A flood forecasting method and system based on big data and artificial intelligence

Abstract

The application discloses a flood forecasting method and system based on big data and artificial intelligence, and relates to the technical field of hydrological forecasting.The method comprises the following steps: using seepage pressure monitoring data, displacement monitoring data, stress monitoring data and vibration monitoring data to respectively calculate seepage regulation indexes, displacement regulation indexes, stress regulation indexes and vibration regulation indexes, and constructing a flood season regulation characteristic vector; according to the flood season regulation characteristic vector and historical flood data, a flood season objective function and a flood season constraint condition are established, and a flood process forecasting model is constructed through deep learning training; based on the flood process forecasting model, real-time hydrological data is input, and a flood forecasting result is output. The application innovatively converts reservoir multi-source monitoring data into quantitative regulation indexes, effectively solving the problem that the flood process obtained by the traditional forecasting method cannot meet the actual flood control demand when extreme weather or special working conditions are encountered.

Description

Technical Field

[0001] This invention relates to the field of hydrological forecasting technology, and more specifically, to a flood forecasting method and system based on big data and artificial intelligence. Background Technology

[0002] With the intensification of global climate change and the increasing frequency of extreme weather events, reservoir flood control scheduling faces unprecedented challenges. As a crucial engineering measure for flood control and disaster reduction, the scientific scheduling of reservoirs is directly related to the flood control safety of river basins and economic and social development. In recent years, the rapid development of big data and artificial intelligence technologies has provided new technical support for reservoir flood control scheduling, especially in flood forecasting and determination of flood control limits. Advanced algorithms such as deep learning and knowledge graphs can better extract value from historical data, improving forecast accuracy and control efficiency. Against this backdrop, how to fully utilize new technologies to enhance the flood control capacity of reservoirs has become an important research direction in the field of water conservancy engineering.

[0003] Currently, reservoir flood forecasting mainly relies on hydrological forecasting systems and flood season control rules. Traditional forecasting methods establish rainfall-runoff models, combining upstream and downstream hydrological conditions and meteorological forecasts to predict the inflow process over a future period, thereby determining the flood control water level scheme. With the improvement of automation and informatization, various monitoring equipment has been widely used in reservoir projects, realizing real-time monitoring of reservoir operation status. These monitoring data include seepage pressure, displacement deformation, stress and strain, and vibration response, providing important basis for dynamically determining the flood control water level. At the same time, machine learning methods based on historical data have also made significant progress in flood forecasting.

[0004] However, existing flood forecasting technologies primarily consider hydrological conditions (such as rainfall and upstream / downstream water levels) and fixed flood season control rules, failing to incorporate real-time reservoir monitoring data (such as seepage pressure, displacement deformation, stress-strain, and seismic response) into the forecasting system. This means that in extreme weather or special operating conditions, the flood events predicted by traditional forecasting methods may not meet actual flood control needs, especially when determining flood control limits based on forecast results. For example, when abnormally high seepage pressure is detected, it may be necessary to appropriately lower the flood control limit; when unconventional seismic responses are observed, it may be necessary to adjust the discharge method. These reservoir conditions all affect the actual flood evolution process. Therefore, there is an urgent need for an intelligent flood forecasting method that can comprehensively consider multi-source monitoring data. Summary of the Invention

[0005] To address the problems in related technologies, this invention proposes a flood forecasting method and system based on big data and artificial intelligence. It possesses the advantage of establishing seepage control indices, displacement control indices, stress control indices, and vibration control indices to quantitatively represent real-time reservoir monitoring data. Furthermore, it uses these control indices to form a flood season control feature vector, which serves as a crucial input feature for a deep learning model. This solves the problem that existing technologies primarily consider hydrological conditions (such as rainfall and upstream / downstream water levels) and fixed flood season control rules, failing to incorporate real-time reservoir monitoring data (such as seepage pressure, displacement deformation, stress and strain, and vibration response) into the forecasting system. Consequently, in extreme weather or special operating conditions, the flood events predicted by traditional forecasting methods may not meet actual flood control needs.

[0006] Therefore, the specific technical solution adopted by the present invention is as follows:

[0007] According to one aspect of the present invention, a flood forecasting method based on big data and artificial intelligence is provided, the method comprising the following steps:

[0008] S1. Using seepage pressure monitoring data, displacement monitoring data, stress monitoring data and vibration monitoring data, calculate the seepage control index, displacement control index, stress control index and vibration control index respectively, and construct the flood season control feature vector.

[0009] S2. Based on the flood season regulation feature vector and historical flood data, establish the flood season objective function and flood season constraints, and construct a flood process forecasting model through deep learning training;

[0010] S3. Based on the flood process forecasting model, input real-time hydrological data and output flood forecasting results.

[0011] Furthermore, using seepage pressure monitoring data, displacement monitoring data, stress monitoring data, and vibration monitoring data, the seepage control index, displacement control index, stress control index, and vibration control index are calculated respectively, and a flood season control feature vector is constructed, including the following steps:

[0012] S11. Based on seepage pressure monitoring data, extract real-time pressure changes and historical average values ​​for the same period, and calculate the seepage control index by combining hydrological element correction coefficients.

[0013] S12. Based on real-time displacement monitoring data, obtain standardized displacement values, median displacement, and standard deviation, and use the seasonal influence coefficient to solve the displacement control index.

[0014] S13. Using real-time stress monitoring data, analyze the standardized stress value and temperature correction coefficient, and calculate the stress control index by combining the material characteristic correction parameters.

[0015] S14. Based on real-time vibration monitoring data, construct a vibration control index, and combine it with the seepage control index, displacement control index and stress control index to establish a flood season control characteristic vector.

[0016] Furthermore, based on seepage pressure monitoring data, real-time pressure changes and historical averages are extracted, and combined with hydrological element correction coefficients, the seepage control index is calculated, including the following steps:

[0017] S111. Based on the continuous time period seepage pressure monitoring data, generate the pressure difference between adjacent time periods, and obtain the real-time seepage pressure change through pressure change calculation;

[0018] S112. Based on historical seepage pressure data for the same period, extract the statistical characteristics of seepage pressure and use mathematical statistics methods to obtain the average seepage pressure for the same period in history.

[0019] S113. Based on the reservoir water level and rainfall data, solve for the seepage correction coefficient, and combine the real-time seepage pressure change with the historical average seepage pressure for the same period to calculate the seepage control index.

[0020] The expression for the seepage control index is:

[0021] PI = ΔP / (P) c ×α)

[0022] In the formula, PI is the seepage control index, ΔP is the real-time seepage pressure change, and P c The mean seepage pressure is the historical average for the same period, and α is the seepage pressure correction factor.

[0023] Furthermore, based on real-time displacement monitoring data, standardized displacement values, median displacement, and standard deviation are obtained. Using the seasonal influence coefficient, the displacement control index is calculated, including the following steps:

[0024] S121. Based on real-time displacement monitoring data, extract the current displacement feature value and synthesize it using displacement components to obtain a standardized displacement value;

[0025] S122. Based on historical displacement monitoring data, establish displacement statistical parameters, and obtain the median and standard deviation of displacement based on probability statistical methods.

[0026] S123. Based on meteorological element monitoring data, solve for the seasonal influence coefficient, and obtain the displacement control index according to the standardized displacement value and displacement statistical parameters.

[0027] The expression for the displacement control index is:

[0028] DI = (DD) m ) / (σ×β)

[0029] In the formula, DI is the displacement control index, D is the standardized displacement value, and D m σ is the median displacement, σ is the standard deviation, and β is the seasonal influence coefficient.

[0030] Furthermore, by utilizing real-time stress monitoring data, analyzing standardized stress values ​​and temperature correction coefficients, and combining them with material characteristic correction parameters, the stress regulation index is calculated, including the following steps:

[0031] S131. Based on real-time stress monitoring data, determine the standardized stress value using the finite element interpolation method;

[0032] S132. Based on continuous temperature monitoring data, analyze the temperature change trend and obtain the temperature correction coefficient through statistical analysis;

[0033] S133. Based on the material characteristics of the stress concentration region, solve for the material correction coefficient, and combine the standardized stress value and the allowable stress value to calculate the stress control index.

[0034] The expression for the stress regulation index is:

[0035] SI=(S×k) / (S a ×γ)

[0036] In the formula, SI is the stress regulation index, S is the standardized stress value in the dam heel region, k is the temperature correction coefficient in the dam heel region, and S a γ represents the allowable stress value of concrete in the dam heel region, and γ is the correction factor for concrete material in the dam heel region.

[0037] Furthermore, based on real-time vibration monitoring data, a vibration control index is constructed, and combined with the seepage control index, displacement control index, and stress control index, a flood season control characteristic vector is established, including the following steps:

[0038] S141. Based on real-time vibration monitoring data, extract the peak acceleration time series to obtain standardized vibration values;

[0039] S142. Based on environmental vibration monitoring data, analyze the background frequency and amplitude parameters, and calculate the vibration correction coefficient;

[0040] S143. Using temperature and frequency monitoring data, solve for the dynamic correction coefficient, and combine the standardized vibration value with the vibration correction coefficient to calculate the vibration control index.

[0041] The expression for the vibration control index is:

[0042] VI=(V×μ) / (V a ×λ)

[0043] In the formula, VI is the vibration control index, V is the standardized vibration value, μ is the vibration correction coefficient, and Va Where λ is the permissible vibration value, and λ is the dynamic correction factor;

[0044] S144. Construct a flood season control feature vector based on the seepage control index, displacement control index, stress control index, and vibration control index;

[0045] The expression for the flood season regulation characteristic vector is:

[0046] F = [PI, DI, SI, VI]

[0047] In the formula, F is the flood season regulation characteristic vector, PI is the seepage regulation index, DI is the displacement regulation index, SI is the stress regulation index, and VI is the vibration regulation index.

[0048] Furthermore, based on the flood season regulation feature vector and historical flood data, a flood season objective function and flood season constraints are established, and a flood process forecasting model is constructed through deep learning training, including the following steps:

[0049] S21. Based on historical flood data and flood season regulation feature vectors, model input features are obtained through data standardization, and a training sample set is constructed.

[0050] S22. Establish a deep neural network structure and establish a flood season objective function based on the root mean square error;

[0051] S23. Based on the regulation feature vector, generate flood season constraints and establish a flood process forecasting model in conjunction with the flood season objective function.

[0052] Furthermore, the model input features include hydrological features, flood season regulation feature vectors, and forecasted rainfall values;

[0053] Hydrological characteristics include inflow, outflow, and water level;

[0054] The expression for the objective function during the flood season is:

[0055]

[0056] In the formula, RM is the objective function for the flood season, n is the forecast period length, t is the time period number within the forecast period, and Y... t Y is the measured value during time period t. t ′ represents the predicted value for time period t.

[0057] Furthermore, the expression for the flood season constraint is as follows:

[0058]

[0059] In the formula, ΔZ is the water level fluctuation during the flood season, a is the normal water level fluctuation threshold during the flood season, b is the water level fluctuation threshold of concern during the flood season, c is the warning water level fluctuation threshold during the flood season, and max(F) is the maximum value of the seepage control index, displacement control index, stress control index and vibration control index in the flood season control characteristic vector.

[0060] According to another aspect of the present invention, a flood forecasting system based on big data and artificial intelligence is also provided, the system comprising:

[0061] The data analysis module is used to calculate the seepage control index, displacement control index, stress control index and vibration control index respectively using seepage pressure monitoring data, displacement monitoring data, stress monitoring data and vibration monitoring data, and to construct the flood season control feature vector.

[0062] The artificial intelligence module is used to establish the flood season objective function and flood season constraints based on the flood season regulation feature vector and historical flood data, and to build a flood process forecasting model through deep learning training.

[0063] The results output module is used to take real-time hydrological data as input and output flood forecast results based on the flood process forecasting model.

[0064] The beneficial effects of this invention are as follows:

[0065] (1) This invention establishes seepage control index, displacement control index, stress control index and vibration control index to realize the quantitative characterization of real-time monitoring data of reservoirs, and uses these control indices to form a flood season control feature vector as an important input feature of deep learning model. This method not only considers traditional hydrological conditions, but also incorporates the safety status indicators of reservoirs into the forecasting system, so that the forecast results can respond to abnormal working conditions of reservoirs in a timely manner. In addition, this multi-source data fusion forecasting technology significantly improves the scientificity and reliability of flood forecasting. Especially when encountering extreme weather or special working conditions, it can adaptively adjust the forecasting strategy according to the real-time status of the reservoir, and provide more reliable decision support for reservoir flood control scheduling.

[0066] (2) This invention uses deep learning technology to construct a flood process forecasting model. Through a four-layer neural network structure, it achieves deep integration of multi-dimensional features such as hydrological characteristics, regulation feature vectors, and forecasted rainfall. By designing a flood season objective function based on root mean square error and introducing flood season constraints based on regulation feature vectors, the flood forecast results can ensure both prediction accuracy and meet the requirements for safe reservoir operation. Experiments show that the forecast error of this invention is controlled within a small range within a 24-hour lead time, and the forecast results meet the water level fluctuation threshold. This organic combination of forecast accuracy and constraints ensures that the forecast results not only conform to the evolution law of hydrological processes but also adapt to the real-time operation status of reservoirs. This effectively solves the technical problem that traditional forecasting methods cannot cope with special working conditions and provides strong support for intelligent flood control scheduling of reservoirs. Attached Figure Description

[0067] Figure 1 This is a flowchart illustrating a flood forecasting method based on big data and artificial intelligence according to an embodiment of the present invention;

[0068] Figure 2 This is a schematic diagram of a flood forecasting system based on big data and artificial intelligence according to an embodiment of the present invention. Detailed Implementation

[0069] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention. The components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0070] According to embodiments of the present invention, a flood forecasting method and system based on big data and artificial intelligence are provided.

[0071] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, according to an embodiment of the present invention, a flood forecasting method based on big data and artificial intelligence is provided, which includes the following steps:

[0072] S1. Using seepage pressure monitoring data, displacement monitoring data, stress monitoring data and vibration monitoring data, calculate the seepage control index, displacement control index, stress control index and vibration control index respectively, and construct the flood season control feature vector.

[0073] S2. Based on dynamic operation status assessment, obtain the main evaluation indicators;

[0074] S3. Based on the main evaluation indicators, an optimized operation plan is obtained.

[0075] In one embodiment, the seepage control index, displacement control index, stress control index, and vibration control index are calculated using seepage pressure monitoring data, displacement monitoring data, stress monitoring data, and vibration monitoring data, respectively, and a flood season control feature vector is constructed, including the following steps:

[0076] S11. Based on seepage pressure monitoring data, extract real-time pressure changes and historical average values ​​for the same period, and calculate the seepage control index by combining hydrological element correction coefficients.

[0077] S12. Based on real-time displacement monitoring data, obtain standardized displacement values, median displacement, and standard deviation, and use the seasonal influence coefficient to calculate the displacement control index.

[0078] S13. Based on the monitoring data of the buried stress gauges and strain gauges, extract the spatial distribution data of the stress and strain field, and use stress concentration analysis to obtain the stress and strain anomaly index.

[0079] In one embodiment, the calculation of the seepage control index, based on seepage pressure monitoring data, extracts real-time pressure changes and historical average values ​​for the same period, and combines these with hydrological element correction coefficients, including the following steps:

[0080] S111. Based on the continuous time period seepage pressure monitoring data, generate the pressure difference between adjacent time periods, and obtain the real-time seepage pressure change through pressure change calculation;

[0081] Specifically, based on the seepage pressure monitoring data within 24 hours, data is collected every 4 hours to form a time series of [0.26, 0.27, 0.28, 0.29, 0.32, 0.31] MPa. The pressure difference ΔP between the two latest time periods is calculated. In this embodiment, the pressure value measured in the current time period (t = 20h) is 0.32 MPa, and the pressure value measured in the previous time period (t = 16h) is 0.29 MPa, so ΔP = 0.03 MPa is obtained.

[0082] S112. Based on historical seepage pressure data for the same period, extract the statistical characteristics of seepage pressure and use mathematical statistics methods to obtain the average seepage pressure for the same period in history.

[0083] Specifically, based on the seepage pressure data for the same period over the past five years, the historical average seepage pressure Pc is calculated using the moving average method. In this embodiment, the data for the same period (second half of June) from 2019 to 2023 are [0.31, 0.29, 0.30, 0.28, 0.32] MPa, respectively. Pc is obtained by weighted averaging (with weights of [0.1, 0.15, 0.2, 0.25, 0.3]). c =0.30MPa.

[0084] S113. Based on the reservoir water level and rainfall data, solve for the seepage correction coefficient, and combine the real-time seepage pressure change with the historical average seepage pressure for the same period to calculate the seepage control index.

[0085] Specifically, the seepage pressure correction coefficient is calculated using the reservoir water level H and the rainfall R. The expression for the seepage pressure correction coefficient is: α = 0.85 + 0.002H + 0.001R. In this embodiment, when the reservoir water level H = 145m and the cumulative rainfall R in the past 24 hours = 50mm, α = 1.15 is calculated.

[0086] The expression for the seepage control index is:

[0087] PI = ΔP / (P) c ×α)

[0088] In the formula, PI is the seepage control index, ΔP is the real-time seepage pressure change, and P c The mean seepage pressure is the historical average for the same period, and α is the seepage pressure correction factor.

[0089] Specifically, the ΔP obtained in the above embodiment is 0.03 MPa, P c Substituting α = 0.30 MPa and α = 1.15 into the calculation formula, we get PI = 0.03 / (0.30 × 1.15) = 0.087.

[0090] In one embodiment, obtaining standardized displacement values, median displacement, and standard deviation based on real-time displacement monitoring data, and calculating the displacement control index using the seasonal influence coefficient includes the following steps:

[0091] S121. Based on real-time displacement monitoring data, extract the current displacement feature value and synthesize it using displacement components to obtain a standardized displacement value;

[0092] Specifically, based on the displacement monitoring data in the X, Y, and Z directions of a measuring point at the center of the dam crest of a reservoir dam within 24 hours, the displacement synthesis formula D = (X... 2 +Y 2 +Z 2 The standardized displacement value is calculated as follows: In this embodiment, the displacement of the measuring point at the center of the dam crest in the current time period (t = 20h) is 1.2mm in the X direction, 0.8mm in the Y direction, and 0.5mm in the Z direction. The standardized displacement value D = 1.5mm is then calculated.

[0093] S122. Based on historical displacement monitoring data, establish displacement statistical parameters, and obtain the median and standard deviation of displacement based on probability statistical methods.

[0094] Specifically, based on the displacement monitoring data of this measuring point for the same period over the past five years (June 15th - June 30th), the displacement parameters were calculated using statistical analysis methods. In this embodiment, the data for the same period from 2019 to 2023 were [1.0, 1.1, 1.2, 1.3, 1.4] mm, respectively. The median displacement D was obtained by weighted averaging (with weights of [0.1, 0.15, 0.2, 0.25, 0.3]). m =1.2mm, standard deviation σ=0.2mm.

[0095] S123. Based on meteorological element monitoring data, solve for the seasonal influence coefficient, and obtain the displacement control index according to the standardized displacement value and displacement statistical parameters.

[0096] Specifically, the seasonal influence coefficient β is calculated using the temperature T, humidity H, and sunshine duration S collected by the on-site meteorological station at the reservoir dam. The expression for the seasonal influence coefficient is β = 0.90 + 0.003T + 0.002H + 0.001S. In this embodiment, when the temperature T = 25℃, the relative humidity H = 85%, and the sunshine duration S = 6h, β = 1.12 is calculated.

[0097] The expression for the displacement control index is:

[0098] DI = (DD) m ) / (σ×β)

[0099] In the formula, DI is the displacement control index, D is the standardized displacement value, and D m σ is the median displacement, σ is the standard deviation, and β is the seasonal influence coefficient.

[0100] Specifically, the D = 1.5 mm obtained in the above embodiment, D m Substituting σ = 0.2 mm and β = 1.12 into the calculation formula, we get DI = (1.5 - 1.2) / (0.2 × 1.12) = 1.34.

[0101] In one embodiment, the stress regulation index is calculated by extracting standardized stress values ​​and temperature correction coefficients based on real-time stress monitoring data and combining them with material characteristic correction parameters, including the following steps:

[0102] S131. Based on real-time stress monitoring data, determine the standardized stress value using the finite element interpolation method;

[0103] Specifically, based on real-time stress monitoring data from 15 monitoring points (5 in the dam heel area, 5 in the middle of the dam body, and 5 in the dam crest area) deployed from upstream to downstream of the dam body, the monitoring data for the dam heel area are [2.7, 2.8, 2.9, 2.7, 2.8] MPa, the monitoring data for the middle of the dam body are [2.5, 2.6, 2.4, 2.5, 2.4] MPa, and the monitoring data for the dam crest area are [2.3, 2.5, 2.6, 2.4, 2.5] MPa.

[0104] It should be noted that the finite element interpolation method is a numerical method that uses the stress values ​​of a finite number of discrete points to perform meshing and numerical approximation on the entire region through shape functions, thereby obtaining the stress value at any point. This is existing technology and will not be elaborated on here.

[0105] Specifically, the average stress value of each region was calculated using finite element interpolation, resulting in a stress value of 2.78 MPa for the dam heel region, 2.48 MPa for the middle of the dam body, and 2.46 MPa for the dam crest region. In this embodiment, the maximum stress value at the measuring point in the dam heel region during the current time period (t = 20 h) was selected as the standardized stress value S = 2.8 MPa.

[0106] S132. Based on continuous temperature monitoring data, analyze the temperature change trend and obtain the temperature correction coefficient through statistical analysis;

[0107] Specifically, based on temperature data collected by temperature sensors installed on the dam surface over the past three days, data was collected every 6 hours at 5 measuring points in the dam heel area (located consistent with the stress monitoring points), forming a time series of [18.5, 19.2, 20.1, 21.5, 22.8, 23.5, 24.2, 25.0, 24.8, 24.5, 24.0, 23.2]℃. Using the temperature correction coefficient calculation formula k = 1.05 - 0.004ΔT, in this embodiment, when the temperature variation ΔT = 25.0 - 18.5 = 6.5℃, k = 0.95 was calculated.

[0108] S133. Based on the material characteristics of the stress concentration region, solve for the material correction coefficient, and combine the standardized stress value and the stress tolerance value to calculate the stress control index.

[0109] Specifically, the elastic modulus Y of the concrete material in the dam heel area and the concrete strength f are used. c Calculate the material correction factor using the formula γ = (0.95 + 0.001Y) × (f c / f c0 In this embodiment, when the elastic modulus of the concrete in the dam heel region is Y = 32 GPa, the measured strength f of the concrete in the dam heel region is... c =35MPa, design strength f c0When the pressure is 30 MPa, γ is calculated to be 1.18.

[0110] The expression for the stress regulation index is:

[0111] SI=(S×k) / (S a ×γ)

[0112] In the formula, SI is the stress regulation index, S is the standardized stress value in the dam heel region, k is the temperature correction coefficient in the dam heel region, and S a γ represents the allowable stress value of concrete in the dam heel region, and γ is the correction factor for concrete material in the dam heel region.

[0113] Specifically, the standardized stress value S = 2.8 MPa in the dam heel region obtained in the above embodiments, the temperature correction factor k = 0.95 in the dam heel region, and the allowable concrete stress value S in the dam heel region are... a =3.5MPa, and the correction factor for concrete material in the dam heel area is γ =1.18. Substituting these values ​​into the calculation formula, we get SI = (2.8 × 0.95) / (3.5 × 1.18) = 0.64.

[0114] In one embodiment, a vibration control index is constructed based on real-time vibration monitoring data, and a flood season control feature vector is established by combining the seepage control index, displacement control index, and stress control index, including the following steps:

[0115] S141. Based on real-time vibration monitoring data, extract the peak acceleration time series to obtain standardized vibration values;

[0116] Specifically, based on 24-hour vibration data collected by five accelerometers deployed on the dam crest, peak acceleration was recorded hourly. The acceleration time series at measuring point 1 was [0.12, 0.13, 0.14, 0.15, 0.13, 0.14, 0.15, 0.14, 0.15, 0.14, 0.13, 0.15, 0.14, 0.15, 0.14, 0.13, 0.15, 0.14, 0.15, 0.14, 0.15, 0.14, 0.15] m / s 2 In this embodiment, the maximum peak acceleration of measuring point 1 in the current time period (t = 20h) is selected as the standardized vibration value V = 0.15m / s². 2 .

[0117] S142. Based on environmental vibration monitoring data, analyze the background frequency and amplitude parameters, and calculate the vibration correction coefficient;

[0118] Specifically, based on data collected from three environmental vibration monitoring points (50m, 100m, and 150m from the dam crest) deployed around the dam, background frequency and amplitude were recorded hourly, forming a background frequency sequence [24,25,25,26,23,25] Hz and an amplitude sequence [0.018,0.020,0.019,0.020,0.021,0.020] m / s. 2 The vibration correction factor is calculated using the formula μ = 1.0 - 0.002F + 0.001A. In this embodiment, when the background frequency F = 25Hz and the amplitude A = 0.020m / s are selected at the monitoring point 100m away from the dam crest for the current time period (t = 20h), the vibration correction factor is calculated using the formula μ = 1.0 - 0.002F + 0.001A. 2 When the time was calculated, μ = 0.92.

[0119] S143. Using temperature and frequency monitoring data, solve for the dynamic correction coefficient, and combine the standardized vibration value with the vibration correction coefficient to calculate the vibration control index.

[0120] Specifically, the dynamic correction coefficient is calculated using the ratio of temperature T to frequency f / f0. The calculation formula is λ=(0.90+0.002T)×(f / f0)^0.5. In this embodiment, when the dam body temperature T=25℃, the measured frequency f=2.8Hz, and the design frequency f0=3.0Hz, the calculated λ=1.08.

[0121] The expression for the vibration control index is:

[0122] VI=(V×μ) / (V a ×λ)

[0123] In the formula, VI is the vibration control index, V is the standardized vibration value, μ is the vibration correction coefficient, and V a λ is the permissible vibration value, and λ is the dynamic correction coefficient.

[0124] The V obtained in the previous steps is 0.15 m / s. 2 μ = 0.92, V a =0.20m / s 2 Substituting λ = 1.08 into the calculation formula, we get VI = (0.15 × 0.92) / (0.20 × 1.08) = 0.64.

[0125] S144. Construct a flood season control feature vector based on the seepage control index, displacement control index, stress control index, and vibration control index.

[0126] The expression for the flood season regulation feature vector is:

[0127] F = [PI, DI, SI, VI]

[0128] In the formula, F is the flood season regulation characteristic vector, PI is the seepage regulation index, DI is the displacement regulation index, SI is the stress regulation index, and VI is the vibration regulation index.

[0129] Specifically, substituting PI = 0.087, DI = 1.34, SI = 0.64, and VI = 0.64 obtained in the previous steps into the calculation formula, we get F = [0.087, 1.34, 0.64, 0.64].

[0130] In one embodiment, the objective function of the forecasting model is constructed based on the flood season regulation feature vector and historical flood data, and the flood process forecasting model is obtained through deep learning training, including the following steps:

[0131] S21. Based on historical flood data and regulation feature vectors, obtain model input features through data standardization and construct a training sample set;

[0132] Specifically, the model input features include:

[0133] ① Hydrological characteristics, including inflow Q t =[2500,2800,3100,3400,3800,4200]m 3 / s;

[0134] Outbound flow q t =[2000,2300,2600,3000,3400,3800]m 3 / s;

[0135] Water level Z t =[175.2,175.8,176.4,177.0,177.6,178.2]m.

[0136] ② Flood season regulation characteristic vector, which includes 4 regulation indices;

[0137] F = [PI,DI,SI,VI] = [0.087,1.34,0.64,0.64].

[0138] ③ Forecast rainfall value, R t +n=[10,15,20]mm (rainfall forecast for the next 3 periods); the normalized feature sequence is obtained by processing it using the min-max standardization method.

[0139] Specifically, the model input features from the flood season (June-September) of 2019-2023 were selected to form a training sample set.

[0140] S22. Establish a deep neural network structure and establish a flood season objective function based on the root mean square error;

[0141] Specifically, a four-layer deep neural network is constructed, with 13 nodes in the input layer (6 hydrological features + 4 regulation indices + 3 forecast rainfall values), [64, 32, 16] nodes in the hidden layer, and 3 nodes in the output layer (forecast water level, inflow, and outflow).

[0142] Specifically, the model input features include hydrological features, flood season regulation feature vectors, and forecasted rainfall values;

[0143] Hydrological characteristics include inflow, outflow, and water level;

[0144] The expression for the objective function during the flood season is:

[0145]

[0146] In the formula, RM is the objective function for the flood season, n is the forecast period length (the response time of reservoirs and watersheds is usually between several hours and a day, so a 24-hour forecast period can effectively capture the key processes of flood evolution; therefore, in this embodiment, the forecast period is set to 24 hours and divided into 6 time periods, i.e., the forecast period length n = 6), t is the time period number within the forecast period, t = 1, 2, ..., 6 (in this embodiment, the 24-hour forecast period is divided into 6 4-hour time periods), Y t The measured values ​​for time period t, including the measured water level Z. t (m), measured inflow rate Q t (m 3 / s) and measured outflow q t (m 3 / s), Y t ′ represents the predicted value for time period t, including the predicted water level Z. t ′(m), Predicted inflow Q t ′(m 3 / s) and predicted outbound flow q t ′(m 3 / s).

[0147] S23. Based on the regulation feature vector, generate flood season constraints and establish a flood process forecasting model in conjunction with the model objective function.

[0148] Specifically, the expression for the flood season constraint is:

[0149]

[0150] In the formula, ΔZ is the water level fluctuation during the flood season, a is the normal water level fluctuation threshold during the flood season (2.0m in this embodiment), b is the water level fluctuation threshold of concern during the flood season (1.5m in this embodiment), c is the warning water level fluctuation threshold during the flood season (1.0m in this embodiment), and max(F) is the maximum value of the seepage control index, displacement control index, stress control index, and vibration control index in the flood season control characteristic vector.

[0151] Specifically, in this embodiment, the seepage control index PI = 0.087 indicates that the seepage state is normal; the displacement control index DI = 1.34 indicates that the displacement exceeds the warning value; the stress control index SI = 0.64 indicates that the stress state is normal; the vibration control index VI = 0.64 indicates that the vibration state is normal; finally, since max(F) = 1.34 > 1.0, the water level fluctuation ΔZ = c = 1.0m is determined.

[0152] Specifically, in the above embodiments, based on the current water level Z t =175.2m, establish water level constraints: Z min ≤Z t ≤Z max In the formula, Z min =175.2-1.0=174.2m; Z max =175.2 + 1.0 = 176.2m.

[0153] Specifically, in the above embodiment, the Adam optimizer is used, the learning rate is set to 0.001, the training epochs are 1000, the flood process forecasting model is trained using the training sample set, and the final RM on the validation set is 0.15.

[0154] In one embodiment, based on a flood process forecasting model, real-time hydrological data is input, and flood forecasting results are output.

[0155] Specifically, real-time hydrological data includes the current inflow, outflow, and water level; flood forecast results include water level predictions, inflow predictions, and outflow predictions for future periods.

[0156] Specifically, when t=1 (i.e., the next 0-4 hours);

[0157] Measured value Y1 = [Z1 = 176.5m, Q1 = 3200m] 3 / s,q1=2800m 3 / s];

[0158] Predicted value Y1' = [Z1' = 176.3m, Q1' = 3150m] 3 / s,q1'=2750m 3 / s];;

[0159] When t = 2 (i.e., 4-8 hours from now);

[0160] Measured value Y2=[Z2=177.0m, Q2=3500m] 3 / s,q2=3100m 3 / s];

[0161] Predicted values ​​Y2' = [Z2' = 176.8m, Q2' = 3450m] 3 / s,q2'=3050m 3 / s];;

[0162] This process continues until t=6; the trained flood process forecasting model can simultaneously optimize the prediction accuracy of flood level and flow, thereby providing more accurate flood forecast results.

[0163] To facilitate understanding of the above-mentioned technical solution of the present invention, the following detailed description is provided using a tributary reservoir in a certain area as an example:

[0164] The reservoir, located on a tributary of the Yangtze River, is a large reservoir primarily for flood control, but also provides comprehensive benefits such as power generation and water supply. It controls a drainage area of ​​2,800 square kilometers, has a total storage capacity of 1.25 billion cubic meters, and a flood control capacity of 420 million cubic meters. The dam is a concrete gravity dam, 145 meters high and 430 meters long at the crest. Completed and put into operation in 2018, the reservoir has a comprehensive dam safety monitoring system, including seepage pressure monitoring, displacement monitoring, stress monitoring, and vibration monitoring.

[0165] In late June 2023, the basin experienced heavy rainfall, necessitating flood control scheduling of the reservoir. Based on the flood forecasting method provided by this invention, real-time monitoring data of the reservoir was first collected, including the seepage pressure monitoring value of 0.32 MPa at the dam heel area, the displacement monitoring value of 1.5 mm at the center measuring point of the dam crest, the stress monitoring value of 2.8 MPa at the dam heel area, and the vibration acceleration value of 0.15 m / s² at the dam crest. 2 The calculated regulation feature vector is F = [0.087, 1.34, 0.64, 0.64]. The displacement regulation index DI = 1.34 exceeds the warning value, indicating that stricter control over the reservoir water level fluctuation is needed.

[0166] Based on the current hydrological conditions (inflow rate 3100 m³ / h) 3 / s, outbound flow rate 2600m³ 3The data (water level 175.2m) and the 24-hour rainfall forecast (estimated rainfall amounts of 10mm, 15mm, and 20mm) are input into a trained deep learning model to obtain the flood forecast for the next 24 hours. The forecast results show that, under the constraint of ensuring the water level fluctuation does not exceed 1.0m, the reservoir water level will fluctuate within 176.2m, and the maximum inflow will not exceed 3500m³. 3 / s provides an important reference for the formulation of reservoir flood control scheduling plans. Practice has shown that the forecasting method proposed in this invention considers both hydrological conditions and incorporates the real-time status of the reservoir into the forecasting system, resulting in forecasts that better meet actual flood control needs.

[0167] like Figure 2 As shown, according to another embodiment of the present invention, a flood forecasting system based on big data and artificial intelligence is also provided, the system comprising:

[0168] Data analysis module 1 is used to calculate the seepage control index, displacement control index, stress control index and vibration control index respectively using seepage pressure monitoring data, displacement monitoring data, stress monitoring data and vibration monitoring data, and to construct the flood season control feature vector;

[0169] Artificial intelligence module 2 is used to establish the flood season objective function and flood season constraints based on the flood season regulation feature vector and historical flood data, and to build a flood process forecasting model through deep learning training;

[0170] Result output module 3 is used to take real-time hydrological data as input and output flood forecast results based on the flood process forecasting model.

[0171] In summary, by utilizing the technical solutions described above, this invention establishes seepage control indices, displacement control indices, stress control indices, and vibration control indices, thereby achieving a quantitative representation of real-time reservoir monitoring data. These control indices are then used to form a flood season control feature vector, serving as a crucial input feature for a deep learning model. This method not only considers traditional hydrological conditions but also incorporates reservoir safety status indicators into the forecasting system, enabling timely responses to abnormal reservoir conditions. Furthermore, this multi-source data fusion forecasting technology significantly improves the scientific rigor and reliability of flood forecasts. Particularly in the event of extreme weather or special operating conditions, it can adaptively adjust forecasting strategies based on the real-time reservoir status, providing more reliable decision support for reservoir flood control scheduling. This invention employs deep learning technology to construct a flood process forecasting model. Through a four-layer neural network structure, it achieves deep fusion of multi-dimensional features such as hydrological characteristics, regulatory feature vectors, and forecasted rainfall. By designing a flood season objective function based on root mean square error and introducing flood season constraints based on regulatory feature vectors, the flood forecast results can ensure both prediction accuracy and meet the requirements for safe reservoir operation. Experiments show that the forecast error of this invention is controlled within a small range within a 24-hour lead time, and the forecast results meet the water level fluctuation threshold. This organic combination of forecast accuracy and constraints ensures that the forecast results not only conform to the evolution law of hydrological processes but also adapt to the real-time operating status of reservoirs. This effectively solves the technical problem that traditional forecasting methods cannot cope with special operating conditions and provides strong support for intelligent flood control scheduling of reservoirs.

[0172] 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 flood forecasting method based on big data and artificial intelligence, characterized in that, Includes the following steps: S1. Using seepage pressure monitoring data, displacement monitoring data, stress monitoring data and vibration monitoring data, calculate the seepage control index, displacement control index, stress control index and vibration control index respectively, and construct the flood season control feature vector. S1 includes the following steps: S11. Based on seepage pressure monitoring data, extract real-time pressure changes and historical average values ​​for the same period, and calculate the seepage control index by combining hydrological element correction coefficients. S12. Based on real-time displacement monitoring data, obtain standardized displacement values, median displacement, and standard deviation, and use the seasonal influence coefficient to solve the displacement control index. S13. Using real-time stress monitoring data, analyze the standardized stress value and temperature correction coefficient, and calculate the stress control index by combining the material characteristic correction parameters. S14. Based on real-time vibration monitoring data, construct a vibration control index, and combine it with the seepage control index, displacement control index, and stress control index to establish a flood season control feature vector; S14 includes the following steps: S141. Based on real-time vibration monitoring data, extract the peak acceleration time series to obtain standardized vibration values; S142. Based on environmental vibration monitoring data, analyze the background frequency and amplitude parameters, and calculate the vibration correction coefficient; S143. Using temperature and frequency monitoring data, solve for the dynamic correction coefficient, and combine the standardized vibration value with the vibration correction coefficient to calculate the vibration control index. The expression for the vibration control index is: VI=(V×μ) / (V a ×λ) In the formula, VI is the vibration control index, V is the standardized vibration value, μ is the vibration correction coefficient, and V a Where λ is the permissible vibration value, and λ is the dynamic correction factor; S144. Construct a flood season control feature vector based on the seepage control index, displacement control index, stress control index, and vibration control index; The expression for the flood season regulation feature vector is: F = [PI, DI, SI, VI] In the formula, F is the flood season regulation characteristic vector, PI is the seepage regulation index, DI is the displacement regulation index, SI is the stress regulation index, and VI is the vibration regulation index. S2. Based on the flood season regulation feature vector and historical flood data, establish the flood season objective function and flood season constraints, and construct a flood process forecasting model through deep learning training; S3. Based on the flood process forecasting model, input real-time hydrological data and output flood forecasting results.

2. The flood forecasting method based on big data and artificial intelligence according to claim 1, characterized in that, S11 includes the following steps: S111. Based on the continuous time period seepage pressure monitoring data, generate the pressure difference between adjacent time periods, and obtain the real-time seepage pressure change through pressure change calculation; S112. Based on historical seepage pressure data for the same period, extract the statistical characteristics of seepage pressure and use mathematical statistics methods to obtain the average seepage pressure for the same period in history. S113. Based on the reservoir water level and rainfall data, solve for the seepage correction coefficient, and combine the real-time seepage pressure change with the historical average seepage pressure for the same period to calculate the seepage control index. The expression for the seepage control index is: PI = ΔP / (P c ×α) In the formula, PI is the seepage control index, ΔP is the real-time seepage pressure change, and P c The mean seepage pressure is the historical average for the same period, and α is the seepage pressure correction factor.

3. The flood forecasting method based on big data and artificial intelligence according to claim 1, characterized in that, S12 includes the following steps: S121. Based on real-time displacement monitoring data, extract the current displacement feature value and synthesize it using displacement components to obtain a standardized displacement value; S122. Based on historical displacement monitoring data, establish displacement statistical parameters, and obtain the median and standard deviation of displacement based on probability statistical methods. S123. Based on meteorological element monitoring data, solve for the seasonal influence coefficient, and obtain the displacement control index according to the standardized displacement value and displacement statistical parameters. The expression for the displacement control index is: DI=(DD m ) / (σ×β) In the formula, DI is the displacement control index, D is the standardized displacement value, and D m σ is the median displacement, σ is the standard deviation, and β is the seasonal influence coefficient.

4. The flood forecasting method based on big data and artificial intelligence according to claim 1, characterized in that, S13 includes the following steps: S131. Based on real-time stress monitoring data, determine the standardized stress value using the finite element interpolation method; S132. Based on continuous temperature monitoring data, analyze the temperature change trend and obtain the temperature correction coefficient through statistical analysis; S133. Based on the material characteristics of the stress concentration region, solve for the material correction coefficient, and combine the standardized stress value and the allowable stress value to calculate the stress control index. The expression for the stress regulation index is: SI=(S×k) / (S a ×γ) In the formula, SI is the stress regulation index, S is the standardized stress value in the dam heel region, k is the temperature correction coefficient in the dam heel region, and S a γ represents the allowable stress value of concrete in the dam heel region, and γ is the correction factor for concrete material in the dam heel region.

5. The flood forecasting method based on big data and artificial intelligence according to claim 1, characterized in that, S2 includes the following steps: S21. Based on historical flood data and flood season regulation feature vectors, model input features are obtained through data standardization, and a training sample set is constructed. S22. Establish a deep neural network structure and establish a flood season objective function based on the root mean square error; S23. Based on the regulation feature vector, generate flood season constraints and establish a flood process forecasting model in conjunction with the flood season objective function.

6. The flood forecasting method based on big data and artificial intelligence according to claim 5, characterized in that, The model input features include hydrological features, flood season regulation feature vectors, and forecasted rainfall values; The hydrological characteristics include inflow, outflow and water level; The expression for the objective function during the flood season is: In the formula, RM is the objective function for the flood season, n is the forecast period length, t is the time period number within the forecast period, and Y... t Y is the measured value during time period t. t ′ represents the predicted value for time period t.

7. The flood forecasting method based on big data and artificial intelligence according to claim 5, characterized in that, The expression for the flood season constraint is: In the formula, ΔZ is the water level fluctuation during the flood season, a is the normal water level fluctuation threshold during the flood season, b is the water level fluctuation threshold of concern during the flood season, c is the warning water level fluctuation threshold during the flood season, and max(F) is the maximum value of the seepage control index, displacement control index, stress control index and vibration control index in the flood season control characteristic vector.

8. A flood forecasting system based on big data and artificial intelligence, used to implement the flood forecasting method based on big data and artificial intelligence as described in any one of claims 1-7, characterized in that, This flood forecasting system, based on big data and artificial intelligence, includes: The data analysis module is used to calculate the seepage control index, displacement control index, stress control index and vibration control index respectively using seepage pressure monitoring data, displacement monitoring data, stress monitoring data and vibration monitoring data, and to construct the flood season control feature vector. The artificial intelligence module is used to establish the flood season objective function and flood season constraints based on the flood season regulation feature vector and historical flood data, and to build a flood process forecasting model through deep learning training. The results output module is used to take real-time hydrological data as input and output flood forecast results based on the flood process forecasting model.

Images