A Multi-Sensor-Based Early Warning and Dredging Scheduling Method for Urban Road Sedimentation Wells

Abstract

This invention provides a multi-sensor-based method for early warning and dredging scheduling of silt in urban road sedimentation wells, relating to the field of urban drainage system management technology. The method includes collecting silt layer thickness, fluid flow characteristics, and well structure response signals from sensors within the sedimentation well; extracting features through time-frequency domain decomposition to identify the siltation state; constructing a risk propagation graph model for dynamic diffusion calculation to generate regional early warning information; and matching dredging process parameters based on siltation characteristics to generate a dredging scheduling plan. This invention enables real-time monitoring and accurate early warning of silt status in sedimentation wells, improving drainage system safety and dredging operation efficiency.

Description

Technical Field

[0001] This invention relates to the field of urban drainage system management technology, and in particular to a method for early warning and dredging scheduling of silt in urban road sedimentation wells based on multiple sensors. Background Technology

[0002] Urban road sedimentation wells are critical infrastructure in urban drainage systems, primarily used to collect rainwater, settle pollutants, and prevent solid waste from entering the drainage network. With accelerating urbanization, the problem of silt accumulation in road sedimentation wells is becoming increasingly serious. Excessive siltation not only reduces the collection and sedimentation efficiency of the wells but can also lead to serious consequences such as urban flooding and pollutant overflows. Traditional sedimentation well management mainly relies on regular manual inspections and experience-based judgment, lacking a scientific and effective early warning and scheduling mechanism.

[0003] In recent years, with the development of the Internet of Things and sensing technology, some cities have begun to deploy single-type sensors in sedimentation wells to monitor siltation conditions. Some areas have adopted fixed-cycle dredging plans or made dredging decisions based on simple siltation depth data. However, these methods still have many shortcomings in practical applications. Existing technologies mostly use single-parameter monitoring methods, focusing primarily on the intuitive indicator of siltation depth, neglecting important information such as fluid dynamics characteristics and structural response. This leads to a one-sided assessment of siltation conditions, failing to comprehensively reflect the actual working condition and potential risks of the sedimentation wells. Summary of the Invention

[0004] This invention provides a method for early warning and dredging scheduling of silt in urban road sedimentation wells based on multiple sensors, which can solve the problems in the prior art.

[0005] A first aspect of the present invention provides a method for early warning and dredging scheduling of silt in urban road sedimentation wells based on multiple sensors, comprising:

[0006] Physical quantity signals are collected by multiple types of sensors inside the sedimentation well, including silt layer thickness, fluid flow characteristics, and well structure response.

[0007] The physical quantity signal is decomposed in the time-frequency domain and a multi-scale feature spectrum is extracted. The multi-scale feature spectrum is then nonlinearly mapped and siltation state features are identified to obtain a siltation feature vector.

[0008] A risk propagation graph is constructed based on the topological connectivity of the urban drainage pipe network, where each node corresponds to a sedimentation well, and the corresponding risk propagation coefficient is calculated based on the sedimentation feature vector.

[0009] Dynamic diffusion calculations are performed on the risk propagation graph model to determine the set of downstream associated wells affected by the current sedimentation well and the risk accumulation value. When the risk accumulation value exceeds the trigger threshold, regional early warning information including the risk propagation path and the predicted failure time series is generated.

[0010] Based on the siltation feature vectors of each sedimentation well in the regional early warning information, the corresponding dredging process parameters are matched, and combined with the risk propagation path and the spatial distribution of dredging resources, a dredging operation scheduling plan is generated and sent to the dredging execution unit.

[0011] The physical quantity signal is decomposed in the time-frequency domain and a multi-scale feature spectrum is extracted. The multi-scale feature spectrum is then nonlinearly mapped and siltation state features are identified to obtain a siltation feature vector, including:

[0012] Multi-resolution time-frequency decomposition is performed on the physical quantity signal, dividing the physical quantity signal into several scale intervals on the time axis and frequency axis. Subspace projection decomposition is performed on the frequency domain component of each scale interval, decomposing the frequency domain component into a dominant subspace component and a residual subspace component. The energy proportion of the dominant subspace component and the fluctuation amplitude of the residual subspace component are calculated respectively. The energy proportion and fluctuation amplitude of all scale intervals are combined to form the multi-scale feature spectrum.

[0013] Construct a siltation state space model, map the energy proportion of the dominant subspace component and the fluctuation amplitude of the residual subspace component in the multi-scale feature spectrum to the siltation state space model, and determine the current state node position of the physical quantity signal in the siltation state space model;

[0014] Based on the multi-scale feature spectrum sequence within the historical time window, the migration trajectory of the state node position is tracked in the siltation state space model, the direction vector and migration rate of the migration trajectory are extracted, and the state node position, the direction vector and the migration rate are encapsulated into the siltation feature vector.

[0015] Constructing a siltation state space model, mapping the energy proportion of the dominant subspace component and the fluctuation amplitude of the residual subspace component in the multi-scale feature spectrum to the siltation state space model, and determining the current state node position of the physical quantity signal in the siltation state space model, including:

[0016] A two-dimensional feature space is established with the energy proportion of the dominant subspace component and the fluctuation amplitude of the residual subspace component as coordinate axes. According to the feature distribution law of different stages in the accumulation evolution process, multiple topological division boundaries are set in the two-dimensional feature space. The topological division boundaries divide the two-dimensional feature space into several non-overlapping state subspaces.

[0017] For each state subspace, a corresponding discrete state node is set. Based on the unidirectional and reversible characteristics of siltation evolution, a directed migration path is established between the state nodes corresponding to adjacent state subspaces that share the topological partition boundary, thus forming the siltation state space model.

[0018] Extract the energy proportion of the dominant subspace component and the fluctuation amplitude of the residual subspace component at the current moment, and mark them as the current feature point in the two-dimensional feature space; calculate the shortest vertical distance from the current feature point to the topological segmentation boundary of each state subspace, determine the state subspace to which the current feature point belongs, and determine the state node corresponding to the state subspace as the state node position.

[0019] A risk propagation graph is constructed based on the topological connectivity of the urban drainage network, where each node corresponds to a sedimentation well, and the corresponding risk propagation coefficient is calculated based on the sedimentation feature vector, including:

[0020] Based on the physical connectivity of the urban drainage network, the upstream and downstream connection relationships between each sedimentation well are extracted. Each sedimentation well is abstracted as a node in the risk propagation graph model. Directed connection edges are established between the nodes corresponding to sedimentation wells with upstream and downstream connection relationships to construct the risk propagation graph.

[0021] Based on the siltation feature vector, the trajectory of the change of state node positions within the historical time window is extracted, a nonlinear trend curve is fitted, and the instantaneous evolution rate at the current moment is calculated; according to the siltation feature vector, the predicted sequence of state node positions within the future time step is deduced in the siltation state space model, and the risk increment corresponding to each predicted position is calculated.

[0022] For each node, the instantaneous evolution rate and the risk increment are summed in a time-weighted manner to calculate the local risk intensity of each node. For each directed connection edge, the time delay of risk signal propagation is calculated based on the hydraulic parameters of the pipe segment between the upstream and downstream nodes.

[0023] Based on the time delay, the risk increment value at the corresponding moment is selected from the risk increment sequence of the upstream node, and coupled with the local risk intensity of the upstream node to obtain the risk propagation coefficient corresponding to each directed connection edge, which is then superimposed on the risk propagation graph.

[0024] Dynamic diffusion calculations are performed on the risk propagation graph model to determine the set of downstream associated wells affected by the current sedimentation well and the cumulative risk value. When the cumulative risk value exceeds a trigger threshold, regional early warning information including the risk propagation path and predicted failure time series is generated, including:

[0025] Using the node corresponding to the current sedimentation well as the initial diffusion source node, the risk propagation coefficients of each directed connection edge on the path between each node and the initial diffusion source node are extracted and multiplied together to obtain the path attenuation factor, which is then multiplied with the local risk intensity of each node to obtain the risk potential energy value.

[0026] For each incoming edge of each node, the difference between the risk potential energy value of the upstream node and each node is calculated as the potential energy gradient. The potential energy gradients of each incoming edge are vector-synthesized to obtain the synthetic potential energy gradient vector. Based on the vector magnitude of the synthetic potential energy gradient vector, the risk accumulation value of each node is calculated. Nodes whose risk accumulation value exceeds the trigger threshold are selected to form a downstream associated well set.

[0027] For each node in the downstream associated well set, backtrack hop by hop along the opposite direction of the synthetic potential energy gradient vector, selecting the upstream node with the largest potential energy gradient magnitude each time, until the initial diffusion source node is reached, and the backtracking path is recorded to form the fastest risk propagation path;

[0028] The total path delay is obtained by summing the time delays of each directed connection edge in the fastest risk propagation path. The predicted arrival time is obtained by adding the current time to the total path delay. The predicted failure time is calculated by combining the risk accumulation value. The predicted failure times of each node are arranged to form a predicted failure time sequence, and integrated to generate regional early warning information.

[0029] Based on the siltation feature vectors of each sedimentation well in the regional early warning information, the corresponding dredging process parameters are matched. Combined with the risk propagation path and the spatial distribution of dredging resources, a dredging operation scheduling plan is generated and distributed to the dredging execution unit, including:

[0030] Calculate the siltation feature vector of each sedimentation well in the regional early warning information and the similarity with the historical siltation feature vector in the historical process implementation record. Select the historical process implementation record with the highest similarity and extract the corresponding process parameters.

[0031] From the risk propagation path, identify the critical propagation nodes, which are nodes with an out-degree greater than a set value and a risk accumulation value within the target range. Elevate the operation priority of the critical propagation nodes to the beginning of the sequence. For non-critical propagation nodes, establish a time-constraint network, using the predicted failure time as the upper time constraint of the node and the hydraulic propagation time of the pipe segment between adjacent nodes as the lower time constraint of the edge. Solve the topology sorting scheme that satisfies all time constraints in the time-constraint network, and merge the topology sorting scheme with the critical propagation node sequence to form an operation sequence.

[0032] Obtain the spatial location and operational capacity parameters of dredging resources, calculate the total cost of each dredging resource using a rolling time-domain optimization method according to the operational sequence, select the resource allocation strategy with the lowest cost, generate a dredging operation scheduling plan, and send it to the dredging execution unit.

[0033] Obtain the spatial location and operational capacity parameters of dredging resources. Following the operational sequence, calculate the total cost of each dredging resource using a rolling time-domain optimization method. Select the resource allocation strategy with the lowest cost, generate a dredging operation scheduling plan, and distribute it to the dredging execution unit. This includes:

[0034] Obtain the spatial location coordinates and operational capability parameters of each dredging resource, construct a resource task spatiotemporal correlation matrix, where the rows of the spatiotemporal correlation matrix correspond to each dredging resource and the columns correspond to each sedimentation well node in the operational sequence, calculate the travel time of each dredging resource to each sedimentation well node according to the operational sequence, calculate the operation duration based on the operational capability parameters and process parameters, accumulate the travel time and operation duration and fill them into the corresponding elements of the spatiotemporal correlation matrix, and sum the rows of the spatiotemporal correlation matrix to obtain the total operation duration;

[0035] Set the time window length and rolling step size of the rolling time domain optimization method, extract the subset of nodes to be allocated within the time window from the job sequence, construct a state transition graph for the subset of nodes to be allocated, where nodes represent resource allocation status and edges represent allocation decisions, and use dynamic programming to calculate the sum of the immediate allocation cost and the expected cost in the future period as the total cost weight of the state transition edge.

[0036] The minimum total cost path is searched in the state transition graph. The allocation scheme within the current rolling step of the path is extracted as the execution strategy. The decision time is advanced according to the rolling step and the above calculation is repeated until all nodes complete resource allocation. The dredging operation scheduling scheme is generated and sent to the dredging execution unit.

[0037] A second aspect of the present invention provides a multi-sensor-based urban road sedimentation well sludge early warning and dredging scheduling system, comprising:

[0038] The signal acquisition unit is used to acquire physical quantity signals through multiple types of sensors in the sedimentation well. The physical quantity signals include the thickness of the silt layer, fluid flow characteristics, and well body structural response.

[0039] The feature extraction unit is used to perform time-frequency domain decomposition on the physical quantity signal and extract multi-scale feature spectrum, perform nonlinear mapping on the multi-scale feature spectrum and identify siltation state features to obtain siltation feature vector;

[0040] The graph model unit is used to construct a risk propagation graph based on the topological connectivity of the urban drainage pipe network, wherein each node corresponds to a sedimentation well, and the corresponding risk propagation coefficient is calculated based on the sedimentation feature vector.

[0041] The risk warning unit is used to perform dynamic diffusion calculations on the risk propagation graph model, determine the set of downstream associated wells affected by the current sedimentation well and the risk accumulation value, and generate regional warning information including risk propagation path and predicted failure time series when the risk accumulation value exceeds the trigger threshold.

[0042] The scheduling generation unit is used to match the corresponding dredging process parameters based on the siltation feature vectors of each sedimentation well in the regional early warning information, and generate a dredging operation scheduling plan by combining the risk propagation path and the spatial distribution of dredging resources, and then send it to the dredging execution unit.

[0043] A third aspect of the present invention provides an electronic device, comprising:

[0044] processor;

[0045] Memory used to store processor-executable instructions;

[0046] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0047] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0048] The beneficial effects of this application are as follows:

[0049] By integrating multiple types of sensors to simultaneously collect physical quantity signals such as silt layer thickness, fluid flow characteristics, and well structure response, comprehensive monitoring of sedimentation well status is achieved, which significantly improves the sensing accuracy and reliability of siltation status compared to single parameter monitoring.

[0050] By employing time-frequency domain decomposition technology to extract multi-scale feature spectra and performing nonlinear mapping, the characteristics of siltation in complex environments can be effectively identified, overcoming the technical shortcomings of traditional methods in poor identification performance under complex working conditions. A risk propagation graph model is constructed based on the topological relationship of the drainage pipe network, which not only considers single-point siltation risk but also assesses and predicts the network cascading effect of siltation risk, thus expanding the risk assessment dimension from "point" to "network".

[0051] By using dynamic diffusion calculations on the risk propagation map, the affected downstream associated well sets can be accurately identified, and potential system failure time series can be predicted in advance, providing precise regional early warnings for urban drainage systems. Based on the siltation feature vector, the optimal dredging process parameters are intelligently matched, and combined with the risk propagation path and resource distribution status, an optimized dredging operation scheduling scheme is generated, significantly improving dredging efficiency and reducing scheduling complexity. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating the method for early warning and dredging scheduling of urban road sedimentation wells based on multiple sensors, according to an embodiment of the present invention.

[0053] Figure 2 This is a flowchart illustrating the method for constructing a risk propagation graph according to an embodiment of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0055] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0056] Figure 1 This is a flowchart illustrating the multi-sensor-based urban road sedimentation well sludge early warning and dredging scheduling method according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0057] Physical quantity signals are collected by multiple types of sensors inside the sedimentation well, including silt layer thickness, fluid flow characteristics, and well structure response.

[0058] The physical quantity signal is decomposed in the time-frequency domain and a multi-scale feature spectrum is extracted. The multi-scale feature spectrum is then nonlinearly mapped and siltation state features are identified to obtain a siltation feature vector.

[0059] A risk propagation graph is constructed based on the topological connectivity of the urban drainage pipe network, where each node corresponds to a sedimentation well, and the corresponding risk propagation coefficient is calculated based on the sedimentation feature vector.

[0060] Dynamic diffusion calculations are performed on the risk propagation graph model to determine the set of downstream associated wells affected by the current sedimentation well and the risk accumulation value. When the risk accumulation value exceeds the trigger threshold, regional early warning information including the risk propagation path and the predicted failure time series is generated.

[0061] Based on the siltation feature vectors of each sedimentation well in the regional early warning information, the corresponding dredging process parameters are matched, and combined with the risk propagation path and the spatial distribution of dredging resources, a dredging operation scheduling plan is generated and sent to the dredging execution unit.

[0062] The physical quantity signal is decomposed in the time-frequency domain and a multi-scale feature spectrum is extracted. The multi-scale feature spectrum is then nonlinearly mapped and siltation state features are identified to obtain a siltation feature vector, including:

[0063] Multi-resolution time-frequency decomposition is performed on the physical quantity signal, dividing the physical quantity signal into several scale intervals on the time axis and frequency axis. Subspace projection decomposition is performed on the frequency domain component of each scale interval, decomposing the frequency domain component into a dominant subspace component and a residual subspace component. The energy proportion of the dominant subspace component and the fluctuation amplitude of the residual subspace component are calculated respectively. The energy proportion and fluctuation amplitude of all scale intervals are combined to form the multi-scale feature spectrum.

[0064] Construct a siltation state space model, map the energy proportion of the dominant subspace component and the fluctuation amplitude of the residual subspace component in the multi-scale feature spectrum to the siltation state space model, and determine the current state node position of the physical quantity signal in the siltation state space model;

[0065] Based on the multi-scale feature spectrum sequence within the historical time window, the migration trajectory of the state node position is tracked in the siltation state space model, the direction vector and migration rate of the migration trajectory are extracted, and the state node position, the direction vector and the migration rate are encapsulated into the siltation feature vector.

[0066] Physical quantity signals collected from various types of sensors within sedimentation wells typically manifest as multi-dimensional time-series signals. These signals carry rich information about the sedimentation state in both the time and frequency dimensions. To fully extract this information, multi-resolution time-frequency decomposition technology is employed to process the raw signals. Specifically, the collected physical quantity signals are divided along the time axis according to preset time windows. The length of each time window is determined based on the signal sampling frequency and the dynamic characteristics of the sedimentation process, typically set to cover the period of significant changes in the sedimentation state. Simultaneously, along the frequency axis, the frequency range is divided into several scale intervals based on the signal's spectral characteristics. Each scale interval corresponds to different physical process characteristics; for example, the low-frequency band reflects the slow accumulation process of the silt layer, the mid-frequency band reflects the fluid disturbance characteristics, and the high-frequency band reflects the transient response of the well structure.

[0067] For each scale interval, the frequency domain components within that interval are extracted and subspace projection decomposition is performed. This decomposition process is based on the eigenvalue decomposition of the signal covariance matrix. The frequency domain components are projected onto the dominant subspace spanned by the eigenvectors corresponding to the first few largest eigenvalues ​​of the covariance matrix, and the residual subspace spanned by the remaining eigenvectors. The dominant subspace captures the most significant periodic or quasi-periodic components of the signal. These components are usually related to the main driving factors of siltation, such as tidal drainage patterns or flow changes caused by periodic rainfall. The residual subspace contains the effects of irregular disturbances, random fluctuations, and sudden events. These components are of great value in identifying abnormal siltation events.

[0068] The energy proportion of the dominant subspace component is calculated, defined as the ratio of the energy of the dominant subspace component to the total energy within that scale interval. Energy can be measured by the sum of squares of the signal amplitude or the sum of the corresponding eigenvalues ​​of the covariance matrix. A high energy proportion indicates that the signal within that scale interval has obvious regularity, reflecting a relatively stable accumulation phase in the accumulation process. The fluctuation amplitude of the residual subspace component is calculated, measured by the standard deviation or range of the residual component within the time window. An increase in fluctuation amplitude often indicates that the accumulation process has entered an unstable phase or is affected by external disturbances. The energy proportions and fluctuation amplitudes of all scale intervals are arranged in scale order to form a two-dimensional array structured multi-scale characteristic spectrum. This characteristic spectrum comprehensively characterizes the energy distribution and fluctuation characteristics of the signal at different time and frequency scales.

[0069] To transform multi-scale feature spectra into siltation state information usable for early warning and judgment, a siltation state space model is constructed. This model is a multi-dimensional feature space, the dimensions of which are determined by the types of physical quantities involved in the modeling and the number of scale intervals. In this space, each coordinate axis corresponds to the feature value of a specific physical quantity in a specific scale interval. For example, the energy proportion of the silt layer thickness sensor signal in the low-frequency band constitutes one coordinate axis, and the fluctuation amplitude of the fluid flow velocity sensor signal in the mid-frequency band constitutes another coordinate axis. Through historical data calibration, regions corresponding to different degrees of siltation are divided in this state space, including normal areas, lightly silted areas, moderately silted areas, and heavily silted areas. The boundaries of these regions are determined by historical dredging records and expert experience, and are quantitatively described using methods such as clustering algorithms or support vector machines.

[0070] When mapping the multi-scale feature spectrum of the current moment to the siltation state space model, feature normalization is required to ensure that the feature values ​​of different physical quantities and different scale intervals are within a comparable numerical range. Normalization methods can include maximum-minimum standardization or standardization based on historical statistical distribution. After normalization, the energy proportion of the dominant subspace component and the fluctuation amplitude of the residual subspace component are used as coordinate values ​​to determine the position of the current state node in the state space. This position not only reflects the current degree of siltation but also implies information about the stability of the siltation process. For example, a state node located in a region with low energy proportion and high fluctuation amplitude indicates that although siltation is not yet severe, the siltation rate is accelerating.

[0071] Based on obtaining the position of state nodes at a single moment, the concept of a historical time window is introduced. Multi-scale feature spectrum sequences from several past sampling periods are collected and sequentially mapped onto the state space, forming a series of state nodes. By connecting these nodes, the migration trajectory of the state nodes in space can be tracked. The geometric characteristics of this trajectory directly reflect the dynamic evolution pattern of the siltation process. When the trajectory shows a stable and slow movement, it indicates that the siltation is in a state of uniform accumulation; when the trajectory shows acceleration or abrupt changes in direction, it indicates that the siltation process has been significantly disturbed or has entered a rapid deterioration stage.

[0072] Quantitative analysis of the migration trajectory is performed to calculate a direction vector. This vector points from the earliest state node within the historical time window to the latest state node, and its direction indicates the evolution trend of the siltation state. If the direction vector points to a heavily silted-up area, the warning level should be raised even if the current state node is in a normal or lightly silted-up area. The direction vector can be calculated using a linear regression method, fitting the historical state node sequence; the slope vector of the regression line is the direction vector. Simultaneously, the migration rate is calculated, defined as the distance a state node moves in space per unit time. The distance can be calculated using the Euclidean norm or weighted norm, with the weighting coefficients determined based on the importance of different feature dimensions in the siltation assessment. The migration rate reflects the speed of change in the siltation state; a high migration rate means a shorter warning response time is needed.

[0073] The current state node position, direction vector, and migration rate are encapsulated into a siltation feature vector. This vector is presented as a fixed-dimensional numerical array for easy use in subsequent risk assessment and scheduling decisions. To ensure the vector's robustness, outliers are filtered. When a feature value in a sampling period experiences a jump exceeding a reasonable range, median filtering or Kalman filtering is used for smoothing. Furthermore, for signals acquired by multiple sensors, the siltation feature vectors for each sensor are calculated separately, and then a comprehensive siltation feature vector is formed through weighted fusion or feature concatenation. The fusion weights are adaptively adjusted based on the warning accuracy of different sensors in historical data.

[0074] Constructing a siltation state space model, mapping the energy proportion of the dominant subspace component and the fluctuation amplitude of the residual subspace component in the multi-scale feature spectrum to the siltation state space model, and determining the current state node position of the physical quantity signal in the siltation state space model, including:

[0075] A two-dimensional feature space is established with the energy proportion of the dominant subspace component and the fluctuation amplitude of the residual subspace component as coordinate axes. According to the feature distribution law of different stages in the accumulation evolution process, multiple topological division boundaries are set in the two-dimensional feature space. The topological division boundaries divide the two-dimensional feature space into several non-overlapping state subspaces.

[0076] For each state subspace, a corresponding discrete state node is set. Based on the unidirectional and reversible characteristics of siltation evolution, a directed migration path is established between the state nodes corresponding to adjacent state subspaces that share the topological partition boundary, thus forming the siltation state space model.

[0077] Extract the energy proportion of the dominant subspace component and the fluctuation amplitude of the residual subspace component at the current moment, and mark them as the current feature points in the two-dimensional feature space;

[0078] Calculate the shortest vertical distance from the current feature point to the topological segmentation boundary of each state subspace, determine the state subspace to which the current feature point belongs, and determine the state node corresponding to the state subspace as the state node position.

[0079] After obtaining the multi-scale feature spectrum, it is necessary to transform the multi-dimensional feature vectors obtained by frequency domain decomposition into a state expression form that can intuitively reflect the stage of sedimentation evolution. Through statistical analysis of historical sedimentation data, it was found that the energy proportion of the dominant subspace component can characterize the main sedimentary characteristics of the silt layer, while the fluctuation amplitude of the residual subspace component reflects the degree of influence of fluid disturbance and anomalous sedimentation events. Using these two key parameters as the basis for modeling, the characteristic differences of different sedimentation stages can be effectively distinguished.

[0080] When establishing a two-dimensional feature space, the horizontal axis is defined as the energy proportion of the dominant subspace component, with a value ranging from 0 to 1. The larger the value, the more stable the silt layer morphology and the more concentrated the sedimentation structure. The vertical axis is defined as the fluctuation amplitude of the residual subspace component. This parameter is obtained by calculating the standard deviation of the residual component within a set time window, reflecting the disturbance intensity of non-dominant factors on the siltation process. According to the actual operation law of urban road sedimentation wells, the siltation evolution process can be divided into four typical stages: initial deposition, steady-state accumulation, critical siltation, and severe blockage. Each stage exhibits significant clustering distribution characteristics in the two-dimensional feature space.

[0081] The topological boundary settings are determined based on statistical analysis of a large amount of measured data. The boundary between the initial deposition stage and the steady-state accumulation stage is typically characterized by a dominant energy percentage of 0.4 to 0.5, at which point the silt layer begins to form a relatively stable layered structure, and the residual fluctuation amplitude remains at a low level. The boundary between steady-state accumulation and critical blockage is mainly determined by the abrupt change in the residual fluctuation amplitude. When the fluctuation amplitude exceeds 1.5 to 2 times the historical baseline value, it indicates a significant change in the flow regime within the well, with enhanced interaction between the silt layer and the fluid. The boundary between critical blockage and severe blockage depends simultaneously on a decrease in the dominant energy percentage and a continuous increase in residual fluctuation. In this stage, the dominant energy percentage typically decreases to below 0.3, and the residual fluctuation amplitude reaches its peak level. These boundary settings are not strictly straight lines or curves, but rather banded regions set according to the actual data distribution characteristics, allowing for a certain transition range to accommodate differences in well conditions.

[0082] The division of the state subspace results in four main regions. The first state subspace corresponds to the initial deposition stage, located in the lower left region of the two-dimensional space, characterized by a dominant energy percentage below 0.4 and residual fluctuations at the baseline level. The second state subspace represents the steady-state accumulation stage, located in the central region, with a dominant energy percentage between 0.4 and 0.6 and stable residual fluctuations. The third state subspace marks the critical clogging stage, where the dominant energy percentage begins to decrease but remains above 0.3, and residual fluctuations significantly increase. The fourth state subspace indicates the severe clogging stage, where the dominant energy percentage is below 0.3 and residual fluctuations reach a warning level. The center point of each state subspace is defined as the corresponding discrete state node, and the coordinates of this node are obtained by calculating the centroid of historical samples within that subspace.

[0083] The establishment of directed migration paths reflects the physical laws of the sedimentation process. Normal sedimentation evolution follows a unidirectional trend, progressing from initial deposition to steady-state accumulation, critical siltation, and severe blockage. Correspondingly, unidirectional connections are established between state nodes from lower to higher levels. However, during dredging operations or heavy rainfall, the sedimentation state undergoes reverse migration. In this case, a retreat path from higher-level state nodes to lower-level state nodes needs to be established. This bidirectional path design allows the sedimentation state space model to fully describe the evolution and recovery process of sedimentation. The migration path weights between adjacent state subspaces are assigned based on the frequency and rate of state transitions in historical data. Frequent transitions correspond to higher path weights, providing a probabilistic basis for subsequent predictions.

[0084] Upon receiving new sensor data, time-frequency domain decomposition and subspace analysis are used to extract the energy proportion of the dominant subspace component and the fluctuation amplitude of the residual subspace component at the current moment. The energy proportion of the dominant component is calculated by summing the energy contribution rates of each principal component, typically selecting the top few principal components with a cumulative contribution rate of over 85% as the dominant subspace. The fluctuation amplitude of the residual component is a statistical measure of the variation of the remaining higher-order components along the time axis. These two values ​​are used as coordinates to pinpoint the location of the current feature point in the established two-dimensional feature space.

[0085] To determine the state subspace to which a feature point belongs, a geometric distance discrimination method is used to calculate the perpendicular distance from the current feature point to each topological partition boundary. Since the partition boundaries are straight lines, polylines, or curves, corresponding distance calculation algorithms are needed for different boundary shapes. For straight boundaries, the distance is directly calculated using the point-to-line distance formula; for curved boundaries, the boundary is discretized into several line segments, and the minimum value is taken after calculating the shortest distance from the point to each line segment. The set of distances from the current feature point to all partition boundaries is statistically analyzed. If the point is located inside a certain state subspace, then the distances to all boundaries of that subspace are positive; if the point happens to fall near a boundary, then the nearest boundary is selected as a reference, and the point is assigned to the subspace side that best matches the dominant feature.

[0086] After determining the state subspace to which the current feature point belongs, the corresponding discrete state node is extracted. The identifier and attribute parameters of this node are output as the state node location information. The state node location not only indicates the current accumulation stage but also carries the characteristic statistical information of this stage, the typical range of physical parameters, and the probability distribution of migration to adjacent states. This information provides structured input for subsequent accumulation trend prediction and risk assessment, enabling the entire early warning system to reason and make decisions within a unified state space framework. Through this state space modeling method, complex multidimensional sensor signals are transformed into clear discrete state representations, significantly improving the interpretability of the accumulation evolution process and the timeliness of the early warning response.

[0087] Figure 2 This is a flowchart illustrating the method for constructing a risk propagation graph according to an embodiment of the present invention. The risk propagation graph is constructed based on the topological connectivity of the urban drainage network, where each node corresponds to a sedimentation well, and the corresponding risk propagation coefficient is calculated based on the sedimentation feature vector, including:

[0088] Based on the physical connectivity of the urban drainage network, the upstream and downstream connection relationships between each sedimentation well are extracted. Each sedimentation well is abstracted as a node in the risk propagation graph model. Directed connection edges are established between the nodes corresponding to sedimentation wells with upstream and downstream connection relationships to construct the risk propagation graph.

[0089] Based on the siltation feature vector, the trajectory of the change of state node position within the historical time window is extracted, a nonlinear trend curve is fitted, and the instantaneous evolution rate at the current moment is calculated.

[0090] Based on the siltation feature vector, the predicted sequence of state node positions within a future time step is deduced in the siltation state space model, and the risk increment corresponding to each predicted position is calculated.

[0091] For each node, the instantaneous evolution rate and the risk increment are summed in a time-weighted manner to calculate the local risk intensity of each node. For each directed connection edge, the time delay of risk signal propagation is calculated based on the hydraulic parameters of the pipe segment between the upstream and downstream nodes.

[0092] Based on the time delay, the risk increment value at the corresponding moment is selected from the risk increment sequence of the upstream node, and coupled with the local risk intensity of the upstream node to obtain the risk propagation coefficient corresponding to each directed connection edge, which is then superimposed on the risk propagation graph.

[0093] In actual urban drainage systems, sedimentation wells are interconnected through underground pipe networks, forming a complex topology. To accurately depict the risk propagation process within the network, the physical pipe network needs to be mapped into a mathematical graph model. Specifically, by querying the urban drainage pipe network geographic information system database, the spatial coordinates, depth, diameter, and other physical parameters of each sedimentation well, as well as the information of the pipe segments connecting each well, including hydraulic characteristics such as pipe diameter, length, slope, and material, are obtained. Based on the natural flow direction of the fluid in the pipe network, the upstream and downstream relationships between sedimentation wells are determined. For example, if the outlet pipe of sedimentation well A is connected to the inlet of sedimentation well B, then A is the upstream well and B is the downstream well. Each sedimentation well is represented as a node in the risk propagation graph, with the node number corresponding one-to-one with the physical number of the sedimentation well. For sedimentation well pairs with direct connections, directed edges are established between the corresponding nodes, with the edge pointing downstream. This directed graph structure can accurately reflect the unidirectional flow characteristics of sewage under gravity, providing a topological basis for subsequent risk diffusion calculations.

[0094] After the graph model is constructed, dynamic risk attributes need to be assigned to each node. For the obtained siltation feature vectors, key components reflecting the degree of silt accumulation are extracted, such as the normalized value of silt layer thickness and siltation rate indicators. A historical time window is selected, the length of which is determined based on the pipeline network operation cycle, typically continuous data from the past 30 to 90 days. In the pre-established siltation state space model, the degree of siltation can be mapped to state points in a multi-dimensional space. By tracing back multiple siltation feature vectors corresponding to the node within the historical window, a series of historical positions of the state point in space are obtained. These discrete position points are arranged in chronological order, and a smooth state evolution trajectory curve is fitted using cubic spline interpolation or local weighted regression. The derivative of this curve at the current moment yields the instantaneous velocity vector of the state point moving along the trajectory; its magnitude is the instantaneous evolution rate. The larger the rate value, the faster the siltation deterioration.

[0095] Simultaneously, based on the current sedimentation feature vector as the initial state, a forward extrapolation is performed using a sedimentation state space model. This model incorporates a coupling mechanism between the sedimentation dynamics equation and the fluid scouring effect. A time step is set for future extrapolation, such as the next 7 days, with each hour as the step unit. The movement trajectory of the state points in space is calculated progressively, forming a position prediction sequence for the future time period. For each position point in the prediction sequence, the Euclidean distance between it and the current state point, or the risk metric distance defined in the state space, is calculated. The larger this distance, the more severe the deviation of the sedimentation state from the current safe state. The distance value corresponding to each predicted position is divided by the time interval between that position and the current time to obtain the risk increment per unit time, constituting a risk increment time series.

[0096] For each node in the risk propagation graph, the aforementioned two types of information are integrated into a local risk intensity index. In the specific calculation, an exponentially decaying weight is applied to the risk increment sequence according to the time distance, and the predicted risk increment that is closer to the current time is given a higher weight. The weighted risk increment sequence is summed and coupled with the instantaneous evolution rate by multiplication to obtain a scalar value reflecting the current threat level of the node, which is denoted as the local risk intensity. This value takes into account the inertia of historical trends and the prediction of future states, and can more accurately quantify the danger of the node itself.

[0097] After determining the node risks, it is necessary to calculate the propagation characteristics of the risks along the directed edges. For the directed edge connecting node i and node j, the physical parameters of the pipe segment between the two nodes are extracted. Based on the pipe segment length L and the average flow velocity v, the hydraulic residence time required for the fluid to flow from the upstream well to the downstream well is calculated. ,Right now This time delay reflects that the siltation risk signal from the upstream well needs a certain amount of time to affect the downstream well. Therefore, when calculating the risk propagation coefficient of an edge, the current risk value of the upstream node should not be used directly, but rather backtracked from its risk increment sequence. Duration: Extract the risk increment value at the corresponding moment. This value represents the current strength of the risk signal propagating from upstream to downstream.

[0098] Furthermore, the amplification or inhibition effect of the local risk intensity of upstream nodes on downstream propagation is considered. The risk increment value obtained from backtracking is nonlinearly coupled with the local risk intensity of upstream nodes, for example, by using the square root of their product, or by introducing hydraulic parameters such as pipe resistance coefficient and flow coefficient for correction. This coupled calculation can simulate the chain reaction in actual physical processes where severe upstream siltation leads to reduced flow velocity and intensified sediment deposition, thereby accelerating the downstream spread of risk. The calculated value is the risk propagation coefficient of the directed edge, which describes the amount of risk transmitted from upstream nodes to downstream nodes per unit time.

[0099] The local risk intensity of all nodes is stored as node attributes, and the risk propagation coefficients of all directed edges are assigned as edge weights to the risk propagation graph. This weighted directed graph structure comprehensively depicts the siltation risk status of each sedimentation well in the urban drainage network and its propagation dynamics within the network. Each node in the graph not only has its own risk assessment but also establishes a quantified influence relationship with its neighboring nodes through edge weights, providing a computational foundation for subsequent execution of dynamic diffusion algorithms, tracking risk propagation paths, and identifying high-risk areas on the graph. This risk propagation graph model can be updated in real time. When new sensor data is collected, the siltation feature vector and local risk intensity of the nodes are recalculated, and the risk propagation coefficients of the edges are dynamically adjusted, ensuring that the risk assessment always reflects the latest operating status of the pipeline network and guaranteeing the timeliness and accuracy of early warning and scheduling decisions.

[0100] Dynamic diffusion calculations are performed on the risk propagation graph model to determine the set of downstream associated wells affected by the current sedimentation well and the cumulative risk value. When the cumulative risk value exceeds a trigger threshold, regional early warning information including the risk propagation path and predicted failure time series is generated, including:

[0101] Using the node corresponding to the current sedimentation well as the initial diffusion source node, the risk propagation coefficients of each directed connection edge on the path between each node and the initial diffusion source node are extracted and multiplied together to obtain the path attenuation factor, which is then multiplied with the local risk intensity of each node to obtain the risk potential energy value.

[0102] For each incoming edge of each node, the difference between the risk potential energy value of the upstream node and each node is calculated as the potential energy gradient. The potential energy gradients of each incoming edge are vector-synthesized to obtain the synthetic potential energy gradient vector. Based on the vector magnitude of the synthetic potential energy gradient vector, the risk accumulation value of each node is calculated. Nodes whose risk accumulation value exceeds the trigger threshold are selected to form a downstream associated well set.

[0103] For each node in the downstream associated well set, backtrack hop by hop along the opposite direction of the synthetic potential energy gradient vector, selecting the upstream node with the largest potential energy gradient magnitude each time, until the initial diffusion source node is reached, and the backtracking path is recorded to form the fastest risk propagation path;

[0104] The total path delay is obtained by summing the time delays of each directed connection edge in the fastest risk propagation path. The predicted arrival time is obtained by adding the current time to the total path delay. The predicted failure time is calculated by combining the risk accumulation value. The predicted failure times of each node are arranged to form a predicted failure time sequence, and integrated to generate regional early warning information.

[0105] After establishing the risk propagation graph model, dynamic diffusion calculations need to be performed based on this model to accurately identify downstream sedimentation wells affected by siltation risk and quantify the path and temporal characteristics of risk propagation. In actual execution, when the siltation feature vector of a sedimentation well shows that its risk level has reached the point of triggering propagation, the node corresponding to that sedimentation well is marked as the initial diffusion source node. Starting from this node, the risk propagation coefficient carried by the directed connection edges in the graph model is extracted to quantitatively characterize the diffusion of risk in the topological structure.

[0106] Specifically, for any node in the risk propagation graph model, calculating the path from that node to the initial diffusion source node requires identifying all directed edges along the path. Each directed edge corresponds to a risk propagation coefficient, which reflects the attenuation or enhancement characteristics of risk during its transmission from upstream to downstream nodes. Multiplying the risk propagation coefficients of all directed edges along the path sequentially yields the path's attenuation factor. This attenuation factor reflects the cumulative attenuation effect of risk during multi-level propagation; the longer the path, the smaller the attenuation factor typically is, indicating that the risk propagation capability weakens with increasing distance. Simultaneously, each node possesses a local risk intensity, calculated using the node's accumulation feature vector, reflecting the contribution of the node's own accumulation state to risk propagation. Multiplying the path attenuation factor by the node's local risk intensity yields the node's risk potential value. This risk potential value comprehensively considers the node's own state and its position in the topology; a higher value indicates a more significant risk to the node in the current diffusion scenario.

[0107] After obtaining the risk potential energy values ​​of each node, the propagation trend of risk along the topology is further calculated. For each node in the risk propagation graph model, there are several incoming edges, each connecting to an upstream node. For a given incoming edge, the difference between the risk potential energy value of the upstream node connected to that edge and the risk potential energy value of the current node is calculated. This difference is defined as the potential energy gradient. A positive potential energy gradient indicates that there is a propagation driving force along the direction of that incoming edge, and the larger the value, the stronger the propagation trend. The potential energy gradients of all incoming edges of a node are treated as vectors, with each potential energy gradient corresponding to a vector component. The vector direction points from the incoming edge to the current node. The potential energy gradient vectors of all incoming edges are vector synthesized to obtain the composite potential energy gradient vector. The direction of this composite vector indicates the main source direction of risk convergence towards that node, and its vector magnitude reflects the cumulative intensity of risk at that node.

[0108] Based on the vector magnitude of the synthetic potential gradient vector, the risk accumulation value of each node is quantified. This risk accumulation value characterizes the overall risk level that the node bears under the current diffusion scenario. The larger the magnitude, the more severe the risk accumulation of the node. The risk accumulation values ​​of all nodes are compared with the preset trigger threshold, and nodes whose risk accumulation values ​​exceed the trigger threshold are selected. These nodes constitute the downstream associated well set, indicating that these sedimentation wells are affected by the initial diffusion source nodes under the current siltation state, and there is a risk of functional failure or accelerated siltation. The setting of the trigger threshold is usually determined by combining the safety margin of the urban drainage system and historical fault data to ensure the sensitivity and accuracy of the early warning.

[0109] For each node in the downstream associated well set, it is necessary to identify the most significant risk propagation path. This is achieved using a potential energy gradient backtracking strategy. Starting from a node in the downstream associated well set, backtracking is performed along the opposite direction of the synthetic potential energy gradient vector. Since the synthetic potential energy gradient vector points in the direction of risk convergence, its opposite direction points to the main source of risk. During the backtracking process, the upstream node with the largest potential energy gradient magnitude among all incoming edges of that node is selected as the next hop node. Selecting the incoming edge with the largest potential energy gradient magnitude means that the driving force of risk propagation along that direction is the strongest, corresponding to the fastest risk propagation path. The hop-by-hop backtracking process continues until the initial diffusion source node is reached. All nodes and directed edges traversed during the backtracking process are recorded sequentially, forming a fastest risk propagation path from the initial diffusion source node to the target downstream node. This path reveals the most likely propagation trajectory of risk in the topology, providing crucial information for subsequent dredging scheduling.

[0110] After identifying the fastest risk propagation path, it is necessary to quantify the temporal characteristics of risk propagation along this path. Each directed connection edge carries a risk propagation coefficient and is associated with a time delay, which characterizes the time required for fluid to flow from an upstream node to a downstream node, reflecting the combined influence of factors such as pipeline structure, flow rate, and distance. The time delays of all directed connections along the fastest risk propagation path are sequentially accumulated to obtain the total time delay of the path. Adding the current time to the total path time delay yields the predicted arrival time of the risk propagation to the target downstream node. Combining the cumulative risk value of the target node with its own carrying capacity parameters, the predicted failure time of the node is calculated. For all nodes in the downstream associated well set, their predicted failure times are calculated separately and arranged in chronological order to form a predicted failure time series. This time series clearly shows the time windows for failure of different sedimentation wells, providing a quantitative basis for prioritizing dredging operations.

[0111] By integrating the fastest risk propagation path, downstream associated well set, predicted failure time series, and risk accumulation values ​​at each node, regional early warning information is generated. This early warning information not only identifies which sedimentation wells are at risk, but also clarifies the propagation path and temporal characteristics of the risk, enabling managers to fully grasp the diffusion status of sedimentation risk in the drainage network and providing a scientific basis for timely initiation of dredging scheduling and emergency response.

[0112] Based on the siltation feature vectors of each sedimentation well in the regional early warning information, the corresponding dredging process parameters are matched. Combined with the risk propagation path and the spatial distribution of dredging resources, a dredging operation scheduling plan is generated and distributed to the dredging execution unit, including:

[0113] Calculate the siltation feature vector of each sedimentation well in the regional early warning information and the similarity with the historical siltation feature vector in the historical process implementation record. Select the historical process implementation record with the highest similarity and extract the corresponding process parameters.

[0114] From the risk propagation path, identify the critical propagation nodes, which are nodes with an out-degree greater than a set value and a risk accumulation value within the target range. Elevate the operation priority of the critical propagation nodes to the beginning of the sequence. For non-critical propagation nodes, establish a time-constraint network, using the predicted failure time as the upper time constraint of the node and the hydraulic propagation time of the pipe segment between adjacent nodes as the lower time constraint of the edge. Solve the topology sorting scheme that satisfies all time constraints in the time-constraint network, and merge the topology sorting scheme with the critical propagation node sequence to form an operation sequence.

[0115] Obtain the spatial location and operational capacity parameters of dredging resources, calculate the total cost of each dredging resource using a rolling time-domain optimization method according to the operational sequence, select the resource allocation strategy with the lowest cost, generate a dredging operation scheduling plan, and send it to the dredging execution unit.

[0116] After obtaining regional early warning information, the siltation status needs to be transformed into an executable dredging operation task. The siltation feature vector of each sedimentation well in the regional early warning information is denoted as... Where i represents the sedimentation well number and m represents the feature dimension, a historical process implementation record database is established. This database stores the siltation status and corresponding process parameters used in past dredging operations, including key parameters such as suction pressure, flushing flow rate, and operation duration. The siltation feature vector of the j-th record in the historical records is denoted as . The corresponding set of process parameters is denoted as .

[0117] Calculate the current sedimentation feature vector of the sedimentation well. With each historical feature vector The similarity between them is measured using the weighted cosine similarity method, which assigns different weight coefficients to different feature dimensions to reflect the importance of each physical quantity in process selection. The similarity calculation formula is as follows:

[0118] ,

[0119] in Let be the weight coefficient of the k-th dimension feature. Among all historical records, select the similarity... Extract the corresponding process parameters from the largest record. As a preliminary process configuration for current sedimentation wells, for cases with thick silt layers, it is necessary to increase the suction pressure and pretreatment time. For siltation conditions with high fluid viscosity, a high-pressure flushing process should be selected in conjunction with a chemical dispersant. This process matching method based on historical experience can avoid blindly setting process parameters and ensure the effectiveness of dredging operations.

[0120] After determining the process parameters for each sedimentation well, it is necessary to rationally arrange the operation sequence to control risk propagation. Key propagation nodes are identified from the risk propagation path; these nodes have a significant propagation influence in the pipeline network. The out-degree of a node is defined as the number of downstream nodes directly connected to that node. A larger out-degree indicates a wider downstream range of influence from that node. An out-degree threshold is then set. The risk accumulation value is typically determined based on 1.5 to 2 times the average outlet value of the pipeline network, while simultaneously defining the target range for the risk accumulation value. The lower bound of this interval This indicates that the risk level has reached a point where it must be addressed; upper bound. This represents the critical value that has not yet caused a systemic failure, satisfying the condition that the out-degree is greater than 1. Furthermore, nodes with accumulated risk values ​​within this range are marked as critical propagation nodes. If these nodes are not dealt with in a timely manner, they will cause a large-scale chain of blockages in a short period of time. Therefore, the operation priority of critical propagation nodes is raised to the beginning of the sequence to ensure that dredging resources are allocated first for disposal.

[0121] For non-critical propagation nodes, a time-constrained network is established to prioritize tasks. This network is a directed graph structure, where nodes represent sedimentation wells to be dredged, edges represent temporal dependencies between nodes, and each node is subject to an upper time bound constraint, the value of which is taken from the predicted failure time. This indicates that the dredging operation of sedimentation well i must be completed before this time; otherwise, overflow or pipe blockage will occur. Edges between nodes are subject to a lower time bound constraint, the value of which is the hydraulic propagation time of the pipe segment between adjacent sedimentation wells. The calculation method is to divide the pipe section length by the average fluid velocity. The hydraulic propagation time reflects the delay in the impact of the upstream sedimentation well's sedimentation state on the downstream sedimentation well. If the upstream node is at time [time missing]... Once the dredging is completed, the downstream nodes must be at least at time [time missing]. Only after this can the work begin, in order to avoid the upstream water from interfering with the downstream work during the dredging process.

[0122] In a time-constrained network, a topology sorting scheme satisfying all time constraints is solved. Topology sorting guarantees that in a directed acyclic graph, for any directed edge (u, v), node u always appears before node v in the sorting result. The Kahn algorithm is used to implement topology sorting with time window constraints: calculate the in-degree of each node, and add nodes with an in-degree of zero to the candidate queue; select the node with the most pressing time upper bound constraint from the candidate queue and add it to the sorting result, and update the in-degree of all its successor nodes; when the in-degree of a successor node decreases to zero, check whether the earliest possible start time of the node satisfies the time lower bound constraint of the predecessor node. If it does, add it to the candidate queue; repeat the above process until all nodes are sorted. If the earliest possible start time of a node is found to be later than its time upper bound during the sorting process, it indicates a constraint conflict, and the predicted failure time needs to be adjusted or the dredging resource input needs to be increased. The calculated topology sorting scheme is merged with the critical propagation node sequence to form a complete operation sequence.

[0123] Information on currently available dredging resources is obtained, including the real-time spatial coordinates of each dredging vehicle or equipment, operational capacity parameters (such as single-operation processing capacity and movement speed), and current task status. Among the operational capacity parameters, single-operation processing capacity represents the volume of sludge that the equipment can clean per unit time under standard operating conditions, and movement speed affects the transfer time of the equipment between different sedimentation wells. According to the operation sequence, a rolling time-domain optimization method is used for resource allocation. The rolling time-domain optimization divides the entire operation period into multiple time windows, and solves the locally optimal resource allocation scheme within each time window. The optimization results are continuously updated as time progresses.

[0124] Within each time window, calculate the total cost of tasks in each dredging resource execution sequence. The total cost includes movement costs, operation costs, and delay penalty costs. Movement costs are proportional to the distance from the resource's current location to the target sedimentation well, reflecting the transportation overhead of equipment scheduling.

[0125] Obtain the spatial location and operational capacity parameters of dredging resources. Following the operational sequence, calculate the total cost of each dredging resource using a rolling time-domain optimization method. Select the resource allocation strategy with the lowest cost, generate a dredging operation scheduling plan, and distribute it to the dredging execution unit. This includes:

[0126] Obtain the spatial location coordinates and operational capability parameters of each dredging resource, construct a resource task spatiotemporal correlation matrix, where the rows of the spatiotemporal correlation matrix correspond to each dredging resource and the columns correspond to each sedimentation well node in the operational sequence, calculate the travel time of each dredging resource to each sedimentation well node according to the operational sequence, calculate the operation duration based on the operational capability parameters and process parameters, accumulate the travel time and operation duration and fill them into the corresponding elements of the spatiotemporal correlation matrix, and sum the rows of the spatiotemporal correlation matrix to obtain the total operation duration;

[0127] Set the time window length and rolling step size of the rolling time domain optimization method, extract the subset of nodes to be allocated within the time window from the job sequence, construct a state transition graph for the subset of nodes to be allocated, where nodes represent resource allocation status and edges represent allocation decisions, and use dynamic programming to calculate the sum of the immediate allocation cost and the expected cost in the future period as the total cost weight of the state transition edge.

[0128] The minimum total cost path is searched in the state transition graph. The allocation scheme within the current rolling step of the path is extracted as the execution strategy. The decision time is advanced according to the rolling step and the above calculation is repeated until all nodes complete resource allocation. The dredging operation scheduling scheme is generated and sent to the dredging execution unit.

[0129] During the dredging resource scheduling process, real-time location information of each dredging resource is obtained from the urban management system, including spatial location data such as the GPS coordinates of dredging vehicles and the fixed addresses of sewage suction pump stations. Simultaneously, operational capacity parameters for each dredging resource are collected, specifically including the maximum volume of sludge that can be processed in a single operation, the rated power of the equipment, the continuous operating time, and the equipment's moving speed. These parameters constitute a resource description vector, providing basic data support for subsequent task allocation.

[0130] When constructing the spatiotemporal correlation matrix of resource tasks, the number of rows in the matrix equals the total number of schedulable dredging resources, and the number of columns corresponds to the number of sedimentation well nodes in the operation sequence. For the element in the i-th row and j-th column of the matrix, the total time cost of the i-th dredging resource executing the j-th sedimentation well task is calculated. The calculation of travel time needs to consider the actual road network distance from the current location of the dredging resource to the target sedimentation well. The shortest path algorithm of the urban road network is used to obtain the travel distance, and the travel time is calculated by combining the movement speed and the road congestion coefficient. The operation duration is determined based on the silt layer thickness and silt volume in the siltation feature vector of the target sedimentation well, combined with the pumping rate in the matched dredging process parameters. For example, when the silt volume is V cubic meters and the pumping rate of the dredging resource is R cubic meters per hour, the operation duration can be estimated as V / R hours. At the same time, the equipment preparation time and the final cleaning time are added together. The travel time and operation duration are then added together and filled into the corresponding position in the matrix to form a complete spatiotemporal correlation matrix. Summing each row of the matrix yields the total operation time required for each dredging resource to complete the entire operation sequence. This indicator can provide a preliminary assessment of the balance of resource load.

[0131] The implementation of the rolling time-domain optimization method requires pre-setting the time window length and rolling step size parameters. The time window length determines the range of future time periods covered by each optimization calculation, and is usually set to include a time span of 3 to 5 sedimentation well nodes, which balances computational efficiency and captures the mutual influence between recent tasks. The rolling step size represents the time interval for decision updates, and is generally set to the average working cycle of one or two sedimentation well nodes. Based on the set time window length, a subsequence starting from the current moment is extracted from the work sequence to form a subset of nodes to be allocated. A state transition graph is constructed for this subset, where the nodes represent resource allocation states. Each state node is described by a vector describing the set of tasks currently undertaken by each dredging resource and its expected completion time. The state transition edge represents an allocation decision, that is, assigning a sedimentation well task in the subset of nodes to be allocated to a dredging resource.

[0132] In the construction of the state transition graph, the edge weight consists of two cost components. The immediate allocation cost reflects the direct expenses caused by the current decision, mainly including the travel cost of moving dredging resources from the current task point to the new task point, the direct cost of operation execution, and the penalty cost caused by task delay. The travel cost is related to the travel distance and fuel consumption per unit distance, the operation execution cost is related to equipment running time and operating cost per unit time, and the delay penalty cost is calculated based on the deviation between the risk accumulation value of the sedimentation well and the task start time. The expected cost in the future period is estimated through a heuristic function to assess the impact of the current allocation decision on the subsequent task allocation space. For example, if a high-efficiency dredging resource is pre-allocated to a low-risk node, it will lead to a lack of suitable resources for subsequent high-risk nodes. This potential risk is reflected in the expected cost. The immediate allocation cost and the expected cost in the future period are added together as the total cost weight of the state transition edge.

[0133] A dynamic programming approach is used to calculate the minimum total cost path in the state transition graph. A state value function is defined, representing the minimum cumulative cost to reach the current state from the initial state. The value function of each state is updated by traversing the nodes of the state transition graph layer by layer. For each node to be assigned, all feasible resource allocation options are enumerated, the total cost of transitioning from the previous state to the new state is calculated, and the transition path that minimizes the state value function is retained. After traversing all nodes within the time window, the optimal path is backtracked, and the allocation scheme within the current rolling step size is extracted as the execution strategy for this round. This execution strategy explicitly specifies which sedimentation well tasks are allocated to which dredging resources at the current decision moment.

[0134] The decision-making process advances in a rolling step, updating the location status and available time window of dredging resources, extracting a new subset of nodes to be allocated from the work sequence, reconstructing the state transition diagram and performing dynamic programming calculations. Through iterative rolling processes, the resource allocation of all sedimentation well nodes is gradually completed. This rolling optimization mechanism can dynamically respond to uncertainties in the task execution process. For example, if a dredging resource is delayed due to a fault, the allocation scheme can be readjusted in subsequent rolling cycles to maintain the robustness of the overall scheduling.

[0135] The generated dredging operation scheduling plan is organized in structured data format, including the allocated resource number, planned start time, estimated operation duration, dredging process type, and process parameter settings for each sedimentation well node. The plan also provides a sequence of resource movement paths, indicating the sedimentation well locations and arrival times for each dredging resource. This plan is transmitted to the vehicle-mounted terminals or control center of each dredging execution unit via a wireless communication network. The execution unit navigates and performs operations based on the received command parameters. During plan execution, the dredging execution unit transmits operation progress and actual completion time back in real time. The scheduling system dynamically adjusts the rolling optimization parameters for subsequent periods based on the feedback information, achieving closed-loop management. Through the above multi-stage calculation and rolling optimization mechanism, it is ensured that the allocation of dredging resources satisfies both the priority constraints of risk propagation paths and minimizes the overall operation cost.

[0136] This invention provides a multi-sensor-based urban road sedimentation well sludge early warning and dredging scheduling system, comprising:

[0137] The signal acquisition unit is used to acquire physical quantity signals through multiple types of sensors in the sedimentation well. The physical quantity signals include the thickness of the silt layer, fluid flow characteristics, and well body structural response.

[0138] The feature extraction unit is used to perform time-frequency domain decomposition on the physical quantity signal and extract multi-scale feature spectrum, perform nonlinear mapping on the multi-scale feature spectrum and identify siltation state features to obtain siltation feature vector;

[0139] The graph model unit is used to construct a risk propagation graph based on the topological connectivity of the urban drainage pipe network, wherein each node corresponds to a sedimentation well, and the corresponding risk propagation coefficient is calculated based on the sedimentation feature vector.

[0140] The risk warning unit is used to perform dynamic diffusion calculations on the risk propagation graph model, determine the set of downstream associated wells affected by the current sedimentation well and the risk accumulation value, and generate regional warning information including risk propagation path and predicted failure time series when the risk accumulation value exceeds the trigger threshold.

[0141] The scheduling generation unit is used to match the corresponding dredging process parameters based on the siltation feature vectors of each sedimentation well in the regional early warning information, and generate a dredging operation scheduling plan by combining the risk propagation path and the spatial distribution of dredging resources, and then send it to the dredging execution unit.

[0142] A third aspect of the present invention provides an electronic device, comprising:

[0143] processor;

[0144] Memory used to store processor-executable instructions;

[0145] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0146] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0147] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0148] 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 early warning and dredging scheduling of silt in urban road sedimentation wells based on multiple sensors, characterized in that, include: Physical quantity signals are collected by multiple types of sensors inside the sedimentation well, including silt layer thickness, fluid flow characteristics, and well structure response. The physical quantity signal is decomposed in the time-frequency domain and a multi-scale feature spectrum is extracted. The multi-scale feature spectrum is then nonlinearly mapped and siltation state features are identified to obtain a siltation feature vector. A risk propagation graph is constructed based on the topological connectivity of the urban drainage pipe network, where each node corresponds to a sedimentation well, and the corresponding risk propagation coefficient is calculated based on the sedimentation feature vector. Dynamic diffusion calculations are performed on the risk propagation graph model to determine the set of downstream associated wells affected by the current sedimentation well and the risk accumulation value. When the risk accumulation value exceeds the trigger threshold, regional early warning information including the risk propagation path and the predicted failure time series is generated. Based on the siltation feature vectors of each sedimentation well in the regional early warning information, the corresponding dredging process parameters are matched, and combined with the risk propagation path and the spatial distribution of dredging resources, a dredging operation scheduling plan is generated and sent to the dredging execution unit.

2. The method according to claim 1, characterized in that, The physical quantity signal is decomposed in the time-frequency domain and a multi-scale feature spectrum is extracted. The multi-scale feature spectrum is then nonlinearly mapped and siltation state features are identified to obtain a siltation feature vector, including: Multi-resolution time-frequency decomposition is performed on the physical quantity signal, dividing the physical quantity signal into several scale intervals on the time axis and frequency axis. Subspace projection decomposition is performed on the frequency domain component of each scale interval, decomposing the frequency domain component into a dominant subspace component and a residual subspace component. The energy proportion of the dominant subspace component and the fluctuation amplitude of the residual subspace component are calculated respectively. The energy proportion and fluctuation amplitude of all scale intervals are combined to form the multi-scale feature spectrum. Construct a siltation state space model, map the energy proportion of the dominant subspace component and the fluctuation amplitude of the residual subspace component in the multi-scale feature spectrum to the siltation state space model, and determine the current state node position of the physical quantity signal in the siltation state space model; Based on the multi-scale feature spectrum sequence within the historical time window, the migration trajectory of the state node position is tracked in the siltation state space model, the direction vector and migration rate of the migration trajectory are extracted, and the state node position, the direction vector and the migration rate are encapsulated into the siltation feature vector.

3. The method according to claim 2, characterized in that, Constructing a siltation state space model, mapping the energy proportion of the dominant subspace component and the fluctuation amplitude of the residual subspace component in the multi-scale feature spectrum to the siltation state space model, and determining the current state node position of the physical quantity signal in the siltation state space model, including: A two-dimensional feature space is established with the energy proportion of the dominant subspace component and the fluctuation amplitude of the residual subspace component as coordinate axes. According to the feature distribution law of different stages in the accumulation evolution process, multiple topological division boundaries are set in the two-dimensional feature space. The topological division boundaries divide the two-dimensional feature space into several non-overlapping state subspaces. For each state subspace, a corresponding discrete state node is set. Based on the unidirectional and reversible characteristics of siltation evolution, a directed migration path is established between the state nodes corresponding to adjacent state subspaces that share the topological partition boundary, thus forming the siltation state space model. Extract the energy proportion of the dominant subspace component and the fluctuation amplitude of the residual subspace component at the current moment, and mark them as the current feature point in the two-dimensional feature space; calculate the shortest vertical distance from the current feature point to the topological segmentation boundary of each state subspace, determine the state subspace to which the current feature point belongs, and determine the state node corresponding to the state subspace as the state node position.

4. The method according to claim 1, characterized in that, A risk propagation graph is constructed based on the topological connectivity of the urban drainage pipe network, where each node corresponds to a sedimentation well, and the corresponding risk propagation coefficient is calculated based on the sedimentation feature vector, including: Based on the physical connectivity of the urban drainage network, the upstream and downstream connection relationships between each sedimentation well are extracted. Each sedimentation well is abstracted as a node in the risk propagation graph model. Directed connection edges are established between the nodes corresponding to sedimentation wells with upstream and downstream connection relationships to construct the risk propagation graph. Based on the siltation feature vector, the trajectory of the change of state node positions within the historical time window is extracted, a nonlinear trend curve is fitted, and the instantaneous evolution rate at the current moment is calculated; according to the siltation feature vector, the predicted sequence of state node positions within the future time step is deduced in the siltation state space model, and the risk increment corresponding to each predicted position is calculated. For each node, the instantaneous evolution rate and the risk increment are summed in a time-weighted manner to calculate the local risk intensity of each node. For each directed connection edge, the time delay of risk signal propagation is calculated based on the hydraulic parameters of the pipe segment between the upstream and downstream nodes. Based on the time delay, the risk increment value at the corresponding moment is selected from the risk increment sequence of the upstream node, and coupled with the local risk intensity of the upstream node to obtain the risk propagation coefficient corresponding to each directed connection edge, which is then superimposed on the risk propagation graph.

5. The method according to claim 1, characterized in that, Dynamic diffusion calculations are performed on the risk propagation graph model to determine the set of downstream associated wells affected by the current sedimentation well and the cumulative risk value. When the cumulative risk value exceeds a trigger threshold, regional early warning information including the risk propagation path and predicted failure time series is generated, including: Using the node corresponding to the current sedimentation well as the initial diffusion source node, the risk propagation coefficients of each directed connection edge on the path between each node and the initial diffusion source node are extracted and multiplied together to obtain the path attenuation factor, which is then multiplied with the local risk intensity of each node to obtain the risk potential energy value. For each incoming edge of each node, the difference between the risk potential energy value of the upstream node and each node is calculated as the potential energy gradient. The potential energy gradients of each incoming edge are vector-synthesized to obtain the synthetic potential energy gradient vector. Based on the vector magnitude of the synthetic potential energy gradient vector, the risk accumulation value of each node is calculated. Nodes whose risk accumulation value exceeds the trigger threshold are selected to form a downstream associated well set. For each node in the downstream associated well set, backtrack hop by hop along the opposite direction of the synthetic potential energy gradient vector, selecting the upstream node with the largest potential energy gradient magnitude each time, until the initial diffusion source node is reached, and the backtracking path is recorded to form the fastest risk propagation path; The total path delay is obtained by summing the time delays of each directed connection edge in the fastest risk propagation path. The predicted arrival time is obtained by adding the current time to the total path delay. The predicted failure time is calculated by combining the risk accumulation value. The predicted failure times of each node are arranged to form a predicted failure time sequence, and integrated to generate regional early warning information.

6. The method according to claim 1, characterized in that, Based on the siltation feature vectors of each sedimentation well in the regional early warning information, the corresponding dredging process parameters are matched. Combined with the risk propagation path and the spatial distribution of dredging resources, a dredging operation scheduling plan is generated and distributed to the dredging execution unit, including: Calculate the siltation feature vector of each sedimentation well in the regional early warning information and the similarity with the historical siltation feature vector in the historical process implementation record. Select the historical process implementation record with the highest similarity and extract the corresponding process parameters. From the risk propagation path, identify the critical propagation nodes, which are nodes with an out-degree greater than a set value and a risk accumulation value within the target range. Elevate the operation priority of the critical propagation nodes to the beginning of the sequence. For non-critical propagation nodes, establish a time-constraint network, using the predicted failure time as the upper time constraint of the node and the hydraulic propagation time of the pipe segment between adjacent nodes as the lower time constraint of the edge. Solve the topology sorting scheme that satisfies all time constraints in the time-constraint network, and merge the topology sorting scheme with the critical propagation node sequence to form an operation sequence. Obtain the spatial location and operational capacity parameters of dredging resources, calculate the total cost of each dredging resource using a rolling time-domain optimization method according to the operational sequence, select the resource allocation strategy with the lowest cost, generate a dredging operation scheduling plan, and send it to the dredging execution unit.

7. The method according to claim 6, characterized in that, Obtain the spatial location and operational capacity parameters of dredging resources. Following the operational sequence, calculate the total cost of each dredging resource using a rolling time-domain optimization method. Select the resource allocation strategy with the lowest cost, generate a dredging operation scheduling plan, and distribute it to the dredging execution unit. This includes: Obtain the spatial location coordinates and operational capability parameters of each dredging resource, construct a resource task spatiotemporal correlation matrix, where the rows of the spatiotemporal correlation matrix correspond to each dredging resource and the columns correspond to each sedimentation well node in the operational sequence, calculate the travel time of each dredging resource to each sedimentation well node according to the operational sequence, calculate the operation duration based on the operational capability parameters and process parameters, accumulate the travel time and operation duration and fill them into the corresponding elements of the spatiotemporal correlation matrix, and sum the rows of the spatiotemporal correlation matrix to obtain the total operation duration; Set the time window length and rolling step size of the rolling time domain optimization method, extract the subset of nodes to be allocated within the time window from the job sequence, construct a state transition graph for the subset of nodes to be allocated, where nodes represent resource allocation status and edges represent allocation decisions, and use dynamic programming to calculate the sum of the immediate allocation cost and the expected cost in the future period as the total cost weight of the state transition edge. The minimum total cost path is searched in the state transition graph. The allocation scheme within the current rolling step of the path is extracted as the execution strategy. The decision time is advanced according to the rolling step and the above calculation is repeated until all nodes complete resource allocation. The dredging operation scheduling scheme is generated and sent to the dredging execution unit.

8. A multi-sensor-based urban road sedimentation well sludge early warning and dredging scheduling system, used to implement the method as described in any one of claims 1-7, characterized in that, include: The signal acquisition unit is used to acquire physical quantity signals through multiple types of sensors in the sedimentation well. The physical quantity signals include the thickness of the silt layer, fluid flow characteristics, and well body structural response. The feature extraction unit is used to perform time-frequency domain decomposition on the physical quantity signal and extract multi-scale feature spectrum, perform nonlinear mapping on the multi-scale feature spectrum and identify siltation state features to obtain siltation feature vector; The graph model unit is used to construct a risk propagation graph based on the topological connectivity of the urban drainage pipe network, wherein each node corresponds to a sedimentation well, and the corresponding risk propagation coefficient is calculated based on the sedimentation feature vector. The risk warning unit is used to perform dynamic diffusion calculations on the risk propagation graph model, determine the set of downstream associated wells affected by the current sedimentation well and the risk accumulation value, and generate regional warning information including risk propagation path and predicted failure time series when the risk accumulation value exceeds the trigger threshold. The scheduling generation unit is used to match the corresponding dredging process parameters based on the siltation feature vectors of each sedimentation well in the regional early warning information, and generate a dredging operation scheduling plan by combining the risk propagation path and the spatial distribution of dredging resources, and then send it to the dredging execution unit.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images