Water level prediction method and device, electronic equipment and storage medium

Abstract

The invention relates to a water level prediction method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring a multi-source drainage basin feature at a current moment; inputting the multi-source drainage basin features into a plurality of pre-trained prediction models for water line prediction to obtain a plurality of pieces of candidate water line information; and determining target water level line information according to the weight coefficient of each prediction model and the multiple pieces of candidate water level line information. Therefore, firstly, each prediction model performs prediction by using the multi-source drainage basin characteristics, so that the prediction process can comprehensively reflect the complex drainage basin state, and the water level prediction accuracy of each prediction model can be improved. And secondly, a plurality of prediction models are integrated, and a plurality of generated candidate water level lines are fused based on a weight coefficient mechanism to obtain the target water level line, so that the limitation of a single model is effectively overcome, and the accuracy and robustness of water level prediction are remarkably improved.

Description

Technical Field

[0001] This disclosure relates to the field of water resource management technology, and more specifically, to a water level prediction method, apparatus, electronic device, and storage medium. Background Technology

[0002] In the fields of hydrology, water conservancy engineering, and water resources management, accurate prediction of water levels in rivers, reservoirs, and other water bodies is a crucial and fundamental task. Accurate water level information is the core basis for decision-making in areas such as flood control and drought relief, navigation safety, hydropower generation, and ecological regulation.

[0003] Therefore, how to more accurately predict water levels has become an urgent problem to be solved. Summary of the Invention

[0004] In order to overcome the problems existing in the related technologies, this disclosure provides a water level prediction method, apparatus, electronic device and storage medium.

[0005] In a first aspect, this disclosure provides a water level prediction method, the method comprising: acquiring multi-source watershed features at the current time; inputting the multi-source watershed features into multiple pre-trained prediction models to predict water levels, thereby obtaining multiple candidate water level information; and determining target water level information based on the weight coefficients of each prediction model and the multiple candidate water level information.

[0006] Secondly, this disclosure provides a water level prediction device, the device comprising: a watershed feature acquisition module for acquiring multi-source watershed features at the current time; a water level prediction module for inputting the multi-source watershed features into multiple pre-trained prediction models to predict water levels and obtain multiple candidate water level information; and a fusion module for determining target water level information based on the weight coefficients of each prediction model and the multiple candidate water level information.

[0007] Thirdly, this disclosure provides an electronic device comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to implement the steps of the first aspect.

[0008] Fourthly, this disclosure provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0009] In this disclosure, firstly, each prediction model utilizes multi-source watershed characteristics for prediction, ensuring that the prediction process comprehensively reflects the complex watershed conditions, thereby improving the accuracy of water level prediction for each model. Secondly, by integrating multiple prediction models and fusing the multiple candidate water levels generated based on a weighting coefficient mechanism to obtain the target water level, the limitations of a single model are effectively overcome, significantly improving the accuracy and robustness of water level prediction.

[0010] Other features and advantages of this disclosure will be described in detail in the following detailed description section. Attached Figure Description

[0011] The accompanying drawings are provided to further illustrate the present disclosure and form part of the specification. They are used together with the following detailed description to explain the present disclosure, but do not constitute a limitation thereof. In the drawings: Figure 1 A schematic flowchart of a water level prediction method provided in an embodiment of this application is shown.

[0012] Figure 2 A flowchart illustrating a water level prediction method provided in another embodiment of this application is shown.

[0013] Figure 3 This illustration shows a flowchart of a feedforward-feedback hierarchical weight fusion method provided in an embodiment of this application.

[0014] Figure 4 A comparative line graph showing the water level prediction results provided in an embodiment of this application is shown.

[0015] Figure 5 A schematic diagram of the weight trajectory provided in the embodiments of this application is shown.

[0016] Figure 6 A schematic diagram of a calibration curve provided in an embodiment of this application is shown.

[0017] Figure 7 The diagram shows a line graph comparing quantile non-crossing correction before and after, according to an embodiment of this application.

[0018] Figure 8 This is a block diagram of a water level prediction device according to an embodiment of this application.

[0019] Figure 9 This is a block diagram of an electronic device for performing a water level prediction method according to an embodiment of this application.

[0020] Figure 10 This is a storage unit in this application embodiment for storing or carrying program code that implements the water level prediction method according to this application embodiment. Detailed Implementation

[0021] The specific embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit this disclosure.

[0022] It should be noted that all actions involving the acquisition of signals, information, or data in this disclosure are carried out in compliance with the relevant data protection laws and policies of the country where the location is situated, and with authorization from the owner of the relevant device.

[0023] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0024] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0025] In related technologies, the water level of run-of-river hydropower stations is affected by multiple factors such as rainfall pulses, upstream inflow, operation scheduling, and backwater, resulting in strong non-stationarity and multi-timescale coupling in the sequence. Single data-driven models (such as single LSTM, XGBoost, and ARIMA) show significant performance differences under different scenarios; traditional ensemble methods are mostly static weights or simple adaptive methods based solely on recent errors, making it difficult to simultaneously consider offline "scenario differences" (prior advantages) and online "recent performance," and rapid weight switching can easily lead to forecast jitter, affecting scheduling stability and availability.

[0026] Based on this, and to address the shortcomings of the existing technology, the inventors propose a water level prediction method, apparatus, electronic device, storage medium, and program product. The water level prediction method provided in the embodiments of this application will be described in detail below.

[0027] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a water level prediction method according to an embodiment of this application. The following will be combined with... Figure 1 The water level prediction method provided in this application is described in detail. This water level prediction method may include the following steps: Step S110: Obtain the multi-source watershed features at the current moment.

[0028] In this embodiment, a unified timeline can be defined first. Multi-source observation data is collected based on this timeline. For example, at 5-minute intervals; subsequently, a resampling operator is used. This aligns observation data from different sources, which may have different original acquisition frequencies, onto this time axis. This can be represented by the following formula:

[0029] in, Represents a unified timestamp Multi-source observation data collected from above.

[0030] Optionally, multi-source observation data may include at least one of the following: upstream flow, rainfall grid / station data, meteorological elements (such as temperature, humidity, and wind speed), and generator / gate operation.

[0031] Furthermore, data cleaning and repair can be performed on the collected multi-source observation data. Specifically, for missing values ​​in the data, a cascaded repair strategy is adopted, using "nearest neighbor preservation," "linear interpolation," and "robust spline" methods sequentially to fill in the missing values, ensuring data continuity. For outliers in the data, machine learning algorithms such as "robust Z-score method" (using the median and mean absolute deviation (MAD) instead of the mean and standard deviation to improve anti-interference ability) or "isolation forest" are used for identification and truncation to ensure data quality and reliability.

[0032] Finally, after obtaining the cleaned and repaired multi-source observation data... Based on this, through feature constructor Extract multi-source watershed features from the data that are truly useful for model input. This multi-source watershed characteristic It can be represented in the following form:

[0033] Among them, the characteristics of a multi-source watershed include at least one of the following: lag term characteristics, sliding statistical characteristics, cumulative hydrological characteristics, and backwater / tidal indicators and operating condition characteristics.

[0034] Step S120: Input the multi-source watershed features into multiple pre-trained prediction models to predict water levels and obtain multiple candidate water level information.

[0035] In this embodiment, a base model library covering different methodological principles can be pre-constructed. This base model library can include the aforementioned multiple prediction models, all of which are pre-trained offline. These multiple prediction models can include at least one of classical statistical models, machine learning models, and deep time series models. Classical statistical models can include ARIMAX, VARX, etc. Taking ARIMAX as an example, its model form can be expressed as follows:

[0036] Machine learning models can include SVR, XGBoost, LightGBM, CatBoost, etc. These models support quantile regression objectives, and their training process aims to minimize a loss function of the following form:

[0037] This allows for the direct output of water level predictions at different quantiles, thus quantifying the uncertainty of the prediction.

[0038] Deep time series models can include TCN, LSTM, GRU, Transformer, Informer, etc. These models can capture complex temporal dependencies in long sequences.

[0039] Based on this, after completing offline training of multiple prediction models, the multi-source watershed features at the current moment can be obtained. The data is input in parallel to each prediction model in the aforementioned base model library. Each model outputs its predicted value for the target water level, i.e., candidate water level information, thus obtaining multiple candidate water level information.

[0040] Optionally, the candidate water level information may include at least one of the following: point prediction value (i.e., specific water level value), prediction interval (i.e., water level value interval), and prediction distribution.

[0041] Step S130: Determine the target water level information based on the weight coefficients of each prediction model and the multiple candidate water level information.

[0042] Optionally, weighting coefficients can be pre-set for each prediction model. After obtaining candidate water levels from multiple prediction models, the candidate water level information can be weighted and fused according to the weighting coefficients of each model to obtain the target water level information. This provides more accurate and reliable water level information compared to using a single model for water level prediction. It should be noted that the sum of the weighting coefficients of multiple prediction models is 1.

[0043] In some implementations, considering that the results of multiple model fusions may still contain random fluctuations or "jitter," which are not reflections of the actual hydrological process but rather stem from the model's own prediction errors and uncertainties; and that unprocessed water level prediction curves may exhibit physically impossible abrupt changes, a drastic change in water level would interfere with operators' judgment and be unusable for stable control of gates or generating units, the target water level information can be smoothed after step S130. Smoothing methods include, but are not limited to, moving average filtering, low-pass filtering, or constraint-based optimization smoothing. This makes the final smoothed water level curve more consistent with the physical laws of continuous, inertial water movement in hydrology, and provides a stable and reliable water level change trajectory for downstream scheduling decision-making modules (such as gate control and power generation planning), avoiding frequent or oscillating equipment operations caused by data jitter.

[0044] In this embodiment, firstly, each prediction model utilizes multi-source watershed features for prediction, ensuring that the prediction process comprehensively reflects the complex watershed conditions, thereby improving the accuracy of water level prediction for each model. Secondly, by integrating multiple prediction models and fusing the multiple candidate water levels generated based on a weighting coefficient mechanism to obtain the target water level, the limitations of a single model are effectively overcome, significantly improving the accuracy and robustness of water level prediction.

[0045] Please refer to Figure 2 , Figure 2 This is a flowchart illustrating a water level prediction method according to another embodiment of this application. The following will be combined with... Figure 2 The water level prediction method provided in this application is described in detail. This water level prediction method may include the following steps: Step S210: Obtain the multi-source watershed features at the current moment.

[0046] Step S220: Input the multi-source watershed features into multiple pre-trained prediction models to predict water levels and obtain multiple candidate water level information.

[0047] In this embodiment, steps S210 to S220 can refer to the content in the previous embodiments, and will not be repeated here.

[0048] Step S230: Obtain the watershed scenario label at the current moment.

[0049] In this embodiment, after the aforementioned acquisition of multi-source observation data... Then, a scenario constructor can be used. Based on the current moment and a window of time in the past Multi-source observation data is used to generate a scenario context vector that can characterize the current comprehensive state of the watershed. The context vector of this scenario can be regarded as a watershed scenario label. The watershed scenario label can include time period labels (such as season, day / night, weekday / weekend), hydrological dynamic labels (such as the rising edge of heavy rainfall, the falling edge of heavy rainfall, and the dry and stable period), external water stress labels (such as backwater and tidal effects), and engineering operation condition labels (such as gate opening and unit full operation).

[0050] Step S240: Obtain the preset weight coefficients of each prediction model under the watershed scenario label.

[0051] Understandably, during the offline training phase of each prediction model, a preset weight coefficient is set for each prediction model under that watershed scenario label based on its water level prediction capability under different watershed scenario labels.

[0052] For example, the watershed scenario labels include Label 1 and Label 2, and the multiple prediction models include Model 1, Model 2, and Model 3. In the case of Label 1, Model 1 and Model 2 have stronger and more accurate predictive capabilities, so the preset weight coefficients for Model 1, Model 2, and Model 3 can be set to 0.4.

[0053] Step S250: Obtain the predicted water level information for each prediction model at the current time.

[0054] In this embodiment, the specific implementation of step S250 can be found in the content of the foregoing embodiments, and will not be repeated here.

[0055] Step S260: Monitor the real-time water level information at the current moment.

[0056] In this embodiment, physical sensors can be installed at key sections of the watershed (such as in front of a reservoir dam or at an important river control station) to monitor the actual water level at the current moment, i.e., to monitor the real-time water level. These physical sensors include, but are not limited to, radar level gauges, pressure level gauges, ultrasonic level gauges, and float level gauges.

[0057] Step S270: Compare the predicted water level information predicted by each prediction model with the real-time water level information to obtain the comparison result corresponding to each prediction model.

[0058] Specifically, the water level prediction deviation between the predicted water level information and the real-time water level information of each prediction model can be obtained as the comparison result for each prediction model.

[0059] Step S280: Adjust the preset weight coefficient of each prediction model according to the comparison results corresponding to each prediction model.

[0060] Specifically, based on the water level prediction deviation value corresponding to each prediction model, the preset weight coefficient of each prediction model is adjusted. The magnitude of the adjusted preset weight coefficient is negatively correlated with the water level prediction deviation value. That is, the larger the deviation value of the prediction model for the water level, the lower the current prediction accuracy of the prediction model. Therefore, the preset weight coefficient of the prediction model can be reduced. Conversely, the smaller the deviation value of the prediction model for the water level, the higher the current prediction accuracy of the prediction model. Therefore, the preset weight coefficient of the prediction model can be increased.

[0061] In other words, it utilizes a feedforward-feedback hierarchical weight fusion mechanism to combine the long-term scenario advantages of the models with their short-term real-time performance, dynamically and robustly calculating the optimal weights for each model in the final fusion. Simply put, the feedforward answers "in this scenario, who should be stronger?"; the feedback answers "during this period, who is actually stronger?". Gating coefficients allow the relative weights of the two to automatically change with the environment. The formation paths and fusion / gating mechanisms of the feedforward and feedback signals are as follows... Figure 3 As shown, where: The feedforward weights are based on the current The nearest cluster or The nearest neighbor set is used as the score, with the negative of the conditional expected loss of each model as the value, and then processed by softmax. ; based on situational characteristics Obtained through similar scenario retrieval / gating When a peak is identified as "rising edge," a model that can quickly follow the peak is preferred. The feedforward weights can be expressed as:

[0062] Or use a gated network (softmax).

[0063] Feedback weight window (Approximately 6 hours), Exponential weighting coefficient The weighted sum of CRPS and Pinball Loss is used as the online loss, and normalized to... ; and using Ledoit–Wolf contraction to construct the residual covariance, solving for the minimum variance combination as an alternative. Feedback is "recent fact". In the downstream branch of 3, EWMA / covariance contraction generates .like Figure 5 As shown, the weights shift smoothly across different stages without skipping steps. Specifically, the feedback weights can be expressed as:

[0064] Robust option: Residual covariance minimum variance combination

[0065] Fusion and Gating:

[0066] Gating coefficient Adaptable to uncertainty and anomalies: ,in This refers to the weight entropy or quantile dispersion. This is an anomaly indication. In other words, multiplicative gating is used. Adaptive to anomalies and uncertainties; assume With minimum residence ;right Apply total variation regularization, and use Kalman smoothing if necessary. Illustration: The fusion / gating node is located at the center of Figure 3; the smoothing trajectory in Figure 5 intuitively demonstrates the combined effect of "rate constraint + dwell + TV" - the curve is not jittery, but can still keep up with process changes.

[0067] Inertia / rate limitations: .

[0068] Hysteresis and Retention: Top-1 Weighted Entry Threshold Exit threshold and minimum dwell step size .

[0069] Total variation regularization:

[0070] Secure projection: When an exception is triggered , .

[0071] Step S290: Determine the target water level information based on the preset weight coefficients of each of the adjusted prediction models and the multiple candidate water level information.

[0072] Finally, based on the preset weight coefficients of each adjusted prediction model, the multiple candidate water level information are weighted and fused to obtain the target water level information. Based on this, it is possible to... Figure 4 , Figure 5 , Figure 6 as well as Figure 7 This demonstrates the technical effectiveness of the water level prediction method proposed in this application.

[0073] like Figure 4 As shown, the water level forecast curves of different methods are compared: the single model has a large deviation between the peak and the fall section; the static integration tends to be central but the response is lagging; the results and phases of the prediction of the peak part of this application are closer to the reality.

[0074] like Figure 5 As shown, the weight trajectory of this method is as follows: weight switching has hysteresis and minimum dwell time, avoiding "jitter" while retaining the traceability of mutations.

[0075] like Figure 6 As shown, the reliability (calibration curve) is as follows: a curve that is closer to the nominal 90% range indicates that the actual 90% coverage is closer to the true 90% coverage.

[0076] like Figure 7 As shown, quantile non-cross-correction (before / after): the correction is monotonic throughout, avoiding physical inconsistencies caused by confidence band "cross-through".

[0077] In this embodiment, firstly, multiple prediction models are integrated, leveraging their complementary advantages to generate the final water level line through weighted fusion. This effectively overcomes the limitation of unstable prediction performance of a single model in complex watershed environments, significantly improving the accuracy of the prediction results. Secondly, an innovative scenario-aware and dynamic weight adjustment mechanism is introduced. This mechanism can invoke corresponding preset weights based on real-time scenario labels of the watershed and continuously optimize the weight allocation based on the comparison between the prediction results and the measured water level. This allows the model to maintain its adaptive capability under different hydrological conditions, enhancing the intelligence level and scenario adaptability of the prediction system.

[0078] Please refer to Figure 8 The diagram illustrates a structural block diagram of a water level prediction device 300 according to an embodiment of this application. The device 300 may include: a watershed feature acquisition module 310, a water level prediction module 320, and a fusion module 330.

[0079] The watershed feature acquisition module 310 is used to acquire the multi-source watershed features at the current moment.

[0080] The water level prediction module 320 is used to input the multi-source watershed features into multiple pre-trained prediction models to predict the water level and obtain multiple candidate water level information.

[0081] The fusion module 330 is used to determine the target water level information based on the weight coefficients of each prediction model and the multiple candidate water level information.

[0082] Optionally, the fusion module 330 may be specifically used to: perform weighted fusion of the multiple candidate water level information according to the weight coefficients of each of the prediction models to obtain the target water level information.

[0083] Optionally, the water level prediction device 300 may further include a weighting coefficient determination module. Specifically, the weighting coefficient determination module may be used to obtain the current watershed scenario label before determining the target water level information based on the weighting coefficients of each prediction model and the multiple candidate water level information; and to obtain the preset weighting coefficients of each prediction model under the watershed scenario label.

[0084] Optionally, the water level prediction device 300 may further include a weighting coefficient adjustment module. Specifically, after obtaining the preset weighting coefficients of each prediction model under the watershed scenario label, the weighting coefficient adjustment module may: obtain the predicted water level information predicted by each prediction model for the current time; monitor the real-time water level information at the current time; compare the predicted water level information predicted by each prediction model with the real-time water level information to obtain a comparison result corresponding to each prediction model; and adjust the preset weighting coefficients of each prediction model according to the comparison result corresponding to each prediction model.

[0085] Optionally, the fusion module 330 can be specifically used to: determine the target water level information based on the preset weight coefficients of each of the adjusted prediction models and the multiple candidate water level information.

[0086] Optionally, the comparison result includes the water level prediction deviation value; the weight coefficient adjustment module can be specifically used to adjust the preset weight coefficient of each prediction model according to the water level prediction deviation value corresponding to each prediction model, and the magnitude of the adjusted preset weight coefficient is negatively correlated with the water level prediction deviation value.

[0087] Optionally, the water level prediction device 300 may further include a smoothing module. Specifically, the smoothing module may be used to smooth the target water level information after determining the target water level information based on the weight coefficients of each prediction model and the multiple candidate water level information.

[0088] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described device and module can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0089] In the several embodiments provided in this application, the coupling between modules can be electrical, mechanical, or other forms of coupling.

[0090] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0091] In summary, firstly, by integrating multiple prediction models and leveraging their complementary advantages, a weighted fusion method is used to generate the final water level, effectively overcoming the limitation of unstable prediction performance of a single model in complex watershed environments and significantly improving the accuracy of the prediction results. Secondly, an innovative scenario-aware and dynamic weight adjustment mechanism is introduced, which can invoke corresponding preset weights based on real-time scenario labels of the watershed and continuously optimize the weight allocation based on the comparison between the prediction results and the measured water level. This enables the model to maintain its adaptability under different hydrological conditions, enhancing the intelligence level and scenario adaptability of the prediction system.

[0092] Please see Figure 9 , Figure 9 This is a block diagram illustrating an electronic device 400 according to an exemplary embodiment. Figure 9 As shown, the electronic device 400 may include: a processor 401 and a memory 402.

[0093] The processor 401 controls the overall operation of the electronic device 400 to complete all or part of the steps in the water level prediction method described above. The memory 402 stores various types of data to support the operation of the electronic device 400. This data may include, for example, instructions for any application or method operating on the electronic device 400, and application-related data such as contact data, sent and received messages, pictures, audio, video, etc. The memory 402 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0094] In an exemplary embodiment, the electronic device 400 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the water level prediction method described above.

[0095] Please refer to Figure 10 This diagram illustrates a structural block diagram of a computer-readable storage medium provided in an embodiment of this application. The computer-readable medium 500 stores program code that can be called by a processor to execute the methods described in the above method embodiments.

[0096] The computer-readable storage medium 500 may be an electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. Optionally, the computer-readable storage medium 500 includes a non-transitory computer-readable storage medium. The computer-readable storage medium 500 has storage space for program code 510 that performs any of the method steps described above. This program code can be read from or written to one or more computer program products. The program code 510 may be compressed, for example, in a suitable form.

[0097] In some embodiments, a computer program product or computer program is provided, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the steps in the above-described method embodiments.

[0098] The preferred embodiments of this disclosure have been described in detail above with reference to the accompanying drawings. However, this disclosure is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this disclosure, various simple modifications can be made to the technical solutions of this disclosure, and these simple modifications all fall within the protection scope of this disclosure.

[0099] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, this disclosure will not describe the various possible combinations separately.

[0100] Furthermore, various different embodiments of this disclosure can be combined in any way, as long as they do not violate the spirit of this disclosure, they should also be regarded as the content disclosed in this disclosure.

Claims

1. A method for predicting water levels, characterized in that, The method includes: Obtain the multi-source watershed characteristics at the current moment; The multi-source watershed features are input into multiple pre-trained prediction models to predict water levels, thereby obtaining multiple candidate water level information. The target water level information is determined based on the weight coefficients of each prediction model and the information of the multiple candidate water levels.

2. The method according to claim 1, characterized in that, The step of determining the target water level information based on the weight coefficients of each prediction model and the multiple candidate water level information includes: Based on the weight coefficients of each prediction model, the multiple candidate water level information are weighted and fused to obtain the target water level information.

3. The method according to claim 1, characterized in that, Before determining the target water level information based on the weight coefficients of each prediction model and the multiple candidate water level information, the method further includes: Obtain the current watershed scenario label; Obtain the preset weight coefficients for each prediction model under the watershed scenario label.

4. The method according to claim 3, characterized in that, After obtaining the preset weight coefficients of each prediction model under the watershed scenario label, the method includes: Obtain the predicted water level information for each of the prediction models at the current time; Monitor the real-time water level information at the current moment; The predicted water level information predicted by each prediction model is compared with the real-time water level information to obtain the comparison result corresponding to each prediction model. Based on the comparison results corresponding to each prediction model, the preset weight coefficients of each prediction model are adjusted.

5. The method according to claim 4, characterized in that, The step of determining the target water level information based on the weight coefficients of each prediction model and the multiple candidate water level information includes: The target water level information is determined based on the preset weight coefficients of each of the adjusted prediction models and the information of the multiple candidate water levels.

6. The method according to claim 4, characterized in that, The comparison results include the water level prediction deviation value; The step of adjusting the preset weight coefficients of each prediction model based on the comparison results corresponding to each prediction model includes: Based on the water level prediction deviation value corresponding to each prediction model, the preset weight coefficient of each prediction model is adjusted, and the magnitude of the adjusted preset weight coefficient is negatively correlated with the water level prediction deviation value.

7. The method according to any one of claims 1-6, characterized in that, After determining the target water level information based on the weight coefficients of each prediction model and the multiple candidate water level information, the method further includes: The target water level information is smoothed.

8. A water level prediction device, characterized in that, The device includes: The watershed feature acquisition module is used to acquire the multi-source watershed features at the current moment; The water level prediction module is used to input the multi-source watershed features into multiple pre-trained prediction models to predict the water level and obtain multiple candidate water level information. The fusion module is used to determine the target water level information based on the weight coefficients of each prediction model and the multiple candidate water level information.

9. An electronic device, characterized in that, include: A memory on which computer programs are stored; A processor for executing the computer program in the memory to implement the steps of the method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 7.

Images