Karst spring flow prediction method based on hydro-meteorological data and physical mechanism

Abstract

The application discloses a karst spring flow prediction method based on hydro-meteorological data and physical mechanism, relates to the technical field of hydrological prediction, and comprises the following steps: obtaining historical hydro-meteorological data and karst spring flow data, performing rainfall event correlation processing to obtain rainfall event data; calculating initial water storage based on spring flow response data and groundwater level data, calculating recharge based on rainfall data, and establishing water storage records; calculating candidate release amounts according to the water storage records, calculating total release amounts according to the spring flow response data, distributing the total release amounts according to the proportions of the candidate release amounts, and updating the water storage records; inputting predicted hydro-meteorological data and the updated water storage records into a pre-trained karst spring flow prediction model to obtain karst spring flow prediction data; and the application can reduce the attribution mismatch of residual water storage release amounts and newly added recharge amounts when adjacent rainfall event responses overlap, and improve the karst spring flow prediction accuracy.

Description

Technical Field

[0001] This application relates to the field of hydrological prediction technology, and more specifically, to a method for predicting karst spring flow based on hydrological and meteorological data and physical mechanisms. Background Technology

[0002] Karst spring flow rates reflect the replenishment, storage, and discharge status of karst aquifer systems. Predicting karst spring flow rates can provide a basis for water resource allocation, water supply security, and rainstorm early warning. Existing methods for predicting karst spring flow rates based on hydrological and meteorological data and physical mechanisms typically use rainfall, temperature, evapotranspiration, groundwater level, and historical karst spring flow rates, combined with rainfall infiltration, groundwater storage, and spring discharge processes, to predict future karst spring flow rates.

[0003] Karst aquifers typically contain karst conduits, fissures, and rock matrix. After rainfall enters the karst aquifer, some water reaches the spring through the karst conduits, while some is stored in the fissures and rock matrix and continues to be released after the rainfall ends.

[0004] Existing methods typically construct forecast data based on fixed time intervals or rainfall events. However, when the time interval between adjacent rainfall events is less than the release time of water stored in fractures and the rock matrix, the residual water formed by the previous rainfall event continues to be released during the current rainfall event. In this case, the spring discharge includes both the release of residual water from the previous rainfall event and the new replenishment from the current rainfall event.

[0005] Existing methods struggle to separate the two types of water volume, easily attributing the remaining water release from the previous rainfall event to the current rainfall event. This leads to an overestimation of the replenishment contribution of the current rainfall event and an underestimation of the continued impact of the previous rainfall event, affecting the prediction of remaining water volume, the receding process, and future karst spring flow. Summary of the Invention

[0006] To address the problem of karst spring flow prediction bias caused by the difficulty in distinguishing between the water release process and the new replenishment process when responses to adjacent rainfall events overlap, this application provides the following technical solution: a karst spring flow prediction method based on hydrometeorological data and physical mechanisms, including:

[0007] Acquire historical hydrological and meteorological data and karst spring flow data; perform rainfall event correlation processing on the historical hydrological and meteorological data and karst spring flow data to obtain rainfall event data;

[0008] The initial water storage volume was calculated based on spring flow response data and groundwater level data from rainfall event data.

[0009] The replenishment water volume is calculated based on the rainfall data in the rainfall event data, and a water storage record is established based on the initial water storage volume and the replenishment water volume;

[0010] Calculate the corresponding candidate release amount based on each water storage record, and calculate the total release amount based on the spring flow response data; allocate the total release amount to the corresponding water storage record according to the proportion of each candidate release amount to obtain the allocated release amount, and update the water storage record based on the allocated release amount;

[0011] The acquired predicted hydrological and meteorological data and the updated water storage records are input into the pre-trained karst spring flow prediction model to obtain karst spring flow prediction data.

[0012] Furthermore, methods for obtaining rainfall event data include:

[0013] The sampling interval time unit for rainfall data, groundwater level data, and karst spring flow data is unified, and the least common multiple of the sampling interval is used as the evaluation period length.

[0014] Historical data are divided according to the length of the evaluation period. Rainfall data within each evaluation period are accumulated. Groundwater level data and karst spring flow data are time-weighted averaged. The time-weighted average of karst spring flow data is used as spring flow response data.

[0015] The preset event interval is obtained. If the consecutive rainless duration of adjacent rainfall events is less than the preset event interval, they are classified into the same rainfall event. If the duration is greater than or equal to the preset event interval, they are classified into different rainfall events.

[0016] Only the evaluation period with rainfall greater than zero is assigned a rainfall event identifier;

[0017] Rainfall event data is obtained by arranging the rainfall data, groundwater level data, spring flow response data, and corresponding rainfall event identifiers for each evaluation period in chronological order.

[0018] Furthermore, methods for obtaining preset event intervals include:

[0019] Extract multiple historical rainfall events from rainfall data;

[0020] For each historical rainfall event, the evaluation period is identified where the spring flow response data first exceeds the spring flow response data of the previous evaluation period after the start time of the historical rainfall event.

[0021] The duration between the start time of the historical rainfall event and the start time of the identified evaluation period is defined as the spring flow response delay, and the median of multiple spring flow response delays is defined as the preset event interval.

[0022] Furthermore, methods for calculating the initial water storage volume include:

[0023] Set the prediction start time at the boundary of the evaluation period, and take the largest continuous data segment in which the rainfall is continuously equal to zero and the spring flow response data continuously decreases before the prediction start time as the complete water receding process;

[0024] The time and spring flow response data of each complete drainage process are fitted according to the exponential drainage relationship to obtain the drainage coefficient. The calculation relationship between groundwater level and drainage coefficient is established based on the groundwater level and drainage coefficient of the first evaluation period of each complete drainage process.

[0025] The first evaluation period is designated as the initial water storage calculation period, selecting the complete water recession process whose end time is closest to the prediction start time.

[0026] The drainage coefficient is obtained based on the groundwater level during the initial water storage calculation period and the relationship between the groundwater level and the drainage coefficient. The drainage ratio is calculated based on the drainage coefficient and the evaluation period length. The initial water storage is obtained by dividing the product of the spring flow response data during the initial water storage calculation period and the evaluation period length by the drainage ratio.

[0027] Furthermore, methods for establishing water storage records based on initial water storage and replenishment include:

[0028] An initial water storage record is established based on the initial water storage volume. The initial water storage volume, the calculation period of the initial water storage volume, and the corresponding water receding coefficient are respectively used as the remaining water storage volume, the formation period, and the water receding coefficient of the initial water storage record, and no rainfall event identifier is set.

[0029] For the evaluation period when the rainfall is greater than zero, a water storage record is established based on the corresponding recharge amount. The recharge amount and the evaluation period are respectively used as the remaining water storage and the formation period of the corresponding water storage record. The drainage coefficient is obtained based on the groundwater level during the formation period and the calculation relationship between the groundwater level and the drainage coefficient. The rainfall event identifier of the evaluation period is used as the rainfall event identifier of the corresponding water storage record.

[0030] When the same rainfall event includes multiple evaluation periods with rainfall greater than zero, water storage records are established based on the replenishment water volume corresponding to each evaluation period to obtain the water storage record.

[0031] Furthermore, the method for calculating the corresponding candidate release amount based on each water storage record and calculating the total release amount based on the spring flow response data includes:

[0032] The first and second rainfall events are determined sequentially based on their occurrence time.

[0033] For the current evaluation period, select water storage records whose formation time is earlier than or equal to the current evaluation period and whose remaining water storage is greater than zero;

[0034] When the second rainfall event occurs, the water storage records corresponding to the first rainfall event and the second rainfall event will be selected together as the water storage records to participate in the allocation of the total release during the current evaluation period.

[0035] The water discharge ratio is calculated based on the water discharge coefficient and evaluation period length of each participating water storage record. The product of the corresponding remaining water volume and the water discharge ratio is determined as the candidate release volume. The product of the spring flow response data of the current evaluation period and the evaluation period length is determined as the total release volume. Thus, the candidate release volume corresponding to each participating water storage record and the total release volume corresponding to the current evaluation period are obtained.

[0036] Furthermore, the total release amount is allocated to the corresponding water storage records according to the proportion of each candidate release amount. The methods for obtaining the allocated release amount include:

[0037] The remaining water volume in each water storage record is used as the remaining water volume before allocation and the current available remaining water volume, and the cumulative allocation and release volume is set to zero;

[0038] If the total release amount is not greater than the sum of the remaining water storage before each allocation, the total release amount will be taken as the total release amount before the allocation is completed.

[0039] The total unallocated amount of water is allocated according to the proportion of the candidate release amount of the storage records with a current available remaining water volume greater than zero to the sum of the candidate release amounts, thus obtaining the water volume to be allocated in this round;

[0040] The smaller value between the water volume to be allocated in this round and the current available remaining water volume is added to the cumulative allocated release amount. The current available remaining water volume is updated by subtracting the cumulative allocated release amount from the remaining water volume before allocation. The total release amount is updated by subtracting the sum of the cumulative allocated release amount from the total release amount.

[0041] Repeat until the total unallocated release amount is zero, then use the cumulative allocated release amount as the allocated release amount.

[0042] Furthermore, the methods for obtaining the water volume to be allocated in this round include:

[0043] Exclude water storage records where the current available water volume is zero;

[0044] Calculate the proportion of the candidate release amount corresponding to each water storage record with a current available remaining water volume greater than zero to the sum of all candidate release amounts, and obtain the recalculated proportion of candidate release amounts.

[0045] The total unallocated amount is allocated according to the recalculated candidate release ratio, and the amount of water to be allocated in this round is obtained for each water storage record with a current available water volume greater than zero.

[0046] Furthermore, methods for updating water storage records based on allocated release amounts include:

[0047] The updated remaining water volume is obtained by subtracting the corresponding allocated release amount from the remaining water volume before allocation in each water storage record;

[0048] Storage records with a remaining water volume of zero will no longer participate in the total release allocation for subsequent evaluation periods, while storage records with a remaining water volume greater than zero will be retained for the next evaluation period.

[0049] The process of calculating the replenishment volume, establishing the storage record, calculating the candidate release volume, allocating the total release volume, and updating the remaining storage volume is repeated in chronological order of the evaluation period until the storage record of the last evaluation period before the start of the prediction is completed, thus obtaining the updated storage record.

[0050] Furthermore, the methods for obtaining karst spring flow prediction data by inputting the acquired predicted hydro-meteorological data and updated water storage records into a pre-trained karst spring flow prediction model include:

[0051] When the updated water storage record is not empty, calculate the time difference between the start time of the formation period of each updated water storage record and the start time of the prediction to obtain the interval duration;

[0052] The remaining water storage volume, interval duration, and water receding coefficient are combined to form the water storage record input vector, and arranged from morning to night according to the formation time period to obtain the water storage record input sequence.

[0053] The predicted rainfall and predicted temperature for each prediction period are extracted from the predicted hydrometeorological data. The predicted rainfall and predicted temperature for each prediction period are combined to form a hydrometeorological input vector, and then arranged in the time order of the prediction period to obtain the hydrometeorological input sequence.

[0054] The water storage record input sequence and the hydrological and meteorological input sequence are input into the pre-trained karst spring flow prediction model to obtain karst spring flow prediction data.

[0055] Compared with existing technologies, the technical effects and advantages of the karst spring flow prediction method based on hydrological and meteorological data and physical mechanisms proposed in this application are as follows:

[0056] This application performs rainfall event correlation processing on historical hydrological and meteorological data and karst spring flow data. Based on spring flow response data and groundwater level data, it calculates the initial water storage volume and establishes water storage records according to the recharge volume formed in different evaluation periods. Then, it allocates the total spring release volume according to the candidate release volume ratio corresponding to each water storage record and updates the remaining water storage volume. This allows the water storage volume formed in the previous rainfall event that has not yet been released to be recorded separately from the recharge volume formed in the subsequent rainfall event and participate in the karst spring flow prediction. This reduces the mismatch between the attribution of the remaining water storage release volume and the newly added recharge volume when the responses of adjacent rainfall events overlap, improves the accuracy of the representation of the water storage state at the start of the prediction, and thus improves the accuracy of the karst spring flow prediction results. Attached Figure Description

[0057] Figure 1 This is a schematic diagram of an implementation environment of an embodiment of this application;

[0058] Figure 2 This is a schematic diagram of another implementation environment of the present application;

[0059] Figure 3 This is a flowchart of a karst spring flow prediction method based on hydrological and meteorological data and physical mechanisms, according to an embodiment of this application.

[0060] Figure 4 This is a flowchart illustrating the process of establishing water storage records, allocating total release volume, and updating water storage records in an embodiment of this application. Detailed Implementation

[0061] The technical solutions of this application will be described in detail, clearly, and completely below with reference to the accompanying drawings of the embodiments. It should be particularly noted that the specific embodiments described below are only used to better illustrate and explain the technical solutions of this application, and are intended to enable those skilled in the art to better understand and implement this application, and should not be construed as limiting the scope of protection of this application. Without departing from the spirit and substance of this application, those skilled in the art can modify, adjust, or make equivalent substitutions based on the content disclosed in this application, and these modifications, adjustments, or equivalent substitutions should all be considered within the scope of protection of this application.

[0062] Example 1:

[0063] To make the technical solutions and effects of the embodiments of this application clearer, the implementation methods of this application will be further described below with reference to the accompanying drawings.

[0064] Figure 1 This is a schematic diagram illustrating an implementation environment provided in an embodiment of this application. See also... Figure 1Rainfall monitoring equipment, groundwater level monitoring equipment, and spring flow monitoring equipment are installed within the karst spring area. The rainfall monitoring equipment collects rainfall data within the karst spring area; the groundwater level monitoring equipment is located in observation wells within the karst aquifer to collect groundwater level data; and the spring flow monitoring equipment is located at the karst spring outlet to collect spring flow data. Forecasted hydrological and meteorological data are obtained through a hydrological and meteorological forecast data interface.

[0065] After rainfall enters the karst aquifer, part of the recharge water reaches the karst spring outlet through karst conduits, while the rest is stored in fissures and the rock matrix, and continues to be released to the karst spring outlet after the rainfall ends. Therefore, the karst spring flow rate at the predicted start time may be affected not only by the current rainfall but also by the water volume formed before the predicted start time that has not yet been released.

[0066] Figure 2 This is a schematic diagram illustrating another implementation environment provided for an embodiment of this application. See also... Figure 2 After the recharge water from the first rainfall event enters the karst aquifer, there is still unreleased water remaining at the end of the first rainfall event. If a second rainfall event occurs before the water stored in the first rainfall event has been fully released, the second rainfall event will generate new recharge water. Therefore, the total release during the second rainfall event is affected by both the remaining water stored in the first rainfall event and the recharge water generated by the second rainfall event.

[0067] for Figure 2 In the implementation environment shown, if spring flow response data is directly attributed to a single rainfall event, it is easy to include the release of water generated during the second rainfall event from the remaining water storage formed in the first rainfall event in the second rainfall event. Therefore, this application does not directly assign spring flow response data to a single rainfall event. Instead, it establishes separate water storage records based on the recharge volume corresponding to different evaluation periods. Based on the candidate release volume corresponding to each water storage record, the total release volume corresponding to the same evaluation period is allocated to the corresponding water storage record, and each water storage record is updated according to the allocated release volume. If the water storage record corresponding to the first rainfall event still has a remaining water storage volume greater than zero when the second rainfall event occurs, the water storage record corresponding to the first rainfall event and the water storage record corresponding to the second rainfall event jointly participate in the allocation of the total release volume.

[0068] Figure 3 A flowchart illustrating a method for predicting karst spring flow based on hydrological and meteorological data and physical mechanisms, provided as an embodiment of this application. See also... Figure 3The karst spring flow prediction method provided in this embodiment takes historical hydrological and meteorological data and karst spring flow data of the same karst spring area as the processing objects. It calculates the recharge water volume formed in different evaluation periods based on rainfall data, calculates the total release volume corresponding to different evaluation periods based on spring flow response data, allocates the total release volume according to the candidate release volume ratio corresponding to each water storage record, and uses the predicted hydrological and meteorological data and the updated water storage records for karst spring flow prediction.

[0069] The method for predicting karst spring flow based on hydrological and meteorological data and physical mechanisms includes steps S101 to S105.

[0070] S101: Obtain historical hydrological and meteorological data and karst spring flow data; perform rainfall event correlation processing on the historical hydrological and meteorological data and karst spring flow data to obtain rainfall event data.

[0071] Historical hydrological and meteorological data include rainfall data and groundwater level data. Rainfall data records the sampling time and corresponding rainfall amount, groundwater level data records the sampling time and corresponding groundwater level, and karst spring flow data records the sampling time and corresponding karst spring flow.

[0072] Rainfall data can be collected by rainfall monitoring equipment installed within the karst spring area, groundwater level data can be collected by groundwater level monitoring equipment installed in observation wells within the karst aquifer, and karst spring flow data can be directly collected by flow monitoring equipment installed at the spring mouth. When using the spring mouth water level monitoring method, the karst spring flow data is calculated based on the spring mouth water level monitoring data and the pre-calibrated water level-flow relationship.

[0073] This embodiment uses fixed sampling intervals to collect rainfall data, groundwater level data, and karst spring flow data. Each fixed sampling interval is converted to the same time unit, the least common multiple of each fixed sampling interval is calculated, and this least common multiple is determined as the length of the evaluation period.

[0074] For example, when the sampling interval for rainfall data is ten minutes, the sampling interval for groundwater level data is thirty minutes, and the sampling interval for karst spring flow data is one hour, the least common multiple of each fixed sampling interval is one hour, and one hour can be determined as the length of the evaluation period.

[0075] In one possible implementation, when the preset time resolution of the karst spring flow prediction data output is an integer multiple of the least common multiple of the fixed sampling intervals, the preset time resolution of the karst spring flow prediction data output can be determined as the evaluation period length; when the preset time resolution of the karst spring flow prediction data output is not an integer multiple of the least common multiple of the fixed sampling intervals, the least common multiple of the fixed sampling intervals is determined as the evaluation period length. When the evaluation period length is greater than the least common multiple of the fixed sampling intervals, the rainfall sampling values ​​within the evaluation period are still accumulated according to the start and end times of the evaluation period, and the groundwater level sampling values ​​and karst spring flow sampling values ​​within the evaluation period are respectively time-weighted averaged.

[0076] Historical data is continuously divided according to the length of the evaluation period, so that each evaluation period includes an integer number of sampling intervals for rainfall data, groundwater level data, and karst spring flow data, and the rainfall data, groundwater level data, and karst spring flow data within the same evaluation period have the same time start and time end.

[0077] When multiple rainfall samples exist within a given evaluation period, these samples are summed to obtain the total rainfall for that period. Similarly, when multiple groundwater level samples exist within a given evaluation period, a time-weighted average is calculated based on the duration of each sample within the evaluation period to obtain the groundwater level for that period. Likewise, when multiple karst spring flow samples exist within a given evaluation period, a time-weighted average is calculated based on the duration of each karst spring flow sample within the evaluation period to obtain the average karst spring flow for that period.

[0078] Before dividing rainfall data into rainfall events, a preset event interval is set. The preset event interval is the boundary of continuous rainless duration used when classifying adjacent rainfall events into the same rainfall event or different rainfall events.

[0079] The method for setting the preset event interval includes: extracting multiple historical rainfall events from historical rainfall data, each historical rainfall event including one or more consecutive evaluation periods with rainfall greater than zero; for each historical rainfall event, determining the start time of the historical rainfall event, and identifying the evaluation period after the start time of the historical rainfall event where the spring flow response data first exceeds the spring flow response data of the previous evaluation period; using the duration between the start time of the historical rainfall event and the start time of the identified evaluation period as the spring flow response delay of the corresponding historical rainfall event; calculating the median of multiple spring flow response delays, and determining the median of the spring flow response delays as the preset event interval.

[0080] For example, the median can be calculated using the spring flow response delays corresponding to three or more historical rainfall events. If no spring flow response data exceeding the spring flow response data of the previous evaluation period occurs after a historical rainfall event, the corresponding historical rainfall event is not used to calculate the preset event interval.

[0081] In one possible implementation, when historical hydrogeological survey data or existing rainfall-spring flow response analysis results record spring flow response delays in karst spring areas, these recorded delays can be determined as preset event intervals. If historical hydrogeological survey data or existing rainfall-spring flow response analysis results record multiple spring flow response delays, the median of these delays is calculated, and this median is determined as the preset event interval.

[0082] The preset event interval is used to characterize the duration from rainfall input to the start of spring flow response in a karst spring area. When the continuous dry period between adjacent rainfall events is less than the preset event interval, the adjacent rainfall events are considered temporally related and classified into the same rainfall event. When the continuous dry period between adjacent rainfall events is greater than or equal to the preset event interval, the adjacent rainfall events are classified into different rainfall events.

[0083] Rainfall events are assigned rainfall event identifiers sequentially based on their occurrence time. The rainfall event identifier corresponding to the evaluation period with rainfall greater than zero is used as the rainfall event identifier for that evaluation period. Evaluation periods with zero rainfall are retained at their corresponding time positions, but no rainfall event identifier is assigned.

[0084] The groundwater level data and karst spring flow data are aligned with the rainfall data according to the evaluation period. The average karst spring flow corresponding to the evaluation period is used as the spring flow response data. The evaluation period, rainfall data, groundwater level data, spring flow response data and the corresponding rainfall event identifier are arranged continuously in chronological order to obtain the rainfall event data.

[0085] It should be noted that the rainfall event association processing is used to establish the correspondence between evaluation time periods and rainfall data, groundwater level data, and spring flow response data, and to establish the correspondence between rainfall data and the corresponding rainfall event identifier; the rainfall event association processing does not directly classify spring flow response data to a single rainfall event.

[0086] Before performing the initial water storage calculation, the prediction start time is determined and set at the time boundary of the evaluation period. The complete water recession process before the prediction start time is identified based on rainfall data and spring flow response data. The largest continuous data segment where rainfall is continuously equal to zero and spring flow response data continuously decreases over the evaluation period is defined as the complete water recession process. The largest continuous data segment is a continuous data segment that cannot be extended forward or backward.

[0087] Each complete receding event includes at least three adjacent evaluation periods. The complete receding events before the prediction start time are arranged in order of their end times from latest to earliest, and the complete receding events whose end times are closest to the prediction start times are given priority for determining the initial water storage calculation period.

[0088] S102, based on spring flow response data and groundwater level data from rainfall event data, calculates the initial water storage volume.

[0089] Multiple complete recession processes are extracted from rainfall event data, and the recession coefficient corresponding to each complete recession process is calculated.

[0090] In the complete recession process used to calculate the recession coefficient, the flow response data of each spring is greater than zero. If there are evaluation periods in the complete recession process where the flow response data of a spring is less than or equal to zero, the corresponding complete recession process is not used to calculate the recession coefficient.

[0091] For any complete water receding process, the start time of the first evaluation period of the complete water receding process is taken as the starting point, and the time and spring flow response data corresponding to each evaluation period in the complete water receding process are obtained.

[0092] For example, this embodiment can use an exponential recession relationship to describe the change in karst spring flow over time during a complete recession process. One form of the exponential recession relationship is as follows:

[0093] ;

[0094] In the formula, This indicates the time elapsed since the start of the complete receding process. The flow rate of the karst spring at that time This indicates the flow rate of the karst spring at the beginning of the complete receding process. Indicates the drainage coefficient. Time When using hourly water discharge coefficient The unit is per hour. It is a dimensionless quantity.

[0095] Using the aforementioned exponential recession relation, taking the natural logarithm of the exponential recession relation yields:

[0096] ;

[0097] For each complete water receding process, the time and spring flow response data corresponding to each evaluation period are obtained. The linear relationship between the time and the natural logarithm of the spring flow response data is calculated by least squares, and the absolute value of the slope of the linear relationship is determined as the water receding coefficient of the corresponding complete water receding process.

[0098] Obtain the groundwater level corresponding to the first evaluation period of each complete water receding process, and combine the groundwater level corresponding to the first evaluation period of each complete water receding process with the water receding coefficient of the corresponding complete water receding process to form a groundwater level and water receding coefficient data pair.

[0099] Groundwater level and drainage coefficient data with a drainage coefficient greater than zero are used to establish the calculation relationship between groundwater level and drainage coefficient. When the drainage coefficient is less than or equal to zero, the corresponding groundwater level and drainage coefficient data are not used to establish the calculation relationship between groundwater level and drainage coefficient.

[0100] When multiple groundwater level and drainage coefficient data pairs have the same groundwater level, calculate the average value of the multiple drainage coefficients corresponding to the same groundwater level, and use the average value of the drainage coefficients as the drainage coefficient of the corresponding groundwater level.

[0101] This embodiment uses at least two different groundwater level and drainage coefficient data to establish the calculation relationship between groundwater level and drainage coefficient.

[0102] By arranging groundwater level and drainage coefficient data pairs with different groundwater levels in ascending order, the calculation relationship between groundwater level and drainage coefficient is obtained. The applicable groundwater level range for establishing the calculation relationship between groundwater level and drainage coefficient is the range between the minimum and maximum groundwater levels used to establish this relationship.

[0103] When the groundwater level to be calculated is within the applicable groundwater level range, select two pairs of groundwater level and drainage coefficient data located on both sides of the groundwater level to be calculated and adjacent to the groundwater level to be calculated. Perform piecewise linear interpolation according to the distance ratio between the groundwater level to be calculated and the two adjacent groundwater levels to obtain the drainage coefficient corresponding to the groundwater level to be calculated.

[0104] When the groundwater level to be calculated is lower than the lower limit of the applicable groundwater level range, the drainage coefficient corresponding to the lower limit of the applicable groundwater level range is determined as the drainage coefficient corresponding to the groundwater level to be calculated. When the groundwater level to be calculated is higher than the upper limit of the applicable groundwater level range, the drainage coefficient corresponding to the upper limit of the applicable groundwater level range is determined as the drainage coefficient corresponding to the groundwater level to be calculated.

[0105] In one possible implementation, the calculation relationship between groundwater level and drainage coefficient can be set as a correspondence table between groundwater level intervals and drainage coefficients. Specifically, groundwater level and drainage coefficient data pairs with different groundwater levels are arranged in ascending order of groundwater level. The average value of any two adjacent groundwater levels is determined as the boundary between the corresponding two groundwater level intervals. The range from the minimum groundwater level to the boundary between the minimum groundwater level and the next minimum groundwater level is determined as the first groundwater level interval. The range between the boundaries of two adjacent intervals is determined as the middle groundwater level interval, and the range from the last interval boundary to the maximum groundwater level is determined as the last groundwater level interval. When the groundwater level to be calculated equals any interval boundary, the groundwater level to be calculated is divided into groundwater level intervals with the interval boundary as the lower limit. The drainage coefficient in each groundwater level and drainage coefficient data pair is determined as the drainage coefficient of the corresponding groundwater level interval.

[0106] When the groundwater level to be calculated is located within any groundwater level interval, the drainage coefficient of the corresponding groundwater level interval is determined as the drainage coefficient corresponding to the groundwater level to be calculated. When the groundwater level to be calculated is lower than the lower limit of the applicable groundwater level range, the drainage coefficient corresponding to the first groundwater level interval is used; when the groundwater level to be calculated is higher than the upper limit of the applicable groundwater level range, the drainage coefficient corresponding to the last groundwater level interval is used.

[0107] The first evaluation period of the currently selected complete receding water process is determined as the initial water storage calculation period, and the continuous data with time boundaries between the initial water storage calculation period and the last evaluation period before the prediction start time are determined as the continuous historical period. When performing the initial water storage calculation for the first time, the currently selected complete receding water process is the one whose end time is closest to the prediction start time. The prediction start time is the same as the end time of the last evaluation period in the continuous historical period.

[0108] Obtain the groundwater level corresponding to the initial water storage calculation period, and based on the groundwater level corresponding to the initial water storage calculation period and the calculation relationship between the groundwater level and the drainage coefficient, obtain the drainage coefficient corresponding to the initial water storage calculation period.

[0109] Obtain the spring flow response data corresponding to the initial water storage calculation period, and use the spring flow response data corresponding to the initial water storage calculation period as the average karst spring flow during the initial water storage calculation period.

[0110] In the example employing the aforementioned exponential recession relationship, the initial water storage is characterized as the equivalent groundwater storage that decays exponentially according to the recession coefficient corresponding to the calculation period of the initial water storage. The amount of water released during the initial water storage calculation period is equal to the product of the initial water storage and the recession ratio, where the recession ratio is... The amount of water released during the initial water storage calculation period is also equal to the product of the average karst spring flow rate corresponding to the initial water storage calculation period and the length of the evaluation period.

[0111] For example, when using the aforementioned exponential receding water relationship, one way to calculate the initial water storage is as follows:

[0112] ;

[0113] In the formula, Indicates the initial water storage volume. This represents the average karst spring flow rate corresponding to the period in which the initial water storage was calculated. This represents the water receding coefficient corresponding to the period in which the initial water storage was calculated. Indicates the length of the evaluation period.

[0114] When the average karst spring flow rate is expressed in cubic meters per hour, the evaluation period length is expressed in hours, and the drainage coefficient is expressed in hours, the unit of the initial water storage is cubic meters.

[0115] Initial storage capacity represents the equivalent groundwater storage that existed at the start of a continuous historical period and is capable of generating karst spring flows in subsequent evaluation periods. Initial storage capacity is not attributed to subsequent rainfall events within the continuous historical period.

[0116] S103, calculate the replenishment water volume based on the rainfall data in the rainfall event data, and establish a water storage record based on the initial water storage volume and the replenishment water volume.

[0117] The recharge area and rainfall infiltration coefficient of the spring area were obtained from the hydrogeological survey data of the spring area. The recharge area of ​​the spring area was determined based on the groundwater watershed, the distribution range of karst aquifers, and the hydrogeological survey results, while the rainfall infiltration coefficient was determined based on the water balance calculation results of the spring area.

[0118] In this embodiment, the same rainfall infiltration coefficient is used within the same continuous historical period. For different continuous historical periods, the rainfall infiltration coefficient can be determined separately based on the spring water balance calculation results for the corresponding historical period.

[0119] In another possible approach, when a continuous historical period spans multiple hydrological periods, the rainfall infiltration coefficient corresponding to each hydrological period can be obtained separately, and the recharge volume can be calculated based on the rainfall infiltration coefficient corresponding to the hydrological period to which the evaluation period belongs. Hydrological periods include high-water periods, normal-water periods, and low-water periods. The start and end times of each hydrological period are determined according to the start and end times based on the multi-year rainfall and seasonal variations in groundwater level in the hydrogeological survey data. The rainfall infiltration coefficients corresponding to different hydrological periods are determined based on the spring water balance calculation results for the corresponding hydrological period, and the same rainfall infiltration coefficient is used for each evaluation period within the same hydrological period.

[0120] Standardize the units of measurement for rainfall data, spring recharge area, and water recharge. For example, when rainfall is measured in millimeters, spring recharge area in square meters, and water recharge in cubic meters, the unit of measurement for the first rainfall is... One method for calculating the replenishment water volume corresponding to each evaluation period is as follows:

[0121] ;

[0122] In the formula, Indicates the first The amount of water replenishment corresponding to each evaluation period. Indicates the first Rainfall corresponding to each evaluation period Indicates the area replenished by the spring. Indicates the first The rainfall infiltration coefficient used in each evaluation period is a dimensionless quantity.

[0123] When the rainfall corresponding to the evaluation period is zero, the corresponding water supply is set to zero, and no new water storage record is established based on the water supply. When the rainfall corresponding to the evaluation period is greater than zero, a corresponding water storage record is established based on the calculated water supply.

[0124] Water storage records are used to record groundwater storage information corresponding to the initial water storage volume or the recharge volume within an evaluation period. Each water storage record includes at least the remaining water storage volume, the formation period, and the receding coefficient; water storage records established based on the recharge volume also include the identifier of the corresponding rainfall event.

[0125] When establishing an initial water storage record based on the initial water storage volume, the initial water storage volume is used as the remaining water storage volume in the initial water storage record, the period for calculating the initial water storage volume is used as the formation period of the initial water storage record, and the recession coefficient corresponding to the period for calculating the initial water storage volume is used as the recession coefficient of the initial water storage record. No rainfall event identifier is set for the initial water storage record.

[0126] It should be noted that the formation period in the initial water storage record is used to indicate the evaluation period when the initial water storage record begins to participate in the distribution of release volume, and does not indicate the actual time when the initial water storage volume is formed.

[0127] The recharge volume corresponding to each evaluation period within a continuous historical period is processed according to the time sequence of the evaluation period. When the recharge volume corresponding to an evaluation period is greater than zero, the recharge volume is used as the remaining water storage volume of the new water storage record, the evaluation period corresponding to the recharge volume is used as the formation period of the new water storage record, and the rainfall event identifier corresponding to the rainfall data is used as the rainfall event identifier of the new water storage record.

[0128] Obtain the groundwater level corresponding to the formation period of the new water storage record. Based on the groundwater level corresponding to the formation period and the calculation relationship between the groundwater level and the drainage coefficient, obtain the drainage coefficient of the new water storage record.

[0129] A single rainfall event can include multiple assessment periods with rainfall greater than zero. When the recharge volume corresponding to multiple assessment periods within the same rainfall event is all greater than zero, a water storage record is established based on the recharge volume corresponding to each assessment period. Multiple water storage records corresponding to the same rainfall event have the same rainfall event identifier, and the formation period, remaining water storage volume, and drainage coefficient of each water storage record are determined according to the corresponding assessment period.

[0130] The remaining water volume in the water storage record is used to record the amount of water that has not yet been released after the release allocation for the evaluation period is completed. The formation period is used to determine the evaluation period in which the water storage record begins to participate in the release allocation. The water receding coefficient is used to calculate the candidate release amount corresponding to the water storage record. The rainfall event identifier is used to maintain the correspondence between the replenishment water volume, the water storage record and the rainfall event.

[0131] This embodiment uses the evaluation period as the smallest time calculation unit, no longer distinguishing between the specific rainfall occurrence time and the karst spring flow release time within an evaluation period. For each evaluation period, the replenishment water volume is first calculated based on rainfall data and a water storage record is established. Then, the total release volume is calculated based on spring flow response data and the total release volume is allocated. The water storage record whose formation period is equal to the current evaluation period participates in the total release volume allocation starting from the current evaluation period.

[0132] S104. Calculate the corresponding candidate release amount based on each water storage record, and calculate the total release amount based on the spring flow response data; allocate the total release amount to the corresponding water storage record according to the proportion of each candidate release amount to obtain the allocated release amount, and update the water storage record based on the allocated release amount.

[0133] It should be noted that the continuous historical time period used in this embodiment meets the following data conditions: each evaluation period in the continuous historical time period has rainfall data, groundwater level data, and spring flow response data, and the total release corresponding to each evaluation period is less than or equal to the sum of the remaining water storage volumes of each water storage record participating in the allocation of the total release volume of the corresponding evaluation period before allocation. If the continuous historical time period determined by the complete receding process whose end time is closest to the prediction start time does not meet the aforementioned data conditions, the previous complete receding process is selected in order of the end time of the complete receding process from latest to earliest, and the initial water storage calculation, water storage record establishment, and data condition verification are re-executed until a continuous historical time period that meets the aforementioned data conditions is obtained. If there is no complete receding process that meets the aforementioned data conditions before the prediction start time, karst spring flow prediction is not performed at the prediction start time. When used to construct model training samples, historical prediction start times that do not meet the aforementioned data conditions are not used to construct model training samples.

[0134] Starting from the initial water storage calculation period, each evaluation period in the continuous historical period is processed in chronological order.

[0135] For the current evaluation period, water storage records with formation times earlier than or equal to the current evaluation period and remaining water volume greater than zero are selected from the established water storage records. These selected water storage records are designated as participating in the allocation of the total release volume for the current evaluation period. Water storage records with formation times later than the current evaluation period are not included in the allocation of the total release volume for the current evaluation period.

[0136] Based on the remaining water volume, receding coefficient, and evaluation period length of each water storage record participating in the allocation of the total release volume for the current evaluation period, the candidate release volume corresponding to each water storage record is calculated.

[0137] For example, when using the aforementioned exponential recession relationship, the first Within the evaluation period, the first One method for calculating the candidate release amount corresponding to each water storage record is as follows:

[0138] ;

[0139] In the formula, Indicates the first Within the evaluation period, the first The candidate release amount corresponding to each water storage record Indicates the first Before allocating the total release amount in the evaluation period, the [number]th [period] was [number]. The remaining water storage volume recorded in the water storage record. Indicates the first The water discharge coefficient of each water storage record. Indicates the length of the evaluation period.

[0140] Candidate release amounts are used to characterize the relative release proportions of different water storage records within the current evaluation period. Candidate release amounts are only used to calculate the distribution ratio of total release among different water storage records and are not used as an upper limit for the allocated release amount. The upper limit for the allocated release amount is the remaining water volume of the corresponding water storage record before the total release allocation is performed within the current evaluation period.

[0141] The remaining water storage volume of the water storage records participating in the allocation of the total release volume for the current evaluation period is greater than zero, the water receding coefficient is greater than zero, and the evaluation period length is greater than zero. Therefore, the release volume of each candidate is greater than zero, and the sum of the release volumes of each candidate is greater than zero.

[0142] The average karst spring flow rate is obtained from the spring flow response data corresponding to the current evaluation period. The average karst spring flow rate is multiplied by the evaluation period length to obtain the total release amount corresponding to the current evaluation period. The units of measurement for the average karst spring flow rate, the evaluation period length, and the total release amount are standardized.

[0143] For each water storage record participating in the allocation of the total release volume for the current evaluation period, a cumulative allocation release volume is set. The initial value of the cumulative allocation release volume is zero, and the remaining water storage volume of each water storage record before the allocation of the total release volume for the current evaluation period is taken as the remaining water storage volume of the corresponding water storage record before allocation.

[0144] For example, the initial allocation volume of each water storage record can be calculated according to the proportion of each candidate release volume to the sum of all candidate release volumes participating in the allocation of the total release volume for the current evaluation period. One calculation method is as follows:

[0145] ;

[0146] In the formula, Indicates the first Within the evaluation period, the first The first round of water to be allocated in the water storage record. Indicates the first Total release amount corresponding to each evaluation period Indicates participation in the The number of water storage records allocated to the total release volume for each evaluation period.

[0147] Compare the initial water volume to be allocated in each water storage record with the remaining water volume before allocation in the corresponding water storage record. Add the smaller value between the initial water volume to be allocated and the remaining water volume before allocation to the cumulative allocation release amount of the corresponding water storage record. Subtract the cumulative allocation release amount from the remaining water volume before allocation in the corresponding water storage record to obtain the current available remaining water volume for allocation in the corresponding water storage record.

[0148] The total unallocated release amount is obtained by subtracting the sum of the current cumulative allocated release amounts from each water storage record from the total release amount.

[0149] When the total unallocated release amount is zero, the allocation of the total release amount for the current evaluation period ends. When the total unallocated release amount is greater than zero, the water storage records with zero remaining allocable water are excluded, and the candidate release amount ratio is recalculated based on the candidate release amounts already calculated for each water storage record with a greater remaining allocable water in the current evaluation period.

[0150] The total unallocated release volume is allocated according to the recalculated candidate release volume ratio, resulting in the water volume to be allocated in this round for each storage record with a current allocable remaining water volume greater than zero. The water volume to be allocated in this round is compared with the current allocable remaining water volume of the corresponding storage record. The smaller value between the water volume to be allocated in this round and the current allocable remaining water volume is added to the cumulative allocated release volume of the corresponding storage record. The current allocable remaining water volume of the corresponding storage record is then recalculated by subtracting the cumulative allocated release volume from the remaining storage volume before allocation.

[0151] The total unallocated release volume is recalculated by subtracting the sum of the cumulative allocated release volumes from each updated water storage record. If the total unallocated release volume is still greater than zero, the process of excluding water storage records with zero remaining allocated water volume, recalculating the candidate release volume ratio, and updating the cumulative allocated release volume is repeated until the total unallocated release volume equals zero.

[0152] The final cumulative distribution release amount of each water storage record is determined as the distribution release amount of the corresponding water storage record. The corresponding distribution release amount is then deducted from the remaining water storage amount before distribution of each water storage record to obtain the updated remaining water storage amount of each water storage record.

[0153] After the total release allocation for the current evaluation period is completed, water storage records with zero remaining water volume will no longer be included in the set of water storage records participating in the total release allocation for subsequent evaluation periods. Initial water storage records with a remaining water volume greater than zero retain their formation period and recession coefficient. Water storage records with a remaining water volume greater than zero and established based on the recharge water volume retain their formation period, recession coefficient, and the identifier of the rainfall event to which they belong, and are included in the set of water storage records participating in the total release allocation for the next evaluation period.

[0154] The water storage records are processed sequentially according to the time order of the evaluation periods, up to the last evaluation period before the start of the prediction, to obtain the updated water storage records.

[0155] The updated water storage records are those with a remaining water storage volume greater than zero after allocating the total release volume for each evaluation period within a continuous historical timeframe. Each updated water storage record includes the remaining water storage volume, the formation period, and the recession coefficient; updated water storage records established based on recharge volume also include the identifier of the associated rainfall event.

[0156] S105, the acquired predicted hydrological and meteorological data and the updated water storage records are input into the pre-trained karst spring flow prediction model to obtain karst spring flow prediction data.

[0157] Based on the determined prediction start time, the number of prediction periods is determined. The prediction start time is the same as the end time of the last evaluation period in the continuous historical period, and the length of each prediction period is the same as the evaluation period length used in step S101.

[0158] Hydrological and meteorological forecast data that were formed before the start of the forecast and correspond to each forecast period are obtained through the hydrological and meteorological forecast data interface. The hydrological and meteorological forecast data are then organized according to the length of the evaluation period to obtain the predicted hydrological and meteorological data.

[0159] The predicted hydrometeorological data includes predicted rainfall data and predicted temperature data. The predicted hydrometeorological data are arranged in chronological order according to the prediction period, with each prediction period corresponding to predicted rainfall and predicted temperature.

[0160] When the updated water storage records are not empty, they are arranged in order of formation time from earliest to latest. For each updated water storage record, the remaining water volume and the water receding coefficient are extracted, and the time difference between the start time of the formation time of the updated water storage record and the start time of the prediction is calculated to obtain the interval duration corresponding to the updated water storage record.

[0161] The remaining water storage volume, interval duration, and water receding coefficient corresponding to each updated water storage record are combined to form a water storage record input vector. The water storage record input vectors are then arranged according to the formation time period to obtain the water storage record input sequence.

[0162] The associated rainfall event identifier is used to maintain the correspondence between the updated water storage record and historical rainfall events. In this embodiment, the associated rainfall event identifier is not used as input to the pre-trained karst spring flow prediction model.

[0163] In another possible implementation, the rainfall event identifier can be converted into a rainfall event sequence feature, and this sequence feature can be added to the corresponding water storage record input vector. Specifically, the rainfall event sequence feature corresponding to the initial water storage record is set to zero, and the rainfall event sequence features are sequentially set for the water storage records established based on the replenishment volume, according to the order of occurrence of each rainfall event within a continuous historical period. The same rainfall event sequence feature setting method is used in both the model training and prediction phases.

[0164] The predicted rainfall and predicted temperature for each prediction period are combined to form a hydrometeorological input vector, and the hydrometeorological input vectors are arranged in chronological order according to the prediction period to obtain the hydrometeorological input sequence.

[0165] In some embodiments, the pre-trained karst spring flow prediction model may include a first long short-term memory network, a second long short-term memory network, and a prediction layer implemented using fully connected layers. The following uses a karst spring flow prediction model including a first long short-term memory network, a second long short-term memory network, and a prediction layer as an example to illustrate the model processing procedure.

[0166] When the updated water storage record is not empty, the water storage record input sequence is fed into the first long short-term memory network, and the hidden state output by the first long short-term memory network when processing the last water storage record input vector is used as the water storage record feature vector. When the number of updated water storage records is different at different prediction start times, the first long short-term memory network processes the water storage record input sequence sequentially according to the actual number of updated water storage records.

[0167] When the updated water storage record is empty, the zero vector with the same dimension as the output of the first Long Short-Term Memory network is used as the feature vector of the water storage record. The same empty water storage record processing method is used in both the model training and prediction phases.

[0168] The hydrological and meteorological input sequences are sequentially input into the second long short-term memory network according to the prediction time period to obtain the hydrological and meteorological feature vectors corresponding to each prediction time period.

[0169] The water storage record feature vector is concatenated with the hydrological and meteorological feature vectors corresponding to each prediction period, and the concatenation results for each prediction period are sequentially input into the prediction layer. The prediction layer maps the concatenation result for each prediction period to a karst spring flow prediction value, thus obtaining the karst spring flow prediction value for each prediction period.

[0170] Optionally, the first and second Long Short-Term Memory (LSTM) networks can be replaced by gated recurrent unit networks, temporal convolutional networks, or attention-based temporal coding networks, respectively. The temporal network processing the water storage record input sequence converts it into a water storage record feature vector, while the temporal network processing the hydrological and meteorological input sequence converts it into hydrological and meteorological feature vectors corresponding to each prediction period. Regardless of the temporal network used, the water storage record feature vector and the corresponding hydrological and meteorological feature vector for each prediction period are concatenated. The concatenation results for each prediction period are then sequentially input into the prediction layer implemented using fully connected layers to obtain the predicted karst spring flow values ​​for each prediction period.

[0171] The predicted karst spring flow rates for each prediction period are arranged in chronological order to form the karst spring flow rate prediction data.

[0172] The method for obtaining the pre-trained karst spring flow prediction model is described in Example 3.

[0173] This embodiment obtains continuous rainfall event data arranged according to the evaluation period by organizing historical hydrological and meteorological data and karst spring flow data and associating them with rainfall events. Initial water storage is calculated based on spring flow response data and groundwater level data, and recharge water volume is calculated based on rainfall data. Water storage records are established based on the initial water storage and recharge water volume. The total release volume is allocated according to the proportion of candidate release volumes corresponding to each water storage record, and the allocated release volume is updated based on the allocated release volume. The predicted hydrological and meteorological data and the updated water storage records are input into a pre-trained karst spring flow prediction model to obtain predicted karst spring flow data.

[0174] Therefore, the water storage record corresponding to the previous rainfall event and with a remaining water storage volume greater than zero can coexist with the water storage record formed by subsequent rainfall and participate in the total release allocation. This preserves the historical water storage state corresponding to the prediction start time when the spring flow response times of adjacent rainfall events overlap, thereby improving the correspondence between the predicted input of karst spring flow and the actual replenishment and release process.

[0175] Example 2:

[0176] Based on the above embodiment 1, Figure 4 This is a flowchart illustrating the process of establishing water storage records, allocating total release volume, and updating water storage records in a karst spring flow prediction method based on hydrological and meteorological data and physical mechanisms.

[0177] This embodiment takes the sequential occurrence of a first rainfall event and a second rainfall event within a continuous historical period as an example. When the second rainfall event occurs, some of the water stored during the first rainfall event has not yet been released. The water storage records corresponding to the first and second rainfall events coexist within the same evaluation period. For the same evaluation period, candidate release amounts are calculated for each water storage record. The total release amount calculated based on the spring flow response data is then allocated to the corresponding water storage records according to the proportion of each candidate release amount. Each water storage record is then updated based on the allocated release amounts.

[0178] The acquisition of historical hydrological and meteorological data and karst spring flow data, the correlation processing of rainfall events, the calculation of initial water storage, the calculation of water replenishment, and the prediction of karst spring flow, which are not specifically described in this embodiment, can be performed in the manner described in Embodiment 1.

[0179] See Figure 4 The process of establishing water storage records, allocating total release volume, and updating water storage records includes:

[0180] S201, establish water storage records corresponding to different rainfall events based on rainfall event data within a continuous historical period.

[0181] A continuous historical period includes multiple evaluation periods with consecutive time boundaries, and the end time of the continuous historical period is the same as the start time of the prediction. According to the rainfall event division method described in Example 1, two rainfall processes that occur sequentially within the continuous historical period are respectively identified as the first rainfall event and the second rainfall event, with the first rainfall event occurring earlier than the second rainfall event.

[0182] For each evaluation period in the first and second rainfall events where the rainfall was greater than zero, the recharge volume was calculated based on the rainfall data for the corresponding evaluation period, and a water storage record was established based on the recharge volume. When establishing the water storage record, the recharge volume was used as the remaining water storage volume, the evaluation period corresponding to the recharge volume was used as the formation period, and the drainage coefficient was obtained based on the groundwater level corresponding to the formation period and the calculation relationship between the groundwater level and the drainage coefficient. The rainfall event identifier of the corresponding rainfall event was used as the rainfall event identifier.

[0183] Before the second rainfall event occurs, the total release volume is allocated to the initial water storage record and the water storage record corresponding to the first rainfall event according to the time sequence of the evaluation period, and the remaining water storage volume of each water storage record is updated according to the allocated release volume.

[0184] When the second rainfall event occurs, at least one water storage record corresponding to the first rainfall event still has a remaining water storage volume greater than zero. After the recharge volume for the current evaluation period in the second rainfall event is greater than zero and the water storage record corresponding to the second rainfall event is established, the water storage record corresponding to the first rainfall event with a remaining water storage volume greater than zero and the water storage record corresponding to the second rainfall event coexist.

[0185] It should be noted that the distinction between the first and second rainfall events is used to differentiate the supply sources corresponding to different rainfall processes, and is not used to attribute spring flow response data to the first or second rainfall event respectively.

[0186] S202, determine the water storage records that participate in the allocation of the total release volume for the current evaluation period, and calculate the candidate release volume corresponding to each water storage record.

[0187] From the established water storage records, select those whose formation period is earlier than or equal to the current evaluation period and whose remaining water storage volume is greater than zero, to obtain the water storage records that participate in the allocation of the total release volume for the current evaluation period.

[0188] After the water storage record corresponding to the second rainfall event is established, the water storage records participating in the allocation of the total release for the current evaluation period include those corresponding to the first rainfall event and having a remaining water storage volume greater than zero, as well as those from the second rainfall event whose formation time is earlier than or equal to the current evaluation period and whose remaining water storage volume is greater than zero. If the remaining water storage volume of the initial water storage record is greater than zero, the initial water storage record also participates in the allocation of the total release for the current evaluation period.

[0189] According to the candidate release calculation method described in Example 1, the candidate release amount corresponding to each water storage record is calculated based on the remaining water storage volume, water receding coefficient, and evaluation period length of each water storage record participating in the allocation of the total release amount for the current evaluation period.

[0190] Candidate release amounts are used to determine the proportion of the total release amount corresponding to the current evaluation period among the various water storage records. Candidate release amounts are not used as the upper limit of the allocated release amount for a corresponding water storage record; the upper limit of the allocated release amount for a corresponding water storage record is the remaining water volume of that record before the total release amount is allocated during the current evaluation period.

[0191] S203, calculate the total release amount based on the spring flow response data corresponding to the current evaluation period, and allocate the total release amount according to the proportion of each candidate release amount.

[0192] This embodiment focuses on the case where the total release volume corresponding to the current evaluation period is less than or equal to the sum of the remaining water volume of each water storage record before the allocation of the total release volume of the current evaluation period.

[0193] Calculate the total release amount corresponding to the current evaluation period based on the average karst spring flow rate and the length of the evaluation period.

[0194] For each water storage record identified in S202 that participates in the allocation of the total release volume for the current evaluation period, a cumulative allocation release volume is set, and the initial value of the cumulative allocation release volume is set to zero. The remaining water storage volume of each water storage record before the allocation of the total release volume for the current evaluation period is taken as the remaining water storage volume of the corresponding water storage record before allocation.

[0195] Based on the proportion of the candidate release amount corresponding to each water storage record to the sum of all candidate release amounts as determined in S202, the total release amount corresponding to the current evaluation period is allocated to obtain the first round of water volume to be allocated for each water storage record.

[0196] Compare the initial unallocated water volume for each water storage record with the remaining water volume before allocation for that record. Add the smaller of these two values ​​to the cumulative allocation release amount for the corresponding water storage record. Subtract the cumulative allocation release amount for the corresponding water storage record from its remaining water volume before allocation to obtain the current allocable remaining water volume for that record.

[0197] The total unallocated release amount is obtained by subtracting the sum of the cumulative allocated release amounts from each water storage record from the total release amount corresponding to the current evaluation period.

[0198] When the total unallocated release amount is zero, the allocation of the total release amount for the current evaluation period ends. When the total unallocated release amount is greater than zero, the water storage records with zero remaining allocable water are excluded, and the candidate release amount ratio is recalculated according to the candidate release amounts corresponding to each water storage record with a greater than zero remaining allocable water.

[0199] The total unallocated amount is allocated according to the recalculated candidate release ratio, and the amount of water to be allocated in this round is obtained for each water storage record with a current available water volume greater than zero.

[0200] Compare the water volume to be allocated in the current round for each water storage record with the current allocable remaining water volume for that record. Add the smaller of the water volume to be allocated in the current round and the current allocable remaining water volume to the cumulative allocation release amount for the corresponding water storage record. Subtract the updated cumulative allocation release amount from the remaining water volume before allocation for the corresponding water storage record to obtain the updated current allocable remaining water volume for that record.

[0201] The total unallocated release volume is recalculated by subtracting the sum of the cumulative allocated release volumes after the updates of each water storage record from the total release volume corresponding to the current evaluation period. If the total unallocated release volume is still greater than zero, the process of excluding water storage records with zero current allocable remaining water volume, recalculating the candidate release volume ratio, allocating the total unallocated release volume, updating the cumulative allocated release volume, and updating the current allocable remaining water volume is repeated until the total unallocated release volume equals zero.

[0202] The final cumulative distribution release amount of each water storage record is determined as the distribution release amount of the corresponding water storage record.

[0203] S204, update the water storage records according to the allocated release amount corresponding to each water storage record, and obtain the updated water storage records corresponding to the predicted start time.

[0204] For each water storage record that participates in the allocation of the total release amount for the current evaluation period, the corresponding allocated release amount is deducted from the remaining water storage amount before allocation of the corresponding water storage record to obtain the remaining water storage amount after the update of the corresponding water storage record.

[0205] Storage records with zero remaining water volume will no longer participate in the total release allocation for subsequent evaluation periods. Initial storage records with a greater than zero remaining water volume retain their formation period and recession coefficient. Storage records with a greater than zero remaining water volume and established based on the replenishment water volume retain their formation period, recession coefficient, and the corresponding rainfall event identifier, and participate in the total release allocation for the next evaluation period.

[0206] According to the time sequence of the evaluation period, the calculation of water supply, establishment of water storage records, calculation of candidate release volume, allocation of total release volume and update of remaining water storage volume are repeated until the water storage record corresponding to the last evaluation period before the prediction start time is completed, and the updated water storage record corresponding to the prediction start time is obtained.

[0207] When the water storage records corresponding to the first rainfall event and the second rainfall event both have a remaining water storage volume greater than zero at the start of the prediction, the updated water storage records at the start of the prediction simultaneously retain the remaining water storage status formed by the first and second rainfall events.

[0208] The updated water storage record and predicted hydrological and meteorological data corresponding to the prediction start time are input into the pre-trained karst spring flow prediction model to obtain karst spring flow prediction data. The specific processing method can be performed in accordance with the method described in Example 1.

[0209] This embodiment retains the water storage records corresponding to the first rainfall event with a remaining water storage volume greater than zero, and allows the water storage records corresponding to the first rainfall event and the second rainfall event to participate in the total release allocation for the same evaluation period, thereby updating the remaining water storage volume formed by different rainfall events separately. When the water storage records corresponding to different rainfall events still have a remaining water storage volume greater than zero at the start of the prediction, the updated water storage records can retain the historical water storage state formed by the corresponding rainfall event.

[0210] Example 3:

[0211] This embodiment, based on Embodiments 1 and 2, further illustrates the method for obtaining the pre-trained karst spring flow prediction model. The acquisition of historical hydrological and meteorological data and karst spring flow data, the correlation processing of rainfall events, the establishment of water storage records, the allocation of total release, the updating of water storage records, and the organization of model input data, which are not specifically described in this embodiment, can be performed in accordance with the methods described in Embodiments 1 and 2.

[0212] The steps to obtain a pre-trained karst spring flow prediction model include:

[0213] Multiple historical forecast start times are obtained. Each historical forecast start time is located at the time boundary of the evaluation period and uses the same evaluation period length and number of forecast periods. Before the historical forecast start time, there is historical hydrological and meteorological data and karst spring flow data for obtaining updated water storage records. Before the historical forecast start time, historical hydrological and meteorological forecast data corresponding to each forecast period has been formed and stored. After the historical forecast start time, there is karst spring flow data corresponding to each forecast period.

[0214] For each historical prediction start time, the historical hydrological and meteorological data and karst spring flow data prior to the historical prediction start time are processed in the manner described in Examples 1 and 2 to obtain the updated water storage record corresponding to the historical prediction start time.

[0215] When the updated water storage records are not empty, arrange the updated water storage records in order of formation time from early to late, calculate the time difference between the start time of the formation time of each updated water storage record and the start time of the historical prediction, and obtain the interval duration of the corresponding updated water storage record; form the remaining water storage volume, interval duration and water receding coefficient corresponding to each updated water storage record into a water storage record input vector, and arrange the water storage record input vectors in order of formation time from early to late to obtain the water storage record input sequence.

[0216] When the updated water storage record is empty, the input sequence of the water storage record at the corresponding historical prediction start time is set to an empty sequence.

[0217] Historical hydrological and meteorological forecast data are organized according to the evaluation period length to obtain the predicted rainfall and temperature for each forecast period. The predicted rainfall and temperature for each forecast period are combined to form a hydrological and meteorological input vector, and the hydrological and meteorological input vectors are arranged in chronological order according to the forecast period to obtain the hydrological and meteorological input sequence.

[0218] Organize the actual karst spring flow data collected after the start time of the historical prediction according to the evaluation period length, obtain the karst spring flow corresponding to each prediction period, and arrange the karst spring flow corresponding to each prediction period in chronological order to obtain training labels.

[0219] Historical training samples are composed of the water storage record input sequence, hydrological and meteorological input sequence, and training labels corresponding to the historical prediction start time. A historical training sample set is then established based on these historical training samples corresponding to multiple historical prediction start times. Each historical training sample in the historical training sample set includes at least the water storage record input sequence, the hydrological and meteorological input sequence, and the training labels. When the updated water storage record is empty, the corresponding water storage record input sequence in the historical training sample is an empty sequence.

[0220] It should be noted that the hydrological and meteorological input sequence uses historical hydrological and meteorological forecast data that were formed and saved before the start time of the corresponding historical forecast, and does not use hydrological and meteorological data actually collected after the start time of the historical forecast to replace the historical hydrological and meteorological forecast data.

[0221] The water storage record input sequence and hydro-meteorological input sequence are used as input features of the karst spring flow prediction model. The training labels composed of karst spring flow corresponding to each prediction period are used as supervision targets. The model parameters of the karst spring flow prediction model are iteratively updated by a supervised learning regression training method.

[0222] This embodiment employs a karst spring flow prediction model comprising a first long short-term memory network, a second long short-term memory network, and a prediction layer implemented with fully connected layers. The first long short-term memory network is used to obtain the water storage record feature vector based on the input sequence of water storage records, and the second long short-term memory network is used to obtain the hydrological and meteorological feature vectors corresponding to each prediction period based on the hydrological and meteorological input sequence. The water storage record feature vectors and the hydrological and meteorological feature vectors corresponding to each prediction period are concatenated, and the concatenation results for each prediction period are input into the prediction layer to obtain the training prediction values ​​of the karst spring flow for each prediction period.

[0223] When the number of updated water storage records corresponding to different historical training samples is different, the first long short-term memory network processes the water storage record input sequence according to the actual number of updated water storage records in the corresponding historical training samples, and keeps the input vectors of each water storage record arranged from early to late according to the formation time period.

[0224] When the input sequence of water storage records in the historical training samples is an empty sequence, the zero vector with the same dimension as the feature vector of the water storage record is used as the feature vector of the corresponding historical training sample, according to the empty water storage record processing method described in Example 1.

[0225] The model prediction error is calculated based on the difference between the predicted karst spring flow rates for each prediction period in the historical training samples and the karst spring flow rates for the same prediction period in the training labels. This embodiment uses the mean squared error as the loss function, and the average of the squared differences for each prediction period in the historical training samples is taken as the model prediction error.

[0226] Based on the model prediction error, the adaptive moment estimation optimization algorithm is used to update the model parameters of the first long short-term memory network, the second long short-term memory network, and the prediction layer, and the model prediction error after updating the model parameters is recalculated.

[0227] The model prediction error is compared with a preset error threshold. When the model prediction error is greater than the preset error threshold and the number of training iterations is less than the preset maximum number of training iterations, the model parameters of the karst spring flow prediction model continue to be updated. When the model prediction error is less than or equal to the preset error threshold, or the number of training iterations reaches the preset maximum number of training iterations, the model parameter updates are stopped, and the model parameters at the time of stopping the model parameter updates are fixed, thus obtaining the pre-trained karst spring flow prediction model.

[0228] The preset error threshold is the mean square error threshold, which is determined based on the upper limit of the allowable mean square error specified in advance for the karst spring flow prediction task, and remains unchanged during the model training process.

[0229] The preset maximum number of training iterations is the fallback stopping condition for model training. It is determined based on the allowed training time and the time required for the model training device to complete one round of model parameter updates, and is set before model training begins.

[0230] The same evaluation period length, prediction period number, water storage record input vector composition, water storage record arrangement, hydrological and meteorological data types, and empty water storage record processing methods are used in both the model training phase and the actual prediction phase.

[0231] The specific method for inputting the updated water storage record and predicted hydrological and meteorological data corresponding to the prediction start time into the pre-trained karst spring flow prediction model to obtain karst spring flow prediction data can be performed in accordance with the method described in Example 1.

[0232] This embodiment uses historical hydrological and meteorological forecast data formed and saved before the historical prediction start time and updated water storage records corresponding to the historical prediction start time to construct historical training samples, ensuring that the model training phase and the actual prediction phase use the same data acquisition boundaries and input structure. When the number of updated water storage records corresponding to different historical prediction start times is different, the karst spring flow prediction model extracts water storage state features according to the actual number of updated water storage records and the order of their formation time, so that water storage states corresponding to a remaining water storage volume greater than zero at the prediction start time participate in the karst spring flow prediction.

[0233] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for predicting karst spring flow based on hydrological and meteorological data and physical mechanisms, characterized in that, include: Acquire historical hydrological and meteorological data and karst spring flow data; Historical hydrological and meteorological data and karst spring flow data are correlated with rainfall events to obtain rainfall event data; The initial water storage volume was calculated based on spring flow response data and groundwater level data from rainfall event data. The replenishment water volume is calculated based on the rainfall data in the rainfall event data, and a water storage record is established based on the initial water storage volume and the replenishment water volume; Calculate the corresponding candidate release amount based on each water storage record, and calculate the total release amount based on the spring flow response data; allocate the total release amount to the corresponding water storage record according to the proportion of each candidate release amount to obtain the allocated release amount, and update the water storage record based on the allocated release amount; The acquired predicted hydrological and meteorological data and the updated water storage records are input into the pre-trained karst spring flow prediction model to obtain karst spring flow prediction data.

2. The method for predicting karst spring flow based on hydrological and meteorological data and physical mechanisms according to claim 1, characterized in that, Methods for obtaining rainfall event data include: The sampling interval time unit for rainfall data, groundwater level data, and karst spring flow data is unified, and the least common multiple of the sampling interval is used as the evaluation period length. Historical data are divided according to the length of the evaluation period. Rainfall data within each evaluation period are accumulated. Groundwater level data and karst spring flow data are time-weighted averaged. The time-weighted average of karst spring flow data is used as spring flow response data. The preset event interval is obtained. If the consecutive rainless duration of adjacent rainfall events is less than the preset event interval, they are classified into the same rainfall event. If the duration is greater than or equal to the preset event interval, they are classified into different rainfall events. Only the evaluation period with rainfall greater than zero is assigned a rainfall event identifier; Rainfall event data is obtained by arranging the rainfall data, groundwater level data, spring flow response data, and corresponding rainfall event identifiers for each evaluation period in chronological order.

3. The method for predicting karst spring flow based on hydrological and meteorological data and physical mechanisms according to claim 2, characterized in that, Methods for obtaining preset event intervals include: Extract multiple historical rainfall events from rainfall data; For each historical rainfall event, the evaluation period is identified where the spring flow response data first exceeds the spring flow response data of the previous evaluation period after the start time of the historical rainfall event. The duration between the start time of the historical rainfall event and the start time of the identified evaluation period is defined as the spring flow response delay, and the median of multiple spring flow response delays is defined as the preset event interval.

4. The method for predicting karst spring flow based on hydrological and meteorological data and physical mechanisms according to claim 2, characterized in that, Methods for calculating the initial water storage include: Set the prediction start time at the boundary of the evaluation period, and take the largest continuous data segment in which the rainfall is continuously equal to zero and the spring flow response data continuously decreases before the prediction start time as the complete water receding process; The time and spring flow response data of each complete drainage process are fitted according to the exponential drainage relationship to obtain the drainage coefficient. The calculation relationship between groundwater level and drainage coefficient is established based on the groundwater level and drainage coefficient of the first evaluation period of each complete drainage process. The first evaluation period is designated as the initial water storage calculation period, selecting the complete water recession process whose end time is closest to the prediction start time. The drainage coefficient is obtained based on the groundwater level during the initial water storage calculation period and the relationship between the groundwater level and the drainage coefficient. The drainage ratio is calculated based on the drainage coefficient and the evaluation period length. The initial water storage is obtained by dividing the product of the spring flow response data during the initial water storage calculation period and the evaluation period length by the drainage ratio.

5. The method for predicting karst spring flow based on hydrological and meteorological data and physical mechanisms according to claim 4, characterized in that, Methods for establishing water storage records based on initial water storage and replenishment include: An initial water storage record is established based on the initial water storage volume. The initial water storage volume, the calculation period of the initial water storage volume, and the corresponding water receding coefficient are respectively used as the remaining water storage volume, the formation period, and the water receding coefficient of the initial water storage record, and no rainfall event identifier is set. For the evaluation period when the rainfall is greater than zero, a water storage record is established based on the corresponding recharge amount. The recharge amount and the evaluation period are respectively used as the remaining water storage and the formation period of the corresponding water storage record. The drainage coefficient is obtained based on the groundwater level during the formation period and the calculation relationship between the groundwater level and the drainage coefficient. The rainfall event identifier of the evaluation period is used as the rainfall event identifier of the corresponding water storage record. When the same rainfall event includes multiple evaluation periods with rainfall greater than zero, water storage records are established based on the replenishment water volume corresponding to each evaluation period to obtain the water storage record.

6. The method for predicting karst spring flow based on hydrological and meteorological data and physical mechanisms according to claim 5, characterized in that, The methods for calculating the corresponding candidate release volume based on each water storage record and calculating the total release volume based on spring flow response data include: The first and second rainfall events are determined sequentially based on their occurrence time. For the current evaluation period, select water storage records whose formation time is earlier than or equal to the current evaluation period and whose remaining water storage is greater than zero; When the second rainfall event occurs, the water storage records corresponding to the first rainfall event and the second rainfall event will be selected together as the water storage records to participate in the allocation of the total release during the current evaluation period. The water discharge ratio is calculated based on the water discharge coefficient and evaluation period length of each participating water storage record. The product of the corresponding remaining water volume and the water discharge ratio is determined as the candidate release volume. The product of the spring flow response data of the current evaluation period and the evaluation period length is determined as the total release volume. Thus, the candidate release volume corresponding to each participating water storage record and the total release volume corresponding to the current evaluation period are obtained.

7. The method for predicting karst spring flow based on hydrological and meteorological data and physical mechanisms according to claim 6, characterized in that, The total release amount is allocated to the corresponding water storage records according to the proportion of each candidate release amount. Methods for obtaining the allocated release amount include: The remaining water volume in each water storage record is used as the remaining water volume before allocation and the current available remaining water volume, and the cumulative allocation and release volume is set to zero; If the total release amount is not greater than the sum of the remaining water storage before each allocation, the total release amount will be taken as the total release amount before the allocation is completed. The total unallocated amount of water is allocated according to the proportion of the candidate release amount of the storage records with a current available remaining water volume greater than zero to the sum of the candidate release amounts, thus obtaining the water volume to be allocated in this round; The smaller value between the water volume to be allocated in this round and the current available remaining water volume is added to the cumulative allocated release amount. The current available remaining water volume is updated by subtracting the cumulative allocated release amount from the remaining water volume before allocation. The total release amount is updated by subtracting the sum of the cumulative allocated release amount from the total release amount. Repeat until the total unallocated release amount is zero, then use the cumulative allocated release amount as the allocated release amount.

8. The method for predicting karst spring flow based on hydrological and meteorological data and physical mechanisms according to claim 7, characterized in that, The methods for obtaining the water volume to be allocated in this round include: Exclude water storage records where the current available water volume is zero; Calculate the proportion of the candidate release amount corresponding to each water storage record with a current available remaining water volume greater than zero to the sum of all candidate release amounts, and obtain the recalculated proportion of candidate release amounts. The total unallocated amount is allocated according to the recalculated candidate release ratio, and the amount of water to be allocated in this round is obtained for each water storage record with a current available water volume greater than zero.

9. The method for predicting karst spring flow based on hydrological and meteorological data and physical mechanisms according to claim 7, characterized in that, Methods for updating water storage records based on allocated release volumes include: The updated remaining water volume is obtained by subtracting the corresponding allocated release amount from the remaining water volume before allocation in each water storage record; Storage records with a remaining water volume of zero will no longer participate in the total release allocation for subsequent evaluation periods, while storage records with a remaining water volume greater than zero will be retained for the next evaluation period. The process of calculating the replenishment volume, establishing the storage record, calculating the candidate release volume, allocating the total release volume, and updating the remaining storage volume is repeated in chronological order of the evaluation period until the storage record of the last evaluation period before the start of the prediction is completed, thus obtaining the updated storage record.

10. The method for predicting karst spring flow based on hydrological and meteorological data and physical mechanisms according to claim 9, characterized in that, Methods for obtaining karst spring flow prediction data by inputting the acquired predicted hydro-meteorological data and updated water storage records into a pre-trained karst spring flow prediction model include: When the updated water storage record is not empty, calculate the time difference between the start time of the formation period of each updated water storage record and the start time of the prediction to obtain the interval duration; The remaining water storage volume, interval duration, and water receding coefficient are combined to form the water storage record input vector, and arranged from morning to night according to the formation time period to obtain the water storage record input sequence. The predicted rainfall and predicted temperature for each prediction period are extracted from the predicted hydrometeorological data. The predicted rainfall and predicted temperature for each prediction period are combined to form a hydrometeorological input vector, and then arranged in the time order of the prediction period to obtain the hydrometeorological input sequence. The water storage record input sequence and the hydrological and meteorological input sequence are input into the pre-trained karst spring flow prediction model to obtain karst spring flow prediction data.

Images