A small and medium-sized reservoir operation safety evaluation system based on sky-ground-water engineering

Abstract

The application provides a small and medium-sized reservoir operation safety evaluation system based on sky-ground-waterworks, and belongs to the technical field of hydrological monitoring.The application comprises a time-space alignment layer, a feature weighting layer and a dynamic coupling layer connected in sequence;the time-space alignment layer performs time alignment and space alignment on the input data of sky-ground-waterworks;the feature weighting layer quantifies the contribution degree of different monitoring features to safety evaluation according to the data after time-space alignment;the dynamic coupling layer adopts a dynamic weight distribution strategy for different modules through a multi-module coordination engine, fuses the feature weighting, obtains a fusion result, and outputs a small and medium-sized reservoir operation safety index according to the fusion result.The evaluation result of the application is more accurate and has better timeliness.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of hydrological monitoring, and in particular to a small and medium-sized reservoir operation safety evaluation system based on sky-ground-water engineering. BACKGROUND

[0002] Small and medium-sized reservoirs play an important role in the water conservancy engineering system in China, especially the earth-rock dams, which are numerous and widely distributed, and bear multiple functions such as flood control, irrigation, water supply, power generation, etc., playing an indispensable role in regional economic development and people's life. However, due to factors such as construction age, technical level and capital investment, some small and medium-sized reservoirs face many safety hazards in the operation process. Traditional monitoring methods often have problems such as untimely information acquisition, insufficient data accuracy, limited monitoring range, etc., and are difficult to meet the real-time, comprehensive monitoring needs of reservoir safe operation. Under the background of frequent extreme weather events, the safety risk of small and medium-sized reservoirs is further highlighted, and once a dam break or other accidents occur, it will cause great loss to the life and property of the downstream people, seriously affecting social stability and economic sustainable development.

[0003] The current evaluation method for the operation safety of small and medium-sized reservoirs mainly uses image recognition, and a dam safety evaluation model system is constructed based on the static basic data of the reservoir dam, the real-time monitoring data of the engineering entity and the runoff conditions. The source data is relatively single, only reflects the local working condition information of the project, and has not been coordinated and fused in terms of time and space, range, precision, frequency, etc., and has not realized comprehensive evaluation of the whole factor, whole process and whole time domain of the reservoir project.

[0004] The traditional dam safety evaluation model has two defects:

[0005] Single data dimension: only relying on limited factors such as dam body conditions (displacement, settlement) and water conditions (reservoir water level, rainfall), ignoring the systematic correlation of the basin-reservoir-dam body;

[0006] Poor monitoring timeliness: relying on manual inspection and fixed sensors, unable to realize high-frequency, high-precision and full-space coverage monitoring, leading to hidden danger discovery lag. SUMMARY

[0007] To solve the technical problems existing in the prior art, the present application provides a small and medium-sized reservoir operation safety evaluation system based on sky-ground-water engineering, which realizes.

[0008] The present application provides a small and medium-sized reservoir operation safety evaluation system based on sky-ground-water engineering, which comprises a time-space alignment layer, a feature weighting layer and a dynamic coupling layer connected in sequence;

[0009] The time-space alignment layer performs time alignment and space alignment on the input sky-ground-water engineering data;

[0010] a feature weighting layer, configured to quantify the contribution degrees of different monitoring features to the safety evaluation according to the spatio-temporally aligned data;

[0011] a dynamic coupling layer, configured to fuse the feature weighting by a dynamic weight distribution strategy of different modules through a multi-module coordination engine to obtain a fusion result, and output a medium and small-sized reservoir operation safety index according to the fusion result, wherein the multi-module includes a short-term risk prediction module, a medium and long-term stability prediction module, and an extreme working condition deduction module; the short-term risk prediction module is used for safety evaluation of the dam within a first time of real time or preset; the medium and long-term stability prediction module is used for evaluation of the stability of the dam within a preset year; and the extreme working condition deduction module is used for evaluation of the structural safety under extreme conditions by simulating the influence of an overload event on the dam body.

[0012] Further, the time alignment of the input data is specifically that, for data of different sampling frequencies, time series alignment is achieved by a dynamic curved time axis, and for low-frequency data lower than a preset frequency threshold, cubic spline interpolation is used to generate preset high-frequency time point data.

[0013] Further, for data of different sampling frequencies, dynamic time warping is used, and the specific formula is as follows:

[0014]

[0015] wherein,

[0016] is a weight coefficient of a time point of the data to be aligned; m

[0017] is a time decay coefficient;

[0018] is a time point of a reference time series; m

[0019] is a time point of the data to be aligned;

[0020] is a time point of the reference time series;

[0021] is a total number of time points of the reference sequence.

[0022] Further, for low-frequency data lower than a preset frequency threshold, the following formula is used to generate preset high-frequency time point data:

[0023]

[0024] wherein, ​​

[0025] 、 、 、 are coefficients of the cubic polynomial, determined by cubic spline interpolation, ensuring that the interpolated function is continuous and the first derivative, the second derivative is continuous within the interval [ , ];

[0026] is the target time point to be interpolated, located within the interval [ , ];

[0027] is the previous time point in the original low-frequency data;

[0028] is the next time point adjacent to in the original low-frequency data;

[0029] is the estimated value obtained by cubic spline interpolation at time .

[0030] Further, the spatial alignment of data is performed by the following method: all data are converted to the same coordinate system, deformation is eliminated by Gauss-Kruger projection, and continuous spatial distribution is generated by using spatial interpolation kernel function on discrete monitoring point data.

[0031] Further, the spatial interpolation kernel function is as follows:

[0032]

[0033] wherein,

[0034] is the normalized spatial distance;

[0035] is the spatial interpolation weight.

[0036] Further, the feature weighting layer adopts feature selection based on attention mechanism, and dynamically adjusts the feature weight by combining context-aware attention and environmental working conditions.

[0037] Further, the features in the sky level are compressed in weight by the following formula, and the original features are reduced in dimension:

[0038]

[0039] wherein,

[0040] is the PCA projection matrix, which is obtained by eigen-decomposition of the covariance matrix of the original data;

[0041] is the transpose of the aligned original eigenvector, is the total number of original eigenvectors;

[0042] is the compressed eigenvector in the sky level, l k represents the dimension after dimension reduction.

[0043] Further, in the cross-level of the sky + ground + environment context vector, the feature is weighted and compressed by the following formula:

[0044]

[0045] wherein,

[0046] is the attention weight of the j level;

[0047] is the environmental context vector;

[0048] is the compressed eigenvector of the j level;

[0049] is the summation index, taking a value in the range of 1 to 4, corresponding to different levels;

[0050] is the learnable weight vector, is the transpose of ;

[0051] is a concatenation operator, concatenating and by dimension to form a comprehensive vector containing level features and environmental information.

[0052] Further, the feature weighting layer adopts a static and dynamic mixed weighting guided by expert knowledge. The static weight is preset according to the expert experience, and the dynamic weight is adjusted according to the real-time working condition.

[0053] ​​Further, the short-term prediction module uses an LSTM-Transformer model to mine local features and global features in the reservoir operation data. The LSTM-Transformer model includes an LSTM layer, a Transformer self-attention layer, and a fully connected layer. The LSTM layer captures the time sequence relationship of data features, the Transformer self-attention layer identifies the spatial attention weight between multi-sensor data features, and the fully connected layer outputs the dam collapse probability using a Sigmoid activation function.

[0054] Further, the medium and long-term stability prediction module constructs a physically guided neural network architecture, inputs low-frequency cumulative data and environmental parameters, embeds the constitutive equation of earth-rock dams as a physical constraint, and learns the nonlinear relationship between historical monitoring data and geological parameters using a multi-layer perceptron as data-driven.

[0055] Further, the extreme working condition deduction module constructs a dam-foundation-reservoir three-dimensional model based on BIM, integrates geometric models, material properties, and safety boundary conditions, and establishes different parameterized scene libraries, then performs coupled solving of multiple scenes to obtain pore water pressure distribution, update effective stress field, and iteratively solve displacement and strain based on elastic-plastic constitutive equation, and extract the safety factor FS.

[0056] Further, the geometric model includes the geometric model of the terrain, dam structure, and flood discharge facility, the material property includes the constitutive parameters of earth-rock materials and the time-varying characteristics of the permeability coefficient, and the safety boundary condition includes the reservoir water pressure, seismic input, and seepage boundary.

[0057] Further, the dynamic weight distribution strategy defines a rule base through a fuzzy logic controller to achieve adaptive weighting of different models, and updates the feature weighting parameters in real time based on evaluation errors.

[0058] Further, if the reservoir water level > warning water level, then the weight of the short-term prediction module = 0.8; if the seismic intensity ≥ 6, then the weight of the extreme working condition deduction module = 0.7; if the reservoir water level is too high and the rainfall intensity is heavy rain, then the weight of the short-term prediction module = 0.8, the weight of the medium and long-term stability prediction module = 0.1, and the weight of the extreme working condition deduction module = 0.1.

[0059] Further, after outputting the results, the predicted results are compared with the actual monitoring data to trigger model retraining in each module, new monitoring data is injected into the training set, and the feature weighting parameters are updated. The adaptive correction equation based on model parameter adjustment is as follows:

[0060]

[0061] wherein,

[0062] For index, used to identify the first data point or the first monitoring index;

[0063] For the first original monitoring data;

[0064] For the first prediction value based on Kalman filtering;

[0065] σ For the historical data standard deviation;

[0066] For the first corrected sensor data value;

[0067] For the weighting coefficient, when the data deviation exceeds 3σ, the weights of the original data and the prediction value are balanced.

[0068] Compared with the prior art, the beneficial effects of the present application are as follows:

[0069] 1. The present application establishes a three-level calculation system of "spatiotemporal alignment-feature weighting-dynamic coupling", breaks through the traditional single-point static evaluation mode, and the evaluation result is more accurate and the timeliness is better.

[0070] 2. The present application aligns the data in all dimensions: it supports second-level to annual scale data fusion in time, and the spatial resolution reaches sub-micron level, and can eliminate more than 75% of data conflicts caused by spatiotemporal differences. For example, it realizes real-time fusion of satellite rainfall data (0.5 hour delay) and ground radar data to generate minute-level rainfall distribution map.

[0071] 3. The present application intelligently adjusts the feature weight: the response time of feature weight dynamic adjustment is <100ms, the accuracy rate of key feature identification is >92%, and it also supports mixed decision of expert rules and data-driven, such as realizing automatic promotion of weight of infiltration flow and wetting line pressure to 0.45.

[0072] 4. The present application uses multi-module coupling evaluation, and the efficiency is high: the multi-module collaborative evaluation error is reduced by 40%-60% compared with single model, and the extreme working condition deduction speed is increased by 5 times. BRIEF DESCRIPTION OF DRAWINGS

[0073] In order to make the technical solutions in the embodiments of the present application or the prior art clearer, the accompanying drawings needed in the embodiments or prior art description will be briefly introduced. Obviously, the accompanying drawings in the following description only only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor based on the drawings shown.

[0074] Figure 1 The system block diagram of the small and medium-sized reservoir operation safety evaluation system based on sky-ground-water engineering according to an embodiment of the present application;

[0075] Figure 2 The processing flow schematic diagram of the short-term risk prediction module according to an embodiment of the present application;

[0076] Figure 3 The processing flow schematic diagram of the medium and long-term stability prediction module according to an embodiment of the present application;

[0077] Figure 4 The processing flow schematic diagram of the extreme working condition deduction module according to an embodiment of the present application. DETAILED DESCRIPTION

[0078] The specific embodiments of the present application will be described in detail below.

[0079] The present application provides a small and medium-sized reservoir operation safety evaluation system based on sky-ground-water engineering, comprising a time-space alignment layer, a feature weighting layer and a dynamic coupling layer connected in sequence;

[0080] The time-space alignment layer performs time alignment and space alignment on the input sky-ground-water engineering data;

[0081] The feature weighting layer quantifies the contribution of different monitoring features to safety evaluation according to the time-space aligned data;

[0082] The dynamic coupling layer adopts a dynamic weight distribution strategy for different modules through a multi-module coordination engine, fuses the feature weighting, obtains a fusion result, and outputs a small and medium-sized reservoir operation safety index according to the fusion result, wherein the multi-module includes a short-term risk prediction module, a medium and long-term stability prediction module and an extreme working condition deduction module; the short-term risk prediction module is used for safety evaluation of the dam within a first time of real time or preset; the medium and long-term stability prediction module evaluates the stability of the dam within a preset year; and the extreme working condition deduction module evaluates the structural safety under extreme conditions by simulating the influence of an overload event on the dam body.

[0083] Furthermore, time alignment of the input data specifically involves aligning the time series by dynamically bending the time axis for data with different sampling frequencies, and using cubic spline interpolation for low-frequency data below a preset frequency threshold to generate preset high-frequency time point data.

[0084] Furthermore, dynamic time warping is employed for data with different sampling frequencies, as detailed in the following formula:

[0085]

[0086] in,

[0087] For the data to be aligned m Weighting coefficients for each time point;

[0088] This is the time decay coefficient;

[0089] For the reference time series m A point in time;

[0090] The time point for the data to be aligned;

[0091] These are the time points used as a reference time series.

[0092] This represents the total number of time points in the reference sequence.

[0093] Furthermore, for low-frequency data below a preset frequency threshold, the following formula is used to generate preset high-frequency time point data:

[0094]

[0095] in,

[0096] , , , The coefficients of the cubic polynomial are determined by cubic spline interpolation, ensuring that the interpolation function is within the interval [ ]. , It is continuous within the interval and its first and second derivatives are continuous;

[0097] The target time point to be interpolated is located in the interval [ , ]Inside;

[0098] This refers to the previous time point in the original low-frequency data;

[0099] the original low-frequency data at the time point adjacent to the previous time point;

[0100] the estimated value obtained by cubic spline interpolation at time .

[0101] Further, the cubic spline interpolation condition is determined , , , , so as to realize smooth high-frequency data interpolation, , , , The calculation method is as follows:

[0102] When t = , the formula is substituted to obtain d = X ( t k ), that is, d is the value of the original low-frequency data at , and then a three-moment equation is constructed to solve . The cubic spline interpolation requires that the second derivative is continuous, and the second derivative value of each node is obtained by the three-moment equation ( is the second derivative value at , and the interval length );

[0103] Then, the following formula is used to calculate , , respectively:

[0104]

[0105]

[0106] The quadratic term coefficient is directly determined by the second derivative value at t k :

[0107]

[0108] The linear term coefficient is determined in combination with the interval function value difference and the second derivative value.

[0109] Further, the following method is used for spatial alignment of data: all data are converted to the same coordinate system, deformation is eliminated by using Gauss-Krueger projection, and continuous spatial distribution is generated by using a spatial interpolation kernel function for discrete monitoring point data.

[0110] Further, the spatial interpolation kernel function is as follows:

[0111]

[0112] wherein,

[0113] is a normalized spatial distance;

[0114] is a spatial interpolation weight.

[0115] Further, the feature weighting layer adopts feature selection based on attention mechanism, and dynamically adjusts the feature weight by combining context-aware attention with environmental working conditions.

[0116] Further, the features in the sky level are compressed in weight by the following formula, and the original features are reduced in dimension:

[0117]

[0118] wherein,

[0119] is a PCA projection matrix, which is obtained by eigen decomposition of the covariance matrix of the original data;

[0120] is the transpose of the aligned th original feature vector, is the total number of original features;

[0121] is the l th compressed feature in the sky level, k represents the dimension after dimension reduction.

[0122] Weight compression is to reduce the dimension of high-dimensional features through mathematical transformation, reducing the feature dimension while retaining key information, so as to more efficiently perform subsequent analysis (such as weight allocation).

[0123] Further, composed of a number of principal components (feature vectors) with the largest variance in the data, used to map high-dimensional original features to low-dimensional space, achieving dimension reduction.

[0124] Further, The calculation method is as follows:

[0125] Standardize the original feature data;

[0126] Calculate the covariance matrix of the standardized data;

[0127] The covariance matrix is ​​decomposed into eigenvalues ​​and eigenvectors.

[0128] Sort the eigenvalues ​​from largest to smallest, and select the eigenvectors corresponding to the first predetermined number of largest eigenvalues ​​to form... .

[0129] Furthermore, the raw data includes features such as surface deformation of the sky monitoring layer, thermal infrared anomalies, and the width of cracks caused by drones.

[0130] The original features at the sky level (such as satellite surface deformation values, UAV crack widths, etc.) have high dimensionality, and can be obtained through W l PCA Projection can compress these features into low-dimensional principal components (such as...) This reduces computational complexity while preserving key information.

[0131] Furthermore, across the context vectors of sky, ground, and environment, features are weighted and compressed using the following formula:

[0132]

[0133] in,

[0134] For the first j Attention weights at different levels; used for weight allocation between the sky layer (such as satellites / drones) and the ground layer (such as sensors);

[0135] For the environment context vector;

[0136] For the first j Hierarchical compression features; for example, principal components of surface deformation and temporal characteristics of seepage flow after dimensionality reduction;

[0137] The summation index ranges from 1 to 4, corresponding to different levels; the denominator normalizes the weights by summing the exponents of these four levels.

[0138] This is a learnable weight vector used to calculate the score in the attention mechanism. The weights are determined by multiplying the weights by the concatenated vector and then taking the exponent. yes transpose;

[0139] It is the concatenation operator, which concatenates... and By concatenating the elements along the dimensions, a comprehensive vector containing hierarchical features and environmental information is formed.

[0140] Furthermore, the environmental context vector contains environmental information such as real-time reservoir water level (unit: m) and rainfall intensity (unit: mm / h).

[0141] Furthermore, the feature weighting layer adopts a hybrid static and dynamic weighting guided by expert knowledge. The static weights are preset based on expert experience, while the dynamic weights are adjusted according to real-time operating conditions.

[0142] Furthermore, the short-term prediction module uses the LSTM-Transformer model to mine local and global features in reservoir operation data. The LSTM-Transformer model includes an LSTM layer, a Transformer self-attention layer, and a fully connected layer. The LSTM layer captures the temporal relationship of data features, the Transformer self-attention layer identifies the spatial attention weights between multi-sensor data features, and the fully connected layer uses the Sigmoid activation function to output the dam failure probability.

[0143] Furthermore, the medium- and long-term stability prediction module constructs a physical-guided neural network architecture, inputs low-frequency cumulative data and environmental parameters, uses the embedded constitutive equation of the earth-rock dam as a physical constraint, and uses the nonlinear relationship between historical monitoring data and geological parameters learned by the multilayer perceptron as a data driver.

[0144] Furthermore, the extreme working condition simulation module constructs a three-dimensional model of the dam body, foundation, and reservoir area based on BIM, integrates geometric models, material properties, and safety boundary conditions, and establishes different parametric scenario libraries. Then, it performs coupled solution of multiple scenarios to obtain the pore water pressure distribution, updates the effective stress field, and iteratively solves displacement and strain based on the elastoplastic constitutive equation to extract the safety factor FS.

[0145] Furthermore, the pore water pressure distribution is solved based on the finite element method (FEM). The three-dimensional model of dam body-foundation-reservoir area is discretized into elements, and the seepage control equation is established. For steady seepage, the Laplace equation is satisfied; for unsteady seepage, the diffusion equation is satisfied.

[0146] Further, the solution is obtained through the following steps:

[0147] Discretization: Divide the data into units and derive the penetration matrix for each unit.

[0148] Boundary conditions: Set upstream and downstream water levels (known head), impermeable boundary (zero flow rate), etc.

[0149] Assembly equation: The integration unit penetration matrix is ​​the overall matrix, forming a system of algebraic equations. K h ]{ h}={ Q}([ K his the total permeability matrix, Q} is the flow vector);

[0150] Solving: solve the equation group by numerical method (such as Gaussian elimination method) to obtain the water head of each node h , and then calculate the pore water pressure by u = γ w h ( γ w is the specific weight of water u .

[0151] Further, the geometric model includes the geometric model of the terrain, the dam body structure and the flood discharge facility, the material attribute includes the constitutive parameter of the soil and stone material, the time-varying characteristic of the permeability coefficient, and the safety boundary condition includes the reservoir water pressure, the earthquake input and the seepage boundary.

[0152] Further, the dynamic weight distribution strategy defines a rule base through a fuzzy logic controller, realizes the adaptive weight of different models, and updates the feature weighting parameter in real time based on the evaluation error.

[0153] Further, if the reservoir water level is greater than the warning water level, then the weight of the short-term prediction module is 0.8; if the earthquake intensity is greater than or equal to 6, then the weight of the extreme working condition deduction module is 0.7; if the reservoir water level is over high and the rainfall intensity is storm, then the weight of the short-term prediction module is 0.8, the weight of the medium and long-term stability prediction module is 0.1, and the weight of the extreme working condition deduction module is 0.1.

[0154] Further, after the output result, the predicted result is compared with the actual monitoring data, the model retraining in each module is triggered, the new monitoring data is injected into the training set, the feature weighting parameter is updated, and the adaptive correction equation based on the model parameter adjustment is as follows:

[0155]

[0156] wherein,

[0157] is an index, used to identify the th data point or the th monitoring index;

[0158] is the th original monitoring data;

[0159] is the th predicted value based on Kalman filtering;

[0160] σ is the standard deviation of historical data;

[0161] for the first modified sensor data value;

[0162] for a weighting factor that balances the original data with the predicted value when the data deviation exceeds 3σ.

[0163] Embodiment 1

[0164] The application provides a small and medium-sized reservoir operation safety evaluation system based on sky-ground-water engineering, comprising a time-space alignment layer, a feature weighting layer and a dynamic coupling layer connected in sequence;

[0165] The time-space alignment layer performs time alignment and space alignment on the input sky-ground-water engineering data;

[0166] The feature weighting layer quantifies the contribution of different monitoring features to safety evaluation according to the time-space aligned data;

[0167] The dynamic coupling layer adopts a dynamic weight distribution strategy for different modules through a multi-module coordination engine, fuses the feature weighting, obtains a fusion result, and outputs a small and medium-sized reservoir operation safety index according to the fusion result, wherein the multi-module includes a short-term risk prediction module, a medium and long-term stability prediction module and an extreme working condition deduction module; the short-term risk prediction module is used for safety evaluation of the dam within a first time of real time or preset; the medium and long-term stability prediction module evaluates the stability of the dam within a preset year; and the extreme working condition deduction module evaluates the structural safety under extreme conditions by simulating the influence of an overload event on the dam body.

[0168] Embodiment 2

[0169] The application provides a small and medium-sized reservoir operation safety evaluation system based on sky-ground-water engineering, comprising a time-space alignment layer, a feature weighting layer and a dynamic coupling layer connected in sequence;

[0170] The time-space alignment layer performs time alignment and space alignment on the input sky-ground-water engineering data;

[0171] The feature weighting layer quantifies the contribution of different monitoring features to safety evaluation according to the time-space aligned data;

[0172] The dynamic coupling layer, through a multi-module coordination engine, employs a dynamic weight allocation strategy to fuse features using weighted averages, obtaining a fusion result. Based on the fusion result, it outputs a safety index for the operation of small and medium-sized reservoirs. The multi-module includes a short-term risk prediction module, a medium-to-long-term stability prediction module, and an extreme condition simulation module. The short-term risk prediction module is used for real-time or preset first-time safety assessment of the dam. The medium-to-long-term stability prediction module assesses the stability of the dam within a preset number of years. The extreme condition simulation module assesses the structural safety under extreme conditions by simulating the impact of overload events on the dam body.

[0173] Furthermore, time alignment of the input data specifically involves aligning the time series by dynamically bending the time axis for data with different sampling frequencies, and using cubic spline interpolation for low-frequency data below a preset frequency threshold to generate preset high-frequency time point data.

[0174] Furthermore, dynamic time warping is employed for data with different sampling frequencies, as detailed in the following formula:

[0175]

[0176] in,

[0177] For the data to be aligned m Weighting coefficients for each time point;

[0178] This is the time decay coefficient;

[0179] For the reference time series m A point in time;

[0180] The time point for the data to be aligned;

[0181] These are the time points used as a reference time series.

[0182] This represents the total number of time points in the reference sequence.

[0183] Furthermore, for low-frequency data below a preset frequency threshold, the following formula is used to generate preset high-frequency time point data:

[0184]

[0185] in,

[0186] , , , are coefficients of cubic polynomial determined by cubic spline interpolation, ensuring that the interpolated function is continuous and the first derivative and the second derivative are continuous within the interval [ , ];

[0187] is a target time point to be interpolated, located within the interval [ , ];

[0188] is a previous time point in the original low-frequency data;

[0189] is a next time point adjacent to in the original low-frequency data;

[0190] is an estimated value obtained by cubic spline interpolation at time .

[0191] Further, the cubic spline interpolation condition determines , , , , so as to achieve smooth high-frequency data interpolation, , , , The calculation method is as follows:

[0192] When t = , substituting the formula gives d = X ( t k ), that is, d is the value of the original low-frequency data at , and then a three-moment equation is constructed to solve The cubic spline interpolation requires the second derivative to be continuous, and the second derivative value of each node is obtained by the three-moment equation ( is the second derivative value at , and the interval length );

[0193] The following formula is used to calculate , , respectively:

[0194]

[0195]

[0196] Byt k The second derivative value at the point directly determines the quadratic term coefficient:

[0197]

[0198] The first term coefficient is determined by combining the interval function value difference and the second derivative value.

[0199] Further, the spatial alignment of data is performed by the following method: converting all data to the same coordinate system, eliminating deformation by Gauss-Krueger projection, and generating continuous spatial distribution by using a spatial interpolation kernel function on discrete monitoring point data.

[0200] Further, the spatial interpolation kernel function is as follows:

[0201]

[0202] wherein,

[0203] is the normalized spatial distance;

[0204] is the spatial interpolation weight.

[0205] Further, the feature weighting layer adopts feature selection based on the attention mechanism, and dynamically adjusts the feature weight by combining the context-aware attention and the environmental working condition.

[0206] Further, the features in the sky level are compressed in weight by the following formula, and the original features are reduced in dimension:

[0207]

[0208] wherein,

[0209] is the PCA projection matrix, which is obtained by eigen decomposition on the covariance matrix of the original data;

[0210] is the transpose of the aligned first original feature vector, is the total number of original features;

[0211] is the compressed feature in the sky level, l k represents the dimension after dimension reduction.

[0212] Weight compression is to reduce the dimension of high-dimensional features by mathematical transformation, while retaining key information, so as to more efficiently perform subsequent analysis (such as weight allocation).

[0213] ​Further, The several principal components (eigenvectors) with the largest variance from the data are used to map high-dimensional original features to low-dimensional space, realizing dimension reduction.

[0214] Further, The calculation method is as follows:

[0215] The original feature data is standardized;

[0216] The covariance matrix of the standardized data is calculated;

[0217] The covariance matrix is decomposed to obtain the eigenvalues and eigenvectors;

[0218] The eigenvalues are sorted from large to small, and the eigenvectors corresponding to the first preset number of largest eigenvalues are selected to form .

[0219] Further, the original data includes features such as ground deformation, thermal infrared anomaly, and unmanned aerial vehicle crack width in the sky monitoring layer.

[0220] The original features of the sky layer (such as satellite ground deformation values, unmanned aerial vehicle crack width, etc.) have high dimensions, and through W l PCA Projection, these features can be compressed into low-dimensional principal components (such as ), while retaining the main information and reducing the computational complexity.

[0221] Further, in the cross-layer of the sky + ground + environment context vector, the features are compressed by weight using the following formula:

[0222]

[0223] Wherein,

[0224] is the attention weight of the j layer; used for weight distribution between the sky layer (such as satellite / unmanned aerial vehicle) and the ground layer (such as sensor) etc.

[0225] is the environmental context vector;

[0226] is the compressed feature of the j layer; for example, the principal component of ground deformation after dimension reduction, time series features of infiltration capacity, etc.

[0227] is the summation index, with a value range of 1 to 4, corresponding to different layers; the denominator is obtained by summing the exponential terms of the four layers, realizing the normalization of the weight;

[0228] is a learnable weight vector for calculating the score in the attention mechanism, which determines the weight by multiplying the concatenated vector and taking the exponential is the transpose of

[0229] is a concatenation operator that concatenates and by dimension to form a comprehensive vector containing hierarchical features and environmental information.

[0230] Further, the environmental context vector contains real-time reservoir water level (unit: m), rainfall intensity (unit: mm / h), and other environmental information.

[0231] Further, the feature weighting layer adopts a static and dynamic hybrid weighting guided by expert knowledge. The static weight is preset based on expert experience, and the dynamic weight is adjusted according to real-time working conditions.

[0232] Further, the short-term prediction module uses the LSTM-Transformer model to mine local and global features in reservoir operation data. The LSTM-Transformer model includes LSTM layers, Transformer self-attention layers, and fully connected layers. The LSTM layers capture the time sequence relationship of data features, the Transformer self-attention layers identify the spatial attention weights between multi-sensor data features, and the fully connected layers output the dam failure probability using the Sigmoid activation function.

[0233] Further, the medium and long-term stability prediction module constructs a physically guided neural network architecture, inputs low-frequency cumulative data and environmental parameters, embeds the constitutive equation of earth-rockfill dams as a physical constraint, and learns the nonlinear relationship between historical monitoring data and geological parameters using a multi-layer perceptron as data-driven.

[0234] Further, the extreme working condition deduction module constructs a dam-foundation-reservoir three-dimensional model based on BIM, integrates geometric models, material properties, and safety boundary conditions, and establishes different parameterized scene libraries. Then, it performs coupled solving of multiple scenes to obtain the pore water pressure distribution, updates the effective stress field, iteratively solves the displacement and strain based on the elastoplastic constitutive equation, and extracts the safety factor FS.

[0235] Further, the solution of the pore water pressure distribution is based on the finite element method (FEM), which discretizes the dam-foundation-reservoir three-dimensional model into elements and establishes seepage control equations. For stable seepage, it satisfies the Laplace equation, and for non-stable seepage, it satisfies the diffusion equation.

[0236] Further, the solution is obtained by the following steps:

[0237] Discretization: divide the domain into elements, and derive the permeability matrix for each element.

[0238] Boundary conditions: set upstream and downstream water levels (known water heads), impermeable boundaries (zero flux), etc.

[0239] Assembly of equations: integrate the element permeability matrices into a global matrix, forming a system of algebraic equations K h ]{ h}={ Q}([( K h ] is the total permeability matrix, and Q} is the flux vector);

[0240] Solution: solve the system of equations by numerical methods (e.g., Gaussian elimination) to obtain the water heads at each node h , and then calculate the pore water pressure u γ w h ( γ w is the unit weight of water). u .

[0241] Further, the geometric model includes the terrain, dam structure, and spillway facility geometric models, material properties include soil and rock constitutive parameters, time-varying characteristics of permeability coefficients, and safety boundary conditions include reservoir water pressure, seismic input, and seepage boundary.

[0242] Further, the dynamic weight distribution strategy defines a rule base through a fuzzy logic controller, realizes adaptive weights for different models, and updates the feature weighting parameters in real time based on evaluation errors.

[0243] Further, if the reservoir water level > alarm water level, then the weight of the short-term prediction module = 0.8; if the earthquake intensity ≥ 6, then the weight of the extreme case deduction module = 0.7; if the reservoir water level is super high and the rainfall intensity is heavy rain, then the weight of the short-term prediction module = 0.8, the weight of the medium and long-term stability prediction module = 0.1, and the weight of the extreme case deduction module = 0.1.

[0244] Further, after outputting the results, compare the prediction results with the actual monitoring data, trigger model retraining in each module, inject new monitoring data into the training set, update the feature weighting parameters, and adjust the adaptive correction equation based on the model parameters as follows:

[0245]

[0246] where

[0247] is the index, used to identify the​ data points or the monitoring indicators;

[0248] the original monitoring data;

[0249] the prediction value based on Kalman filtering;

[0250] σ the standard deviation of historical data;

[0251] the corrected sensor data value;

[0252] the weighting coefficient, when the data deviation exceeds 3σ, the weights of the original data and the prediction value are balanced.

[0253] Embodiment 3

[0254] The application provides a small and medium-sized reservoir operation safety evaluation system based on sky-ground-water engineering:

[0255] (1) System architecture

[0256] The application adopts a hierarchical progressive fusion strategy, including a time-space alignment layer, a feature weighting layer and a dynamic coupling layer connected in turn; as shown in Figure 1 ;

[0257] The time-space alignment layer performs time alignment and space alignment on the input sky-ground-water engineering data.

[0258] The feature weighting layer quantifies the contribution of different monitoring features to safety evaluation according to the time-space aligned data.

[0259] The dynamic coupling layer adopts a dynamic weight distribution strategy for different modules through a multi-module coordination engine, fuses the feature weighting, obtains a fusion result, and outputs a small and medium-sized reservoir operation safety index according to the fusion result, wherein the multi-module includes a short-term risk prediction module, a medium and long-term stability prediction module and an extreme working condition deduction module; the short-term risk prediction module is used for safety evaluation of the dam within a first time of real time or preset; the medium and long-term stability prediction module evaluates the stability of the dam within a preset year; and the extreme working condition deduction module evaluates the structural safety under extreme conditions by simulating the influence of an overload event on the dam body.

[0260] (2) Fusion of multi-source data

[0261] 2.1, multi-source data input

[0262] This invention uses multi-source data input, as shown in Table 1.

[0263] Table 1. Input Multi-Source Data Table

[0264]

[0265] 2.2 Spatiotemporal Alignment Processing

[0266] The spatiotemporal alignment layer addresses the inconsistencies in time and space between multi-source heterogeneous monitoring data, and constructs a unified data benchmark.

[0267] Regarding time alignment:

[0268] For data with different sampling frequencies, time series matching is achieved by dynamically bending the time axis. For example, dynamic time warping (DTW) is used to generate minute-level continuous datasets by setting the time decay coefficient γ=0.3 for satellite data (6 hours / sample) and sensor data (5 minutes / sample).

[0269] The Dynamic Time Warping (DTW) calculation is as follows:

[0270]

[0271] in, For the data to be aligned m Weighting coefficients for each time point;

[0272] This is the time decay coefficient; For the reference time series m A point in time; The time point for the data to be aligned;

[0273] These are the time points used as a reference time series. This represents the total number of time points in the reference sequence.

[0274] Table 2. Input Data Table for Dynamic Time Warping

[0275]

[0276] For low-frequency data (such as satellite data) below a preset frequency threshold, the following formula is used to generate preset high-frequency time point data, employing time interpolation based on a cubic polynomial:

[0277]

[0278] in,

[0279] , , , are coefficients of cubic polynomial, determined by cubic spline interpolation, ensuring that the interpolation function is continuous and the first derivative and the second derivative are continuous within the interval [ , ];

[0280] is the target time point to be interpolated, located within the interval [ , ];

[0281] is the previous time point in the original low-frequency data;

[0282] is the next time point adjacent to in the original low-frequency data;

[0283] is the estimated value obtained by cubic spline interpolation at time .

[0284] Further, the cubic spline interpolation condition determines , , , , so as to achieve smooth high-frequency data interpolation, , , , The calculation method is as follows:

[0285] When t = , substituting the formula gives d = X ( t k ), that is, d is the value of the original low-frequency data at , and then a three-moment equation is constructed to solve The cubic spline interpolation requires the second derivative to be continuous, and the second derivative value of each node is obtained by the three-moment equation ( is the second derivative value at , and the interval length );

[0286] Then, the following formula is used to calculate , , respectively:

[0287]

[0288]

[0289] By t k The quadratic term coefficient is determined directly by the second derivative value at the point:

[0290]

[0291] The linear term coefficient is determined by combining the interval function value difference and the second derivative value.

[0292] In space, it is:

[0293] All data is converted to the same coordinate system (such as WGS-84), and the Gauss-Kruger projection is used to eliminate deformation to achieve the unification of geographic coordinates. The spatial interpolation kernel function is used to generate continuous spatial distribution for discrete monitoring point data. For example, for dam seepage pressure data, discrete piezometer data is converted into a two-dimensional distribution map of the dam body phreatic line with a grid precision of 0.5m x 0.5m. For abnormal data processing, the Mahalanobis distance-based abnormal data recognition method is used for labeling processing.

[0294] The spatial interpolation kernel function is calculated as follows:

[0295]

[0296] Table 3. Spatial interpolation kernel function input data table

[0297]

[0298] The feature weighting layer mainly quantifies the contribution of different monitoring features to safety evaluation and realizes dynamic weight allocation. Two methods are mainly used, the first of which is based on attention mechanism feature selection, which dynamically adjusts feature weights by combining context-aware attention with environmental conditions (such as reservoir water level, rainfall intensity).

[0299] Feature weight compression in the sky level:

[0300] The features in the sky level are compressed in weight using the following formula, which reduces the dimensionality of the original features:

[0301]

[0302] where,

[0303] is the PCA projection matrix, which is obtained by eigen decomposition of the covariance matrix of the original data;

[0304] is the transpose of the aligned th original feature vector, is the total number of original features;

[0305] is the l th compressed feature in the sky level, k represents the dimension after dimension reduction.

[0306] Table 4. Sky level feature dimension reduction input data table

[0307]

[0308] In the cross-level of the sky + ground + environment context vector, the features are weighted compressed as follows:

[0309]

[0310] where,

[0311] is the attention weight of the j th level; used for weight distribution between the sky level (such as satellite / unmanned aerial vehicle) and the ground level (such as sensor) and the like;

[0312] is the environmental context vector;

[0313] is the compressed feature of the j th level; for example, the principal component of ground deformation after dimension reduction, time series feature of infiltration capacity and the like;

[0314] is the summation index, taking the value range of 1 to 4, corresponding to different levels; the denominator is obtained by summing the exponential terms of the four levels, realizing the normalization of the weight;

[0315] is the learnable weight vector, used to calculate the score in the attention mechanism, and the weight is determined by multiplying the spliced vector and taking the exponential is the transpose of

[0316] is the splicing operator, which splices and by dimension to form a comprehensive vector containing level features and environmental information.

[0317] Table 5. Sky + ground + environment context vector dimension reduction input data

[0318]

[0319] The second kind adopts a hybrid weighting guided by expert knowledge. The static weight is preset with a basic weight according to expert experience (such as infiltration capacity weight 0.3, displacement weight 0.4), and the dynamic correction is adjusted by real-time working conditions (such as when the earthquake is triggered, the vibration frequency spectrum weight is improved).​

[0320] The dynamic coupling layer is an effective fusion of feature weighting, which uses a dynamic weight distribution strategy for different modules through a multi-module coordination engine to fuse the feature weighting and obtain a fusion result, and outputs a small and medium-sized reservoir operation safety index according to the fusion result to realize a dynamic iteration of risk assessment, wherein the multi-module includes a short-term risk prediction module, a medium and long-term stability prediction module, and an extreme working condition deduction module; the short-term risk prediction module is used for safety assessment of the dam within a first time of real time or pre-setting, the medium and long-term stability prediction module is used for assessment of the stability of the dam within a pre-set year, and the extreme working condition deduction module is used for assessment of the structural safety under extreme conditions by simulating the influence of an overload event on the dam body.

[0321] Table 6. List of dynamic coupling layer processing ideas

[0322]

[0323] The short-term risk prediction module is used for dynamic assessment of the safety risk of the earth and rockfill dam in real time or in the near future, can quickly identify sudden hazards, predict the dam failure probability and weak parts within future hours, and in the embodiment, an LSTM-Transformer model is built to fully excavate local and global features in the reservoir operation data, accurately capture the change law of indexes such as water level and flow with time and the complex correlation therebetween, and predict the future short-term (such as 1-3 days in the future) operation state of the reservoir.

[0324] The LSTM-Transformer model is composed of an LSTM layer, a Transformer self-attention layer, and an output layer, wherein the LSTM layer captures the time-dependent relationship (such as continuous change of displacement and seepage flow), the Transformer self-attention layer identifies the spatial correlation between multi-sensor data (such as linkage between crack propagation and rising of the wetting line), and the output layer outputs the dam failure probability (0-1) by using a Sigmoid activation function. The running process is as shown in Figure 2 .

[0325] The medium and long-term stability prediction module is used for assessment of the stability of the dam within a time scale of several years or even decades, and focuses on material aging, geological subsidence, and progressive risks such as seepage erosion. In the embodiment, a PINNs model is built to predict the future medium and long-term (such as 1 month to 2 years in the future) operation state of the reservoir. In the construction of the physical guided neural network (PINNs) architecture design, the constitutive equation of the earth and rockfill dam is embedded as a physical constraint, the multi-layer perceptron (MLP) learns the nonlinear relationship between the historical monitoring data and the geological parameters as data-driven, and the data fitting error and the physical equation residual (weight λ=1.0) are used as the loss function. The running process is as shown in Figure 3 .

[0326] The extreme working condition deduction module assesses the structural safety under extreme conditions by simulating the impact of rare or extreme load events (such as the possible maximum flood, VIII earthquake) on the dam body, and generates emergency plans. In this embodiment, by building a digital twin engine for the reservoir project, a three-dimensional refined model of the dam-foundation-reservoir area is constructed based on BIM, integrating geometric models of topography, dam structure, flood discharge facilities, etc., material properties such as soil and rock constitutive parameters, time-varying characteristics of permeability coefficients, and boundary conditions such as reservoir water pressure, seismic input, seepage boundary, etc., and establishing different parameterized scene libraries, as shown in the following table, then performing multi-scene coupling solution, solving the pore water pressure distribution, updating the effective stress field, based on the elastic-plastic constitutive equation, iteratively solving displacement and strain, etc., finally extracting the safety factor FS, as well as the maximum principal stress, strain distribution cloud map, critical sliding surface three-dimensional coordinates, emergency flood discharge scheme recommendation, etc. The running process is as shown in Figure 4 .

[0327] Table 7. Extreme working condition parameter setting and use example table

[0328]

[0329] (3) Dynamic weight allocation strategy

[0330] For the dynamic weight allocation strategy, the rule base is defined by the fuzzy logic controller to realize model weight self-adaptation, and the weight is updated in real time based on the evaluation error. In this embodiment:

[0331] IF reservoir water level > warning water level, TNHE short-term model weight = 0.8

[0332] IF earthquake intensity ≥ 6, TNHE extreme working condition model weight = 0.7

[0333] IF reservoir water level exceeds high AND rainfall intensity is heavy rain, TNHE short-term module weight = 0.8, medium and long term = 0.1, extreme = 0.1

[0334] Then the feedback optimization loop checks the evaluation results, compares the model prediction with the actual monitoring data, triggers model retraining, and injects new monitoring data into the training set to update the feature weighting parameters.

[0335] Adaptive correction equation based on model parameter adjustment:

[0336]

[0337] where,

[0338] is the index, used to identify the th data point or the th monitoring index;

[0339] is the first original monitoring data;

[0340] is the first prediction value based on Kalman filter;

[0341] σ is the historical data standard deviation;

[0342] is the first corrected sensor data value;

[0343] is the weighting coefficient, when the data deviation exceeds 3σ, the weight of the original data and the prediction value is balanced.

[0344] Table 7. Adaptive correction equation input data list

[0345]

[0346] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. The present application can have various modifications and changes for those skilled in the art. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application is included in the protection scope of the present application.

Claims

1. A small and medium-sized reservoir operation safety evaluation system based on sky-ground-water engineering, characterized by, The method comprises a space-time alignment layer, a feature weighting layer and a dynamic coupling layer connected in sequence. The space-time alignment layer performs time alignment and space alignment on the input sky-ground-water data; the time alignment of the input data is specifically that, for data of different sampling frequencies, time series alignment is achieved by a dynamic bending time axis, and for low-frequency data below a preset frequency threshold, cubic spline interpolation is used to generate preset high-frequency time point data. The feature weighting layer quantifies the contribution of different monitoring features to safety evaluation according to the data after space-time alignment. The feature weighting layer adopts feature selection based on an attention mechanism, dynamically adjusts feature weights by combining context-aware attention with environmental working conditions. Or the feature weighting layer adopts a static and dynamic hybrid weighting guided by expert knowledge, the static weight is a preset basic weight according to expert experience, and the dynamic weight is adjusted according to real-time working conditions. The dynamic coupling layer uses a multi-module coordination engine to adopt a dynamic weight distribution strategy for different modules, fuses the results of different modules of the feature weighting, obtains a fusion result, and outputs a medium and small-sized reservoir operation safety index according to the fusion result, wherein the multi-module includes a short-term risk prediction module, a medium and long-term stability prediction module and an extreme working condition deduction module; the short-term risk prediction module is used for safety evaluation of the dam within a first time of real time or preset; the medium and long-term stability prediction module evaluates the stability of the dam within a preset year; and the extreme working condition deduction module evaluates the structural safety under extreme conditions by simulating the influence of an overload event on the dam body. The short-term prediction module uses an LSTM-Transformer model to mine local features and global features in the reservoir operation data, the LSTM-Transformer model includes an LSTM layer, a Transformer self-attention layer and a full connection layer, the LSTM layer captures the time sequence relationship of data features, the Transformer self-attention layer identifies the spatial attention weight between multi-sensor data features, and the full connection layer outputs a dam breach probability using a Sigmoid activation function. The medium and long-term stability prediction module constructs a physically guided neural network architecture, inputs low-frequency cumulative data and environmental parameters, embeds the constitutive equation of earth-rockfill dam as a physical constraint, and learns the nonlinear relationship between historical monitoring data and geological parameters by a multilayer perceptron as data-driven; The extreme working condition deduction module constructs a dam-foundation-reservoir three-dimensional model based on BIM, integrates geometric models, material properties and safety boundary conditions, establishes different parameterized scene libraries, and then performs coupled solving of multiple scenes to obtain pore water pressure distribution, update effective stress field, iteratively solve displacement and strain based on elastic-plastic constitutive equation, and extract safety factor FS.

2. The sky-ground-water-based small and medium-sized reservoir operation safety assessment system according to claim 1, characterized in that, For data of different sampling frequencies, dynamic time warping is used, and the specific formula is as follows: wherein, a weight coefficient for a time point of the data to be aligned; and m a weight coefficient for a time point of the data to be aligned; and is the time decay coefficient; for a reference time series of the m th time point; a time point for data to be aligned; t is a time point of the reference time series; is the total number of time points for the reference sequence.

3. The sky-ground-water-based small and medium-sized reservoir operation safety assessment system according to claim 1, characterized in that, For low-frequency data below a preset frequency threshold, the following formula is used to generate preset high-frequency time point data: wherein, , , , are coefficients of a cubic polynomial determined by cubic spline interpolation, ensuring that the interpolated function is continuous and the first derivative, the second derivative is continuous within the interval [ , ]. is the target time point to be interpolated, which lies within the interval [ , ] ; is the previous time point in the original low frequency data; the original low frequency data at the time point adjacent to the next time point; The estimate at time t is obtained by cubic spline interpolation of the values. The estimate at time t is obtained by cubic spline interpolation of the values.

4. The sky-ground-water-based small and medium-sized reservoir operation safety assessment system according to claim 1, characterized in that, The following method is used for spatial alignment of data: all data are converted to the same coordinate system, Gaussian-Krueger projection is used to eliminate deformation, and a spatial interpolation kernel function is used to generate continuous spatial distribution for discrete monitoring point data.

5. The sky-ground-water-based small and medium-sized reservoir operation safety assessment system according to claim 4, characterized in that, The spatial interpolation kernel function is as follows: wherein, is the normalized spatial distance; Spatial interpolation weights.

6. The sky-ground-water-based small and medium-sized reservoir operation safety assessment system according to claim 1, characterized in that, The features in the sky level are compressed in weight by the following formula, and the original features are reduced in dimension: wherein, is the PCA projection matrix, obtained by eigen decomposition of the covariance matrix of the original data; is the transpose of the aligned original eigenvector, is the transpose of the aligned original eigenvector, is the total number of original eigenvectors; is the number of compressed features in the sky level l k represents the dimensionality after dimensionality reduction.​ 7. The sky-ground-water-based small and medium-sized reservoir operation safety assessment system according to claim 1, characterized in that, The geometric model includes the geometric models of the terrain, dam structure, and flood discharge facilities, the material properties include the constitutive parameters of soil and stone materials and the time-varying characteristics of the permeability coefficient, and the safety boundary conditions include the reservoir water pressure, seismic input, and seepage boundary.

8. The sky-ground-water-based small and medium-sized reservoir operation safety assessment system according to claim 1, characterized in that, The dynamic weight distribution strategy defines a rule base through a fuzzy logic controller, realizes the adaptive weight of different models, and updates the feature weighting parameters in real time based on the evaluation error.

9. The sky-ground-water-based small and medium-sized reservoir operation safety assessment system according to claim 8, characterized in that, If the reservoir water level > warning water level, then the weight of the short-term prediction module = 0.8; if the seismic intensity ≥ 6, then the weight of the extreme case deduction module = 0.7; if the reservoir water level is over high and the rainfall intensity is storm, then the weight of the short-term prediction module = 0.8, the weight of the medium and long-term stability prediction module = 0.1, and the weight of the extreme case deduction module = 0.

1.

10. The sky-ground-water-based small and medium-sized reservoir operation safety assessment system according to claim 8, characterized in that, After the output result, the prediction result is compared with the actual monitoring data, the model in each module is triggered for retraining, the new monitoring data is injected into the training set, the feature weighting parameters are updated, and the adaptive correction equation based on the model parameter adjustment is as follows: wherein, Index, used to identify the data point or the monitoring indicator; for the first original monitoring data; For the Kalman filter-based first prediction value; σ is the historical data standard deviation; the first modified sensor data value; and the second modified sensor data value. For the weighting factor, when the data deviation exceeds 3σ, the original data is balanced with the predicted value weight.

Images