Prediction method of irrigation district outflow considering hysteresis and digital twin irrigation district system

Abstract

This invention belongs to the field of water resource management technology and discloses a method for predicting irrigation runoff volume in irrigation districts that considers lag time, as well as a digital twin irrigation district system. The method includes determining a first lag time between runoff and precipitation based on measured runoff and precipitation sequences, and determining a second lag time between runoff and irrigation based on measured runoff and irrigation water sequences. The measured sequences are proportionally divided into training and validation sequences. The runoff training sequence is used as the output, and the precipitation training sequence (corresponding to the first lag time) and the irrigation water training sequence (corresponding to the second lag time) are used as inputs to train the runoff volume prediction model. The precipitation validation sequence and the irrigation water validation sequence are then input into the trained runoff volume prediction model to output the runoff volume. This invention can improve the accuracy of runoff volume prediction and, combined with a digital twin irrigation district system, provides effective technical support for irrigation district water resource scheduling and refined management.

Description

Technical Field

[0001] This invention belongs to the field of water resource management technology, and specifically discloses a method for predicting irrigation runoff volume and a digital twin irrigation district system that takes into account lag. Background Technology

[0002] Irrigation runoff is a common phenomenon in agricultural irrigation, mainly due to the continued use of traditional surface irrigation methods and a lack of scientific management measures in most irrigation districts. Runoff results in significant water waste and reduces water resource utilization efficiency. However, the complexity of the composition and influencing factors of irrigation runoff makes its prediction increasingly difficult.

[0003] Current methods for predicting irrigation runoff mainly include the runoff coefficient method, numerical simulation, and statistical analysis. The runoff coefficient method characterizes the relationship between runoff and irrigation water volume using the ratio of runoff to irrigation water volume. This method is simple, practical, and provides some guidance for agricultural production. However, it only uses a single ratio to broadly represent the relationship between runoff and irrigation water, often neglecting the composition and spatiotemporal characteristics of runoff. Furthermore, it is affected by the cumulative errors of meteorological conditions, user water usage, irrigation canal water utilization coefficients, and field water utilization coefficients, leading to significant deviations between estimated and actual runoff volumes. Numerical simulation can describe the runoff formation process and its underlying mechanisms, providing a new approach for runoff simulation and prediction. However, the model input requires distributed parameters from hydrology, meteorology, soil, crops, and irrigation and drainage systems, and model construction is relatively complex, with simulation accuracy greatly affected by the parameters. Statistical analysis, based on actual monitoring data, predicts irrigation runoff through regression analysis or machine learning models, directly reflecting the runoff patterns and the relationship between runoff and various influencing factors. However, the current study did not consider the dynamic relationship between water runoff and precipitation and irrigation water volume, and ignored the lag characteristics of water runoff in irrigation areas relative to precipitation and irrigation, resulting in low accuracy of water runoff prediction using statistical analysis methods. Summary of the Invention

[0004] The purpose of this invention is to provide a method for predicting irrigation runoff volume in irrigation districts that takes into account lag, and a digital twin irrigation district system.

[0005] A first aspect of the present invention provides a method for predicting irrigation runoff volume considering lag, comprising:

[0006] Step 1: Obtain the measured sequence of the irrigation area, which includes the measured sequence of precipitation, the measured sequence of irrigation water, and the measured sequence of water discharge.

[0007] Step 2: Determine the first lag time of water discharge relative to precipitation based on the measured water discharge sequence and the measured precipitation sequence; determine the second lag time of water discharge relative to irrigation based on the measured water discharge sequence and the measured irrigation water sequence.

[0008] Step 3: Divide the measured sequence into training sequence and validation sequence according to the ratio. Take the water receding volume training sequence as the output and take the precipitation training sequence with the first lag time corresponding to the water receding volume training sequence and the irrigation water training sequence with the second lag time corresponding to the water receding volume training sequence as the input to train the water receding volume prediction model.

[0009] Step 4: Determine the time period for predicting the amount of water receding, and denot it as the prediction period. Input the precipitation verification sequence with the first lag time corresponding to the prediction period and the irrigation water verification sequence with the second lag time corresponding to the prediction period into the trained water receding volume prediction model, output the water receding volume prediction sequence for the prediction period, and correct the water receding volume prediction model by combining the water receding volume prediction sequence and the water receding volume verification sequence for the prediction period.

[0010] Preferably, the first lag time between the receding water and the precipitation is determined based on the measured receding water sequence and the measured precipitation sequence, specifically as follows:

[0011] Determine the first continuous wavelet transform coefficients of the measured water receding volume sequence, and determine the second continuous wavelet transform coefficients of the measured precipitation volume sequence;

[0012] The cross-wavelet spectrum between the measured sequence of receding water volume and the measured sequence of precipitation is determined based on the first continuous wavelet transform coefficient and the second continuous wavelet transform coefficient.

[0013] The first lag time between the receding water and the precipitation is determined based on the cross-wavelet spectrum.

[0014] Preferably, the first lag time between the receding water and the precipitation is determined based on the cross-wavelet spectrum, specifically as follows:

[0015] Determine the cross wavelet power corresponding to each time-frequency grid point of the cross wavelet spectrum;

[0016] Determine the phase of the time-frequency grid point corresponding to the maximum power of the cross wavelet;

[0017] The first lag time between the receding water and the precipitation is determined based on the phase.

[0018] Preferably, the second lag time between drainage and irrigation is determined based on the measured drainage volume sequence and the measured irrigation volume sequence, specifically as follows:

[0019] Determine the first continuous wavelet transform coefficients of the measured water discharge sequence and the third continuous wavelet transform coefficients of the measured irrigation water sequence;

[0020] The cross-wavelet spectrum between the measured sequence of water discharge and the measured sequence of irrigation water is determined based on the first continuous wavelet transform coefficient and the third continuous wavelet transform coefficient.

[0021] The second lag time of the drainage relative to the irrigation is determined by the cross-wavelet spectrum between the measured drainage volume sequence and the measured irrigation volume sequence.

[0022] Preferably, the water discharge prediction model is corrected by combining the predicted water discharge sequence and the water discharge verification sequence for the predicted period, specifically as follows:

[0023] The absolute value of the error between the predicted water volume sequence and the verified water volume sequence for the predicted period is determined. When the absolute value of the error is greater than a preset threshold, the parameters of the water volume prediction model are adjusted.

[0024] Preferably, the parameters of the water discharge prediction model are adjusted as follows:

[0025] Based on the water receding volume verification sequence for the predicted period, the parameters in the water receding volume prediction model are iteratively corrected using an optimization algorithm, so that the error between the water receding volume prediction sequence output by the water receding volume prediction model and the corresponding water receding volume verification sequence is less than or equal to the preset threshold.

[0026] Preferably, after step 4, the method further includes:

[0027] Step 5: Obtain the measured sequence for the next time period in the irrigation area, and add the measured sequence for the next time period to the original measured sequence. Repeat steps 1 to 5.

[0028] Preferably, the water discharge prediction model is a recurrent neural network model.

[0029] Preferably, the recurrent neural network model is a long short-term memory neural network model.

[0030] A second aspect of the present invention provides a digital twin irrigation district system, including a data monitoring module and a water discharge prediction module connected to the data monitoring module;

[0031] The data monitoring module is used to acquire measured precipitation sequences, measured irrigation water sequences, and measured drainage water sequences.

[0032] The drainage volume prediction module uses the above-mentioned irrigation area drainage volume prediction method that takes into account lag, and is used to predict the drainage volume based on the measured precipitation sequence and measured irrigation water sequence obtained by the data monitoring module.

[0033] The irrigation district drainage prediction method and digital twin irrigation district system of the present invention, which take into account the lag effect, have the following advantages compared with the prior art:

[0034] On the one hand, this invention addresses the phenomenon that drainage in irrigation areas lags significantly behind influencing factors such as precipitation and irrigation. First, it calculates the lag time of drainage relative to each influencing factor and constructs an explicit relationship that integrates the lag time, using drainage volume as the output factor and precipitation and irrigation volume as input factors. This intuitively reflects the dynamic relationship between drainage and factors such as precipitation and irrigation, while also improving the computational efficiency of the next step in model prediction. Secondly, this invention employs LSTM, combining the explicit structural formulas of the input drainage and factors such as precipitation and irrigation, to effectively capture the temporal dependence between drainage and influencing factors. This more accurately handles the characteristics of continuous and periodic changes and the nonlinear relationships between factors, improving the accuracy of model prediction.

[0035] On the other hand, this invention embeds a drainage volume prediction model into a digital twin irrigation district system. By periodically iteratively optimizing the prediction model using real-time monitoring data, and combining it with irrigation district scheduling plans for scenario-based prediction and simulation of the drainage process, the reliability of drainage volume prediction is improved. This provides effective technical support for irrigation district water resource scheduling and refined management. Ultimately, this invention can improve the prediction accuracy of irrigation district drainage, enhance the analysis and prediction capabilities of the drainage process, improve the efficiency of irrigation district water resource management and scheduling, and achieve refined management and optimized control of the irrigation district operation process. Attached Figure Description

[0036] Figure 1 This is a flowchart of a method for predicting irrigation runoff volume that takes into account lag, as an embodiment of the present invention.

[0037] Figure 2 The cross-wavelet spectrograms in the irrigation district drainage prediction method considering lag in the embodiments of the present invention are shown in (a) and (b) respectively. (a) is the cross-wavelet spectrogram between the monthly drainage sequence and the monthly precipitation sequence, and (b) is the cross-wavelet spectrogram between the monthly drainage sequence and the monthly irrigation water sequence.

[0038] Figure 3 The figure shows the result of the predicted water discharge sequence obtained by the water discharge prediction model during the training and verification phases in the water discharge prediction method for irrigation districts that takes into account the lag in the embodiments of the present invention. Detailed Implementation

[0039] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0040] The first aspect of this invention provides a method for predicting irrigation runoff volume that takes into account lag, such as... Figure 1 As shown, it includes:

[0041] Step 1: Obtain the measured sequence of the irrigation area, which includes the measured sequence of precipitation, the measured sequence of irrigation water, and the measured sequence of water discharge.

[0042] In this embodiment of the invention, the measured precipitation sequence is derived from the historical daily records of the meteorological station, the measured irrigation water sequence is derived from the daily flow of each branch canal, and the measured drainage water sequence is derived from the flow monitored at the outlet of the drainage ditch in the irrigation area.

[0043] After obtaining the above-mentioned measured precipitation sequence, measured irrigation water sequence, and measured drainage sequence, the embodiments of the present invention process the above data into the same time scale.

[0044] Step 2: Determine the first lag time of water discharge relative to precipitation based on the measured sequence of water discharge and precipitation, and determine the second lag time of water discharge relative to irrigation based on the measured sequence of water discharge and irrigation water.

[0045] Step 2.1: Determine the first lag time between the receding water volume and the precipitation volume based on the measured receding water volume sequence and the measured precipitation volume sequence, specifically as follows:

[0046] Step 2.1.1: Determine the first continuous wavelet transform coefficients of the measured water receding volume sequence and the second continuous wavelet transform coefficients of the measured precipitation volume sequence.

[0047] In this embodiment of the invention, a continuous wavelet transform is performed on the measured sequence of receding water volume to obtain the first continuous wavelet transform coefficients.

[0048] In this embodiment of the invention, a continuous wavelet transform is performed on the measured precipitation sequence to obtain the second continuous wavelet transform coefficients.

[0049] Step 2.1.2: Determine the cross-wavelet spectrum between the measured receding water volume sequence and the measured precipitation sequence based on the first and second continuous wavelet transform coefficients, as shown in formula (1):

[0050] (1)

[0051] In the formula, for Measured sequence of water receding during the time period With the measured precipitation sequence Cross-wavelet spectrum between; for Measured sequence of water receding during the time period The first continuous wavelet transform coefficients, The scaling parameter in wavelet transform; The time position parameter represents the translation of the wavelet function in the time domain; for Measured precipitation sequence over a period of time Second continuous wavelet transform coefficients .

[0052] Step 2.1.3: Determine the first lag time of receding water relative to precipitation based on the cross-wavelet spectrum, specifically as follows:

[0053] Step 2.1.3.1: Determine the cross-wavelet spectrum Cross wavelet power corresponding to each time-frequency grid point :

[0054] (2)

[0055] In the formula, for Measured sequence of water receding during the time period and measured precipitation sequence At the corresponding scale Time and location The resonant energy below, i.e., the cross wavelet power.

[0056] Step 2.1.3.2: Determine the phase of the time-frequency grid point corresponding to the maximum power of the cross wavelet.

[0057] This invention utilizes a computer to traverse all time-frequency grid points and automatically searches for cross-wavelet power through numerical comparison. The optimal time-frequency grid point corresponding to the global maximum value There is no need for manual image reading and grid location identification; the phase at the grid point can then be obtained using formula (3).

[0058] (3)

[0059] In the formula, Optimal time-frequency grid points phase, for Measured sequence of water receding during the time period With the measured precipitation sequence At the optimal time-frequency grid point Cross wavelet spectrum at the location.

[0060] Then based on phase Determine the first lag time of receding water relative to precipitation. As shown in formula (4):

[0061] (4)

[0062] In the formula, For frequency, its relationship with Inversely proportional.

[0063] Step 2.2: Determine the second lag time of drainage relative to irrigation based on the measured drainage volume sequence and the measured irrigation volume sequence, specifically as follows:

[0064] Determine the first continuous wavelet transform coefficients of the measured sequence of water discharge and the third continuous wavelet transform coefficients of the measured sequence of irrigation water.

[0065] The cross-wavelet spectrum between the measured sequence of water discharge and the measured sequence of irrigation water is determined based on the first and third continuous wavelet transform coefficients.

[0066] The second lag time of the drainage relative to the irrigation was determined by cross-wavelet spectrum between the measured drainage volume sequence and the measured irrigation volume sequence.

[0067] The method for determining the second lag time in this embodiment of the invention is the same as the method for determining the first lag time, and will not be described again here.

[0068] Step 3: Divide the measured sequence into training sequence and validation sequence according to the ratio. Take the water receding volume training sequence as the output and take the precipitation training sequence corresponding to the first lag time and the irrigation water training sequence corresponding to the second lag time as the input to train the water receding volume prediction model.

[0069] The aforementioned water discharge prediction model is a recurrent neural network model. Exemplary models include Long Short-Term Memory (LSTM) neural network models and Bidirectional Recurrent Neural Network (Bi-RNN) models. In this embodiment of the invention, LSTM is preferred.

[0070] This invention considers dynamic factors such as precipitation and irrigation, and leverages the advantages of LSTM in constructing complex nonlinear relationships, handling long-term dependencies, and adaptive learning capabilities to develop a recession flow prediction model that incorporates explicit structural hysteresis features, thereby improving prediction accuracy. The constructed explicit relationship is as follows:

[0071] (5)

[0072] In the formula, For output elements; For the first The first period One input element; The output elements lag behind the input elements by a certain time.

[0073] Because the units and magnitudes of elements such as water runoff, precipitation, and irrigation are different, to avoid the influence of weighting and to improve the calculation speed of the water runoff prediction model, this embodiment of the invention normalizes the measured sequences of water runoff, precipitation, and irrigation water volume using the min-max normalization method. After normalization, the dataset containing the measured sequences of water runoff, precipitation, and irrigation water volume is divided into training and validation sets proportionally. Training parameters (such as the number of hidden layers, the number of neurons in the hidden layers, the learning rate, the dropout rate, and the batch size) are set, and the parameters are continuously adjusted through cross-validation.

[0074] In this embodiment of the invention, the water receding volume training sequence in the training set is used as the output, and the precipitation training sequence with the first lag time corresponding to the water receding volume training sequence and the irrigation water training sequence with the second lag time corresponding to the water receding volume training sequence are used as the input to train the LSTM model.

[0075] The LSTM model has a memory block, which includes storage space. Cell state information at specific times Used to control The Forget Gate of Time-Based Information Transmission Input gate and output gate The forget gate is controlled by the previous cell state. Which part of the information will be forgotten; the input gate controls the latent cell state. Which information is allowed to pass through to update the current cell state? The output gate controls the current cell state. Which information is allowed to flow into the new ( (Time) Hidden Cell State The specific process is as follows:

[0076] First, calculate the forget gate:

[0077] (6)

[0078] In the formula, It is the sigmoid activation function; , and These are the input weight matrix, cyclic weight matrix, and bias vector of the forget gate, respectively; for The input vector at time step 1. The hidden cell state in the LSTM. initial value Calculate the latent cell state based on the current input and the previous hidden cell state. :

[0079] (7)

[0080] in , , These are the input weight matrix, cyclic weight matrix, and bias vector for the potential cell state, respectively.

[0081] Next, calculate the input gate:

[0082] (8)

[0083] in , , These are the input weight matrix, cyclic weight matrix, and bias vector of the input gate, respectively; the current cell state is updated based on the above formulas. :

[0084] (9)

[0085] In the formula, This indicates element-wise multiplication.

[0086] Finally, calculate the output gate:

[0087] (10)

[0088] in , , Let the input weight matrix, cyclic weight matrix, and bias vector of the output gate be represented, respectively. Based on the output gate, the current hidden cell state is calculated. :

[0089] (11)

[0090] Connect the traditional dense layer network to the last hidden cell state and compute the final prediction result of the LSTM. :

[0091] (12)

[0092] in and These are the weight matrix and bias vector from the hidden layer to the output layer, respectively; This refers to the hidden cellular state of complete temporal long-term memory.

[0093] Step 4: Determine the time period for the predicted water volume, denoted as the prediction period. Input the precipitation verification sequence corresponding to the first lag time of the prediction period and the irrigation water verification sequence corresponding to the second lag time of the prediction period into the trained water volume prediction model. Output the water volume prediction sequence for the prediction period. Combine the water volume prediction sequence and the water volume verification sequence for the prediction period to correct the water volume prediction model.

[0094] In this embodiment of the invention, after the drainage volume prediction model is trained, the precipitation verification sequence and irrigation water verification sequence from the validation set are input into the trained drainage volume prediction model, which outputs a drainage volume prediction sequence. The drainage volume prediction sequence is then compared with the drainage volume validation sequence to evaluate the accuracy of the drainage volume prediction model. Exemplarily, this embodiment of the invention uses MRE, RMSE, and NSE to evaluate the accuracy of the drainage volume prediction model.

[0095] (13)

[0096] (14)

[0097] (15)

[0098] In the formula, The number of samples; This is the actual value; This is a predicted value; To verify the sequence mean.

[0099] This invention's embodiments combine the predicted water discharge sequence and the water discharge verification sequence for the predicted period to correct the water discharge prediction model, specifically as follows:

[0100] Determine the absolute value of the error between the predicted water volume sequence and the verified water volume sequence for the predicted period. When the absolute value of the error exceeds a preset threshold, adjust the parameters of the water volume prediction model.

[0101] The parameters of the water discharge prediction model were adjusted as follows:

[0102] Using the flood discharge verification sequence for the predicted period as a benchmark, the parameters in the flood discharge prediction model are iteratively corrected using an optimization algorithm, so that the error between the flood discharge prediction sequence output by the model and the corresponding flood discharge verification sequence is less than or equal to a preset threshold. For example, the optimization algorithm used in this embodiment of the invention is a genetic algorithm or a simulated annealing algorithm, etc.

[0103] To ensure that the parameters in the water discharge prediction model can be continuously optimized with updated measured sequences, this embodiment of the invention further includes the following after step 4:

[0104] Step 5: Obtain the measured sequence for the next time period in the irrigation area, and add the measured sequence for the next time period to the original measured sequence. Repeat steps 1 to 5.

[0105] Existing water discharge prediction models are usually used as independent calculation tools and lack deep integration with irrigation district information management systems. It is difficult to achieve periodic updates and iterative optimization of water discharge prediction models based on real-time monitoring data (i.e., measured sequence data), and it is also difficult to carry out scenario-based prediction and simulation of the water discharge process in conjunction with irrigation district scheduling schemes. This limits the application effect of water discharge prediction models in irrigation district operation and management.

[0106] With the application of digital twin technology in irrigation district operation and management, some irrigation districts have established digital twin systems for water diversion scheduling, water distribution management and engineering operation monitoring. However, the existing digital twin irrigation district systems mainly focus on monitoring and scheduling water diversion volume, water transmission process and water use process. The simulation and prediction functions for water discharge process are relatively insufficient, and a technical system for deep integration of water discharge prediction model and digital twin system has not yet been formed.

[0107] Meanwhile, existing digital twin irrigation district systems mainly rely on monitoring statistics or post-event analysis for water discharge volume, lacking a prediction mechanism driven by continuously updated monitoring data, making it difficult to achieve periodic iterative optimization of the water discharge volume prediction model. On the other hand, existing systems also struggle to combine water discharge volume prediction results with irrigation district scheduling plans to conduct scenario-based prediction and simulation of the water discharge process, lacking intuitive display and analysis methods, thus limiting the application effect of the water discharge volume prediction model in the refined scheduling of irrigation districts.

[0108] Therefore, it is necessary to construct an irrigation district system based on digital twin technology, combining a multi-factor and time-lag-integrated drainage volume prediction model with the digital twin irrigation district system. This enables scenario-based prediction and simulation of the drainage process, and uses monitoring data to correct the prediction results. By periodically updating the model parameters, iterative optimization of the drainage volume prediction model is achieved, thereby improving the accuracy of drainage volume prediction. Therefore, the second aspect of this invention provides a digital twin irrigation district system, the construction process of which is as follows:

[0109] Acquire basic data for the irrigation district and establish a digital twin data base. The basic data includes the irrigation district canal system topology, canal section / node parameters, engineering facility parameters such as gates and pumping stations, measurement section and monitoring point information, irrigation district zoning and water use unit information, etc.

[0110] Construct a spatial and object model of the digital twin irrigation district to form a digital representation of the irrigation district's engineering objects, canal connectivity, and operational status, which will support subsequent monitoring data mapping, simulation calculations, and business applications.

[0111] A digital twin irrigation district system platform is constructed. Based on the digital twin data base and the irrigation district space and object model, the various functional modules of the system are integrated, developed and deployed to form a unified digital twin irrigation district operation platform.

[0112] The digital twin irrigation district system of this invention includes a data monitoring module and a water discharge prediction module connected to the data monitoring module.

[0113] The data monitoring module acquires measured sequences of precipitation, irrigation water, and drainage, and preprocesses these sequences. The drainage prediction module uses the aforementioned drainage prediction method for irrigation districts, which considers lag, to predict drainage volume based on the measured precipitation and irrigation water sequences obtained by the data monitoring module. The drainage prediction module is encapsulated as a callable model service or model component, defining input data items, output data items, and invocation methods, and is deployed into the digital twin irrigation district system.

[0114] This invention embodiment organizes and stores monitoring data, business data, and model data in a unified manner.

[0115] The digital twin irrigation district system of this invention also includes a supply and demand analysis module, a water resource allocation module, a water distribution scheduling module, and a channel water flow simulation module. These modules, together with the data monitoring module and the water discharge prediction module, jointly realize the analysis, scheduling, and prediction functions of the irrigation district operation process.

[0116] Based on the irrigation district operation data obtained by the data monitoring module, the present invention calculates and analyzes the data through the supply and demand analysis module, the water resource allocation module, and the water distribution scheduling module to generate an irrigation district scheduling plan under the corresponding scenario.

[0117] Based on the scheduling scheme, the irrigation area water flow process is simulated and calculated using the channel water flow simulation module to obtain the simulation results of the irrigation area water conveyance and distribution process.

[0118] Furthermore, the water flow simulation results are used as input to the drainage volume prediction model. The drainage volume prediction model, which has been periodically iteratively optimized, is then called to predict the drainage volume under the corresponding scheduling scheme. Finally, the water flow process and drainage volume prediction results under the scheduling scheme are comprehensively evaluated to realize the simulation and analysis of the drainage process under different scheduling scenarios.

[0119] This invention addresses the problems of existing methods for predicting water discharge volume lacking deep integration with digital management systems for irrigation districts and struggling to achieve scenario-based prediction and iterative model optimization. It provides a digital twin irrigation district system based on digital twin technology, which embeds a water discharge volume prediction method that considers lag into the digital twin irrigation district system. Through real-time data-driven operation, synchronous virtual and real operation, and feedback correction, it achieves scenario-based prediction and simulation of the water discharge process, thereby improving the accuracy of water discharge volume prediction and the efficiency of irrigation district water resource management.

[0120] The effectiveness of the method of the present invention will be described in detail below with more specific embodiments.

[0121] The irrigation district drainage prediction method considering lag in this invention specifically includes the following steps:

[0122] S1. Data collection and preprocessing.

[0123] S1.1 Collect measured sequences of precipitation, irrigation water, and drainage within the irrigation district, ensuring that the lengths of the precipitation and irrigation measured sequences include the drainage measured sequences, to form the original dataset.

[0124] S1.2. Organize the raw data into monthly-scale data.

[0125] S2. Construct a cross-wavelet model and calculate the lag time.

[0126] S2.1. Based on the processed monthly-scale data, establish cross-wavelet models between monthly water runoff and monthly precipitation, and between monthly water runoff and monthly irrigation, respectively, to calculate the lag time of water runoff relative to precipitation and irrigation. The specific process in step S2.1 is as follows:

[0127] S2.1.1 Calculate the cross-wavelet spectra between the measured monthly water loss series and the measured monthly precipitation series, and between the measured monthly water loss series and the measured monthly irrigation water series, respectively. Figure 2 As shown.

[0128] S2.1.2 Calculate the phase in the corresponding cross-wavelet spectrum of the measured monthly water loss sequence and the measured monthly precipitation sequence, and the phase in the corresponding cross-wavelet spectrum of the measured monthly water loss sequence and the measured monthly irrigation water sequence.

[0129] S2.1.3. Based on the phase, calculate the lag time of the receding water relative to the precipitation and irrigation water, respectively.

[0130] S3. Establish an explicit structural formula for the lag relationship between water runoff and factors such as precipitation and irrigation.

[0131] S3.1. Based on the calculated lag time, establish an explicit relationship for the fusion lag time, taking the discharge volume as the output element and the precipitation and irrigation water volume as the input elements, as shown in formula (16):

[0132] (16)

[0133] In the formula, For the first Measured sequence of water receding volume over a given period; , The first Measured precipitation and irrigation water volumes for different time periods; , These represent the lag time of water discharge relative to precipitation and irrigation water, respectively.

[0134] S4. Construct the LSTM model.

[0135] S4.1 Normalize the measured sequences of water discharge, precipitation, and irrigation water volume that have lag characteristics.

[0136] S4.2, Building a LSTM-based model for predicting water discharge.

[0137] S4.2.1 Select the Tensorflow deep learning framework.

[0138] S4.2.2 Constructing an LSTM model: Using monthly precipitation and irrigation water volume as input features, and monthly drainage volume as the output variable, a 3-year monthly time series dataset (36 samples) is used to construct a drainage volume prediction model. The network structure consists of two input neurons. Each feature is extracted through three LSTM hidden layers (8 neurons). After feature fusion, it is fitted through one fully connected layer (4 neurons), and finally, one output neuron completes the drainage volume prediction. The model uses the Adam optimizer with a learning rate of 0.001, a batch size of 8 samples, 50 iterations, and a mean squared error loss function. Other parameters remain at the framework defaults.

[0139] S4.3 Divide the measured sequence into training and validation sequences in a 7:3 ratio, and find the optimal hyperparameter settings through cross-validation during the training phase.

[0140] S4.4. Input the precipitation verification sequence and irrigation water verification sequence with lag characteristics into the optimized drainage volume prediction model to obtain the drainage volume prediction sequence for the verification stage. Compare the predicted sequence with the drainage volume verification sequence, and calculate the MRE, RMSE, and NSE indices to evaluate the prediction accuracy of the drainage volume prediction model. The drainage volume prediction sequences obtained in the training and verification stages of this embodiment of the invention are as follows: Figure 3 As shown, according to Figure 3 It can be seen that the predicted water discharge sequence obtained by the present invention, which takes into account the lag time, is closer to the measured value and has higher accuracy.

[0141] S5, Construction and Model Embedding of Digital Twin Irrigation District System.

[0142] In this embodiment, after completing steps S1 to S4 to establish the water discharge prediction model, a digital twin irrigation district system is constructed so that the established water discharge prediction model can run in the digital twin environment and be used for subsequent model iteration optimization, scenario-based prediction and simulation.

[0143] S5.1 First, acquire basic data of the irrigation area and establish a digital twin data base. The basic data includes canal system topology data, engineering facility parameters, control node information, monitoring station information and irrigation area zoning information.

[0144] S5.2 After obtaining the basic data, construct a spatial and object model of the digital twin irrigation district. By modeling the engineering objects, canal connectivity, and operational status of the irrigation district, a digital representation of the irrigation district's operation process can be achieved.

[0145] S5.3. Based on the digital twin data base and the spatial and object model of the irrigation district, a digital twin irrigation district system platform is constructed. The various functional modules are integrated, developed, and deployed to form a unified digital twin irrigation district operation platform. The system platform includes a data monitoring module, a supply and demand analysis module, a water resource allocation module, a water distribution scheduling module, a channel flow simulation module, and a drainage volume prediction module.

[0146] S5.4 After the system platform is built, establish system data access and management functions to uniformly access and manage monitoring data, business data, and model data during the irrigation district operation, and preprocess and organize the data for storage. The data includes precipitation data, irrigation data, drainage data, and engineering operation status data.

[0147] S5.5 Deploy the water discharge prediction model established in steps S1 to S4 into the digital twin irrigation district system, and encapsulate it as a callable model component, defining the model input data items, output data items and calling methods.

[0148] S6. Periodic iterative optimization process of the water discharge prediction model.

[0149] In this embodiment, the digital twin irrigation district system updates and optimizes the water discharge prediction model on a monthly basis during operation.

[0150] S6.1. In each cycle, the system continuously collects irrigation district operation monitoring data through the data monitoring module, including precipitation, irrigation water volume and drainage volume, and after processing the data, inputs it into the drainage volume prediction model to generate the drainage volume prediction result for that cycle.

[0151] S6.2 After obtaining the actual water discharge monitoring data, the system compares the predicted water discharge volume with the actual water discharge volume and calculates the prediction error.

[0152] S6.3 When the prediction error exceeds the preset threshold, the system adjusts the parameters or updates the model of the water discharge prediction model through the model optimization module to correct the model deviation.

[0153] S6.4 The updated drainage volume prediction model is used for subsequent drainage volume prediction calculations. The system repeats the above process in each cycle, thereby forming a periodic iterative optimization mechanism driven by monitoring data, so that the drainage volume prediction model can gradually adapt to changes in irrigation district operating conditions.

[0154] S7. Contextualized prediction and simulation pre-playing process based on digital twin system.

[0155] In this embodiment, after completing the iterative optimization of the water discharge prediction model, the system can conduct simulations of irrigation district operation scenarios for different irrigation scheduling needs.

[0156] S7.1. Set or select different irrigation scheduling scenarios (such as different water supply schemes or water distribution schemes) in the system. Based on the current monitoring data, the system generates the corresponding scheduling scheme through the supply and demand analysis module, water resource allocation module and water distribution scheduling module.

[0157] S7.2 The system calls the channel water flow simulation module to simulate the irrigation area water flow process under the scheduling scheme and obtains simulation results such as flow rate, water level change and water conveyance process of each channel section.

[0158] S7.3 The system inputs the water flow simulation results into the water discharge prediction model and calculates the water discharge prediction results under the corresponding scheduling scenario.

[0159] S7.4 The system compares and analyzes the water discharge and water conveyance processes under different scheduling schemes through the scheduling scheme evaluation module, and displays the simulation results in a graphical way to assist irrigation district scheduling decisions.

[0160] The irrigation district drainage prediction method and digital twin irrigation district system considering the lag effect of the present invention have the following beneficial effects:

[0161] On the one hand, this invention addresses the phenomenon that drainage in irrigation areas lags significantly behind influencing factors such as precipitation and irrigation. First, it calculates the lag time of drainage relative to each influencing factor and constructs an explicit relationship that integrates the lag time, using drainage volume as the output factor and precipitation and irrigation volume as input factors. This intuitively reflects the dynamic relationship between drainage and factors such as precipitation and irrigation, while also improving the computational efficiency of the next step in model prediction. Secondly, this invention employs LSTM, combining the explicit structural formulas of the input drainage and factors such as precipitation and irrigation, to effectively capture the temporal dependence between drainage and influencing factors. This more accurately handles the characteristics of continuous and periodic changes and the nonlinear relationships between factors, improving the accuracy of model prediction.

[0162] On the other hand, this invention embeds a drainage volume prediction model into a digital twin irrigation district system. By periodically iteratively optimizing the prediction model using real-time monitoring data, and combining it with irrigation district scheduling plans for scenario-based prediction and simulation of the drainage process, the reliability of drainage volume prediction is improved. This provides effective technical support for irrigation district water resource scheduling and refined management. Ultimately, this invention can improve the prediction accuracy of irrigation district drainage, enhance the analysis and prediction capabilities of the drainage process, improve the efficiency of irrigation district water resource management and scheduling, and achieve refined management and optimized control of the irrigation district operation process.

[0163] The above descriptions are merely a few embodiments of the present invention and are not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any modifications or alterations made by those skilled in the art without departing from the scope of the technical solution of the present invention using the disclosed technical content are equivalent to equivalent implementation cases and fall within the scope of the technical solution.

Claims

1. A method for predicting irrigation runoff volume considering lag, characterized in that, include: Step 1: Obtain the measured sequence of the irrigation area, which includes the measured sequence of precipitation, the measured sequence of irrigation water, and the measured sequence of water discharge. Step 2: Determine the first lag time of water discharge relative to precipitation based on the measured water discharge sequence and the measured precipitation sequence; determine the second lag time of water discharge relative to irrigation based on the measured water discharge sequence and the measured irrigation water sequence. Step 3: Divide the measured sequence into training sequence and validation sequence according to the ratio. Take the water receding volume training sequence as the output and take the precipitation training sequence with the first lag time corresponding to the water receding volume training sequence and the irrigation water training sequence with the second lag time corresponding to the water receding volume training sequence as the input to train the water receding volume prediction model. Step 4: Determine the time period for predicting the amount of water receding, and denot it as the prediction period. Input the precipitation verification sequence with the first lag time corresponding to the prediction period and the irrigation water verification sequence with the second lag time corresponding to the prediction period into the trained water receding volume prediction model, output the water receding volume prediction sequence for the prediction period, and correct the water receding volume prediction model by combining the water receding volume prediction sequence and the water receding volume verification sequence for the prediction period.

2. The irrigation district drainage prediction method considering lag as described in claim 1, characterized in that, The first lag time between the receding water volume and the precipitation volume is determined based on the measured receding water volume sequence and the measured precipitation volume sequence, specifically as follows: Determine the first continuous wavelet transform coefficients of the measured water receding volume sequence, and determine the second continuous wavelet transform coefficients of the measured precipitation volume sequence; The cross-wavelet spectrum between the measured sequence of receding water volume and the measured sequence of precipitation is determined based on the first continuous wavelet transform coefficient and the second continuous wavelet transform coefficient. The first lag time between the receding water and the precipitation is determined based on the cross-wavelet spectrum.

3. The irrigation district drainage prediction method considering lag as described in claim 2, characterized in that, The first lag time between receding water and precipitation is determined based on the cross-wavelet spectrum, specifically as follows: Determine the cross wavelet power corresponding to each time-frequency grid point of the cross wavelet spectrum; Determine the phase of the time-frequency grid point corresponding to the maximum power of the cross wavelet; The first lag time between the receding water and the precipitation is determined based on the phase.

4. The irrigation district drainage prediction method considering lag as described in claim 1, characterized in that, The second lag time between drainage and irrigation is determined based on the measured sequence of drainage volume and the measured sequence of irrigation volume, specifically as follows: Determine the first continuous wavelet transform coefficients of the measured water discharge sequence and the third continuous wavelet transform coefficients of the measured irrigation water sequence; The cross-wavelet spectrum between the measured sequence of water discharge and the measured sequence of irrigation water is determined based on the first continuous wavelet transform coefficient and the third continuous wavelet transform coefficient. The second lag time of the drainage relative to the irrigation is determined by the cross-wavelet spectrum between the measured drainage volume sequence and the measured irrigation volume sequence.

5. The irrigation district drainage prediction method considering lag as described in claim 1, characterized in that, The water receding volume prediction model is corrected by combining the predicted water receding volume sequence and the water receding volume verification sequence for the predicted period, specifically as follows: The absolute value of the error between the predicted water volume sequence and the verified water volume sequence for the predicted period is determined. When the absolute value of the error is greater than a preset threshold, the parameters of the water volume prediction model are adjusted.

6. The irrigation district runoff prediction method considering lag as described in claim 5, characterized in that, The parameters of the aforementioned water discharge prediction model are adjusted as follows: Based on the water receding volume verification sequence for the predicted period, the parameters in the water receding volume prediction model are iteratively corrected using an optimization algorithm, so that the error between the water receding volume prediction sequence output by the water receding volume prediction model and the corresponding water receding volume verification sequence is less than or equal to the preset threshold.

7. The irrigation district drainage prediction method considering lag as described in claim 1, characterized in that, Following step 4, the following is also included: Step 5: Obtain the measured sequence for the next time period in the irrigation area, and add the measured sequence for the next time period to the original measured sequence. Repeat steps 1 to 5.

8. The irrigation district drainage prediction method considering lag as described in claim 1, characterized in that, The water discharge prediction model is a recurrent neural network model.

9. The irrigation district drainage prediction method considering lag as described in claim 8, characterized in that, The recurrent neural network model is specifically a long short-term memory neural network model.

10. A digital twin irrigation district system, characterized in that, It includes a data monitoring module and a water discharge prediction module connected to the data monitoring module; The data monitoring module is used to acquire measured precipitation sequences, measured irrigation water sequences, and measured drainage water sequences. The drainage volume prediction module uses the irrigation area drainage volume prediction method considering lag as described in any one of claims 1-9, to predict the drainage volume based on the measured precipitation sequence and measured irrigation water sequence obtained by the data monitoring module.

Images