Well-canal joint scheduling method coupled with long short memory deep learning

Abstract

The application provides a well-canal combined scheduling method coupled with long short memory deep learning, first, a real-time hydrological forecast model is constructed, historical N-day meteorological, hydrological and different water source irrigation amounts are taken as inputs through a long short memory deep learning method to forecast runoff in a short future period, the different water source irrigation amounts include well water and canal water irrigation amounts; then, a real-time water amount scheduling model is constructed, a minimum runoff is taken as a constraint to maximize irrigation area water supply benefit to solve optimal irrigation amounts of different water sources in a short period; finally, the real-time hydrological forecast model and the real-time water amount scheduling model are iterated and simulated until balance is reached, so that optimal well water and canal water irrigation amounts are obtained. The application can realize precise irrigation of an irrigation area, fully utilize water supply capacity of a canal, exchange water supply time for water supply space, fully play water storage engineering capacity of the irrigation area and maximize reduction of drought.

Description

Technical Field

[0001] This invention belongs to the field of hydrology and water resources engineering technology, and relates to a method for optimizing water resource scheduling in irrigation districts, and particularly to a well-canal joint scheduling method coupled with long short memory deep learning. Background Technology

[0002] The Ningxia Yellow River irrigation area has an irrigation history of over two thousand years. The current canal system layout is largely a continuation and continuous modification of historical systems. Combined with the area's unique geographical features, this has resulted in a distinctive irrigation and drainage system. The vastness of the irrigation basin, the complex and intersecting canal system, and the complexity of influencing factors make it extremely difficult to establish an accurate mathematical model for water allocation. Currently, the irrigation area primarily uses Case-Based Reasoning (CBR) to address water allocation issues. However, this method can only provide a basis for future water resource allocation using limited historical data. When future water inflow scenarios exceed the scope of historical data, this method becomes ineffective for reasonable allocation. Furthermore, the allocation schemes included in historical data may not be optimal, leaving room for optimization.

[0003] Purpose of the invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies, optimize current irrigation district water diversion and scheduling schemes, and achieve real-time optimization of irrigation district water allocation, including maximizing irrigation benefits and reducing water supply conflicts within irrigation districts. By fully utilizing the water supply capacity of canals, the invention achieves "storage during wet seasons and regulation during dry seasons" within irrigation districts, providing a guarantee for long-term stable water resource allocation in irrigation districts. Summary of the Invention

[0005] This invention provides a well-ditch joint scheduling method coupled with long short-term memory deep learning, comprising the following steps:

[0006] Step 1: Construct a real-time hydrological forecasting module model; The real-time hydrological forecasting module uses the long short-term memory deep learning method to forecast the runoff in the near future by taking the meteorological and hydrological data of N historical days and the irrigation volume of different water sources as input. The irrigation volume of different water sources includes well water irrigation volume and canal water irrigation volume.

[0007] Step 2: Construct a real-time water allocation module model. The real-time water allocation module aims to maximize the water supply benefits of the irrigation area and uses the minimum runoff as a constraint to solve for the optimal irrigation volume for different water sources in the short term.

[0008] Step 3: Iterate the simulation between the real-time hydrological forecasting module and the real-time water volume scheduling module until they reach equilibrium, thereby obtaining the optimal well water irrigation volume and canal water irrigation volume.

[0009] Preferably, the scheduling objective and constraints of the joint scheduling method are represented as shown in equations (1)-(4):

[0010]

[0011]

[0012]

[0013]

[0014] Where P represents the maximum benefit during the irrigation period. Let t be the amount of water injected into the well. Let t be the benefit of well water irrigation. Let t be the irrigation water volume in the canal. Let t be the irrigation efficiency of the canal, and n be the total duration. Let be the water inflow at time t. Let t be the amount of water receding from the farmland. Let t be the water volume at the irrigation district outlet. P represents the constrained water volume at the irrigation district outlet at time t. t T t G t t represents the rainfall, temperature, and groundwater level at time t, respectively; f(·) is the LSTM function of the Long Short-Term Memory deep learning model.

[0015] Preferably, step 3 is as follows: when the weather factors for the next n days are known, the simulated runoff for the next n days is obtained by iterating through formula (4), and then the simulated runoff is input into formulas (1)-(3) to solve for the corresponding well water irrigation volume and canal water irrigation volume; then formulas (1) and (2) are iterated continuously until the well water irrigation volume and canal water irrigation volume obtained by the solution no longer change, then the optimal well water irrigation volume and canal water irrigation volume are obtained. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the well-channel joint scheduling method coupled with long short memory deep learning as described in this invention. Detailed Implementation

[0017] The present invention will be described in detail below with reference to the accompanying drawings. Those skilled in the art should understand that the specific embodiments are only used to illustrate the present invention and should not be regarded as limiting the present invention. Any variation or element substitution technical solution that does not depart from the spirit of the present invention falls within the scope of the present invention.

[0018] A schematic diagram of the well-channel joint scheduling method coupled with long short-term memory deep learning described in this invention is shown below. Figure 1As shown. This invention provides a well-ditch joint scheduling method coupled with long short-term memory deep learning, comprising the following steps:

[0019] Step 1: Construct a real-time hydrological forecasting module model; The real-time hydrological forecasting module uses the long short-term memory deep learning method to forecast the runoff in the near future by taking the meteorological and hydrological data of N historical days and the irrigation volume of different water sources as input. The irrigation volume of different water sources includes well water irrigation volume and canal water irrigation volume.

[0020] Step 2: Construct a real-time water allocation module model. The real-time water allocation module aims to maximize the water supply benefits of the irrigation area and uses the minimum runoff as a constraint to solve for the optimal irrigation volume for different water sources in the short term.

[0021] Step 3: Iterate the simulation between the real-time hydrological forecasting module and the real-time water volume scheduling module until they reach equilibrium, thereby obtaining the optimal well water irrigation volume and canal water irrigation volume.

[0022] The scheduling objective and constraints of the joint scheduling method are represented as shown in equations (1)-(4):

[0023]

[0024]

[0025]

[0026]

[0027] Where P represents the maximum benefit during the irrigation period. Let t be the amount of water injected into the well. Let t be the benefit of well water irrigation. Let t be the irrigation water volume in the canal. Let t be the irrigation efficiency of the canal, and n be the total duration. Let be the water inflow at time t. Let t be the amount of water receding from the farmland. Let t be the water volume at the irrigation district outlet. P represents the constrained water volume at the irrigation district outlet at time t. t T t G t t represents the rainfall, temperature, and groundwater level at time t, respectively; f(·) is the LSTM function of the Long Short-Term Memory deep learning model.

[0028] When the weather factors for the next n days are known, the simulated runoff for the next n days is obtained by iterating through equation (4). Then, the simulated runoff is input into equations (1)-(3) to solve for the corresponding well water irrigation volume and canal water irrigation volume. Then, equations (1) and (2) are iterated continuously until the well water irrigation volume and canal water irrigation volume obtained by the solution no longer change. Then, the optimal well water irrigation volume and canal water irrigation volume are obtained.

[0029] Compared with existing technologies, this invention combines cutting-edge deep learning models with stochastic optimization models. Through dynamic solution, the water supply plan can be continuously adjusted according to weather conditions, so as to maintain the optimal dynamic balance between water supply and demand.

[0030] This invention can help achieve precise irrigation in irrigation districts, fully utilize the water supply capacity of canals, exchange water supply time for water supply space, give full play to the water storage capacity of irrigation districts, minimize drought, and provide a guarantee for ecological construction. The combination of deep learning-based hydrological forecasting and a stochastic uncertain optimization model achieves the dynamic optimal irrigation volume for different water sources.

Claims

1. A well-ditch joint scheduling method coupled with long short-term memory deep learning, characterized in that, Includes the following steps: Step 1: Construct a real-time hydrological forecasting module model; The real-time hydrological forecasting module uses the long short-term memory deep learning method to forecast the runoff in the near future by taking the meteorological and hydrological data of N historical days and the irrigation volume of different water sources as input. The irrigation volume of different water sources includes well water irrigation volume and canal water irrigation volume. Step 2: Construct a real-time water allocation module model. The real-time water allocation module aims to maximize the water supply benefits of the irrigation area and uses the minimum runoff as a constraint to solve for the optimal irrigation volume for different water sources in the short term. Step 3: Iterate the simulation between the real-time hydrological forecasting module and the real-time water volume scheduling module until they reach equilibrium, thereby obtaining the optimal well water irrigation volume and canal water irrigation volume. The scheduling objectives and constraints are represented as shown in equations (1)-(4): Max: (1), （2）， （3）， （4）， Where P represents the maximum benefit during the irrigation period. Let t be the amount of water injected into the well. Let t be the benefit of well water irrigation. Let t be the irrigation water volume in the canal. Let t be the irrigation efficiency of the canal, and n be the total duration. Let be the water inflow at time t. Let t be the amount of water receding from the farmland. Let t be the water volume at the irrigation district outlet. Let t be the constrained water volume at the irrigation district outlet. , , The values ​​represent rainfall, temperature, and groundwater level at time t, respectively. For the Long Short-Term Memory (LSTM) deep learning model; The specific process of step 3 is as follows: when the weather factors for the next n days are known, the simulated runoff for the next n days is obtained by iterating through formula (4), and then the simulated runoff is input into formula (1)-(3) to solve for the corresponding well water irrigation volume and canal water irrigation volume; then formula (1) and (2) are iterated continuously until the well water irrigation volume and canal water irrigation volume obtained by the solution no longer change, then the optimal well water irrigation volume and canal water irrigation volume are obtained.

Images