A channel siltation probability early warning method and system based on water flow characteristic identification

Abstract

The present invention discloses a channel siltation probability early warning method and system based on flow feature identification, relating to the field of channel maintenance technology. The method comprises: obtaining channel characteristic parameters and inputting them into an interactive knowledge graph to determine a set of key flow characteristic indicators; performing random feature combination based on this indicator set, screening K coordinated feature groups based on historical operating data, training K benchmark siltation prediction models, and integrating them to generate a siltation probability prediction model; collecting real-time flow data through sensors, processing it, and inputting it into the model to predict the channel siltation probability; and triggering a siltation warning if the prediction results meet the warning conditions. This method thereby achieves the technical effects of improving the foresight of early warnings, reducing the error rate, and enhancing adaptability to complex environments.

Description

Technical Field

[0001] The present invention relates to the technical field of channel maintenance, and in particular to a channel siltation probability early warning method and system based on water flow feature identification. Background Art

[0002] Traditional channel monitoring methods primarily focus on monitoring basic channel parameters such as water level and flow rate, often lacking comprehensive analysis of the interrelationships between these parameters and their deeper connections to channel siltation. For example, some existing channel monitoring systems simply install simple water level gauges and flow meters within the channel to obtain single-point water level and flow rate data, then perform simple siltation assessments based on empirical thresholds. This method struggles to accurately identify early, mild channel siltation, often only detecting it after it has become severe and is significantly impacting the channel's normal function.

[0003] Existing technologies have many shortcomings in monitoring channel siltation. On the one hand, they only rely on a small number of single water flow parameters for analysis, which cannot fully reflect the complex water flow behavior in the channel. They are easily affected by changes in local water flow conditions or accidental factors, resulting in frequent misjudgments. On the other hand, the lack of effective use and in-depth mining of historical data makes the model's predictive ability and adaptability poor, and cannot meet the urgent needs of modern water conservancy projects for early and accurate warning of channel siltation. Summary of the Invention

[0004] The present invention provides a channel siltation probability warning method and system based on water flow feature identification to solve the technical problems in the existing technology, such as lack of early warning capability, high misjudgment rate and inability to adapt to complex and changeable channel operating conditions, and achieve the technical effects of improving the foresight of warning, reducing the misjudgment rate and enhancing adaptability to complex environments.

[0005] In a first aspect, the present invention provides a method for early warning of channel siltation probability based on water flow feature identification, wherein the method comprises:

[0006] The characteristic parameters of the target channel are obtained and input into a pre-built interactive knowledge graph to determine a set of key water flow characteristic indicators used to characterize the water flow behavior in the channel.

[0007] Taking the key water flow characteristic index set as the feature space, random feature combination with replacement is performed, and the feature combination results are screened based on historical channel operation data to determine K coordinated feature groups.

[0008] Taking the K coordinated feature groups as input variable groups, K benchmark sedimentation prediction models are constructed and trained, and the K benchmark sedimentation prediction models are integrated to obtain a sedimentation probability prediction model.

[0009] Based on the key water flow characteristic indicator set, sensors are deployed and water flow monitoring is performed to obtain real-time water flow monitoring data, which is then processed through feature engineering and input into the siltation probability prediction model to predict the siltation probability of the target channel.

[0010] Based on the siltation probability prediction results, it is determined whether the target channel has siltation risk, and a siltation probability warning is issued when the risk meets the preset conditions.

[0011] In one feasible implementation, characteristic parameters of the target channel are obtained and input into a pre-built interactive knowledge graph to determine a set of key flow characteristic indicators for characterizing the channel flow behavior, including:

[0012] The channel structure parameters and hydraulic design parameters in the target area are obtained and input into the interactive knowledge graph, triggering the knowledge graph reasoning process based on semantic reasoning and entity relationship matching.

[0013] According to the knowledge graph reasoning result, the key water flow behavior characteristic indicators corresponding to the target channel are identified to form the key water flow characteristic indicator set.

[0014] In a feasible implementation, the key flow characteristic indicator set is used as the feature space, and random feature combinations with replacement are performed. The feature combination results are screened based on historical channel operation data to determine K coordinated feature groups, including:

[0015] A principal component analysis is performed on the key water flow characteristic indicator set, and the number of characteristic indicators that meet a preset contribution threshold is defined as a random selection constraint.

[0016] A random number generator is configured based on the random selection constraint.

[0017] The random number generator is activated to generate a single selection quantity n, and n key water flow characteristic indicators are randomly selected from the key water flow characteristic indicator set and stored as a characteristic combination.

[0018] Iteratively obtain N feature combinations, and evaluate the combined synergy of the N feature combinations in combination with the historical channel operation data, select the top K feature combinations with high combined synergy, and output them as K coordinated feature groups, where N is greater than n and K is less than N.

[0019] In a feasible implementation, K coordinated feature groups are used as input variable groups to construct and train K benchmark sedimentation prediction models, and the K benchmark sedimentation prediction models are integrated to obtain a sedimentation probability prediction model, including:

[0020] Extracting K coordination promotion degrees of the K coordination feature groups.

[0021] Based on the K coordination promotion degrees and combined with the historical channel operation data, K groups of lightweight prediction models are correspondingly constructed and trained.

[0022] The lightweight prediction model with the best performance in each group is selected, K benchmark sedimentation prediction models are obtained for ensemble learning, and the sedimentation probability prediction model is constructed.

[0023] In a feasible implementation, sensor deployment and water flow monitoring are performed based on the key water flow characteristic indicator set to obtain real-time water flow monitoring data, including:

[0024] Based on the key water flow characteristic indicator set, the monitoring requirements of each characteristic indicator are determined, and the corresponding type of sensors are configured and established.

[0025] The flow velocity data of each cross-sectional direction of the target channel is obtained through the flow velocity sensor, the water level change data of each typical section is obtained through the water level sensor, the pressure pulsation signal is collected through the turbulence sensor to construct a turbulence dynamic feature sequence, and the suspended matter concentration and bed sand concentration in the water body are obtained through the sediment sensor, and the output is the real-time water flow monitoring data.

[0026] In a feasible implementation, the feature engineering process is performed and then input into the siltation probability prediction model to predict the siltation probability of the target channel, including:

[0027] The key water flow characteristic indicator set is decomposed and classified into intuitive water flow characteristic indicators and inferred water flow characteristic indicators.

[0028] Based on the pre-processed real-time water flow monitoring data, intuitive water flow characteristic parameters corresponding to the intuitive water flow characteristic indicators are extracted.

[0029] The estimated water flow characteristic parameters corresponding to the estimated water flow characteristic index are calculated based on the intuitive water flow characteristic parameters.

[0030] The intuitive water flow characteristic parameters and the inferred water flow characteristic parameters are fused, and the fusion results are processed through a feature engineering method to obtain real-time input variables.

[0031] The real-time input variable is transmitted to the input end of the sedimentation probability prediction model, and the sedimentation probability prediction result is obtained from the output end.

[0032] In a feasible implementation, based on the siltation probability prediction results, it is determined whether the target channel has siltation risk, and a siltation probability warning is issued when the risk meets the preset conditions, including:

[0033] Set a siltation probability risk classification threshold that is compatible with the channel operation safety level, and establish siltation risk judgment criteria.

[0034] The sedimentation probability prediction result is compared with the sedimentation risk judgment criterion.

[0035] When the sedimentation probability result exceeds the threshold value corresponding to the discrimination criterion, it is determined that there is a sedimentation risk.

[0036] An early warning instruction is generated according to the siltation risk level, and siltation probability early warning information is sent to the channel operation management end through the early warning interface.

[0037] In a second aspect, the present invention further provides a channel siltation probability early warning system based on water flow feature identification, wherein the channel siltation probability early warning system based on water flow feature identification comprises:

[0038] The characteristic index determination module is used to obtain the characteristic parameters of the target channel and input the characteristic parameters into a pre-built interactive knowledge graph to determine the key water flow characteristic index set used to characterize the channel water flow behavior.

[0039] The coordinated feature group screening module is used to perform random feature combination with replacement based on the key water flow characteristic indicator set as the feature space, and screen the feature combination results based on historical channel operation data to determine K coordinated feature groups.

[0040] The prediction model construction and integration module is used to construct and train K benchmark sedimentation prediction models using the K coordinated feature groups as input variable groups, and integrate the K benchmark sedimentation prediction models to obtain a sedimentation probability prediction model.

[0041] The water flow monitoring and siltation probability prediction module is used to deploy sensors and monitor water flow based on the key water flow characteristic indicator set, obtain real-time water flow monitoring data, and input it into the siltation probability prediction model after feature engineering processing to predict the siltation probability of the target channel.

[0042] The sedimentation risk judgment and early warning module is used to determine whether there is a sedimentation risk in the target channel based on the sedimentation probability prediction results, and to issue a sedimentation probability early warning when the risk meets the preset conditions.

[0043] The present invention discloses a channel sedimentation probability early warning method and system based on water flow feature identification, comprising: obtaining characteristic parameters of a target channel, inputting the characteristic parameters into a pre-constructed interactive knowledge graph, and determining a key water flow characteristic indicator set for characterizing channel water flow behavior; using the key water flow characteristic indicator set as a feature space, performing a random feature combination with replacement, and screening the feature combination results based on historical channel operation data to determine K coordinated feature groups; using the K coordinated feature groups as input variable groups, respectively, constructing and training K benchmark sedimentation prediction models, and integrating the K benchmark sedimentation prediction models to obtain a sedimentation probability prediction model. ; Based on the key water flow characteristic indicator set, sensors are deployed and water flow monitoring is carried out to obtain real-time water flow monitoring data, and the data is input into the siltation probability prediction model after feature engineering processing to predict the siltation probability of the target channel; based on the siltation probability prediction result, it is judged whether there is a siltation risk in the target channel, and a siltation probability warning is issued when the risk meets the preset conditions. The channel siltation probability warning method and system for water flow characteristic identification disclosed in the present invention solve the technical problems of lack of early warning capability, high misjudgment rate and inability to adapt to complex and changeable channel operating conditions, and achieve the technical effects of improving the foresight of warning, reducing the misjudgment rate and enhancing adaptability to complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 The figure is a flow chart of a method for early warning of channel siltation probability based on water flow characteristic identification according to the present invention.

[0045] Figure 2 This is a structural diagram of a channel siltation probability early warning system for water flow feature identification according to the present invention.

[0046] Explanation of the accompanying symbols: characteristic index determination module 11, coordination feature group screening module 12, prediction model construction and integration module 13, water flow monitoring and siltation probability prediction module 14, siltation risk judgment and early warning module 15. DETAILED DESCRIPTION

[0047] The above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods of the specification to better understand the above technical solution. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments of the present invention. It should be understood that the present invention is not limited to the example embodiments used only to explain the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In addition, it should be noted that, for the convenience of description, only the parts related to the present invention, rather than all, are shown in the drawings.

[0048] Example 1, as Figure 1The flowchart of a method for early warning of channel siltation probability based on water flow feature identification according to the present invention is as follows:

[0049] The characteristic parameters of the target channel are obtained and input into a pre-built interactive knowledge graph to determine a set of key water flow characteristic indicators used to characterize the water flow behavior in the channel.

[0050] Specifically, "channel characteristic parameters" refer to a set of parameters that can reflect the physical structure and operating status of the target channel, including but not limited to channel cross-sectional shape, channel bottom slope, channel width, channel roughness, water depth, water flow velocity, flow rate, water level, sediment concentration, etc.

[0051] Specifically, the "interactive knowledge graph" refers to a multi-entity, multi-relationship knowledge network structure constructed based on channel operation data, monitoring data, and historical sedimentation events. Nodes represent entities such as channels, monitoring points, sensors, and historical events, while edges represent physical, logical, or causal relationships between entities. Various characteristic parameters can be stored and retrieved. By treating the various characteristic parameters of channels and water bodies as entity nodes and defining the relationships between these entities (such as semantic connections like "influence" and "association"), a semantic network capable of logical reasoning is formed.

[0052] The specific "key water flow characteristic indicator set" refers to a set of characteristic indicators that can sensitively reflect changes in channel water flow behavior and are highly correlated with siltation risks, such as flow velocity distribution, water level change rate, turbulent pulsation intensity, sediment concentration, etc.

[0053] Through the above-mentioned interactive knowledge graph, efficient screening from massive channel characteristic parameters to precise key water flow characteristics is achieved, avoiding the one-sidedness of manual experience judgment, and providing a highly relevant and high-value data foundation for the subsequent model construction based on these characteristics, thereby ensuring the accuracy and reliability of the entire siltation probability warning method.

[0054] In some embodiments, characteristic parameters of a target channel are obtained and input into a pre-built interactive knowledge graph to determine a set of key flow characteristic indicators for characterizing the flow behavior of the channel, including:

[0055] The channel structure parameters and hydraulic design parameters in the target area are obtained and input into the interactive knowledge graph, triggering the knowledge graph reasoning process based on semantic reasoning and entity relationship matching; according to the knowledge graph reasoning results, the key water flow behavior characteristic indicators corresponding to the target channel are identified to form the key water flow characteristic indicator set.

[0056] Specifically, channel structural parameters refer to parameters related to the channel's geometry, such as width, depth, and slope coefficient, while hydraulic design parameters cover the channel's design flow rate, design velocity, and roughness. These two parameters are used to describe the static characteristics of the channel and the expected dynamic flow characteristics from different dimensions.

[0057] Specifically, the interactive knowledge graph contains a vast amount of water conservancy knowledge and logical rules. It pre-constructs and organizes and stores various water conservancy concepts and their interrelationships, such as the relationship between channel slope and flow velocity, and the relationship between flow rate and channel capacity. This interactive knowledge graph efficiently identifies characteristic indicators applicable to channels of different types and parameters.

[0058] Specifically, the system first obtains the structural parameters of the target area's channel, such as a channel width of 5m, a depth of 3m, and a slope coefficient of 1.2. It also obtains hydraulic design parameters, such as a design flow of 10m³ / s, a design velocity of 1.5m / s, and a roughness of 0.02. These parameters are then input as attribute information into a pre-built interactive knowledge graph. After receiving this information, the knowledge graph automatically retrieves historical data, expert rules, and typical water flow behavior patterns associated with these parameters through semantic reasoning and entity relationship matching mechanisms, triggering the knowledge graph's reasoning process. For example, the system performs semantic reasoning based on the relationship between "small channel bottom slope + large roughness" and "historical high incidence of siltation." Based on internally stored hydraulic knowledge rules, it determines which locations, given a given channel width, depth, and design velocity, are prone to abnormally low flow velocity, leading to sediment accumulation. It then further identifies key water flow behavior characteristic indicators, such as velocity distribution characteristics at bends and the frequency of water level fluctuations at specific sections. This constitutes a set of key water flow characteristic indicators, providing a feature foundation for subsequent siltation probability warnings.

[0059] Through this process, we can achieve structured management and semantic association of channel structural parameters and hydraulic design parameters. Leveraging the reasoning capabilities of knowledge graphs, we can automatically filter and output a set of key flow characteristic indicators highly correlated with channel siltation risk. This helps improve the intelligent and scientific nature of channel flow behavior feature identification, provides efficient and accurate feature input for channel siltation probability warning models, and enhances the accuracy and reliability of siltation risk warnings.

[0060] Taking the key water flow characteristic index set as the feature space, random feature combination with replacement is performed, and the feature combination results are screened based on historical channel operation data to determine K coordinated feature groups.

[0061] Specifically, the key water flow characteristic indicator set is a group of parameters that have a significant characterization effect on channel water flow behavior, which are screened out through the interactive knowledge graph. These parameters can effectively reflect the channel water flow state and the potential degree of siltation risk, such as the characteristic value of flow velocity distribution, water level change rate, and mean turbulence pulsation intensity.

[0062] Specifically, the feature space is a multidimensional parameter space constructed with key water flow characteristic indicators as dimensions, that is, a selection space for random feature combinations that can be replaced. When constructing feature combinations, the same feature indicator is allowed to be selected multiple times in different combinations, thereby generating diversified feature combinations from limited feature indicators, thereby increasing the probability of screening out high-quality feature combinations.

[0063] Specifically, the coordinated feature group refers to those feature combinations that perform well in the combined synergy evaluation. Their combination can demonstrate good characterization ability and prediction potential for channel siltation conditions under the verification of historical channel operation data.

[0064] In some embodiments, the key water flow characteristic indicator set is used as the feature space, and random feature combination with replacement is performed. The feature combination results are screened based on historical channel operation data to determine K coordinated feature groups, including:

[0065] Perform principal component analysis on the key water flow characteristic indicator set, define the number of characteristic indicators that meet a preset contribution threshold as a random selection constraint; configure a random number generator based on the random selection constraint; activate the random number generator to generate a single selection quantity n, and randomly select n key water flow characteristic indicators from the key water flow characteristic indicator set, and store them as a feature combination; iteratively obtain N of the feature combinations, and evaluate the combined synergy of the N feature combinations in combination with the historical channel operation data, select the top K feature combinations with high combined synergy, and output them as K coordinated feature groups, where N is greater than n and K is less than N.

[0066] Specifically, the synergy of feature combinations refers to the synergistic improvement effect of each feature combination in the channel siltation probability prediction task, which is usually measured by the proportion of prediction performance improvement after combination; K coordinated feature groups refer to the top K groups of feature indicator subsets screened from all random feature combinations and showing high synergy in historical channel operation data.

[0067] Specifically, during implementation, principal component analysis is first performed on a set of key flow characteristic indicators (e.g., including eight items: flow coefficient of variation, velocity distribution, water level gradient, and sediment concentration) to calculate the contribution of each feature. If the number of features required to achieve a cumulative contribution of 90% is 5, "n ≤ 5" is used as a random selection constraint. Then, based on this constraint, a random number generator is configured to generate a selection quantity n (e.g., n = 3, 4, or 5) each time it is activated. Then, n indicators are randomly selected from the set of key flow characteristic indicators to form a feature combination. These selected indicators can be reused in subsequent selections, allowing the same feature to be included in different combinations multiple times. This process is repeated to iteratively obtain N feature combinations (e.g., N = 50, n = 3 to 5).

[0068] Furthermore, using historical channel operation data, the performance of each feature combination in the sedimentation probability prediction model is evaluated and the combination synergy is calculated. This synergy can be measured through metrics such as the increase in area under the curve, accuracy, and correlation coefficient. Finally, based on these synergy evaluation results and the satisfaction of screening constraints (e.g., selecting the top K feature combinations with the highest synergy), K coordinated feature groups are output. These coordinated feature groups best reflect the inherent connection between channel flow characteristics and sedimentation conditions.

[0069] Through the above process, using a random feature combination method with replacement, we can efficiently explore multiple potential feature combination patterns within a set of key flow characteristic indicators. Through screening and verification with historical data, we can accurately identify feature groups that are highly coordinated and representative for channel sedimentation prediction. This not only enriches the diversity of feature combinations and improves the comprehensiveness of feature screening, but also provides high-quality, multi-dimensional feature input for the subsequent construction of a benchmark sedimentation prediction model, helping to improve the model's prediction accuracy and generalization capabilities, thereby achieving early and accurate warning of sedimentation risks under complex and changing channel operating conditions.

[0070] Taking the K coordinated feature groups as input variable groups, K benchmark sedimentation prediction models are constructed and trained, and the K benchmark sedimentation prediction models are integrated to obtain a sedimentation probability prediction model.

[0071] Specifically, the coordinated feature group refers to a feature group obtained by random feature combination and historical data screening. The features within each group are highly synergistic and complementary in characterizing the channel siltation condition, and can comprehensively reflect the relationship between channel water flow behavior and siltation risk from different dimensions. In other words, the feature combination relationship of the coordinated feature group helps to improve prediction accuracy.

[0072] Specifically, the baseline sedimentation prediction model is constructed based on a single coordinated feature set. It utilizes lightweight machine learning algorithms (such as logistic regression, decision trees, K-nearest neighbors, and shallow neural networks) to ensure rapid model training and efficient prediction capabilities. By integrating the prediction results of multiple baseline models (e.g., using weighted averaging and voting mechanisms), a comprehensive sedimentation probability prediction model is generated. This model, through the fusion of diverse models, improves prediction accuracy and stability.

[0073] In some embodiments, K coordination feature groups are used as input variable groups to construct and train K benchmark sedimentation prediction models, and the K benchmark sedimentation prediction models are integrated to obtain a sedimentation probability prediction model, including:

[0074] Extract K coordination promotion degrees of K coordination feature groups; based on the K coordination promotion degrees and combined with the historical channel operation data, construct and train K groups of lightweight prediction models accordingly; select the lightweight prediction model with the best performance in each group, obtain K benchmark sedimentation prediction models for ensemble learning, and construct the sedimentation probability prediction model.

[0075] Specifically, the coordination promotion degree is first calculated for each coordination feature group. Assuming K = 3, the three coordination feature groups are Group 1 (flow velocity, water level), Group 2 (sediment concentration, turbulence intensity), and Group 3 (flow velocity, sediment concentration). A preliminary model is trained using historical data, and the explained variance contribution or other performance indicators of each group to sedimentation prediction are calculated. The coordination promotion degree for Group 1 is 0.4, for Group 2 is 0.3, and for Group 3 is 0.3.

[0076] Next, based on the obtained coordination promotion degree and combined with historical channel operation data, multiple lightweight prediction models are constructed and trained for each coordination feature group. The number of lightweight prediction models corresponding to each coordination feature group is determined by the coordination promotion degree. The higher the promotion degree, the fewer configurable models are, which helps to reduce model costs.

[0077] Next, the lightweight models within each group are evaluated for performance (e.g., accuracy and recall), and the best-performing model in each group is selected as a baseline sedimentation prediction model. For example, if the decision tree model in group 1 performs best, the support vector machine model in group 2, and the logistic regression model in group 3, then three baseline sedimentation prediction models are obtained. Finally, these three models are combined through ensemble learning (e.g., using a weighted average method) to construct a sedimentation probability prediction model.

[0078] Optionally, when using the weighted average method for ensemble learning, the weights may be normalized and allocated according to the degree of coordination promotion.

[0079] Through the above process, the complementary information of different feature groups can be fully utilized, thereby improving the model's sensitivity and recognition accuracy to channel siltation risks, reducing the risk of overfitting or feature bias of a single model, and providing more reliable decision-making support for channel operation and management.

[0080] Based on the key water flow characteristic indicator set, sensors are deployed and water flow monitoring is performed to obtain real-time water flow monitoring data, which is then processed through feature engineering and input into the siltation probability prediction model to predict the siltation probability of the target channel.

[0081] Specifically, based on the monitoring requirements of the key water flow characteristic indicator set, the appropriate sensor type (such as flow rate sensor, water level meter, etc.) can be selected and its reasonable installation location on the channel can be determined.

[0082] Sensors can include: velocity profile monitoring units, such as ADCPs and electromagnetic current meter arrays, to collect vertical and lateral velocity distributions; water level monitoring units, such as high-precision pressure sensors and radar level gauges, to obtain real-time water level changes; turbulence monitoring units, such as high-frequency pressure pulsation sensors, to analyze turbulent energy and vortex spectrum characteristics; and sediment monitoring units, such as turbidity meters and laser particle size analyzers.

[0083] Specifically, real-time water flow monitoring data refers to various data reflecting water flow conditions collected by sensors during channel operation. Feature engineering involves preprocessing, feature extraction, and transformation of the collected raw monitoring data to generate feature inputs that can be efficiently utilized by prediction models.

[0084] In some embodiments, sensor deployment and water flow monitoring are performed based on the key water flow characteristic indicator set to obtain real-time water flow monitoring data, including:

[0085] Based on the key water flow characteristic indicator set, the monitoring requirements of each characteristic indicator are determined, and the corresponding type of sensors are configured to establish; the flow velocity data of each cross-section direction of the target channel is obtained through the flow velocity sensor, the water level change data of each typical section is obtained through the water level sensor, the pressure pulsation signal is collected through the turbulence sensor, and a turbulence dynamic characteristic sequence is constructed; the suspended matter concentration and bed sand concentration in the water body are obtained through the sediment sensor, and the output is the real-time water flow monitoring data.

[0086] Specifically, monitoring needs refer to determining the required monitoring accuracy, frequency, and location requirements based on the characteristics of a set of key water flow characteristic indicators to ensure that the collected data can accurately reflect water flow behavior.

[0087] Specifically, sensors include flow rate sensors, water level sensors, turbulence sensors, and sediment sensors, which measure flow velocity, water level fluctuations, turbulence pulsation signals, and sediment concentration distribution in the water, respectively. Real-time flow monitoring data refers to the raw data collected and transmitted by these sensors during channel operation, serving as a crucial basis for subsequent siltation probability prediction.

[0088] Specifically, based on the key flow characteristic indicator set (such as flow velocity, water level, turbulence, suspended load concentration, and bed sediment concentration), the spatial distribution and dynamic change requirements of each characteristic are analyzed to determine the monitoring sections and sampling frequency. For example, flow velocity requires multiple points to be deployed upstream, midstream, and downstream, as well as at key sections, with a sampling frequency of 1 Hz. Turbulence characteristics require high-frequency turbulence sensors to be deployed at bends or near structures prone to turbulence, with a sampling frequency of at least 10 Hz. Suspended load and bed sediment concentration require sediment sensors to be deployed at typical sections, with a sampling frequency of 0.1 Hz.

[0089] Specifically, flow velocity data in each cross-sectional direction of the target channel are obtained through flow velocity sensors (such as ultrasonic flow meters) to form a cross-sectional flow velocity distribution matrix; water level change data of each typical section are obtained through water level sensors (such as pressure water level gauges, radar water level gauges) to form a water level time series curve; pressure pulsation signals are collected through turbulence sensors (such as pressure pulsation sensors, particle image velocimeters), and turbulence dynamic feature sequences (such as turbulence intensity, energy spectrum, etc.) are constructed through signal processing (such as wavelet analysis, Fourier transform); suspended matter concentration and bed sand concentration in the water body are obtained through sediment sensors (such as optical turbidity meters or ultrasonic sediment meters), and sediment temporal and spatial distribution data are output.

[0090] Furthermore, the multi-source data collected by the above-mentioned various sensors are integrated and output as the real-time water flow monitoring data, providing high-quality input for subsequent feature engineering and sedimentation probability prediction models.

[0091] Through scientific sensor deployment and multi-dimensional water flow monitoring based on a set of key water flow characteristic indicators, the water flow and sediment transport status in the channel can be obtained comprehensively, dynamically and accurately, providing a rich and accurate data basis for subsequent feature engineering processing and siltation probability prediction, thereby improving the timeliness and reliability of channel siltation risk monitoring and intelligent early warning.

[0092] In some embodiments, the prediction of the siltation probability of the target channel is performed by inputting the feature engineering data into the siltation probability prediction model, including:

[0093] The key water flow characteristic indicator set is decomposed and classified into intuitive water flow characteristic indicators and inferred water flow characteristic indicators; based on the pre-processed real-time water flow monitoring data, the intuitive water flow characteristic parameters corresponding to the intuitive water flow characteristic indicators are extracted; the inferred water flow characteristic parameters corresponding to the inferred water flow characteristic indicators are calculated according to the intuitive water flow characteristic parameters; the intuitive water flow characteristic parameters and the inferred water flow characteristic parameters are fused, and the fusion result is processed by a feature engineering method to obtain real-time input variables; the real-time input variables are transmitted to the input end of the siltation probability prediction model, and the siltation probability prediction result is obtained from the output end.

[0094] Specifically, intuitive water flow characteristic indicators refer to water flow characteristic parameters that can be directly measured by sensors, such as flow velocity, water level, turbidity, pressure pulsation, etc. These indicators can directly reflect the basic state of the water flow. Inferred water flow characteristic indicators are indirect characteristics derived from intuitive water flow characteristic parameters through mathematical models, physical formulas or empirical relationships, such as the kinetic energy of the water flow, shear force, sediment transport rate, Reynolds number, Froude number, etc., which are used to provide more in-depth information on water flow behavior. For example, the flow rate can be calculated based on flow velocity and cross-sectional area; the shear force can be calculated based on flow velocity gradient and fluid viscosity; dimensionless parameters such as the Reynolds number and Froude number can be calculated based on flow velocity and water depth; and the sediment transport rate can be inferred based on flow velocity and sediment concentration.

[0095] During the specific implementation process, the key water flow characteristic indicator set is first decomposed and classified into intuitive water flow characteristic indicators and inferred water flow characteristic indicators. For example, parameters that can be directly measured, such as flow velocity and water level, are classified as intuitive water flow characteristic indicators; turbulence intensity calculated based on flow velocity, and flow rate calculated based on water level and flow velocity are classified as inferred water flow characteristic indicators. Then, based on the pre-processed real-time water flow monitoring data, the intuitive water flow characteristic parameters corresponding to the intuitive water flow characteristic indicators are extracted. For example, the flow velocity values ​​in each cross-sectional direction are extracted from the data collected by the flow velocity sensor, and the water level values ​​of each typical section are extracted from the data collected by the water level sensor.

[0096] Furthermore, the inferred flow characteristic parameters corresponding to the inferred flow characteristic indicators are calculated based on the extracted intuitive flow characteristic parameters. The intuitive and inferred flow characteristic parameters are fused, and the fusion results are processed through feature engineering methods to obtain real-time input variables. For example, parameters such as flow velocity, water level, turbulence intensity, and flow rate are standardized to eliminate dimensionality effects. Feature construction (e.g., feature concatenation) is then performed to form a feature vector or feature matrix that comprehensively characterizes the channel flow state.

[0097] Finally, the processed real-time input variables are transmitted to the input end of the siltation probability prediction model. The model performs calculations based on the input feature vector and obtains the siltation probability prediction results of the target channel from the output end, such as the output siltation probability P (for example, P = 0.82, indicating that the current channel siltation probability is 82%).

[0098] The above process can significantly improve the representativeness and effectiveness of the model input variables through scientific decomposition, parameter extraction and fusion of the key water flow characteristic indicator set, combined with feature engineering methods, and enhance the model's ability to characterize complex hydrodynamics and sediment transport processes, thereby greatly improving the accuracy and timeliness of sedimentation probability prediction.

[0099] Based on the siltation probability prediction results, it is determined whether the target channel has siltation risk, and a siltation probability warning is issued when the risk meets the preset conditions.

[0100] In some embodiments, based on the siltation probability prediction result, determining whether the target channel has a siltation risk and issuing a siltation probability warning when the risk meets a preset condition includes:

[0101] A siltation probability risk grading threshold that is compatible with the channel operation safety level is set, and a siltation risk discrimination criterion is established; the siltation probability prediction result is compared with the siltation risk discrimination criterion; when the siltation probability result exceeds the threshold corresponding to the discrimination criterion, it is determined that there is a siltation risk; an early warning instruction is generated according to the siltation risk level, and a siltation probability early warning information is sent to the channel operation management end through the early warning interface.

[0102] Specifically, the siltation probability risk grading thresholds refer to the probability thresholds used to distinguish different siltation risk levels based on the channel's operational safety level, such as low risk (0%-30%), medium risk (30%-60%), high risk (60%-90%), and extremely high risk (above 90%). The siltation risk discrimination criteria, based on the siltation probability risk grading thresholds, are used to determine the current siltation risk and its level within the target channel.

[0103] Specifically, the early warning interface is a communication interface used to transmit siltation risk warning information to the channel operation and management end. It can be implemented in various forms, such as wired, wireless, or platform API. When a siltation risk is determined, the system automatically generates an early warning instruction and pushes the risk warning information through the early warning interface.

[0104] Specifically, the model's output of the siltation probability prediction is compared with pre-set criteria. For example, if the model predicts a siltation probability of 65% for a particular section, and the high-risk threshold for that channel is 50%, then that section is considered to have a high risk of siltation. Furthermore, an early warning instruction is generated based on the determined risk level, and this siltation probability warning information is sent to the channel operation management terminal via an early warning interface, prompting management personnel to take appropriate measures. For example, the warning information may include fields such as the channel number, time, current siltation probability, risk level, and recommended measures.

[0105] Through this process, risk grading thresholds adapted to channel safety levels are set, and scientific risk identification criteria are established, enabling automated, graded assessment and real-time early warning of channel siltation risks. This effectively enhances the risk perception and response efficiency of channel operations and maintenance, significantly reduces operational safety hazards caused by siltation, and ensures the safe and stable operation of the channel system.

[0106] In summary, the method for early warning of channel siltation probability based on water flow feature identification provided by the present invention has the following technical effects:

[0107] By obtaining the characteristic parameters of the target channel and inputting the characteristic parameters into a pre-built interactive knowledge graph, a set of key water flow characteristic indicators for characterizing the channel water flow behavior is determined; using the key water flow characteristic indicator set as the feature space, a random feature combination with replacement is performed, and the feature combination results are screened based on historical channel operation data to determine K coordinated feature groups; using the K coordinated feature groups as input variable groups, K benchmark siltation prediction models are constructed and trained, and the K benchmark siltation prediction models are integrated to obtain a siltation probability prediction model; based on the key water flow characteristic indicator set, sensors are deployed and water flow monitoring is performed to obtain real-time water flow monitoring data, and after feature engineering processing, the data is input into the siltation probability prediction model to predict the siltation probability of the target channel; based on the siltation probability prediction result, it is determined whether the target channel has a siltation risk, and a siltation probability warning is issued when the risk meets the preset conditions, thereby achieving the technical effects of improving the foresight of the warning, reducing the misjudgment rate, and enhancing adaptability to complex environments.

[0108] Example 2, as Figure 2 This is a schematic diagram of the structure of a channel sedimentation probability early warning system based on water flow feature recognition according to the present invention. Figure 1 The flow chart of the channel sedimentation probability early warning method based on water flow feature recognition in the present invention can be shown as follows: Figure 2 The structure shown is implemented.

[0109] Based on the same concept as the channel siltation probability early warning method based on water flow feature recognition in the above embodiment, the present invention also provides a channel siltation probability early warning system based on water flow feature recognition, comprising:

[0110] The characteristic index determination module 11 is used to obtain characteristic parameters of the target channel and input the characteristic parameters into a pre-built interactive knowledge graph to determine a set of key water flow characteristic indicators for characterizing the water flow behavior in the channel.

[0111] The coordination feature group screening module 12 is used to perform random feature combination with replacement using the key water flow feature index set as the feature space, and screen the feature combination results based on historical channel operation data to determine K coordination feature groups.

[0112] The prediction model construction and integration module 13 is used to construct and train K benchmark sedimentation prediction models using the K coordination feature groups as input variable groups, and integrate the K benchmark sedimentation prediction models to obtain a sedimentation probability prediction model.

[0113] The water flow monitoring and siltation probability prediction module 14 is used to perform sensor deployment and water flow monitoring based on the key water flow characteristic indicator set, obtain real-time water flow monitoring data, and input it into the siltation probability prediction model after feature engineering processing to predict the siltation probability of the target channel.

[0114] The sedimentation risk judgment and warning module 15 is used to judge whether the target channel has sedimentation risk based on the sedimentation probability prediction result, and issue a sedimentation probability warning when the risk meets the preset conditions.

[0115] In some embodiments, the characteristic indicator determination module 11 includes:

[0116] The channel parameter acquisition and knowledge graph input unit is used to obtain the channel structure parameters and hydraulic design parameters in the target area, and input them into the interactive knowledge graph to trigger the knowledge graph reasoning process based on semantic reasoning and entity relationship matching.

[0117] The key water flow characteristic indicator identification unit is used to identify the key water flow behavior characteristic indicators corresponding to the target channel based on the knowledge graph reasoning result, and form the key water flow characteristic indicator set.

[0118] In some embodiments, the coordination feature group screening module 12 includes:

[0119] The principal component analysis and random selection constraint definition unit is used to perform principal component analysis on the key water flow characteristic index set and define the number of characteristic indicators that meet a preset contribution threshold as a random selection constraint.

[0120] A random number generator configuration unit is used to configure the random number generator based on the random selection constraint.

[0121] The feature combination generating and storing unit is used to activate the random number generator to generate a single selection quantity n, and randomly select n key water flow characteristic indicators from the key water flow characteristic indicator set and store them as a feature combination.

[0122] The feature combination iteration and synergy evaluation unit is used to iteratively obtain N feature combinations, and evaluate the combined synergy of the N feature combinations in combination with the historical channel operation data, select the top K feature combinations with high combined synergy, and output them as K coordinated feature groups, where N is greater than n and K is less than N.

[0123] In some embodiments, the prediction model building and integration module 13 includes:

[0124] The coordination promotion degree extraction unit is used to extract K coordination promotion degrees of the K coordination feature groups.

[0125] The lightweight prediction model construction and training unit is used to construct and train K groups of lightweight prediction models based on the K coordination promotion degrees and in combination with the historical channel operation data.

[0126] The benchmark sedimentation prediction model integration and prediction model construction unit is used to select the lightweight prediction model with the best performance in each group, obtain K benchmark sedimentation prediction models for integrated learning, and construct the sedimentation probability prediction model.

[0127] In some embodiments, the water flow monitoring and sedimentation probability prediction module 14 includes:

[0128] The monitoring requirement determination and sensor configuration unit is used to determine the monitoring requirements of each characteristic indicator based on the key water flow characteristic indicator set, configure the corresponding type of sensor, and establish a monitoring system.

[0129] The real-time water flow monitoring data acquisition unit is used to obtain the flow velocity data in each cross-sectional direction of the target channel through the flow velocity sensor, obtain the water level change data of each typical section through the water level sensor, collect the pressure pulsation signal through the turbulence sensor, construct the turbulence dynamic feature sequence, and obtain the suspended matter concentration and bed sand concentration in the water body through the sediment sensor, and output the real-time water flow monitoring data.

[0130] In some embodiments, the water flow monitoring and sedimentation probability prediction module 14 further includes:

[0131] The key water flow characteristic index decomposition and classification unit is used to decompose the key water flow characteristic index set and classify it into intuitive water flow characteristic indexes and inferred water flow characteristic indexes.

[0132] The intuitive water flow characteristic parameter extraction unit is used to extract the intuitive water flow characteristic parameters corresponding to the intuitive water flow characteristic index based on the pre-processed real-time water flow monitoring data.

[0133] The inferred water flow characteristic parameter calculation unit is used to calculate the inferred water flow characteristic parameter corresponding to the inferred water flow characteristic index based on the intuitive water flow characteristic parameter.

[0134] The real-time input variable acquisition unit is used to fuse the intuitive water flow characteristic parameters with the inferred water flow characteristic parameters, and process the fusion result through a feature engineering method to obtain real-time input variables.

[0135] The sedimentation probability prediction result acquisition unit is used to transmit the real-time input variable to the input end of the sedimentation probability prediction model and obtain the sedimentation probability prediction result from the output end.

[0136] In some embodiments, the sedimentation risk judgment and warning module 15 includes:

[0137] The sedimentation probability risk classification threshold setting and judgment criterion establishment unit is used to set the sedimentation probability risk classification threshold that is compatible with the channel operation safety level and establish the sedimentation risk judgment criterion.

[0138] The sedimentation probability prediction result comparison unit is used to compare the sedimentation probability prediction result with the sedimentation risk judgment criterion.

[0139] The sedimentation risk determination unit is configured to determine that there is a sedimentation risk when the sedimentation probability result exceeds a threshold value corresponding to the discrimination criterion.

[0140] The warning instruction generation and warning information sending unit is used to generate warning instructions according to the siltation risk level, and send siltation probability warning information to the channel operation management end through the warning interface.

[0141] It should be understood that the embodiments mentioned in this specification focus on their differences from other embodiments. The specific embodiments in the aforementioned embodiment one are also applicable to the channel siltation probability warning system for water flow feature identification described in embodiment two. For the sake of brevity of the specification, no further elaboration will be given here.

[0142] It should be understood that the embodiments disclosed in the present invention and the above description can enable those skilled in the art to use the present invention to implement the present invention. At the same time, the present invention is not limited to the embodiments mentioned above. It should be understood that those skilled in the art can still modify the technical solutions described in the above embodiments or replace some of the technical features therein with equivalents; and such modifications or replacements do not deviate from the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention and are all included in the scope of protection of the present invention.

Claims

1. A channel sedimentation probability early warning method based on water flow feature identification, characterized in that: include: Obtain characteristic parameters of the target channel and input the characteristic parameters into a pre-built interactive knowledge graph to determine a set of key flow characteristic indicators for characterizing the channel water flow behavior; Taking the key water flow characteristic indicator set as the feature space, performing random feature combination with replacement, and screening the feature combination results based on historical channel operation data to determine K coordinated feature groups; Using the K coordinated feature groups as input variable groups, constructing and training K benchmark sedimentation prediction models, integrating the K benchmark sedimentation prediction models to obtain a sedimentation probability prediction model; Deploy sensors and monitor water flow based on the key water flow characteristic indicator set to obtain real-time water flow monitoring data, and input the data into the siltation probability prediction model after feature engineering processing to predict the siltation probability of the target channel; Based on the siltation probability prediction results, determine whether the target channel has siltation risk and issue a siltation probability warning when the risk meets the preset conditions; Among them, K coordinated feature groups are determined, including: Performing principal component analysis on the key water flow characteristic indicator set, defining the number of characteristic indicators that meet a preset contribution threshold as a random selection constraint; configuring a random number generator based on the random selection constraint; activating the random number generator to generate a single selection quantity n, and randomly selecting n key water flow characteristic indicators from the key water flow characteristic indicator set, and storing them as a characteristic combination; Iteratively obtain N feature combinations, and evaluate the performance improvement ratio of the N feature combinations for siltation probability prediction in combination with the historical channel operation data, obtain the combination synergy, select the top K feature combinations with the highest combination synergy, and output them as K coordinated feature groups, where N is greater than n and K is less than N; Among them, the sedimentation probability prediction model is obtained, including: Calculating the contribution rates of the K coordination feature groups to the prediction of the sedimentation probability, and obtaining K coordination promotion degrees; Based on the K coordination promotion degrees and combined with the historical channel operation data, K groups of lightweight prediction models are constructed and trained accordingly; The lightweight prediction model with the best performance in each group is selected, K benchmark sedimentation prediction models are obtained for ensemble learning, and the sedimentation probability prediction model is constructed.

2. The channel sedimentation probability early warning method based on water flow feature recognition according to claim 1, characterized in that: Obtain the characteristic parameters of the target channel and input them into the pre-built interactive knowledge graph to determine the key flow characteristic indicator set used to characterize the channel water flow behavior, including: Obtain channel structure parameters and hydraulic design parameters within the target area and input them into the interactive knowledge graph, triggering a knowledge graph reasoning process based on semantic reasoning and entity relationship matching; According to the knowledge graph reasoning result, the key water flow behavior characteristic indicators corresponding to the target channel are identified to form the key water flow characteristic indicator set.

3. A channel sedimentation probability early warning method based on water flow feature recognition as claimed in claim 2, characterized in that: Based on the key water flow characteristic indicator set, sensors are deployed and water flow monitoring is performed to obtain real-time water flow monitoring data, including: Based on the key water flow characteristic indicator set, determine the monitoring requirements of each characteristic indicator, configure the corresponding type of sensor, and establish; The flow velocity data of each cross-sectional direction of the target channel is obtained through the flow velocity sensor, the water level change data of each typical section is obtained through the water level sensor, the pressure pulsation signal is collected through the turbulence sensor to construct a turbulence dynamic feature sequence, and the suspended matter concentration and bed sand concentration in the water body are obtained through the sediment sensor, and the output is the real-time water flow monitoring data.

4. A channel sedimentation probability early warning method based on water flow feature recognition as claimed in claim 3, characterized in that: After feature engineering processing, the data is input into the siltation probability prediction model to predict the siltation probability of the target channel, including: Decomposing the key water flow characteristic indicator set and classifying them into intuitive water flow characteristic indicators and inferred water flow characteristic indicators; Extracting intuitive water flow characteristic parameters corresponding to the intuitive water flow characteristic indicators based on the pre-processed real-time water flow monitoring data; Calculating the estimated water flow characteristic parameter corresponding to the estimated water flow characteristic index according to the intuitive water flow characteristic parameter; fusing the intuitive water flow characteristic parameters with the inferred water flow characteristic parameters, and processing the fusion results through a feature engineering method to obtain real-time input variables; The real-time input variable is transmitted to the input end of the sedimentation probability prediction model, and the sedimentation probability prediction result is obtained from the output end.

5. The method according to claim 1, wherein Based on the siltation probability prediction results, determine whether the target channel has siltation risk and issue a siltation probability warning when the risk meets the preset conditions, including: Set a siltation probability risk classification threshold that is compatible with the channel operation safety level, and establish siltation risk identification criteria; comparing the sedimentation probability prediction result with the sedimentation risk judgment criterion; When the predicted result of the siltation probability exceeds the threshold corresponding to the discrimination criterion, it is determined that there is a siltation risk; An early warning instruction is generated according to the siltation risk level, and siltation probability early warning information is sent to the channel operation management end through the early warning interface.

6. A channel sedimentation probability early warning system based on water flow feature recognition, characterized in that: A channel sedimentation probability early warning method for realizing the water flow characteristic identification according to any one of claims 1 to 5, comprising: A characteristic index determination module is used to obtain characteristic parameters of the target channel and input the characteristic parameters into a pre-built interactive knowledge graph to determine a set of key water flow characteristic indicators used to characterize the water flow behavior in the channel; A coordinated feature group screening module is used to perform random feature combination with replacement using the key water flow characteristic indicator set as the feature space, and screen the feature combination results based on historical channel operation data to determine K coordinated feature groups; A prediction model construction and integration module is used to construct and train K benchmark sedimentation prediction models using the K coordinated feature groups as input variable groups, and integrate the K benchmark sedimentation prediction models to obtain a sedimentation probability prediction model; A water flow monitoring and siltation probability prediction module is used to perform sensor deployment and water flow monitoring based on the key water flow characteristic indicator set, obtain real-time water flow monitoring data, and input the data into the siltation probability prediction model after feature engineering processing to predict the siltation probability of the target channel; The sedimentation risk judgment and early warning module is used to determine whether there is a sedimentation risk in the target channel based on the sedimentation probability prediction results, and to issue a sedimentation probability early warning when the risk meets the preset conditions.

Images