Estuary water source water intake salt tide invasion real-time prediction method and system
Through the physical mechanism of seawater intrusion in estuaries and the Bi-LSTM deep learning model, the timeliness and accuracy issues of saltwater intrusion at water intakes in estuaries were solved, and fast and accurate saltwater intrusion prediction and warning were achieved.
Patent Information
- Application Number
- CN202510909799.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-02
- Publication Date
- 2025-10-21
AI Technical Summary
Traditional methods for predicting saltwater intrusion at estuary water intakes have the disadvantages of poor timeliness, high cost, large model calculation volume, time-consuming prediction, and poor real-time prediction and early warning capabilities.
Based on the physical mechanism of seawater intrusion in estuaries, data from meteorological and hydrological monitoring stations are used to make real-time predictions through the Bi-LSTM deep learning model, determine the physical factors affecting chloride concentrations, and issue early warnings based on the forecast factors.
It has achieved fast and accurate saltwater intrusion prediction and warning, with prediction accuracy improved by more than 20%. The calculation time is only 1.25% of the traditional method. It has risk analysis function and improves decision-making level.
Smart Images
Figure CN120823896A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of urban water supply monitoring, and in particular to a method and system for real-time prediction of saltwater intrusion at a water intake at an estuary water source. Background Art
[0002] Estuarine regions are often economically developed, densely populated, and possess enormous development potential. The supply of freshwater resources is crucial for ensuring high-quality regional economic and social development. Estuarine water sources are crucial suppliers of freshwater to cities. However, due to factors such as tides, runoff, and seawater intrusion, these areas are susceptible to saltwater intrusion. This causes significant fluctuations in chloride concentrations at water intakes, often exceeding recommended critical values, severely impacting urban water supply security.
[0003] The diagnosis of saltwater intrusion at estuarine water source intakes is generally based on chloride concentrations at the intake. Traditional chloride concentration monitoring methods rely primarily on on-site sampling and laboratory analysis, which suffer from poor timeliness, high costs, and an inability to provide real-time predictions. Furthermore, traditional mathematical models of saltwater intrusion are based on water diffusion theory and require numerical calculations to solve numerous equations. This results in high computational complexity, time-consuming simulations, and limited real-time prediction capabilities.
[0004] Therefore, the traditional method of saltwater intrusion at water intakes has the problems of poor timeliness, high cost, large model calculation volume, long prediction time, and poor real-time prediction and early warning capabilities. Summary of the Invention
[0005] The purpose of the present invention is to provide a real-time prediction method and system for saltwater intrusion at water intakes of estuary water sources, which can solve the technical problems of traditional saltwater intrusion methods at water intakes, such as poor timeliness, high cost, large model calculation amount, long prediction time, and poor real-time prediction and early warning capabilities.
[0006] To achieve the above object, the present invention provides the following technical solutions: The present invention designs a real-time prediction method for saltwater intrusion at a water intake of an estuary water source, comprising the following steps: By analyzing the physical mechanism of seawater intrusion at the estuary, we determined the physical factors that affect chloride concentrations at the water intake of the estuary source, including the intensity of seawater intrusion at the water intake, the wind speed in the water area, the tidal intensity of the estuary, and the dilution capacity of the river flowing into the sea. We then converted these physical factors into specific monitoring data. Collect monitoring data and the chloride concentration of the day; The monitoring data and the chloride concentration of the day are processed to form a deep learning dataset; Input the deep learning dataset into the Bi-LSTM deep learning model for training, and adjust the parameters of the Bi-LSTM deep learning model based on the performance of the validation set; Determine the prediction factors of saltwater intrusion at the water intake of the estuary water source, which are hourly chloride concentration data, hourly wind direction and wind speed data of nearby meteorological stations, hourly tide level forecast data of the open sea, middle reaches of the estuary, and the estuary entrance waters, and hourly flow data; The prediction factors are input into the trained Bi-LSTM deep learning model to obtain the real-time prediction results of chloride concentration at the water intake; Based on the real-time prediction data of chloride concentration at the water intake, the intensity of saltwater intrusion at the water intake is judged and an early warning is issued depending on the situation.
[0007] As a preferred option, physical factors are converted into specific monitoring data. The specific method is to express the intensity of seawater intrusion at the water intake in the early stage by the chloride concentration at the water intake of the estuary water source in the meteorological and hydrological monitoring station network, the wind force in the water area is expressed by wind direction and wind speed, the tidal intensity at the estuary is expressed by the estuary tide level, and the dilution capacity of the river entering the sea is expressed by the early flow into the sea.
[0008] As a preferred option, the monitoring data collected are the chloride concentration in the previous 3 days, 2 days and 1 day, the wind direction and speed in the water area, the tide level at the estuary and the flow rate of the main rivers flowing into the sea in the previous 4 days. The duration of these monitoring data is more than 30 days, and the monitoring data time interval is 1 hour.
[0009] As a preferred solution, the steps for forming a deep learning dataset from monitoring data and the chloride concentration of the day are as follows: The monitoring data are processed with gap filling method, and the gap filling method is cubic spline interpolation method; The monitoring data after gap filling processing are combined to form an impact factor vector and a target factor vector; Normalize the influencing factor vector and the target factor vector. After normalization, the value of each vector is between -1 and 1. The normalized influencing factor vector and target factor vector are divided into training set and validation set according to the proportion to form a deep learning dataset.
[0010] As the preferred solution, the Bi-LSTM deep learning model is a model with bidirectional long short-term memory classification. The model includes a forward LSTM model and a backward LSTM model. The two work in parallel to process data features input in opposite directions. It not only utilizes the properties of the current time step, but also utilizes the properties of previous and subsequent time steps, fully considering the periodicity of hydrological factors in estuary waters, making the prediction more accurate.
[0011] Furthermore, when using the Bi-LSTM deep learning model for training, the samples are shuffled to reduce the risk of overfitting and improve the generalization ability and model accuracy on unknown data.
[0012] As a preferred solution, the steps for inputting the prediction factors into the trained Bi-LSTM deep learning model for prediction are as follows: Collect hourly chloride concentration data at the water intake for the previous three days, two days, and one day; Collect hourly wind direction and speed data from nearby weather stations for the next day; Collect hourly tide level forecast data for the open sea, middle reaches of the estuary, and the estuary entrance waters for the next day; Collect hourly flow data for the first four days of the main rivers flowing into the sea at the estuary; The above data are normalized to form a prediction factor vector, which is then input into the trained Bi-LSTM deep learning model to predict the chloride concentration at the water intake hourly for the next day in real time.
[0013] As a preferred solution, the process of judging the intensity of saltwater intrusion at the water intake based on the real-time prediction results of chloride concentration at the water intake and issuing an early warning as appropriate is as follows: The hourly chloride concentration at the water intake for the next day is represented by y; When 200≤y<250mg / L, it is a weak saltwater intrusion, an orange warning is issued, and pump gate facilities are opened to take water; When y≥250mg / L, it is a strong saltwater intrusion, a red alert is issued, and the pump gate facilities are closed and water intake is stopped; When y drops from above 250mg / L to y<250mg / L, it is a transition from strong saltwater intrusion to weak saltwater intrusion, a blue warning is issued, and the pump gate facilities are opened to take water; When y<200mg / L, there is no saltwater intrusion, no warning is issued, and the pump and gate facilities draw water normally.
[0014] The present invention also designs a real-time prediction system for saltwater intrusion at the water intake of an estuary water source, including the following modules: The monitoring data module is used to analyze the physical mechanism of seawater intrusion at the estuary and determine the physical factors that affect the chloride concentration at the water intake of the estuary water source, including the intensity of seawater intrusion at the water intake, the wind speed in the water area, the tidal intensity of the estuary, and the dilution capacity of the river entering the sea; and convert the physical factors into specific monitoring data; Data acquisition module, used to collect monitoring data and the chloride concentration of the day; The deep learning dataset module is used to process the monitoring data and the chloride concentration of the day to form a deep learning dataset; The model training module is used to input the deep learning dataset into the Bi-LSTM deep learning model for training and adjust the parameters of the Bi-LSTM deep learning model based on the performance of the validation set; The prediction factor module is used to determine the prediction factors of saltwater intrusion at the water intake of the estuary water source, which include hourly chloride concentration data, hourly wind direction and wind speed data of nearby meteorological stations, hourly tide level forecast data of the open sea, middle reaches of the estuary, and waters at the entrance of the estuary, and hourly flow data; The prediction module is used to input the prediction factors into the trained Bi-LSTM deep learning model for prediction, and obtain the real-time prediction results of the chloride concentration at the water intake; The early warning module is used to judge the intensity of saltwater intrusion at the water intake based on the real-time prediction data of chloride concentration at the water intake and issue an early warning as appropriate.
[0015] Beneficial effects of the present invention: First, based on the study of the physical mechanism of seawater intrusion in estuaries, the present invention determined that the main physical factors affecting the chloride concentration at the water intake of the estuary water source are the intensity of seawater intrusion in the early stage of the water intake, the strength of the wind in the water area, the tidal intensity of the estuary and the dilution capacity of the river entering the sea. The physical mechanism is clear.
[0016] Secondly, the present invention clarifies that the intensity of seawater intrusion in the early stage at the water intake can be represented by the chloride concentration 3 days, 2 days and 1 day before the water intake, the wind force in the water area can be represented by the wind direction and wind speed of the adjacent meteorological station, the tidal intensity in the estuary can be represented by the tide levels in the open sea, the middle reaches of the estuary and the estuary import waters, and the dilution capacity of the river entering the sea can be represented by the flow rate of the main river entering the sea at the estuary 4 days before. The physical factors are comprehensive and the data are easy to obtain or collect, which effectively reduces the difficulty of prediction and early warning.
[0017] Third, the present invention designs a deep learning model Bi-LSTM with bidirectional long-short-term memory classification. This model fully considers the periodicity of hydrological factors in estuarine waters and can utilize not only the attributes of the current time step, but also the attributes of previous and subsequent time steps, effectively improving the accuracy of forecasts and warnings. Compared with the traditional estuarine salt tide forecasting mathematical model prediction method, the forecast and warning accuracy is improved by more than 20%, and the calculation time is only 1.25% of the traditional estuarine salt tide forecasting mathematical model, which greatly improves the forecast and warning speed.
[0018] Fourthly, the present invention can predict the hourly chloride concentration y at the water intake of the estuary water source for the next day, and also provides the confidence interval of the predicted conclusion, which has the function of risk analysis and effectively improves the decision-making level.
[0019] Fifth, the present invention provides the standards for blue warning, orange warning and red warning according to the hourly chloride concentration of the water intake of the estuary water source in the next day, has the function of early warning, and is highly practical. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] Figure 1 is a flow chart of the real-time prediction and early warning method of the present invention; Figure 2 It is a schematic diagram of real-time prediction and early warning result data of the present invention. DETAILED DESCRIPTION
[0021] To make the technical problems solved by the present invention, the technical solutions adopted, and the technical effects achieved more clearly, the technical solutions of the present invention are further described below with reference to the accompanying drawings and through specific embodiments. It should be understood that the specific embodiments described herein are intended only to illustrate the present invention and are not intended to limit the present invention. It should also be noted that, for ease of description, the drawings only show portions relevant to the present invention, not all of them.
[0022] In the description of the present invention, it should be noted that the terms "center," "up," "down," "left," "right," "vertical," "horizontal," "inside," and "outside" and the like, indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings and are intended only to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or components referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limitations on the present invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. The terms "first position" and "second position" refer to two different positions.
[0023] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood broadly. For example, they may refer to fixed or detachable connections; mechanical or electrical connections; direct or indirect connections through an intermediary; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention.
[0024] Traditional methods for predicting saltwater intrusion at water intakes have problems such as poor timeliness, high cost, large model calculation volume, long prediction time consumption, and poor real-time prediction and early warning capabilities. Therefore, it is of great significance to develop a method for predicting saltwater intrusion at water intakes of estuaries and water sources that can quickly, accurately, and predict in real time. The purpose of the present invention is to provide a real-time prediction method and system for saltwater intrusion at water intakes of estuaries and water sources. Based on the study of the physical mechanism of seawater intrusion at estuaries, the method uses a deep learning model to realize the real-time prediction of chloride concentration at water intakes of estuaries and water sources through easily accessible or collected monitoring data from meteorological and hydrological monitoring stations. Based on this, relevant early warnings of saltwater intrusion at water intakes of estuaries and water sources are issued, thereby effectively improving the forecast accuracy and early warning capabilities of saltwater intrusion at water intakes of estuaries and water sources.
[0025] To achieve the above object, the present invention provides the following technical solutions: The present invention provides a method and system for real-time prediction of saltwater intrusion at a water intake of an estuary water source, comprising the following steps: Step 1: By analyzing the physical mechanism of seawater intrusion at the estuary, determine the main physical factors affecting the chloride concentration at the water intake of the estuary water source, namely the initial seawater intrusion intensity at the water intake, the wind speed in the water area, the tidal strength of the estuary, and the dilution capacity of the river entering the sea; and convert these main physical factors into specific monitoring data. The initial seawater intrusion intensity at the water intake is represented by the initial chloride concentration at the water intake of the estuary water source in the meteorological and hydrological monitoring station network, the wind speed in the water area is represented by wind direction and wind speed, the tidal strength of the estuary is represented by the tidal level at the estuary, and the dilution capacity of the river entering the sea is represented by the initial discharge into the sea. Step 2: Collect monitoring data such as the previous chloride concentration at the water intake of the estuary water source, wind direction and speed in the water area, tidal level at the estuary, and previous flow rate of major rivers flowing into the sea from the meteorological and hydrological monitoring station network, and also collect the chloride concentration of the day; Step 3: Process the monitoring data and the chloride concentration of the day to form a deep learning dataset; Step 4: Input the deep learning dataset into the Bi-LSTM deep learning model for training. Based on the performance of the validation set, adjust the parameters of the Bi-LSTM deep learning model. The Bi-LSTM deep learning model is an existing technology. Step 5: Determine the prediction factors for saltwater intrusion at the water intake of the estuary water source, which are hourly chloride concentration data, hourly wind direction and wind speed data of nearby meteorological stations, hourly tide level forecast data of the open sea, the middle reaches of the estuary, and the estuary entrance waters, and hourly flow data; The prediction factors are input into the trained Bi-LSTM deep learning model for prediction to obtain the real-time prediction results of chloride concentration at the water intake.
[0026] Step 6: Based on the real-time prediction data of chloride concentration at the water intake, determine the intensity of saltwater intrusion at the water intake and issue an early warning as appropriate.
[0027] In step 1, based on the study of the physical mechanism of seawater intrusion in estuaries, the present invention considers 9 main physical factors that affect the chloride concentration at the water intake of the estuary water source, and converts these 9 main physical factors into specific monitoring data, namely: the intensity of seawater intrusion at the water intake in the early stage is represented by the chloride concentrations X1, X2, and X3 at the water intake 3 days before, 2 days before, and 1 day before; the wind force in the water area is represented by the wind direction of the adjacent weather station ( o ) X4 and wind speed (m / s) X5; the tidal intensity of the estuary is represented by the tide levels X6, X7, and X8 in the offshore, middle reaches of the estuary, and the estuary entrance waters; the dilution capacity of the river entering the sea is represented by the flow rate X9 of the main river entering the sea in the previous four days.
[0028] In the step 2, the monitoring data collected are the chloride concentrations X1-X3 of the previous 3 days, the previous 2 days, and the previous day, the wind direction X4 and wind speed X5 of the water area, the tide level X6-X8 of the estuary, and the flow rate X9 of the main rivers entering the sea in the previous 4 days. The duration of these monitoring data is preferably more than 30 days, and the monitoring data time interval is 1 hour, taking into account the periodicity of the tidal cycle of salt water intrusion in the estuary; at the same time, the chloride concentration Y of the day is collected.
[0029] In step 3, the steps for forming a deep learning dataset from the monitoring data and the chloride concentration of the day are as follows: (1) Perform gap filling processing on the monitoring data. The gap filling processing method is the cubic spline interpolation method, which is a mathematical method for constructing a smooth curve through a series of shape value points and belongs to the existing technology; (2) Combining the monitoring data after gap filling to form the influencing factor vector and target factor vector; (3) Normalize the influencing factor vector and the target factor vector. After normalization, the value of each vector is between -1 and 1. (4) The normalized influencing factor vector and target factor vector are divided into a training set and a validation set in a ratio of 8:2 to form a deep learning dataset.
[0030] In the step 4, the Bi-LSTM deep learning model used is a model with bidirectional long short-term memory classification. The model includes a forward LSTM model and a backward LSTM model. The two models work in parallel to process data features input in opposite directions. Not only can the attributes of the current time step be utilized, but also the attributes of the previous and subsequent time steps can be utilized, fully considering the periodicity of hydrological factors in the estuary waters, making the prediction more accurate.
[0031] In step 4, when the Bi-LSTM deep learning model is used for training, the samples are shuffled to reduce the risk of overfitting, improve the generalization ability of the model on unknown data, and improve the model accuracy.
[0032] In step 5, the steps of inputting the prediction factors into the trained Bi-LSTM deep learning model for prediction are as follows: (1) Collect hourly chloride concentration data x1, x2, and x3 at the water intake three days before, two days before, and one day before; “hourly” means collecting data once every hour; (2) Collect hourly wind direction and wind speed data x4 and x5 from nearby weather stations for the next day; (3) Collect hourly tide level forecast data x6, x7, and x8 for the open sea, the middle reaches of the estuary, and the estuary entrance waters for the next day; (4) Collect hourly flow data x9 for the first four days of the main rivers flowing into the sea at the estuary; (5) The above data are normalized to form a prediction factor vector, which is then input into the trained Bi-LSTM deep learning model to predict the hourly chloride concentration y at the water intake in real time for the next day.
[0033] In step five, the prediction factor vector is input into the trained Bi-LSTM deep learning model for prediction. This model can predict the hourly chloride concentration y at the water intake for the next day. It also provides a confidence interval for the predicted conclusion, providing risk analysis capabilities. A confidence interval is a form of parameter estimation. Using samples drawn from a population, an appropriate interval is constructed based on certain accuracy and precision requirements to estimate the range of the true value of the distribution parameter (or function of the parameter) of the population. A distance on the number axis, or a data interval, is used to represent the possible range of the population parameter. This distance or data interval is called the confidence interval of the interval estimate.
[0034] In step 6, the intensity of saltwater intrusion at the water intake is determined based on the predicted hourly chloride concentration y at the water intake of the estuary source for the next day, and an early warning is issued as appropriate: When 200≤y<250mg / L, it is a weak saltwater intrusion, and an orange warning is issued to the water source management department, and the pump and gate facilities are opened to take water as soon as possible; When y≥250mg / L, it is a strong saltwater intrusion, and a red alert is issued to the water source management department, and the pump gate facilities are closed and water extraction is stopped; When y drops from above 250mg / L to y<250mg / L, it is a transition from strong saltwater intrusion to weak saltwater intrusion, and a blue warning is issued to the water source management department, and the pump gate facilities are opened to take water; When y<200mg / L, there is no saltwater intrusion, no warning is issued, and the pump and gate facilities draw water normally.
[0035] It should be understood that the specific order or hierarchy of steps in the processes disclosed herein are examples of exemplary methods. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the processes may be rearranged without departing from the scope of the present disclosure. The accompanying method claims present elements of the various steps in an exemplary order and are not intended to be limited to the specific order or hierarchy described.
[0036] The following further illustrates the specific embodiments of the present invention in conjunction with the accompanying drawings. The present invention is not limited to the following embodiments.
[0037] Example: This embodiment provides a specific structure of a method and system for real-time prediction of saltwater intrusion at a water source in an estuary. Figure 1 As shown, a real-time prediction method and system for saltwater intrusion at a water intake of an estuary water source includes the following steps: Step 1: Analyze the physical mechanism of seawater intrusion at the estuary to determine the main physical factors affecting chloride concentrations at the estuary water source intake, including the intensity of seawater intrusion at the water intake, wind speed, tidal intensity at the estuary, and dilution capacity of the river flowing into the sea. These main physical factors are then converted into specific monitoring data. Step 2: Collect monitoring data such as the previous chloride concentration at the water intake of the estuary water source, wind direction and speed in the water area, tidal level at the estuary, and previous flow rate of major rivers flowing into the sea from the meteorological and hydrological monitoring station network, and also collect the chloride concentration of the day; Step 3: Process the monitoring data and the chloride concentration of the day to form a deep learning dataset; Step 4: Input the deep learning dataset into the bidirectional long short-term memory classification model, that is, the Bi-LSTM deep learning model for training. Based on the performance of the validation set, adjust the parameters of the Bi-LSTM deep learning model. Step 5: Input the prediction factors into the trained Bi-LSTM deep learning model for prediction to obtain the real-time prediction results of chloride concentration at the water intake; Step 6: Based on the real-time prediction data of chloride concentration at the water intake, determine the intensity of saltwater intrusion at the water intake and issue an early warning as appropriate.
[0038] like Figure 1In step 1, based on the study of the physical mechanism of seawater intrusion in estuaries, the present invention considers 9 main physical factors that affect the chloride concentration at the water intake of the estuary water source, and converts these 9 main physical factors into specific monitoring data, namely: the intensity of seawater intrusion at the water intake in the early stage is represented by the chloride concentrations X1, X2, and X3 at the water intake 3 days before, 2 days before, and 1 day before; the wind force in the water area is represented by the wind direction of the adjacent weather station ( o ) X4 and wind speed (m / s) X5; the tidal intensity of the estuary is represented by the tide levels X6, X7, and X8 in the offshore, middle reaches of the estuary, and the estuary entrance waters; the dilution capacity of the river entering the sea is represented by the flow rate X9 of the main river entering the sea in the previous four days.
[0039] like Figure 1 In the step 2, the monitoring data collected are the chloride concentrations X1-X3 of the previous 3 days, the previous 2 days, and the previous day, the wind direction X4 and wind speed X5 of the water area, the tide level X6-X8 of the estuary, and the flow rate X9 of the main rivers entering the sea in the previous 4 days. The duration of these monitoring data is preferably more than 30 days, and the monitoring data time interval is 1 hour, taking into account the periodicity of the tidal cycle of salt water intrusion in the estuary; at the same time, the chloride concentration Y of the day is collected.
[0040] like Figure 1 In step 3, the steps of forming a deep learning dataset from the monitoring data and the chloride concentration of the day are as follows: (1) The monitoring data are processed by filling in the gaps. The method of filling in the gaps is cubic spline interpolation, which is a mathematical method that constructs a smooth curve through a series of shape value points; (2) Combining the monitoring data after gap filling to form the influencing factor vector and target factor vector; (3) Normalize the influencing factor vector and the target factor vector. After normalization, the value of each vector is between -1 and 1. (4) The normalized influencing factor vector and target factor vector are divided into a training set and a validation set in a ratio of 8:2 to form a deep learning dataset.
[0041] like Figure 1 In the step 4, the Bi-LSTM deep learning model used is a model with bidirectional long short-term memory classification. The model includes a forward LSTM model and a backward LSTM model. The two models work in parallel to process data features input in opposite directions. Not only can the attributes of the current time step be utilized, but also the attributes of the previous and subsequent time steps, fully considering the periodicity of hydrological factors in the estuary waters, making the prediction more accurate.
[0042] like Figure 1In the fourth step, when using the Bi-LSTM deep learning model for training, the sample is shuffled, which can reduce the risk of overfitting, improve the generalization ability of the model on unknown data, and improve the model accuracy.
[0043] like Figure 1 In step 5, the prediction factors include hourly chloride concentration data, hourly tide level forecast data for the open sea, the middle reaches of the estuary, and the estuary entrance waters, and hourly flow data. The steps of inputting the prediction factors into the trained Bi-LSTM deep learning model for prediction are as follows: (1) Collect hourly chloride concentration data x1, x2, and x3 at the water intake three days before, two days before, and one day before; (2) Collect hourly wind direction and wind speed data x4 and x5 from nearby weather stations for the next day; (3) Collect hourly tide level forecast data x6, x7, and x8 for the open sea, the middle reaches of the estuary, and the estuary entrance waters for the next day; (4) Collect hourly flow data x9 for the first four days of the main rivers flowing into the sea at the estuary; (5) The above data are normalized to form a prediction factor vector, which is input into the trained Bi-LSTM deep learning model to predict the hourly chloride concentration y at the water intake in real time for the next day.
[0044] like Figure 1 In step five, the prediction factor vector is input into the trained Bi-LSTM deep learning model for prediction, which can predict the hourly chloride concentration y at the water intake for the next day. At the same time, the confidence interval of the predicted conclusion is also given, which has the function of risk analysis.
[0045] like Figure 1 In step 6, the intensity of saltwater intrusion at the water intake is determined based on the predicted hourly chloride concentration y at the water intake of the estuary source for the next day, and an early warning is issued depending on the situation: When 200≤y<250mg / L, it is a weak saltwater invasion, and an orange warning is issued to the water source management department, and the pump and gate facilities are opened to draw water as soon as possible; when y≥250mg / L, it is a strong saltwater invasion, and a red warning is issued to the water source management department, and the pump and gate facilities are closed to stop drawing water; when y drops from above 250mg / L to y<250mg / L, it is a transition from strong saltwater invasion to weak saltwater invasion, and a blue warning is issued to the water source management department, and the pump and gate facilities are opened to draw water; when y<200mg / L, there is no saltwater invasion, no warning is issued, and the pump and gate facilities draw water normally.
[0046] like Figure 2As shown in the figure, taking the Qingcaosha water source, the largest salt-avoiding freshwater source at a certain estuary, as an example, 45 days of monitoring data were collected for 9 factors, including the chloride concentrations X1-X3 in the water area of the water source's water intake 3 days, 2 days, and 1 day before, the chloride concentration Y on the day, the wind direction X4 and wind speed X5 at the Sheshan meteorological station at the estuary, the tide levels X6-X8 at the Gongqingwei station in the outer sea of the estuary, the Lingdian Port station in the middle reaches, and the Xuliujing station at the inlet, and the flow X9 in the first 4 days at the Datong station on the Yangtze River. After gap filling and normalization processing, the influencing factor vector and the target factor vector were divided into training and validation sets in a ratio of 8:2, forming a deep learning dataset for chloride concentration prediction and early warning in the water area of the Qingcaosha water source.
[0047] Based on this data, a Bi-LSTM deep learning model was constructed and trained for chloride concentration prediction and early warning at the Qingcaosha water source inlet at a certain estuary. The validation results on the validation set showed a correlation coefficient of 0.91, a MAE of 30.07 mg / L, and an RMSE of 38.42 mg / L. MAE refers to mean absolute error, and RMSE refers to root mean square error.
[0048] In order to verify the established Bi-LSTM deep learning model for chloride concentration prediction and early warning in the water area of the Qingcaosha water source in a certain estuary, the traditional estuary salt tide forecasting mathematical model was used for simulation under the same prediction conditions and verification period. The simulation results showed that the correlation coefficient was 0.82, the MAE was 60.32 mg / L, and the RMSE was 72.46 mg / L.
[0049] Comparison of the prediction results from the two models shows that the Bi-LSTM deep learning model's computation time is only 1.25% of that of traditional mathematical models for estuarine saltwater tide forecasting, significantly improving forecasting speed. Prediction accuracy is also significantly improved by over 20% compared to traditional mathematical models for estuarine saltwater tide forecasting, demonstrating significant improvements. Compared to traditional mathematical models for estuarine saltwater tide forecasting, the Bi-LSTM deep learning model offers significant improvements, demonstrating greater accuracy and practicality.
[0050] Based on the verification of the Bi-LSTM deep learning model for chloride concentration prediction and early warning in the water area of the Qingcaosha water source at a certain estuary, the hourly chloride concentration data x1, x2, and x3 for the water area of the Qingcaosha water source were collected for the previous 3 days, 2 days, and 1 day, the hourly wind direction and wind speed forecast data x4 and x5 for the next day were collected from the meteorological station, the hourly tide level forecast data x6, x7, and x8 for the next day were collected, and the hourly flow rate data x9 for the previous 4 days were collected. The above data were normalized to form a prediction factor vector, which was input into the trained Bi-LSTM deep learning model to predict the hourly chloride concentration y in the water area of the Qingcaosha water source at the water source for the next day in real time. At the same time, the confidence intervals of the forecast conclusions were given, with three confidence intervals of 85%, 90%, and 95% respectively. The predicted data basically fell within the 85% confidence interval, as shown in the attached figure. Figure 2 shown.
[0051] Based on the real-time forecast data, i.e. the hourly chloride concentration y in the water area of a certain estuary water source intake for the next day, the intensity of saltwater intrusion at the Qingcaosha water source intake at the Yangtze River estuary is determined, and an early warning is issued depending on the situation: no early warning, blue early warning, orange early warning, or red early warning. Figure 2 When 200≤y<250mg / L, it is a weak saltwater intrusion, and an orange warning is issued to the water source management department, and the pump gate facilities are opened to take water as soon as possible; when y≥250mg / L, it is a strong saltwater intrusion, and a red warning is issued to the water source management department, and the pump gate facilities are closed to stop taking water; when y drops from above 250mg / L to y<250mg / L, it is a transition from strong saltwater intrusion to weak saltwater intrusion, and a blue warning is issued to the water source management department, and the pump gate facilities are opened to take water; when y<200mg / L, there is no saltwater intrusion, no warning is issued, and the pump gate facilities can take water normally.
[0052] The present invention has the following beneficial effects: First, based on the study of the physical mechanism of seawater intrusion in estuaries, the present invention determined that the main physical factors affecting the chloride concentration at the water intake of the estuary water source are the intensity of seawater intrusion in the early stage of the water intake, the strength of the wind in the water area, the tidal intensity of the estuary and the dilution capacity of the river entering the sea. The physical mechanism is clear.
[0053] Secondly, the present invention clarifies that the intensity of seawater intrusion in the early stage at the water intake can be represented by the chloride concentration 3 days, 2 days and 1 day before the water intake, the wind force in the water area can be represented by the wind direction and wind speed of the adjacent meteorological station, the tidal intensity in the estuary can be represented by the tide levels in the open sea, the middle reaches of the estuary and the estuary import waters, and the dilution capacity of the river entering the sea can be represented by the flow rate of the main river entering the sea at the estuary 4 days before. The physical factors are comprehensive and the data are easy to obtain or collect, which effectively reduces the difficulty of prediction and early warning.
[0054] Third, the present invention has designed a deep learning model with bidirectional long-short-term memory (Bi-LSTM) classification. This model fully considers the periodicity of hydrological factors in estuary waters and can utilize the attributes of not only the current time step but also those of previous and subsequent time steps, effectively improving the accuracy of forecasts and warnings. Compared with traditional mathematical model prediction methods, the accuracy of forecasts and warnings has increased by more than 20%, and the speed of forecasts and warnings has been greatly improved.
[0055] Fourthly, the present invention can predict the hourly chloride concentration y at the water intake of the estuary water source for the next day, and also provides the confidence interval of the predicted conclusion, which has the function of risk analysis and effectively improves the decision-making level.
[0056] Fifth, the present invention provides the standards for blue warning, orange warning and red warning according to the hourly chloride concentration of the water intake of the estuary water source in the next day, has the function of early warning, and is highly practical.
[0057] Although the present invention is disclosed above in terms of preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art may make possible changes and modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be based on the scope defined by the claims of the present invention.
Claims
1. A real-time prediction method for saltwater intrusion at a water intake of an estuary, characterized by: The following steps are included: By analyzing the physical mechanism of seawater intrusion at the estuary, we determined the physical factors that affect chloride concentrations at the water intake of the estuary source, including the intensity of seawater intrusion at the water intake, the wind speed in the water area, the tidal intensity of the estuary, and the dilution capacity of the river flowing into the sea. We then converted these physical factors into specific monitoring data. Collect monitoring data and the chloride concentration of the day; The monitoring data and the chloride concentration of the day are processed to form a deep learning dataset; Input the deep learning dataset into the Bi-LSTM deep learning model for training, and adjust the parameters of the Bi-LSTM deep learning model based on the performance of the validation set; Determine the prediction factors of saltwater intrusion at the water intake of the estuary water source, which are hourly chloride concentration data, hourly wind direction and wind speed data of nearby meteorological stations, hourly tide level forecast data of the open sea, middle reaches of the estuary, and the estuary entrance waters, and hourly flow data; The prediction factors are input into the trained Bi-LSTM deep learning model to obtain the real-time prediction results of chloride concentration at the water intake; Based on the real-time prediction data of chloride concentration at the water intake, the intensity of saltwater intrusion at the water intake is judged and an early warning is issued depending on the situation.
2. The method for real-time prediction of saltwater intrusion at an estuary water source intake according to claim 1, characterized in that: Physical factors are converted into specific monitoring data. The specific method is to express the intensity of seawater intrusion at the water intake in the early stage by the chloride concentration at the water intake of the estuary water source in the meteorological and hydrological monitoring station network, the wind force in the water area is expressed by wind direction and wind speed, the tidal intensity at the estuary is expressed by the estuary tide level, and the dilution capacity of the river entering the sea is expressed by the early flow into the sea.
3. The method for real-time prediction of saltwater intrusion at an estuary water source water intake according to claim 2, characterized in that: The monitoring data collected include chloride concentrations in the previous three days, two days, and one day, wind direction and speed in the waters, tide levels at the estuary, and flow rates of major rivers flowing into the sea in the previous four days. The duration of these monitoring data is more than 30 days, and the monitoring data time interval is 1 hour.
4. The method for real-time prediction of saltwater intrusion at an estuary water source intake according to claim 3, characterized in that: The steps to form a deep learning dataset from monitoring data and the chloride concentration of the day are as follows: The monitoring data are processed with gap filling method, and the gap filling method is cubic spline interpolation method; The monitoring data after gap filling processing are combined to form an impact factor vector and a target factor vector; Normalize the influencing factor vector and the target factor vector. After normalization, the value of each vector is between -1 and 1. The normalized influencing factor vector and target factor vector are divided into training set and validation set according to the proportion to form a deep learning dataset.
5. A real-time prediction method for saltwater intrusion at an estuary water source intake according to any one of claims 1 to 4, characterized in that: The Bi-LSTM deep learning model is a model with bidirectional long short-term memory classification. The model includes a forward LSTM model and a backward LSTM model. The two work in parallel to process data features input in opposite directions. It not only utilizes the properties of the current time step, but also the properties of previous and subsequent time steps, fully considering the periodicity of hydrological factors in estuary waters to make predictions more accurate.
6. The method for real-time prediction of saltwater intrusion at an estuary water source water intake according to claim 5, characterized in that: When using the Bi-LSTM deep learning model for training, the samples are shuffled to reduce the risk of overfitting and improve the generalization ability and model accuracy on unknown data.
7. The method for real-time prediction of saltwater intrusion at an estuary water source intake according to claim 6, characterized in that: The steps to input the prediction factors into the trained Bi-LSTM deep learning model for prediction are as follows: Collect hourly chloride concentration data at the water intake for the previous three days, two days, and one day; Collect hourly wind direction and speed data from nearby weather stations for the next day; Collect hourly tide level forecast data for the open sea, middle reaches of the estuary, and the estuary entrance waters for the next day; Collect hourly flow data for the first four days of the main rivers flowing into the sea at the estuary; The above data are normalized to form a prediction factor vector, which is then input into the trained Bi-LSTM deep learning model to predict the chloride concentration at the water intake hourly for the next day in real time.
8. The method for real-time prediction of saltwater intrusion at an estuary water source intake according to claim 7, characterized in that: The process of judging the intensity of saltwater intrusion at the water intake based on the real-time prediction results of chloride concentration at the water intake and issuing early warnings as appropriate is as follows: The hourly chloride concentration at the water intake for the next day is represented by y; When 200≤y<250mg / L, it is a weak saltwater intrusion, an orange warning is issued, and pump gate facilities are opened to take water; When y≥250mg / L, it is a strong saltwater intrusion, a red alert is issued, and the pump gate facilities are closed and water intake is stopped; When y drops from above 250mg / L to y<250mg / L, it is a transition from strong saltwater intrusion to weak saltwater intrusion, a blue warning is issued, and the pump gate facilities are opened to take water; When y<200mg / L, there is no saltwater intrusion, no warning is issued, and the pump and gate facilities draw water normally.
9. A real-time prediction system for saltwater intrusion at an estuary water source intake, characterized by: Includes the following modules, The monitoring data module is used to analyze the physical mechanism of seawater intrusion at the estuary and determine the physical factors that affect the chloride concentration at the water intake of the estuary water source, including the intensity of seawater intrusion at the water intake, the wind speed in the water area, the tidal intensity of the estuary, and the dilution capacity of the river entering the sea; and convert the physical factors into specific monitoring data; Data acquisition module, used to collect monitoring data and the chloride concentration of the day; The deep learning dataset module is used to process the monitoring data and the chloride concentration of the day to form a deep learning dataset; The model training module is used to input the deep learning dataset into the Bi-LSTM deep learning model for training and adjust the parameters of the Bi-LSTM deep learning model based on the performance of the validation set; The prediction factor module is used to determine the prediction factors of saltwater intrusion at the water intake of the estuary water source, which include hourly chloride concentration data, hourly wind direction and wind speed data of nearby meteorological stations, hourly tide level forecast data of the open sea, middle reaches of the estuary, and waters at the entrance of the estuary, and hourly flow data; The prediction module is used to input the prediction factors into the trained Bi-LSTM deep learning model for prediction and obtain the real-time prediction results of the chloride concentration at the water intake; The early warning module is used to judge the intensity of saltwater intrusion at the water intake based on the real-time prediction data of chloride concentration at the water intake and issue an early warning as appropriate.