A polder waterlogging forecast and early warning method and system

Abstract

The application discloses a polder waterlogging disaster forecasting and early warning method and system. The application relates to the technical field of polder waterlogging disaster early warning, and aims to solve the problems that the forecasting accuracy of the prior art is insufficient, the connection between early warning information and actual pump station scheduling needs to rely on manual experience, and the grading response characteristics of crops in different growth periods to waterlogging depth and duration cannot be fully utilized. The application obtains multi-source weather forecasting data and real-time field monitoring data and performs weighted fusion; future field water level and future ditch water level are predicted; the growth period is determined, the warning water level is calculated, and the early warning level is determined; when the early warning level exceeds a threshold value, the amount of water to be pre-pumped is calculated, and a joint scheduling scheme is generated through a multi-objective optimization algorithm to minimize the operation energy consumption of a pump station group and the yield reduction loss of crops, and the joint scheduling scheme is executed. The application realizes full-chain automatic decision-making closed loop from weather forecasting to pump station group scheduling, and improves the waterlogging disaster forecasting accuracy and emergency scheduling timeliness.

Description

Technical Field

[0001] This invention relates to the field of flood warning technology for polder areas, and in particular to a method and system for forecasting and warning of floods in polder areas. Background Technology

[0002] Polder areas are enclosed flood control and drainage areas formed by dikes in low-lying, flood-prone areas along plains, river networks, or lakes. They are also important agricultural production areas. Polder areas are characterized by flat terrain and low elevations, making them highly susceptible to flooding during heavy rainfall. While research on farmland flood control has reached a certain level, it suffers from the following shortcomings: First, existing technologies primarily focus on improving drainage devices. For example, patent CN102031768B discloses a novel farmland waterlogging prevention device, but this requires the construction of multi-stage drainage systems and groundwater level control devices, resulting in high investment costs and limited short-term widespread adoption. Furthermore, it lacks the ability to forecast and warn of future waterlogging disasters. Second, some early warning methods only involve issuing warning information. For instance, patent CN118114827A discloses a flood disaster forecasting and early warning method for plains and river networks based on hydrological and hydrodynamic models, but it does not provide a closed-loop decision-making process from early warning to engineering scheduling. The connection between early warning information and actual pump station scheduling still relies on human experience. Third, existing methods often rely on a single data source or a single model, and their prediction accuracy and robustness need improvement. Furthermore, existing studies on crop yield reduction assessments often employ fixed piecewise function forms, failing to fully utilize the graded response characteristics of crops at different growth stages to flood depth and duration. Therefore, it is necessary to propose a method and system for forecasting and warning of flooding in polder areas to address these issues. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for forecasting and warning of flooding in polder areas, in order to solve the problems of insufficient forecast accuracy in existing technologies, reliance on human experience for the connection between early warning information and actual pump station scheduling, and failure to fully utilize the graded response characteristics of crops at different growth stages to flood depth and duration.

[0004] In a first aspect, the present invention provides a method for forecasting and early warning of flooding in polder areas, comprising the following steps: Step 1: Obtain multi-source weather forecast data and real-time field monitoring data. The multi-source weather forecast data includes the rainfall amount and duration for the next few days, and the real-time field monitoring data includes field water levels and ditch water levels. Step 2: The multi-source meteorological forecast data and the real-time field monitoring data are weighted and fused to obtain a fused forecast dataset; Step 3: Based on the rainfall and rainfall duration in the fused forecast dataset and the field water level in the real-time field monitoring data, predict the future field water level using a pre-trained long short-term memory network model; based on the rainfall and rainfall duration in the fused forecast dataset and the ditch water level and drainage flow in the real-time field monitoring data, predict the future ditch water level using a pre-trained random forest model. Step 4: Determine the current growth stage based on the rice transplanting date, calculate the current warning water level based on the plant height growth model, determine the flooding depth range based on the ratio of the future field water level to the current warning water level, and determine the warning level based on the number of consecutive flooding days within the flooding depth range and the preset yield reduction classification matrix of growth stage-flooding depth-consecutive flooding days. Step 5: When the warning level exceeds the preset threshold, calculate the pre-discharge volume according to the pre-discharge volume formula; Step 6: With the optimization objectives of minimizing the total operating energy consumption of the pump station group and minimizing crop yield loss, and with the constraints of the rated flow of each pump station, the upper limit of the number of units that can be started, and the pre-drainage volume, a multi-objective optimization algorithm is used to generate a joint scheduling scheme that includes the start-up and shutdown sequence, the number of units in operation, and the running time of each pump station, and then the scheme is issued for execution.

[0005] Furthermore, in step three, the training process of the long short-term memory network model includes: Historical rainfall data, field water level data, ditch water level data, and drainage flow data were collected as training samples; The long short-term memory network model was trained using historical rainfall duration, rainfall amount, and pre-rain water depth as input features and historical field water level as a label. After training, the model is adaptively corrected based on feedback from real-time monitoring data.

[0006] Furthermore, in step three, the training process of the random forest model includes: Historical rainfall data, field water level data, ditch water level data, and drainage flow data were collected as training samples; The random forest model was trained using historical pre-rainwater levels, rainfall, rainfall duration, and drainage flow as input features and historical water levels as labels. After training, the model is adaptively corrected based on feedback from real-time monitoring data.

[0007] Furthermore, in step four, the plant height growth model... y = 44.46 e 0.0113x ; in, y Plant height, xThis is the number of days since transplanting; the current warning water level is 1 / 4 of the plant's height.

[0008] Furthermore, in step four, the yield reduction grading matrix is ​​set separately for the jointing and booting stage, the heading and flowering stage, and the milk stage. Under each growth stage, it is divided into five levels according to the flooding depth range and the number of consecutive flooding days: mild yield reduction, moderate yield reduction, relatively severe yield reduction, severe yield reduction, and serious yield reduction.

[0009] Furthermore, in step four, the flooding depth range is determined based on the ratio of future field water level to plant height. Specifically, if the future field water level is above the warning level but does not reach 1 / 2 of the plant height, it is determined to be the 1 / 4 plant height range; if the future field water level reaches 1 / 2 of the plant height but does not reach 3 / 4 of the plant height, it is determined to be the 1 / 2 plant height range; if the future field water level reaches 3 / 4 of the plant height but does not reach full flooding, it is determined to be the 3 / 4 plant height range; and if the future field water level reaches or exceeds the plant height, it is determined to be the full flooding range.

[0010] Furthermore, the number of consecutive flooded days is calculated based on the real-time field water level data for the current day and the previous 15 days, combined with the future field water level for the next 5 days.

[0011] Furthermore, in step five, the formula for the pre-drainage volume is:

[0012] In the formula, A t For farmland area, h pred This represents the predicted average water depth in the fields after rain. h current To warn of water levels, V upstream This refers to the upstream water volume. A river The average cross-sectional area of ​​the river channel. h safe For the safe water level of the river, h river_current This refers to the current real-time water level of the river channel; when V pre No drainage is required when ≤ 0. V pre When the value is >0, drainage is required and the drainage volume is: V pre .

[0013] Further, in step six, the multi-objective optimization algorithm is any one of the following: NSGA-II non-dominated sorting genetic algorithm, multi-objective particle swarm optimization algorithm, or simulated annealing-swarm intelligence hybrid algorithm; the constraints include the rated flow rate of a single pump in each pumping station, the upper limit of the number of units that can be started, and the total amount of pre-drainage; the optimization objective function is:

[0014] In the formula, f 1 represents the total operating energy consumption target for the pump station group. C i For the first i Energy consumption coefficient per unit time of the pump station unit. P i In start / stop state. T i Runtime; f 2 is the target for crop yield reduction, based on the predicted field water level. h pred With warning water level h current The difference and duration are used to determine the result.

[0015] Secondly, the present invention provides a flood forecasting and early warning system for polder areas, comprising: The data acquisition module is used to acquire multi-source weather forecast data and real-time field monitoring data. The multi-source weather forecast data includes the rainfall amount and duration for the next few days, and the real-time field monitoring data includes field water level and ditch water level. The multi-source weather forecast data and the real-time field monitoring data are weighted and fused to obtain a fused forecast dataset. The water level prediction module is used to predict future field water levels based on rainfall, rainfall duration, and field water levels in the fused forecast dataset and real-time field monitoring data using a pre-trained long short-term memory network model; and to predict future ditch water levels based on rainfall, rainfall duration, and ditch water levels and drainage flow in the fused forecast dataset and real-time field monitoring data using a pre-trained random forest model. The early warning and classification module is used to determine the current growth stage based on the crop transplanting date, calculate the current warning water level based on the plant height growth model, determine the flooding depth range based on the ratio of the future field water level to the current warning water level, and determine the early warning level based on the number of consecutive flooding days within the flooding depth range and by referring to the preset yield reduction classification matrix of growth stage-flooding depth-consecutive flooding days. The intelligent scheduling module is used to calculate the pre-drainage volume according to the pre-drainage volume formula when the warning level exceeds the preset threshold. With minimizing the total operating energy consumption of the pump station group and minimizing crop yield loss as optimization objectives, and with the rated flow of each pump station, the upper limit of the number of units that can be started, and the required pre-drainage volume as constraints, the module uses a multi-objective optimization algorithm to generate a joint scheduling scheme that includes the start-up and shutdown sequence of each pump station, the number of units in operation, and the running time, and then issues it for execution. The front-end visualization platform is used for the visualization and interactive operation of early warning information, risk assessment results, and scheduling plans. The backend service module is used for data management, business logic processing, and API interface services.

[0016] The beneficial effects of this invention are as follows: The polder flood forecasting and early warning method and system of this invention, by constructing a complete closed loop from multi-source meteorological data acquisition, field and ditch water level dual-model prediction, crop growth stage-based hierarchical early warning assessment to multi-objective optimization scheduling of pump station groups, realizes a universal automated decision-making closed loop from meteorological forecasting to joint scheduling of pump station groups. This replaces fixed engineering rules that are only applicable to specific configurations, enabling the method to adaptively generate optimal scheduling schemes for polder pump station groups with different numbers and configurations of units, thus broadening its application scope. The use of a dual model of long short-term memory network and random forest to predict field and ditch water levels respectively fully leverages the advantages of deep learning and ensemble learning in predicting different water bodies, significantly improving prediction accuracy. The early warning assessment method based on the plant height growth model and the growth stage-flooding depth-continuous flooding days hierarchical matrix is ​​more in line with the actual flood tolerance characteristics of crops, resulting in higher early warning accuracy. The system adopts a front-end and back-end separation architecture, which facilitates deployment and maintenance. Attached Figure Description

[0017] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the system architecture according to an embodiment of the present invention; Figure 3 This is a flowchart of the multi-objective optimization scheduling algorithm for pump station groups according to an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention. The technical solutions provided by various embodiments of this invention will be described in detail below with reference to the accompanying drawings.

[0020] Please see Figures 1 to 3 The flowchart of the method used in this embodiment is as follows: Figure 1 As shown, the system architecture is as follows: Figure 2 As shown.

[0021] A specific polder area was selected as the research subject. This polder area has a subtropical monsoon climate, characterized by distinct seasons, simultaneous rainfall and heat, and abundant but uneven spatial and temporal distribution of precipitation. The total drainage area is approximately 972 km². 2 Of this, 31,600 mu (approximately 2,400 hectares) are protected arable land. Agriculture in the region is primarily focused on grain and oilseed production, with rice cultivation being the main grain crop. The overall terrain of the study area is relatively flat, with the surface elevation gradually decreasing from west to east, resulting in some low-lying areas and a shallow drainage slope. The area relies on a single drainage ditch for flood control, making it prone to waterlogging and flooding during heavy rainfall, thus facing significant drainage challenges.

[0022] I. Data Collection and Preprocessing Collect basic geographic information data such as topography, water system, water conservancy project layout, and land use of the study area, as well as historical meteorological data, field water level observation data, ditch water level observation data, and pump station operation records.

[0023] Historical meteorological data selected from a continuous 30-year (1994-2023) time series, including temperature, humidity, wind speed, and precipitation, were analyzed from four dimensions: total amount, duration, intensity, and extremes. The average annual precipitation over the past 30 years was 1725.9 mm, with an average of 1197.2 mm during the flood season (April-September), accounting for an average of 69.4% of the annual precipitation. Significant interannual fluctuations were observed, with a difference of 76 days between the year with the most rainy days (1997, 202 rainy days) and the year with the fewest (2003, 2007, 126 rainy days). During a typical flood disaster from June to August 2024, based on Sentinel-1 SAR remote sensing image inversion, the maximum inundated area reached 2.0291 km². 2 .

[0024] Field water level data were obtained from five field water level monitoring points within the study area, with a sampling interval of one hour. Ditch water level data were obtained from 15 existing IoT monitoring points (equipment model: JH-WL100-FC) in the polder area and drainage ditch water level monitoring stations. Pump station operation records were obtained from the electric pumping station, which is equipped with four drainage pump sets, each with a rated flow rate of 64.4 m³ / h. 3 Due to limitations imposed by power configuration and operational management regulations, the pumping station can only support a combined operation of 2 or 4 pumps in actual operation. Upstream water inflow data is acquired through an IoT platform.

[0025] II. Model Construction and Training Based on historical monitoring data, we comprehensively adopted methods such as random forest, support vector machine and neural network to carry out modeling and comparative analysis. We used rainfall data, drainage data and pre-rain water level as input values ​​and post-rain water level as output values ​​to select the models with the highest accuracy.

[0026] (1) Long Short-Term Memory (LSTM) Field Water Level Prediction Model A field water level prediction model based on a long short-term memory network was constructed. The model's input features are rainfall duration (min) and rainfall amount ( ). mm ) and water depth before rain ( m The pre-rain water depth is taken as the average of the most recent field water level gauge readings. The model output is the predicted future field water level. m ).

[0027] The root mean square error of the LSTM model on the validation set ( RMSE The coefficient of determination is 0.0950. R 2 It is 0.8406. Pearson The correlation coefficient is 0.9174. Compared to random forest ( RMSE =0.1045, R 2 =0.8073), Support Vector Machine ( RMSE =0.1024, R 2 =0.8105), decision tree (RMSE=0.1665, R 2 Compared to traditional machine learning models such as (e.g., 0.5105), the LSTM model performs best across all metrics and is better able to characterize the nonlinear relationship and time-series dependence features between rainfall processes and field water levels.

[0028] (2) Random forest ditch water level prediction model A ditch water level prediction model based on random forest regression was constructed. The input feature of the model is the pre-rain ditch water level (…). m ), rainfall ( mm ), duration of rainfall ( h ) and drainage flow (m 3 / s). The pre-rain ditch water level is taken as the average value of the ditch water level monitoring points at the current time. The model output is the predicted future ditch water level ( m ).

[0029] After parameter optimization, the training set of the random forest model... R 2 It is 0.9148. MAE The value is 0.1693; Test set R 2 It is 0.8348. MAE The value was 0.2207, demonstrating strong generalization ability. Compared to linear regression (test set) R 2=0.3804), Support Vector Regression ( SVR-RBF test set R 2 =0.1153), Neural Network (Test Set) R 2 Compared to other models such as (e.g., 0.2723), the random forest model outperforms other models in all indicators and can more accurately reflect the changing trend of water level elevation in drainage ditches.

[0030] (3) Adaptive correction of the model In actual operation, the system regularly feeds back the latest accumulated measured data to the model and uses incremental learning to update the model parameters in order to maintain the model's adaptability to seasonal changes and changes in farmland underlying surface conditions.

[0031] III. Acquisition and Fusion of Multi-Source Forecast Data The system utilizes the standardized global weather forecasting API provided by OpenWeatherMap and builds an automated data acquisition module on Alibaba Cloud Elastic Compute Service (ECS). Data acquisition is automatically executed daily at 02:00 via a Cron scheduled task on the Linux system. The acquired forecast elements include seven meteorological parameters: rainfall (mm), temperature (°C), humidity (%), air pressure (hPa), and wind speed (m / s) every 3 hours for the next 5 days. The system dynamically evaluates the forecast accuracy of the data source and assigns fusion weights based on the forecast accuracy over the past 30 days to generate a fused forecast dataset.

[0032] IV. Water Level Prediction For field water level prediction, the predicted rainfall and rainfall duration for each 3-hour period in the future are extracted from the fused forecast dataset, combined with the measured field water level at the current moment, and then input into the LSTM model to predict the future field water level sequence in a rolling manner for each period.

[0033] For ditch water level prediction, the forecast rainfall and rainfall duration (converted to hours) are extracted from the fused forecast dataset, combined with the current measured ditch water level and drainage flow monitoring value, and input into the random forest model to predict the future ditch water level.

[0034] V. Early Warning Classification and Issuance (1) Calculation of plant height and warning water level The system has a built-in rice plant height growth model: y = 44.46 e 0.0113xThe exponential growth function is constructed based on field data from paddy fields in the region. It exhibits a high R² fit and effectively reflects the relationship between rice plant height and the number of days after transplanting. Administrators input the transplanting dates for early and late rice within the system interface. Early rice is typically transplanted in mid-April, with a growth period from April to July; late rice is typically transplanted in late July, with a growth period from July to November. Starting from the transplanting date (referred to as day 0), the system automatically calculates the number of days after transplanting and uses this data to call the plant height growth model to obtain the corresponding rice plant height value for that period.

[0035] The field warning water level is set at 1 / 4 of the current rice plant height. When the field water level exceeds this warning level, it indicates that the rice has entered a waterlogging-sensitive state, requiring further analysis of the duration of waterlogging and the growth stage to conduct flood warning identification. According to the "Irrigation and Drainage Engineering Design Standard" (GB50288-2018), the flood-tolerant depth for rice at maturity is 30-35 cm. The plant height growth model constructed in this invention calculates a warning water level of approximately 22-27 cm at the milk stage, which is within the flood-tolerant range specified in the standard, and retains a safety zone of approximately 5-10 cm.

[0036] (2) Determination of reproductive period The system automatically determines the crop's growth stage based on the transplanting date and the current date. In the study area, single-season rice is grown. Referring to field data from the region, the growth stages are divided as follows: jointing and booting stage is 35-51 days after transplanting; heading and flowering stage is 52-61 days after transplanting; and milk stage is 62-88 days after transplanting.

[0037] (3) Determination of flood depth range and number of consecutive days The system determines the flooding depth range based on the ratio of predicted field water level to plant height: if the predicted water level is above the warning level but does not reach 1 / 2 plant height, it is determined to be in the "1 / 4 plant height" range; if it reaches 1 / 2 plant height but does not reach 3 / 4 plant height, it is determined to be in the "1 / 2 plant height" range; if it reaches 3 / 4 plant height but does not reach full flooding, it is determined to be in the "3 / 4 plant height" range; if it reaches or exceeds the plant height, it is determined to be in the "full flooding" range.

[0038] The system saves real-time field water level data for the current day and the previous 15 days, and combines this with the predicted field water level for the next 5 days to calculate the number of consecutive flooded days T for each flood depth range. When the field water depth is continuously within the same flood depth range in historical monitoring and future forecasts, the corresponding number of days is accumulated; if the water depth is lower than the depth threshold, the continuous counting is interrupted and recalculated.

[0039] (4) Warning levels and issuance The warning levels are divided into six levels according to the severity of yield reduction: no warning (not reaching the level of mild yield reduction), Level 1 warning (mild yield reduction), Level 2 warning (moderate yield reduction), Level 3 warning (relatively severe yield reduction), Level 4 warning (severe yield reduction), and Level 5 warning (very severe yield reduction). The grading matrix is ​​set separately for the jointing and booting stage, the heading and flowering stage, and the milk stage. Different continuous flooding duration threshold ranges are used for each growth stage due to the different waterlogging tolerance of rice.

[0040] The following example illustrates a typical rainfall event from June 16th to 25th, 2022. This period coincides with the milk-ripe stage of early-season single-cropping rice in the polder area, a critical period for yield formation. The simulated cumulative rainfall reached 166.4 mm, with a maximum single-day rainfall of 48.92 mm on June 17th, followed by continuous moderate to heavy rain from June 18th to 20th. On June 16th, the initial field water depth was 3.0 cm, and the initial drainage ditch water level was 35.8 m. With the occurrence of heavy rainfall, by June 19th, the predicted field water level gradually approached and exceeded the warning water level threshold corresponding to the milk-ripe stage, with continuous flooding lasting for 6 days. Based on a comprehensive analysis of field water depth, duration of continuous flooding, and growth stage conditions, the platform identified a Level 1 warning (mild yield reduction).

[0041] VI. Generation of Optimized Scheduling Scheme for Pump Station Group When the warning level reaches the preset trigger condition, the system initiates optimized scheduling decision-making for the pump station group. The algorithm flow is as follows: Figure 3 As shown.

[0042] First, calculate the total amount of water that needs to be pre-discharged according to the pre-discharge formula. V pre Taking the parameters of this polder area as an example: farmland area A t = 35.64×10 6 m 2 Average cross-sectional area of ​​the river channel A river = 1.2×10 6 m 2 River safety water level h safe = 40m. Calculate the water level difference based on the predicted average water depth in the fields after the rain and the current warning water level. Combine this with the upstream inflow and the current available storage capacity of the river channel to calculate the required pre-discharge volume.

[0043] when V pre When the water level is ≤ 0, the existing water storage capacity of the river channel is sufficient to absorb the water flowing into the farmland, and there is no need to initiate drainage operations. V pre When the value is greater than 0, the system enters the multi-objective optimization scheduling stage of the pump station group.

[0044] The system employs a dual objective of minimizing the total operating energy consumption of the pumping station group and minimizing crop yield loss, establishing a multi-objective optimization mathematical model. Constraints include the rated flow rate of each pump at each station, the maximum number of units that can be operated, and the total required drainage volume. The optimization objective function is:

[0045] In the formula, f 1 represents the total operating energy consumption target for the pump station group. C i For the first i Energy consumption coefficient per unit time of the pump station unit. P i In start / stop state. T i Runtime; f 2 is the target for crop yield reduction, based on the predicted field water level. h pred With warning water level h current The difference and duration are used to determine the result.

[0046] Taking a certain electric pumping station as an example, a total of 4 drainage pump sets are configured (each pump has a rated flow rate of 64.4 m³ / h). 3 / s). The system automatically transforms the scheduling problem into an optimization problem: the algorithm automatically evaluates the trade-off between "operating 2 pumps for a longer period of time" and "operating 4 pumps for a shorter period of time" in terms of energy consumption and production reduction. A multi-objective optimization algorithm (such as the NSGA-II non-dominated sorting genetic algorithm) searches for the Pareto front between the two options, automatically generates the start-up and shutdown sequence, the number of pumps to be operated, and the specific running time of each pump station unit based on decision preferences, and outputs an optimal joint scheduling scheme that balances economy and safety and issues it for execution.

[0047] Taking a typical rainfall event in June 2022 as an example, the system calculation results showed that drainage measures needed to be activated, and the generated contingency plan result was: drainage volume of 1,393,428 m³. 3 The pump start-up time is 3 hours, and the number of pumps is 2.

[0048] VII. System Architecture and Deployment The forecasting, early warning and scheduling system of the present invention adopts a B / S architecture and follows the design pattern of front-end and back-end separation.

[0049] The front-end platform is developed using TypeScript, employing the Vue 3 framework, Vite build tool, and Ant DesignVue component library to build the user interface. Pinia is integrated for status management, and the ECharts chart library provides visualization of warning levels, water level change trends, and pump station operating status. Users can access the system through a browser without installing client software.

[0050] The backend service is developed using Java, employing the Spring Boot framework and Maven project management tool. The data persistence layer uses the MyBatis-Plus framework. The database is a PostgreSQL relational database, storing structured data such as historical meteorological data, water level monitoring data, and pump station operation records. Redis is used as a distributed cache to store frequently accessed data such as forecast data and early warning results, improving system response speed. The XXL-JOB distributed task scheduling framework is integrated for periodically triggering weather forecast data acquisition tasks and distributing scheduling schemes. File storage uses MinIO object storage service to store unstructured data such as system logs and reports.

[0051] The platform is divided into a business function module and a basic information module. The business function module includes functional pages such as homepage overview, rainfall management, water level management, flood forecast, flood assessment, early warning plan, and pump station management; the basic information module includes content such as polder management department, polder basic information, pump station basic information, pump station equipment information, rain gauge information, and equipment management.

[0052] The system uses a RESTful API to achieve data communication between the front-end and back-end. The default port configuration for the front-end is 3100, and the default port configuration for the back-end is 30001. The system has multi-platform adaptability and can be adapted to PCs, tablets, and mobile devices.

[0053] VIII. System Performance Using a typical heavy rainfall event from June 16th to 25th, 2022 as a simulated application scenario, the system underwent complete business process verification. This period coincided with the milk-ripe stage of early and mid-season single-season rice in the polder area, with a cumulative rainfall of 166.4 mm and a maximum daily rainfall of 48.92 mm. Simulation results show that the platform can automatically complete the entire business operation process of meteorological data acquisition, field water level forecasting, early warning level identification, and emergency drainage plan generation without intervention, verifying the collaborative operation capability of the forecast-early warning-plan business chain.

[0054] The embodiments of the present invention described above do not constitute a limitation on the scope of protection of the present invention.

Claims

1. A method for forecasting and early warning of flooding in polder areas, characterized in that, Includes the following steps: Step 1: Obtain multi-source weather forecast data and real-time field monitoring data. The multi-source weather forecast data includes the rainfall amount and duration for the next few days, and the real-time field monitoring data includes field water levels and ditch water levels. Step 2: The multi-source meteorological forecast data and the real-time field monitoring data are weighted and fused to obtain a fused forecast dataset; Step 3: Based on the rainfall and rainfall duration in the fused forecast dataset and the field water level in the real-time field monitoring data, predict the future field water level using a pre-trained long short-term memory network model; based on the rainfall and rainfall duration in the fused forecast dataset and the ditch water level and drainage flow in the real-time field monitoring data, predict the future ditch water level using a pre-trained random forest model. Step 4: Determine the current growth stage based on the rice transplanting date, calculate the current warning water level based on the plant height growth model, determine the flooding depth range based on the ratio of the future field water level to the current warning water level, and determine the warning level based on the number of consecutive flooding days within the flooding depth range and the preset yield reduction classification matrix of growth stage-flooding depth-consecutive flooding days. Step 5: When the warning level exceeds the preset threshold, calculate the pre-discharge volume according to the pre-discharge volume formula; Step 6: With the optimization objectives of minimizing the total operating energy consumption of the pump station group and minimizing crop yield loss, and with the constraints of the rated flow of each pump station, the upper limit of the number of units that can be started, and the pre-drainage volume, a multi-objective optimization algorithm is used to generate a joint scheduling scheme that includes the start-up and shutdown sequence, the number of units in operation, and the running time of each pump station, and then the scheme is issued for execution.

2. The method for forecasting and early warning of flooding in polder areas according to claim 1, characterized in that, Step three, the training process of the long short-term memory network model includes: Historical rainfall data, field water level data, ditch water level data, and drainage flow data were collected as training samples; The long short-term memory network model was trained using historical rainfall duration, rainfall amount, and pre-rain water depth as input features and historical field water level as a label. After training, the model is adaptively corrected based on feedback from real-time monitoring data.

3. The method for forecasting and early warning of flooding in polder areas according to claim 1, characterized in that, Step three, the training process of the random forest model includes: Historical rainfall data, field water level data, ditch water level data, and drainage flow data were collected as training samples; The random forest model was trained using historical pre-rainwater levels, rainfall, rainfall duration, and drainage flow as input features and historical water levels as labels. After training, the model is adaptively corrected based on feedback from real-time monitoring data.

4. The method for forecasting and early warning of flooding in polder areas according to claim 1, characterized in that, In step four, the plant height growth model y = 44.46 e 0.0113x ; in, y Plant height, x This is the number of days since transplanting; the current warning water level is 1 / 4 of the plant's height.

5. The method for forecasting and early warning of flooding in polder areas according to claim 1, characterized in that, In step four, the yield reduction grading matrix is ​​set separately for the jointing and booting stage, the heading and flowering stage, and the milk stage. Under each growth stage, it is divided into five levels according to the flooding depth range and the number of consecutive flooding days: mild yield reduction, moderate yield reduction, relatively severe yield reduction, severe yield reduction, and serious yield reduction.

6. The method for forecasting and early warning of flooding in polder areas according to claim 1, characterized in that, In step four, the flooding depth range is determined based on the ratio of future field water level to plant height. Specifically, if the future field water level is above the warning level but does not reach 1 / 2 of the plant height, it is determined to be the 1 / 4 plant height range; if the future field water level reaches 1 / 2 of the plant height but does not reach 3 / 4 of the plant height, it is determined to be the 1 / 2 plant height range; if the future field water level reaches 3 / 4 of the plant height but does not reach full flooding, it is determined to be the 3 / 4 plant height range; and if the future field water level reaches or exceeds the plant height, it is determined to be the full flooding range.

7. The method for forecasting and early warning of flooding in polder areas according to claim 1, characterized in that, The number of consecutive flooded days is calculated based on the real-time field water level data for the current day and the previous 15 days, combined with the future field water level for the next 5 days.

8. The method for forecasting and early warning of flooding in polder areas according to claim 1, characterized in that, In step five, the formula for pre-drainage volume is: In the formula, A t For farmland area, h pred This represents the predicted average water depth in the fields after rain. h current To warn of water levels, V upstream This refers to the upstream water volume. A river The average cross-sectional area of ​​the river channel. h safe For the safe water level of the river, h river_current This refers to the current real-time water level of the river channel; when V pre No drainage is required when ≤ 0. V pre When the value is >0, drainage is required and the drainage volume is: V pre .

9. The method for forecasting and early warning of flooding in polder areas according to claim 1, characterized in that, In step six, the multi-objective optimization algorithm is any one of the following: NSGA-II non-dominated sorting genetic algorithm, multi-objective particle swarm optimization algorithm, or simulated annealing-swarm intelligence hybrid algorithm; the constraints include the rated flow rate of a single pump in each pumping station, the upper limit of the number of units that can be started, and the total amount of pre-drainage; the optimization objective function is: In the formula, f 1 represents the total operating energy consumption target for the pump station group. C i For the first i Energy consumption coefficient per unit time of the pump station unit. P i In start / stop state. T i Runtime; f 2 is the target for crop yield reduction, based on the predicted field water level. h pred With warning water level h current The difference and duration are used to determine the result.

10. A flood forecasting and early warning system for polder areas, characterized in that, include: The data acquisition module is used to acquire multi-source weather forecast data and real-time field monitoring data. The multi-source weather forecast data includes the rainfall amount and duration for the next few days, and the real-time field monitoring data includes field water level and ditch water level. The multi-source weather forecast data and the real-time field monitoring data are weighted and fused to obtain a fused forecast dataset. The water level prediction module is used to predict future field water levels based on rainfall, rainfall duration, and field water levels in the fused forecast dataset and real-time field monitoring data using a pre-trained long short-term memory network model; and to predict future ditch water levels based on rainfall, rainfall duration, and ditch water levels and drainage flow in the fused forecast dataset and real-time field monitoring data using a pre-trained random forest model. The early warning and classification module is used to determine the current growth stage based on the crop transplanting date, calculate the current warning water level based on the plant height growth model, determine the flooding depth range based on the ratio of the future field water level to the current warning water level, and determine the early warning level based on the number of consecutive flooding days within the flooding depth range and by referring to the preset yield reduction classification matrix of growth stage-flooding depth-consecutive flooding days. The intelligent scheduling module is used to calculate the pre-discharge volume according to the pre-discharge volume formula when the warning level exceeds the preset threshold. With the optimization objectives of minimizing the total operating energy consumption of the pump station group and minimizing crop yield loss, and with the constraints of the rated flow of each pump station, the upper limit of the number of units that can be started, and the required pre-drainage volume, a multi-objective optimization algorithm is used to solve the problem and generate a joint scheduling scheme that includes the start-up and shutdown sequence of each pump station, the number of units in operation, and the running time, and then issues it for execution. The front-end visualization platform is used for the visualization and interactive operation of early warning information, risk assessment results, and scheduling plans. The backend service module is used for data management, business logic processing, and API interface services.

Images