Canal system dynamic water distribution scheduling method considering short-term rainfall forecast

Abstract

The invention discloses a canal system dynamic water distribution scheduling method considering short-term rainfall forecast, and the method comprises the steps: employing a grey wolf optimization GWO algorithm to optimize an SVR model according to meteorological and hydrological data; performing short-term forecasting on rainfall through a GWO optimized SVR model and a long short-term memory LSTM prediction model; the predicted short-term rainfall is used for constructing a canal system water distribution multi-target dynamic optimization model comprehensively considering irrigation time and irrigation water efficiency; a dynamic multi-target non-dominated sorting genetic algorithm is used for solving a multi-target canal system water distribution dynamic optimization model, and meanwhile, maximum utilization of rainfall resources and an optimal control strategy of canal system water distribution are realized through a prediction-decision-prediction rolling optimization mechanism. On the premise of ensuring agricultural safe and stable production, the potential of saving water for agricultural irrigation is released as much as possible.

Description

Technical Field

[0001] The invention belongs to the technical field of agricultural water management, and in particular relates to a canal system dynamic water distribution scheduling method considering short-term rainfall forecast. Background Art

[0002] With the acceleration of the modernization process of irrigation areas, various computer hardware and software infrastructures are constantly upgraded and improved. Under the premise of ensuring agricultural grain output, the current stage of agricultural development has also put forward new requirements for the utilization efficiency of irrigation water. The canal optimization water distribution scheduling model is a very important part of the irrigation area water distribution work, and it is also an important construction content of the digital twin irrigation area model library. It is urgent to build an efficient, reasonable and economical model. The scientific scheduling of irrigation canals is of great significance to improving the efficiency of irrigation water use, saving water resources and ensuring agricultural production. However, the water distribution process of irrigation canals is often affected by natural rainfall. Especially in the case of short-term rainfall, the traditional fixed water distribution scheme is difficult to adjust in time, resulting in water waste or insufficient irrigation. At the same time, climate change has exacerbated the uncertainty of rainfall distribution, making the applicability of existing regulation methods face challenges. How to improve the adaptability of canal scheduling schemes to climate change on the basis of ensuring food production and social stability, maximize the use of natural rainfall resources and improve the utilization efficiency of surface water resources, is crucial to achieving the economic benefits of irrigation water use.

[0003] Traditional canal water distribution scheduling schemes mainly rely on surface water and groundwater, and the development and utilization model of natural rainfall resources is also relatively rough, which is prone to farmland waterlogging caused by the superposition of rainfall and irrigation. At the same time, the irrigation canal system may be subjected to excessive water delivery pressure, increasing the risk of channel overflow and dam breach. In addition, some irrigation areas will rely too much on groundwater due to imperfect canal structure and insufficient water supply capacity, which is easy to cause groundwater overexploitation. Timely response to rainfall forecast information and dynamic optimization of canal water supply scheduling are effective solutions to give full play to the effectiveness of irrigation scheduling system, reduce the risk of farmland waterlogging, save irrigation water resources and protect groundwater resources.

[0004] In traditional open channel irrigation systems, irrigation district managers usually only pay attention to the water level changes in the main canal to ensure that the operating water level does not exceed the design water level of the main canal, so as to ensure the safety and stability of the channel hardware facilities. In addition, the flow and water distribution time of the lower-level channels are not considered. When the short-term rainfall forecast shows that the rainfall reaches a certain standard and the rainfall probability is met, the irrigation district manager chooses to suspend the entire channel water supply system, and resumes water supply according to the previous irrigation plan after the rainfall ends. In summary, the traditional scheduling plan cannot be dynamically adjusted according to historical water supply data and future rainfall levels. Although it ensures the safe operation of the project, it leads to a waste of water resources and an extension of the irrigation cycle to a certain extent. In the past, most irrigation canal water distribution scheduling only considered the optimal allocation of surface and groundwater resources, and rarely explored the dynamic response of the entire irrigation cycle to short-term forecasts and the dynamic correction of scheduling plans. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a dynamic water distribution scheduling method for a canal system taking into account short-term rainfall forecasts, by maximizing the use of rainwater resources, realizing the joint scheduling of rainwater flood water, surface water and groundwater, and dynamically correcting the canal system water distribution scheduling plan, thereby alleviating the contradiction between supply and demand of agricultural water resources, improving irrigation water efficiency and enhancing agricultural economic benefits, which has become an important solution to promote the sustainable development of agriculture.

[0006] To achieve the above object, the present invention adopts the following technical solution:

[0007] A method for dynamic water distribution scheduling of canal system considering short-term rainfall forecast, comprising:

[0008] Obtain meteorological and hydrological data;

[0009] According to meteorological and hydrological data, the SVR model is optimized using the Grey Wolf Optimization GWO algorithm;

[0010] According to meteorological and hydrological data, the short-term rainfall forecast is carried out through the SVR model optimized by GWO and the long short-term memory LSTM prediction model;

[0011] The predicted short-term rainfall is used to construct a multi-objective dynamic optimization model for canal water distribution that comprehensively considers irrigation time and irrigation water efficiency.

[0012] A dynamic multi-objective non-dominated sorting genetic algorithm is used to solve the multi-objective canal water distribution dynamic optimization model. At the same time, the maximum utilization of rainfall resources and the optimal control strategy of canal water distribution are achieved through the prediction-decision-prediction rolling optimization mechanism.

[0013] Preferably, the meteorological and hydrological data include: long-term series of rainfall, temperature, wind speed, humidity, air pressure, solar radiation, runoff, evaporation, and water demand of each branch canal.

[0014] As a preferred option, the flow rate of each branch canal, the irrigation start time, and the irrigation end time are used as decision variables, the design flow rate and design water level of the canal system are used as constraints, and multi-objective modeling is performed with maximizing irrigation water utilization efficiency, the shortest irrigation cycle, and minimizing flood risk as optimization directions. The irrigation strategy is dynamically corrected in combination with the predicted future rainfall to obtain a multi-objective dynamic optimization model for canal water distribution.

[0015] The present invention uses the Gray Wolf Optimization GWO algorithm to optimize the SVR model based on meteorological and hydrological data; according to meteorological and hydrological data, the SVR model optimized by GWO and the long short-term memory LSTM prediction model are used to make short-term rainfall forecasts; the predicted short-term rainfall is used to construct a multi-objective dynamic optimization model for canal water distribution that comprehensively considers irrigation time and irrigation water efficiency; a dynamic multi-objective non-dominated sorting genetic algorithm is used to solve the multi-objective dynamic optimization model for canal water distribution, and at the same time, the rolling optimization mechanism of prediction-decision-prediction is used to realize the maximum utilization of rainfall resources and the optimal control strategy for canal water distribution. The potential for agricultural irrigation water conservation is released as much as possible under the premise of ensuring safe and stable agricultural production. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0017] Figure 1 This is a flow chart of a method for dynamic water distribution scheduling of a canal system taking into account short-term rainfall forecast according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] Embodiment 1:

[0021] like Figure 1 As shown, an embodiment of the present invention provides a canal system dynamic water distribution scheduling method considering short-term rainfall forecast, comprising the following steps:

[0022] S1: Basic data collection: The meteorological and hydrological data required by the model include long-term series of rainfall, temperature (maximum temperature, minimum temperature and average temperature), wind speed, humidity (relative humidity and absolute humidity), air pressure, solar radiation (total radiation, direct radiation and diffuse radiation), runoff, calculated evaporation, water demand of each branch canal, etc.; channel characteristic data include channel length, channel top elevation, channel bottom elevation, slope coefficient, lining type, leakage coefficient, design flow, design water level, channel system topology, controlled irrigation area, etc.

[0023] S2: Collect historical meteorological data required for rainfall forecast and perform data preprocessing, using interpolation or filling methods (such as mean filling or KNN interpolation). Remove abnormal data points through statistical analysis (such as box plots). Align the data to a uniform time resolution (daily). Standardize the data to the range of [0,1] according to the following formula, which is suitable for the input of LSTM and SVR.

[0024]

[0025] Next, we construct input and output samples, taking historical meteorological data as input and future rainfall as output; finally, we split the dataset into training set, validation set, and test set, and divide the dataset in chronological order, with 80% used for training, 10% for validation, and 10% for testing.

[0026] S3: Build and train the LSTM model. First, design the LSTM model structure, use the time series data (shape is (number of time steps, number of features)) as the input layer, define the number of hidden units (such as 64 or 128) and the fully connected layer (predict the rainfall at the next moment). Secondly, set the hyperparameters of the model, including the learning rate, optimizer (such as Adam), training batch size (batchsize), and number of iterations (epochs). Set the mean square error (MSE) as the loss function. Finally, save the best model for testing and deployment. The core calculation formula of LSTM is as follows:

[0027] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0028] In the formula, h t-1 is the hidden state at the previous moment; x t is the input at the current moment; W f and b f are the weight and bias of the forget gate; σ is the Sigmoid activation function.

[0029] it =σ(W i ·[h t-1 ,x t ]+b i )

[0030]

[0031] In the formula, i t is the output of the input gate, which determines what information will be written into the memory cell; is the new candidate memory value, used to update the memory unit; W i , W C and b i , b C are the weight and bias of the input gate; tanh is the tanh activation function.

[0032]

[0033] In the formula, C t-1 is the state of the memory unit at the previous moment; C t is the state of the memory unit at the current moment.

[0034] o t =σ(W o [h t-1 ,x t ]+b o )

[0035] h t =o t tanh(C t )

[0036] In the formula, o t is the output of the output gate; h t is the hidden state at the current moment.

[0037] S4: Initialize the hyperparameters of the SVR model, such as kernel function (such as RBF), C (penalty parameter) and γ (kernel width). Then input the training set to fit the SVR model and output the predicted value of future rainfall. The objective function of SVR is as follows:

[0038] f(x)=w T φ(x)+b

[0039] Where w is the weight vector, φ(x) is the nonlinear mapping function of the input sample x, which maps the data from the low-dimensional input space to the high-dimensional feature space. b is the bias term.

[0040] The loss function of SVR is as follows:

[0041]

[0042] In the formula, y i is the actual observed value; f(x i ) is the predicted value; |y i -f(x i )| is the absolute error; ε is the insensitive interval threshold, which is used to control the tolerance range of the error. Losses will only occur when the error exceeds ε; L ε (y i ,f(x i )) is the loss function value.

[0043] The optimization objectives and constraints of SVR are as follows:

[0044]

[0045] y i -f(x i )≤ε+ξ i ,i=1,2,...,n

[0046]

[0047]

[0048] In the formula, ||w|| 2 is the complexity of the model (regularization term); C is the penalty coefficient, which controls the trade-off between loss and model complexity; ξ i , is a slack variable, representing positive and negative errors beyond the range of ε.

[0049] S5: GWO is used to optimize the hyperparameters of the SVR model and retrain the data set. The GWO algorithm updates the position of individuals by simulating the process of gray wolves besieging prey when hunting. The position update formula of each wolf is as follows:

[0050] X i (t+1)=X i (t)+A·D

[0051] In the formula, X i (t) is the position of the ith wolf at the tth iteration; A is the coefficient vector that affects the direction of the wolf, and the calculation formula is:

[0052] A=2·a·r1-a

[0053] In the formula, a is a linearly decreasing vector (decreasing from 2 to 0); r1 is a random vector uniformly distributed between [0,1]. D is the position vector of the prey (i.e. the optimal solution), and the calculation formula is:

[0054] D=|C·X p -Xi |

[0055] In the formula, X p is the optimal position of the current individual; C is the coefficient vector, and the calculation formula is:

[0056] C=2·r2

[0057] Where r2 is a random vector.

[0058] The process of combining the hyperparameter optimization of SVR with GWO can be expressed as:

[0059]

[0060] Where C is the penalty coefficient, ε is the tolerance error of SVR, and γ is the parameter of the kernel function.

[0061] S6: Evaluate the performance of the SVR model optimized by LSTM and GWO to ensure that the accuracy of model prediction is sufficient to support the canal system optimization water distribution model and obtain a reasonable water distribution plan. The evaluation indicators of model error are as follows:

[0062] Mean Squared Error (MSE):

[0063]

[0064] Root Mean Square Error (RMSE):

[0065]

[0066] Mean Absolute Error (MAE):

[0067]

[0068] Coefficient of determination (R 2 ):

[0069]

[0070] Where: n is the total number of data points (sample number); y i is the true value of the i-th sample; is the predicted value of the i-th sample; is the mean of the true values.

[0071] S7: Construct a dynamic optimization model for canal water distribution, with the flow of each branch canal, the start time of irrigation, and the end time of irrigation as decision variables, and the design flow and design water level of the canal system as constraints. Multi-objective modeling is performed with maximizing irrigation water utilization efficiency, the shortest irrigation cycle, and the minimum risk of flooding as optimization directions. The irrigation strategy is dynamically modified in combination with the predicted future rainfall. The constructed dynamic multi-objective planning model is as follows:

[0072] Objective function 1: Maximize irrigation water resource utilization efficiency

[0073]

[0074]

[0075]

[0076] Where W total is the total water supply, m 3 ; S is the leakage loss of the channel, m 3 ; is the irrigation start time of branch canal i, h; is the end time of irrigation of branch canal i, h; n is the total number of branch canals; q i is the water distribution flow of branch canal i, m 3 / s; β i A is the reduction coefficient of water leakage after anti-seepage measures are taken for branch canal i; i is the permeability coefficient of the channel bed; l i is the water conveyance length of branch canal i, m; m is the channel bed permeability index of branch canal i;

[0077] Objective function 2: Minimize irrigation cycle

[0078]

[0079] Objective function 3: Minimize flood risk

[0080]

[0081] In the formula, is the design flow rate of branch channel i, m 3 / s;

[0082] Constraints:

[0083] (1) Canal system design flow constraints:

[0084]

[0085] (2) Irrigation demand constraints:

[0086]

[0087]

[0088] In the formula, is the irrigation demand of branch canal i at the current moment, m 3 ER t is the predicted rainfall for the next time step after area conversion, m3 ; is the actual irrigation amount in the previous time step, m 3 .

[0089] (3) Irrigation start time constraints:

[0090]

[0091] Where, T is the irrigation period, d;

[0092] (4) Branch canal flow constraints:

[0093]

[0094] (5) Traffic balance constraints:

[0095]

[0096] In the formula, Q t is the water distribution flow of the main canal, m 3 / s.

[0097] S8: Use the dynamic multi-objective non-dominated sorting genetic algorithm (DNSGA-II) to solve the multi-objective canal water distribution dynamic optimization model, input the future rainfall predicted by LSTM and SVR into the model, make scheduling decisions based on the current state variables, and add the actual rainfall observation data to the prediction model data set for retraining every time the time step is updated, and predict the next long future rainfall. The new actual irrigation water demand is obtained through the next long rainfall and the decision is iterated again. When the irrigation cycle ends, the canal water distribution plan for the entire irrigation cycle can be obtained after rolling correction.

[0098] The present invention combines support vector regression (SVR) and long short-term memory (LSTM) models, and uses the gray wolf optimization (GWO) algorithm to optimize SVR parameters, thereby improving the accuracy of short-term rainfall prediction and providing reliable data support for irrigation scheduling. By constructing a multi-objective dynamic optimization model that comprehensively considers irrigation time and water use efficiency, combined with the dynamic non-dominated sorting genetic algorithm (DNSGA-II), intelligent optimization of canal water distribution is achieved, and the utilization efficiency of agricultural water resources is improved. The rolling optimization method of "prediction-decision-prediction" is adopted to dynamically respond to rainfall changes, flexibly adjust water distribution plans, maximize the use of rainfall resources, and reduce the risks of over-irrigation and waste of water resources.

[0099] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for dynamic water distribution scheduling of a canal system considering short-term rainfall forecast, characterized in that: include: Obtain meteorological and hydrological data; According to meteorological and hydrological data, the SVR model is optimized using the Grey Wolf Optimization GWO algorithm; According to meteorological and hydrological data, the short-term rainfall forecast is carried out through the SVR model optimized by GWO and the long short-term memory LSTM prediction model; The predicted short-term rainfall is used to construct a multi-objective dynamic optimization model for canal water distribution that comprehensively considers irrigation time and irrigation water efficiency. A dynamic multi-objective non-dominated sorting genetic algorithm is used to solve the multi-objective canal water distribution dynamic optimization model. At the same time, the maximum utilization of rainfall resources and the optimal control strategy of canal water distribution are achieved through the prediction-decision-prediction rolling optimization mechanism.

2. The canal system dynamic water distribution scheduling method considering short-term rainfall forecast as claimed in claim 1, characterized in that: Meteorological and hydrological data include: long-term series of rainfall, temperature, wind speed, humidity, air pressure, solar radiation, runoff, evaporation, and water demand of each branch canal.

3. The method for dynamic water distribution scheduling of a canal system considering short-term rainfall forecast as claimed in claim 2, characterized in that: The flow rate of each branch canal, the start time of irrigation, and the end time of irrigation are used as decision variables, the design flow rate and design water level of the canal system are used as constraints, and multi-objective modeling is carried out with maximizing irrigation water utilization efficiency, shortest irrigation cycle, and minimum flood risk as optimization directions. The irrigation strategy is dynamically corrected based on the predicted future rainfall, and a multi-objective dynamic optimization model for canal water distribution is obtained.

Images