Hydrological multi-element online sensing intelligent early warning method

Abstract

The present application relates to the technical field of hydrological element early warning, and specifically discloses a hydrological multi-element online sensing intelligent early warning method, which obtains a sequence of overturning time points of a tipping-bucket rain gauge at a hydrological section, synchronously collects soil moisture content, water level elevation and surface water flow rate, and forms a multi-element time sequence record with a time stamp. When the instantaneous rain intensity exceeds a critical threshold, a basic corrected rainfall is first obtained by matching a compensation coefficient according to a rain intensity loss mapping curve, and then a turbulent flow is identified by using a continuous interval fluctuation variance, thereby constructing a collaborative discrimination model for capturing rainfall dynamics and the response of the underlying surface runoff generation. The core of the model is that when the turbulence is severe, a dynamic collaborative coefficient is constructed by using the ratio of the actual runoff increment to the predicted runoff increment based on the infiltration saturation mapping, the basic corrected rainfall is compensated twice, and the true storm intensity is inverted. The model breaks through the limitation of the fixed conversion coefficient of the traditional tipping-bucket rain gauge, effectively avoids the underestimation under strong storms, and improves the fidelity and foresight.

Description

Technical Field

[0001] This invention belongs to the field of hydrological element early warning technology, and relates to an online sensing and intelligent early warning method for multiple hydrological elements. Background Technology

[0002] Tipping bucket rain gauges are widely used metering devices in field precipitation observation. Due to their simple structure, convenient deployment, and low maintenance costs, they have long been responsible for recording precipitation at the minute and hourly levels. Under normal rainfall conditions, they can well meet the basic needs of watershed monitoring, urban drainage, and weather forecasting. Meanwhile, rainfall intensity, as a core parameter characterizing short-term precipitation processes, not only affects surface runoff processes but also directly relates to flash flood warnings, urban flooding assessments, and the reliability of hydrological model inputs. Therefore, continuous observation and stable recording of rainfall processes are of great significance.

[0003] It is important to note that the metering characteristics of tipping bucket rain gauges are significantly correlated with rainfall intensity: under light to moderate rainfall conditions, the tipping action is relatively smooth, and the recorded results usually reflect changes in precipitation. However, as the intensity of heavy rainfall increases, the tipping frequency also increases. During rapid pouring, rainwater inside the bucket is prone to splashing, sticking to the sides, and short-term incomplete tipping, leading to a tendency for the metering results in high rainfall intensity ranges to be lower than the actual values. This deviation is often amplified, especially during the peak rainfall phase, potentially impacting subsequent flood peak assessments and risk level determinations. Therefore, it is evident that there are often discrepancies between the recorded results of tipping bucket rain gauges under extreme rainfall scenarios and the actual precipitation process, requiring further correction.

[0004] However, existing monitoring methods mostly rely on fixed resolution and conventional operating conditions for calibration, primarily establishing measurement rules for uniform or low-to-medium intensity rainfall scenarios, with insufficient attention paid to the nonlinear losses of tipping buckets under high-frequency tipping conditions. Current data processing typically uses fixed rainfall values ​​corresponding to a single tipping to summarize the data, lacking a combined analysis of instantaneous rainfall intensity changes, continuous tipping stability, and loss characteristics across different intensity ranges. Therefore, during sudden heavy rainfall events, intensity jumps, or sustained high rainfall, issues such as underestimating cumulative rainfall, delayed peak value identification, and distorted hourly rainfall intensity are prone to occur. These problems limit the data availability of tipping bucket rain gauges under extreme conditions and make it difficult for rainstorm assessments based on this data to fully reflect the actual precipitation process. Summary of the Invention

[0005] In view of the problems existing in the prior art, the present invention provides an online sensing and intelligent early warning method for multiple hydrological elements to solve the above-mentioned technical problems.

[0006] To achieve the above and other objectives, the technical solution adopted by the present invention is as follows: This invention provides an online sensing and intelligent early warning method for multiple hydrological elements, the method comprising: The sequence of flipping time points triggered by the tipping bucket rain gauge at the hydrological monitoring section is obtained, and the adjacent time intervals in the flipping time point sequence are extracted. The instantaneous rainfall intensity is obtained by dividing the preset water volume by the adjacent time interval. In response to the instantaneous rainfall intensity exceeding the critical threshold, a basic compensation coefficient corresponding to the instantaneous rainfall intensity is matched according to the preset rainfall intensity loss mapping curve, and the basic compensation coefficient is multiplied by the instantaneous rainfall intensity to obtain the basic corrected rainfall. Extracting multiple consecutive adjacent time intervals to calculate the variance of fluctuation; If the variance of the fluctuation is less than the preset turbulence threshold, the base correction rainfall will be set to the actual rainfall intensity to be calculated. If the variance of the fluctuation is greater than or equal to the preset turbulence threshold, the soil moisture content, water level elevation and surface water velocity within the same monitoring section are extracted. The watershed infiltration saturation is calculated based on soil moisture content, and the predicted runoff increment is output by combining the basic corrected rainfall with the preset runoff generation mapping relationship. Simultaneously, cross-sectional flow field calculations are performed based on water level elevation and surface water velocity to extract the actual runoff increment; The ratio of actual runoff increment to predicted runoff increment is calculated to construct a dynamic coordination coefficient. The base correction rainfall is multiplied by the dynamic coordination coefficient for secondary compensation to obtain the true rainfall intensity. The obtained real rainfall intensity, water level elevation, surface water flow velocity, and soil moisture content are input into the early warning discrimination matrix, and the predicted peak water level is extracted and multi-element early warning signal is generated.

[0007] As described above, the online sensing and intelligent early warning method for multiple hydrological elements provided by the present invention has at least the following beneficial effects: This invention calculates instantaneous rainfall intensity by acquiring the flipping time sequence of a tipping bucket rain gauge and extracting adjacent time intervals. When the instantaneous rainfall intensity exceeds a critical threshold, a corresponding basic compensation coefficient is matched based on a preset rainfall intensity loss mapping curve to multiply and correct the instantaneous rainfall to obtain the basic corrected rainfall. Simultaneously, this invention uses multiple consecutive flipping time intervals to calculate the fluctuation variance, thereby identifying the severity of turbulence in the rainfall process. This method, combining dynamic compensation of instantaneous rainfall intensity with time series fluctuation characteristics, ensures that the initial correction of rainstorm intensity no longer relies on fixed empirical conversion coefficients. Instead, it automatically adapts the compensation intensity based on the transient characteristics of rainfall intensity, effectively avoiding the systematic underestimation introduced by traditional tipping bucket rain gauges under high-intensity, non-steady rainfall conditions due to mechanical inertia and splash loss. Based on this, by comparing the variance of fluctuation with the preset turbulence threshold, stable rainfall and turbulent rainfall are distinguished and processed: when the variance of fluctuation is low, the base corrected rainfall is directly used as the true rainfall intensity, which greatly simplifies the calculation process and ensures the high timeliness of monitoring data during stable rainfall; while when the turbulence is significant, a deeper multi-element collaborative correction mechanism is triggered, which provides an accurate trigger criterion for subsequent runoff dynamic correction and significantly improves the robustness and adaptability of rainfall intensity forecasting under complex meteorological conditions.

[0008] This invention, when the variance exceeds the turbulence threshold, further incorporates soil moisture content, water level, and surface water velocity from the same monitoring section. Based on soil moisture content, the watershed infiltration saturation is calculated, and a predicted runoff increment is output through mapping transformation. Simultaneously, the actual runoff increment is extracted through cross-sectional flow field calculations based on water level and surface water velocity. Then, a dynamic coordination coefficient is constructed using the ratio of the actual runoff increment to the predicted runoff increment, which is used to perform secondary compensation on the baseline rainfall correction, thereby obtaining the final true rainfall intensity. By using the actual runoff increment to inversely constrain and correct the rainfall measurement values, the final rainfall intensity can dynamically reflect the impact of changes in soil moisture content in the early stages, surface saturation, and nonlinear confluence effects. Therefore, even in extreme situations where sudden, exceptionally heavy rainfall causes turbulent tipping action or abrupt changes in soil infiltration capacity, the bias in rainfall intensity retrieval can be greatly reduced, ensuring the physical consistency of the true rainfall intensity data. Furthermore, when these highly corrected real rainfall intensities, along with water level elevations, surface water flow velocities, and soil moisture content, are input into the early warning discrimination matrix, the calculated predicted peak water level can more accurately predict flood risks. The generated multi-element early warning signals also have higher timeliness and reliability, significantly reducing the risk of false alarms and delayed reports caused by inaccurate rainfall intensity estimations, and providing a solid data foundation for flash flood disaster prevention and urban waterlogging early warning. Attached Figure Description

[0009] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a schematic diagram showing the connections between the steps of the method of the present invention.

[0011] Figure 2 This is a schematic diagram illustrating the relationship between the dynamic coordination coefficient and the ratio of actual runoff increment to predicted runoff increment provided by this invention.

[0012] Figure 3 This is a schematic diagram of the predicted peak water level data for future target time points provided by the present invention. Detailed Implementation

[0013] The present invention will be further described in detail below with reference to specific embodiments. The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the inventive concept or exceed the scope defined by the claims, all of which should fall within the protection scope of the present invention.

[0014] Please see Figure 1-3 As shown, a multi-element online sensing and intelligent early warning method for hydrological data includes: The sequence of flipping time points triggered by the tipping bucket rain gauge at the hydrological monitoring section is obtained, and the adjacent time intervals in the flipping time point sequence are extracted. The instantaneous rainfall intensity is obtained by dividing the preset water volume by the adjacent time interval.

[0015] Preferably, the process involves extracting adjacent time intervals from the flipping time sequence and calculating the instantaneous rainfall intensity by dividing a preset water-flipping amount by the adjacent time interval, including: The first, second, and third time points triggered by the completion of the overturning action of the small room are continuously captured; The preceding time distance parameter is calculated by subtracting the first time point from the second time point, and the following time distance parameter is calculated by subtracting the second time point from the third time point. The transient rainfall intensity of the first stage is obtained by calculating the ratio of the preset water volume to the preceding time distance parameter, and the transient rainfall intensity of the second stage is obtained by calculating the ratio of the preset water volume to the subsequent time distance parameter. The attenuation ratio weights calculated based on time proximity are assigned to the transient rainfall intensity in the first and second phases, respectively. Perform a weighted summation calculation on the initial and subsequent transient rainfall intensities after allocating attenuation ratio weights, and output the instantaneous rainfall intensity at this moment.

[0016] Specifically, the sensor inside the rain gauge closes each time the chamber tipps over and touches the bottom. The edge computing gateway captures this level transition signal and records the time point using its internal clock. In one specific embodiment, during rainfall, the times of the three most recent consecutive flipping events are retained and sequentially marked as the first time point in chronological order. Second time point And the latest third time point The purpose of obtaining these three consecutive nodes is to construct two time intervals that are connected end to end, so as not only to measure the rainfall value, but also to perceive the transient evolution trend of the rainfall.

[0017] Based on the three acquired time points, the system first calculates the interval between two adjacent actions. (Previous time distance parameter) It is obtained by subtracting the first time point from the second time point; similarly, it is followed by the time distance parameter. The result is obtained by subtracting the second time point from the third time point. Subsequently, the system calls the preset water volume parameters in the rain gauge firmware. This parameter represents the equivalent physical water depth of a single pour in the equipment's chamber, and is typically preset according to the factory specifications. Combining the aforementioned time intervals, the system calculates the transient precipitation intensity for both intervals. The formula for calculating the transient rainfall intensity in the initial stage is: The constant 60 is used to convert the rate (in seconds) into rainfall per minute as measured by hydrological monitoring standards; following the same calculation logic, the formula for calculating the transient rainfall intensity in the latter part is: .

[0018] Since rainfall is a highly fluctuating natural physical process, the rainfall state closer to the current moment better represents the actual environmental conditions. Simply averaging two transient rainfall intensities would mask the rainfall peaks caused by sudden, severe convective weather. Therefore, this embodiment modifies the formula based on the exponentially weighted moving average theory in classical time series analysis, changing the attenuation from the original model based on the number of events to attenuation based on the actual physical time span. The advantage of this modification is that it effectively eliminates statistical distortion caused by differences in the tipping frequency under different rainfall intensities. In the specific weight allocation logic, the system first determines the distance between the midpoint of the two tipping action intervals and the most recently occurred third time point. The time span. The first part of the time span. The span of the later period Next, the attenuation ratio weight corresponding to the transient rainfall intensity in the later stage is calculated using the natural exponential decay function. The specific formula is as follows: ; in, The preset time-memory smoothing constant represents the system's rate of forgetting historical data. The preset logic for this parameter aims to find a balance between anti-interference filtering and sensitive response; a range of 60 to 120 seconds is recommended. If the value is too large, the system will over-rely on old data, causing the warning system to become less sensitive to sudden thunderstorm peaks; if the value is too small, it loses its smoothing effect, making the calculation results highly susceptible to large fluctuations due to single mechanical errors in the rain gauge. To balance the timeliness of the warning and the stability of the data, a preferred setting of 60 seconds is recommended. (The text abruptly ends here.) Then, according to the normalization principle of weights, the weight of the attenuation ratio corresponding to the transient rainfall intensity in the preceding period is... .

[0019] Finally, the system uses the calculated weighting coefficients to weight and merge the two rainfall states. The transient rainfall intensity of the preceding period is then considered. Multiply by its weight In addition to the transient rainfall intensity in the later period Its weight The product of these factors is summed and the instantaneous rainfall intensity is output. .

[0020] In response to an instantaneous rainfall intensity exceeding a critical threshold, a basic compensation coefficient corresponding to the instantaneous rainfall intensity is matched based on a preset rainfall intensity loss mapping curve, and the basic compensation coefficient is multiplied by the instantaneous rainfall intensity to obtain the basic corrected rainfall.

[0021] Preferably, the method also involves an adaptive iterative replacement of the cumulative precipitation mapping curve with a preset rainfall intensity loss curve, which includes: The total precipitation is theoretically estimated by integrating and accumulating the actual rainfall intensity based on discrete time slices. Obtain the cumulative precipitation generated based on independent physical mechanisms within the same time slice; Calculate the relative deviation rate between the cumulative precipitation and the theoretically projected total precipitation; In response to the relative deviation rate exceeding the confidence threshold in multiple consecutive discrete time slices, the extracted relative deviation rate sequence is input into the Kalman filter equation to filter and update the baseline offset parameter. The updated baseline offset parameter is used to calculate the offset compensation of the tangent slope of the current preset rainfall intensity loss mapping curve in the corresponding rainfall intensity interval, and the iteratively updated preset rainfall intensity loss mapping curve is output.

[0022] In this embodiment of the invention, when the equipment determines that the instantaneous rainfall intensity exceeds a preset critical threshold during precipitation monitoring, it indicates that the tipping bucket mechanism has generated significant splashing due to high-frequency tipping. At this time, the system will use the factory-preset rainfall intensity loss mapping curve to find the basic compensation coefficient corresponding to the current instantaneous rainfall intensity, and multiply this basic compensation coefficient by the instantaneous rainfall intensity to replenish the water lost through splashing, thus obtaining the basic corrected rainfall. It is recommended that the critical threshold be set within an adjustable range of 1.5 mm / min to 3.0 mm / min, which corresponds to the natural boundary between heavy rain and torrential rain. To prevent misjudging normal equipment mechanical lag as splashing loss, a setting of 2.0 mm / min is preferred.

[0023] However, after long-term operation in harsh field environments, wear on the tipping bucket shaft and sediment buildup on the inner wall of the funnel alter the original hydrodynamic characteristics of the equipment, causing a systematic deviation from the static preset curve calibrated in the laboratory environment. Therefore, this embodiment introduces an adaptive iterative replacement mechanism based on macroscopic cumulative precipitation. The system first divides the continuous monitoring process into fixed discrete time slices, multiplies the generated actual rainfall intensity by the corresponding sampling time step, where the sampling time step is the time interval between two consecutive data acquisitions at a preset sampling frequency. Within each discrete time slice, all product results are summed to complete the integral accumulation over time, thus obtaining the theoretically projected total precipitation for that period.

[0024] The length of discrete time slices is recommended to be set to a range of 1 to 6 hours. To avoid the slight water level fluctuations being amplified as errors due to the short time, and to capture the curve drift characteristics during the rainstorm process in time, it is recommended to use 2 hours as the periodic slice for accumulation calculation.

[0025] At the end of each time slice, the system reads the cumulative precipitation from a weighing rain gauge deployed at the same monitoring station. The weighing rain gauge directly weighs rainwater using a high-precision mechanical sensor at its bottom. Its physical mechanism does not rely on water flow, therefore it is largely unaffected by splashing and mechanical hysteresis during heavy rainfall, making it a reliable macroscopic comparison benchmark. After acquiring these two cumulative data points, the system calculates the difference between the cumulative precipitation obtained from the weighing rain gauge and the theoretically extrapolated total precipitation, and divides this difference by the cumulative precipitation obtained from the weighing rain gauge to obtain the relative deviation rate. If the relative deviation rate calculated by the system exceeds a preset confidence threshold in multiple consecutive discrete time slices, it indicates that the error of the tipping bucket rain gauge has exceeded the random disturbance caused by environmental wind. The number of consecutive discrete time slices is typically set to 3 to 5, preferably 3; and the preset confidence threshold is determined based on reasonable mechanical noise under steady-state precipitation over the years at the station, and is recommended to be set to 5% to 8%, preferably 5%.

[0026] The system then extracts multiple consecutive relative deviation rate values ​​that exceed the limit to form an observation sequence, which is then input into the first-order Kalman filter equation for state estimation. The Kalman filter calculates the gain coefficient using the prior estimate from the previous time step and the observation sequence of the current slice. Specifically, the error covariance of the previous state and the system process noise are introduced to calculate the prior error covariance of the current time step. Specifically, the system uses the built-in state transition matrix to perform matrix multiplication mapping and update the error covariance of the previous state. Then, the updated result is added to the covariance matrix representing the system process noise to obtain the prior error covariance of the current time step. Since the estimation object in this embodiment is a one-dimensional scalar, the state transition matrix is ​​precisely preset as a one-dimensional constant matrix with a value of 1, based on the prior law of slow physical wear of equipment at adjacent time steps. The covariance matrix of the system process noise is a preset empirical value generated based on historical data noise floor calibration, specifically configured to take values ​​within a certain range. to A fixed numerical constant within a certain range; the error covariance of the previous state is assigned a preset initial tolerance value when the system is first started, and in subsequent runs, it is directly used as the recursive retention value that is output and cached at the end of the previous calculation cycle.

[0027] Next, the prior error covariance is divided by its sum with the measurement noise variance to obtain the current Kalman gain coefficient. The measurement noise variance is specifically calculated by extracting historical samples of relative deviation rates from uniform and stable precipitation periods over a long historical time series at the station, calculating the statistical variance of these historical samples, and fixing it as the measurement noise variance. After calculating the gain coefficient, the system multiplies the optimal posterior estimate output from the previous calculation cycle with the state transition matrix (value 1) to directly inherit the current prior estimate. Subsequently, the relative deviation rate of the current slice (i.e., the observed value) is subtracted from the current prior estimate, multiplied by the Kalman gain coefficient, and then compensated to the prior estimate to obtain the current optimal posterior estimate. Through the above formula fusion, which is recursively applied according to the discrete time step, the measurement noise variance parameter is used to filter out occasional measurement noise caused by gusts blowing through the funnel, extracting a smooth trend reflecting the actual physical wear of the equipment. The system outputs the finally converged optimal posterior estimate as the baseline offset parameter.

[0028] After obtaining the baseline offset parameter, it needs to be applied back to the original preset curve. Because the aggravation of physical splashing is mainly concentrated in the high-rainfall-intensity frequency band, simply performing global translation or subtraction on the curve would lead to severe distortion of the compensation data in the low-rainfall-intensity range. Therefore, this embodiment uses a modified local tangent slope offset method to update and reshape the curve. The system calculates the average instantaneous rainfall intensity within the current time slice and uses it as the reshaping center point. The modified calculation formula for curve updating is: ; In the formula, f(R) represents the dynamic compensation coefficient after iterative replacement; f(R) represents the original preset compensation coefficient. R is the baseline offset parameter output by the Kalman filter equation; R is the instantaneous rainfall intensity of the current input. To calculate the average instantaneous rainfall intensity; The broadening factor is used to reshape the preset curve. The further the curve deviates from the average rainfall intensity zone within the slice, the smaller the correction should be to ensure a smooth transition of the physical curve and prevent abrupt distortion points. The broadening factor is... To control the rate at which the correction effect decays to both sides, it is recommended to set it to an adjustable range of 0.1 to 0.5. To prevent excessive slope distortion that could lead to divergence in later calculations, a value of 0.2 is preferred. Through this calculation process, the system outputs an iteratively updated rainfall intensity loss mapping curve, which is then used to overwrite the old curve for future rainfall events.

[0029] Extract multiple consecutive adjacent time intervals to calculate the fluctuation variance; if the fluctuation variance is less than the preset turbulence threshold, set the base correction rainfall to the actual rainfall intensity to be calculated; if the fluctuation variance is greater than or equal to the preset turbulence threshold, extract the soil moisture content, water level elevation, and surface water velocity within the same monitoring section.

[0030] Preferably, the variance of fluctuation is calculated by extracting multiple consecutive adjacent time intervals, including: Extract the time interval observation window by backtracking a preset number of sequentially adjacent time intervals from the flipped time point sequence; The arithmetic mean of all elements within the observation window is accumulated over a time interval. Calculate the sum of squares of the differences between each element and the arithmetic mean within the observation window over the time interval; The initial expected dispersion of observations is obtained by dividing the sum of squares by the degree of freedom parameter. The adjacent time intervals with the largest values ​​within the time interval observation window are selected as the maximum time element, and the adjacent time intervals with the smallest values ​​are selected as the minimum time element. The difference between the two is calculated to obtain the local range data. The local range data is multiplied by the expected dispersion of the initial observations and then amplified by the endpoint mutation factor to output the fluctuation variance representing the hydrological variation disturbance.

[0031] In this embodiment of the invention, severe convective rainfall in nature is often accompanied by strong gusts and uneven distribution of raindrops. When such hydrological variability disturbances fall into the equipment funnel, they cause the time interval between two consecutive tipping actions to fluctuate erratically. To quantitatively measure the severity of such extreme fluctuations, the system extracts multiple consecutive adjacent time intervals to calculate the variance of the fluctuations, which includes historical memory. In specific operation, the system first traces back from the most recent to the oldest time intervals in the historical cached tipping time sequence, extracting a preset number of sequentially adjacent time intervals according to the chronological order of occurrence, thus forming the time interval observation window for the current moment. Here, the preset number refers to the number of interval samples entering the observation window, and it is recommended to set it in the capacity range of 5 to 15. If this number is too large, the observation window will span a long time span, and the past stable rainfall characteristics will dilute and weaken the current climate change signal; if it is too small, there will be insufficient statistical sample support, and the objective laws of data dispersion cannot be reflected. Considering both the hydrological computational load and the sensitivity to capture abrupt changes, a preset number of 10 is recommended. After the fixed-length observation window is constructed, the system sums up all time interval elements within the window one by one, and divides the sum by the total number of samples included to obtain the arithmetic mean reflecting the frequency baseline of this precipitation.

[0032] Next, the system estimates the baseline deviation of the data. It uses a traversal calculation method to calculate the difference between each time element within the observation window and the arithmetic mean. These differences are then squared and summed to obtain a sum of squares. This sum of squares is then divided by the calibrated degrees of freedom parameter to obtain the initial expected dispersion. The degrees of freedom parameter is preset to the total number of samples within the observation window minus 1. This division by degrees of freedom, rather than directly by the total number of samples, aims to address the downward bias that can easily occur when estimating population variance in a small sample space. This ensures that the calculated initial expected dispersion fairly and unbiasedly reflects the theoretical variance in the real-world context.

[0033] Based on this, since traditional variance calculation methods tend to smooth out outlier distortion points, when strong gusts cause the tipping bucket meter to experience an occasional prolonged mechanical jam or a rapid series of tippings due to sudden flooding, relying on the conventional mean of squared deviations is unlikely to immediately reveal signs of an emergency. Therefore, a local variation of the classic variance formula with added penalty weights was designed. The system iterates through and arranges the ten time elements within the observation window, directly selecting and extracting the largest and smallest time elements and calculating their difference. To maintain dimensionality conservation during subsequent feature fusion, the system divides this extracted time difference by a pre-set time reference scale constant, thereby obtaining local range data in a relative form. The introduced time reference scale constant represents the theoretical calibration time of the tipping bucket under a baseline rainstorm in a laboratory calm wind environment, typically set to 1 second, and used as the normalization denominator. The system then multiplies this local range data by the previously calculated initial expected dispersion, essentially using local maxima deviations to amplify the conventional dispersion through endpoint mutation factors. The advantages of this variant calculation strategy are self-evident: In the event of sudden, extremely uneven rainfall, even if the overall mean fluctuation across the entire window is minimal, the isolated extreme differences will amplify the fundamental deviation index exponentially, thus providing a highly sensitive output of the variance representing the severity of the current hydrological environment. The aforementioned calculations can be integrated and expressed as the following variant formula: ; In the formula, This represents the variance of the final output of the system. The initial observation expected dispersion is calculated by dividing the sum of squares by the degrees of freedom. and These represent the maximum and minimum time elements within the observation window, respectively. The time reference scale constant participating in the constraint serves as a normalization denominator, converting the calculated absolute time range into a dimensionless abrupt change multiple, thereby maintaining dimensional conservation.

[0034] After extracting the variance reflecting the intensity of hydrological disturbances, the system substitutes it into the logical channel and directly compares it with the preset turbulence threshold of the equipment. This turbulence threshold represents the critical threshold at which the hydrological station's inherent static calibration curve can still withstand error fluctuations. If the comparison shows that the calculated variance is indeed less than the preset turbulence threshold, it proves that even if the current storm intensity is high, the overall pattern is still uniform and stable, and the rainfall has not been violently damaged by wind shear. In this case, the system decides to trust and retain the previous compensation calculation results, directly determining the basic corrected rainfall from the previous process as the actual storm intensity to be calculated. Conversely, if the variance is lower, it indicates that the station's hydrological and meteorological conditions are facing extremely complex destructive forces. In this situation, the system immediately abandons the original conservative processing approach and, exceeding its authority, extracts the soil moisture content, the elevation of the downcut water level, and the surface water velocity on the slope within the same monitoring section. These three factors represent the overall picture of runoff generation and confluence when the ground source receives and reflects these distorted rainwater. They will serve as multi-dimensional hydrological environmental variables and participate in the fusion calculation of the next in-depth flood control judgment and early warning model.

[0035] The watershed infiltration saturation is calculated based on soil moisture content, and the predicted runoff increment is output by converting the baseline corrected rainfall according to a preset runoff generation mapping relationship. The predicted runoff increment refers to the theoretical surface runoff increment obtained by subtracting the infiltration loss reflected by soil moisture content from the baseline corrected rainfall, based on the preset runoff generation mapping relationship. This includes: Calculate the porosity difference parameter by comparing the soil moisture content with the preset extreme values ​​of saturation. The porosity difference parameter is used as an index of watershed infiltration saturation. It is multiplied by the preset effective soil infiltration layer thickness to calculate and determine the current soil carrying capacity limit. The porosity difference parameter is the difference between the preset extreme value of water saturation and the current soil moisture content, which is used to characterize the remaining pore space of the current watershed soil that can accommodate rainwater infiltration. The preset effective soil infiltration layer thickness represents the depth of the active soil layer that can actually participate in and receive rainwater infiltration during a single rainfall event. Its preset logic is as follows: In the early stage of the hydrological monitoring station construction, the stratigraphic profile is extracted through geological exploration, and the average root retention depth of the surface vegetation around the station, the vertical hydraulic conductivity of the soil texture, and the burial depth of the underlying impermeable rock layer are comprehensively evaluated for static calibration. Specifically, when the surface vegetation at the monitoring station is dense, the root system is well-developed, resulting in numerous large pores in the soil, and the soil texture is loose sandy loam, a larger value should be used for the thickness of the infiltration layer; conversely, if the surface is bare and compacted, the vegetation is sparse, or there are impermeable rock slabs in the shallow layer, a smaller value should be used. This thickness parameter is conventionally taken as 300 mm to 800 mm, and 500 mm is recommended as the preferred value.

[0036] The baseline corrected rainfall is accumulated over time according to a preset calculation step to obtain the hypothetical total precipitation for the period. By comparing the hypothetical total precipitation over a given period with the current soil carrying capacity limit, the excess precipitation exceeding the current soil carrying capacity limit in the hypothetical total precipitation over a given period is extracted as the net rain source. The net rainfall source is multiplied by the catchment area coefficient to output the calculated predicted runoff increment.

[0037] In this embodiment of the invention, when the fluctuation variance is greater than or equal to the preset turbulence threshold, it means that the wind shear and disturbance caused by the current severe convective weather have disrupted the original calibration state of the tipping bucket meter. First, the current soil volumetric water content fed back by the soil water potential sensor buried at the same monitoring section is read, and the preset extreme value of the saturation rate of that stratum is retrieved from the system. This extreme value of the saturation rate represents the critical point of saturation when all soil pores are filled with water. Its preset logic is based on the actual soil texture of the survey site and is pre-fixed. For example, for a conventional sandy loam profile, this value is usually set in an adjustable range of 0.35 to 0.45, and 0.40 is recommended as the preferred value. Subtracting the current soil water content from the preset extreme value of the saturation rate, the difference constitutes the porosity difference parameter. Subsequently, the effective infiltration layer thickness of the soil is directly multiplied by the porosity difference parameter, thereby using this difference parameter as a watershed infiltration saturation index to determine the current soil carrying capacity limit value. The thickness of the effective soil infiltration layer is also preset based on the root system of the vegetation around the station and the characteristics of the topsoil. The conventional value is 300 mm to 800 mm, with 500 mm being the preferred value.

[0038] Using the previously derived baseline rainfall correction, an integral time series accumulation is performed according to a preset calculation step size. This preset calculation step size is a calculation window defined to handle transient responses to heavy rainfall; a 1-minute step size is recommended. Because rainfall intensity changes rapidly during heavy rain periods, a large step size could easily obscure the center of short-duration heavy precipitation. To balance computational load and high-frequency capture requirements, a 1-minute step size is also recommended. The baseline rainfall correction is continuously multiplied by its corresponding duration and accumulated to obtain the hypothetical total precipitation injected into the surface within this time window.

[0039] Water falling to the ground either seeps into the ground or remains on the surface as needed clean rain. Traditional single-point hydrological estimations often simply subtract the total precipitation from the soil carrying capacity, treating the excess as runoff. However, natural surface undulations and uneven local soil properties mean that puddles often saturate prematurely, overflowing and generating runoff, while slopes continue to absorb it. Therefore, this embodiment, in comparing the hypothetical total precipitation over a given period with the current soil carrying capacity value, introduces an infiltration stripping formula based on the spatial water storage capacity index distribution assumption to extract overflow precipitation. The specific calculation is as follows: ; Represents net rainfall; This represents the hypothetical total precipitation for the period just calculated; It is the measured current soil carrying capacity limit value.

[0040] After extracting this portion of net rainfall that rolls purely on the surface, the system multiplies it by the catchment area coefficient of the cross section. This outputs the final calculated predicted runoff increment.

[0041] Simultaneously, cross-sectional flow field calculations are performed based on water level elevation and surface water velocity to extract the actual runoff increment, including: Substitute the water level elevation into the geometric mapping equation of the station section to calculate the current cross-sectional water area; The instantaneous flow vector is obtained by multiplying the current cross-sectional area of ​​the water passage by the surface water velocity. Extract the current instantaneous flow vector for two consecutive calculation cycles to calculate the time-weighted evenly distributed flow; The total runoff for the current period is derived by multiplying the time-weighted average distribution flow by the time span of the corresponding calculation period. Calculate the absolute value of the difference between the total runoff in the current cycle and the total runoff in the immediately preceding cycle, and output this absolute value directly as the actual runoff increment.

[0042] In one specific embodiment, the system receives real-time water level data from a radar level gauge fixed at the monitoring section and substitutes it into the geometric mapping equation of the station section to calculate the current cross-sectional water flow area. This geometric mapping equation is a spatial integral polynomial obtained by a hydrological survey team during the dry season using a total station to conduct physical topographic mapping of the specific river channel cross-section and fitting irregular riverbed elevation points. Using this polynomial pre-written into the gateway's underlying layer, the system can calculate the two-dimensional spatial cross-sectional area that can accommodate the water flow below the current water level by simply inputting the absolute elevation of the surface water surface. Subsequently, the system retrieves the surface water velocity detected in real-time by an acoustic Doppler current meter deployed below the water surface, multiplies the calculated current cross-sectional water flow area by this surface water velocity, and extends the two-dimensional area to depth to obtain the instantaneous flow vector describing the current transient water potential throughput.

[0043] Simultaneously, the system traces backward in the time dimension, extracting the current instantaneous flow vector from two consecutive calculation cycles. This involves combining the flow measured at the immediately preceding moment with the currently obtained calculated flow, and combining these two to deduce a more representative time-weighted average flow. Considering that the conventional arithmetic mean method ignores the acceleration differences caused by water potential changes and cannot distinguish the nonlinearity between rapidly rising and slowly receding water, a nonlinear compensation variant formula based on the rate of flow change is introduced: ; In the formula, This represents the final calculated time-weighted average flow rate; The instantaneous flow vector obtained for the current calculation cycle; This is the instantaneous flow vector from the previous calculation cycle; It is the actual sampling time interval between these two flow measurements by the hardware facilities; The preset cross-sectional hydrodynamic hysteresis constant; It is a preset confidence weight parameter for the current state; This is the preset dynamic compensation correction coefficient. Regarding the preset part in the formula, The preset logic is to balance the signal-to-noise ratio between the old and new hydrological states. To allow the model to respond more sensitively to the latest rainfall developments, it is recommended to set it to an adjustable range of 0.6 to 0.8, with 0.7 being the preferred value; the cross-sectional hydrodynamic hysteresis constant... This is a hydraulic empirical value estimated by combining local riverbed roughness and water surface width-to-shallowness ratio, usually set to 5 to 15 seconds, with 10 seconds being the recommended optimum; while the dynamic compensation correction coefficient This is to prevent the derivative term from diverging abnormally due to obstruction of the flow meter by debris on the water surface, thus playing a damping constraint role. The value is usually between 0.1 and 0.3, with 0.2 being the preferred value.

[0044] The time-weighted average flow rate is further multiplied by the corresponding calculation period's time span to convert the dynamic flow velocity into a static total water volume, thus deriving the current cycle's total runoff for the entire time period. Since flood control assessment needs to consider both the power load of field monitoring equipment and the timeliness of disaster warnings, the calculation period's time span is recommended to be between 30 and 120 minutes, with 60 minutes being the preferred option. Next, the system horizontally extracts the total runoff from the previous cycle with identical data, calculates the difference between the current cycle's total runoff and the immediately preceding cycle's total runoff, and directly takes the absolute value of this difference. The logic behind using the absolute value of the difference instead of simply retaining the positive or negative sign is that the backend fusion warning model does not need to distinguish whether the river is rising or falling at this point, but rather to obtain the magnitude of the fluctuation in the intensity of rainfall disturbance to the actual water body. Finally, the system directly outputs this absolute value of the difference as the actual runoff increment.

[0045] A dynamic coordination coefficient is constructed by calculating the ratio of actual runoff increment to predicted runoff increment. The baseline corrected rainfall is then multiplied by the dynamic coordination coefficient for secondary compensation to obtain the true rainfall intensity, including: Compare the obtained ratio with the preset tolerance value range; If the ratio exceeds the upper limit of the preset tolerance range, extract the excess overflow ratio. Input the difference overflow ratio into the natural logarithmic smoothing equation to perform dimensionality reduction mapping and obtain the mutation penalty weight; Based on the preset allocation ratio, a dynamic synergy coefficient is generated by superimposing a preset benchmark constant and a mutation penalty weight. The corrected rainfall intensity is calculated by multiplying the base corrected rainfall amount with the dynamic synergy coefficient compensated for by extreme turbulence characteristics.

[0046] In practice, the system first extracts the actual runoff increment calculated in the previous steps and divides it by the predicted runoff increment calculated based on the theoretical model to obtain the ratio between the two. At the physical dimension level, both the actual and predicted runoff increments represent specific water volumes, with the same dimension of cubic meters. After dividing, the units cancel each other out, resulting in a dimensionless, purely proportional parameter. This ratio quantifies the deviation of the actual surface runoff scale from the theoretically projected rainfall scale. Considering the uneven distribution of soil pores in nature and the reasonable background noise of conventional sensors, the obtained ratio is compared horizontally with a preset tolerance range. The preset logic for this tolerance range is derived through statistical convergence of historical hydrological records under stable, windless precipitation conditions at the station over the years, typically defined as between 0.85 and 1.15. Here, the upper limit of the tolerance range is preferably set at 1.15, meaning that a maximum of 15% positive hydrological fluctuation is allowed as normal background noise.

[0047] When the comparison process finds that the calculated ratio exceeds the upper limit of the preset tolerance range of 1.15, it indirectly confirms that the current rainstorm is accompanied by extremely strong turbulence and wind shear, causing a precipitous reduction in the data recorded by the rain gauge. At this point, the system subtracts the preset upper limit from the current ratio to extract the excess overflow ratio specifically induced by extreme weather. Faced with this overflow data representing the degree of under-measurement, if a simple linear ratio is used to directly multiply it back into the original rainfall, in the event of a localized short-term flash flood, the transient surge in the runoff multiple will cause the derived rainfall intensity value to diverge without limit, leading to the collapse of the subsequent early warning model. Therefore, this operation node specifically introduces a modified natural logarithmic smoothing equation, substituting the excess overflow ratio into it for dimensionality reduction mapping, thereby obtaining a relatively mild and physically limit-compliant abrupt change penalty weight. The specific calculation formula is as follows: ; In the formula, This represents the final calculated mutation penalty weight; This is the overflow ratio value extracted by the subtraction just now; `ln` is the preset terrain confluence amplification factor; `ln` is the natural logarithm operator. The constant 1 in the formula ensures that when the overflow difference returns to zero, the entire logarithmic term smoothly returns to zero without negative numbers or mathematically meaningless terms. Regarding the preset terrain confluence amplification factor... The setting logic is to adjust the sensitivity to water flow convergence based on the average slope and vegetation cover of the watershed where the station is located. For steep, bare rock terrain with extremely rapid water flow convergence, it is recommended to set this value lower to prevent overcompensation; for gentle, flat woodlands with slow water flow and easy water concealment, it is recommended to set it higher. For conventional hilly monitoring scenarios, a value of 1.2 is recommended.

[0048] After obtaining the mutation penalty weights, the system begins synthesizing the final compensation multipliers. A preset baseline constant, representing the uncorrected original state, is pre-set in system memory and is always 1.0. Subsequently, the system reads a preset allocation ratio and linearly superimposes the preset baseline constant and the mutation penalty weights according to this ratio to generate a dynamic synergy coefficient. Specifically, the superposition calculation logic is as follows: the dynamic synergy coefficient equals the preset baseline constant plus the product of the allocation ratio and the mutation penalty weights. It is recommended that the allocation ratio be set between 0.3 and 0.6, with 0.4 being the preferred value.

[0049] The final step involves retrieving the transmitted baseline correction rainfall and multiplying it by the compensated dynamic coordination coefficient. The original dimension of the baseline correction rainfall is millimeters per minute. After multiplying by the dimensionless dynamic coordination coefficient, the value itself undergoes a corresponding compensatory expansion, but its physical dimension remains unchanged and anchored at millimeters per minute.

[0050] The obtained actual rainfall intensity, water level elevation, surface water flow velocity, and soil moisture content are input into the early warning discrimination matrix. The predicted peak water level is then extracted and a multi-element early warning signal is generated, including: Feature subsets of atmospheric input layer, runoff distribution layer and hydrodynamic convergence layer are constructed in the internal dimension of the early warning discrimination matrix; The actual rainfall intensity is mapped to the atmospheric input layer, the soil moisture content is mapped to the runoff distribution layer, and the water level and surface water velocity are mapped to the hydrodynamic collection layer. The extreme value reduction decay operator is used to reduce and filter the data sequence of the atmospheric input layer with the runoff distribution layer index to extract the deep net rainfall set; the extreme value reduction decay operator is a nonlinear reduction function constructed based on the interception capacity of surface vegetation and the soil pre-wet index, which is used to attenuate and correct the actual rainfall intensity in the atmospheric input layer in order to extract the deep net rainfall set that enters the surface runoff stage. The deep net rainwater harvesting layer combined with the hydrodynamic catchment layer status is imported into the watershed time delay evolution calculation table to extrapolate and calculate the predicted peak water level at the future target time node; Based on the predicted peak water level, the corresponding hazard label is extracted from the flood control dictionary, and a corresponding multi-element early warning signal is generated according to its severity level.

[0051] Preferably, the predicted peak water level at future target time points is calculated by incorporating the deep net rainwater harvesting layer state into the watershed time delay evolution calculation table, including: Extract the impedance coefficient table and surface slope matrix within the mapping range; The deep net rainwater harvesting is converted into a water source injection, and the water flow propulsion vector path is calculated according to the surface slope matrix. The input time data sequence is calculated by multiplying the elements in the impedance coefficient table along the water flow vector path; Based on the evolution time data sequence, the flow volume that converges to the same target section is accumulated and integrated to derive the flow duration curve; The peak horizontal and vertical coordinate vectors are extracted from the flow duration curve, and the equivalent liquid level elevation is back-mapped in the extreme value distribution model as the predicted water level peak.

[0052] In practical implementation, the system first initializes a multi-dimensional data structure in memory, namely the early warning discrimination matrix. Within this matrix, three logically isolated but interconnected feature subsets are defined: the atmospheric input layer, the runoff distribution layer, and the water flow collection layer. Specifically, the atmospheric input layer is a one-dimensional time-series array that records water supply over time; the runoff distribution layer is constructed as a two-dimensional spatial matrix corresponding to the watershed's geographical topology, with each grid node representing the soil infiltration potential of the corresponding plot; and the water flow collection layer uses monitoring sections as its core, constructing node state vectors containing water level and flow velocity attributes.

[0053] After defining the underlying matrix architecture, the system reads the compensated true rainfall intensity in mm / min and fills it into the time series array of the atmospheric input layer according to the sampling timestamp. Simultaneously, it reads the soil moisture content transmitted back by the underground sensors, interpolates and matches the latitude and longitude coordinates of each sensor with the grid coordinates of the two-dimensional spatial matrix of the runoff distribution layer, and writes the values ​​into the corresponding grid nodes. At the same time, it directly assigns the water level elevation and surface water velocity measured by the river cross-section radar sensor to the current node state vector of the hydrodynamic collection layer.

[0054] The system extracts soil moisture content from the runoff distribution layer and uses it as a suppressor to reduce and filter the actual rainfall data sequence in the atmospheric input layer, thereby extracting the deep net rain collection that can truly flow to the surface. An extreme value reduction decay operator is introduced here, and the specific calculation formula is as follows: ; In the formula, This represents the net deep rainfall calculated at the current moment. This represents the actual intensity of the heavy rainfall transmitted from the atmospheric input layer. This represents the current soil moisture content mapped onto the runoff distribution layer grid. This represents the upper limit of the saturated water content of the soil corresponding to this grid. This is a preset nonlinear attenuation constant for infiltration. The logic of this formula is that when the soil is extremely dry... Approaching 0, and the value within parentheses approaching 1, rainwater is almost entirely absorbed, resulting in minimal net surface rainfall; as the soil gradually becomes saturated, its infiltration capacity increases exponentially. The nonlinear shape decreases sharply. Preset constant. The value depends on the degree of compaction of the local topsoil. For conventional yellow soil, it is recommended to set it between 1.2 and 1.8. In this embodiment, 1.5 is preferred.

[0055] After obtaining the actual surface water yield, the system imports this portion of deep net rainfall, combined with the basic state of the hydrodynamic catchment layer, into the watershed time-delay evolution calculation table to begin simulating the crawling process of water flow on the physical surface. Specifically, the system retrieves the impedance coefficient table and surface slope matrix within the current mapping range from the underlying geographic information database. The surface slope matrix records the slope drop ratio of each grid to adjacent low-lying grids, while the impedance coefficient table records the drag resistance coefficients of surface vegetation and obstacles on the water flow, with the dimension set to s / m. The system multiplies the deep net rainfall by the physical area of ​​each grid to convert it into a volumetric water source, with the unit being... Then, following the direction of the fastest gradient descent in the surface slope matrix, a water flow propulsion vector path is calculated.

[0056] The system calculates the time required for water flow to pass through each grid cell by multiplying the grid path length by the elements in the impedance coefficient table along these vector paths and dividing by the square root of the surface slope matrix elements. This accumulated time sequence yields the evolution time data for each water flow to reach the main channel. Based on this time sequence, the system performs integral accumulation on the flow rates expected to converge at the target cross-section at the same future time. Since early rainfall and distant rainfall may arrive at the same time, this physical time-based integral derivation ultimately plots a flow-time curve on the coordinate axis reflecting future flow changes at the target cross-section, with time on the horizontal axis and flow rate on the vertical axis. .

[0057] To obtain a clearer picture of the flood control water level, the system scans and extracts the point with the largest vertical coordinate value from the aforementioned flow duration curve. The flow vector corresponding to this point is the predicted peak flow. Subsequently, the system uses an extreme value distribution model to inversely calculate the water level based on this peak flow. Here, the classic power function empirical formula for water level-flow conversion is used: .in The extracted predicted peak flow rate is expressed in units of... ; 1 represents the baseline elevation of the riverbed at the station; a and b are the characteristic constants of the river channel section that were previously fitted and solidified using measured data from previous years.

[0058] Finally, based on the predicted peak water level The system then searches the flood control dictionary for comparison. The dictionary contains pre-entered critical labels for various water levels, such as warning level, guaranteed level, and overflow level. Once the system finds a matching severity level range, it extracts the corresponding response strategy and combines it to generate a multi-element early warning signal containing the estimated arrival time and peak water level.

[0059] It should be noted that the interval and threshold sizes are set for ease of comparison. The size of the threshold depends on the amount of sample data and the base number set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless calculations, and the formulas are derived from software simulations using a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0060] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0061] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

[0062] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for online sensing and intelligent early warning of multiple hydrological elements, characterized in that, include: The sequence of flipping time points triggered by the tipping bucket rain gauge at the hydrological monitoring section is obtained, and the adjacent time intervals in the flipping time point sequence are extracted. The instantaneous rainfall intensity is obtained by dividing the preset water volume by the adjacent time interval. In response to the instantaneous rainfall intensity exceeding the critical threshold, a basic compensation coefficient corresponding to the instantaneous rainfall intensity is matched according to the preset rainfall intensity loss mapping curve, and the basic compensation coefficient is multiplied by the instantaneous rainfall intensity to obtain the basic corrected rainfall. Extracting multiple consecutive adjacent time intervals to calculate the variance of fluctuation; If the variance of the fluctuation is less than the preset turbulence threshold, the base correction rainfall will be set to the actual rainfall intensity to be calculated. If the variance of the fluctuation is greater than or equal to the preset turbulence threshold, the soil moisture content, water level elevation and surface water velocity within the same monitoring section are extracted. The watershed infiltration saturation is calculated based on soil moisture content, and the predicted runoff increment is output by combining the basic corrected rainfall with the preset runoff generation mapping relationship. Simultaneously, cross-sectional flow field calculations are performed based on water level elevation and surface water velocity to extract the actual runoff increment; The ratio of actual runoff increment to predicted runoff increment is calculated to construct a dynamic coordination coefficient. The base correction rainfall is multiplied by the dynamic coordination coefficient for secondary compensation to obtain the true rainfall intensity. The obtained real rainfall intensity, water level elevation, surface water flow velocity, and soil moisture content are input into the early warning discrimination matrix, and the predicted peak water level is extracted and multi-element early warning signal is generated.

2. The online sensing and intelligent early warning method for multiple hydrological elements according to claim 1, characterized in that, The watershed infiltration saturation is calculated based on soil moisture content, and the predicted runoff increment is output by converting the baseline corrected rainfall according to a preset runoff generation mapping relationship, including: Calculate the porosity difference parameter by comparing the soil moisture content with the preset extreme values ​​of saturation. The porosity difference parameter is used as the watershed infiltration saturation index. It is multiplied by the preset effective soil infiltration layer thickness to calculate and determine the current soil carrying capacity limit value. The baseline corrected rainfall is accumulated over time according to a preset calculation step to obtain the hypothetical total precipitation for the period. By comparing the hypothetical total precipitation over a given period with the current soil carrying capacity limit, the excess precipitation exceeding the current soil carrying capacity limit in the hypothetical total precipitation over a given period is extracted as the net rain source. The net rainfall source is multiplied by the catchment area coefficient to output the calculated predicted runoff increment.

3. The online sensing and intelligent early warning method for multiple hydrological elements according to claim 1, characterized in that, Simultaneously, cross-sectional flow field calculations are performed based on water level elevation and surface water velocity to extract the actual runoff increment, including: Substitute the water level elevation into the geometric mapping equation of the station section to calculate the current cross-sectional water area; The instantaneous flow vector is obtained by multiplying the current cross-sectional area of ​​the water passage by the surface water velocity. Extract the current instantaneous flow vector for two consecutive calculation cycles to calculate the time-weighted evenly distributed flow; The total runoff for the current period is derived by multiplying the time-weighted average distribution flow by the time span of the corresponding calculation period. Calculate the absolute value of the difference between the total runoff in the current cycle and the total runoff in the immediately preceding cycle, and output this absolute value directly as the actual runoff increment.

4. The online sensing and intelligent early warning method for multiple hydrological elements according to claim 1, characterized in that, A dynamic coordination coefficient is constructed by calculating the ratio of actual runoff increment to predicted runoff increment. The baseline corrected rainfall is then multiplied by the dynamic coordination coefficient for secondary compensation to obtain the true rainfall intensity, including: Compare the obtained ratio with the preset tolerance value range; If the ratio exceeds the upper limit of the preset tolerance range, extract the excess overflow ratio. Input the difference overflow ratio into the natural logarithmic smoothing equation to perform dimensionality reduction mapping and obtain the mutation penalty weight; Based on the preset allocation ratio, a dynamic synergy coefficient is generated by superimposing a preset benchmark constant and a mutation penalty weight. The corrected rainfall intensity is calculated by multiplying the base corrected rainfall amount with the dynamic synergy coefficient compensated for by extreme turbulence characteristics.

5. The online sensing and intelligent early warning method for multiple hydrological elements according to claim 1, characterized in that, Extracting multiple consecutive adjacent time intervals to calculate the variance of fluctuation includes: Extract the time interval observation window by backtracking a preset number of sequentially adjacent time intervals from the flipped time point sequence; The arithmetic mean of all elements within the observation window is accumulated over a time interval. Calculate the sum of squares of the differences between each element and the arithmetic mean within the observation window over the time interval; The initial expected dispersion of observations is obtained by dividing the sum of squares by the degree of freedom parameter. The adjacent time intervals with the largest values ​​within the time interval observation window are selected as the maximum time element, and the adjacent time intervals with the smallest values ​​are selected as the minimum time element. The difference between the two is calculated to obtain the local range data. The local range data is multiplied by the expected dispersion of the initial observations and then amplified by the endpoint mutation factor to output the fluctuation variance representing the hydrological variation disturbance.

6. The online sensing and intelligent early warning method for multiple hydrological elements according to claim 1, characterized in that, The method also involves an adaptive iterative replacement of the cumulative precipitation mapping curve with a preset rainfall intensity loss curve, which includes: The total precipitation is theoretically estimated by integrating and accumulating the actual rainfall intensity based on discrete time slices. Obtain the cumulative precipitation generated based on independent physical mechanisms within the same time slice; Calculate the relative deviation rate between the cumulative precipitation and the theoretically projected total precipitation; In response to the relative deviation rate exceeding the confidence threshold in multiple consecutive discrete time slices, the extracted relative deviation rate sequence is input into the Kalman filter equation to filter and update the baseline offset parameter. The updated baseline offset parameter is used to calculate the offset compensation of the tangent slope of the current preset rainfall intensity loss mapping curve in the corresponding rainfall intensity interval, and the iteratively updated preset rainfall intensity loss mapping curve is output.

7. The online sensing and intelligent early warning method for multiple hydrological elements according to claim 1, characterized in that, The obtained actual rainfall intensity, water level elevation, surface water flow velocity, and soil moisture content are input into the early warning discrimination matrix. The predicted peak water level is then extracted and a multi-element early warning signal is generated, including: Feature subsets of atmospheric input layer, runoff distribution layer and hydrodynamic convergence layer are constructed in the internal dimension of the early warning discrimination matrix; The actual rainfall intensity is mapped to the atmospheric input layer, the soil moisture content is mapped to the runoff distribution layer, and the water level and surface water velocity are mapped to the hydrodynamic collection layer. The extreme value reduction decay operator is used to reduce and filter the data sequence of the atmospheric input layer with the runoff distribution layer index to extract the deep net rain collection; The deep net rainwater harvesting layer combined with the hydrodynamic catchment layer status is imported into the watershed time delay evolution calculation table to extrapolate and calculate the predicted peak water level at the future target time node; Based on the predicted peak water level, the corresponding hazard label is extracted from the flood control dictionary, and a corresponding multi-element early warning signal is generated according to its severity level.

8. The online sensing and intelligent early warning method for multiple hydrological elements according to claim 7, characterized in that, By incorporating the deep net rainwater harvesting layer state into the watershed time delay evolution calculation table, the predicted peak water level at future target time points is calculated, including: Extract the impedance coefficient table and surface slope matrix within the mapping range; The deep net rainwater harvesting is converted into a water source injection, and the water flow propulsion vector path is calculated according to the surface slope matrix. The input time data sequence is calculated by multiplying the elements in the impedance coefficient table along the water flow vector path; Based on the evolution time data sequence, the flow volume that converges to the same target section is accumulated and integrated to derive the flow duration curve; The peak horizontal and vertical coordinate vectors are extracted from the flow duration curve, and the equivalent liquid level elevation is back-mapped in the extreme value distribution model as the predicted water level peak.

9. The online sensing and intelligent early warning method for multiple hydrological elements according to claim 1, characterized in that, Extracting adjacent time intervals from the flipping time sequence, and calculating the instantaneous rainfall intensity by dividing the preset water-flipping amount by the adjacent time interval, includes: The first, second, and third time points triggered by the completion of the overturning action of the small room are continuously captured; The preceding time distance parameter is calculated by subtracting the first time point from the second time point, and the following time distance parameter is calculated by subtracting the second time point from the third time point. The transient rainfall intensity of the first stage is obtained by calculating the ratio of the preset water volume to the preceding time distance parameter, and the transient rainfall intensity of the second stage is obtained by calculating the ratio of the preset water volume to the subsequent time distance parameter. The attenuation ratio weights calculated based on time proximity are assigned to the transient rainfall intensity in the first and second phases, respectively. Perform a weighted summation calculation on the initial and subsequent transient rainfall intensities after allocating attenuation ratio weights, and output the instantaneous rainfall intensity at this moment.

Images