Intelligent early warning and fault diagnosis method for diversion desilting and flushing system

Abstract

This invention relates to the field of intelligent monitoring and fault diagnosis technology for water conservancy projects, and in particular to an intelligent early warning and fault diagnosis method for a water diversion sediment flushing system. The method includes: S1, constructing a comprehensive air-space-ground intelligent water-based monitoring system; S2, building a multi-source data fusion preprocessing and digital twin modeling platform; S3, a multi-factor coupled intelligent early warning method; S4, a precise fault diagnosis method for all operating conditions; S5, adopting an integrated closed-loop collaborative handling mechanism encompassing early warning, diagnosis, control, and operation and maintenance; and S6, system security protection and data encryption. This invention employs a comprehensive monitoring system, breaking through the limitations of existing single monitoring modes, and achieving all-round, blind-spot-free, and full-parameter monitoring of the water diversion sediment flushing system. It particularly solves the technical challenges of monitoring underwater components and hidden sediment accumulation, achieving leading domestic levels in monitoring coverage, real-time performance, and accuracy, filling the technical gap in comprehensive monitoring of this type of system.

Description

Technical Field

[0002] This invention relates to the field of intelligent monitoring and fault diagnosis technology for water conservancy projects, and in particular to an intelligent early warning and fault diagnosis method for a water diversion sediment flushing system. Background Technology

[0004] Existing river dams and water diversion projects suffer from severe siltation during the flood season. To address this, relevant technical literature describes a water diversion sediment flushing structure (system). This system includes a diversion channel located on one side of the river; a side weir on the side of the diversion channel closest to the river, and a bank protection structure on the side furthest from the river; a river dam structure at the upstream end of the diversion channel, comprising a diversion channel inlet sand retainer, a pneumatic steel dam, and a spillway dam arranged in parallel; an intake gate at the downstream end of the diversion channel, with an intake gate sand retainer upstream of the intake gate; and a flushing gate between the intake gate sand retainer and the end of the diversion channel side weir. Through the coordinated operation of components such as the diversion channel inlet sand retainer, pneumatic steel dam, spillway dam, intake gate, flushing gate, and ecological gate, the system effectively solves the technical challenges of water diversion, sediment flushing, flood discharge, and ecological water supply, achieving preliminary control of siltation and dynamic regulation of water flow.

[0005] However, this type of water diversion sedimentation and flushing system only focuses on structural design and does not address intelligent monitoring, risk warning, and fault diagnosis functions during system operation. Currently, this type of water diversion sedimentation and flushing system still has the following technical problems in actual operation:

[0006] (1) Outdated monitoring methods: The existing water diversion sediment flushing system mostly adopts the manual inspection + single-point instrument monitoring mode, which has a limited monitoring range and poor real-time performance. It cannot achieve all-round and full-parameter synchronous monitoring of core components such as water diversion channels, pneumatic steel dams, sand flushing gates, and river overflow dams. In particular, it cannot capture hidden risks such as sediment accumulation thickness, component stress and strain, and abnormal water flow patterns in a timely manner, which can easily lead to the accumulation of risks and cause system failure.

[0007] (2) Low accuracy and strong lag in early warning: Existing early warning methods are mostly based on fixed threshold triggering, without considering the coupled effects of multiple factors such as water flow rate, sediment content, seasonal changes, and component aging. This makes it easy to have false or missed early warnings. Moreover, the early warning can only indicate the existence of risks, but cannot predict the development trend of risks or locate the source of risks, which makes it impossible for maintenance personnel to take timely and targeted measures.

[0008] (3) Low efficiency and poor accuracy of fault diagnosis: During system operation, faults such as leakage of pneumatic steel dam airbags, jamming of sand flushing gate, silt compaction in water diversion channel, scouring and damage of sand retaining wall, and abnormal flow of ecological gate are prone to occur. Existing technologies mostly rely on the experience of operation and maintenance personnel to make judgments, resulting in long diagnosis cycles and large errors. It is impossible to achieve accurate fault location, type identification and root cause analysis, and it is difficult to achieve early prediction and proactive prevention of faults.

[0009] Currently, there is no integrated intelligent early warning and fault diagnosis method for the "two-stage sedimentation + three-dimensional flushing + dynamic control" type water diversion sedimentation and flushing system in the field of water conservancy engineering in China. Existing technologies are mostly limited to the monitoring or fault diagnosis of single components and single parameters, and have not formed an intelligent early warning and fault diagnosis system covering all components, all working conditions and the entire process of the system, which cannot meet the needs of efficient, safe and stable operation of such systems.

[0010] The information disclosed in this background section is intended only to enhance the understanding of the general background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0012] The purpose of this invention is to provide an intelligent early warning and fault diagnosis method for a water diversion sedimentation and flushing system, so as to solve the technical problems existing in the prior art.

[0013] To achieve the above objectives, the present invention adopts the following technical solution:

[0014] This invention provides an intelligent early warning and fault diagnosis method for a water diversion sedimentation and flushing system, comprising:

[0015] S1. Construct a comprehensive monitoring system covering air, land, sea, and water: Deploy various types of high-precision monitoring equipment in the core components and surrounding environment of the water diversion sediment flushing system to construct a comprehensive, all-round, and parameter-free monitoring system covering air, land, sea, water, and equipment operation, so as to realize the real-time acquisition, filtering, and stable transmission of data related to the system's operating status.

[0016] S2. Establish a multi-source data fusion preprocessing and digital twin modeling platform: perform fusion preprocessing on the multi-source monitoring data collected in S1, and at the same time build a full-scale digital twin model of the system based on the structural parameters of the water diversion sediment flushing system, and build a supporting operation and maintenance database to provide data and model support for subsequent early warning and diagnosis.

[0017] S3, Multi-factor Coupled Intelligent Early Warning Method: Based on the digital twin model built by S2 and the preprocessed multi-source monitoring data, a multi-factor coupled early warning system is constructed to realize hierarchical early warning, trend prediction and precise positioning of system operation risks, and trigger corresponding linkage mechanisms;

[0018] S4. Full-condition accurate fault diagnosis method: Based on the early warning information of S3, the digital twin model of S2, and multi-source monitoring data and operation and maintenance database, a full-condition fault diagnosis system is constructed to achieve accurate fault identification, type determination, root cause tracing and component life prediction.

[0019] S5: Adopts an integrated closed-loop collaborative handling mechanism encompassing early warning, diagnosis, control, and operation and maintenance, including:

[0020] Based on the early warning information of S3 and the fault diagnosis results of S4, combined with the digital twin model of S2 and the operation and maintenance database, an integrated closed-loop collaborative system of early warning, diagnosis, control and operation and maintenance is constructed to achieve rapid fault handling, effective risk prevention and control and continuous system optimization.

[0021] S6: System Security Protection and Data Encryption: Establish a multi-layered system security protection system to protect monitoring equipment, data transmission, digital twin platform and database. At the same time, encrypt and back up all types of data throughout the process to ensure stable system operation and controllable data security.

[0022] Preferably, step S1 includes:

[0023] S11. Aerial Monitoring: Deploy an autonomous drone inspection system, equipped with high-definition cameras, lidar, and infrared thermal imagers. Following a preset inspection route, it covers the entire area of ​​the water diversion channel, river overflow dam, pneumatic steel dam, sand flushing gate, and bank protection. It regularly inspects system components, collecting image data including the appearance of components, changes in the surrounding topography, and floating objects on the water surface. The inspection frequency can be dynamically adjusted according to the flood season and non-flood season. During the flood season, it is once every 2 hours, and during the non-flood season, it is once every 6 hours. In case of emergency, it can trigger real-time inspection.

[0024] S12, Ground-to-ground monitoring: Deploy rain-measuring radar and satellite remote sensing receiving modules to collect environmental data in real time, including regional rainfall, precipitation intensity, river water level, and watershed inflow. Combined with satellite remote sensing images, identify potential landslide hazards on riverbanks to achieve watershed-scale environmental and safety monitoring, providing macro-data support for system operation control and risk warning.

[0025] S13. Ground monitoring: Stress-strain sensors, vibration sensors, displacement sensors, and temperature sensors are installed on the dam / gate bodies of the water diversion channel revetment, overflow dam, pneumatic steel dam, sand flushing gate, intake gate, and ecological gate to collect structural health data, including stress, strain, vibration frequency, displacement, and surface temperature, in real time, and to monitor whether there are scour, damage, deformation, or aging problems in the components; liquid level sensors and flow velocity sensors are installed at the side weirs and sand retaining walls of the water diversion channel to collect hydrological data, including the overflow of the side weir, the water level difference between the upstream and downstream of the sand retaining wall, and the water flow velocity, in real time.

[0026] S14. Underwater monitoring: Deploy underwater high-definition cameras, acoustic Doppler current meters, sediment concentration sensors, and underwater topographic surveying instruments in the water diversion channel, stilling basin section, seawall section, and downstream of the sand flushing gate to collect hydrological data in real time on underwater sediment thickness, sediment distribution, water flow pattern, sediment particle size distribution, and the integrity of underwater components. Focus on monitoring whether sediment accumulation in the water diversion channel reaches the warning threshold and the sand flushing effect of the sand flushing gate to avoid sand discharge problems caused by sediment compaction.

[0027] S15. Intelligent Monitoring Terminal: Pressure sensors, current sensors, voltage sensors, and stroke sensors are installed on the air filling and emptying devices of the pneumatic steel dam and the opening and closing devices of the sand flushing gate / intake gate / ecological gate to collect equipment operating parameters in real time, including air filling and emptying pressure, opening and closing current, voltage, stroke position, and operating power; current sensors are installed at the light strips in the water diversion channel to monitor the operating status of the lighting system; all monitoring devices are connected to the intelligent monitoring terminal to realize real-time data acquisition, filtering, encoding, and transmission. The transmission adopts 5G + fiber optic dual-mode redundant transmission to ensure the stability and real-time performance of data transmission and avoid data loss or delay.

[0028] Preferably, step S2 includes:

[0029] S21. Multi-source data fusion preprocessing: Receive multi-type data transmitted from various monitoring devices in step S1, including image data, structural health data, hydrological data, and environmental data. Adaptive filtering algorithms are used to remove data noise. Preprocessing operations such as data alignment, missing value completion, and outlier removal ensure data accuracy and completeness. Feature extraction algorithms are used to extract characteristic parameters such as component crack length, damaged area, and siltation thickness from image data; component fatigue damage characteristics from structural health data; and abnormal flow patterns from hydrological data, achieving standardization and characterization of multi-source data. A three-level multi-source data fusion algorithm is employed, including the following:

[0030] S211, Level 1 data preprocessing, used for noise suppression and anomaly removal:

[0031] An adaptive Kalman filter algorithm is used to suppress data noise, and specific filter formulas are designed for different types of monitoring data.

[0032] For structural health data, namely stress σ, strain ε, and displacement d, the filtering formula is:

[0033] ;

[0034] ;

[0035] ;

[0036] in: The data value after filtering at time k. Let A be the filtered data value at time k-1, A be the state transition matrix (A=1.02 for structural health data), and B be the control matrix (B=0.01). The control quantity at time k. For Kalman gain, Let be the original monitoring data at time k, Q be the process noise covariance (taken as Q=0.001), P be the covariance matrix, and I be the identity matrix.

[0037] For hydrological data, namely sediment load S, flow velocity v, and water level h, outliers are removed using the improved 3σ criterion, and the formula is as follows:

[0038] If satisfied If so, then the data will be retained;

[0039] If satisfied ,but ;

[0040] in: This is the average of 100 consecutive historical data sets for this monitoring parameter. Standard deviation, This represents the original monitoring data at time k;

[0041] S212, Second-level feature fusion, used for feature extraction and standardization:

[0042] To extract core features from different types of data, min-max standardization is used to eliminate the influence of units. The formula is as follows:

[0043] ;

[0044] in: Here, x represents the standardized feature values, and x represents the original feature values. This is the historical maximum value of this feature parameter. This is the historical minimum value of this feature parameter;

[0045] The following strategy is used for feature extraction:

[0046] (1) Image data, i.e., data from drones and underwater cameras:

[0047] The CNN algorithm is used to extract features, with core features including crack length L, damaged area S, and sediment thickness H. The feature extraction discriminant is:

[0048] ;

[0049] ;

[0050] in: The pixel coordinates of the crack edge; , where I = 1 for damaged areas and I = 0 for undamaged areas, and m and n are the number of rows and columns of the image pixel matrix;

[0051] (2) Structural health data: Extract fatigue damage feature D, using Miner's fatigue damage accumulation theory, the formula is:

[0052] ;

[0053] in: Let i be the actual number of cycles under stress level i. Let D be the fatigue life corresponding to the i-th stress level, and k be the number of stress levels. When D≥1, the component is determined to have a risk of fatigue damage.

[0054] S213, Three-level decision-level fusion, used for weight allocation and fusion output: Design a multi-source data fusion weight allocation strategy based on the analytic hierarchy process, clarify the priority of each monitoring data, and the fusion formula is:

[0055] ;

[0056] Where: F is the integrated feature value after fusion. Let w represent the weights of the i-th type of monitoring data, where w1 = 0.2 for aerial monitoring, w2 = 0.15 for ground-to-ground monitoring, w3 = 0.3 for ground monitoring, w4 = 0.25 for underwater monitoring, and w5 = 0.1 for equipment operation monitoring. The weights are determined using the Analytic Hierarchy Process (AHP), and the consistency check CR < 0.1. It is the sum of standardized eigenvalues ​​of the i-th type of monitoring data;

[0057] S22, Digital Twin Modeling:

[0058] Based on the structural design parameters of the water diversion sediment flushing system and combined with pre-processed monitoring data, a full-scale digital twin model of the system was built to achieve a 1:1 high-fidelity mapping between the physical entity and the digital model. The model covers all core components, including the water diversion channel, pneumatic steel dam, overflow dam, flushing gate, and intake gate, as well as the surrounding environment. Real-time monitoring data, historical operation data, component material parameters, and operation and maintenance records were imported into the digital twin model to achieve real-time synchronous updates between the model and the physical system, intuitively displaying the system's operating status, sediment distribution, and component health status information.

[0059] Synchronization update strategy between digital twin model and physical system:

[0060] The synchronization error control formula adopts a "real-time interpolation + dynamic correction" mechanism:

[0061] ;

[0062] in: For synchronization error, Let k be the parameter values ​​of the digital twin model at time k. Let k be the monitoring value of the physical system at time k; when At that time, the model parameters are corrected using the following formula:

[0063] ;

[0064] in: To correct the parameter values ​​of the digital twin model and ensure high-fidelity synchronization between the model and the physical system;

[0065] S23. Historical Database Construction: Collect long-term monitoring data, fault records, operation and maintenance data, and hydrological and meteorological data of the system to construct a historical database. Combine this with fault cases and operation and maintenance experience of similar water conservancy projects at home and abroad to form a fault knowledge base and an operation and maintenance knowledge base, providing data support for intelligent early warning, fault diagnosis and operation and maintenance optimization.

[0066] Preferably, step S3 includes:

[0067] S31. Construction of an early warning indicator system: Based on the operational characteristics of the water diversion sedimentation and flushing system, an early warning indicator system covering 5 major categories and 28 early warning indicators is constructed, including:

[0068] Sedimentation indicators: sedimentation thickness in the water diversion channel, sedimentation rate, water level difference between upstream and downstream of the sand retaining wall, sediment content in the discharge from the sand flushing gate, and degree of sediment compaction;

[0069] Health indicators of components: stress and strain, displacement, and airbag pressure of pneumatic steel dams; crack length and damaged area of ​​overflow dams and water diversion channel revetments; opening and closing travel deviation and gate vibration frequency of sand flushing gates / intake gates / ecological gates; and scour depth of sand retaining walls.

[0070] Equipment operation indicators: pneumatic steel dam filling and venting pressure, filling and venting time, sand flushing gate / intake gate / ecological gate opening and closing current, voltage, and power, filling and venting device operating status, and LED strip operating current;

[0071] Hydrological and environmental indicators: river level, inflow, sediment content, rainfall, precipitation intensity, overflow of the diversion channel side weir, stilling basin level, and flow velocity of the floodplain.

[0072] Ecological indicators: ecological gate discharge flow, downstream ecological flow compliance rate, dissolved oxygen concentration in the water diversion channel, and water quality parameters;

[0073] S32. Dynamic calibration of early warning thresholds: Utilizing machine learning algorithms and historical database data, dynamic threshold calibration is performed on each early warning indicator, overcoming the limitations of traditional fixed thresholds. Based on seasonal changes, inflow sediment content, and system operating conditions (including water diversion, sediment flushing, flood discharge, and ecological water supply), the early warning thresholds for each indicator are automatically adjusted to achieve adaptive optimization. The adaptive threshold calibration formula, using the sediment deposition thickness early warning threshold as an example, is as follows:

[0074] ;

[0075] in: The dynamic early warning threshold for sediment deposition thickness at time t, in meters; The baseline threshold for siltation thickness is 1.2m during the non-flood season and 0.8m during the flood season. The sediment content of the incoming water at time t, in kg / m³. The inflow rate at time t is expressed in m³ / s. The correction coefficients are: θ=0.1 for water diversion, θ=0.3 for sand flushing, θ=0.2 for flood discharge, and θ=0.15 for ecological water supply; α, β, and γ are weighting coefficients: α=0.0002, β=0.0001, and γ=0.1, determined through training with historical data.

[0076] The dynamic threshold calibration of other indicators adopts a similar logic as described above, with only the weighting coefficients and benchmark thresholds adjusted according to the indicator type. For example, the formula for the pressure warning threshold of the pneumatic steel dam airbag is:

[0077] ;

[0078] in: The dynamic warning threshold for airbag pressure at time t (unit: kPa). The reference pressure threshold is 60 kPa, and t is the running time (in hours). Let t be the pressure change at time t (unit: kPa), α = 0.0001, β = 0.05.

[0079] S33. Multi-factor Coupled Early Warning Model: Based on the fusion algorithm of BP neural network and LSTM, a multi-factor coupled early warning model is constructed. The input is preprocessed multi-source monitoring data and real-time status data of digital twin model. The output is the risk level of each early warning indicator (no risk, general risk, relatively high risk, major risk). The model can capture the coupling relationship between various indicators (such as the increase in siltation thickness will lead to a decrease in the flow velocity of the water diversion channel, which will aggravate siltation and increase the stress of the sand retaining wall), so as to achieve a comprehensive risk assessment.

[0080] The core formulas and operating mechanism of the model are as follows:

[0081] (1) Output layer of BP neural network (comprehensive evaluation of multiple indicators):

[0082] ;

[0083] in: Let y_j be the comprehensive evaluation value of the j-th early warning indicator (0≤y_j≤1). Let be the connection weight between the i-th neuron in the input layer and the j-th neuron in the output layer. This represents the input value (fused feature value) of the i-th neuron in the input layer. Let σ be the bias of the j-th neuron in the output layer, and σ be the activation function, using the sigmoid function. ;

[0084] (2) LSTM trend prediction layer (risk trend prediction):

[0085] ;

[0086] ;

[0087] ;

[0088] ;

[0089] ;

[0090] in: For input gate, For the Gate of Oblivion For output gate, In cellular state, For output of the hidden layer, The input data is at time t. For the input weight matrix, The hidden layer weight matrix is... The term is the bias term, and ⊙ represents the Hadamard product;

[0091] (3) Risk Level Determination Mechanism: Combining the comprehensive assessment value output by BP and the trend prediction value output by LSTM, a risk level determination rule is designed, and the formula is as follows:

[0092] ;

[0093] Where R is the risk level quantification value, and λ is the weighting coefficient (λ=0.6, highlighting the impact of the current state). Risk trend change rate; risk level is determined based on R value:

[0094] ① No risk: R < 0.2; ② Moderate risk: 0.2 ≤ R < 0.4; ③ Significant risk: 0.4 ≤ R < 0.7; ④ Major risk: R ≥ 0.7;

[0095] S34. Trend Early Warning and Precise Positioning: The LSTM algorithm is used to predict the changing trends of various early warning indicators, and the direction and speed of risk development can be predicted 12-24 hours in advance to avoid early warning lag. Combined with digital twin models, the specific location of the risk, the components involved and the scope of impact can be precisely located. For example, the siltation early warning can accurately locate the specific section of siltation in the water diversion channel, and the component damage early warning can accurately locate the specific location and length of the crack.

[0096] The formula for controlling the prediction accuracy of trend early warning is:

[0097] ;

[0098] Where: δ is the prediction error; Δt is the prediction lead time, Δt∈[12,24], in hours;

[0099] This is the predicted value Δt hours in advance; The actual monitoring value is t+Δt hours; when δ>5%, the prediction accuracy is improved by adjusting the hidden layer weights of the LSTM model.

[0100] Precise positioning mechanism: Based on the spatial coordinate mapping of the digital twin model, the coordinates (x, y, z) of the monitoring point corresponding to the early warning indicator are matched with the coordinates of the model components. The positioning formula is as follows:

[0101] ;

[0102] Where: D is the spatial distance between the monitoring point and the model component, (x,y,z) are the actual coordinates of the monitoring point ... m ,y m ,z m () represents the coordinates of the component in the digital twin model; when D≤0.5m, the component is determined to be a risky component, achieving precise positioning;

[0103] S35. Early Warning Issuance and Linkage: Based on the risk level, early warning information is automatically generated (including risk level, location, scope of impact, predicted trend, and preliminary handling suggestions), and issued to maintenance personnel through multiple channels such as SMS, audible and visual alarms, central control room pop-ups, and mobile APP; for major risks, an emergency linkage mechanism is automatically triggered to suspend the operation of relevant equipment (such as stopping the raising and lowering of pneumatic steel dams and emergency opening of sand flushing gates), and to activate emergency water supply and emergency sand flushing plans to prevent the risk from escalating; at the same time, the early warning information is integrated into a digital twin model to intuitively display the risk distribution and development trend;

[0104] Emergency Response Trigger Logic and Control Parameter Calculation: Taking the major risk of siltation as an example, the formula for adjusting the opening and closing degree of the silt flushing gate is as follows:

[0105] ;

[0106] in: The opening and closing degree of the sand flushing gate at time t (%) The baseline opening / closing degree is 50%, and κ is the adjustment coefficient (κ=20). Let t be the actual thickness of the sediment deposit. The dynamic early warning threshold ensures that the sand flushing flow rate meets the sand discharge requirements and quickly alleviates the risk of siltation.

[0107] Preferably, step S4 includes:

[0108] S41. Fault Type Classification; Based on the structure and operational characteristics of the water diversion sedimentation and flushing system, common fault types of the system are identified and classified into 6 major categories and 32 specific faults, including:

[0109] Failures of pneumatic steel dams: airbag leakage, inflation and deflation device malfunction, shield plate jamming, limit band damage, excessive stress, and abnormal displacement;

[0110] Gate malfunctions: Sand flushing gate / intake gate / ecological gate jamming, gate body damage, aging seals, opening and closing device malfunction, excessive travel deviation;

[0111] Sediment-related faults: excessive sediment accumulation in the water diversion channel, sediment compaction, erosion and damage to the sand retaining wall, and poor sand flushing effect of the sand flushing gate;

[0112] Structural damage and failures: cracks and damage to the revetment of the water diversion channel, scouring and damage to the overflow dam, and damage to the stilling basin and the anti-scouring structure of the seawall section;

[0113] Equipment malfunctions: power supply failure of the charging and discharging device, motor failure of the opening and closing device, monitoring equipment failure, and light strip failure of the water diversion channel;

[0114] Ecological and hydrological anomalies: abnormal flow at the ecological gate, insufficient downstream ecological flow, abnormal water quality in the water diversion channel, and turbulent water flow.

[0115] S42. Fault Feature Matching: Using deep learning algorithms, the monitoring data features under fault conditions (such as abnormal inflation and deflation pressure drop rate and abnormal current change when the pneumatic steel dam airbag leaks) are extracted and compared with the fault feature templates in the fault knowledge base to achieve accurate identification of fault types with an accuracy rate of no less than 98%. For new faults, the fault features are automatically learned through machine learning algorithms to update the fault knowledge base and achieve self-evolution of fault identification.

[0116] This step uses a fault feature similarity matching algorithm to clarify the specific details and judgment criteria for fault identification. The formula is as follows:

[0117] ;

[0118] in: The similarity between the fault features to be identified and the fault knowledge base template is used. Let be the i-th feature parameter of the fault to be identified (after standardization). Let be the i-th feature parameter of the fault template (after standardization), and n be the number of feature parameters; the discrimination rule is:

[0119] ①When When the fault is identified as this type, it is successfully identified.

[0120] ② When 0.7≤ When the fault is suspected to be of this type, manual review is triggered.

[0121] ③When When a new type of fault is identified, its characteristics are automatically learned and the fault knowledge base is updated.

[0122] Details of typical fault identification:

[0123] (1) Pneumatic steel dam airbag leakage fault: The characteristic parameter is the rate of change of inflation and deflation pressure. Current change The discriminant is: and At the same time, satisfying similarity The cause was determined to be an airbag leak;

[0124] (2) Sand flushing gate opening and closing jamming fault: The characteristic parameter is the opening and closing stroke deviation. Power of start and stop motor The discriminant is: and ( (Rated power), while also satisfying similarity. The condition was determined to be a jammed opening / closing mechanism.

[0125] (3) Sedimentation failure in the water diversion channel: The characteristic parameter is the rate of change of sediment thickness. The discriminant for the water flow velocity v is: and At the same time, satisfying similarity It was determined to be mud and sand compaction;

[0126] S43. Trace the root cause of the fault;

[0127] Based on digital twin models, the entire process of a fault is simulated. Combined with multi-source monitoring data, the root cause of the fault can be traced, avoiding only checking surface faults and ignoring the root cause. For example, if the sand flushing gate is stuck, model simulation and data analysis can be used to determine whether it is caused by siltation of the gate body, mechanical wear of the opening and closing device, or power failure, and to identify the inducing factors of the fault (such as excessive silt content during the flood season or untimely operation and maintenance).

[0128] Fault root cause tracing employs causal chain analysis combined with a data verification mechanism, using the following formula:

[0129] ;

[0130] Where: C represents the correlation degree of the root factor. Let i be the weight of the i-th potential root cause factor. Let be the correlation coefficient between the i-th potential root cause factor and the fault (determined through training with historical fault data); the factor with the highest correlation is the root cause of the fault; taking the jamming of the sand flushing gate as an example, the potential root causes factors include: siltation (C1), mechanical wear (C2), and power failure (C3), with weights w1=0.5, w2=0.3, and w3=0.2 respectively; the correlation coefficient is calculated through monitoring data: if , , Then C = 0.5 × 0.8 + 0.3 × 0.3 + 0.2 × 0.2 = 0.53, which indicates that siltation is the root cause. At the same time, by combining the digital twin model to simulate the siltation location, it is clear that the inducing factors are excessively high sediment content during the flood season and untimely sand flushing.

[0131] S44. Fault Level Determination and Impact Assessment;

[0132] Based on the severity, scope of impact, and effect on the core functions of the system (water diversion, sedimentation, sand flushing, and ecological water supply), the faults are classified into four levels: general faults, major faults, critical faults, and emergency faults. The consequences of the continued development of the fault are simulated through a digital twin model to assess the impact of the fault on system operation, the surrounding environment, and downstream water supply, providing a basis for the formulation of disposal plans.

[0133] The formula for determining the fault level is:

[0134] ;

[0135] Where: G is the quantified fault level value, S is the fault impact range, where 0≤S≤1, the larger the range, the larger the S value, I is the degree of impact of the fault on core functions, where 0≤I≤1, the larger the impact, the larger the I value, T is the fault duration, after standardization 0≤T≤1, α=0.4, β=0.4, γ=0.2 are weighting coefficients; fault levels are classified according to the G value:

[0136] ① General fault: G < 0.3; ② Major fault: 0.3 ≤ G < 0.6; ③ Critical fault: 0.6 ≤ G < 0.9; ④ Emergency fault: G ≥ 0.9;

[0137] The impact assessment of the failure was conducted using a digital twin model, with the core assessment indicator being the water diversion efficiency loss rate. Sand washing effect loss rate Ecological flow compliance rate The evaluation formula is:

[0138] ;

[0139] ;

[0140] ;

[0141] in: This is the normal operating water flow rate. This refers to the water diversion flow rate under fault conditions. This represents the sand flushing efficiency under normal operating conditions. For sand flushing efficiency under fault conditions; For standard ecological flow, Ecological flow under fault conditions;

[0142] S45. Component life prediction: Based on the structural health data (stress, strain, vibration, displacement), operating time, and environmental factors (water erosion, sediment abrasion, temperature change) of the components, a fatigue damage accumulation algorithm is used to predict the remaining service life of each core component (pneumatic steel dam, overflow dam, sand flushing gate, sand retaining wall), providing early warning of component aging risks and supporting the formulation of operation and maintenance plans; for example, predicting the remaining service life of the airbags in the pneumatic steel dam allows for early replacement, avoiding system failures caused by airbag rupture;

[0143] The formula for predicting the remaining service life of a component is as follows:

[0144] ;

[0145] in: The remaining service life of the component (unit: hours). The design service life of the component is given by (unit: h), D is the fatigue damage characteristic (calculated by the formula in step S212), t is the actual operating time (unit: h), E is the water flow scouring intensity (unit: N / m²), and T is the change in ambient temperature (unit: ℃). The attenuation coefficients are determined based on the material of the components; for pneumatic steel dam airbags, k1=0.00001, k2=0.00002, and k3=0.000005.

[0146] Taking the pneumatic steel dam airbag as an example, if (10 years), D=0.3, t=21900h (2.5 years), E=500N / m², T=20℃, then This means the remaining service life is approximately 5.8 years, and an airbag replacement plan will be developed accordingly.

[0147] Preferably, step S5 includes:

[0148] S51, Adaptive Control Strategy: Based on the warning level and fault type, an adaptive control scheme is automatically generated, linking all components of the system for coordinated control, achieving rapid fault handling without manual intervention.

[0149] (1) Siltation warning / fault: Automatically adjust the opening and closing degree of the flushing gate and the lifting height of the pneumatic steel dam to increase the flushing flow rate and optimize the flushing time to avoid siltation; for areas with severe siltation, link with drones to provide precise dredging operation guidance, or start mechanical dredging equipment.

[0150] (2) Pneumatic steel dam failure: If the airbag leaks, the inflation and deflation device will be automatically shut down, the backup airbag will be activated, and the operating status of the pneumatic steel dam will be adjusted to ensure the safety of river flood discharge and water diversion; if the inflation and deflation device fails, it will automatically switch to manual control mode and issue an operation and maintenance warning.

[0151] (3) Gate failure: If the sand flushing gate / inlet gate is stuck in the opening and closing, the emergency opening and closing device will be automatically activated to clear the mud and sand around the gate and adjust the opening and closing force; if the gate is damaged, the operating load of the gate will be automatically reduced to prevent the damage from expanding.

[0152] (4) Abnormal ecological flow: Automatically adjust the opening and closing degree of the ecological gate to ensure that the downstream ecological flow meets the standard, and optimize the ecological water supply plan in combination with the water quality monitoring data of the water diversion channel;

[0153] This step supplements the specific strategies and parameter calculations for adaptive control. Taking a pneumatic steel dam airbag leakage fault as an example, the pressure control formula after the backup airbag is activated is:

[0154] ;

[0155] in: Let t be the pressure of the standby airbag. Standard operating pressure (60 kPa). The pressure stabilization time is 10 minutes. The current pressure of the faulty airbag; ensure the pressure of the backup airbag rises steadily to avoid sudden pressure changes that could damage the equipment.

[0156] S52, intelligent generation of operation and maintenance solutions;

[0157] Based on fault diagnosis results and component life prediction data, personalized operation and maintenance plans are automatically generated, specifying the operation and maintenance content, operation and maintenance time, operation and maintenance process, and required consumables and equipment, so as to achieve refined and intelligent operation and maintenance; for example, for aging seals, a replacement plan is automatically generated to remind operation and maintenance personnel to replace them in time; for minor erosion of sand retaining walls, a reinforcement plan is automatically generated to prevent the erosion from worsening.

[0158] The formula for optimizing operation and maintenance time is:

[0159] ;

[0160] in: For optimal operation and maintenance time, For maintenance costs, To mitigate system downtime losses, the optimal maintenance time window for different fault types is determined through training with historical data (e.g., 2-4 AM during non-flood seasons to minimize the impact on water diversion and ecological water supply).

[0161] S53, Closed-loop verification and optimization;

[0162] After fault handling and maintenance are completed, system operation data is collected through the monitoring system to verify the handling effect. If the fault is not completely resolved, the handling plan is automatically adjusted until the fault is eliminated. At the same time, the fault handling process and maintenance data are entered into the historical database and fault knowledge base to optimize the early warning model and fault diagnosis model, realizing the system's self-learning and self-optimization, and continuously improving the accuracy of early warning and diagnosis.

[0163] The formula for determining closed-loop verification is:

[0164]

[0165] Where: ε represents the deviation of the treatment effect. This represents the comprehensive characteristic value after fault handling. This represents the comprehensive characteristic value under normal operating conditions. When ε > 3%, the treatment plan is adjusted, and the treatment process is repeated until the requirements are met.

[0166] Preferably, step S6 includes:

[0167] S61 Security Protection: Establishes a multi-layered security protection system to protect monitoring equipment, data transmission, and digital twin platform, preventing equipment intrusion and data tampering; sets up equipment access control, assigning different operation permissions to different maintenance personnel to avoid misoperation; establishes an emergency response mechanism to automatically activate emergency protection plans in response to extreme weather (rainstorms, floods, typhoons) and sudden failures, ensuring the safe and stable operation of the system;

[0168] S62 Data Encryption: Employs the AES encryption algorithm to encrypt monitoring data, early warning information, fault data, and operation and maintenance data throughout the entire process, ensuring data security and confidentiality; establishes a data backup mechanism to regularly back up data and prevent data loss; and implements hierarchical data management to ensure the security and controllability of core data.

[0169] Preferably, the UAV autonomous inspection system supports autonomous path planning and automatic alarm for anomalies, and can identify cracks with a minimum width of not less than 0.1 mm and a lidar measurement accuracy of not less than ±1 cm; the underwater monitoring equipment can operate stably in a water depth range of 0-10 m, the sediment concentration sensor has a measurement range of 0-500 kg / m³ and a measurement accuracy of not less than ±5%; the sampling frequency of the monitoring equipment can be dynamically adjusted according to the operating conditions, with a sampling frequency of 1 time / minute during the flood season and 1 time / 5 minutes during the non-flood season.

[0170] Preferably, the multi-source data fusion preprocessing uses an adaptive Kalman filter algorithm to remove noise, interpolation is used to fill in missing values, and the 3σ criterion is used to remove outliers; the digital twin model supports real-time rendering, dynamic simulation, and fault reproduction, and can realize the visualization of system operation status and multi-scenario simulation (water diversion, sand flushing, flood discharge, and fault handling).

[0171] Preferably, the multi-factor coupled early warning model adopts a BP neural network and LSTM fusion algorithm, wherein the BP neural network is used for comprehensive evaluation of multiple indicators, and the LSTM is used for trend prediction. The model training uses data from the historical database, and the training accuracy is not less than 97%. The early warning information includes risk level, location of occurrence, scope of impact, predicted trend, and preliminary handling suggestions. The response time for major risk early warnings does not exceed 10 seconds.

[0172] Preferably, the fault diagnosis adopts the CNN-LSTM deep learning algorithm, with a fault identification accuracy of not less than 98% and a fault location accuracy of not more than 1m; the component life prediction adopts the Miner fatigue damage accumulation theory, with a prediction error of not more than 5%.

[0173] Preferably, the response time of the adaptive control strategy is no more than 30 seconds, and the control accuracy can reach ±0.1m (gate opening and closing degree, pneumatic steel dam lifting height); the operation and maintenance plan can be dynamically optimized according to the system operating status, component life, and operation and maintenance cost to achieve a balance between operation and maintenance cost and system reliability.

[0174] By adopting the above technical solution, the present invention has the following beneficial effects:

[0175] This invention adopts a full-parameter monitoring system, breaking through the limitations of the existing single monitoring mode, and realizing all-round, blind-spot-free, and full-parameter monitoring of the water diversion sediment flushing system. In particular, it solves the technical problems of underwater components and hidden siltation that are difficult to monitor. The monitoring coverage, real-time performance and accuracy have reached the leading level in China, filling the technical gap in all-round monitoring of this type of system.

[0176] This invention constructs a full-scale digital twin model adapted to the structure of a water diversion sedimentation and flushing system, achieving a 1:1 high-fidelity mapping between the physical system and the digital model. It can intuitively display the system's operating status, risk distribution, and fault conditions, and supports fault reproduction, multi-scenario simulation, and trend prediction. It breaks through the limitations of existing technologies that "emphasize monitoring but neglect simulation" and promotes the transformation of water conservancy projects from "visualization" to "digitalization and intelligence".

[0177] This invention proposes a multi-factor coupled intelligent early warning method, which abandons the drawbacks of traditional fixed threshold early warning, and innovatively designs an adaptive threshold calibration formula for working conditions and a multi-factor coupled early warning model. It clarifies the logic of early warning data fusion and the mechanism of trend prediction, and can predict the development trend of risks 12-24 hours in advance and accurately locate the source of risks. It solves the technical pain points of existing early warnings, such as low accuracy, strong lag, false early warning and missed early warning. The early warning accuracy and response speed have reached the leading level in China.

[0178] This invention establishes a precise fault diagnosis system for all working conditions, innovatively designs a fault feature similarity matching algorithm, a fault root cause tracing formula, and a component remaining life prediction formula, clarifies the judgment details, discrimination criteria, and root cause analysis logic for 32 specific faults, achieves a fault identification accuracy of no less than 98%, and a fault location accuracy of no more than 1 meter, solves the problems of existing technologies relying on manual experience, low diagnostic efficiency, and poor accuracy, and realizes the transformation of fault diagnosis from "passive investigation" to "proactive prediction and precise handling". Attached Figure Description

[0180] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0181] Figure 1 A flowchart of an intelligent early warning and fault diagnosis method for a water diversion and sediment flushing system provided in an embodiment of the present invention; Detailed Implementation

[0183] The following will be combined with the appendix Figure 1The technical solutions of the present invention have been clearly and completely described. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0184] The intelligent early warning and fault diagnosis method of the water diversion sedimentation and flushing system of the present invention will be described in detail below with reference to specific embodiments. This embodiment is only used to explain the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made without departing from the principle of the present invention should be included within the protection scope of the present invention.

[0185] The intelligent early warning and fault diagnosis method for a water diversion sedimentation and flushing system provided by this invention addresses the technical pain points of existing water diversion sedimentation and flushing systems, such as incomplete monitoring, delayed early warning, low fault diagnosis accuracy, and poor operation and maintenance efficiency, by constructing a comprehensive monitoring system, building a digital twin platform, designing multi-factor coupled early warning, realizing full-condition fault diagnosis, establishing a closed-loop collaborative handling mechanism, and improving safety protection. This achieves intelligent and refined management and control of the system operation. The specific implementation steps are as follows:

[0186] Step S1: Construct a comprehensive monitoring system for air, space, water, and intelligent systems;

[0187] The core objective of this step is to break through the limitations of the existing single monitoring mode by deploying multiple types of high-precision monitoring equipment on the core components and surrounding environment of the water diversion sedimentation and flushing system. This will construct a comprehensive, seamless, and all-parameter monitoring system covering the air, ground, underwater, and equipment operation, enabling real-time acquisition, filtering, and stable transmission of system operation status data. This will provide comprehensive and accurate basic data support for subsequent early warning and diagnosis. Specifically, this step consists of the following five sub-steps:

[0188] S11: Aerial surveillance;

[0189] A drone-based autonomous inspection system is deployed. This system supports autonomous path planning and automatic anomaly alarm functions. Inspection routes can be preset according to the actual system layout, ensuring comprehensive coverage of all core areas and surrounding environments, including water diversion channels, overflow dams, pneumatic steel dams, sand flushing gates, and revetments. The drones are equipped with three core devices: high-definition cameras, LiDAR, and infrared thermal imagers. The high-definition cameras can identify cracks with a minimum width of 0.1mm, the LiDAR measurement accuracy is no less than ±1cm, and the infrared thermal imager can capture abnormal surface temperatures of components, enabling precise monitoring of the component's appearance.

[0190] During the inspection, the drone collects three types of core data in real time: first, the appearance status data of components, including cracks, damage, aging and other appearance information of dams, gates and revetments; second, the surrounding topographical changes data, monitoring whether there are potential hazards such as landslides and collapses on the banks and around the river; and third, the floating object data, identifying debris, garbage and other obstacles that may affect the operation of the system.

[0191] The inspection frequency is dynamically adjusted according to the system's operating conditions, taking into account both real-time monitoring and equipment energy consumption: during the flood season (high rainfall, high sediment content in incoming water, and high risk level), an inspection is carried out every 2 hours; during the non-flood season, an inspection is carried out every 6 hours; when the system issues early warning information or sudden abnormal situations (such as rainstorms, floods, or equipment alarms), a real-time inspection mode can be triggered, and the drone will immediately start and go to the abnormal area to carry out a special inspection to ensure that abnormal situations are detected in a timely manner.

[0192] S12: Ground-to-ground monitoring;

[0193] By deploying rainfall-measuring radar and satellite remote sensing receiving modules, a watershed-scale macro-monitoring network is constructed to achieve comprehensive macro-monitoring of the environment and safety. The rainfall-measuring radar collects meteorological data such as regional rainfall and precipitation intensity in real time, with a sampling frequency synchronized with ground monitoring equipment: once per minute during the flood season and once every 5 minutes during the non-flood season, accurately capturing precipitation change trends. The satellite remote sensing receiving module receives satellite remote sensing images in real time and simultaneously collects hydrological environmental data such as river water levels and watershed inflow. Combining the spatial resolution advantages of remote sensing images, it identifies potential hazards such as bank landslides and river channel changes, providing macro-data support for system operation control and risk early warning, and compensating for the spatial limitations of ground and underwater monitoring.

[0194] The monitoring data from the ground and underwater are linked with other monitoring data. When the rain-measuring radar detects that the precipitation reaches the warning threshold, or when satellite remote sensing identifies potential landslide hazards on the bank slope, the sampling frequency of the ground and underwater monitoring equipment is automatically increased. This achieves coordinated linkage between macro and micro monitoring and improves the comprehensiveness of risk identification.

[0195] S13: Ground monitoring;

[0196] The ground monitoring system focuses on the structural health and key hydrological parameters of the core components. Stress-strain sensors, vibration sensors, displacement sensors, and temperature sensors are evenly deployed on the dam / gate bodies of water diversion revetments, overflow dams, pneumatic steel dams, sand flushing gates, intake gates, and ecological gates. The sensor deployment density is determined according to the component size and stress characteristics, with denser deployment at key stress-bearing locations (such as the dam root and gate opening / closing points) to ensure the representativeness of the monitoring data.

[0197] The aforementioned sensors collect structural health data of the components in real time, including stress σ, strain ε, vibration frequency, displacement, and surface temperature. By analyzing this data, it is possible to determine in real time whether the components have problems such as erosion, damage, deformation, or aging. For example, when the stress and strain data continuously exceed the normal range, it indicates that the components may have structural damage; when the vibration frequency fluctuates abnormally, it may be due to loosening or wear of the components.

[0198] Meanwhile, level sensors and flow velocity sensors are deployed at key hydrological monitoring points such as the side weirs and sand-retaining embankments of the water diversion channel to collect hydrological data such as the overflow of the side weirs, the water level difference between the upstream and downstream of the sand-retaining embankment, and the flow velocity in real time. This monitors the water flow status in the water diversion channel and provides basic data for judging siltation and sand flushing effects. For example, an excessively large water level difference between the upstream and downstream of the sand-retaining embankment may indicate that the sand-retaining embankment is blocked or has siltation problems.

[0199] S14: Underwater monitoring;

[0200] To address the challenges of monitoring underwater components due to their high degree of concealment, underwater high-definition cameras, acoustic Doppler current meters, sediment concentration sensors, and underwater topographic surveying instruments have been deployed in key underwater areas such as the water diversion channel, stilling basin section, seawall section, and downstream of the sand flushing gate. All underwater monitoring equipment can operate stably within a water depth range of 0-10m and is adaptable to complex underwater environments (such as turbidity and water flow impact).

[0201] The underwater monitoring equipment collects four types of core hydrological data in real time: First, underwater sediment deposition data, including sediment thickness and distribution, is collected collaboratively by an underwater topographic surveying instrument and a sediment concentration sensor. The sediment concentration sensor has a measurement range of 0-500 kg / m³ and a measurement accuracy of no less than ±5%. Second, water flow data is collected by an acoustic Doppler current meter to monitor changes in water flow velocity and direction and determine whether there are any abnormalities such as water flow turbulence. Third, sediment particle size distribution data is used to analyze the distribution of sediment particle size and provide a basis for optimizing the sand flushing scheme. Fourth, underwater component integrity data is collected by an underwater high-definition camera to monitor whether underwater dams, gates, sand retainers, and other components have problems such as damage, erosion, and sediment adhesion.

[0202] The focus of underwater monitoring is on the siltation in the water intake channel and the sand flushing effect of the flushing gate. When the siltation thickness reaches the warning threshold, or the silt concentration downstream of the flushing gate is abnormally low (indicating poor sand flushing effect), an early warning is triggered immediately to avoid silt compaction leading to poor sand discharge, which in turn affects the system's water intake and sand flushing functions.

[0203] S15: Intelligent monitoring terminal;

[0204] Dedicated monitoring sensors are installed at key operating points of the system's core equipment to enable real-time monitoring of equipment operating parameters: pressure sensors, current sensors, and voltage sensors are installed on the inflation and deflation devices of the pneumatic steel dam to collect parameters such as inflation and deflation pressure, operating current, and voltage in real time; current sensors, voltage sensors, and travel sensors are installed on the opening and closing devices of the sand flushing gate / intake gate / ecological gate to collect parameters such as opening and closing current, voltage, travel position, and operating power in real time; and current sensors are installed at the light strips in the water diversion channel to monitor the operating status of the lighting system and prevent safety hazards caused by light strip malfunctions.

[0205] All monitoring equipment (airborne, ground-based, terrestrial, and underwater) is uniformly connected to the intelligent monitoring terminal. This terminal has real-time data acquisition, filtering, encoding, and transmission functions. It can perform preliminary filtering of the collected raw data to remove obviously abnormal data (such as invalid data caused by sensor malfunctions) before encoding to ensure the standardization of data transmission. Data transmission adopts a 5G + fiber optic dual-mode redundant transmission mode. 5G transmission ensures the real-time performance of the data, while fiber optic transmission ensures the stability of the data. The dual transmission mode can effectively avoid data loss or delay, ensuring that the monitoring data can be transmitted to the subsequent digital twin platform in a timely and accurate manner.

[0206] In addition, the sampling frequency of the monitoring equipment can be dynamically adjusted according to the operating conditions. The sampling frequency is once per minute during the flood season and once every 5 minutes during the non-flood season. This ensures the real-time data under high-risk conditions during the flood season while reducing equipment energy consumption and data redundancy during the non-flood season.

[0207] Step S2: Build a multi-source data fusion preprocessing and digital twin modeling platform;

[0208] This step, based on the multi-source monitoring data collected in step S1, performs data fusion preprocessing to eliminate data noise, fill in missing data, and extract core features. Simultaneously, it builds a full-scale digital twin model of the system and a supporting operation and maintenance database to standardize and characterize the monitoring data, providing solid data and model support for subsequent intelligent early warning and fault diagnosis. This step is specifically divided into three sub-steps:

[0209] S21: Multi-source data fusion preprocessing;

[0210] The system receives various types of data transmitted from monitoring devices in step S1, including image data (acquired by drones and underwater cameras), structural health data (acquired by ground sensors), hydrological data (acquired by ground and underwater sensors), and environmental data (acquired by ground and space monitoring). Due to issues such as inconsistent dimensions, noise interference, missing data, and outliers in these diverse data types, multi-source data fusion preprocessing is required to ensure data accuracy, completeness, and standardization. Specifically, a three-level multi-source data fusion algorithm is employed, with preprocessing performed at three levels:

[0211] S211: Primary data preprocessing, used for noise suppression and anomaly removal;

[0212] Specific noise suppression and anomaly removal methods were designed to address the characteristics of different types of monitoring data, ensuring the reliability of the raw data.

[0213] For structural health data, namely stress σ, strain ε, and displacement d, an adaptive Kalman filter algorithm is used to suppress data noise. This algorithm can dynamically adjust the filtering parameters according to data changes to adapt to the fluctuation characteristics of structural health data. The specific filtering formula is as follows:

[0214]

[0215]

[0216]

[0217] in: The data value after filtering at time k. Let A be the filtered data value at time k-1, A be the state transition matrix (A=1.02 for structural health data), and B be the control matrix (B=0.01). The control quantity at time k. For Kalman gain, Let be the original monitoring data at time k, Q be the process noise covariance (taken as Q=0.001), P be the covariance matrix, and I be the identity matrix. Using the above formula, random noise in structural health data can be effectively filtered out, preserving the true trend of data change.

[0218] For hydrological data, namely sediment load S, flow velocity v, and water level h, an improved 3σ criterion is used to remove outliers. This method, based on the statistical characteristics of the data, can accurately identify abnormal data that exceeds the normal fluctuation range (such as outliers caused by sensor malfunctions or sudden interference). The specific formula is as follows:

[0219] If satisfied If so, then the data will be retained;

[0220] If satisfied ,but (Use the mean to replace outliers to avoid missing data);

[0221] in: This is the average of 100 consecutive historical data sets for this monitoring parameter. Standard deviation, This represents the original monitoring data at time k. By improving the 3σ criterion, outliers in the hydrological data can be effectively removed, while avoiding data loss due to outlier removal and ensuring the continuity of hydrological data.

[0222] In addition, for the problem of missing data, interpolation is used to fill in the missing data. Based on the valid data before and after the missing data, the missing values ​​are filled in by linear interpolation or polynomial interpolation to ensure the integrity of the data. For image data, denoising algorithms (such as Gaussian denoising) are used to filter out noise in the image, improve the image clarity, and lay the foundation for subsequent feature extraction.

[0223] S212: Second-level feature fusion, used for feature extraction and standardization;

[0224] For different types of data, core parameters that reflect the system's operating status, potential risks, and fault characteristics are extracted. Min-max standardization is then employed to eliminate the influence of dimensions, achieving standardization and characterization of multi-source data, which facilitates subsequent data fusion and model calculation.

[0225] First, min-max standardization is used to eliminate the influence of dimensions. The standardization formula is:

[0226] ;

[0227] in: Here, x represents the standardized feature values, and x represents the original feature values. This is the historical maximum value of this feature parameter. This is the historical minimum value of the feature parameter. Through standardization, feature parameters with different dimensions and numerical ranges are uniformly mapped to the [0,1] interval, eliminating the impact of dimensional differences on subsequent fusion and modeling.

[0228] Secondly, specific strategies are used to extract core features for different types of data, as follows:

[0229] (1) Image data, i.e., image data collected by UAVs and underwater cameras: Features are extracted using CNN (convolutional neural network) algorithm. The core features include crack length L, damaged area S, and siltation thickness H. The feature extraction discriminant is as follows:

[0230] ;

[0231] ;

[0232] in: The pixel coordinates of the crack edge; Let I represent the pixel value of the damaged area, where I=1 for the damaged area and I=0 for the undamaged area. m and n are the number of rows and columns in the image pixel matrix, respectively. Using this discriminant, core features in the image can be accurately extracted, quantifying conditions such as component damage and siltation.

[0233] (2) Structural health data: Fatigue damage feature D is extracted, and Miner's fatigue damage accumulation theory is used. This theory can accurately calculate the degree of fatigue damage accumulation of components under long-term stress. The formula is:

[0234] ;

[0235] in: Let i be the actual number of cycles under stress level i. Let D be the fatigue life corresponding to the i-th stress level, and k be the number of stress levels. When D ≥ 1, the component is determined to have a risk of fatigue damage and timely maintenance is required.

[0236] (3) Hydrological data and environmental data: Extract abnormal characteristics of water flow pattern, characteristics of rainfall change, characteristics of sediment concentration change, etc. For example, extract the fluctuation coefficient of water flow velocity and the rate of change of sediment concentration as the core characteristics for judging water flow turbulence and sediment deposition anomalies.

[0237] S213: Three-level decision-making fusion, used for weight allocation and fusion output;

[0238] We design a multi-source data fusion weight allocation strategy based on the Analytic Hierarchy Process (AHP), clarify the priority of each type of monitoring data, and combine it with standardized feature values ​​to achieve the fusion output of multi-source data, obtaining a comprehensive feature value F that can comprehensively reflect the system's operating status. The specific fusion formula is as follows:

[0239] ;

[0240] Where: F is the integrated feature value after fusion. The weights of the i-th type of monitoring data are determined by the AHP method, with a consistency check CR < 0.1 (ensuring the rationality of the weight allocation). The specific weights of each type of monitoring data are as follows: aerial monitoring w1 = 0.2, ground monitoring w2 = 0.15, ground monitoring w3 = 0.3, underwater monitoring w4 = 0.25, and equipment operation monitoring w5 = 0.1. It is the sum of the standardized eigenvalues ​​of the j-th feature parameter in the i-th type of monitoring data.

[0241] Through a three-level multi-source data fusion algorithm, scattered and multi-type monitoring data can be transformed into unified and comprehensive feature values, eliminating data redundancy, highlighting key information, and providing standardized data support for subsequent digital twin modeling, intelligent early warning, and fault diagnosis.

[0242] S22: Digital Twin Modeling;

[0243] Based on the structural design parameters of the water diversion sediment flushing system (such as dam dimensions, gate specifications, water diversion channel length, component materials, etc.), and combined with the monitoring data after preprocessing in step S21, a full-scale digital twin model of the water diversion sediment flushing system is built to achieve a 1:1 high-fidelity mapping between the system's physical entity and the digital model. The model covers all core components of the water diversion channel, pneumatic steel dam, overflow dam, flushing gate, and intake gate, as well as the surrounding environment (such as bank slope, river channel, and surrounding topography), thus restoring the system's true structure and operating environment.

[0244] Real-time monitoring data, historical operation data, component material parameters, operation and maintenance records, and other types of data are uniformly imported into the digital twin model to achieve real-time synchronous updates between the model and the physical system. This allows for an intuitive display of information such as system operation status, sediment distribution, and component health status. Operation and maintenance personnel can use the digital twin model to monitor the system's operation in real time and understand the status of core components without on-site inspection.

[0245] To ensure high-fidelity synchronization between the digital twin model and the physical system, a "real-time interpolation + dynamic correction" mechanism is adopted. The synchronization error control formula is as follows:

[0246] ;

[0247] in: For synchronization error, Let k be the parameter values ​​of the digital twin model at time k. Let k be the monitoring value of the physical system at time k; when At that time, the model parameters are corrected using the following formula:

[0248] ;

[0249] in: To correct the parameter values ​​of the digital twin model and ensure high-fidelity synchronization between the model and the physical system, the synchronization error is controlled within the allowable range.

[0250] In addition, the digital twin model supports real-time rendering, dynamic simulation, and fault reproduction, enabling the visualization of system operation status and simulation of multiple scenarios (water diversion, sand flushing, flood discharge, and fault handling). For example, it can simulate the sedimentation process under different sediment contents of incoming water and simulate the changes in system operation after a fault occurs, providing visual support for early warning and prediction, fault tracing, and optimization of handling plans.

[0251] S23: Historical database construction;

[0252] Collect various types of data from the long-term operation of the system, including monitoring data (raw data collected in step S1 and feature data after preprocessing in step S21), fault records (fault type, occurrence time, occurrence location, inducing factors, and handling process), operation and maintenance data (operation and maintenance content, operation and maintenance time, operation and maintenance effect, and consumable usage), and hydrological and meteorological data (historical rainfall, inflow, sediment content, etc.) to build a dedicated historical database for the system.

[0253] Meanwhile, we collected failure cases and operation and maintenance experience from similar water conservancy projects at home and abroad, organized and classified the cases and experience to form a failure knowledge base and an operation and maintenance knowledge base. The failure knowledge base includes characteristic templates, judgment criteria and root cause analysis methods for various failures, while the operation and maintenance knowledge base includes operation and maintenance processes, operation and maintenance cycles and optimization schemes for various components.

[0254] The historical database, fault knowledge base, and operation and maintenance knowledge base are interconnected, providing data support for intelligent early warning, fault diagnosis, and operation and maintenance optimization. For example, early warning models can be trained and optimized using historical data, fault diagnosis can be performed using feature matching through the fault knowledge base, and operation and maintenance solutions can be generated by combining the operation and maintenance knowledge base with historical operation and maintenance data.

[0255] Step S3: Multi-factor coupled intelligent early warning method;

[0256] This step, based on the digital twin model built in step S2 and the preprocessed multi-source monitoring data, constructs a multi-factor coupled early warning system. It overcomes the drawbacks of traditional fixed-threshold early warning systems, achieving tiered early warning, trend prediction, and precise location of system operational risks, and triggering corresponding linkage mechanisms to prevent risks in advance. Specifically, it consists of 5 sub-steps:

[0257] S31: Construction of Early Warning Indicator System;

[0258] Based on the operational characteristics of the water diversion sedimentation and flushing system, and fully considering five core dimensions—sedimentation, component health, equipment operation, hydrological environment, and ecological protection—a warning indicator system covering 5 categories and 28 warning indicators has been constructed. This system comprehensively covers all types of risks and hidden dangers in system operation. The specific warning indicators for each type are as follows:

[0259] (1) Sedimentation indicators: sedimentation thickness in the water diversion channel, sedimentation rate, water level difference between upstream and downstream of the sand retaining wall, sediment content in the flushing gate discharge, and degree of sediment compaction; these indicators are mainly used to monitor the risks related to sedimentation and avoid problems such as water diversion channel blockage and poor flushing caused by sedimentation.

[0260] (2) Component health indicators: stress and strain, displacement and airbag pressure of pneumatic steel dam, crack length and damaged area of ​​overflow dam and water diversion channel bank protection, opening and closing stroke deviation of sand flushing gate / intake gate / ecological gate, gate body vibration frequency, and sand retaining wall scour depth; these indicators are mainly used to monitor the structural health status of core components and to warn of risks such as component damage, deformation and aging.

[0261] (3) Equipment operation indicators: pneumatic steel dam filling and venting pressure, filling and venting time, sand flushing gate / water intake gate / ecological gate opening and closing current, voltage and power, filling and venting device operation status, and light strip operation current; these indicators are mainly used to monitor the operation status of system equipment and to warn of equipment failure, abnormal operation and other risks.

[0262] (4) Hydrological and environmental indicators: river level, inflow, sediment content, rainfall, precipitation intensity, overflow of the diversion channel side weir, stilling basin level, and flow velocity of the sea embankment section; these indicators are mainly used to monitor changes in the surrounding hydrological environment and to warn of risks such as floods, rainstorms, and water flow disturbances.

[0263] (5) Ecological indicators: ecological gate discharge flow, downstream ecological flow compliance rate, dissolved oxygen concentration in water diversion channel, and water quality parameters; these indicators are mainly used to monitor the ecological water supply effect, provide early warning of ecological and environmental risks, and ensure that the system operation takes into account both water conservancy functions and ecological protection.

[0264] S32: Dynamic calibration of early warning threshold

[0265] Machine learning algorithms (such as random forest) are used in conjunction with data from historical databases to dynamically calibrate the thresholds of various early warning indicators, overcoming the limitations of traditional fixed thresholds and achieving adaptive optimization of early warning thresholds. Based on seasonal changes, sediment content in incoming water, and system operating conditions (water diversion, sediment flushing, flood discharge, ecological water supply), the early warning thresholds for each indicator are automatically adjusted to ensure the accuracy and relevance of the early warnings.

[0266] Taking the early warning threshold for siltation thickness as an example, the adaptive threshold calibration formula for operating conditions is as follows:

[0267] ;

[0268] in: The dynamic early warning threshold for sediment deposition thickness at time t, in meters; The baseline threshold for siltation thickness is 1.2m during the non-flood season and 0.8m during the flood season. The sediment content of the incoming water at time t, in kg / m³. The inflow rate at time t is expressed in m³ / s. The correction coefficients are: θ=0.1 for water diversion, θ=0.3 for sand flushing, θ=0.2 for flood discharge, and θ=0.15 for ecological water supply; α, β, and γ are weighting coefficients, α=0.0002, β=0.0001, and γ=0.1, which are determined through training with historical data.

[0269] The dynamic threshold calibration of other indicators adopts a similar logic as described above, with only the weighting coefficients and benchmark thresholds adjusted according to the indicator type. For example, the formula for the pressure warning threshold of the pneumatic steel dam airbag is:

[0270] ;

[0271] in: The dynamic warning threshold for airbag pressure at time t (unit: kPa). The reference pressure threshold is 60 kPa, and t is the running time (in hours). Let t be the pressure change at time t (unit: kPa), α = 0.0001, β = 0.05.

[0272] Dynamic threshold calibration allows warning thresholds to be adapted to different operating conditions, avoiding false or missed warnings caused by fixed thresholds and improving the accuracy of warnings.

[0273] S33: Multi-factor coupled early warning model;

[0274] A multi-factor coupled early warning model is constructed based on a fusion algorithm of BP neural network and LSTM. This model fully leverages the comprehensive evaluation capability of BP neural network and the trend prediction capability of LSTM algorithm to achieve comprehensive assessment and trend prediction of system operation risks. The model input consists of multi-source monitoring data preprocessed in step S21 and real-time status data from the digital twin model. The output is the risk level of each early warning indicator (no risk, moderate risk, significant risk, major risk). Simultaneously, it captures the coupling relationships between indicators (e.g., increased siltation thickness leads to decreased flow velocity in the water diversion channel, thus exacerbating siltation and increasing stress on the sand-retaining embankment), achieving a comprehensive risk assessment.

[0275] The core formulas and operating mechanism of the model are as follows:

[0276] (1) BP neural network output layer (multi-indicator comprehensive evaluation): used to comprehensively evaluate the current state of each early warning indicator and output the comprehensive evaluation value. The formula is:

[0277] ;

[0278] in: Let y_j be the comprehensive evaluation value of the j-th early warning indicator (0≤y_j≤1). Let be the connection weight between the i-th neuron in the input layer and the j-th neuron in the output layer. This represents the input value (fused feature value) of the i-th neuron in the input layer. Let σ be the bias of the j-th neuron in the output layer, and σ be the activation function, using the sigmoid function. .

[0279] (2) LSTM Trend Prediction Layer (Risk Trend Prediction): Used to predict the changing trends of various early warning indicators and capture the time series characteristics of the data. The formula is:

[0280] ;

[0281] ;

[0282] ;

[0283] ;

[0284] ;

[0285] ;

[0286] in: For input gate, For the Gate of Oblivion For output gate, In cellular state, For output of the hidden layer, The input data is at time t. The input weight matrix and the hidden layer weight matrix are respectively. Equal terms are bias terms, and ⊙ is the Hadamard product.

[0287] (3) Risk Level Determination Mechanism: Combining the comprehensive assessment value output by BP and the trend prediction value output by LSTM, a risk level determination rule is designed to quantify the risk level. The formula is as follows:

[0288] ;

[0289] Where R is the risk level quantification value, and λ is the weighting coefficient (λ=0.6, highlighting the impact of the current state). This refers to the rate of change of risk trend; risk levels are classified based on the R-value, as follows:

[0290] ① No risk: R < 0.2; ② Moderate risk: 0.2 ≤ R < 0.4; ③ Significant risk: 0.4 ≤ R < 0.7; ④ Major risk: R ≥ 0.7.

[0291] The model is trained using data from a historical database, achieving a training accuracy of no less than 97%. It can accurately assess the risk level of each early warning indicator, capture the coupling relationship between indicators, and avoid the limitations of single indicator early warning.

[0292] S34: Trend Early Warning and Precise Positioning;

[0293] By using the LSTM algorithm to predict the changing trends of various early warning indicators, the direction and speed of risk development can be predicted 12-24 hours in advance, avoiding early warning delays and buying time for risk prevention and control. Simultaneously, combined with the spatial coordinate mapping function of the digital twin model, the specific location of the risk, the components involved, and the scope of impact can be accurately pinpointed, achieving precise risk control.

[0294] The formula for controlling the prediction accuracy of trend early warning is:

[0295] ;

[0296] Where: δ is the prediction error; Δt is the prediction lead time, Δt∈[12,24], in hours; This is the predicted value Δt hours in advance; The actual monitoring value is t+Δt hours; when δ>5%, the prediction accuracy is improved by adjusting the hidden layer weights of the LSTM model, ensuring the reliability of trend warning.

[0297] Precise positioning mechanism: Based on the spatial coordinate mapping of the digital twin model, the coordinates (x, y, z) of the monitoring point corresponding to the early warning indicator are matched with the coordinates of the model components. The positioning formula is as follows:

[0298] ;

[0299] Where: D is the spatial distance between the monitoring point and the model component, (x,y,z) are the actual coordinates of the monitoring point ... m ,y m ,z m (D) represents the coordinates of a component in the digital twin model; when D ≤ 0.5m, the component is determined to be a component at risk, achieving precise positioning. For example, siltation early warning can accurately locate the specific section of siltation in the water diversion channel, and component damage early warning can accurately locate the specific location and length of cracks, providing precise guidance for subsequent treatment.

[0300] S35: Early Warning Issuance and Coordination;

[0301] Based on the risk level, the system automatically generates early warning information, which includes the risk level, location, scope of impact, predicted trend, and preliminary handling suggestions, ensuring that maintenance personnel can quickly grasp the risk situation. Early warning information is disseminated through multiple channels, including SMS, audible and visual alarms, control room pop-ups, and mobile apps, ensuring that maintenance personnel can receive early warning information in a timely manner. The response time for major risk warnings is no more than 10 seconds.

[0302] Different linkage mechanisms are triggered for different levels of risk: for general and significant risks, early warning information is issued and maintenance personnel are reminded to strengthen inspections and pay close attention; for major risks, an emergency linkage mechanism is automatically triggered to suspend the operation of relevant equipment (such as stopping the raising and lowering of pneumatic steel dams and the emergency opening of sand flushing gates), and to activate emergency water supply and emergency sand flushing plans to prevent the risk from escalating. At the same time, the early warning information is integrated into a digital twin model to intuitively display the risk distribution and development trend, making it easier for maintenance personnel to formulate response plans.

[0303] Taking the major risk of siltation as an example, the formula for adjusting the opening and closing degree of the sluice gate in emergency response is as follows:

[0304] ;

[0305] in: The opening and closing degree of the sand flushing gate at time t (%) The baseline opening / closing degree is 50%, and κ is the adjustment coefficient (κ=20). Let t be the actual thickness of the sediment deposit. This is a dynamic early warning threshold; the opening and closing degree of the flushing gate is adjusted by this formula to ensure that the flushing flow meets the sand discharge requirements and quickly alleviates the risk of siltation.

[0306] Step S4: Accurate Fault Diagnosis Method for All Operating Conditions;

[0307] This step, based on the early warning information from step S3, the digital twin model from step S2, and multi-source monitoring data and the operation and maintenance database, constructs a full-condition fault diagnosis system to achieve accurate fault identification, type determination, root cause tracing, and component life prediction. It addresses the problems of existing technologies relying on manual experience, low diagnostic efficiency, and poor accuracy. Specifically, it consists of 5 sub-steps:

[0308] S41: Fault type classification;

[0309] Based on the structural and operational characteristics of the water diversion sedimentation and flushing system, common system fault types were identified and clearly categorized into 6 major categories and 32 specific faults, covering all core components and operational aspects of the system to ensure comprehensive fault diagnosis. The specific fault types are as follows:

[0310] (1) Failure of pneumatic steel dam: airbag leakage, inflation and deflation device failure, shield plate jamming, limit band damage, stress exceeding the standard, and abnormal displacement;

[0311] (2) Gate body failure: Sand flushing gate / inlet gate / ecological gate is stuck in opening and closing, gate body is damaged, seals are aged, opening and closing device is faulty, and the stroke deviation is too large;

[0312] (3) Sediment-related faults: excessive sediment accumulation in the water diversion channel, sediment compaction, erosion and damage to the sand retaining wall, and poor sand flushing effect of the sand flushing gate;

[0313] (4) Structural damage: cracks and damage to the water diversion channel revetment, scouring and damage to the overflow dam, and damage to the stilling basin and the anti-scouring structure of the seawall section;

[0314] (5) Equipment operation failures: power supply failure of the charging and discharging device, motor failure of the opening and closing device, monitoring equipment failure, and light strip failure of the water diversion channel;

[0315] (6) Ecological and hydrological anomalies: abnormal flow at the ecological gate, failure of downstream ecological flow to meet standards, abnormal water quality in the water diversion channel, and disordered water flow.

[0316] S42: Fault Feature Matching

[0317] The CNN-LSTM deep learning algorithm is used to extract monitoring data features under fault conditions (such as abnormal inflation / deflation pressure drop rate and abnormal current changes when the pneumatic steel dam airbag leaks). These features are then compared with fault feature templates in the fault knowledge base to achieve accurate fault type identification with an accuracy rate of no less than 98%. For novel faults (faults not included in the fault knowledge base), machine learning algorithms are used to automatically learn fault features and update the fault knowledge base, enabling self-evolution of fault identification and improving the system's adaptability to novel faults.

[0318] This step employs a fault feature similarity matching algorithm to clarify the specific details and criteria for fault identification, ensuring the accuracy of fault identification. The similarity calculation formula is as follows:

[0319] ;

[0320] Where: S represents the similarity between the fault feature to be identified and the fault knowledge base template. Let be the i-th feature parameter of the fault to be identified (after standardization). Let be the i-th feature parameter of the fault template (after standardization), and n be the number of feature parameters; the discrimination rule is:

[0321] ① When S≥0.85, it is determined to be this type of fault, and the identification is successful;

[0322] ② When 0.7≤S<0.85, it is judged as a suspected fault of this type, triggering manual review to ensure the accuracy of diagnosis;

[0323] ③ When S < 0.7, it is determined to be a new type of fault. The characteristic parameters of the fault are automatically learned, the fault knowledge base is updated, and support is provided for the diagnosis of similar faults in the future.

[0324] The characteristics and identification details of typical faults are as follows, clarifying the core characteristics and identification criteria of various fault types:

[0325] (1) Pneumatic steel dam airbag leakage fault: The core characteristic parameter is the rate of change of inflation and deflation pressure. Current change The discriminant is: and If the similarity S≥0.85, it is determined to be an airbag leak;

[0326] (2) Sand flushing gate opening and closing jamming fault: The core characteristic parameter is the opening and closing stroke deviation. Power of start and stop motor The discriminant is: and ( If the rated power is met, and the similarity S≥0.85, it is determined to be an opening / closing malfunction.

[0327] (3) Sediment compaction fault in water diversion channel: The core characteristic parameter is the rate of change of sediment thickness. The discriminant for the water flow velocity v is: and If the similarity S≥0.85 is satisfied, it is determined to be mud and sand compaction.

[0328] S43: Trace the root cause of the fault;

[0329] Based on a digital twin model, the entire process of a fault is simulated. By combining multi-source monitoring data and historical fault cases, the root cause of the fault can be traced, avoiding only checking surface faults and ignoring the root cause. This ensures that the fault can be thoroughly resolved and prevents recurrence. For example, if a sand flushing gate is stuck, the opening and closing process can be simulated using a digital twin model. Combined with monitoring data (such as the concentration of sediment around the gate and the current of the opening and closing motor), it can be determined whether the fault is caused by sediment accumulation in the gate, mechanical wear of the opening and closing device, or power failure. The inducing factors of the fault (such as excessive sediment content during the flood season or untimely operation and maintenance) can also be identified.

[0330] The root cause tracing of faults employs causal chain analysis combined with a data verification mechanism. By quantifying the correlation between potential root cause factors and the fault, the root cause is determined. The correlation calculation formula is as follows:

[0331]

[0332] Where: C represents the correlation degree of the root factor. Let i be the weight of the i-th potential root cause factor. Let be the correlation coefficient between the i-th potential root cause factor and the fault (determined through training with historical fault data); the factor with the highest correlation is the root cause of the fault.

[0333] Taking the jamming of the sand flushing gate as an example, the potential root causes include: siltation (C1), mechanical wear (C2), and power failure (C3), with weights of w1=0.5, w2=0.3, and w3=0.2, respectively; the correlation coefficient is calculated based on monitoring data: if , , Therefore, C = 0.5 × 0.8 + 0.3 × 0.3 + 0.2 × 0.2 = 0.53, indicating that siltation is the root cause. At the same time, by combining the digital twin model to simulate the siltation location, it is clear that the inducing factors are the excessively high sand content during the flood season and the untimely sand flushing.

[0334] S44: Fault Level Determination and Impact Assessment

[0335] Based on the severity, scope of impact, and effect on the core functions of the system (water diversion, sedimentation, sand flushing, and ecological water supply), the faults are classified into four levels: general faults, major faults, critical faults, and emergency faults, and the priority of handling each level of fault is clearly defined. At the same time, the consequences of the continued development of the fault are simulated through a digital twin model to assess the impact of the fault on system operation, the surrounding environment, and downstream water supply, providing a basis for the formulation of handling plans.

[0336] The formula for determining the fault level is:

[0337] ;

[0338] Where: G is the quantified fault level value, S is the fault impact range, where 0≤S≤1, the larger the range, the larger the S value, I is the degree of impact of the fault on core functions, where 0≤I≤1, the larger the impact, the larger the I value, T is the fault duration (after standardization 0≤T≤1), and α=0.4, β=0.4, γ=0.2 are weighting coefficients; the fault level is classified according to the G value as follows:

[0339] ① General fault: G < 0.3; ② Major fault: 0.3 ≤ G < 0.6; ③ Serious fault: 0.6 ≤ G < 0.9; ④ Emergency fault: G ≥ 0.9.

[0340] The impact assessment of the failure was conducted using a digital twin model. The core assessment indicators were the water diversion efficiency loss rate, the sand flushing effect loss rate, and the ecological flow compliance rate. The assessment formulas were as follows:

[0341] ;

[0342] ;

[0343] ;

[0344] in: This is the normal operating water flow rate. This refers to the water diversion flow rate under fault conditions. This represents the sand flushing efficiency under normal operating conditions. For sand flushing efficiency under fault conditions; For standard ecological flow, This represents the ecological flow under fault conditions. The above formula quantifies the impact of faults on core system functions, providing data support for optimizing response plans.

[0345] S45: Component life prediction;

[0346] Based on structural health data (stress, strain, vibration, displacement), operating time, and environmental factors (water erosion, sediment abrasion, temperature changes), a fatigue damage accumulation algorithm is used to predict the remaining service life of each core component (pneumatic steel dam, overflow dam, sand flushing gate, sand retaining wall). This provides early warning of component aging risks, supports the development of operation and maintenance plans, and avoids system failures caused by component aging. For example, predicting the remaining service life of the pneumatic steel dam's airbags allows for timely replacement, preventing system downtime due to airbag rupture.

[0347] The formula for predicting the remaining service life of a component is as follows:

[0348] ;

[0349] in: The remaining service life of the component (unit: hours). The design service life of the component is given by (unit: h), D is the fatigue damage characteristic (calculated by the formula in step S212), t is the actual operating time (unit: h), E is the water flow scouring intensity (unit: N / m²), and T is the change in ambient temperature (unit: ℃). The attenuation coefficients are determined based on the material of the components; for pneumatic steel dam airbags, k1=0.00001, k2=0.00002, and k3=0.000005.

[0350] Taking the pneumatic steel dam airbag as an example, if (10 years), D=0.3, t=21900h (2.5 years), E=500N / m², T=20℃, then With a remaining service life of approximately 5.8 years, an airbag replacement plan was developed to ensure that the components are replaced before failure.

[0351] This life prediction method uses Miner's fatigue damage accumulation theory, with a prediction error of no more than 5%, and can accurately predict the remaining service life of components, providing reliable support for operation and maintenance optimization.

[0352] Step S5: Adopt an integrated closed-loop collaborative handling mechanism encompassing early warning, diagnosis, control, and operation and maintenance.

[0353] This step, based on the early warning information from step S3 and the fault diagnosis results from step S4, combines the digital twin model and operation and maintenance database from step S2 to construct an integrated closed-loop collaborative system for early warning, diagnosis, control, and operation and maintenance. This system enables rapid fault handling, effective risk prevention and control, and continuous system optimization, forming a closed-loop management cycle of "early warning-diagnosis-control-operation and maintenance-verification-optimization." Specifically, it consists of three sub-steps:

[0354] S51: Adaptive control strategy;

[0355] Based on the warning level and fault type, the system automatically generates an adaptive control scheme, coordinating the control of all components of the system without manual intervention, enabling rapid fault handling. The control response time is no more than 30 seconds, and the control accuracy can reach ±0.1m (gate opening and closing degree, pneumatic steel dam lifting height). The specific control strategy is as follows:

[0356] (1) Siltation warning / fault: Automatically adjust the opening and closing degree of the flushing gate and the lifting height of the pneumatic steel dam to increase the flushing flow rate and optimize the flushing time (such as extending the flushing time and adjusting the flushing frequency) to avoid siltation; for areas with severe siltation, link with drones to provide precise dredging operation guidance, mark the coordinates of the siltation area, or start mechanical dredging equipment to quickly remove silt.

[0357] (2) Pneumatic steel dam failure: If the airbag leaks, the inflation and deflation device will be automatically shut down, the backup airbag will be activated, and the operating status of the pneumatic steel dam will be adjusted (such as reducing the lifting speed and adjusting the dam body angle) to ensure the safety of river flood discharge and water diversion; if the inflation and deflation device fails, it will automatically switch to manual control mode and issue an operation and maintenance warning to remind operation and maintenance personnel to carry out timely repairs.

[0358] (3) Gate failure: If the sand flushing gate / inlet gate is stuck in the opening and closing, the emergency opening and closing device will be automatically activated to remove mud and sand around the gate (such as starting the high-pressure flushing equipment), and the opening and closing force will be adjusted to avoid damage to the gate due to forced opening and closing; if the gate is damaged, the gate operating load will be automatically reduced (such as reducing the opening and closing frequency and adjusting the operating angle) to avoid the damage from expanding.

[0359] (4) Abnormal ecological flow: Automatically adjust the opening and closing degree of the ecological gate to ensure that the downstream ecological flow meets the standard. Combine the water quality monitoring data of the water diversion channel to optimize the ecological water supply plan (such as adjusting the water supply period and water supply flow) and take into account both ecological protection and water conservancy functions.

[0360] Taking the airbag leakage failure of the pneumatic steel dam as an example, the pressure regulation formula after the backup airbag is activated is:

[0361] ;

[0362] in: The actual pressure of the standby airbag at time t (unit: kPa). The standard working pressure for an airbag is 60 kPa. The formula represents the time required for the backup airbag pressure to stabilize (taken as 5 minutes). This formula ensures a smooth rise in backup airbag pressure, preventing sudden pressure increases that could cause secondary damage to the airbag and ensuring the pneumatic steel dam can quickly return to stable operation.

[0363] S52: Precise Operation and Maintenance Solution;

[0364] By combining fault diagnosis results, component lifespan prediction data, and an operation and maintenance knowledge base, the system automatically generates targeted and precise operation and maintenance solutions. These solutions clearly define the operation and maintenance content, processes, required consumables, timeframes, and technical requirements, avoiding blind operation and maintenance, improving efficiency and quality, and reducing costs. The operation and maintenance solutions are divided into three categories: routine operation and maintenance, fault operation and maintenance, and preventative operation and maintenance, covering the entire lifecycle of the system's operation and maintenance needs.

[0365] (1) Daily operation and maintenance: Based on historical operation and maintenance data and component operation status, a periodic daily operation and maintenance plan is formulated, and the operation and maintenance cycle of different components is clarified (e.g., the airbag of the pneumatic steel dam is inspected once every 6 months, and the sand flushing gate opening and closing device is lubricated and maintained once every 3 months). The operation and maintenance content includes equipment cleaning, parameter calibration, component fastening, consumable inspection, etc. The key operation and maintenance parts and operation specifications are marked by digital twin model. The operation and maintenance personnel can complete the daily operation and maintenance according to the model guidance. After the operation and maintenance is completed, the operation and maintenance record is entered into the historical database to realize the traceability of the operation and maintenance process.

[0366] (2) Fault Operation and Maintenance: For faults that have occurred, a special fault operation and maintenance plan is generated based on the fault level and root cause analysis results, which clarifies the fault handling steps, division of responsibilities, safety precautions and acceptance standards. For example, for the airbag leakage fault of the pneumatic steel dam, the operation and maintenance plan clearly requires that the inflation and deflation device be shut down first, the backup airbag be switched, the damaged airbag be removed and replaced with a new airbag, and finally the pressure test and trial operation be carried out to ensure that the fault is completely resolved; for the silt compaction fault, the plan clarifies the scope of dredging, the dredging method (mechanical dredging or high pressure flushing), and the sand flushing verification steps after dredging to avoid recurrence of compaction.

[0367] (3) Preventive maintenance: Based on the component life prediction results and risk warning information, a preventive maintenance plan is formulated in advance. Components that are about to reach the end of their service life and are at risk of aging are replaced or repaired in advance, and the frequency of maintenance is increased in high-risk areas. For example, when the remaining service life of the pneumatic steel dam airbag is predicted to be less than 1 year, an airbag replacement plan is automatically generated, specifying the replacement time, the required airbag model, and the replacement process. Consumables are prepared in advance to avoid sudden failures due to component failure. For high-risk conditions during the flood season, a comprehensive inspection and reinforcement of core components such as the gate body and revetment are carried out in advance to improve the system's risk resistance.

[0368] During the execution of the operation and maintenance plan, the progress and effect of operation and maintenance are monitored in real time through a digital twin model. After the operation and maintenance is completed, the system automatically collects component operation data to verify the operation and maintenance effect. If the fault is not eliminated or the risk is not mitigated after the operation and maintenance, the operation and maintenance plan is automatically adjusted and the operation and maintenance work is restarted to ensure that the operation and maintenance is handled properly.

[0369] S53: Closed-loop verification and continuous optimization;

[0370] A closed-loop verification mechanism of "control-operation-verification-optimization" is constructed to comprehensively verify the effects of adaptive control and operation and maintenance. Based on the verification results, the early warning model, fault diagnosis model, control strategy and operation and maintenance plan are continuously optimized to achieve continuous improvement in the system's operation and management capabilities.

[0371] (1) Effect Verification: After the control and operation are completed, the system will conduct effect verification from three dimensions through multi-source monitoring data and digital twin model simulation: First, risk mitigation verification, whether the monitoring and early warning indicators have returned to the normal range, and whether the risk level has dropped to no risk or general risk; Second, fault handling verification, checking whether the fault has been completely eliminated, whether the component operation status has returned to normal, and whether the fault recurrence rate is less than 0.5%; Third, functional compliance verification, evaluating whether the core functions of the system such as water diversion, sedimentation, sand flushing, and ecological water supply have met the design standards, and whether the water diversion efficiency and sand flushing effect have returned to normal levels. The verification compliance standards are: early warning indicators have returned to the normal range, faults have been completely eliminated, and the core function compliance rate is ≥99%.

[0372] (2) Continuous optimization: Based on the verification results, we will optimize the technical parameters and model algorithms of each link in a targeted manner: First, we will optimize the early warning model, adjust the early warning threshold calibration parameters and the weight of the multi-factor coupling model, improve the accuracy and timeliness of early warning, and reduce the false and missed early warning rates; Second, we will optimize the fault diagnosis model, update the fault knowledge base, adjust the fault feature matching algorithm parameters, and improve the ability to identify new faults and the accuracy of diagnosis; Third, we will optimize the adaptive control strategy, adjust the control parameters and correction coefficients, and improve the control response speed and control accuracy; Fourth, we will optimize the operation and maintenance plan, adjust the operation and maintenance cycle and operation and maintenance content, and combine operation and maintenance costs and effects to achieve reasonable allocation of operation and maintenance resources.

[0373] Meanwhile, verification data and optimization solutions are entered into the historical database to form a data-driven continuous optimization mechanism. As the system runs for a longer period of time, data and experience are continuously accumulated, gradually improving the system's intelligent early warning, fault diagnosis and operation and maintenance management level, ensuring the long-term stable, efficient and safe operation of the system.

[0374] S6. System security protection and data encryption implementation;

[0375] The core of this step is to build a multi-layered system security protection system to protect monitoring equipment, data transmission, digital twin platform, and database, ensuring stable system operation and controllable data security. Specific implementation details are as follows:

[0376] S61. Implementation of Safety Protection Measures

[0377] Multi-layered protection system construction: A four-tiered security protection system is built, consisting of "device layer - transmission layer - platform layer - data layer". Device layer: Monitoring devices use encrypted chips to prevent intrusion and tampering. Transmission layer: Firewalls and intrusion detection systems (IDS) are used to monitor data transmission and prevent data theft and tampering. Platform layer: The digital twin platform employs access control and permission management to prevent unauthorized access. Data layer: The database uses encrypted storage to prevent data leakage.

[0378] Access Control: Set up a device access control system, assign different operation permissions to different maintenance personnel (such as administrator permissions, maintenance personnel permissions, and monitoring personnel permissions). Administrators have full operation permissions, maintenance personnel only have operation-related operation permissions, and monitoring personnel only have data viewing permissions to avoid misoperation and illegal operation;

[0379] Emergency Response Mechanism: In response to extreme weather (heavy rain, floods, typhoons) and sudden failures (large-scale equipment failure, data transmission interruption), the emergency protection plan is automatically activated. In extreme weather, the power supply to outdoor monitoring equipment is automatically shut off and the equipment is moved to a safe area. In the event of a sudden failure, the system automatically switches to the backup system to ensure the normal operation of the core functions of the system. At the same time, an emergency response plan is formulated, which clarifies the emergency response process, responsible personnel, and response measures, and emergency drills are conducted regularly (once every six months).

[0380] S62, Data Encryption Implementation

[0381] End-to-end encryption: The AES-256 encryption algorithm is used to encrypt monitoring data, early warning information, fault data, and operation and maintenance data throughout the entire process (data collection, transmission, storage, and use). The encryption key is changed regularly (once a month) to ensure data security and confidentiality.

[0382] Data Backup: A dual backup mechanism of "local backup + off-site backup" is established. Local backup is performed once a week and stored on the local server. Off-site backup is performed once a month and stored in an off-site backup center. Backup data is retained for one year to prevent data loss. At the same time, incremental backup is adopted to back up newly added data every day to improve backup efficiency.

[0383] Data hierarchical management: Data is divided into core data (such as system structure parameters, fault knowledge base, operation and maintenance plans), important data (such as monitoring data, early warning information), and general data (such as operation and maintenance records, log data). Core data is stored with the highest level of encryption, important data is stored with medium level of encryption, and general data is stored with basic encryption to ensure the security and controllability of core data.

[0384] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent early warning and fault diagnosis of a water diversion and sediment flushing system, characterized in that, include: S1. Construct a comprehensive monitoring system covering air, land, sea, and water: Deploy various types of high-precision monitoring equipment in the core components and surrounding environment of the water diversion sediment flushing system to construct a comprehensive, all-round, and parameter-free monitoring system covering air, land, sea, water, and equipment operation, so as to realize the real-time acquisition, filtering, and stable transmission of data related to the system's operating status. S2. Establish a multi-source data fusion preprocessing and digital twin modeling platform: perform fusion preprocessing on the multi-source monitoring data collected in S1, and at the same time build a full-scale digital twin model of the system based on the structural parameters of the water diversion sediment flushing system, and build a supporting operation and maintenance database to provide data and model support for subsequent early warning and diagnosis. S3, Multi-factor Coupled Intelligent Early Warning Method: Based on the digital twin model built by S2 and the preprocessed multi-source monitoring data, a multi-factor coupled early warning system is constructed to realize hierarchical early warning, trend prediction and precise positioning of system operation risks, and trigger corresponding linkage mechanisms; S4. Full-condition accurate fault diagnosis method: Based on the early warning information of S3, the digital twin model of S2, and multi-source monitoring data and operation and maintenance database, a full-condition fault diagnosis system is constructed to achieve accurate fault identification, type determination, root cause tracing and component life prediction. S5: Adopts an integrated closed-loop collaborative handling mechanism encompassing early warning, diagnosis, control, and operation and maintenance, including: Based on the early warning information of S3 and the fault diagnosis results of S4, combined with the digital twin model of S2 and the operation and maintenance database, an integrated closed-loop collaborative system of early warning, diagnosis, control and operation and maintenance is constructed to achieve rapid fault handling, effective risk prevention and control and continuous system optimization. S6: System Security Protection and Data Encryption: Establish a multi-layered system security protection system to protect monitoring equipment, data transmission, digital twin platform and database. At the same time, encrypt and back up all types of data throughout the process to ensure stable system operation and controllable data security.

2. The intelligent early warning and fault diagnosis method for the water diversion and sediment flushing system according to claim 1, characterized in that, Step S1 includes: S11. Aerial Monitoring: Deploy an autonomous drone inspection system, equipped with high-definition cameras, lidar, and infrared thermal imagers. Following a preset inspection route, it covers the entire area of ​​the water diversion channel, river overflow dam, pneumatic steel dam, sand flushing gate, and bank protection. It regularly inspects system components, collecting image data including the appearance of components, changes in the surrounding topography, and floating objects on the water surface. The inspection frequency can be dynamically adjusted according to the flood season and non-flood season. During the flood season, it is once every 2 hours, and during the non-flood season, it is once every 6 hours. In case of emergency, it can trigger real-time inspection. S12, Ground-to-ground monitoring: Deploy rain-measuring radar and satellite remote sensing receiving modules to collect environmental data in real time, including regional rainfall, precipitation intensity, river water level, and watershed inflow. Combined with satellite remote sensing images, identify potential landslide hazards on riverbanks to achieve watershed-scale environmental and safety monitoring, providing macro-data support for system operation control and risk warning. S13. Ground monitoring: Stress-strain sensors, vibration sensors, displacement sensors, and temperature sensors are installed on the dam / gate bodies of the water diversion channel revetment, overflow dam, pneumatic steel dam, sand flushing gate, intake gate, and ecological gate to collect structural health data, including stress, strain, vibration frequency, displacement, and surface temperature, in real time, and to monitor whether there are scour, damage, deformation, or aging problems in the components; liquid level sensors and flow velocity sensors are installed at the side weirs and sand retaining walls of the water diversion channel to collect hydrological data, including the overflow of the side weir, the water level difference between the upstream and downstream of the sand retaining wall, and the water flow velocity, in real time. S14. Underwater monitoring: Deploy underwater high-definition cameras, acoustic Doppler current meters, sediment concentration sensors, and underwater topographic surveying instruments in the water diversion channel, stilling basin section, seawall section, and downstream of the sand flushing gate to collect hydrological data in real time on underwater sediment thickness, sediment distribution, water flow pattern, sediment particle size distribution, and the integrity of underwater components. Focus on monitoring whether sediment accumulation in the water diversion channel reaches the warning threshold and the sand flushing effect of the sand flushing gate to avoid sand discharge problems caused by sediment compaction. S15. Intelligent Monitoring Terminal: Pressure sensors, current sensors, voltage sensors, and stroke sensors are installed on the air filling and emptying devices of the pneumatic steel dam and the opening and closing devices of the sand flushing gate / intake gate / ecological gate to collect equipment operating parameters in real time, including air filling and emptying pressure, opening and closing current, voltage, stroke position, and operating power; current sensors are installed at the light strips in the water diversion channel to monitor the operating status of the lighting system; all monitoring devices are connected to the intelligent monitoring terminal to realize real-time data acquisition, filtering, encoding, and transmission. The transmission adopts 5G + fiber optic dual-mode redundant transmission to ensure the stability and real-time performance of data transmission and avoid data loss or delay.

3. The intelligent early warning and fault diagnosis method for the water diversion, sedimentation, and sand flushing system according to claim 1, characterized in that, Step S2 includes: S21. Multi-source data fusion preprocessing: Receive multi-type data transmitted from various monitoring devices in step S1, including image data, structural health data, hydrological data, and environmental data. Adaptive filtering algorithms are used to remove data noise. Preprocessing operations such as data alignment, missing value completion, and outlier removal ensure data accuracy and completeness. Feature extraction algorithms are used to extract characteristic parameters such as component crack length, damaged area, and siltation thickness from image data; component fatigue damage characteristics from structural health data; and abnormal flow patterns from hydrological data, achieving standardization and characterization of multi-source data. A three-level multi-source data fusion algorithm is employed, including the following: S211, Level 1 data preprocessing, used for noise suppression and anomaly removal: An adaptive Kalman filter algorithm is used to suppress data noise, and specific filter formulas are designed for different types of monitoring data. For structural health data, namely stress σ, strain ε, and displacement d, the filtering formula is: ； ； ； in: The data value after filtering at time k. Let A be the filtered data value at time k-1, A be the state transition matrix (A=1.02 for structural health data), and B be the control matrix (B=0.01). The control quantity at time k. For Kalman gain, Let be the original monitoring data at time k, Q be the process noise covariance (taken as Q=0.001), P be the covariance matrix, and I be the identity matrix. For hydrological data, namely sediment load S, flow velocity v, and water level h, outliers are removed using the improved 3σ criterion, and the formula is as follows: If satisfied If so, then the data will be retained; If satisfied ,but ; in: This is the average of 100 consecutive historical data sets for this monitoring parameter. Standard deviation, This represents the original monitoring data at time k; S212, Second-level feature fusion, used for feature extraction and standardization: To extract core features from different types of data, min-max standardization is used to eliminate the influence of units. The formula is as follows: ； in: Here, x represents the standardized feature values, and x represents the original feature values. This is the historical maximum value of this feature parameter. This is the historical minimum value of this feature parameter; The following strategy is used for feature extraction: (1) Image data, i.e., data from drones and underwater cameras: The CNN algorithm is used to extract features, with core features including crack length L, damaged area S, and sediment thickness H. The feature extraction discriminant is: ； ； in: The pixel coordinates of the crack edge; , where I = 1 for damaged areas and I = 0 for undamaged areas, and m and n are the number of rows and columns of the image pixel matrix; (2) Structural health data: Extract fatigue damage feature D, using Miner's fatigue damage accumulation theory, the formula is: ； in: Let i be the actual number of cycles under stress level i. Let D be the fatigue life corresponding to the i-th stress level, and k be the number of stress levels. When D≥1, the component is determined to have a risk of fatigue damage. S213, Three-level decision-level fusion, used for weight allocation and fusion output: Design a multi-source data fusion weight allocation strategy based on the analytic hierarchy process, clarify the priority of each monitoring data, and the fusion formula is: ； Where: F is the integrated feature value after fusion. Let w represent the weights of the i-th type of monitoring data, where w1 = 0.2 for aerial monitoring, w2 = 0.15 for ground-to-ground monitoring, w3 = 0.3 for ground monitoring, w4 = 0.25 for underwater monitoring, and w5 = 0.1 for equipment operation monitoring. The weights are determined using the Analytic Hierarchy Process (AHP), and the consistency check CR < 0.

1. It is the sum of standardized eigenvalues ​​of the i-th type of monitoring data; S22, Digital Twin Modeling: Based on the structural design parameters of the water diversion sediment flushing system and combined with pre-processed monitoring data, a full-scale digital twin model of the system was built to achieve a 1:1 high-fidelity mapping between the physical entity and the digital model. The model covers all core components, including the water diversion channel, pneumatic steel dam, overflow dam, flushing gate, and intake gate, as well as the surrounding environment. Real-time monitoring data, historical operation data, component material parameters, and operation and maintenance records were imported into the digital twin model to achieve real-time synchronous updates between the model and the physical system, intuitively displaying the system's operating status, sediment distribution, and component health status information. Synchronization update strategy between digital twin model and physical system: The synchronization error control formula adopts a "real-time interpolation + dynamic correction" mechanism: ； in: For synchronization error, Let k be the parameter values ​​of the digital twin model at time k. Let k be the monitoring value of the physical system at time k; when At that time, the model parameters are corrected using the following formula: ； in: To correct the parameter values ​​of the digital twin model and ensure high-fidelity synchronization between the model and the physical system; S23. Historical Database Construction: Collect long-term monitoring data, fault records, operation and maintenance data, and hydrological and meteorological data of the system to construct a historical database. Combine this with fault cases and operation and maintenance experience of similar water conservancy projects at home and abroad to form a fault knowledge base and an operation and maintenance knowledge base, providing data support for intelligent early warning, fault diagnosis and operation and maintenance optimization.

4. The intelligent early warning and fault diagnosis method for the water diversion, sedimentation, and sand flushing system according to claim 1, characterized in that, Step S3 includes: S31. Construction of an early warning indicator system: Based on the operational characteristics of the water diversion sedimentation and flushing system, an early warning indicator system covering 5 major categories and 28 early warning indicators is constructed, including: Sedimentation indicators: sedimentation thickness in the water diversion channel, sedimentation rate, water level difference between upstream and downstream of the sand retaining wall, sediment content in the discharge from the sand flushing gate, and degree of sediment compaction; Health indicators of components: stress and strain, displacement, and airbag pressure of pneumatic steel dams; crack length and damaged area of ​​overflow dams and water diversion channel revetments; opening and closing travel deviation and gate vibration frequency of sand flushing gates / intake gates / ecological gates; and scour depth of sand retaining walls. Equipment operation indicators: pneumatic steel dam filling and venting pressure, filling and venting time, sand flushing gate / intake gate / ecological gate opening and closing current, voltage, and power, filling and venting device operating status, and LED strip operating current; Hydrological and environmental indicators: river level, inflow, sediment content, rainfall, precipitation intensity, overflow of the diversion channel side weir, stilling basin level, and flow velocity of the floodplain. Ecological indicators: ecological gate discharge flow, downstream ecological flow compliance rate, dissolved oxygen concentration in the water diversion channel, and water quality parameters; S32. Dynamic calibration of early warning thresholds: Utilizing machine learning algorithms and historical database data, dynamic threshold calibration is performed on each early warning indicator, overcoming the limitations of traditional fixed thresholds. Based on seasonal changes, inflow sediment content, and system operating conditions (including water diversion, sediment flushing, flood discharge, and ecological water supply), the early warning thresholds for each indicator are automatically adjusted to achieve adaptive optimization. The adaptive threshold calibration formula, using the sediment deposition thickness early warning threshold as an example, is as follows: ； in: The dynamic early warning threshold for sediment deposition thickness at time t, in meters; The baseline threshold for siltation thickness is 1.2m during the non-flood season and 0.8m during the flood season. The sediment content of the incoming water at time t, in kg / m³. The inflow rate at time t is expressed in m³ / s. The correction coefficients are: θ=0.1 for water diversion, θ=0.3 for sand flushing, θ=0.2 for flood discharge, and θ=0.15 for ecological water supply; α, β, and γ are weighting coefficients: α=0.0002, β=0.0001, and γ=0.1, determined through training with historical data. S33. Multi-factor Coupled Early Warning Model: Based on the fusion algorithm of BP neural network and LSTM, a multi-factor coupled early warning model is constructed. The input is preprocessed multi-source monitoring data and real-time status data of digital twin model, and the output is the risk level of each early warning indicator. The model can capture the coupling relationship between various indicators and realize the comprehensive assessment of risk. The core formulas and operating mechanism of the model are as follows: (1) Output layer of BP neural network: ； in: Let y_j be the comprehensive evaluation value of the j-th early warning indicator (0≤y_j≤1). Let be the connection weight between the i-th neuron in the input layer and the j-th neuron in the output layer. Let be the input value of the i-th neuron in the input layer. σ is the bias of the j-th neuron in the output layer, and σ is the activation function; (2) LSTM trend prediction layer: ； ； ； ； ； in: For input gate, For the Gate of Oblivion For output gate, In cellular state, For output of the hidden layer, The input data is at time t. For the input weight matrix, The hidden layer weight matrix is... The term is the bias term, and ⊙ represents the Hadamard product; (3) Risk Level Determination Mechanism: Combining the comprehensive assessment value output by BP and the trend prediction value output by LSTM, a risk level determination rule is designed, and the formula is as follows: ； Where: R is the quantified risk level, and λ is the weighting coefficient. Risk trend change rate; risk level is determined based on R value: ① No risk: R < 0.2; ② Moderate risk: 0.2 ≤ R < 0.4; ③ Significant risk: 0.4 ≤ R < 0.7; ④ Major risk: R ≥ 0.7; S34. Trend Early Warning and Precise Positioning: The LSTM algorithm is used to predict the changing trends of various early warning indicators, allowing for the prediction of risk development direction and evolution speed 12-24 hours in advance, avoiding early warning delays. Combined with a digital twin model, the specific location of the risk, the components involved, and the scope of impact are precisely located. The formula for controlling the prediction accuracy of trend early warning is: ； Where: δ is the prediction error; Δt is the prediction lead time, Δt∈[12,24], in hours; This is the predicted value Δt hours in advance; The actual monitoring value is t+Δt hours; when δ>5%, the prediction accuracy is improved by adjusting the hidden layer weights of the LSTM model. Precise positioning mechanism: Based on the spatial coordinate mapping of the digital twin model, the coordinates (x, y, z) of the monitoring point corresponding to the early warning indicator are matched with the coordinates of the model components. The positioning formula is as follows: ； Where: D is the spatial distance between the monitoring point and the model component, (x,y,z) are the actual coordinates of the monitoring point ... m ,y m ,z m () represents the coordinates of the component in the digital twin model; when D≤0.5m, the component is determined to be a risky component, achieving precise positioning; S35. Early Warning Issuance and Linkage: Based on the risk level, early warning information is automatically generated and issued to maintenance personnel through multiple channels, including SMS, audible and visual alarms, central control room pop-ups, and mobile APP. For major risks, an emergency linkage mechanism is automatically triggered to suspend the operation of relevant equipment and initiate emergency water supply and emergency sand flushing plans to prevent the risk from escalating. At the same time, the early warning information is integrated into a digital twin model to intuitively display the risk distribution and development trend. Emergency Response Trigger Logic and Control Parameter Calculation: Taking the major risk of siltation as an example, the formula for adjusting the opening and closing degree of the silt flushing gate is as follows: ； in: Let t be the opening and closing degree of the sand flushing gate. κ is the baseline opening / closing degree, and κ is the adjustment coefficient. Let t be the actual thickness of the sediment deposit. The dynamic early warning threshold ensures that the sand flushing flow rate meets the sand discharge requirements and quickly alleviates the risk of siltation.

5. The intelligent early warning and fault diagnosis method for the water diversion, sedimentation, and sand flushing system according to claim 1, characterized in that, Step S4 includes: S41. Fault Type Classification; Based on the structure and operational characteristics of the water diversion sedimentation and flushing system, common fault types of the system are identified and classified into 6 major categories and 32 specific faults, including: Failures of pneumatic steel dams: airbag leakage, inflation and deflation device malfunction, shield plate jamming, limit band damage, excessive stress, and abnormal displacement; Gate malfunctions: Sand flushing gate / intake gate / ecological gate jamming, gate body damage, aging seals, opening and closing device malfunction, excessive travel deviation; Sediment-related faults: excessive sediment accumulation in the water diversion channel, sediment compaction, erosion and damage to the sand retaining wall, and poor sand flushing effect of the sand flushing gate; Structural damage and failures: cracks and damage to the revetment of the water diversion channel, scouring and damage to the overflow dam, and damage to the stilling basin and the anti-scouring structure of the seawall section; Equipment malfunctions: power supply failure of the charging and discharging device, motor failure of the opening and closing device, monitoring equipment failure, and light strip failure of the water diversion channel; Ecological and hydrological anomalies: abnormal flow at the ecological gate, insufficient downstream ecological flow, abnormal water quality in the water diversion channel, and turbulent water flow. S42. Fault Feature Matching: Using deep learning algorithms, the monitoring data features under fault conditions are extracted and compared with the fault feature templates in the fault knowledge base to achieve accurate identification of fault types with an accuracy rate of no less than 98%. For new faults, the fault features are automatically learned through machine learning algorithms to update the fault knowledge base and achieve self-evolution of fault identification. This step uses a fault feature similarity matching algorithm to clarify the specific details and judgment criteria for fault identification. The formula is as follows: ； in: The similarity between the fault features to be identified and the fault knowledge base template is used. Let i be the i-th feature parameter of the fault to be identified. Let be the i-th feature parameter of the fault template, and n be the number of feature parameters; the discrimination rule is: ①When When the fault is identified as this type, it is successfully identified. ② When 0.7≤ When the fault is suspected to be of this type, manual review is triggered. ③When When a new type of fault is identified, its characteristics are automatically learned and the fault knowledge base is updated. Details of typical fault identification: (1) Pneumatic steel dam airbag leakage fault: The characteristic parameter is the rate of change of inflation and deflation pressure. Current change The discriminant is: and At the same time, satisfying similarity The cause was determined to be an airbag leak; (2) Sand flushing gate opening and closing jamming fault: The characteristic parameter is the opening and closing stroke deviation. Power of start and stop motor The discriminant is: and ( (Rated power), while also satisfying similarity. The condition was determined to be a jammed opening / closing mechanism. (3) Sedimentation failure in the water diversion channel: The characteristic parameter is the rate of change of sediment thickness. The discriminant for the water flow velocity v is: and At the same time, satisfying similarity It was determined to be mud and sand compaction; S43. Trace the root cause of the fault; Based on the digital twin model, the entire process of a fault is simulated. Combined with multi-source monitoring data, the root cause of the fault can be traced, avoiding only checking the surface faults and ignoring the root cause. Fault root cause tracing employs causal chain analysis combined with a data verification mechanism, using the following formula: ； Where: C represents the correlation degree of the root factor. Let i be the weight of the i-th potential root cause factor. Let be the correlation coefficient between the i-th potential root cause factor and the failure; the factor with the highest correlation is the root cause of the failure. S44. Fault Level Determination and Impact Assessment; Based on the severity, scope of impact, and effect on the core functions of the system, the faults are classified into four levels: general faults, major faults, critical faults, and emergency faults. The consequences of the continued development of the fault are simulated through a digital twin model to assess the impact of the fault on system operation, the surrounding environment, and downstream water supply, providing a basis for the formulation of disposal plans. The formula for determining the fault level is: ； Where: G is the quantified fault level value, S is the fault impact range, where 0≤S≤1, the larger the range, the larger the S value, I is the degree of impact of the fault on core functions, where 0≤I≤1, the larger the impact, the larger the I value, T is the fault duration, after standardization 0≤T≤1, α=0.4, β=0.4, γ=0.2 are weighting coefficients; fault levels are classified according to the G value: ① General fault: G < 0.3; ② Major fault: 0.3 ≤ G < 0.6; ③ Critical fault: 0.6 ≤ G < 0.9; ④ Emergency fault: G ≥ 0.9; The impact assessment of the failure was conducted using a digital twin model, with the core assessment indicator being the water diversion efficiency loss rate. Sand washing effect loss rate Ecological flow compliance rate The evaluation formula is: ； ； ； in: This is the normal operating water flow rate. This refers to the water diversion flow rate under fault conditions. This represents the sand flushing efficiency under normal operating conditions. For sand flushing efficiency under fault conditions; For standard ecological flow, Ecological flow under fault conditions; S45. Component life prediction: Based on the structural health data, operating time, and environmental factors of the components, a fatigue damage accumulation algorithm is used to predict the remaining service life of each core component, provide early warning of component aging risks, and support the formulation of operation and maintenance plans. The formula for predicting the remaining service life of a component is as follows: ； in: The remaining service life of the component. Where D is the design service life of the component, t is the fatigue damage characteristic (calculated by the formula in step S212), E is the actual operating time, and T is the water flow scouring intensity. This is the attenuation coefficient.

6. The intelligent early warning and fault diagnosis method for the water diversion, sedimentation, and sand flushing system according to claim 1, characterized in that, Step S5 includes: S51, Adaptive Control Strategy: Based on the warning level and fault type, an adaptive control scheme is automatically generated, linking all components of the system for coordinated control, achieving rapid fault handling without manual intervention. (1) Siltation warning / fault: Automatically adjust the opening and closing degree of the flushing gate and the lifting height of the pneumatic steel dam to increase the flushing flow rate and optimize the flushing time to avoid siltation; for areas with severe siltation, link with drones to provide precise dredging operation guidance, or start mechanical dredging equipment. (2) Pneumatic steel dam failure: If the airbag leaks, the inflation and deflation device will be automatically shut down, the backup airbag will be activated, and the operating status of the pneumatic steel dam will be adjusted to ensure the safety of river flood discharge and water diversion; if the inflation and deflation device fails, it will automatically switch to manual control mode and issue an operation and maintenance warning. (3) Gate failure: If the sand flushing gate / inlet gate is stuck in the opening and closing, the emergency opening and closing device will be automatically activated to clear the mud and sand around the gate and adjust the opening and closing force; if the gate is damaged, the operating load of the gate will be automatically reduced to prevent the damage from expanding. (4) Abnormal ecological flow: Automatically adjust the opening and closing degree of the ecological gate to ensure that the downstream ecological flow meets the standard, and optimize the ecological water supply plan in combination with the water quality monitoring data of the water diversion channel; S52, intelligent generation of operation and maintenance solutions; Based on fault diagnosis results and component life prediction data, a personalized operation and maintenance plan is automatically generated, clearly defining the operation and maintenance content, time, process, and required consumables and equipment, thus achieving refined and intelligent operation and maintenance; the operation and maintenance time optimization formula is: ； in: For optimal operation and maintenance time, For maintenance costs, To mitigate system downtime losses, the optimal maintenance time window for different fault types is determined through training with historical data. S53, Closed-loop verification and optimization; After fault handling and maintenance are completed, system operation data is collected through the monitoring system to verify the handling effect. If the fault is not completely resolved, the handling plan is automatically adjusted until the fault is eliminated. At the same time, the fault handling process and maintenance data are entered into the historical database and fault knowledge base to optimize the early warning model and fault diagnosis model, realizing the system's self-learning and self-optimization, and continuously improving the accuracy of early warning and diagnosis. The formula for determining closed-loop verification is: ； Where: ε represents the deviation of the treatment effect. This represents the comprehensive characteristic value after fault handling. This represents the comprehensive characteristic value under normal operating conditions. When ε > 3%, the treatment plan is adjusted, and the treatment process is repeated until the requirements are met.

7. The intelligent early warning and fault diagnosis method for the water diversion and sediment flushing system according to claim 1, characterized in that, Step S6 includes: S61 Security Protection: Establish a multi-layered security protection system to protect monitoring equipment, data transmission, and digital twin platform, preventing equipment intrusion and data tampering; set up equipment access management, assigning different operation permissions to different maintenance personnel to avoid misoperation; establish an emergency response mechanism to automatically activate emergency protection plans in response to extreme weather and sudden failures, ensuring the safe and stable operation of the system; S62 Data Encryption: Employs the AES encryption algorithm to encrypt monitoring data, early warning information, fault data, and operation and maintenance data throughout the entire process, ensuring data security and confidentiality; establishes a data backup mechanism to regularly back up data and prevent data loss; and implements hierarchical data management to ensure the security and controllability of core data.

Images