood evolution calculations based on a viscoelastic dynamics model and incorporating dynamically identified input delays to obtain original predicted values; constructing confidence ellipses based on ridge regression for model parameters, comparing the residuals of real-time data and predicted values ​​through statistical projection, intelligently distinguishing between measurement noise and river channel physical drift, and smoothly updating the parameters; finally, correcting the original predicted values ​​by solving a safety-constrained convex optimization problem, and outputting a physically reliable final forecast value. The corresponding terminal includes modules that implement the above steps. This application achieves high-precision, adaptive, and intrinsically safe flood forecasting under unattended operation, solving problems such as easy parameter drift, large noise interference, and physically unreliable forecast results in the prior art. Attached Figure Description

[0015] To more clearly illustrate the solution of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating an adaptive flood forecasting method provided in an embodiment of this application.

[0017] Figure 2 This is a flowchart illustrating step S1 of an adaptive flood forecasting method provided in an embodiment of this application.

[0018] Figure 3 This is a flowchart illustrating step S2 of the adaptive flood forecasting method provided in an embodiment of this application.

[0019] Figure 4This is a flowchart illustrating step S3 of the adaptive flood forecasting method provided in an embodiment of this application.

[0020] Figure 5 This is a flowchart illustrating step S4 of the adaptive flood forecasting method provided in an embodiment of this application.

[0021] Figure 6 This is a schematic diagram of the modules of an adaptive flood forecasting terminal provided in an embodiment of this application.

[0022] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0023] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0024] It should be noted that if the embodiments of this application involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0025] Furthermore, if the embodiments of this application involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0026] Firstly, this application provides an adaptive flood forecasting method, the core of which lies in forming a complete technical closed loop from high-fidelity data acquisition, intelligent physical modeling, online parameter self-calibration to secure output of results. This method operates within a dedicated flood forecasting terminal. This terminal is an embedded device integrating sensing, computing, communication, and power management, typically deployed in field environments such as hydrological stations, reservoir dams, or key river sections. As the physical carrier of this application, the terminal is not limited to a specific hardware architecture or component combination; its core lies in containing technical units capable of collaboratively executing the functionalities required for the following method steps. For example, its data acquisition and scheduling functions can be accomplished by connecting sensors such as water level gauges and rain gauges, combined with a scheduler (which can be a software task, timer interrupt, or dedicated logic circuit) to implement adaptive sampling logic; its core computational functions (such as viscoelastic model solving, ridge regression, and convex optimization) can be implemented using general-purpose microprocessors, digital signal processors, field-programmable gate arrays, or customized coprocessors, sequential quadratic programming solvers, and other computational elements; its secure output is published through built-in or external display and communication interfaces. In other words, any physical device that includes functional units capable of implementing the method steps described below should be considered as a terminal protected by this application.

[0027] See Figure 1 This method includes the following steps: S1. Through adaptive sampling and dual-channel differential quantization technology, analog signals from at least one sensor are continuously processed to generate a local hydrological real-time data stream with a high signal-to-noise ratio. This step aims to solve the problem of acquiring high-precision, high-timeliness raw data by field terminals under conditions of strictly limited communication bandwidth and power consumption.

[0028] See Figure 2 Specifically, it includes the following collaborative sub-steps: S11. The analog signal is processed in a dual-channel manner. The first channel performs calculations with a preset voltage threshold as the modulus and performs high-resolution analog-to-digital conversion on the remainder to obtain the remainder value. The second channel counts the number of integer cycles in which the signal exceeds the voltage threshold to obtain the index value. At the same time, the sampling time is dynamically determined based on the changing trend of hydrological physical quantities. S12. The obtained remainder value and index value are combined arithmetically to reconstruct the numerical value of the hydrological physical quantity. S13. The continuously reconstructed digital values ​​are output as the local hydrological real-time data stream, and real-time verification and filtering are performed based on the sensor range and physical limit rules.

[0029] Specifically, firstly, the analog signal from the water level or rainfall sensor is fed into two parallel processing hardware links. In the first channel (fine channel), the signal is first modulo-calculated using a preset reference voltage value (e.g., 1 volt, corresponding to 1 meter of water level), retaining only the remainder portion exceeding an integer multiple threshold. This remainder signal is then amplified with high gain and quantized by a high-resolution analog-to-digital converter to obtain a high-precision remainder value. In the second channel (index channel), a low-precision comparator or counter accumulates the number of integer cycles in which the original signal exceeds the reference voltage, generating a low-data-volume cycle index value. These two data parts, representing the remainder for detail and the index for magnitude, are concatenated in the processor into a highly concise compressed data packet, achieving efficient data representation and transmission at the bit level.

[0030] Then, the terminal abandoned fixed-interval timed sampling. Its built-in scheduler continuously calculates the instantaneous rate of change of local hydrological physical quantities (such as water level) and compares this rate of change with a preset threshold. When the rate of change of water level exceeds the preset high threshold, the terminal determines that the information is of high value and easily expires, immediately shortens the sampling waiting time, and enters a high-frequency acquisition mode to capture transient characteristics such as flood peaks; when the rate of change is lower than the preset low threshold, the terminal determines that the information changes slowly, automatically extends the sampling interval, and puts most circuits into sleep mode, significantly reducing the terminal's power consumption.

[0031] Subsequently, upon reaching the sampling time determined by the scheduler, the terminal reads the sensor data and decompresses the data packet. The processor determines the integer period of the signal amplitude based on the received index value, and then, combining this with a high-precision remainder value, accurately reconstructs the original water level or rainfall value through arithmetic combination (i.e., physical value = index value × reference voltage + remainder value). This value is then precisely time-stamped, and based on the known range of the physical sensor that generated the signal, and the preset maximum reasonable instantaneous rate of change for the hydrological physical quantity, the reconstructed value undergoes instantaneous validity verification. Any instantaneous jumps exceeding the sensor's measurement range or physically impossible (such as a 10-meter rise in water level within 1 second) are automatically filtered out. The verified, continuously generated sequence of digital values ​​constitutes a clean and reliable local real-time hydrological data stream, providing input for subsequent steps.

[0032] S2. Receive upstream input data stream, construct a river channel evolution model based on viscoelastic dynamics to simulate the flow lag effect, and dynamically identify the arrival delay of the upstream input data stream; use the delay-calibrated upstream input data stream and the local hydrological real-time data stream to perform flood evolution calculations and obtain the original predicted value. This step is the core of intelligent computing for forecasting, designed to enable models to automatically fit complex hydrodynamic phenomena and adapt to time-varying external inputs. See also... Figure 3Specifically, it includes the following collaborative sub-steps: S21. Establish a deterministic state-space equation with water level and flow velocity as state variables, and introduce a generalized rheological unit model to dynamically calculate the river friction resistance and reproduce the water level-flow hysteresis relationship. S22. Based on the pre-stored historical upstream input data sequence and local historical hydrological response data sequence, the Krylov subspace projection technique is applied to dynamically identify the effective delay time of upstream water reaching the current section. S23. Use the effective delay time to time align the current upstream input data stream, and input the aligned data stream and the local hydrological real-time data stream into the state space equation to solve for the original predicted value.

[0033] Specifically, a deterministic state-space equation is first established as the computational framework. This equation uses the real-time water level and cross-sectional flow velocity of the river cross-section as the core state variables, and its construction strictly follows the laws of conservation of mass and momentum, ensuring the physical foundation of the model. To accurately reproduce the "loop curve" (hysteresis effect) between water level and flow rate during flood rise and fall, this step introduces a generalized rheological unit model into the equation of state. Specifically, this model employs either the generalized Maxwell model or the generalized Kelvin-Voyt model. This model equates riverbed frictional resistance to a mechanical system containing "springs" (elastic components) and "dampers" (viscous components). At each calculation step, the terminal dynamically updates the state of these internal "virtual components" based on the current flow velocity and its rate of change, thereby calculating the dynamic resistance value related to the flow history in real time. This allows the model to automatically simulate the physical phenomenon of flow rate lags behind water level changes.

[0034] To address the challenge of the nonlinear variation in the arrival time of upstream rainfall or water flow at the station (convergence time) with flood size, the terminal employs a data-driven approach to dynamically identify delays. The processor arranges pre-stored historical upstream input data sequences (such as flow data from upstream stations) and the corresponding historical hydrological response data sequences at the station (i.e., historical water level data generated by S1) in chronological order to construct a Hankel matrix. Subsequently, Krylov subspace projection techniques are applied to perform eigenmode decomposition on this matrix, extracting the dominant modes of terminal dynamics and resolving the current effective input delay time from them.

[0035] Finally, using the effective delay time identified in the above steps, the currently received upstream input data stream is aligned and corrected on the time axis. The time-aligned upstream data stream, along with the local real-time hydrological data stream generated in step S1, is then input into the pre-constructed viscoelastic state-space equations for numerical solution. The solution result is the terminal's original prediction of the future flood evolution process.

[0036] S3. For the parameters of the river evolution model, a statistical confidence ellipse is constructed based on ridge regression to define a reasonable range of values; by statistically projecting and comparing the residuals of the local hydrological real-time data stream and the original predicted values, measurement noise and river physical drift are distinguished, and the parameters are smoothly updated after the drift is confirmed. This step aims to achieve long-term, stable, and human-intervention-free core intelligent decision-making. See also... Figure 4 Specifically, it includes the following collaborative sub-steps: S31. Use the ridge regression algorithm to process the local historical hydrological data series, calculate the baseline estimate of the model parameters, and generate a parameter confidence ellipse covering the preset probability based on the sensor noise distribution. S32. Calculate the residual between the current value of the local hydrological real-time data stream and the original predicted value, map the residual to the model parameter space, and calculate its Mahalanobis distance from the baseline estimate; if the distance falls outside the confidence ellipse for multiple consecutive periods and the direction is consistent, it is determined that the channel physical drift has occurred; otherwise, it is determined to be random noise. S33. When physical drift is determined, ridge regression calculation is re-executed based on the new data stream within the sliding time window, and a smoothing constraint term for the magnitude of parameter mutation is introduced to obtain the updated parameter set, and then the confidence ellipse is regenerated.

[0037] Specifically, the ridge regression algorithm is first used to process a local historical hydrological data series. Considering the inherent multicollinearity problem in hydrological data, ridge regression introduces a regularization coefficient to ensure numerical stability. Based on this, the algorithm calculates the baseline estimates (i.e., the most probable values) of key physical parameters (such as riverbed roughness) in the current channel evolution model. Simultaneously, based on the nominal noise statistics of the sensors (usually assumed to be Gaussian distribution), a confidence ellipse centered at this baseline value is calculated in the multidimensional parameter space. This ellipse statistically defines all reasonable ranges of parameter values ​​(e.g., 95% confidence intervals) at a given noise level.

[0038] Whenever new real-time data arrives, this method does not immediately use it to revise the model. Instead, it calculates the difference (i.e., the residual) between the real-time data (the current value from the S1 data stream) and the original predicted value output by the S2 model. Using the Mahalanobis distance algorithm (a statistical distance that considers parameter covariance), this residual is mapped to the parameter space established in step 1, and the distance between its corresponding parameter solution and the baseline estimate is calculated.

[0039] Subsequently, a strict binary logic judgment is executed. If the calculated distance falls inside the confidence ellipse, the current deviation is determined to be random noise from the sensor or a minor environmental disturbance. All current model parameters are locked without any changes, thus avoiding parameter oscillations caused by tracking noise in traditional adaptive algorithms. If the distance falls outside the confidence ellipse for multiple consecutive sampling periods (e.g., 5 consecutive times) and the direction is consistent (e.g., a positive value indicates a persistently high water level), the terminal determines that the physical properties of the river channel have undergone substantial changes (e.g., siltation has occurred). At this point, parameter update permissions are unlocked.

[0040] Once an update is triggered, this method discards the oldest data frames in the historical data queue that no longer reflect the current river channel state, and reconstructs the ridge regression matrix based on the new data stream within the sliding time window. During the calculation, a penalty term based on the parameter values ​​from the previous time step is introduced as a smoothing constraint to ensure a smooth transition of new parameter values ​​to the true physical values, preventing non-physical, drastic parameter jumps. After the new parameter set is calculated, a confidence ellipse is immediately regenerated around it, and the system immediately reverts to the "parameter-locked" steady-state monitoring mode.

[0041] S4. The original predicted value is corrected by solving a convex optimization problem with safety constraints to output the final predicted value.

[0042] This step aims to provide an inherently safe barrier for flood forecasting results based on mathematical programming, ensuring that the final forecast values ​​strictly comply with physical laws and engineering safety specifications under all circumstances. See also... Figure 5 Specifically, it includes the following collaborative sub-steps: S41. Define a discrete-time high-order safety constraint function that includes the warning water level, the historical highest water level, and the maximum allowable fluctuation rate. S42. At each calculation step, the nonlinear hydrodynamic equations are linearized at the current trajectory point using a first-order Taylor expansion, and the tangent plane of the safety constraint function is calculated, thereby transforming the nonlinear constraint set into a linear half-space constraint set, forming a convex feasible region. S43. Construct a quadratic objective function that simultaneously minimizes the deviation of the output trajectory from the original predicted value and the degree of abrupt change in the trajectory itself; then, within the convex feasible region, solve the optimization problem using a sequential quadratic programming algorithm, and the resulting optimal solution sequence is the final predicted value after safety correction.

[0043] Specifically, firstly, a set of rigid safety rules determined by water conservancy engineering specifications are pre-set in the terminal and mathematically transformed into discrete-time high-order safety constraint functions. These constraints include not only instantaneous value limits (such as water level must not exceed the guaranteed water level) but also trend limits (such as the rate of water level rise in the next 3 hours must not exceed a certain limit value), thereby enabling a forward-looking safety assessment of the predicted trajectory.

[0044] At the start of each forecast calculation step, the original predicted value generated in step S2 and its subsequent extrapolated trajectory serve as the reference trajectory. First, using a first-order Taylor expansion technique, the nonlinear hydrodynamic equation of state is approximated as a time-varying linear equation at the current reference trajectory point. Next, for the equally nonlinear safety constraint function, its tangent plane at the current state point is calculated. Through this linear projection operation, the originally complex nonlinear constraint boundary is transformed into a series of linear half-space constraints. These linear constraints collectively enclose a convex feasible region, within which any solution is mathematically easy to solve efficiently and physically approximately satisfies the original safety requirements.

[0045] Finally, a quadratic objective function is constructed. This function contains two core terms: a "fidelity term," which requires the corrected output trajectory to be as close as possible to the original predicted trajectory generated in step S2; and a "smoothing term," which requires the changes in the corrected trajectory itself to be as gradual as possible, avoiding sharp angles. Subsequently, the terminal's built-in optimization solver (such as a sequential quadratic programming solver) quickly solves the constrained optimization problem within the convex feasible region constructed in step 2. The optimal solution sequence output by the solver is the final prediction value after safety filtering and smoothing. This mechanism ensures that no matter what reason the original prediction of S2 attempts to exceed the safety boundary (such as parameter transients or input anomalies), this step can forcibly and smoothly "pull it back" to the physically and engineering-allowed range, thereby guaranteeing the absolute reliability and security of the published results.

[0046] Through the close coordination of the above four steps, the method and terminal described in this application realize the end-to-end adaptive flood forecasting capability, from high-quality data perception, intelligent physical modeling, accurate parameter self-calibration to absolutely safe output.

[0047] Secondly, this application provides an adaptive flood forecasting terminal 100. The complete technical loop of this method is physically implemented in a dedicated adaptive flood forecasting terminal. As an embedded system, the hardware and software architecture of this terminal is organized around the core steps of this method, as described in [reference needed]. Figure 6 It mainly includes a data acquisition module 10, a dynamic evolution module 20, a parameter calibration module 30, and a safety control module 40. These modules work together to execute the aforementioned method.

[0048] Specifically: The acquisition module 10 is based on a dual-channel parallel analog-to-digital converter circuit, configured as a high-resolution residual quantization channel and a low-resolution period counting channel. Through the coordinated operation of these two channels, combined with its internally implemented adaptive sampling scheduling algorithm, the module performs differential quantization and reconstruction of the sensor's analog signal, ultimately generating a local real-time hydrological data stream.

[0049] The dynamic evolution module 20 is centered on an embedded computing unit that integrates a Krylov subspace projection coprocessor. Using this computing unit as its hardware platform, the module runs a dynamic delay identification algorithm and a viscoelastic dynamic evolution model, performing calculations on upstream input data streams and local real-time hydrological data streams, and outputting raw predicted values.

[0050] The parameter calibration module 30 contains a hardware acceleration unit at its core for performing ridge regression and Mahalanobis distance calculations. This module utilizes this hardware acceleration unit to efficiently construct confidence ellipses for model parameters, calculate statistical projective distances of residuals, and accordingly make decisions on parameter locking or smooth updates.

[0051] The safety control module 40 contains a dedicated sequential quadratic programming solver at its core. Using this solver as its computational engine, the module executes tangent projection and sequential quadratic programming algorithms to optimize and correct the original predicted values ​​under safety constraints, thereby outputting the final, physically reliable prediction value.

[0052] The above four modules interact with each other and synchronize commands through the terminal's internal bus, forming an efficient and reliable adaptive flood forecasting terminal.

[0053] Thirdly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned adaptive flood forecasting method.

[0054] This application not only protects the method itself, but also the specific terminal implementing the method and the computer-readable storage medium, the beneficial effects of which will not be elaborated here.

[0055] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0056] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0057] The above are merely optional embodiments of this application and do not limit the patent scope of this application. All equivalent structural transformations made based on the inventive concept of this application and the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included within the patent protection scope of this application.

Claims

1. An adaptive flood forecasting method, applied to a flood forecasting terminal, characterized in that, Includes the following steps: S1. Through adaptive sampling and dual-channel differential quantization technology, analog signals from at least one sensor are continuously processed to generate a local hydrological real-time data stream with a high signal-to-noise ratio. S2. Receive upstream input data stream, construct a river channel evolution model based on viscoelastic dynamics to simulate the flow lag effect, and dynamically identify the arrival delay of the upstream input data stream; use the delay-calibrated upstream input data stream and the local hydrological real-time data stream to perform flood evolution calculations and obtain the original predicted value. S3. For the parameters of the river evolution model, a statistical confidence ellipse is constructed based on ridge regression to define a reasonable range of values; by statistically projecting and comparing the residuals of the local hydrological real-time data stream and the original predicted values, measurement noise and river physical drift are distinguished, and the parameters are smoothly updated after the drift is confirmed. S4. The original predicted value is corrected by solving a convex optimization problem with safety constraints to output the final predicted value.

2. The method according to claim 1, characterized in that, Step S1 includes: The analog signal is processed in two channels. The first channel performs calculations with a preset voltage threshold as the modulus and performs high-resolution analog-to-digital conversion on the remainder to obtain the remainder value. The second channel counts the number of integer cycles in which the signal exceeds the voltage threshold to obtain the index value. At the same time, the sampling time is dynamically determined based on the changing trend of hydrological physical quantities. The obtained remainder value and index value are arithmetically combined to reconstruct the numerical value of the hydrological physical quantity. The continuously reconstructed digital values ​​are output as a local hydrological real-time data stream, and real-time verification and filtering are performed based on sensor range and physical limit rules.

3. The method according to claim 1, characterized in that, Step S2 includes: A deterministic state-space equation with water level and flow velocity as state variables is established, and a generalized rheological unit model is introduced into it to dynamically calculate the river friction resistance and reproduce the water level-flow hysteresis relationship. Based on the pre-stored historical upstream input data sequence and local historical hydrological response data sequence, the Krylov subspace projection technique is applied to dynamically identify the effective delay time of upstream water reaching the current section. The current upstream input data stream is time-aligned using the effective delay time, and the aligned data stream and the local hydrological real-time data stream are input into the state-space equation to solve for the original predicted value.

4. The method according to claim 1, characterized in that, Step S3 includes: The ridge regression algorithm is used to process local historical hydrological data sequences, calculate baseline estimates of model parameters, and generate parameter confidence ellipses covering preset probabilities based on sensor noise distribution. Calculate the residual between the current value of the local hydrological real-time data stream and the original predicted value, map the residual to the model parameter space, and calculate its Mahalanobis distance from the baseline estimate; if the distance falls outside the confidence ellipse for multiple consecutive periods and is in the same direction, it is determined that the river channel physical drift has occurred; otherwise, it is determined to be random noise. When physical drift is identified, ridge regression calculation is re-executed based on the new data stream within the sliding time window, and a smoothing constraint term for the magnitude of parameter mutation is introduced to obtain an updated parameter set, and then the confidence ellipse is regenerated.

5. The method according to claim 1, characterized in that, Step S4 includes: Define a discrete-time high-order safety constraint function that includes the warning water level, the historical highest water level, and the maximum allowable fluctuation rate. At each computation step, the nonlinear hydrodynamic equations are linearized at the current trajectory point using a first-order Taylor expansion, and the tangent plane of the safety constraint function is computed, thereby transforming the nonlinear constraint set into a linear half-space constraint set, forming a convex feasible region. A quadratic objective function is constructed that simultaneously minimizes the deviation of the output trajectory from the original predicted value and the degree of abrupt change in the trajectory itself. Subsequently, within the convex feasible region, the optimization problem is solved using a sequential quadratic programming algorithm, and the resulting optimal solution sequence is the final predicted value after safety correction.

6. The method according to claim 2, characterized in that, The "dynamic decision-making on sampling time based on the changing trend of hydrological physical quantities" specifically means: determining whether to shorten or extend the waiting time for the next sampling based on whether the rate of change of water level in the previous sampling period exceeds a preset threshold.

7. The method according to claim 3, characterized in that, The generalized rheological unit model is either the generalized Maxwell model or the generalized Kelvin-Voyt model.

8. An adaptive flood forecasting terminal for implementing the method of any one of claims 1-7, Its features are, include: The acquisition module is configured to generate local real-time hydrological data streams through adaptive sampling and dual-channel differential quantization processing; The dynamic evolution module is configured to perform evolution calculations on the upstream input data stream and the local hydrological real-time data stream based on a viscoelastic dynamic model and integrate dynamically identified input delays, and output the original predicted values. The parameter calibration module is configured to construct a confidence ellipse for the model parameters based on ridge regression, and to perform locking or smoothing updates on the parameters of the evolution module according to the position of the residual between the real-time data and the predicted value in the ellipse. The safety control module is configured to optimize and correct the original predicted value under safety constraints using tangent projection and sequential quadratic programming algorithms, and output the final predicted value.

9. The terminal according to claim 8, characterized in that, The acquisition module includes dual parallel analog-to-digital conversion circuits, configured as a high-resolution residual quantization channel and a low-resolution period counting channel, respectively. The dynamic evolution module includes an embedded computing unit that integrates a Krylov subspace projection coprocessor; The parameter calibration module includes a hardware acceleration unit for performing ridge regression and Mahalanobis distance calculations; The safety control module includes a dedicated sequential quadratic programming solver.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the adaptive flood forecasting method as described in any one of claims 1 to 7.

Images