Analog optimization method for urban anti-inflow drainage scheduling system

Abstract

This invention relates to a simulation optimization method for an urban backflow prevention drainage scheduling system, belonging to the field of drainage optimization scheduling technology. The method includes acquiring real-time water level readings and historical backflow event timestamps in the scheduling network. By synergistically applying water level fluctuation phase tracking, hydraulic propagation time delay matrix modeling, and structural causal counterfactual deduction, this invention achieves accurate identification of congestion propagation direction and rapid location of root cause nodes. It quantifies and pre-evaluates the causal effectiveness of candidate scheduling actions using the area under the intervention effect curve, enabling scheduling decisions to shift from passively responding to water level exceeding limits to actively intervening in congestion sources. It constructs an asymmetric scheduling action space centered on root cause nodes and exponentially decaying along the congestion propagation path. By aligning the adjustable drainage capacity of upstream pumping stations with the real-time spare capacity of the downstream network through propagation time delays, it automatically generates a staggered peak-time sequence scheme of upstream discharge followed by downstream acceptance, thereby significantly improving the accuracy and collaborative efficiency of backflow prevention scheduling.

Description

Technical Field

[0001] This invention relates to the field of drainage optimization scheduling technology, and in particular to a simulation optimization method for an urban backflow prevention drainage scheduling system. Background Technology

[0002] During the operation and scheduling of urban drainage systems, drainage pumping stations control their start-stop and flow regulation based on pipeline water level monitoring data to prevent sewage backflow and urban flooding. With the expansion of urban drainage networks and the increase in extreme rainfall, traditional drainage scheduling methods based on fixed threshold triggers are unable to cope with the complex and ever-changing hydraulic conditions of the pipeline network, exposing problems such as response lag, local congestion spread, and miscoordination between upstream and downstream pumping stations. The scheduling and operation of urban drainage systems mainly rely on offline simulation based on hydraulic models and rule-based scheduling based on human experience. Typical technical approaches include: establishing a pipeline hydraulic model using a storm flood management model, triggering pumping station start-stop by setting water level thresholds, or using optimization methods such as linear programming and genetic algorithms to search for better scheduling schemes. Traditional scheduling relies on preset water level thresholds and fixed rules, which remain unchanged for a long time once set.

[0003] Existing technologies, at the level of scheduling simulation optimization frameworks, have the following fundamental shortcomings: All existing methods employ forward time-series simulations. Forward simulations cannot answer the crucial question of where congestion originates and along what path it propagates. This leads to a systematic deviation between the allocation of scheduling resources and the actual origin of congestion. Simulation and scheduling are two independent, sequentially connected modules. The simulator calculates the future state, and the scheduler makes decisions based on the simulation results. This sequential structure means that scheduling decisions always lag behind the future assumed by the simulation. When strong backflow feedback exists in the system, scheduling actions change the outflow from the pipeline network, thereby changing river levels and, in turn, affecting the drainage capacity of the pipeline network. A structural deviation exists between the simulation results and the scheduling execution. Existing methods do not synchronously evolve scheduling actions as endogenous variables in the simulation process. To address the aforementioned technical shortcomings, a solution is proposed. Summary of the Invention

[0004] The purpose of this invention is to achieve a paradigm shift from passively responding to water levels to actively intervening in the source of congestion by constructing an asymmetric scheduling action space centered on the root cause node and decaying exponentially along the congestion propagation path, thereby significantly improving the accuracy and coordination efficiency of backflow prevention scheduling in drainage systems.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a simulation optimization method for an urban backflow prevention drainage scheduling system, comprising the following steps: S1: Obtain real-time water level readings and historical backflow event timestamps in the scheduling pipeline network, extract hydraulic propagation time delay parameters, calculate the instantaneous fluctuation difference between adjacent nodes, and construct the pipeline network water level fluctuation distribution field; S2: Based on the distribution field of water level fluctuations in the pipeline network, calculate the integral of the water level change rate and the cumulative amount of propagation time delay at each node on the propagation path, screen the target propagation path, mark the terminal node of the target propagation path as the root cause node, and use the node sequence on the target propagation path as the congestion propagation path scheme. S3: Based on the congestion propagation path scheme, the risk of water level exceeding the limit at the root cause node is converted into the backflow risk entropy. The adjustable capacity of each pumping station is matched with the spare capacity of the downstream pipe network to generate an asymmetric scheduling action space centered on the root cause node and along the propagation path. S4: For each candidate action in the asymmetric scheduling action space, the counterfactual water level trajectory is inferred by the preset structural causal model, the area under the intervention effect curve of the actual water level trajectory and the counterfactual trajectory is calculated, and the control instruction for selecting candidate actions is generated. S5: Monitor the rate of change in water depth and the amount of reduction in fluctuation time difference after the execution of the monitoring and control command. If the amount of reduction in time difference is lower than the convergence threshold, extend the search area based on the origin point by one monitoring node in the next cycle, and adjust the fluctuation time difference convergence standard according to the ratio of the measured rate to the target rate.

[0006] Furthermore, the instantaneous fluctuation difference between adjacent nodes is calculated to construct the water level fluctuation distribution field of the pipeline network. The specific process is as follows: For adjacent nodes directly connected by pipelines, as node pairs, the water level sequences of the two nodes within multiple historical backflow event windows are extracted, the connection coefficient between the connected points is calculated, and the offset of the peak value of the connection coefficient is selected as the characteristic time delay of water level fluctuations propagating from upstream to downstream. The median of the characteristic time delays calculated for multiple event windows of the same node pair is taken to obtain the hydraulic propagation time delay parameter. If the hydraulic propagation time delay parameter is positive, it indicates that the water level fluctuation of the original node precedes that of the propagation node. The calculated time delay parameter and propagation direction are filled into the propagation time delay matrix to obtain the characteristic propagation time delay matrix. For each node, calculate the instantaneous water level difference and water level change rate difference at the current moment, compare them with the mean and standard deviation of historical normal fluctuations to obtain the normalized fluctuation deviation, extract the instantaneous peak of the water level sequence of each node, calculate the peak difference between adjacent nodes, the positive and negative signs of the peak difference indicate the order of fluctuations, compare them with the direction predicted by the propagation time delay matrix, if they point in opposite directions, they are marked as peak reversal anomalies. The real-time water level, water level change rate, instantaneous peak, and propagation delay, fluctuation deviation, and peak difference between adjacent nodes of each node are integrated into a multi-dimensional feature vector. This vector is then organized into a tensor structure according to nodes and time, generating a three-layer labeled pipeline water level fluctuation distribution field containing normal regions, abnormal fluctuation regions, and abnormal peak regions.

[0007] Furthermore, based on the distribution field of water level fluctuations in the pipeline network, the integral of the rate of change of water level at each node along the propagation path and the cumulative amount of propagation time delay are calculated. The specific process is as follows: From the distribution field of water level fluctuations in the pipeline network, the set of nodes covered by the fluctuation anomaly area and the peak anomaly area is extracted. Starting from each anomaly node, the search proceeds downstream along the direction pointed to by the propagation time delay matrix and the instantaneous phase difference. The path where the water level change rate of the current node is lower than the preset threshold is selected to obtain the candidate propagation path. Each path is recorded as a node sequence. The node where the water level fluctuates first is upstream, and the node where the water level fluctuates later is downstream. The direction of congestion propagation always points from the upstream root cause node to the downstream. For each node on the candidate path, extract its water level change rate time series from the fluctuation distribution field. The water level change rate integral is defined as the total water level rise of the node from the start of the fluctuation anomaly to the current time. The water level change rate integral reflects the degree of water level rise accumulated by the node in the abnormal event. The cumulative propagation delay is defined as the total time taken for the propagation from the starting point of the path to the node.

[0008] Furthermore, the terminal node of the target propagation path is marked as the root cause node, and the node sequence on the target propagation path is used as the congestion propagation path scheme. The specific process is as follows: The integral of the water level change rate at the terminal node of the propagation path is combined with the total propagation time delay to calculate the comprehensive score of each candidate path; The propagation path that minimizes the integral of the water level change rate and has a propagation delay accumulation below a preset threshold is selected as the target propagation path. The terminal node of the target propagation path is marked as the root cause node. The criteria for judgment are: the node has the smallest integral of water level change rate on the path, and the node shows phase lead relative to the downstream node in the peak difference matrix. When there are multiple candidate nodes at the end of the path, the node with the lowest water level change rate and the most significant phase lead feature is selected as the root cause node. The node sequence on the target propagation path is output as the congestion propagation path scheme, with the path direction pointing backward from the root cause node to the fluctuating abnormal area.

[0009] Furthermore, based on the congestion propagation path scheme, the risk of water level exceeding the limit at the root cause node is converted into backflow risk entropy. The specific process is as follows: Based on the root cause node and its water level change rate integral, current water level, historical peak water level, and the historical frequency of backflow at the node, the instantaneous risk coefficient of the root cause node is calculated. The instantaneous risk coefficient is then propagated upstream and downstream along the congestion propagation path to obtain the backflow risk entropy. The backflow risk entropy of each node is normalized so that its value range is mapped to the interval [0,1]. The closer the entropy value is to 1, the closer the node is to the critical state of backflow. The backflow risk entropy of all nodes on the propagation path is aggregated according to spatial distribution to generate a path risk entropy distribution curve. The peak position indicates the currently severely congested section.

[0010] Furthermore, an asymmetric scheduling action space is generated, centered on the root cause node and radiating outward along the propagation path. The specific process is as follows: The current operating status of each pumping station along the congestion propagation path is obtained from the drainage system operation database. The upward and downward adjustment margins of the pumping stations are calculated. The water level of each pipe segment and the full pipe water level are extracted from the position fluctuation distribution field to obtain the downstream pipe network spare capacity. Calculate the propagation time delay of drainage from the upstream pumping station to the downstream pipe network, and align the adjustable capacity of the upstream pumping station with the spare capacity of the downstream pipe network in time sequence to obtain the peak-shaving matching degree. The peak-shaving matching degree reflects the effectiveness of the scheduling efficiency. Scheduling combinations with matching scores exceeding a threshold are entered into the candidate action space. Centered on the root cause node, scheduling actions are generated outward along the propagation path level by level. Each level of action includes: target pump station, action direction, and feasibility and safety verification. Actions that pass the verification are sorted by matching score and output. When the backflow risk entropy of the root cause node increases for two consecutive cycles, the action space is automatically expanded by one hop upstream and downstream.

[0011] Furthermore, the area under the intervention effect curves of the actual water level trajectory and the counterfactual trajectory is calculated to generate candidate control commands. The specific process is as follows: The pre-built structural causal model, based on the current time, fixes all exogenous variables except for the action for each generated candidate action, assumes that the action has not been applied, constructs a counterfactual inference basis, takes the pump station flow adjustment corresponding to the candidate action as the intervention value, and generates the counterfactual water level trajectory of each node through back propagation of the structural causal model. For each candidate action, calculate its intervention effect curve. The curve represents the change of the positive water level reduction effect brought about by the action over time. Sort all candidate actions in descending order and select the action with the largest area as the optimal scheduling decision.

[0012] Furthermore, the rate of change in water depth and the reduction in fluctuation time difference after the execution of monitoring and control commands are specifically analyzed as follows: After executing the control command, continuously collect the water depth change rate of each node within the influence range and the instantaneous phase difference between the root cause node and the downstream adjacent node, calculate the phase difference reduction, and the difference in water depth change rate before and after control, and compare it with the preset convergence threshold. If it is lower than the convergence threshold, it is determined that the control effect is insufficient and adjustment is required. When it is determined that the control effect is insufficient, the congestion origin point search area for the next scheduling cycle will be expanded outward from the current root cause node. The expansion rule is to start from the current root cause node and extend one monitoring node upstream and one downstream along the congestion propagation path. Calculate the ratio of the measured average water depth change rate to the target rate, and use it as the convergence threshold correction coefficient to update the convergence threshold.

[0013] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: This simulation optimization method for urban backflow prevention and drainage scheduling systems achieves accurate identification of congestion propagation direction and rapid location of root cause nodes through the coordinated application of water level fluctuation phase tracking, hydraulic propagation time delay matrix modeling, and structural causal counterfactual inference. This method utilizes the instantaneous phase difference between adjacent monitoring points instead of the traditional comparison of absolute water level values, effectively distinguishing between two fundamentally different water level rise patterns: normal upstream inflow and downstream backflow. It fundamentally solves the problems of misjudgment and response lag caused by the traditional threshold method neglecting water flow direction. By quantitatively pre-evaluating the causal effectiveness of candidate scheduling actions through the area under the intervention effect curve, scheduling decisions are transformed from passively responding to water level exceedances to actively intervening in congestion sources. Furthermore, the entire process does not rely on a historical case database and can address previously unseen but potentially problematic congestion scenarios. Extreme operating conditions are simulated virtually and response strategies are generated in advance to meet the real-time requirements of urban drainage systems. The root cause node location mechanism based on phase difference decreasing backtracking, the risk quantification mechanism based on backflow risk entropy, and the peak-shaving coordination mechanism based on pump station and pipe network capacity time sequence matching are deeply integrated in three layers. An asymmetric scheduling action space centered on the root cause node and exponentially decaying along the congestion propagation path is constructed. By aligning the adjustable drainage capacity of upstream pump stations with the real-time spare capacity of downstream pipe networks with propagation time delay, a peak-shaving time sequence scheme of upstream first discharge and downstream acceptance is automatically generated. It can identify the hidden congestion transmission relationship between upstream and downstream that is difficult to find in traditional hydraulic models, and accurately deliver limited scheduling resources to the pump station nodes that are actually effective, thereby significantly improving the accuracy and coordination efficiency of backflow prevention scheduling. Attached Figure Description

[0014] Figure 1 A schematic diagram of the overall structure of the method steps of the present invention is shown; Figure 2 A schematic diagram of the calculation flow structure of the method data of the present invention is shown; Figure 3 A schematic diagram of the overall structure of the execution flow of step five of the method of the present invention is shown. Detailed Implementation

[0015] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0016] Example 1:

[0017] like Figure 1 As shown, a simulation optimization method for an urban backflow prevention drainage scheduling system includes the following steps: S1: Obtain real-time water level readings and historical backflow event timestamps in the scheduling pipeline network, extract hydraulic propagation time delay parameters, calculate the instantaneous fluctuation difference between adjacent nodes, and construct the pipeline network water level fluctuation distribution field; S2: Based on the distribution field of water level fluctuations in the pipeline network, calculate the integral of the water level change rate and the cumulative amount of propagation time delay at each node on the propagation path, screen the target propagation path, mark the terminal node of the target propagation path as the root cause node, and use the node sequence on the target propagation path as the congestion propagation path scheme. S3: Based on the congestion propagation path scheme, the risk of water level exceeding the limit at the root cause node is converted into the backflow risk entropy. The adjustable capacity of each pumping station is matched with the spare capacity of the downstream pipe network to generate an asymmetric scheduling action space centered on the root cause node and along the propagation path. S4: For each candidate action in the asymmetric scheduling action space, the counterfactual water level trajectory is inferred by the preset structural causal model, the area under the intervention effect curve of the actual water level trajectory and the counterfactual trajectory is calculated, and the control instruction for selecting candidate actions is generated. S5: Monitor the rate of change in water depth and the amount of reduction in fluctuation time difference after the execution of the monitoring and control command. If the amount of reduction in time difference is lower than the convergence threshold, extend the search area based on the origin point by one monitoring node in the next cycle, and adjust the fluctuation time difference convergence standard according to the ratio of the measured rate to the target rate.

[0018] In this scheme, water level sensors are deployed at key nodes of the drainage network, such as rainwater wells, inspection wells, pump station forebays, and river confluences. Real-time water level readings of each node are collected at a fixed sampling frequency of once every 10 seconds. The clocks of all sensors are synchronized with a unified time source to ensure that the time alignment accuracy of cross-node water level data is at the millisecond level. The collected raw data includes node identifiers, timestamps, water level heights, and data quality markers, and is stored in a time series table in a real-time database.

[0019] Backflow event records are retrieved from the historical operation logs of the drainage system. These backflow events include manhole cover overhang, manhole overflow, pump station forebay overflow, and river backflow. The system first automatically marks candidate events through water level change detection, and then manually reviews and confirms them. For each backflow event record, the start and end timestamps of the event and a list of nodes within the event's impact range are extracted. Using the event occurrence time as the center, water level data for a set time period before and after the event is extracted as an analysis window. Each node within the window is marked as having experienced backflow (1 for occurrence, 0 for absence), thus constructing a backflow event annotation dataset. For severe backflow events where the water level exceeds the manhole opening, x hours before and after the event are used; for minor backflow events, a set time period of hours before and after the event is used. The time period and x are set time ranges.

[0020] For adjacent node pairs that have direct pipe connections Extract water level sequences of two nodes within multiple historical backflow event windows. and To eliminate data interference during rainfall, the system accesses meteorological radar rainfall forecast data. When rainfall intensity exceeds the standard, the period is automatically marked as a rainfall interference period, and water level data during this period is not included in the time-lag calibration. The connection coefficient between the two sequences is then calculated. : Take the deviation value at which the connection coefficients of the two sequences reach their maximum values. The characteristic time delay is used to represent the propagation of water level fluctuations from node i to node j. The median of the characteristic time delays calculated for multiple event windows of the same node pair is used to obtain the hydraulic propagation time delay parameters. ,like A positive value indicates that the water level fluctuation at node i precedes that at node j, and the water flow direction is i→j; a negative value indicates the opposite direction. The calculated time delay parameters and propagation direction are then filled into the propagation time delay matrix T, and the matrix elements... ; For the real-time collected water level data, a sliding window median filter is used to eliminate sensor noise and instantaneous disturbances. After filtering, the instantaneous water level difference at the current moment is calculated for each pair of adjacent nodes. and the difference in water level change rate The instantaneous fluctuation difference is compared with the average of historical normal fluctuations. and standard deviation Compare and combine the denominator constants Calculate the normalized fluctuation deviation : when When this happens, mark the node as having abnormal fluctuations.

[0021] Water level time series for each node A sliding window peak detection algorithm is used to extract instantaneous peaks. The detection rule is: when the current water level is greater than the water level values ​​of the three sampling points before and after it, and this value exceeds 1.2 times the average water level within the window, it is marked as a peak point. The occurrence time of each peak is recorded. and peak water level Where k is the peak number, for adjacent node pairs (i, j), calculate the time difference of the occurrence of the same peak event at node i and node j: The sign of the crest difference indicates the order of wave movement; a positive value means the crest at node i appears before that at node j, and a negative value means the crest at node j appears before that at node i. The calculated crest difference... Compare the propagation direction with the direction predicted by the propagation delay matrix T: if the expected peak difference should be positive when the propagation direction is i→j, but the actual calculated peak difference is negative, or vice versa, then it is marked as a peak reversal anomaly. A peak reversal anomaly indicates the presence of a reverse pressure wave or downstream congestion causing a reversal of the propagation direction.

[0022] When constructing the distribution field of water level fluctuations in the pipeline network, the real-time water level, water level change rate, instantaneous peak time, propagation delay, normalized fluctuation deviation, and peak difference between adjacent nodes of each node are integrated into a multi-dimensional feature vector. This vector is organized into a tensor structure by node and time. The Kriging interpolation method is used to extend the water level data of discrete nodes into a continuous water level distribution surface. During the interpolation process, the pipeline topological distance is used instead of Euclidean distance as a spatial correlation measure to ensure that the interpolation results conform to the actual hydraulic connections of the pipeline network. The water level distribution surface, the normalized fluctuation deviation distribution field, and the peak reversal anomaly distribution field are superimposed to generate a three-layer labeled pipeline network water level fluctuation distribution field containing normal regions, fluctuation anomaly regions, and peak anomaly regions. The specific layering rules are as follows: Normal range: fluctuation deviation less than 2 and no abnormal peak reversal; Anomaly zone: fluctuation deviation is greater than 2, but the direction of wave crest propagation is consistent with the predicted direction of time delay; Peak anomaly region: The peak reversal anomaly is marked as true, or the peak propagation direction is opposite to the predicted time delay direction.

[0023] In the distribution field of water level fluctuations in the pipeline network, the set of nodes covered by the fluctuation anomaly region and the phase reversal region is extracted. A bidirectional search strategy is adopted, searching both downstream from the fluctuation anomaly node and upstream from the phase reversal node. The union of the two search paths is taken, and the termination condition is: the water level change rate of the current node is lower than a preset threshold. During the search process, the number of visits to intermediate nodes of each path is recorded. If a node is shared by more than 3 paths, it is marked as a multipath convergence node. Each candidate path is recorded as a node sequence. ,in As the starting point of the propagation path, End point of the propagation path Adjacent nodes and There are direct pipe connections between them and the propagation direction is consistent. For each node on the candidate path P... Extract the time series of water level change rate from the fluctuation distribution field, and integrate the water level change rate. Defined as the node from the start time of the fluctuation anomaly up to the current moment Total rise in water level: The cumulative propagation delay is defined as the amount of time from the starting point of the path. propagation to nodes Cumulative time : Calculate the overall score for each candidate path. Integral of water level change rate at the end node of the comprehensive path Total propagation time delay along the path And the degree of dispersion of the integral of the rate of change of water level along the path: The path with the lowest overall score—meaning the smallest water level rise, shortest time delay, and smallest variance—was selected as the target propagation path. For nodes water level height For time integration, For nodes arrive Hydraulic propagation time delay, The rate of change of water level height over time; The terminal node of the target propagation path is marked as the root cause node. The criteria for determination are: the integral of the water level change rate of this node is the smallest on the path, and the node shows a peak leading its downstream node in the phase difference matrix. The node sequence on the target propagation path is output as the congestion propagation path scheme. The path direction is from the root cause node in the opposite direction to the abnormal fluctuation area. If multiple congestion propagation paths converge at the same root cause node, they are weighted and merged according to the comprehensive score of each path. The path with the higher score has the dominant weight in the control. The path scheme is output once every two water level fluctuation distribution field update cycles. When the position of the root cause node drifts more than a preset distance for two consecutive cycles, a multi-source congestion alarm is triggered.

[0024] Obtain the coordinates of the root cause nodes and their integrals of the rate of change of water level. Current water level Historical high water level and the frequency of backflow history Calculate the instantaneous risk coefficient of the root cause node. : The risk coefficient is propagated along the propagation path, and the formula for calculating the backflow risk entropy of node i is: ; in, The instantaneous risk coefficient of the root cause node at time t. Let i be the propagation delay. The total propagation time delay along the entire propagation path. Let be the topological distance from node ii to the root cause node. Let i be the backflow risk entropy. The backflow risk entropy of each node is normalized. The maximum entropy value of the node in the most recent hour is taken as the upper limit and the minimum entropy value is taken as the lower limit. The closer the entropy value is to 1, the closer the node is to the backflow critical state. The path risk entropy distribution curve is displayed in the form of a line graph, and the peak point is marked with a red circle.

[0025] The process of extracting the adjustable capacity of pumping stations and the downstream spare capacity involves extracting the current flow rate of each pumping station along the propagation path. Rated maximum flow rate Minimum stable flow rate Adjustable capacity of the pumping station Defined as: ; If the pumping station is equipped with multiple pumps, some of which are in standby mode, the adjustable capacity should include the starting capacity of the standby pumps, and the downstream pipe section water level should be extracted from the water level fluctuation distribution field. and full pipe water level Downstream pipeline spare capacity Defined as: Propagation time lag of drainage from upstream pumping stations to downstream pipe networks Off-peak matching degree Defined as: in, Let be the propagation delay from node i to node j. The cross-sectional area of ​​the pipe. For the propagation time delay from the pumping station to the downstream, For scheduling time windows, The time delay decay coefficient is used. Combinations with a matching degree higher than 0.5 enter the candidate action space, while combinations with a matching degree lower than 0.3 do not appear in the candidate action space but will be recorded in the non-peak-shifting log. Starting from the root cause node as the radiation center, scheduling actions are generated level by level along the congestion propagation path. Each level of action includes the target pumping station and the direction of action, which can be to increase drainage, decrease drainage, or maintain flow. Simultaneously, a feasibility check is performed: verifying whether the pumping station flow rate is within acceptable limits after the action. Within the scope, whether upstream and downstream actions conflict is determined. An increase in drainage upstream and a decrease in drainage downstream are considered conflicts. Actions that fail the verification are removed from the action space, and the output is marked as filtered with the reason for filtering recorded. The final output asymmetric scheduling action space includes the root cause node coordinates, the propagation path node sequence, the backflow risk entropy of each node, the adjustable capacity and matching degree of each pumping station, and the list of verified actions.

[0026] The pre-built structural causal model is being loaded. This model describes the causal relationships between nodes of the drainage network in the form of a directed acyclic graph, with causal strength coefficients... The model is calibrated using structural equation modeling based on historical data. For the current moment, candidate actions are used as intervention variables, and the actual water level trajectories at each monitoring point are used as the basis for the model. To construct a counterfactual basis for the observed variable: fix the values ​​of all exogenous variables other than the intervention variable, and only change the state of the intervention variable; For candidate action a, adjust the corresponding pump station flow rate. As an intervention value, the counterfactual water level trajectory under the assumption of no action is inferred through a structural causal model. For node j directly affected by the action, its counterfactual water level is... in The action influence coefficient (calibrated through pump station speed regulation test). Let h be the node response time constant. For upstream node i, its counterfactual water level propagates forward through the causal chain: Among them, the causal propagation attenuation factor .

[0027] For each candidate action a, calculate its intervention effect curve. This curve characterizes the change in the positive water level reduction effect of the action over time, and the area under the intervention effect curve (AUC) is defined as: in, For candidate action a at time t, the counterfactual water level trajectory is... Let be the area under the intervention effect curve for candidate action 'a'. To simulate the time window, The starting time for executing the scheduling decision; When there are multiple monitoring nodes within the range of an action's influence, the weighted sum of the AUCs of each node is taken, with the weight determined by the reciprocal of the distance from the node to the root cause node. All candidate actions are sorted from high to low, and the action with the largest area is selected as the optimal scheduling decision.

[0028] When the AUC values ​​of individual actions are all below 0.5, the top 3 actions with the highest matching degree are combined, and the area under the curve of the combined intervention effect is calculated repeatedly for each combination. The order of actions within the combination is arranged from smallest to largest propagation lag. The formula for calculating the AUC of two action combinations is as follows: ; in, There are two candidate actions. The action interaction items are fitted from historical data. Conflicting combinations that act on the same pumping station or pumping stations directly connected upstream and downstream and whose action directions are opposite are directly eliminated. The combination with the largest AUC is selected as the joint control command output, with suggestions for the start-up sequence of each action. Upstream actions start first, and downstream actions start with a delay. The delay is equal to the hydraulic propagation time delay.

[0029] After executing the generated optimal control command, data from monitoring nodes within the control influence range are continuously collected. The peak difference between the root cause node and its downstream adjacent nodes is recalculated. The reduction in peak time difference is defined as: the reduction in peak time difference between the root cause node and its downstream adjacent nodes. Defined as: in To regulate the peak difference between the precursor node and the downstream node, The peak difference after adjustment.

[0030] Convergence threshold The initial value was determined by analyzing the phase difference fluctuations during historical normal periods. The standard deviation of the phase difference during a continuous 7-day period without abnormal events was taken. If three consecutive sampling periods meet the condition... If the control is deemed effective, the current scheduling strategy is maintained, and a successful case of control is recorded; if If the root cause node water level change rate remains positive for three consecutive sampling cycles, the control effect is deemed insufficient, triggering parameter adjustment.

[0031] When the control effect is deemed insufficient, the congestion origin search area for the next scheduling cycle is expanded outward from the current root cause node. The expansion rule is as follows: starting from the current root cause node, extend one monitoring node upstream and one downstream along the congestion propagation path. If the current root cause node is already located at the end of the path, the expansion only extends in the extendable direction. The first expansion extends to one hop upstream and downstream, the second expansion extends to two hops upstream and downstream, with a maximum expansion step size of 5 hops. The expanded search area includes the original root cause node and one hop upstream and downstream, totaling 3 candidate nodes. The propagation path is re-enumerated in the S2 path tracing of the next cycle.

[0032] The target rate is set as the average rate required for the root cause node water level to drop from its current value to the warning level. Expected recovery time The response time will be dynamically adjusted according to the flood control emergency response level: Level IV response will be 60 minutes, Level III will be 45 minutes, Level II will be 30 minutes, and Level I will be 15 minutes. The measured rate will be the average water level change rate of the most recent 5 sampling periods after the adjustment. Convergence threshold correction coefficient The calculation is as follows: The updated convergence threshold is ; After the dispatching command is executed, an effectiveness evaluation report for this regulation is generated based on indicators such as the improvement in water level change rate, the reduction in phase difference, the magnitude of water level drop at the root cause node, and whether extended adjustment is triggered. The effectiveness is divided into three levels: effective (no extension triggered, phase difference reduction ≥ threshold); partially effective (reaches the target after triggering one extension); and ineffective (still fails to reach the target after triggering extension or is ineffective for two consecutive cycles). The actual data during the regulation process is stored as new samples in the causal model training pool. The causal strength coefficient and action influence coefficient of the structural causal model are recalibrated weekly. For root cause nodes that are determined to be ineffective in three consecutive regulation cycles, they are automatically marked as structural bottleneck points.

[0033] The size of the interval and threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value.

[0034] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation. In the two embodiments provided in this application, it should be understood that the disclosed apparatus and system can be implemented in other ways; for example, the apparatus embodiments described above are merely illustrative, and the division of modules is merely a logical functional division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed; furthermore, the coupling or direct coupling or communication connection between the shown or discussed mutuals can be through some interfaces, and the indirect coupling or communication connection between the apparatus or modules can be electrical, mechanical or other forms. The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A simulation optimization method for urban anti-inflow drainage scheduling system, characterized in that, Includes the following steps: S1: Obtain real-time water level readings and historical backflow event timestamps in the scheduling pipeline network, extract hydraulic propagation time delay parameters, calculate the instantaneous fluctuation difference between adjacent nodes, and construct the pipeline network water level fluctuation distribution field; S2: Based on the distribution field of water level fluctuations in the pipeline network, calculate the integral of the water level change rate and the cumulative amount of propagation time delay at each node on the propagation path, screen the target propagation path, mark the terminal node of the target propagation path as the root cause node, and use the node sequence on the target propagation path as the congestion propagation path scheme. S3: Based on the congestion propagation path scheme, the risk of water level exceeding the limit at the root cause node is converted into the backflow risk entropy. The adjustable capacity of each pumping station is matched with the spare capacity of the downstream pipe network to generate an asymmetric scheduling action space centered on the root cause node and along the propagation path. S4: For each candidate action in the asymmetric scheduling action space, the counterfactual water level trajectory is inferred by the preset structural causal model, the area under the intervention effect curve of the actual water level trajectory and the counterfactual trajectory is calculated, and the control instruction for selecting candidate actions is generated. S5: Monitor the rate of change in water depth and the amount of reduction in fluctuation time difference after the execution of the monitoring and control command. If the amount of reduction in time difference is lower than the convergence threshold, the search area of ​​the next cycle will be extended by one monitoring node according to the origin point, and the fluctuation time difference convergence standard will be adjusted according to the ratio of the measured rate to the target rate.

2. The simulation optimization method of urban anti-inversion drainage scheduling system according to claim 1, characterized in that, The instantaneous fluctuation difference between adjacent nodes is calculated to construct the water level fluctuation distribution field of the pipe network. The specific process is as follows: For adjacent nodes directly connected by pipelines, as node pairs, the water level sequences of the two nodes within multiple historical backflow event windows are extracted, the connection coefficient between the connected points is calculated, and the offset of the peak value of the connection coefficient is selected as the characteristic time delay of water level fluctuations propagating from upstream to downstream. The median of the characteristic time delays calculated for multiple event windows of the same node pair is taken to obtain the hydraulic propagation time delay parameter. If the hydraulic propagation time delay parameter is positive, it indicates that the water level fluctuation of the original node precedes that of the propagation node. The calculated time delay parameter and propagation direction are filled into the propagation time delay matrix to obtain the characteristic propagation time delay matrix. For each node, calculate the instantaneous water level difference and water level change rate difference at the current moment, compare them with the mean and standard deviation of historical normal fluctuations to obtain the normalized fluctuation deviation, extract the instantaneous peak of the water level sequence of each node, calculate the peak difference between adjacent nodes, the positive and negative signs of the peak difference indicate the order of fluctuations, compare them with the direction predicted by the propagation time delay matrix, if they point in opposite directions, they are marked as peak reversal anomalies. The real-time water level, water level change rate, instantaneous peak, and propagation delay, fluctuation deviation, and peak difference between adjacent nodes of each node are integrated into a multi-dimensional feature vector. This vector is then organized into a tensor structure according to nodes and time, generating a three-layer labeled pipeline water level fluctuation distribution field containing normal regions, abnormal fluctuation regions, and abnormal peak regions.

3. The simulation optimization method for an urban backflow prevention drainage scheduling system according to claim 1, characterized in that, Based on the distribution field of water level fluctuations in the pipeline network, the integral of the rate of change of water level at each node along the propagation path and the cumulative propagation time delay are calculated. The specific process is as follows: From the distribution field of water level fluctuations in the pipeline network, the set of nodes covered by the fluctuation anomaly area and the peak anomaly area is extracted. Starting from each anomaly node, the search proceeds downstream along the direction pointed to by the propagation time delay matrix and the instantaneous phase difference. The path where the water level change rate of the current node is lower than the preset threshold is selected to obtain the candidate propagation path. Each path is recorded as a node sequence. The node where the water level fluctuates first is upstream, and the node where the water level fluctuates later is downstream. The direction of congestion propagation always points from the upstream root cause node to the downstream. For each node on the candidate path, extract its water level change rate time series from the fluctuation distribution field. The water level change rate integral is defined as the total water level rise of the node from the start of the fluctuation anomaly to the current time. The water level change rate integral reflects the degree of water level rise accumulated by the node in the abnormal event. The cumulative propagation delay is defined as the total time taken for the propagation from the starting point of the path to the node.

4. The simulation optimization method of urban anti-inversion drainage scheduling system according to claim 1, characterized in that, The terminal node of the target propagation path is marked as the root cause node, and the node sequence on the target propagation path is used as the congestion propagation path scheme. The specific process is as follows: The integral of the water level change rate at the terminal node of the propagation path is combined with the total propagation time delay to calculate the comprehensive score of each candidate path; The propagation path that minimizes the integral of the water level change rate and has a propagation delay accumulation below a preset threshold is selected as the target propagation path. The terminal node of the target propagation path is marked as the root cause node. The criteria for judgment are: the node has the smallest integral of water level change rate on the path, and the node shows phase lead relative to the downstream node in the peak difference matrix. When there are multiple candidate nodes at the end of the path, the node with the lowest water level change rate and the most significant phase lead feature is selected as the root cause node. The node sequence on the target propagation path is output as the congestion propagation path scheme, with the path direction pointing backward from the root cause node to the fluctuating abnormal area.

5. The simulation optimization method of urban anti-inversion drainage scheduling system according to claim 1, characterized in that, According to the congestion propagation path scheme, the risk of water level exceeding the limit at the root cause node is converted into backflow risk entropy. The specific process is as follows: Based on the root cause node and its water level change rate integral, current water level, historical peak water level, and the historical frequency of backflow at the node, the instantaneous risk coefficient of the root cause node is calculated. The instantaneous risk coefficient is then propagated upstream and downstream along the congestion propagation path to obtain the backflow risk entropy. The backflow risk entropy of each node is normalized so that its value range is mapped to the interval [0,1]. The closer the entropy value is to 1, the closer the node is to the critical state of backflow. The backflow risk entropy of all nodes on the propagation path is aggregated according to spatial distribution to generate a path risk entropy distribution curve. The peak position indicates the currently severely congested section.

6. The simulation optimization method of urban anti-inversion drainage scheduling system according to claim 1, characterized in that, The asymmetric scheduling action space is generated, centered on the root cause node and radiating outward along the propagation path. The specific process is as follows: The current operating status of each pumping station along the congestion propagation path is obtained from the drainage system operation database. The upward and downward adjustment margins of the pumping stations are calculated. The water level of each pipe segment and the full pipe water level are extracted from the position fluctuation distribution field to obtain the downstream pipe network spare capacity. Calculate the propagation time delay of drainage from the upstream pumping station to the downstream pipe network, and align the adjustable capacity of the upstream pumping station with the spare capacity of the downstream pipe network in time sequence to obtain the peak-shaving matching degree. The peak-shaving matching degree reflects the effectiveness of the scheduling efficiency. Scheduling combinations with matching scores exceeding a threshold are entered into the candidate action space. Centered on the root cause node, scheduling actions are generated outward along the propagation path level by level. Each level of action includes: target pump station, action direction, and feasibility and safety verification. Actions that pass the verification are sorted by matching score and output. When the backflow risk entropy of the root cause node increases for two consecutive cycles, the action space is automatically expanded by one hop upstream and downstream.

7. The simulation optimization method of urban anti-inversion drainage scheduling system according to claim 1, characterized in that, The area under the intervention effect curves of the actual water level trajectory and the counterfactual trajectory is calculated to generate candidate control commands. The specific process is as follows: The pre-built structural causal model, based on the current time, fixes all exogenous variables except for the action for each generated candidate action, assumes that the action has not been applied, constructs a counterfactual inference basis, takes the pump station flow adjustment corresponding to the candidate action as the intervention value, and generates the counterfactual water level trajectory of each node through back propagation of the structural causal model. For each candidate action, calculate its intervention effect curve. The curve represents the change of the positive water level reduction effect brought about by the action over time. Sort all candidate actions in descending order and select the action with the largest area as the optimal scheduling decision.

8. The simulation optimization method of urban anti-inversion drainage scheduling system according to claim 1, characterized in that, The specific process for monitoring the rate of water depth change and the reduction in fluctuation time difference after the execution of monitoring and control commands is as follows: After executing the control command, continuously collect the water depth change rate of each node within the influence range and the instantaneous phase difference between the root cause node and the downstream adjacent node, calculate the phase difference reduction, and the difference in water depth change rate before and after control, and compare it with the preset convergence threshold. If it is lower than the convergence threshold, it is determined that the control effect is insufficient and adjustment is required. When it is determined that the control effect is insufficient, the congestion origin point search area for the next scheduling cycle will be expanded outward from the current root cause node. The expansion rule is to start from the current root cause node and extend one monitoring node upstream and one downstream along the congestion propagation path. Calculate the ratio of the measured average water depth change rate to the target rate, and use it as the convergence threshold correction coefficient to update the convergence threshold.

Images