Farmland drainage ditch network pre-discharge scheduling method, device and medium fusing short-term prediction

Abstract

This invention discloses a pre-drainage scheduling method for farmland drainage canal networks that integrates short-term forecasts. It dynamically assesses the reliability of each forecast data source and monitoring station based on local monitoring data, and uses interpolation and reconstruction algorithms to recover complete data when monitoring data is missing. This allows the system to automatically reduce the weight of data sources with large errors and strong fluctuations in the event of forecast conflicts or sensor anomalies, enhancing the reliability of overall information fusion. Furthermore, it constructs multiple rainfall scenarios with different spatiotemporal distributions and occurrence probabilities through reliability weighting, and combines these with a digital twin model of the farmland drainage canal network to obtain the probability distribution of key control nodes. This upgrades the system from a single deterministic operating condition to a multi-scenario probability distribution, enabling pre-drainage scheduling to no longer rely on single-point predictions but rather on a holistic assessment based on possible future evolution paths.

Description

Technical Field

[0001] This invention relates to the field of water affairs informatization and intelligent scheduling control technology, specifically a method, equipment and medium for pre-discharge scheduling of farmland drainage canal networks that integrates short-term forecasting. Background Technology

[0002] With the development of high spatiotemporal resolution short-term forecasting technology, some irrigation districts or farmland water conservancy management units have begun to try to introduce refined rainfall forecast data for the next few hours into the operation of farmland drainage canal networks. Most existing systems integrate Internet of Things sensor networks to collect rainfall, water level and flow information of canal networks in real time, and conduct pre-drills in conjunction with canal network hydraulic models or digital twin platforms, so as to empty the canal system storage capacity in advance before the arrival of rainstorms, transforming the traditional passive drainage after the event into proactive pre-drainage scheduling.

[0003] However, short-term rainfall forecasts inherently possess uncertainty, posing a series of practical difficulties for pre-discharge scheduling decisions. When forecasting systems provide probabilistic results such as "the probability of heavy rainfall is about 50%", or when there are significant differences in the intensity and spatial distribution of future rainfall from different forecast sources, existing pre-discharge scheduling methods typically rely on a single deterministic rainfall process line for simulation, lacking an inherent mechanism for handling probabilistic and conflicting forecast information. Furthermore, in actual operation, it often happens that forecasts indicate that short-term heavy rainfall is about to begin, but real-time monitoring data from multiple rain gauges and water level stations in the region have not yet shown significant changes, or even that some monitoring points have data loss due to malfunctions. This leads to inconsistencies between forecast information and local rainfall and water conditions, making it difficult for dispatchers to make timely and reliable decisions.

[0004] To address the aforementioned issues, some existing solutions generate a single rainfall sequence as model input by simply weighting and averaging multi-source forecast results, or reduce the risk of erroneous pre-drainage actions by increasing the rainfall and water level thresholds for initiating pre-drainage. While these methods avoid the blindness of relying solely on a single forecast source to some extent, the uncertainty of forecast results is not explicitly quantified and incorporated into the scheduling decision-making process, and the scheduling strategy cannot be dynamically adjusted according to the forecast risk level. Secondly, the reliability differences between multi-source forecasts and between forecasts and monitoring data are not systematically characterized, making it difficult to achieve a reasonable balance between insufficient pre-drainage and excessive discharge. Therefore, during the critical decision-making window of a heavy rainfall event, the existing farmland drainage network pre-drainage scheduling may still result in problems such as insufficient utilization of drainage capacity or unnecessary excessive discharge. Summary of the Invention

[0005] This invention provides a method, equipment, and medium for pre-discharge scheduling of farmland drainage canal networks that integrates short-term forecasts. By constructing multiple rainfall scenarios with different spatiotemporal distributions and occurrence probabilities through credibility weighting, and then combining them with a digital twin model of the farmland drainage canal network to obtain the probability distribution of key control nodes, it achieves an upgrade from a single deterministic operating condition to a multi-scenario probability distribution, thereby solving the problems in the background technology.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows: A method for pre-discharge scheduling of farmland drainage canal networks that integrates short-term forecasts, comprising the following steps executed via computer equipment: S1, acquire short-term rainfall forecast data composed of multiple forecast data sources, as well as real-time monitoring data and historical fluctuation characteristics of local monitoring stations of farmland drainage network; S2, based on short-term precipitation forecast data and real-time monitoring data, the prediction deviation is obtained. Based on the prediction deviation and historical fluctuation characteristics, the dynamic credibility of each forecast data source and local monitoring station is evaluated to obtain the credibility coefficient of each data source. S3, based on the confidence coefficient of each forecast data source and the short-term precipitation forecast data, a probabilistic precipitation input scenario is obtained. The probabilistic precipitation input scenario is a set of precipitation sequences with occurrence probability weights. S4. Input the probabilistic rainfall input scenario into the digital twin model of the farmland drainage network for hydraulic simulation, and obtain the probability distribution of water level, field surface water depth and groundwater level of key control nodes and field monitoring points of the farmland drainage network in a future predetermined time period. S5. Obtain multiple candidate pre-discharge control strategies. Calculate the risk cost of insufficient pre-discharge and the risk cost of excessive pre-discharge for each pre-discharge control strategy based on the probability distribution. Sum the risk costs under each scenario based on the probability weights of the probabilistic rainfall input scenario to obtain the expected comprehensive risk cost of each pre-discharge control strategy. S6 selects the pre-drainage control strategy with the lowest expected comprehensive risk cost from multiple candidate pre-drainage control strategies and generates the corresponding irrigation and drainage scheduling instructions.

[0007] Preferably, in S1, the real-time monitoring data of the local monitoring station includes local real-time rainfall, canal network water level, flow monitoring data, field surface water depth data, and groundwater level monitoring data; Preferably, the S2 division specifically includes the following steps: S21, assign and maintain a dynamically updated confidence coefficient for each independent forecast data source and local monitoring station; S22. In each evaluation period, for each forecast data source, the rainfall forecast made for the farmland drainage network field group in the previous period is compared with the corresponding local monitoring data after quality control to obtain the forecast deviation. S23. Based on the prediction deviation and the historical fluctuation characteristics of the corresponding data source, a new confidence coefficient of the data source is calculated through a preset confidence update model. The confidence coefficient decreases as the prediction deviation increases and is lowered when the historical fluctuation of the data source intensifies.

[0008] Preferably, in S2, when some of the local monitoring stations have missing data; The corresponding local monitoring data that has undergone quality control is obtained through the following steps: merging effective monitoring data from neighboring monitoring stations, historical data patterns from missing monitoring stations, and equipment status records from the missing monitoring stations, and using data interpolation and reconstruction algorithms to generate a complete monitoring dataset covering all monitoring times and indicators.

[0009] Preferably, in S3, the step of obtaining the probabilistic rainfall input scenario based on the confidence coefficient of each forecast data source and the transformation of multi-source short-term rainfall forecast data includes: Using the confidence coefficient of each forecast data source as the weight or sampling probability basis for the forecast results of the forecast data source, random sampling and combination are performed on the precipitation field or ensemble forecast members of each forecast data source within the forecast spatiotemporal range to generate multiple precipitation scenario sequences that are different in spatiotemporal distribution and are each attached with a probability weight of occurrence.

[0010] Meanwhile, the probabilistic rainfall input scenario is derived from the rainfall forecast results of the same forecast period by the ensemble numerical weather forecast members or different short-term forecast algorithms. Furthermore, when generating the probabilistic rainfall input scenario, the initial confidence coefficient and its allowable range of change are pre-set according to the type of each forecast algorithm or forecast member.

[0011] Preferably, in S4, the probability distribution specifically includes: at several predetermined future time points, providing statistics, confidence intervals, and probability density functions for the corresponding water level, field surface water depth, and groundwater level for each of the key control nodes and field monitoring points.

[0012] Preferably, in S5, the risk cost of insufficient pre-drainage includes the loss cost caused by waterlogging, overflow, or poor drainage due to the water level at key control nodes exceeding the control water level threshold or the water depth on the field exceeding the allowable water accumulation depth threshold. The risks and costs of excessive pre-drainage include the adverse effects of insufficient field water storage, insufficient utilization of canal network regulation and storage, or downstream water level constraints caused by premature drainage, as well as the losses resulting from increased energy consumption of pumping stations.

[0013] The specific division of S5 includes the following steps: S51 defines multiple candidate pre-discharge control strategies, each of which contains the sequence of operations for the gates and pumping station control equipment in the farmland drainage network during the future scheduling period. S52, for each candidate pre-drainage control strategy and each probabilistic rainfall input scenario, the farmland drainage network digital twin model is invoked to simulate, obtain the consequences of insufficient pre-drainage and excessive pre-drainage of the candidate pre-drainage control strategy under the probabilistic rainfall input scenario, and quantify the consequences into corresponding risk costs. S53, for each candidate pre-arrangement control strategy, the risk costs obtained under all the probabilistic rainfall input scenarios are weighted and summed according to the probability weights of the corresponding probabilistic rainfall input scenarios to obtain the expected comprehensive risk cost of the candidate pre-arrangement control strategy.

[0014] Preferably, after S6, a feedback optimization step is included, which includes: After the actual rainfall process ends and the irrigation and drainage scheduling instructions are executed, the prediction effect of the probabilistic rainfall input scenario, the simulation accuracy of the digital twin model of the farmland drainage network, and the assessment results of the expected comprehensive risk cost are verified using actual and complete local monitoring data. Based on the verification results, the parameters in the credibility update model in S2, the probability scenario transformation rule in S3, and the risk cost calculation model in S5 are adaptively adjusted.

[0015] Preferably, in S6, the generated irrigation and drainage scheduling command is sent to the edge intelligent control unit deployed in the field control equipment of the farmland drainage network; The irrigation and drainage scheduling instructions include control logic that allows the edge intelligent control unit to fine-tune the operation of the control equipment within a preset rule range based on local high-frequency ultra-real-time monitoring data; During execution, the edge intelligent control unit transmits its local operating status and fine-tuning records back to the central scheduling system for model and parameter updates in the feedback optimization step.

[0016] In another aspect, the present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method described above.

[0017] In another aspect, the present invention also discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method described above.

[0018] As can be seen from the above technical solution compared with the prior art, the present invention has the following beneficial effects: 1. This invention addresses the problem that existing pre-scheduling methods struggle to handle short-term forecast uncertainties and rely solely on a single deterministic rainfall process line. This invention constructs a complete link from dynamic credibility assessment, probabilistic rainfall scenario construction, probability distribution deduction to minimizing the expected comprehensive risk cost. It transforms multi-source short-term rainfall forecast results, which are originally difficult to use directly for scheduling decisions, into quantifiable and comparable risk indicators, thereby improving the existing passive scheduling mode that relies on a single forecast source and empirical thresholds.

[0019] 2. This invention addresses the problem of difficulty in systematically characterizing the reliability differences between multi-source forecasts and between forecasts and local monitoring data. Based on local monitoring data, this invention performs dynamic reliability assessment on each forecast data source and monitoring station, and uses interpolation and reconstruction algorithms to restore complete data when monitoring data is missing. This enables the system to automatically reduce the weight of data sources with large errors and strong fluctuations in the event of forecast conflicts or sensor anomalies, thereby enhancing the reliability of overall information fusion.

[0020] 3. This invention addresses the problem that traditional pre-drainage strategies struggle to balance flood control safety, storage capacity utilization, downstream constraints, and energy consumption. By utilizing a digital twin model of farmland drainage network, this invention quantifies the risks of insufficient and excessive pre-drainage under different pre-drainage control strategies under multiple probabilistic rainfall scenarios. It calculates the expected comprehensive risk cost using scenario probabilities as weights, enabling the scheduling system to explicitly compare multiple pre-drainage strategies and automatically select the strategy with the lowest expected risk, thus avoiding extreme situations of excessive or insufficient drainage.

[0021] 4. By introducing a feedback optimization mechanism and the local high-frequency fine-tuning capability of the field edge intelligent control unit, this invention can continuously revise the credibility model, scenario construction rules, and risk cost parameters using actual monitoring data after an event. At the same time, during execution, it allows field equipment to make fine-grained adjustments based on the latest water and rainfall information. While taking into account the global nature of central decision-making, it improves the real-time performance and robustness of field control, thereby enhancing the overall proactive defense capability and operational stability of the farmland drainage network under complex and uncertain rainfall conditions. Attached Figure Description

[0022] Figure 1 This is a flowchart of the pre-drainage scheduling method for farmland drainage network integrating short-term forecasts in an embodiment of the present invention; Figure 2 This is a flowchart of the dynamic credibility assessment method in an embodiment of the present invention; Figure 3 This is a flowchart illustrating the expected comprehensive risk cost calculation in an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0024] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate the present invention, but should not be used to limit the scope of the present invention.

[0025] Example 1: In this embodiment, in many regions of my country, farmland drainage network often faces practical problems such as the suddenness of short-term rainfall, the variety of forecast data sources, and the susceptibility of monitoring equipment to environmental interference. A short-term heavy rainfall may cause waterlogging of fields and overflow of the canal system if the pre-drainage is not timely, while blindly pre-draining with large flow rates in advance will cause waste of water resources and a surge in pumping station energy consumption.

[0026] To resolve this core contradiction and achieve precise, intelligent, and risk-controllable drainage scheduling, such as Figure 1 As shown, this application proposes an intelligent pre-drainage scheduling method for farmland drainage network that integrates short-term forecasts, including the following steps: S1, acquire multi-source short-term rainfall forecast data and collect local real-time monitoring data of farmland drainage network; local real-time monitoring data shall include at least the monitoring quantities such as rainfall, water level and flow rate; S2, based on local real-time monitoring data, performs dynamic credibility assessment on each forecast data source and each local monitoring station, and assigns a credibility coefficient to each data source to characterize the credibility status of the data source in the current scheduling period.

[0027] It should be noted that in this embodiment, both short-term forecast data sources and local monitoring stations are included in the credibility assessment framework. Although local monitoring data directly reflects the actual physical processes on site, it may still exhibit systematic errors or short-term distortions due to factors such as sensor calibration deviations, equipment aging, communication interruptions, and local interference. If the monitoring data is simply considered absolutely reliable and its credibility is fixed at 1, then abnormal monitoring equipment will amplify the interference to the fusion of multi-source information.

[0028] Therefore, this embodiment maintains dynamically updated reliability coefficients for each forecast data source and each monitoring station, and continuously corrects the reliability of the monitoring data through comprehensive comparison with neighboring monitoring stations, historical statistical results, and quality control processes.

[0029] S2 can be further subdivided into, for example: Figure 2 S21 to S23 are shown below: S21 assigns and maintains a dynamically updated confidence coefficient for each independent forecast data source and local monitoring station. When the system goes online, the initial confidence coefficients of each data source and monitoring station can be set within a preset range based on historical forecast verification results and the stability of the monitoring station network. In subsequent evaluation cycles, the coefficients are automatically updated through S22 and S23 based on the calculation formulas of prediction deviation and historical fluctuation characteristics. S22. In each evaluation period, the rainfall forecast made by each forecast data source for the farmland drainage network field group in the previous period is compared with the corresponding local monitoring data after quality control to obtain the forecast deviation. S23. Based on the prediction deviation obtained in S22 and the historical fluctuation characteristics calculated based on the prediction deviation sequence of multiple evaluation periods, the prediction deviation index and the historical fluctuation index of prediction deviation are substituted into the preset credibility update model as input, and the new credibility coefficient of each data source in the current evaluation period is output. The credibility coefficient decreases as the prediction deviation increases and is lowered when the historical fluctuation of the data source intensifies. The exponential decay of S23 is updated as follows: Within each evaluation period, a prediction bias index is constructed for each data source; for each data source... During the evaluation period The rainfall forecast is calculated based on the root mean square deviation of the forecast rainfall and the observed rainfall after quality control, as shown in equation (1): (1) In equation (1), Indicates data source During the evaluation period The root mean square bias of rainfall forecasts Indicates the evaluation period The total number of discrete moments used for statistics. Indicates the evaluation period Discrete time index within. Indicates data source During the evaluation period The moment Forecasted rainfall for farmland drainage network field clusters, This indicates that the local monitoring system, after quality control, is in the evaluation cycle. The moment The corresponding rainfall observed; To characterize historical fluctuations, in the data source The degree of fluctuation in statistical prediction bias over the most recent multiple assessment periods; first calculate the most recent Average prediction deviation over the assessment periods: , Then calculate the corresponding standard deviation to obtain the historical fluctuation index of the prediction deviation, as shown in equation (2): (2) In Equation 2, Indicates data source In recent The average prediction deviation over each evaluation period Indicates targeting the data source The number of assessment periods used when statistically analyzing historical fluctuation characteristics. Indicates relative to the current evaluation period Historical cycle offset sequence number, Indicates data source During the evaluation period The root mean square bias of rainfall forecasts Indicates data source Historical fluctuations in prediction bias; After obtaining the prediction deviation index and the historical fluctuation index of prediction deviation, the confidence coefficient is updated in S23 using an exponential decay method, and the update relationship is shown in equation (3); for the data source During the evaluation period The new credibility coefficient is calculated as follows: (3) In Equation 3, Indicates data source During the evaluation period The updated credibility coefficient Indicates data source In the previous assessment cycle The credibility coefficient This represents the preset lower limit of the credibility coefficient, used to ensure that the credibility coefficient does not fall below the allowable lower bound. Indicates targeting the data source The prediction bias attenuation coefficient is used to adjust the prediction bias index. The impact on the intensity of credibility decay Indicates targeting the data source The historical volatility attenuation coefficient is used to adjust the historical volatility index for prediction bias. The impact on the intensity of credibility decay Represents an exponential function with the natural constant as its base; It can be seen that Equation (3) constitutes an exponential decay type credibility update model based on the prediction deviation index and the prediction deviation historical fluctuation index. Its input includes the prediction deviation index obtained by S22 and the prediction deviation historical fluctuation index obtained according to Equation (2), as well as the credibility coefficient of the previous evaluation period. Its output is the credibility coefficient of each data source updated in the current evaluation period. Using the updated formula described above, during the evaluation period... Within the prediction bias When the exponential factor increases, The value decreases, thus making relatively Decline occurs when the fluctuation of prediction bias in recent assessment periods intensifies. When the exponential factor increases, Similarly, reducing the value lowers the overall credibility level of the data source; through parameters , and Differentiated configurations can be used to construct reliability update strategies with varying sensitivities to short-term biases and long-term stability for different types of forecast products or monitoring stations. Maintaining the numerical stability of the confidence coefficient across the global range makes it easy to use it directly as a weight or sampling probability in S3; This model transforms conceptual prediction bias and historical fluctuation characteristics into directly calculable quantitative indicators. By constructing the root mean square deviation (RMS) of predicted rainfall and quality control observed rainfall, the model can uniformly characterize the short-term forecast performance of different data sources in each evaluation period. Simultaneously, it calculates the average and standard deviation using the deviation sequences from the most recent evaluation periods, introducing a historical fluctuation index of prediction bias to constrain long-term stability. Subsequently, it updates the reliability based on the exponential decay formula of prediction bias and historical fluctuation, ensuring that the greater the current bias and the more severe the historical fluctuation, the more significant the decay in the reliability of the corresponding data source within the current period, thus forming an adaptive dynamic distinction among multi-source data. By differentiating the decay coefficient and historical window length, it can construct personalized update trajectories for data streams from different sources and with varying stability, such as radar extrapolation, numerical models, and ground station networks. This provides a physically meaningful weighting basis for subsequent probabilistic rainfall scenario construction and pre-discharge risk cost assessment, while maintaining the boundedness and interpretability of the reliability coefficient numerically, facilitating its embedding and operation in actual farmland drainage network scheduling systems. To ensure the feasibility of the aforementioned reliability update process in the engineering system, the evaluation period can be consistent with the update time of the short-term precipitation forecast product. For example, a 5-minute, 10-minute, or 30-minute evaluation period can be selected, and the whole hour and its integer multiples of the local standard time can be used as the boundary. The forecast sequence and the observation sequence are aligned according to a uniform time grid.

[0030] The number of discrete moments within the evaluation period can be understood as the number of valid moments that pass quality control within that period. When the predicted or observed value at a certain moment is identified as missing or of substandard quality, it is not included in the root mean square deviation statistics to ensure that the prediction deviation index reflects the true performance under valid data.

[0031] Furthermore, the forecasted rainfall can be obtained by area-weighted averaging of the gridded rainfall forecast field within the service area of ​​the farmland drainage network, while the observed rainfall can be obtained by weighted averaging of the monitoring values ​​of several rain gauge stations within the service area according to the catchment area or historical correlation. The specific weights are pre-configured by technical personnel based on existing hydrological analysis results during the system deployment phase.

[0032] When the system goes live, the initial confidence coefficients of various short-term forecast data sources and local monitoring stations can be uniformly set between 0.7 and 0.9, with the lower limit set between 0.1 and 0.3 to avoid the extreme situation where confidence drops excessively to near zero due to large single deviations. The attenuation coefficient related to prediction deviations can be selected between 0.01 and 0.2 based on historical test results, and the attenuation coefficient related to historical fluctuations can be selected between 0.1 and 0.2. The historical window length can be determined according to the number of evaluation periods within 1 to 3 days, so that the confidence update can respond to recent deviations without being overly sensitive to short-term sporadic errors. For example, for short-term forecast products with an update time of ten minutes, the historical window can be set to 6 hours, corresponding to 36 evaluation periods; for numerical weather prediction products with an update time of one hour, the historical window can be set to 24 hours, corresponding to 24 evaluation periods.

[0033] The values ​​for the initial confidence coefficient range, the lower confidence limit range, the attenuation coefficient, and the historical window length can be adjusted by combining the short-term forecast verification results of multiple past rainfall events in the target city and the trial operation data of the farmland drainage network. In engineering practice, technicians can fine-tune the above parameters within a given range based on the typical error level of the forecast products in the region and the long-term stability of the monitoring network, thereby achieving a balance between response speed and robustness, rather than arbitrarily specifying values.

[0034] Furthermore, the prediction deviation index can be understood as the average error level in millimeters, typically falling between 0 and 50 millimeters in drainage scenarios. The historical fluctuation index of prediction deviation can be understood as the standard deviation of the deviation in millimeters, typically falling between 0 and 30. The system can set a reasonable upper limit in the parameter configuration interface to prevent the index from increasing explosively due to abnormal data. Optionally, for data sources with good stability, such as ground station networks with long-term stable quality, a smaller attenuation coefficient and a longer historical window can be configured to highlight their long-term reliability; for unstable or experimental data sources, such as newly launched short-term algorithm products, a larger attenuation coefficient and a shorter historical window can be configured to quickly reduce credibility when significant deviations occur. Optionally, when the number of valid discrete moments is zero due to communication interruption or large-scale failure in a certain evaluation period, the reliability coefficient of the data source in that period can be left unupdated, and the value of the previous evaluation period can be directly used and marked in the log as not updated in this period. When the number of available historical evaluation periods is less than the preset historical window length, the historical fluctuation index of prediction deviation can be calculated based only on the currently available historical periods to avoid the problem of not being able to complete the reliability update due to insufficient historical data at the beginning of system startup.

[0035] When some local monitoring stations have missing data, the corresponding local monitoring data, after quality control, is obtained through the following methods: By integrating effective monitoring data from nearby monitoring stations, historical data patterns from missing monitoring stations, and equipment status records from missing monitoring stations, a complete monitoring dataset covering all monitoring times and indicators is generated using data interpolation and reconstruction algorithms.

[0036] S3, based on the reliability coefficient of each forecast data source, converts the multi-source short-term precipitation forecast data into a set of probabilistic precipitation input scenarios with probability weights; in other words, the probabilistic precipitation input scenario can be understood as a set of multiple precipitation process lines based on risk.

[0037] The specific division of S3 includes the following steps: Using the credibility coefficient of each forecast data source as the weight or sampling probability basis for its forecast results, the precipitation field or ensemble forecast members of each forecast data source within the forecast spatiotemporal range are randomly sampled and combined to generate multiple precipitation scenario sequences that are different in spatiotemporal distribution and each has its own probability of occurrence, which serve as probabilistic precipitation input scenarios. The occurrence probability weights of each rainfall scenario sequence can be statistically obtained based on the frequency of scenario selection during the sampling process and the reliability coefficient of the data sources involved in the combination. After all scenarios are generated, all occurrence probability weights are normalized so that the sum of all scenario probability weights equals 1. These scenario probability weights are the probability weights used in S5 to calculate the expected comprehensive risk cost. .

[0038] S4 uses the probabilistic rainfall input scenario as the driving input and inputs it into the digital twin model of the farmland drainage network for hydraulic simulation, so as to obtain the probability distribution of water level, field surface water depth and groundwater level of key control nodes and field monitoring points of the farmland drainage network in a future predetermined time period.

[0039] In S4, the generated probability distribution specifically includes: at several future predetermined time points, providing statistics, confidence intervals, and probability density functions for each key control node, including water level, field surface water depth, and groundwater level.

[0040] S5. For multiple candidate pre-discharge control strategies, evaluate the risk costs of insufficient pre-discharge and excessive pre-discharge for each pre-discharge control strategy based on probability distribution, and calculate the expected comprehensive risk cost of each pre-discharge control strategy according to the probability weight of the probabilistic rainfall input scenario.

[0041] like Figure 3 As shown, the specific division of S5 includes the following steps: S51. Define multiple candidate pre-dispatch control strategies. Each candidate pre-dispatch control strategy includes the operation sequence of control equipment such as gates and pumping stations in the farmland drainage network during the future scheduling period. S52. For each candidate pre-drainage control strategy and each probabilistic rainfall input scenario, call the digital twin model of the farmland drainage network to simulate, obtain the consequences of insufficient pre-drainage and excessive pre-drainage of the candidate pre-drainage control strategy under the probabilistic rainfall input scenario, and quantify the consequences into the corresponding risk costs. S53. For each candidate pre-arrangement control strategy, the risk costs obtained under all probabilistic rainfall input scenarios are weighted and summed according to the probability weights of the corresponding probabilistic rainfall input scenarios to obtain the expected comprehensive risk cost of the candidate pre-arrangement control strategy. In one implementation, the expected comprehensive risk cost calculation process of S53 is quantified as the expected calculation of the risk cost of the candidate pre-arrangement control strategy under various probabilistic rainfall input scenarios.

[0042] Here, combining the results from S52, for each candidate pre-dispatch control strategy and each probabilistic rainfall input scenario, the comprehensive risk cost under that scenario is first constructed, and then weighted and summed according to the scenario probability weights; specifically including: After completing the hydraulic pre-simulation of the digital twin model of the farmland drainage network in S52, the candidate pre-drainage control strategy was selected with the following sequence number: The strategy and probabilistic rainfall input scenario number is In this scenario, the costs of insufficient and excessive pre-planning risks are weighted and combined to form the overall risk cost for that scenario: (4) In Equation 4, The candidate pre-sorting control strategy number is... The strategy in the probabilistic rainfall input scenario number is The overall risk cost under the scenario, This represents the weighting coefficient for the risk of insufficient pre-arrangement. This represents the weighting coefficient for the risk of over-pre-arrangement. The candidate pre-sorting control strategy number is... The strategy in the probabilistic rainfall input scenario number is The risk cost of insufficient pre-arrangement in the scenario The candidate pre-sorting control strategy number is... The strategy in the probabilistic rainfall input scenario number is The excessive pre-arrangement of risk costs in the context of [the scenario] Indicates the sequence number of the candidate pre-sorting control strategy. Indicates the sequence number of the probabilistic rainfall input scenario; After obtaining the comprehensive risk cost under each scenario, the scenario probability weights generated by S3 are used to perform a weighted summation of the comprehensive risk cost of the candidate pre-arrangement control strategy over all probabilistic rainfall input scenarios to obtain the corresponding expected comprehensive risk cost. If a total of [number] were constructed Given a probabilistic rainfall input scenario, the candidate pre-arrangement control strategy number is: The expected overall risk cost of the strategy is expressed as: (5) in, The candidate pre-sorting control strategy number is... The expected overall risk cost of the strategy This represents the total number of probabilistic rainfall input scenarios. The probabilistic rainfall input scenario number is: The probability weights of the scenarios, The candidate pre-sorting control strategy number is... The strategy in the probabilistic rainfall input scenario number is The overall risk cost under the scenario, Indicates the sequence number of the candidate pre-sorting control strategy. Indicates the sequence number of the probabilistic rainfall input scenario; In one implementation, the weighting coefficient and During the system configuration phase, technicians can pre-set the safety water level constraints, downstream water level constraints, and energy consumption control preferences for farmland drainage network field groups. When the region is extremely sensitive to waterlogging and overtopping, a larger pre-drainage insufficiency risk weight coefficient is selected, so that the losses under pre-drainage insufficiency account for a higher proportion of the overall risk cost; scenario probability weighting. Corresponding to the probabilistic rainfall input scenario converted from multi-source short-term forecasts, calculations are performed in the scenario construction stage of S3 based on the sampling frequency or comprehensive reliability of each scenario. After the reliability coefficient is updated, the scenario probability weights are redistributed and normalized according to the latest reliability, so that the calculation of expected comprehensive risk cost reflects changes in forecast reliability in real time. In S6, the central scheduling system performs pre-arrangement control strategies for all candidate scenarios. By comparing and selecting the strategy with the lowest expected comprehensive risk cost, irrigation and drainage scheduling instructions are generated, forming a complete quantitative link from rainfall forecast scenario construction to pre-drainage scheduling decision-making. Specifically, the above implementation method constructs a complete calculation path from single-scenario risk to cross-scenario expected indicators. For each candidate pre-discharge control strategy and each probabilistic rainfall input scenario, the risks of insufficient and excessive pre-discharge are first weighted and synthesized into a comprehensive cost under the same scenario, so that different types of losses can reflect the management's focus through weight adjustment. Then, across all scenario dimensions, the comprehensive cost is linearly summed using the respective scenario probability weights as coefficients, thereby obtaining the expected comprehensive risk indicator for uncertain rainfall scenarios. In this framework, the probability weights of rainfall scenarios are directly derived from the aforementioned multi-source short-term forecasts and credibility assessment results, while the construction of risk costs comes from the detailed characterization of the hydraulic response of farmland drainage network by the digital twin model. The two are coupled in the weighted summation formula, providing a unified benchmark for the scheduling system to make quantitative comparisons among various pre-discharge strategies. By adjusting the loss weights and scenario probability distributions, differentiated risk preference configurations can be constructed for the safety needs and downstream constraints of different regions, so that the final selected strategy achieves a relatively balanced trade-off between disaster prevention safety and resource utilization. Specifically, to calculate the aforementioned expected comprehensive risk index within the engineering system, probabilistic rainfall input scenarios can be sequentially numbered and stored as a finite sequence of scenarios during the scenario construction phase. The total number of scenarios can be set between 10 and 50, depending on the scale of the farmland drainage network and available computing power. When generating scenarios, the system simultaneously assigns a probability weight between 0 and 1 to each scenario and performs a normalization process after scenario generation, ensuring that the arithmetic sum of all probability weights is strictly equal to one. Candidate pre-discharge control strategies can be automatically enumerated and generated by the central scheduling system before the start of each scheduling cycle based on the current equipment status and operational constraints. For example, several finite strategies can be formed based on the gate opening change step size and pump station start-stop combinations, and each strategy is assigned a unique strategy number for reference during risk calculation and scheduling decisions. For instance, in a medium-sized farmland drainage network, the number of scenarios can be set to around 20, and the number of candidate pre-discharge control strategies can be controlled between several dozen and several hundred, ensuring the accuracy of risk characterization while controlling the number of times the digital twin model is called.

[0043] Furthermore, during the system configuration phase, the costs of insufficient and excessive pre-planning risks can be uniformly converted into the same indicator, such as a loss value with a unified dimension or a standardized dimensionless score, making the comprehensive risk cost and the expected comprehensive risk cost comparable in numerical terms; the relevant conversion coefficients can be determined based on the results of historical event analysis and the weight preferences given by the management department.

[0044] Optionally, in a preferred embodiment, the loss weight coefficient can be limited to a closed interval between zero and one, preferably with the sum of the two being approximately close to one. When the area where the farmland drainage network is located is highly sensitive to urban waterlogging safety, the weight of insufficient pre-drainage risk can be configured to be close to 0.6 to 0.8, while the weight of excessive pre-drainage risk can be configured to be close to 0.2 to 0.4. When downstream water level constraints and energy consumption management objectives are more prominent, the weight of excessive pre-drainage risk can be appropriately increased. Similarly, the scenario probability weight can be directly inherited from the sampling frequency in the probabilistic rainfall input scenario generation stage, or the scenario probability can be redistributed according to the latest credibility of each data source after the credibility coefficient is updated, and normalization is used again to ensure that the total probability is one, so that the expected comprehensive risk index can reflect the latest forecast credibility pattern.

[0045] For example, when the probability weight of certain scenarios is much lower than that of other scenarios and the corresponding comprehensive risk cost is close to that of adjacent scenarios, in engineering implementation, these low-probability scenarios can be merged into adjacent scenarios or not simulated separately to reduce the number of calls to the digital twin model and reduce the online computational load without significantly changing the overall probability distribution. Optionally, in abnormal and boundary conditions, if the digital twin simulation of a certain scenario fails to output an effective comprehensive risk cost due to numerical non-convergence or abnormal input data, the system can assign a preset penalty value to the comprehensive risk cost of that scenario, or use the same strategy to replace it with the maximum comprehensive risk cost calculated in other scenarios. At the same time, the scenario number and the cause of the anomaly are recorded in the log to avoid the failure of a single scenario causing the entire candidate pre-scheduling control strategy to be unable to perform the expected comprehensive risk calculation, thereby ensuring that the scheduling decision-making process can continue to operate under abnormal conditions.

[0046] In S2, the prediction bias and historical fluctuation characteristics can be calculated by formulas (1) and (2) in the embodiment section, respectively. The probabilistic rainfall input scenario and its scenario probability weight in S3 are obtained based on the credibility coefficient of each forecast data source in the scenario construction stage of S3, and are used in S5 to calculate the expected comprehensive risk cost of each candidate pre-arrangement control strategy.

[0047] S6 selects the pre-drainage control strategy with the lowest expected comprehensive risk cost from multiple candidate pre-drainage control strategies, generates the corresponding irrigation and drainage scheduling instructions, and issues them for execution.

[0048] Example 2: Based on Example 1, a feedback optimization step is also included after S6.

[0049] The feedback optimization process includes the following steps: The feedback optimization step takes the probabilistic rainfall input scenario, water level simulation results, irrigation and drainage scheduling instructions, and actual complete local monitoring data generated during the execution of S1 to S6 as inputs. By comparing the model pre-simulation results with the actual observation results, it outputs the adjustment amount used to update the confidence update model parameters, probabilistic scenario transformation rules, and risk cost calculation model parameters, thus forming a closed loop from online pre-schedule scheduling to post-model correction. After the actual rainfall process ends and the irrigation and drainage scheduling instructions are executed, the prediction effect of the probabilistic rainfall input scenario, the simulation accuracy of the digital twin model of the farmland drainage network, and the assessment results of the expected comprehensive risk cost are verified using actual and complete local monitoring data. Based on the verification results, the parameters in the credibility update model in S2, the probability scenario transition rule in S3, and the risk cost calculation model in S5 are adaptively adjusted. The irrigation and drainage scheduling instructions generated in S6 are sent to the edge intelligent control unit deployed in the field control equipment of the farmland drainage network; The irrigation and drainage scheduling instructions include control logic that allows the edge intelligent control unit to fine-tune the operation of the control equipment within a preset rule range based on local high-frequency ultra-real-time monitoring data; During execution, the edge intelligent control unit transmits its local operating status and fine-tuning records back to the central scheduling system for feedback on model and parameter updates in the optimization steps; The risk cost of insufficient pre-drainage includes at least the loss cost caused by the water level at key control nodes exceeding the preset control water level threshold, resulting in overtopping, waterlogging, or overload operation of the canal system. The risk cost of excessive pre-drainage includes at least the loss cost caused by insufficient canal storage capacity utilization, excessively high downstream water levels, and increased energy consumption of pumping stations due to premature emptying. The probabilistic rainfall input scenario is derived at least in part from the rainfall forecast results of the same forecast period from the ensemble numerical weather prediction members and different short-term forecast algorithms. Furthermore, when generating the probabilistic rainfall input scenario, the initial confidence coefficient and its allowable range of change are pre-set according to the type of each forecast algorithm or forecast member, thereby limiting the impact of a single forecast data source on the pre-schedule decision.

[0050] In another aspect, the present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method described above.

[0051] In another aspect, the present invention also discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method described above.

[0052] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to execute any of the above embodiments of the intelligent pre-drainage scheduling method for farmland drainage network that integrates short-term forecasts.

[0053] It is understood that the systems, devices, and storage media provided in the embodiments of the present invention correspond to the methods provided in the embodiments of the present invention, and the explanations, examples, and beneficial effects of the relevant content can be referred to the corresponding parts of the above methods.

[0054] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another.

[0055] For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media.

[0056] The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid state disks (SSDs)).

[0057] It should be noted that in this paper, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0058] Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0059] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0060] The embodiments of the present invention are given for the purposes of illustration and description. Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for pre-discharge scheduling of farmland drainage canal networks integrating short-term forecasts, characterized in that, Perform the following steps using a computer device: S1, acquire short-term rainfall forecast data composed of multiple forecast data sources, as well as real-time monitoring data and historical fluctuation characteristics of local monitoring stations of farmland drainage network; S2, based on short-term precipitation forecast data and real-time monitoring data, the prediction deviation is obtained. Based on the prediction deviation and historical fluctuation characteristics, the dynamic credibility of each forecast data source and local monitoring station is evaluated to obtain the credibility coefficient of each data source. S3, based on the confidence coefficient of each forecast data source and the short-term precipitation forecast data, a probabilistic precipitation input scenario is obtained. The probabilistic precipitation input scenario is a set of precipitation sequences with occurrence probability weights. S4. Input the probabilistic rainfall input scenario into the digital twin model of the farmland drainage network for hydraulic simulation, and obtain the probability distribution of water level, field surface water depth and groundwater level of key control nodes and field monitoring points of the farmland drainage network in a future predetermined time period. S5. Obtain multiple candidate pre-discharge control strategies. Calculate the risk cost of insufficient pre-discharge and the risk cost of excessive pre-discharge for each pre-discharge control strategy based on the probability distribution. Sum the risk costs under each scenario based on the probability weights of the probabilistic rainfall input scenario to obtain the expected comprehensive risk cost of each pre-discharge control strategy. S6 selects the pre-drainage control strategy with the lowest expected comprehensive risk cost from multiple candidate pre-drainage control strategies and generates the corresponding irrigation and drainage scheduling instructions.

2. The method for pre-discharge scheduling of farmland drainage canal networks based on short-term forecasts as described in claim 1, characterized in that: In S1, the real-time monitoring data of the local monitoring station includes local real-time rainfall, canal network water level, flow monitoring data, field surface water depth data, and groundwater level monitoring data; In step S2, when some of the local monitoring stations have missing data; The corresponding local monitoring data that has undergone quality control is obtained through the following methods: By integrating effective monitoring data from nearby monitoring stations, historical data patterns from missing monitoring stations, and equipment status records from the missing monitoring stations, a complete monitoring dataset covering all monitoring times and indicators is generated using data interpolation and reconstruction algorithms. In S5, the risk cost of insufficient pre-drainage includes the loss cost caused by waterlogging, overflow, or poor drainage due to the water level at key control nodes exceeding the control water level threshold or the water depth on the field exceeding the allowable water accumulation depth threshold. The risks and costs associated with excessive pre-drainage include the adverse effects of insufficient field water storage, inadequate canal network regulation and utilization, or downstream water level constraints caused by premature drainage, as well as the losses resulting from increased pump station energy consumption.

3. The method for pre-discharge scheduling of farmland drainage canal networks based on short-term forecasts as described in claim 2, characterized in that: The specific division of S2 includes the following steps: S21, assign and maintain a dynamically updated confidence coefficient for each independent forecast data source and local monitoring station; S22. In each evaluation period, for each forecast data source, the rainfall forecast made for the farmland drainage network field group in the previous period is compared with the corresponding local monitoring data after quality control to obtain the forecast deviation. S23. Based on the prediction deviation and the historical fluctuation characteristics of the corresponding data source, a new confidence coefficient of the data source is calculated through a preset confidence update model. The confidence coefficient decreases as the prediction deviation increases and is lowered when the historical fluctuation of the data source intensifies.

4. The method for pre-discharge scheduling of farmland drainage network based on short-term forecasting as described in claim 1, characterized in that: In step S3, the step of obtaining the probabilistic rainfall input scenario based on the confidence coefficient of each forecast data source and the transformation of multi-source short-term rainfall forecast data includes: Using the confidence coefficient of each forecast data source as the weight or sampling probability basis of the forecast result of the forecast data source, the precipitation field or ensemble forecast members of each forecast data source within the forecast spatiotemporal range are randomly sampled and combined to generate multiple precipitation scenario sequences that are different in spatiotemporal distribution and are each attached with a probability weight of occurrence. Meanwhile, the probabilistic rainfall input scenario is derived from the rainfall forecast results of the same forecast period by the ensemble numerical weather forecast members or different short-term forecast algorithms. Furthermore, when generating the probabilistic rainfall input scenario, the initial confidence coefficient and its allowable range of change are pre-set according to the type of each forecast algorithm or forecast member.

5. The method for pre-discharge scheduling of farmland drainage network based on short-term forecasting as described in claim 1, characterized in that: In S4, the probability distribution specifically includes: at several future predetermined time points, providing statistics, confidence intervals, and probability density functions for the corresponding water level, field surface water depth, and groundwater level for each of the key control nodes and field monitoring points.

6. The method for pre-discharge scheduling of farmland drainage canal networks based on short-term forecasts as described in claim 1, characterized in that: The specific division of S5 includes the following steps: S51 defines multiple candidate pre-discharge control strategies, each of which contains the sequence of operations for the gates and pumping station control equipment in the farmland drainage network during the future scheduling period. S52, for each candidate pre-drainage control strategy and each probabilistic rainfall input scenario, the farmland drainage network digital twin model is invoked to simulate, obtain the consequences of insufficient pre-drainage and excessive pre-drainage of the candidate pre-drainage control strategy under the probabilistic rainfall input scenario, and quantify the consequences into corresponding risk costs. S53, for each candidate pre-arrangement control strategy, the risk costs obtained under all the probabilistic rainfall input scenarios are weighted and summed according to the probability weights of the corresponding probabilistic rainfall input scenarios to obtain the expected comprehensive risk cost of the candidate pre-arrangement control strategy.

7. The method for pre-discharge scheduling of farmland drainage network based on short-term forecasting as described in claim 1, characterized in that: Following step S6, a feedback optimization step is included, which includes: After the actual rainfall process ends and the irrigation and drainage scheduling instructions are executed, the prediction effect of the probabilistic rainfall input scenario, the simulation accuracy of the digital twin model of the farmland drainage network, and the assessment results of the expected comprehensive risk cost are verified using actual and complete local monitoring data. Based on the verification results, the parameters in the credibility update model in S2, the probability scenario transformation rule in S3, and the risk cost calculation model in S5 are adaptively adjusted.

8. The intelligent pre-drainage scheduling method for farmland drainage network integrating short-term forecasts as described in claim 7, characterized in that, In step S6, the generated irrigation and drainage scheduling command is sent to the edge intelligent control unit deployed in the field control equipment of the farmland drainage network; The irrigation and drainage scheduling instructions include control logic that allows the edge intelligent control unit to fine-tune the operation of the control equipment within a preset rule range based on local high-frequency ultra-real-time monitoring data; During execution, the edge intelligent control unit transmits its local operating status and fine-tuning records back to the central scheduling system for model and parameter updates in the feedback optimization step.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it causes the processor to perform the steps of the method as described in any one of claims 1 to 8.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the computer program is executed by the processor, it causes the processor to perform the steps of the method as described in any one of claims 1 to 8.

Images