Intelligent evaluation method and system for flood priority scheduling for multimodal flood early warning

Abstract

This invention discloses an intelligent assessment method for flood priority scheduling in multimodal flood early warning, comprising the following steps: Step 1, collecting rainfall, water level, and flow velocity data and preprocessing them to generate standardized multimodal input; Step 2, constructing a river network graph structure and extracting node and edge features to generate graph structure input data; Step 3, inputting the graph structure into an EvolveGCN evolved graph convolutional network, embedding scheduling path reachability, resource bottleneck, and execution constraint features, and outputting a set of state propagation features; Step 4, generating node state prediction vectors based on the state propagation features; Step 5, calculating three types of risk scores, applying nonlinear constraints to construct a priority score set; Step 6, generating a scheduling action sequence and collecting feedback data; Step 7, updating the scores and scheduling order using feedback, updating the convolutional structure, and outputting the final instruction. This invention achieves a closed-loop intelligent prediction system for flood scheduling, improving response efficiency and scheduling feasibility.

Description

Technical Field

[0001] This invention relates to the field of intelligent scheduling evaluation technology, and in particular to a flood priority scheduling intelligent evaluation method and system for multimodal flood early warning. Background Technology

[0002] Existing flood warning and dispatching systems mainly rely on rule models driven by hydrological monitoring data for risk assessment and dispatching decisions. Most systems adopt static threshold judgment methods, which use trigger conditions such as rainfall, water level or flow velocity to make graded responses. These methods are relatively fixed in terms of response speed and rule configuration, and are difficult to adapt to the dynamic changes in the flood evolution process in complex watershed structures. They are particularly lacking in multi-source data access, multi-node linkage and control executability assessment.

[0003] Some technical solutions attempt to introduce time-series prediction models or multi-source data fusion methods to make more granular judgments on flood risk. However, these methods often lack the ability to model the river network structure and cannot effectively represent the upstream and downstream dependencies and control path structure between nodes. In the scheduling decision-making stage, traditional strategies are executed sequentially based on preset command priorities, lacking dynamic scoring and feasibility assessment mechanisms. This results in some scheduling actions failing to be implemented under conditions of limited resources or path interruption, thus affecting the overall response effect.

[0004] In terms of graph structure modeling, a few existing methods use static graph neural networks to predict node states, but they lack support for temporal evolution, control accessibility, and feedback update processes. The results of instruction execution are disconnected from the priority model, and the system cannot adjust the scoring logic and scheduling order based on feedback. It lacks effective closed-loop scheduling capabilities. Existing technologies are unable to realize a flood priority scheduling evaluation method that integrates multimodal input, structural dynamic propagation, scheduling feasibility assessment, and feedback-driven updates.

[0005] Therefore, how to provide an intelligent assessment method and system for flood priority scheduling for multimodal flood early warning is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an intelligent evaluation method and system for flood priority scheduling for multimodal flood early warning. This invention employs an evolutionary graph convolutional network and a feasibility-enhancing convolutional structure to construct a scheduling strategy based on scoring-driven, structure-aware, and feedback-updated features, which have the advantages of strong decision-making closure, high response accuracy, and good control feasibility.

[0007] The intelligent evaluation method and system for flood priority scheduling for multimodal flood early warning according to embodiments of the present invention includes the following steps:

[0008] Step 1: Collect rainfall, water level, and flow velocity data, and preprocess them to generate standardized multimodal input data;

[0009] Step 2: Construct the river network graph structure, representing observation points as nodes and river channel connections as edges, extract node and edge features, and generate graph structure input data;

[0010] Step 3: Input the graph structure input data into an evolutionary graph convolutional network built on EvolveGCN, use a feasibility-enhancing convolutional structure to propagate node state features, embed scheduling path reachability, resource bottleneck and execution constraint features during convolution, and output a set of state propagation features;

[0011] Step 4: Generate node state prediction vectors based on the state propagation feature set;

[0012] Step 5: Calculate three types of indicators—basic risk, dynamic threat, and regional vulnerability—based on the node state prediction vector. Apply a nonlinear constraint mechanism that includes an upper limit inhibition function and a normalization function to construct a priority scoring feature set and generate a node priority scoring set.

[0013] Step 6: Generate a scheduling action sequence based on the node priority score set, send the scheduling action sequence to the control object for execution, and collect water level change data, flow velocity change data and response status information of the corresponding nodes during the execution process to construct a scheduling execution feedback data set;

[0014] Step 7: Based on the scheduling execution feedback data set, trigger the dynamic update and reordering operation of the node priority score set and the scheduling action sequence. Update the feasibility-enhancing convolutional structure based on the feedback data set to generate the final scheduling action instruction set.

[0015] Optionally, the preprocessing in step one includes: performing time alignment, missing data completion, anomaly removal, and normalization on the rainfall, water level, and flow velocity data to generate standardized multimodal input data.

[0016] Optionally, the evolutionary graph convolutional network constructed based on EvolveGCN in step three includes a graph convolutional parameter initialization unit, a graph convolutional parameter evolution unit, a gated recursive unit, a node state update unit, a historical state fusion unit, a state storage unit, and an output scheduling unit, specifically:

[0017] In the graph convolution initialization stage, the graph convolution parameter initialization unit receives graph structure input data at the input end of the evolved graph convolutional network, divides the graph structure input data into several time-step graph structure snapshots, and initializes the node state propagation weight matrix and bias terms for each layer of graph convolutional structure based on the time-step graph structure snapshots.

[0018] At the beginning of each time step, the graph convolution parameter evolution unit calls the convolution weight set of the previous time step in the parameter evolution path, combines the node state change information of the graph structure snapshot of the current time step, and performs parameter update operation through the gated recursive unit to generate the convolution weight set of the current time step.

[0019] During the convolutional propagation phase at each time step, the node state update unit calls the historical state propagation feature set for each graph structure node, and performs graph structure convolution operation based on the convolution weight set and adjacent edge weight information of the current time step to calculate the node state propagation feature. The operation is completed in parallel by all nodes in the graph structure.

[0020] After state propagation, the historical state fusion unit calls the state propagation features of the node in the state propagation path of the current time step, and uses splicing, linear fusion or residual connection to complete the historical information fusion and generate the fused node state propagation feature representation.

[0021] After each time step, the state storage unit records the state propagation features of all nodes in the current time step in the temporal feature buffer, which serves as the input basis for the graph convolution parameter evolution and state fusion in the next time step.

[0022] After all time steps are executed, the output scheduling unit integrates the node state propagation feature sets generated in each time step and outputs the final state propagation feature set.

[0023] Optionally, the feasibility-enhancing convolutional structure in step three includes a path reachability control unit, a resource bottleneck weighting unit, and an execution constraint activation unit, specifically:

[0024] Before the node state update unit performs the graph structure convolution operation, the path reachability control unit calls the scheduling path reachability features in the graph structure input data to perform path reachability discrimination processing on the adjacent nodes of each graph structure node, removes unreachable adjacent nodes from the feature aggregation path, and retains reachable adjacent nodes to form the state propagation input set.

[0025] During the graph structure convolution operation, the resource bottleneck weight adjustment unit calls the field representing resource capacity in the graph structure input data, performs the propagation channel weight adjustment operation, maps the resource bottleneck feature value corresponding to the edge to the propagation weight adjustment factor, and adjusts the weight value of the corresponding edge in the convolution weight set by edge.

[0026] After the state propagation features are generated, the execution constraint activation unit calls the execution constraint feature field in the graph structure input data, compares the execution constraint state of each node with the set threshold, marks the nodes that do not meet the conditions and stops the output of the state propagation features of the nodes at the current time step, and retains the state propagation features of the nodes that meet the conditions.

[0027] Optionally, the nonlinear constraint mechanism in step five is executed by the scoring construction unit, specifically including:

[0028] After receiving the node state prediction vector, the scoring construction unit extracts the basic risk indicators to describe the flood intensity, the dynamic threat indicators to describe the rate of state change, and the regional vulnerability indicators to describe the degree of population exposure, forming a three-dimensional scoring indicator set.

[0029] The scoring construction unit applies nonlinear numerical suppression to each type of indicator in the three-dimensional scoring indicator set. The basic risk indicator performs interval stretching through a normalization function, the dynamic threat indicator performs gradient suppression through an exponential transformation, and the regional vulnerability indicator performs saturation value limitation through an upper limit suppression function.

[0030] After performing nonlinear processing, the scoring construction unit calls the set of weight factors to perform weighted combination calculation on the three processed indicators, and constructs a set of node priority scoring features.

[0031] Optionally, the controlled objects in step six include gates, pumping stations, dams, and emergency response equipment. The controlled objects perform opening and closing, pumping, water release, and deployment operations according to the scheduling action sequence, and collect corresponding status information to generate a scheduling execution feedback data set.

[0032] Optionally, the dynamic update and reordering operations in step seven include:

[0033] After receiving the scheduling execution feedback data set, the incremental water level change, the magnitude of flow velocity change, and the execution response status identifier corresponding to each node are extracted. The maximum value is updated for the basic risk indicators of each node in the node priority score set, the difference is replaced for the dynamic threat indicators, and the regional vulnerability indicators are scaled and corrected proportionally to generate the updated node priority score set.

[0034] The updated node priority score set is matched with the target node number bound to the action in the original scheduling action sequence. A new score field is added to each action in the scheduling action sequence, and the actions are sorted in descending order according to the score value to regenerate the scheduling action sequence.

[0035] For nodes marked as execution failure in the scheduling execution feedback data set, extract the records with the response status identifier "failure", set the scheduling path reachability feature field in the edge feature vector of the corresponding path to 0, perform a decrement update operation on the resource bottleneck feature field of the corresponding edge, and perform an up adjustment operation on the execution constraint feature field of the target node to form an updated feasibility enhancement convolutional structure. The reordered scheduling action sequence and the updated feasibility enhancement convolutional structure are then input together to generate the final scheduling action instruction set.

[0036] The intelligent assessment system for flood priority scheduling for multimodal flood early warning according to an embodiment of the present invention includes the following modules:

[0037] The multimodal data preprocessing module is used to collect rainfall, water level and flow velocity data, and complete time alignment, anomaly removal and normalization processing;

[0038] The graph structure construction module is used to construct a river network graph structure from the preprocessed data, extract node features and edge features, and generate graph structure input data.

[0039] The evolutionary graph convolution processing module is used to receive graph structure input data, perform graph convolution operations based on EvolveGCN, and output a set of state propagation features.

[0040] The state prediction generation module is used to generate node state prediction vectors based on the state propagation feature set, which represent the future hydrological state of the nodes.

[0041] The priority scoring module is used to calculate three types of indicators and apply nonlinear constraints to generate a set of node priority scores.

[0042] The scheduling action generation module is used to generate a sequence of scheduling actions based on the score set and distribute them to each controlled object for execution.

[0043] The feedback processing and reordering module is used to parse feedback data, update the score set and scheduling sequence, and output the final set of scheduling action instructions.

[0044] The beneficial effects of this invention are:

[0045] (1) An evolutionary graph convolutional network structure is introduced, which combines the node and edge features of the river network graph structure. Based on the modeling of time continuity, the ability to express the evolution process of multi-source hydrological state is enhanced. It can realize the dynamic prediction of node water level, flow and flood peak arrival time, and improve the timeliness and accuracy of flood evolution simulation.

[0046] (2) Design a convolutional structure to enhance feasibility. Embed the reachability features of scheduling paths, resource bottleneck features and execution constraint features into the state propagation process. This avoids the problem of scheduling actions failing on non-executable paths, realizes the structural constraint filtering and path executability guarantee of scheduling actions, and improves the landing rate of control instructions.

[0047] (3) A three-dimensional priority scoring mechanism including basic risks, dynamic threats and regional vulnerabilities is constructed. Combined with nonlinear function processing and feedback data-driven dynamic reordering logic, a complete scoring-scheduling-feedback closed-loop chain is formed, enabling the system to have real-time update, intelligent response and adaptive scheduling capabilities. Attached Figure Description

[0048] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0049] Figure 1 This is an overall flowchart of the intelligent evaluation method for flood priority scheduling for multimodal flood early warning proposed in this invention;

[0050] Figure 2 This is a module connection diagram of the intelligent assessment system for flood priority scheduling for multimodal flood early warning proposed in this invention. Detailed Implementation

[0051] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0052] refer to Figure 1 and Figure 2 A smart assessment method and system for flood priority scheduling for multimodal flood early warning, comprising the following steps:

[0053] Step 1: Collect rainfall, water level, and flow velocity data, and preprocess them to generate standardized multimodal input data;

[0054] Step 2: Construct the river network graph structure, representing observation points as nodes and river channel connections as edges, extract node and edge features, and generate graph structure input data;

[0055] Step 3: Input the graph structure input data into an evolutionary graph convolutional network built on EvolveGCN, use a feasibility-enhancing convolutional structure to propagate node state features, embed scheduling path reachability, resource bottleneck and execution constraint features during convolution, and output a set of state propagation features;

[0056] Step 4: Generate node state prediction vectors based on the state propagation feature set. In the specific implementation process, the state propagation feature set is output by the evolutionary graph convolutional network after multi-time-step structural propagation. The state propagation feature corresponding to each node serves as the graph space embedding representation of the node in the current time series. The system extracts and parses the node feature vectors in the state propagation feature set, and completes the mapping prediction of water level, flow velocity and flood peak arrival time through multi-layer sensing units, outputting node state prediction vectors. The node state prediction vectors contain multi-dimensional hydrological response information within a continuous time window, providing basic support for subsequent risk scoring and scheduling priority judgment. The entire prediction process is time-sensitive and structure-dependent, and can dynamically adapt to graph structure evolution and changes in upstream and downstream influences.

[0057] Step 5: Calculate three types of indicators—basic risk, dynamic threat, and regional vulnerability—based on the node state prediction vector. Apply a nonlinear constraint mechanism that includes an upper limit inhibition function and a normalization function to construct a priority scoring feature set and generate a node priority scoring set.

[0058] Step 6: Generate a scheduling action sequence based on the node priority score set, send the scheduling action sequence to the control object for execution, and collect water level change data, flow velocity change data and response status information of the corresponding nodes during the execution process to construct a scheduling execution feedback data set;

[0059] Step 7: Based on the scheduling execution feedback data set, trigger the dynamic update and reordering operation of the node priority score set and the scheduling action sequence. Update the feasibility-enhancing convolutional structure based on the feedback data set to generate the final scheduling action instruction set.

[0060] In this embodiment, the preprocessing in step one includes: performing time alignment, missing data completion, anomaly removal, and normalization on rainfall, water level, and flow velocity data to generate standardized multimodal input data.

[0061] In this embodiment, the evolutionary graph convolutional network constructed based on EvolveGCN in step three includes a graph convolutional parameter initialization unit, a graph convolutional parameter evolution unit, a gated recursive unit, a node state update unit, a historical state fusion unit, a state storage unit, and an output scheduling unit, specifically:

[0062] In the graph convolution initialization stage, the graph convolution parameter initialization unit receives graph structure input data at the input end of the evolved graph convolutional network, divides the graph structure input data into several time-step graph structure snapshots, and initializes the node state propagation weight matrix and bias terms for each layer of graph convolutional structure based on the time-step graph structure snapshots.

[0063] At the beginning of each time step, the graph convolutional parameter evolution unit calls the convolutional weight set from the previous time step in the parameter evolution path. Combining this with the node state change information from the current time step graph structure snapshot, the gated recursive unit performs parameter update operations to generate the convolutional weight set for the current time step. Specifically, the gated recursive unit receives the convolutional weight set from the previous time step as historical state input, and simultaneously encodes the state change information of each node in the current time step graph structure snapshot as the current input state. Based on the internal structure of the reset gate and update gate, the unit controls the retention ratio of historical weights and the activation level of the current input, respectively. In each update, the gated recursive unit performs weighted combination and state update operations on the parameter matrices of each convolutional layer to generate a convolutional weight set that matches the structural features of the current time step, enabling the graph convolutional network to have temporal adaptability. The update process is executed sequentially in the graph convolutional parameter evolution path to ensure that the convolutional parameters continuously respond to the graph structure state evolution.

[0064] During the convolutional propagation phase at each time step, the node state update unit calls the historical state propagation feature set for each graph structure node. Based on the convolution weight set and adjacent edge weight information of the current time step, it performs graph structure convolution operation to calculate the node state propagation feature. This operation is completed in parallel on all nodes in the graph structure. Specifically, the graph structure convolution operation uses the convolution weight set of the current time step as the kernel function to perform weighted aggregation of the state propagation features of adjacent nodes for each node. During the aggregation process, it combines the adjacent edge weight information and performs weighted cumulative calculation by edge direction to map the state information of adjacent nodes to the target node. After receiving the aggregation result, the target node performs a fusion transformation with its own historical state propagation feature to form the node state propagation feature of the current time step. The convolution operation is synchronously unfolded on all nodes in the graph structure to ensure that the overall state propagation process has spatial consistency and structural constraints.

[0065] After state propagation, the historical state fusion unit calls upon the state propagation features of the node from past time steps within the state propagation path of the current time step. It then uses concatenation, linear fusion, or residual connection methods to complete the historical information fusion, generating a fused node state propagation feature representation. Specifically, during the historical information fusion process, the historical state fusion unit first arranges the state propagation features of the target node across multiple time steps in chronological order. When using the concatenation method, feature vectors from different times are directly connected along the channel dimension, expanding the feature representation dimension. When using the linear fusion method, the feature vectors from each time step are weighted and summed according to set weight coefficients to generate a smoothly transitioning fusion vector. When using the residual connection method, the state propagation features of the current time step are summed element-wise with the feature vectors of the previous time step, preserving the residual changes along the evolution direction. These operations can be flexibly configured to adapt to different structural evolution rates and information decay levels, ensuring that the fused node features retain historical information while also being timely.

[0066] After each time step, the state storage unit records the state propagation features of all nodes in the current time step in the temporal feature buffer, which serves as the input basis for the graph convolution parameter evolution and state fusion in the next time step.

[0067] After all time steps are executed, the output scheduling unit integrates the node state propagation feature sets generated in each time step and outputs the final state propagation feature set.

[0068] In this embodiment, the feasibility-enhancing convolutional structure in step three includes a path reachability control unit, a resource bottleneck weighting unit, and an execution constraint activation unit, specifically:

[0069] Before the node state update unit performs the graph structure convolution operation, the path reachability control unit calls the scheduling path reachability features in the graph structure input data to perform path reachability discrimination processing on the adjacent nodes of each graph structure node. Unreachable adjacent nodes are removed from the feature aggregation path, and reachable adjacent nodes are retained to form the state propagation input set. In specific implementation, the reachability discrimination processing reads and parses the scheduling path reachability feature field associated with each edge in the graph structure input data. The field records whether there is a valid control path connection relationship between nodes. The value range is usually represented by Boolean value or continuous score. During the discrimination process, the path reachability control unit traverses the adjacent edges of each node in turn and performs threshold judgment or logical judgment operation on the reachability field of the edge. If the judgment result is unreachable, the corresponding adjacent node is removed from the feature aggregation path of the current node; if it is reachable, it is retained for subsequent state propagation. This processing step ensures that information transmission is performed only between nodes that have a valid connection in terms of physical or control, improving the structural rationality and scheduling executability of the propagation path.

[0070] During the graph structure convolution operation, the resource bottleneck weight adjustment unit calls the field representing resource capacity in the graph structure input data and performs a propagation channel weight adjustment operation. It maps the resource bottleneck feature values ​​corresponding to each edge to a propagation weight adjustment factor, adjusting the weight values ​​of the corresponding edges in the convolution weight set edge-by-edge. During the propagation channel weight adjustment operation, the resource bottleneck weight adjustment unit extracts resource bottleneck feature values ​​from the resource capacity field of each edge in the graph structure input data. These features reflect the degree of limitation in the transmission capacity of the edge. The system uses a preset mapping function, such as normalization or inverse proportionality, to convert the resource bottleneck values ​​into propagation weight adjustment factors. These factors are applied to the convolution weights of each edge during the graph structure convolution process, performing edge-by-edge multiplication adjustment. The adjusted edge weights directly affect the contribution of adjacent node features in the convolution aggregation, thereby suppressing the information transmission intensity on bottleneck edges and prioritizing the propagation of state information through resource-rich paths, improving the overall scheduling rationality and stability of the propagation path.

[0071] After the state propagation features are generated, the execution constraint activation unit calls the execution constraint feature field in the graph structure input data to compare the execution constraint status of each node with the set threshold. Nodes that do not meet the conditions are marked and the output of the state propagation features for the current time step is stopped, while the state propagation features of nodes that meet the conditions are retained. Specifically, the comparison operation between the execution constraint status and the set threshold is triggered by the execution constraint activation unit after state propagation is completed. The system reads the execution constraint feature field corresponding to each node in the graph structure input data. The field quantifies the controllability level of the node in the current scheduling cycle, represented by an integer score or Boolean identifier. The activation unit compares this value with the preset threshold node by node. When the execution constraint status is lower than the threshold, the node is determined to not meet the scheduling execution conditions, marked as disabled, and the output of the state propagation features for the current time step is stopped. When the threshold conditions are met, the node state features are retained normally and used for subsequent processing. This comparison mechanism ensures that only nodes with controllable resources participate in scheduling decisions, improving the feasibility and reliability of scheduling results.

[0072] In this embodiment, the nonlinear constraint mechanism in step five is executed by the scoring construction unit, specifically including:

[0073] After receiving the node state prediction vector, the scoring construction unit extracts the basic risk indicators to describe the flood intensity, the dynamic threat indicators to describe the rate of state change, and the regional vulnerability indicators to describe the degree of population exposure, forming a three-dimensional scoring indicator set.

[0074] The scoring construction unit applies nonlinear numerical suppression to each indicator in the three-dimensional scoring indicator set. Basic risk indicators undergo interval stretching through a normalization function, dynamic threat indicators undergo gradient suppression through an exponential transformation, and regional vulnerability indicators undergo saturation value limitation through an upper bound suppression function. During the processing of basic risk indicators, the scoring construction unit uses a minimum-maximum normalization method to linearly map the indicator values ​​to a set interval [0,1], eliminating numerical scale differences between different nodes and highlighting differences in risk proportions. Dynamic threat indicators are processed using an exponential transformation, in the form of… ,in The rate of change of state. To adjust the coefficients, this method gives higher response sensitivity to nodes with drastic changes, while suppressing the scoring influence of low-fluctuation nodes. The regional vulnerability index is pruned by setting a fixed upper limit threshold. When the original index value exceeds the upper limit, it is directly assigned the threshold value to prevent nodes with high population exposure or high infrastructure density from being infinitely amplified in the scoring and to avoid scheduling priority imbalance. The three nonlinear processing methods are designed for different types of index characteristics to enhance the stability and recognizability of the scoring system.

[0075] After performing nonlinear processing, the scoring construction unit calls the set of weight factors to perform weighted combination calculation on the three processed indicators, and constructs a set of node priority scoring features.

[0076] In this embodiment, the controlled objects in step six include gates, pumping stations, dams, and emergency response equipment. The controlled objects perform opening and closing, pumping, water release, and deployment operations according to the scheduling action sequence, and collect corresponding status information to generate a scheduling execution feedback data set.

[0077] In this embodiment, the dynamic update and reordering operation in step seven includes:

[0078] After receiving the scheduling execution feedback data set, the incremental water level change, the magnitude of flow velocity change, and the execution response status identifier corresponding to each node are extracted. The maximum value is updated for the basic risk indicators of each node in the node priority score set, the difference is replaced for the dynamic threat indicators, and the regional vulnerability indicators are scaled and corrected proportionally to generate the updated node priority score set.

[0079] The updated node priority score set is matched with the target node number bound to the action in the original scheduling action sequence. A new score field is added to each action in the scheduling action sequence, and the actions are sorted in descending order according to the score value to regenerate the scheduling action sequence.

[0080] For nodes marked as execution failure in the scheduling execution feedback data set, extract the records with the response status identifier "failure", set the scheduling path reachability feature field in the edge feature vector of the corresponding path to 0, perform a decrement update operation on the resource bottleneck feature field of the corresponding edge, and perform an up adjustment operation on the execution constraint feature field of the target node to form an updated feasibility enhancement convolutional structure. The reordered scheduling action sequence and the updated feasibility enhancement convolutional structure are then input together to generate the final scheduling action instruction set.

[0081] The intelligent assessment system for flood priority scheduling for multimodal flood early warning according to an embodiment of the present invention includes the following modules:

[0082] The multimodal data preprocessing module is used to collect rainfall, water level and flow velocity data, and complete time alignment, anomaly removal and normalization processing;

[0083] The graph structure construction module is used to construct a river network graph structure from the preprocessed data, extract node features and edge features, and generate graph structure input data.

[0084] The evolutionary graph convolution processing module is used to receive graph structure input data, perform graph convolution operations based on EvolveGCN, and output a set of state propagation features.

[0085] The state prediction generation module is used to generate node state prediction vectors based on the state propagation feature set, which represent the future hydrological state of the nodes.

[0086] The priority scoring module is used to calculate three types of indicators and apply nonlinear constraints to generate a set of node priority scores.

[0087] The scheduling action generation module is used to generate a sequence of scheduling actions based on the score set and distribute them to each controlled object for execution.

[0088] The feedback processing and reordering module is used to parse feedback data, update the score set and scheduling sequence, and output the final set of scheduling action instructions.

[0089] Example 1: To verify the feasibility of the present invention in practice, the present invention was applied to flood warning and coordinated dispatching tasks in a multi-basin confluence area. The terrain in this area changes dramatically, covering multiple medium-sized tributaries upstream and major downstream drainage channels. During the flood season, there is a risk of multiple sources of water superimposed, resulting in problems such as untimely peak warnings, unreachable dispatching routes, pump station response failures, and delayed command execution. This restricts the operational stability of the regional flood control system.

[0090] During deployment, a multimodal sensing system is used to synchronously collect data from rainfall monitoring points, water level monitoring points, and flow velocity measurement sections within the region. All data undergoes time alignment, anomaly removal, and normalization in the system to form standardized input data. Using the standardized data as input, a river network map structure is constructed. Nodes in the map correspond to monitoring points and control nodes, and edges represent river connectivity. Each node and edge is assigned a feature vector, including real-time observation values, terrain attributes, historical evolution trends, and control attributes.

[0091] After construction, the graph structure is input into an evolutionary graph convolutional network based on EvolveGCN. At each time step, the state propagation path is dynamically updated through the graph convolution parameter evolution unit. During the process, the feasibility enhancement convolutional structure is called to perform real-time suppression and activation processing on the reachability of the scheduling path, the resource bottleneck constraints on the edges, and the execution controllability of the nodes. State changes are propagated only on the adjustable path. Finally, the node state prediction results are output, including the predicted water level, the predicted flood peak arrival time, and the flow velocity evolution trend.

[0092] In the priority scoring construction module, the system calculates three scoring dimensions—basic risk, dynamic threat, and regional vulnerability—based on the state prediction results. To prevent a single indicator from dominating the ranking, the system applies a normalization function and an upper limit suppression function to each indicator to construct a nonlinear weighted priority scoring model. Each scheduling object, such as pumping stations, gates, dams, and emergency equipment, enters the scheduling action sequence according to the scoring results.

[0093] During actual execution, the system continuously receives feedback data, including water level change rate, scheduling response delay, and execution result code, and updates the score set and scheduling sequence in real time. When the water level rise of a node exceeds the original estimate, the system automatically increases its risk score. If a path fails to execute twice consecutively, the system sets the corresponding edge as unreachable, the scheduling path is forcibly interrupted, and it is removed in the next round of generation. When the execution response delay exceeds a set threshold, the system increases the execution constraint strength of the target node to reduce the probability of future repeated scheduling.

[0094] During the actual trial operation phase, 202 historical flood event data points were selected for comparative testing. The comparison method involved a full-process simulation of the system of this invention against existing traditional scheduling methods. The following is a comparison performance data table, demonstrating the advantages of this invention from the perspectives of prediction accuracy, response latency, scheduling coverage, and execution success rate.

[0095] Table 1: Performance Comparison between the Flood Priority Dispatch Intelligent Evaluation System and Traditional Dispatch Schemes

[0096]

[0097] In terms of node status prediction, the mean square error of water level prediction in this invention is 0.172 meters, which is significantly better than the 0.381 meters of the traditional scheme. The error is reduced by more than half, indicating that the simulation of hydrological evolution using evolutionary graph convolutional networks is more accurate. In the prediction of flood peak arrival time, this system controls the error within 6.2 minutes, while the average error of the traditional scheme reaches 15.7 minutes, improving the prediction accuracy by nearly 60%, which is helpful for advance scheduling response.

[0098] The system response speed shows a significant difference. This system achieves an average dynamic score update cycle of 0.44 seconds. Traditional solutions do not support score updates and are difficult to adapt to changes in flood conditions. In terms of average scheduling response delay, the system of this invention is only 32.1 seconds, while the traditional solution is 85.6 seconds, improving response timeliness by more than 60% and ensuring that critical scheduling instructions can be executed in a timely manner within the window period.

[0099] In terms of scheduling reliability, the system of this invention achieves a command execution success rate of 96.7%, far exceeding the 81.2% of the traditional solution. It can also update the feasibility structure field in real time based on the execution results. During the trial operation, the field was updated a total of 1,829 times, demonstrating a strong adaptive scheduling capability. The coverage rate of high-risk nodes and the hit rate of multi-round scheduling reached 93.8% and 87.1% respectively, compared with the 76.5% and 62.9% of the traditional solution, enhancing the scheduling's ability to protect key areas and the accuracy of multi-round linkage.

[0100] In response to sudden failure paths, this system can complete path elimination and structural adjustment within an average of 1.27 seconds. Traditional solutions do not have this capability, demonstrating the advantages of closed-loop structural control. Finally, in terms of system operating efficiency, the average time for a single round of full-map calculation in this invention is 28.4 seconds, which is much lower than the 119.6 seconds of traditional solutions, improving processing efficiency by more than 76%.

[0101] Overall, the system of this invention outperforms traditional scheduling schemes in terms of prediction accuracy, response efficiency, execution stability, and structural adaptability. It can support intelligent scheduling tasks in complex watersheds and improve the scheduling closed-loop capability of multimodal flood early warning.

[0102] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A smart evaluation method for flood priority scheduling for multimodal flood early warning, characterized in that, Includes the following steps: Step 1: Collect rainfall, water level, and flow velocity data, and preprocess them to generate standardized multimodal input data; Step 2: Construct the river network graph structure, representing observation points as nodes and river channel connections as edges, extract node and edge features, and generate graph structure input data; Step 3: Input the graph structure input data into an evolutionary graph convolutional network built on EvolveGCN, use a feasibility-enhancing convolutional structure to propagate node state features, embed scheduling path reachability, resource bottleneck and execution constraint features during convolution, and output a set of state propagation features; The evolutionary graph convolutional network built on EvolveGCN includes a graph convolution parameter initialization unit, a graph convolution parameter evolution unit, a gated recursive unit, a node state update unit, a historical state fusion unit, a state storage unit, and an output scheduling unit. In the graph convolution initialization stage, the graph convolution parameter initialization unit receives graph structure input data at the input end of the evolved graph convolutional network, divides the graph structure input data into several time-step graph structure snapshots, and initializes the node state propagation weight matrix and bias terms for each layer of graph convolutional structure based on the time-step graph structure snapshots. At the beginning of each time step, the graph convolution parameter evolution unit calls the convolution weight set of the previous time step in the parameter evolution path, combines the node state change information of the graph structure snapshot of the current time step, and performs parameter update operation through the gated recursive unit to generate the convolution weight set of the current time step. During the convolutional propagation phase at each time step, the node state update unit calls the historical state propagation feature set for each graph structure node, and performs graph structure convolution operation based on the convolution weight set and adjacent edge weight information of the current time step to calculate the node state propagation feature. The operation is completed in parallel by all nodes in the graph structure. After state propagation, the historical state fusion unit calls the state propagation features of the node in the state propagation path of the current time step, and uses splicing, linear fusion or residual connection to complete the historical information fusion and generate the fused node state propagation feature representation. After each time step, the state storage unit records the state propagation features of all nodes in the current time step in the temporal feature buffer, which serves as the input basis for the graph convolution parameter evolution and state fusion in the next time step. After all time steps are executed, the output scheduling unit integrates the node state propagation feature sets generated in each time step and outputs the final state propagation feature set. Step 4: Generate node state prediction vectors based on the state propagation feature set; Step 5: Calculate three types of indicators—basic risk, dynamic threat, and regional vulnerability—based on the node state prediction vector. Apply a nonlinear constraint mechanism that includes an upper limit inhibition function and a normalization function to construct a priority scoring feature set and generate a node priority scoring set. Step 6: Generate a scheduling action sequence based on the node priority score set, send the scheduling action sequence to the control object for execution, and collect water level change data, flow velocity change data and response status information of the corresponding nodes during the execution process to construct a scheduling execution feedback data set; Step 7: Based on the scheduling execution feedback data set, trigger the dynamic update and reordering operation of the node priority score set and the scheduling action sequence. Update the feasibility-enhancing convolutional structure based on the feedback data set to generate the final scheduling action instruction set.

2. The intelligent evaluation method for flood priority scheduling for multimodal flood early warning as described in claim 1, characterized in that, The preprocessing in step one includes: performing time alignment, missing data completion, anomaly removal, and normalization on rainfall, water level, and flow velocity data to generate standardized multimodal input data.

3. The intelligent evaluation method for flood priority scheduling for multimodal flood early warning as described in claim 2, characterized in that, The feasibility-enhancing convolutional structure in step three includes a path reachability control unit, a resource bottleneck weighting unit, and an execution constraint activation unit, specifically: Before the node state update unit performs the graph structure convolution operation, the path reachability control unit calls the scheduling path reachability features in the graph structure input data to perform path reachability discrimination processing on the adjacent nodes of each graph structure node, removes unreachable adjacent nodes from the feature aggregation path, and retains reachable adjacent nodes to form the state propagation input set. During the graph structure convolution operation, the resource bottleneck weight adjustment unit calls the field representing resource capacity in the graph structure input data, performs the propagation channel weight adjustment operation, maps the resource bottleneck feature value corresponding to the edge to the propagation weight adjustment factor, and adjusts the weight value of the corresponding edge in the convolution weight set by edge. After the state propagation features are generated, the execution constraint activation unit calls the execution constraint feature field in the graph structure input data, compares the execution constraint state of each node with the set threshold, marks the nodes that do not meet the conditions and stops the output of the state propagation features of the nodes at the current time step, and retains the state propagation features of the nodes that meet the conditions.

4. The intelligent evaluation method for flood priority scheduling for multimodal flood early warning as described in claim 3, characterized in that, The nonlinear constraint mechanism in step five is executed by the scoring construction unit, specifically including: After receiving the node state prediction vector, the scoring construction unit extracts the basic risk indicators to describe the flood intensity, the dynamic threat indicators to describe the rate of state change, and the regional vulnerability indicators to describe the degree of population exposure, forming a three-dimensional scoring indicator set. The scoring construction unit applies nonlinear numerical suppression to each type of indicator in the three-dimensional scoring indicator set. The basic risk indicator performs interval stretching through a normalization function, the dynamic threat indicator performs gradient suppression through an exponential transformation, and the regional vulnerability indicator performs saturation value limitation through an upper limit suppression function. After performing nonlinear processing, the scoring construction unit calls the set of weight factors to perform weighted combination calculation on the three processed indicators, and constructs a set of node priority scoring features.

5. The intelligent evaluation method for flood priority scheduling for multimodal flood early warning as described in claim 4, characterized in that, The controlled objects in step six include gates, pumping stations, dams, and emergency response equipment. The controlled objects perform opening and closing, pumping, water release, and deployment operations according to the scheduling action sequence, and collect corresponding status information to generate a scheduling execution feedback data set.

6. The intelligent evaluation method for flood priority scheduling for multimodal flood early warning as described in claim 5, characterized in that, The dynamic update and reordering operations in step seven include: After receiving the scheduling execution feedback data set, the incremental water level change, the magnitude of flow velocity change, and the execution response status identifier corresponding to each node are extracted. The maximum value is updated for the basic risk indicators of each node in the node priority score set, the difference is replaced for the dynamic threat indicators, and the regional vulnerability indicators are scaled and corrected proportionally to generate the updated node priority score set. The updated node priority score set is matched with the target node number bound to the action in the original scheduling action sequence. A new score field is added to each action in the scheduling action sequence, and the actions are sorted in descending order according to the score value to regenerate the scheduling action sequence. For nodes marked as execution failure in the scheduling execution feedback data set, extract the records with the response status identifier "failure", set the scheduling path reachability feature field in the edge feature vector of the corresponding path to 0, perform a decrement update operation on the resource bottleneck feature field of the corresponding edge, and perform an up adjustment operation on the execution constraint feature field of the target node to form an updated feasibility enhancement convolutional structure. The reordered scheduling action sequence and the updated feasibility enhancement convolutional structure are then input together to generate the final scheduling action instruction set.

7. A flood priority scheduling intelligent evaluation system for multimodal flood early warning, applied to the flood priority scheduling intelligent evaluation method for multimodal flood early warning as described in any one of claims 1 to 6, characterized in that, Includes the following modules: The multimodal data preprocessing module is used to collect rainfall, water level and flow velocity data, and complete time alignment, anomaly removal and normalization processing; The graph structure construction module is used to construct a river network graph structure from the preprocessed data, extract node features and edge features, and generate graph structure input data. The evolutionary graph convolution processing module is used to receive graph structure input data, perform graph convolution operations based on EvolveGCN, and output a set of state propagation features. The state prediction generation module is used to generate node state prediction vectors based on the state propagation feature set, which represent the future hydrological state of the nodes. The priority scoring module is used to calculate three types of indicators and apply nonlinear constraints to generate a set of node priority scores. The scheduling action generation module is used to generate a sequence of scheduling actions based on the score set and distribute them to each controlled object for execution. The feedback processing and reordering module is used to parse feedback data, update the score set and scheduling sequence, and output the final set of scheduling action instructions.

Images