Automatic water resource optimal scheduling management method based on irrigation demonstration area

Abstract

The application belongs to the technical field of irrigation scheduling management, and specifically discloses an automatic water resource optimal scheduling management method based on a demonstration area of an irrigation area, which comprises the following steps: acquiring N irrigation areas in a current concerned irrigation area, and setting up a scheduling set containing multiple water irrigation modes for each irrigation area; constructing a selection optimization algorithm for selecting water irrigation modes for all irrigation areas one by one at each time; acquiring a local state set of each irrigation area in real time, inputting all local state sets into the selection optimization algorithm to select water irrigation modes of all irrigation areas at the same time; and having the following advantages: through real-time acquisition of the local state set and application of the selection optimization algorithm, the adaptability of the irrigation system is enhanced, the water resource allocation is optimized, and the irrigation efficiency and crop growth are ensured. Through intelligent management, water resource waste is reduced, cost is reduced, and economic and environmental sustainability of agricultural production is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of irrigation scheduling management, in particular to an automatic water resource optimal scheduling management method based on a demonstration area of an irrigation area. BACKGROUND

[0002] In traditional agricultural irrigation systems, irrigation areas mainly rely on fixed channel water sources for water supply. These systems are designed considering regular climate conditions and can meet basic irrigation needs in years with sufficient water supply. However, these systems show obvious shortcomings in extreme drought conditions. Especially in cases where drought leads to water shortages, scheduling water sources relying on fixed channels often cannot meet the needs of all irrigation areas, because these systems lack the ability to flexibly adjust water sources and optimize water flow paths. In addition, due to the lack of efficient resource allocation strategies and rapid response mechanisms to drought, traditional irrigation systems often cannot reasonably allocate limited water resources, resulting in some areas possibly having excess water while other areas are severely water deficient.

[0003] To this end, an automatic water resource optimal scheduling management method based on a demonstration area of an irrigation area is proposed to solve the above-mentioned problems. SUMMARY

[0004] The present application aims to provide an automatic water resource optimal scheduling management method based on a demonstration area of an irrigation area to solve or improve the above technical problems that traditional agricultural irrigation systems often cannot effectively deal with water shortages due to the lack of flexible water source scheduling and efficient resource allocation capabilities in extreme drought conditions.

[0005] Therefore, the first aspect of the present application is to provide an automatic water resource optimal scheduling management method based on a demonstration area of an irrigation area.

[0006] The first aspect of the present application provides an automatic water resource optimization scheduling management method based on demonstration area of irrigation district, comprising the following steps: obtaining N irrigation areas in the current concerned irrigation district, and setting up a scheduling set containing multiple water irrigation modes for each irrigation area; constructing a selection optimization algorithm for selecting the water irrigation mode of each irrigation area one by one at each time; obtaining the local state set of each irrigation area in real time, inputting all the local state sets into the selection optimization algorithm to select the water irrigation mode of all the irrigation areas at the same time; the step of selecting the water irrigation mode of the irrigation area by the selection optimization algorithm comprises: constructing the action space set of each irrigation area by the irrigation distribution water amount and the type of water irrigation mode, and constructing an independent local network for each irrigation area; obtaining the state set of the irrigation area according to the local state set, calculating the action value of all actions of the irrigation area in the action space set through the local network; retaining the action with the highest action value, and adjusting the water of the current irrigation area according to the irrigation distribution water amount and the water irrigation mode contained in the retained action.

[0007] In any of the above technical solutions, the water irrigation mode of the scheduling set comprises channel water source calling, path optimal calling, water source optimal calling and water amount optimal calling; the step of channel water source calling comprises: obtaining all channel paths in the current irrigation area, opening the currently obtained channel path, and keeping the channel water level of all channel paths consistent; the step of path optimal calling comprises: obtaining the optimal path between the water pump and all the plots in the current irrigation area, and irrigating the plots according to the optimal path; the step of water source optimal calling comprises: obtaining the callable irrigation area adjacent to the current irrigation area, and irrigating the current irrigation area through the callable irrigation area; the step of water amount optimal calling comprises: obtaining the required water amount and the callable water amount of all the irrigation areas, and irrigating all the irrigation areas through external water source according to the difference between the required water amount and the callable water amount.

[0008] In any of the above technical solutions, the step of retaining the action with the highest action value comprises: Wherein, the is the action with the highest action value; the is the action space set, and the is the irrigation distribution water amount, the is the water irrigation mode; the is the number of the irrigation area, and is a positive integer from 1 to N; the is the state set; the parameters for estimating an expected return of each action; the parameters are the action values.

[0009] In any of the above technical solutions, the selection optimization algorithm comprises a local network of all the irrigation areas and a global network constructed by mixing all the local networks, and has the following cases:

[0010] Case one: an integrated action value is obtained according to the action values of all the local networks at each unit time, and the global network is updated through the integrated action value;

[0011] Case two: the update of the local network at the next unit time considers the influence degree of the global network on the current local network.

[0012] In any of the above technical solutions, at each unit time, the local network of each irrigation area is updated by an action value, specifically comprising: wherein the is a learning rate; the is a discount factor; the is a benefit value of the irrigation area adopting a current action; the is a state set at the next unit time; the is an action at the next unit time; the is the maximum action value in all actions at the next state set; the is an adjustment coefficient for controlling the influence degree of the global network on the update of the local network; the is a gradient of the global network relative to the local network, for representing the influence degree of the global network on the local network; the is an integrated action value of the global network; the is a current action value.

[0013] In any of the above technical solutions, the integrated action value is obtained as follows: wherein the is a weight matrix for linearly transforming the action value of the local network; the is a physical state measurement value vector, and comprises at least one item in the local state set; the is a weight matrix for adjusting the influence of the integrated action value in the calculation; the represents a one-hot encoding vector of the water irrigation mode of the local network; the represents a one-hot encoding vector of the water irrigation mode of the local network; the is a weight matrix for adjusting the influence of the water irrigation mode in the comprehensive action value calculation; the is a bias term for the first layer of the global network; the is an activation function for introducing nonlinearity to help the global network capture complex relationships; the is a weight matrix of the second layer, and the is a bias term commonly used to convert the output of the hidden layer h into a comprehensive action value.

[0014] In any of the above technical solutions, the global network is updated by the comprehensive action value at each unit of time, specifically including: calculating the difference value between the comprehensive action value of the current time unit and the predicted comprehensive action value of the next time unit; the modified difference value is updated using a loss function through a gradient descent method to update the network parameters of the global network; the gradient of the loss function with respect to each parameter is calculated by a backpropagation algorithm to reduce the loss.

[0015] In any of the above technical solutions, the loss function includes the following formula: wherein the is a loss function for representing the deviation between the predicted comprehensive action value and the actual obtained value; the is a global reward; the is a discount factor; the is the expected comprehensive action value under all possible actions in the next state set; the is the comprehensive action value estimate of the current state set; the is the network parameter of the current global network; the is the parameter of the target network.

[0016] In any of the above technical solutions, the comprehensive action value is obtained by weighting the action values of all local networks, and the set weight of each action value in the next time unit is determined by the corresponding local state set in the current time unit.

[0017] The present application has the following beneficial effects compared with the prior art:

[0018] By obtaining the local state set of each irrigation area in real time, and dynamically selecting the most suitable irrigation mode using a selection optimization algorithm based on these data, the adaptability and responsiveness of the irrigation system are greatly enhanced. This enables the system to adjust the irrigation strategy in real time, better cope with different environmental conditions, and optimize the use of water resources.

[0019] By constructing a scheduling set, multiple irrigation mode options are provided for each irrigation area, and the current local state set is combined to determine the optimal water resource allocation. This method can ensure the optimal use of water resources in the entire region, reduce waste, and improve irrigation efficiency.

[0020] By precisely controlling the water quantity and irrigation mode of each irrigation area, the present application helps ensure that crops receive adequate water supply, promoting healthy growth and thus improving yield and quality. In addition, by optimizing the use of water resources, it also helps protect the ecological environment and support sustainable agricultural practices.

[0021] By intelligent irrigation management, the possibility of excessive irrigation and water resource waste is reduced, thereby reducing water resource costs and maintenance costs. At the same time, more precise water resource management also helps farmers increase crop yield and increase economic benefits.

[0022] Additional aspects and advantages of embodiments according to the present application will become apparent from the following description with reference to the drawings. BRIEF DESCRIPTION OF DRAWINGS

[0023] The above and / or additional aspects and advantages of the present application will become apparent and more readily appreciated from the following description, taken in conjunction with the accompanying drawings, in which:

[0024] Figure 1 A flowchart of the method steps of the present application. DETAILED DESCRIPTION

[0025] In order to more clearly understand the above-mentioned purposes, features and advantages of the present application, the present application will be further described in detail below in conjunction with the drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0026] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, however, the present application can also be implemented in other ways different from those described herein, therefore, the scope of protection of the present application is not limited by the specific embodiments disclosed below.

[0027] Please refer to Figure 1 , the following describes some embodiments of the present application based on the automatic water resource optimization scheduling management method of the irrigation area demonstration area.

[0028] Embodiments of the first aspect of the present application propose an automatic water resource optimization scheduling management method based on an irrigation area demonstration area. In some embodiments of the present application, as shown in Figure 1 , the automatic water resource optimization scheduling management method comprises:

[0029] S101, obtain N irrigation areas in the current concerned irrigation area, and set up a scheduling set containing multiple water use irrigation methods for each irrigation area.

[0030] As can be seen, all irrigation areas in the current concerned irrigation area are identified and obtained. Specifically, this involves a detailed geographical and resource assessment of the irrigation area, including analysis of soil type, crop demand, and historical irrigation data for each irrigation area. These data provide the necessary basic information for subsequent water resource scheduling, ensuring that all water demand areas are included in the management scope. Subsequently, a scheduling set containing multiple water use irrigation methods is set up for each irrigation area, which is another core component of the invention. The irrigation methods included in the scheduling set may include drip irrigation, sprinkler irrigation, underground seepage irrigation, etc., each of which has its own specific application conditions and efficiency. Through such design, the management system can dynamically select the most suitable irrigation method for the current environment and demand according to real-time monitoring of soil moisture, weather forecast, water resource availability, etc. For example, in times of water shortage, the system may prefer to use drip irrigation technology with high water use efficiency; while in times of abundant water, it may use sprinkler irrigation technology with wider coverage.

[0031] Optimize water source scheduling and irrigation path using information technology and artificial intelligence algorithms. The system integrates sensor networks to monitor water and soil conditions in each irrigation area in real time, while using satellite images and weather data to predict future water demand. Based on these data, the scheduling system can automatically calculate the optimal water resource allocation scheme and issue irrigation instructions through the control center to ensure that each region receives the optimal amount of water it needs. In addition, the system has the ability to quickly respond to drought. When detecting an increased risk of drought, the system can automatically adjust the irrigation strategy, such as increasing irrigation frequency, adjusting irrigation time to night to reduce evaporation, or starting an emergency water source plan. These intelligent response measures greatly enhance the adaptability of the irrigation system to extreme weather conditions and reduce the risk of crop yield reduction due to poor water resource management.

[0032] Specifically, the water use irrigation methods in the scheduling set include channel water source calling, path optimization calling, water source optimization calling, and water quantity optimization calling.

[0033] The steps of channel water source calling include: obtaining all channel paths in the current irrigation area, opening the currently obtained channel path, and synchronizing the channel water level of all channel paths.

[0034] The steps of path optimization calling include: obtaining the optimal path between the water pump and all plots in the current irrigation area, and irrigating the plots according to the optimal path.

[0035] The water source optimal calling step includes: acquiring the callable irrigation area adjacent to the current irrigation area, and irrigating the current irrigation area through the callable irrigation area.

[0036] The water amount optimal calling step includes: acquiring the required water amount and the callable water amount of all irrigation areas, and irrigating all irrigation areas through external water sources according to the difference between the required water amount and the callable water amount.

[0037] As known from the above, the channel water source calling mainly focuses on the effective utilization of the existing channels in the irrigation area. This method includes acquiring all available channel paths in the current irrigation area, ensuring that these paths are effectively opened, and adjusting so that the water levels of each channel reach consistency to achieve uniform distribution of water sources. The specific steps are as follows: first, the system collects all channel path information in the current irrigation area; then the system sends instructions to the relevant control devices to open these channel paths; the system monitors and adjusts the water level in the channel until the water levels of all channels are consistent, ensuring smooth transfer of water flow between channels, avoiding overabundance or deficiency of water resources in some areas. In a large farmland, multiple irrigation channels run through the entire area. When it is detected that the water level in some areas is lower than the standard water level, the system will automatically open the water supply of the corresponding channel until the water level of all channels reaches the preset balanced state.

[0038] The path optimal calling determines the optimal water flow path from the water pump to each grid field through an algorithm. This method aims to accurately calculate and implement the most effective water resource transmission path through technical means to achieve the highest water resource utilization efficiency with the least cost. The specific steps are as follows: the algorithm analyzes the connection paths between the water pump and all grid fields in the current irrigation area, and selects the shortest or lowest cost path; according to these optimal paths, water resources are allocated and irrigation operations are performed. In a drought area, in order to maximize the benefits of limited water sources, the system calculates the shortest distance from the main water pump to each irrigation grid field, and realizes on-demand irrigation through an automatic control system.

[0039] The water source optimal calling focuses on the sharing and scheduling of water resources across regions. This method identifies and utilizes available water sources in adjacent irrigation areas to supplement the water resource demand of the current area, and is particularly suitable for situations where water resources are scarce in some areas. The specific steps are as follows: the system evaluates and determines the available water sources in other areas adjacent to the current irrigation area; through the control system, these additional water sources are introduced into the current area to alleviate the problem of water shortage. When a irrigation area encounters unexpected drought, and the adjacent area has surplus water resources, the system will automatically start the water transfer operation from the adjacent area to ensure that the water demand of the affected area is met.

[0040] Water quantity optimal calling is to implement precise allocation of water resources by comprehensively considering the difference between the demand water quantity and the callable water quantity of all irrigation areas. This way ensures that under the condition of limited resources, the maximum utilization of water resources is realized through scientific calculation and strategy. The specific steps are as follows: the system first obtains the water quantity demand and the existing water quantity of all irrigation areas; the system determines the necessary water quantity allocation strategy according to the difference between demand and supply, and meets the demand of all areas through external water source or internal allocation. In the dry season when the total water quantity in the whole area is not enough to meet all the demand, the system will calculate the gap of each area and prioritize the water quantity demand of key crop areas, while possibly reducing the water supply to non-key areas.

[0041] S102, a selection optimization algorithm for selecting water irrigation mode for each irrigation area at each time is constructed.

[0042] As known above, a selection optimization algorithm is constructed, which is specially used to select the most suitable water irrigation mode for each irrigation area at each time. The core of this algorithm is to intelligently respond to the changing environmental conditions, ensuring that at any given time, each irrigation area can adopt the optimal irrigation mode for water resource scheduling and management, so as to achieve the purpose of water saving and efficiency improvement and promoting crop growth. It can analyze the specific needs and external environmental factors of each irrigation area in real time, such as soil moisture, weather conditions, water source conditions and crop types, and select the most suitable irrigation mode accordingly. This includes but is not limited to drip irrigation, sprinkler irrigation, flooding irrigation, etc., each of which has different performance advantages in terms of water efficiency and crop demand under specific conditions.

[0043] S103, the local state set of each irrigation area is obtained in real time, and all local state sets are input into the selection optimization algorithm to select the water irrigation mode of all irrigation areas at the same time.

[0044] As known above, the real-time data of each irrigation area is continuously monitored and collected, such as soil moisture, temperature, crop growth conditions, water level and other key parameters. These data are the basis for understanding the current demand of each area and are the key basis for making irrigation decisions. In a wide agricultural production area, sensors are deployed at key points to collect soil moisture and temperature data, which are sent to the central processing system in real time every few minutes through wireless network.

[0045] The collected data is aggregated and inputted into a selection optimization algorithm. This algorithm utilizes these data to evaluate and determine the most suitable irrigation method for each region, taking into account the weights and priorities of various factors in processing the data to ensure the rational allocation and use of water resources. If the soil moisture of a certain region is below a predetermined threshold and the weather forecast shows no rainfall in the future, the selection optimization algorithm will automatically recommend starting the drip irrigation system to precisely control water directly to the roots while reducing water waste.

[0046] By acquiring the local state set of each irrigation region in real time and utilizing the selection optimization algorithm for data analysis and decision-making, this process provides strong data support and intelligent decision-making capabilities for irrigation management. This approach enables the irrigation system to dynamically adjust irrigation strategies to adapt to changing environmental conditions and crop needs, thereby maximizing the efficiency of water resource utilization and optimizing the growth environment for crops.

[0047] Specifically, the steps of selecting the water irrigation method for the irrigation region by the selection optimization algorithm include:

[0048] The action space set of each irrigation region is constructed by the irrigation water allocation and the type of water irrigation method, and an independent local network is constructed for each irrigation region.

[0049] According to the local state set, the state set of the irrigation region is obtained, and the action value of all actions of the irrigation region in the action space set is calculated by the local network.

[0050] The action with the highest action value is retained, and the current irrigation region is adjusted according to the irrigation water allocation and water irrigation method contained in the retained action.

[0051] According to the above specific description, define and set the range of all irrigation actions that each irrigation region can take, including different irrigation methods (such as drip irrigation, sprinkler irrigation, etc.) and their corresponding water allocation. The construction of the action space set is based on the water efficiency and adaptability of specific irrigation methods, providing all feasible options for subsequent decision-making. For example, for an irrigation region that is often affected by drought, its action space set may include efficient water-saving drip irrigation and low-water-demand spray irrigation, each with specific water allocation strategies.

[0052] After obtaining the action space set, the algorithm will calculate the expected effect of each possible action, i.e., the action value, based on the current local state set, including soil moisture, weather conditions, crop growth stage, and other data. This step is accomplished by independent local networks for each irrigation area, which use machine learning techniques to evaluate the performance of various actions under current conditions. In one irrigation area, the local network may analyze the action value of drip irrigation and fixed water supply under current high-temperature and drought conditions to determine which irrigation method can more effectively utilize water resources and support crop growth.

[0053] Based on the calculation results of the local network, the algorithm will select the irrigation strategy with the highest action value and implement the corresponding water allocation and irrigation method. This ensures that each irrigation decision is data-driven, achieving optimal resource utilization efficiency and crop growth conditions. If the analysis shows that drip irrigation has a higher action value than traditional surface irrigation in continuous high-temperature weather, the system will automatically adjust the irrigation settings, implement the drip irrigation strategy, and irrigate according to the calculated optimal water quantity.

[0054] As can be seen from the above, the selection of the optimization algorithm not only optimizes the allocation of water resources, but also improves irrigation efficiency and crop productivity. This process significantly enhances the adaptability of the irrigation system through real-time data analysis and intelligent decision-making, enabling it to flexibly respond to various environmental changes. In addition, by automatically selecting the most suitable irrigation method, the algorithm reduces water waste and supports sustainable agricultural practices.

[0055] Specifically, the step of retaining the action with the highest action value includes:

[0056]

[0057] wherein, is the action with the highest action value; is the action space set, and is the irrigation water allocation, is the irrigation method; is the number of irrigation areas, and is a positive integer from 1 to N; is the state set; is the parameter used to estimate the expected return of each action; is the action value.

[0058] As can be seen from the above, a set of action spaces containing different irrigation methods and water allocation configurations is constructed for each irrigation area. The action space set of each irrigation area consists of two parts: one is different irrigation methods, and the other is the water allocation configuration corresponding to each irrigation method. A comprehensive decision basis is provided for the algorithm, including the selectable irrigation methods and the corresponding water allocation quota. Next, the value of each action under the given local state set is evaluated by the local network. The local state set includes key parameters such as soil moisture, crop demand, weather conditions, etc., which are crucial for determining the most suitable irrigation method. The local network evaluates the potential benefits of each action based on these parameters, and the value of each action is calculated. Finally, the action with the highest value is retained, and the irrigation area is watered according to the irrigation water allocation and water use irrigation method of this action. This means that the system will automatically execute the irrigation configuration with the highest action value, ensuring that each irrigation area uses water resources in the most efficient way.

[0059] By intelligently analyzing the specific needs and environmental conditions of each irrigation area and then selecting the most suitable irrigation method and water allocation configuration, not only the efficiency of water resource use is improved, but also the crops can grow in the best conditions, thereby improving the overall benefits and sustainability of agricultural production.

[0060] Further, the system state is , and each local state includes: soil moisture , current water supply , crop water demand , precipitation , and historical irrigation amount . The global state includes: channel state , pump station state and other water source state .

[0061] Further, the irrigation water allocation represents the irrigation water allocation strategy for the area; the scheduling method selection selects one of the four scheduling methods: channel water source priority use, path optimal calculation, water source optimal calculation, and water quantity optimal calculation.

[0062] Further, the selection optimization algorithm includes the local network of all irrigation areas and the global network mixed constructed according to the local network of all irrigation areas, and has the following cases:

[0063] Case one, according to the action value of all local networks in each unit time, the comprehensive action value is obtained, and the global network is updated through the comprehensive action value.

[0064] Case two, the local network updates at the next time unit considering the degree of influence of the global network on the current local network.

[0065] As can be seen from the above, each irrigation area has a local network that calculates the value of different irrigation actions based on the specific environmental conditions and needs of the area, such as soil moisture, crop type, weather conditions, etc. The local network provides an action value by evaluating the potential benefits of each irrigation method and its supporting water volume, guiding the formulation of irrigation decisions. The global network synthesizes the action values of all local networks to build a comprehensive perspective, evaluating the overall action value of the entire irrigation area. This not only considers the optimal solution for a single area, but also considers how to adjust the allocation of resources throughout the system to achieve the optimization of water resource use in the entire irrigation area.

[0066] For the description of case one, at each unit time, the global network aggregates the action values of all local networks for their respective irrigation areas. These action values are calculated based on the adaptability and efficiency of each irrigation method, including the availability of channel water sources, the optimal irrigation path, the available water source, and the optimal allocation of water volume. The global network calculates a comprehensive action value based on these action values, which guides the irrigation strategy for the entire irrigation area, ensuring the highest water resource utilization efficiency and crop growth support in the entire area.

[0067] For the description of case two, the local network not only reflects the immediate status and needs of its own area, but also considers feedback from the global network at each time unit. This design allows each local decision to be optimized under the guidance of the global network, ensuring that local decisions are consistent with the overall water resource management strategy. If the global network indicates that a certain area needs to reduce water use to support the stability of the entire system, the corresponding local network will adjust its strategy, possibly switching from optimal water source calling to optimal water volume calling to adapt to this change.

[0068] Specifically, at each unit time, the local network of each irrigation area updates the action value, which includes:

[0069]

[0070] wherein, is the learning rate; is the discount factor; is the benefit value of the current action adopted by the irrigation area; is the state set of the next unit time; is the action of the next unit time; is the maximum action value among all actions in the next state set; is the adjustment coefficient, used to control the degree of influence of the global network on the update of the local network; is the gradient of the global network with respect to the local network, indicating the degree of influence of the global network on the local network; is the comprehensive action value of the global network; is the current action value.

[0071] For the above specific description, the decision of the irrigation area is updated at each time unit. Specifically, this includes calculating and updating the expected effect of each irrigation area, that is, the action value of the area, which is evaluated based on the current state and possible irrigation actions. This calculation not only considers the immediate reward (such as the direct impact of the current action on crop growth or water resource efficiency), but also considers the future expected reward, which is achieved by predicting the state of the next time unit and selecting the irrigation action that can bring the maximum expected benefit.

[0072] In addition, the algorithm particularly introduces an innovative adjustment term, namely the influence factor of the global network. This factor adjusts the local decision according to the evaluation of the global network, so that the irrigation strategy of each irrigation area not only responds to its local conditions, but also considers the water resource management and efficiency optimization of the entire irrigation area. Specifically, if the global network indicates that the irrigation action of a certain area may have a greater impact on the water resource balance of the entire system, this factor will adjust the action value calculation of the area to encourage it to make decisions that are more beneficial to the overall resource management.

[0073] As can be seen from the above, the core of the entire process lies in the coordinated work of the two main parts: the local network ensures that each irrigation area can make the most appropriate decision based on real-time data; while the global network monitors and adjusts the actions of the entire system, ensuring that all decisions collectively promote the maximum overall benefit of the entire irrigation area. Through this method, the operation of the irrigation system becomes more intelligent and automated, greatly improving the efficiency of water resources, while also optimizing the growth conditions of crops.

[0074] Further, the action value calculation formula is:

[0075]

[0076] wherein, is a neural network with parameters ; the input features are: ; and the action input is: .

[0077] Specifically, the comprehensive action value is obtained as:

[0078]

[0079] wherein, is a weight matrix for linearly transforming the action value of the local network; and is a vector of physical state measurements, and includes at least one item in the local state set; is a weight matrix used to adjust the influence of in the integrated action value calculation; is a one-hot encoding vector representing the water irrigation mode of the local network; is a weight matrix used to adjust the influence of water irrigation mode in the integrated action value calculation; is a bias term for the first layer of the global network; is an activation function used to introduce nonlinearity to help the global network capture complex relationships; is a weight matrix for the second layer, and is a bias term used to jointly convert the output of the hidden layer h into the integrated action value.

[0080] As can be seen from the above, first, the algorithm model constructs a hidden layer, and the purpose of this layer is to effectively integrate input data from different sources. These data include the expected benefit values of each irrigation area, which reflect the expected irrigation effect under specific operating conditions. In addition, it also includes physical state monitoring data such as soil moisture, air temperature, etc., which provides real-time environmental background information for the model. At the same time, the model also considers the operation mode, which may involve specific irrigation strategies or preset behavior patterns, such as automatic or manual control of the irrigation system.

[0081] In this hidden layer, the expected benefit values of each irrigation area are first weighted and synthesized with the corresponding physical state monitoring data and operation mode data. Each type of data has its corresponding weight, which is obtained through training, and the purpose is to adjust the influence degree of each input in the final output. After weighting, these data are aggregated and nonlinearly transformed through an activation function, usually a ReLU function. The role of the ReLU function is to enhance the model's ability to handle nonlinear problems, so that the model can better simulate complex irrigation environments and decision-making processes.

[0082] Next, the output of the hidden layer will be further processed to generate a global irrigation strategy. This step involves another weight matrix that converts the output of the hidden layer into the final global irrigation decision value. This conversion process also includes a bias term, which is used to adjust the output value to adapt to different operation requirements or optimization goals.

[0083] Specifically, at each unit time, the global network is updated through the integrated action value, which includes:

[0084] Calculate the difference value between the integrated action value of the current time unit and the predicted integrated action value of the next time unit.

[0085] The modified difference value is used as an input to the loss function to update the network parameters of the global network through the gradient descent method.

[0086] The gradient of the loss function with respect to each parameter is calculated through the backpropagation algorithm to reduce the loss.

[0087] In the above specific description, at each unit time, the system first needs to evaluate the effect of the current decision and predict the future situation. This involves calculating the comprehensive action value of the irrigation strategy implemented at the current time unit, and predicting the possible results at the next time unit if the current strategy is continued. By comparing these two values, a difference value can be obtained, which reflects the deviation between the immediate effect of the irrigation strategy and the future expectation.

[0088] To optimize the performance of the global network, the difference value will be used as an input to the loss function. The purpose of the loss function is to quantify the error between the current irrigation strategy and the ideal strategy, so that the system can make self-corrections. By applying the gradient descent method, the system will adjust the network parameters, i.e. automatically fine-tune the weights and biases that affect the quality of prediction and decision-making, to minimize the value of the loss function.

[0089] Backpropagation passes error information through network layers layer by layer, calculating the contribution of each parameter to the final error. This calculation process ensures that each update is targeted and can effectively improve the decision-making ability of the network, thereby maximizing the action value of the entire system.

[0090] As can be seen from the above detailed update process, the global network can adapt to complex environmental changes and changing irrigation needs. For example, in the face of sudden drought events, the global network can quickly adjust and optimize water resource allocation strategies to cope with the impact of reduced precipitation. This rapid response capability is the result of continuous learning and adaptation to actual irrigation effects and the differences between them and expected goals.

[0091] Specifically, the loss function includes the following formula:

[0092]

[0093] where, is the loss function, used to represent the deviation between the predicted comprehensive action value and the actual obtained value; is the global reward; is the discount factor; is the expected comprehensive action value under all possible actions in the next state set; is the comprehensive action value estimate of the current state set; is the network parameter of the current global network; is the parameter of the target network.

[0094] As can be seen, the loss function is constructed based on the difference between the irrigation effect predicted by the global network at the current and future states. Specifically, the function calculates the square of the difference between the actual effect after the execution of the current irrigation decision and the optimal effect that can be achieved in the future. This design enables the loss function to reflect the deviation between the immediate performance and long-term sustainability of the irrigation strategy.

[0095] At each evaluation period, the system calculates an effect value based on the current irrigation strategy, including the selected irrigation method and water allocation. At the same time, the system also predicts the highest effect that can be achieved if the current optimal strategy is continued in the next period. By comparing these two effect values, a difference value can be obtained, indicating the degree of deviation between the current strategy and the potential optimal strategy.

[0096] To optimize the prediction ability of the global network and the quality of the decision, the system will use gradient descent to adjust the network parameters. This process involves calculating the gradient of the loss function with respect to the global network parameters, and then adjusting these parameters to reduce the loss. Through this method, the network can gradually learn how to more accurately predict irrigation effects and more effectively allocate resources.

[0097] The backpropagation algorithm is used here to effectively calculate the gradient of the loss function with respect to each network parameter. This algorithm passes error information back through the network layers to find the key parameter adjustment direction that improves prediction accuracy and decision quality. This not only helps the system improve the current irrigation strategy, but also optimizes the response to future irrigation needs.

[0098] Further, the global reward reflects the benefits of the entire irrigation system, defined as the sum of the rewards of each region:

[0099]

[0100] The reward of each region is defined based on the following physical quantities:

[0101]

[0102] where, is the importance coefficient of soil moisture; is the importance coefficient of crop water requirement; is the water resource use cost coefficient; is the positive impact coefficient of precipitation on reward; is the penalty coefficient for violating physical constraints; is the penalty term based on physical constraints (such as water source capacity exceeding limit).

[0103] Further, the comprehensive action value is obtained by weighted aggregation of all local network action values, and the aggregation weight of each action value in the next time unit is determined by the corresponding local state set in the current time unit.

[0104] As can be seen from the above, in each time unit, the local network of each irrigation area calculates an action value according to the current irrigation condition and environmental state. These action values reflect the expected benefits of taking a specific irrigation action under the current conditions. For example, when a certain area is facing drought conditions, the local network may calculate a higher action value for increasing irrigation.

[0105] All local network calculated action values will be aggregated to form a comprehensive action value. The calculation of this comprehensive value is not simply a simple addition, but a weighted method, in which the weight of each action value is determined by the corresponding local state set. This means that if the irrigation demand of a certain area is urgent or has a greater impact on the overall irrigation effect, the action value of this area will obtain a higher weight in the aggregation.

[0106] The aggregation weight of each action value is determined based on the local state set in the current time unit. This includes soil moisture, crop growth stage, weather forecast and other information. This method ensures that the weight distribution can reflect the actual needs and priorities of each area in real time, so that the global irrigation strategy is more in line with the actual situation.

[0107] The automatic water resource optimization scheduling management method based on irrigation area demonstration area provided by the present application greatly increases the flexibility of the system. Each area can choose the most suitable irrigation method according to the actual situation, so as to effectively cope with different environmental changes. In the case of insufficient global resources to meet all needs, the selection of optimization algorithm can evaluate and determine the best irrigation strategy for each area in real time. Not only based on the immediate needs of each irrigation area, but also considering the water resource situation of the whole irrigation area, to ensure that water resources are reasonably allocated and utilized where they are most needed.

[0108] By real-time acquisition and analysis of the local state set of each irrigation area, the system can accurately control the distribution of water quantity and reduce water resource waste. This fine management not only improves the overall efficiency of water resources, but also ensures the effective use of water resources under extreme climate conditions.

[0109] Allowing the irrigation system to dynamically adjust the irrigation strategy according to the actual soil moisture, crop growth conditions and weather forecast and other local states. This adaptive ability enables the irrigation system to better cope with climate change and reduce crop losses caused by improper irrigation.

[0110] Irrigation is a key link in agricultural production, and the application helps agricultural producers improve crop yield and reduce environmental risk by ensuring that each decision is based on the latest data and algorithm optimization. Not only does it increase agricultural output, but it also helps achieve sustainable development of agriculture.

[0111] In the description of the application, it should be understood that the terms "longitudinal", "transverse", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship shown in the drawings, and are only used to facilitate the description of the application, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the application.

[0112] The above examples only describe the preferred mode of the application, and do not limit the scope of the application. Without departing from the design spirit of the application, various modifications and improvements to the technical solutions of the application made by those skilled in the art shall fall within the protection scope determined by the claims of the application.

Claims

1. An automated water resource optimal scheduling management method based on a demonstration area of an irrigation district, characterized in that, The method comprises the following steps: acquiring N irrigation areas in a current irrigation area of interest, and setting up a scheduling set comprising a plurality of water-using irrigation modes for each of the irrigation areas; constructing a global network of the irrigation area according to the action value of a local network of each of the irrigation areas generated by the adaptability and efficiency of each irrigation mode of all the irrigation areas in the same unit time, constructing the local network of each of the irrigation areas according to the instant state and demand of the respective irrigation area and the feedback from the global network in each unit time, and updating the action value of the local network of each of the irrigation areas by the following formula: In the formula, the is the learning rate of the local network of the irrigation area; the is the action of the corresponding local network of the irrigation area in the current unit time; the is the number of the irrigation area, and is a positive integer from 1 to N; the is a state set; the is a discount factor; the is the benefit value of the irrigation area adopting the current action; the is a state set of the next unit time; the is an action of the next unit time; the is the maximum action value of all actions in the next state set; the is an adjustment coefficient for controlling the influence degree of the global network on the local network update; the is the gradient of the global network relative to the local network, for indicating the influence degree of the global network on the local network; the is the comprehensive action value of the global network; and the is the current action value. acquiring a local state set of each of the irrigation areas in real time, inputting all the local state sets into a selection optimization algorithm comprising the local network and the global network to select the water-using irrigation mode of all the irrigation areas at the same time; which comprises: constructing an action space set of each of the irrigation areas by the irrigation distribution water amount and the type of water-using irrigation mode; acquiring a state set of the irrigation area according to the local state set, calculating the action value of all actions in the action space set of the irrigation area by the local network; retaining the action with the highest action value, and adjusting the water of the current irrigation area according to the irrigation distribution water amount and the water-using irrigation mode contained in the retained action; wherein the selection optimization algorithm comprises the local network of all the irrigation areas and the global network mixedly constructed according to the local network of all the irrigation areas, and has the following cases: Case One: acquiring a comprehensive action value according to the action value of all the local networks in each unit time, and updating the global network by the comprehensive action value; Case Two: considering the influence degree of the global network on the current local network in the update of the local network in the next unit time.

2. The method for automated water resources optimal scheduling management according to claim 1, wherein, The water-using irrigation mode of the scheduling set comprises channel water source calling, path optimal calling, water source optimal calling and water amount optimal calling; The step of the channel water source calling comprises: acquiring all channel paths in the current irrigation area, and opening the currently acquired channel paths until the channel water levels of all the channel paths in the current irrigation area are consistent; The step of the path optimal calling comprises: acquiring the optimal path between the water pump and all the plots in the current irrigation area, and irrigating the plots according to the optimal path; The step of the water source optimal calling comprises: acquiring the callable irrigation area adjacent to the current irrigation area, and irrigating the current irrigation area through the callable irrigation area; The step of the water amount optimal calling comprises: acquiring the demand water amount and the callable water amount of all the irrigation areas, and irrigating all the irrigation areas through an external water source according to the difference between the demand water amount and the callable water amount.

3. The method of claim 1, wherein, The step of retaining the action with the highest action value comprises: wherein the is the action with the highest action value; the is the number of the irrigation area, and is a positive integer from 1 to N; the is the set of action spaces, and the is the irrigation water allocation, the is the irrigation water use mode; the is the parameter used to estimate the expected return of each action; the is the action value.

4. The method for automated water resources optimal scheduling management according to claim 1, wherein, The acquisition of the comprehensive action value is: wherein the is a weight matrix for linearly transforming action values of the local network; the is a physical state measurement vector and comprises at least one of the local state set; the is a weight matrix for adjusting influence in the integrated action value calculation; the represents a one-hot encoding vector of water irrigation modes of the local network; the is a weight matrix for adjusting influence of water irrigation modes in the integrated action value calculation; the is a bias term for the first layer of the global network; the is an activation function for introducing non-linearity to help the global network to capture complex relationships; the is a weight matrix of the second layer, and the is a bias term, jointly used to convert the output of the hidden layer h to the integrated action value.

5. The method for automated water resources optimal scheduling management according to claim 1, wherein, In each unit time, the global network is updated by the comprehensive action value, which specifically comprises: calculating the difference value between the comprehensive action value of the current time unit and the predicted comprehensive action value of the next time unit; The modified difference value is used to update network parameters of the global network by a gradient descent method through a loss function; The gradient of the loss function with respect to each parameter is calculated by a back propagation algorithm to reduce the loss.

6. The method for automated water resources optimal scheduling management according to claim 5, wherein, The loss function includes the following formula: Wherein, the is a loss function, used to characterize the deviation between the predicted comprehensive action value and the actual obtained value; the is a global reward; the is a discount factor; the is the expected comprehensive action value under all possible actions in the next state set; the is the comprehensive action value estimate of the current state set; the is the network parameter of the current global network; and the is the parameter of the target network.

7. The method for automated water resources optimal scheduling management according to claim 1, wherein, The comprehensive action value is obtained by a weighted set of action values of all local networks, and the set weight of each action value in the next time unit is determined by the corresponding local state set in the current time unit.

Images