Water distribution optimization method and system for well-canal combined irrigation area based on deep reinforcement learning

Abstract

The present invention discloses a method and system for optimizing water distribution in a well-canal combined irrigation area based on deep reinforcement learning. The method includes: obtaining geographic satellite data, motor well information, canal information, soil moisture forecast data, and meteorological forecast data of a target area; dividing the target area into a well irrigation sub-area, a canal irrigation sub-area, and a well-canal combined irrigation sub-area according to the motor well information and the canal information; combining the soil moisture forecast data and the meteorological forecast data, predicting the irrigation water demand of each sub-area in each period of the future water distribution cycle; constructing and training a real-time water distribution model for a well-canal combined irrigation area based on deep reinforcement learning; for each well irrigation sub-area and canal irrigation sub-area, calculating the water distribution amount according to the irrigation water demand, for each well-canal combined irrigation sub-area, obtaining the optimal strategy according to the trained model, and calculating the water distribution amount, and finally synthesizing the water distribution amounts of all sub-areas as the optimal water distribution plan output. The present invention can obtain a real-time water distribution plan according to meteorological data, and the water distribution plan is more accurate.

Description

Technical Field

[0001] The present invention relates to water distribution optimization technology, and in particular to a method and system for optimizing water distribution in a well-channel combined irrigation area based on deep reinforcement learning. Background Art

[0002] The well-channel combined water intake mode in irrigation areas is an effective way to improve the efficiency of water resource utilization in irrigation areas. During peak water use, groundwater can be used to replenish irrigation gaps, and the seepage replenishment of canal water diversion irrigation enables groundwater to be reused for a second time. However, in daily use, on the one hand, it is common that well water irrigation is given priority because of the untimely canal water diversion. The well-channel combined irrigation mode without reasonable water allocation is prone to excessive groundwater exploitation. On the other hand, insufficient groundwater exploitation is prone to excessive groundwater uplift and land salinization. At present, the research on well-channel combined irrigation mainly allocates the medium- and long-term water volume ratio of well irrigation water sources and canal irrigation water sources, and fails to guide the real-time water distribution process considering rainfall. The existing allocation methods are mainly based on physical models and intelligent optimization algorithms to calculate the allocation ratio of well-channel irrigation in irrigation areas, or to derive an ex ante allocation plan, but cannot derive a water distribution plan based on real-time meteorological conditions, such as rainfall. Summary of the invention

[0003] In view of the problems existing in the prior art, the purpose of the present invention is to provide a method and system for optimizing water distribution in a well-channel combined irrigation area based on deep reinforcement learning, which can obtain a water distribution plan according to real-time meteorological conditions.

[0004] In order to achieve the above-mentioned object of the invention, the present invention provides the following technical solutions:

[0005] A method for optimizing water distribution in a well-canal combined irrigation area based on deep reinforcement learning, comprising the following steps:

[0006] (1) Obtain geographic satellite data, well information, channel information, soil moisture forecast data, and meteorological forecast data of the target area;

[0007] (2) In combination with the geographic satellite data of the target area, the target area is divided into several well irrigation sub-areas, several canal irrigation sub-areas, and several well-canal combined irrigation sub-areas according to the well information and canal information;

[0008] (3) Combine soil moisture forecast data and meteorological forecast data to predict the irrigation water demand of each sub-area in each period of the future water distribution cycle;

[0009] (4) Constructing and training a real-time water distribution model for well-canal irrigation areas based on deep reinforcement learning, wherein the real-time water distribution model for well-canal irrigation areas takes the influencing factors that affect the real-time water distribution of the well-canal irrigation sub-area as the state, takes the start and stop of each well in the well-canal irrigation sub-area as the action, takes the quantification of the difference between the groundwater level and the exploitable water level after the action is executed as the immediate reward, takes the quantification of the influencing factors that exceed the preset irrigation constraints after the action is executed as the penalty, and outputs the set of actions that maximize the cumulative reward in the future water distribution cycle as the optimal strategy;

[0010] (5) For each well-irrigation sub-area and canal-irrigation sub-area, the water allocation is calculated based on the irrigation water demand in each period of the future water allocation cycle. For each well-canal combined irrigation sub-area, the optimal strategy is obtained based on the trained real-time water allocation model of the well-canal combined irrigation area, and the water allocation is calculated based on the optimal strategy. Finally, the water allocation of all sub-areas is combined as the optimal water allocation plan.

[0011] A well-canal combined irrigation area water distribution optimization system based on deep reinforcement learning, including:

[0012] Data acquisition module, used to obtain geographic satellite data, well information, channel information, soil moisture forecast data, and meteorological forecast data of the target area;

[0013] The regional division module is used to combine the geographic satellite data of the target area and divide the target area into several well irrigation sub-areas, several canal irrigation sub-areas and several well-canal combined irrigation sub-areas according to the well information and channel information;

[0014] The water demand forecasting module is used to combine soil moisture forecast data and meteorological forecast data to forecast the irrigation water demand of each sub-area in each period of the future water distribution cycle;

[0015] A model building and training module is used to build and train a real-time water distribution model for well-canal combined irrigation areas based on deep reinforcement learning, wherein the real-time water distribution model for well-canal combined irrigation areas takes the influencing factors that affect the real-time water distribution of the well-canal combined irrigation sub-area as the state, takes the start and stop of each well in the well-canal combined irrigation sub-area as the action, takes the quantification of the difference between the groundwater level and the exploitable water level after the action is executed as the immediate reward, takes the quantification of the influencing factors that exceed the preset irrigation constraints after the action is executed as the penalty, and outputs the set of actions that maximize the cumulative reward in the future water distribution cycle as the optimal strategy;

[0016] The water distribution optimization module is used to calculate the water distribution amount for each well irrigation sub-area and canal irrigation sub-area based on the irrigation water demand in each period in the future water distribution cycle. For each well-canal combined irrigation sub-area, the optimal strategy is obtained based on the trained well-canal combined irrigation area real-time water distribution model, and the water distribution amount is calculated based on the optimal strategy. Finally, the water distribution amount of all sub-areas is integrated as the optimal water distribution plan.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention provides a method and system for optimizing water distribution in a well-canal combined irrigation area based on deep reinforcement learning, which can realize real-time water distribution according to real-time meteorological data. For the water distribution in the well-canal combined irrigation area, the present invention adopts a real-time water distribution model of a well-canal combined irrigation area based on deep reinforcement learning for real-time water distribution, and the water distribution plan is more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flow chart of a method for optimizing water distribution in a well-canal combined irrigation area based on deep reinforcement learning provided by an embodiment of the present invention;

[0019] Figure 2 This is the regional division and project distribution map of Doufuwo Irrigation District;

[0020] Figure 3 It is a diagram of groundwater depth process;

[0021] Figure 4 It is the daily water supply in each area of ​​the target area. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0023] This embodiment provides a method for optimizing water distribution in a well-canal combined irrigation area based on deep reinforcement learning. Figure 1 As shown, the following steps are included:

[0024] S101. Obtain geographic satellite data, well information, channel information, soil moisture forecast data, water supply data, and meteorological forecast data of the target area.

[0025] Among them, the geographic satellite data can select the MODIS data of the target area. The machine well information includes the number of machine wells, location, pumping flow, and irrigation radius; the channel information includes the channel topology, channel design flow, and water diversion flow; the soil moisture forecast data includes soil moisture content; the water supply data includes surface water, groundwater supply, groundwater level, etc.; the meteorological forecast data includes meteorological data such as rainfall, temperature, humidity, atmospheric pressure, wind speed, wind direction, and solar radiation.

[0026] S102. In combination with the geographic satellite data of the target area, the target area is divided into a number of well irrigation sub-areas, a number of canal irrigation sub-areas and a number of well-canal combined irrigation sub-areas according to the pumped well information and the canal information.

[0027] In specific implementation, this step includes:

[0028] S1021. Divide the target area into a well irrigation area, a canal irrigation area, and a well-canal combined irrigation area according to the irrigation water source based on the pumped well information and the canal information;

[0029] S1022, dividing the area belonging to the irrigation range of each pumped well in the well irrigation area into a well irrigation sub-area, thereby dividing the well irrigation area into X well irrigation sub-areas [O1, ..., O x ,…,O X ], where X is a positive integer greater than 0, x represents the xth well irrigation sub-area;

[0030] S1023, dividing the irrigation area of ​​each canal in the canal irrigation area into a canal irrigation sub-area, thereby dividing the canal irrigation area into Y canal irrigation sub-areas [V1,…,V y ,…,V Y ], where Y is a positive integer greater than 0, V y represents the yth canal irrigation sub-area;

[0031] S1024, dividing the irrigation area of ​​each canal in the well-canal combined irrigation area into a well-canal combined irrigation area sub-area, thereby dividing the well-canal combined irrigation area into Z canal irrigation sub-areas [OV1, ..., OV z ,…,OV Z ], where Z is a positive integer greater than 0, OV z It represents the zth well-canal combined irrigation sub-area. Each well-canal combined irrigation sub-area is also the irrigation range of a single or multiple pumped wells. There are three types of water sources in the well-canal combined irrigation area: pumped water from pumped wells, water diversion from canals, and rainfall.

[0032] S103. Combine soil moisture forecast data and meteorological forecast data to predict the irrigation water demand of each sub-area in each period of the future water distribution cycle.

[0033] In specific implementation, the steps include:

[0034] S1031. For each sub-area, the planting area of ​​the crops planted in the sub-area is obtained; specifically, the planting area can be obtained through supervised classification methods and / or entity survey results. The planting area of ​​the well irrigation sub-area, canal irrigation sub-area, and well-canal combined irrigation sub-area can be expressed as: , , , , Respectively represent well irrigation sub-area O x , Quguanzi District V y , well-canal combined irrigation sub-area OV z of planted area;

[0035] S1032. For each sub-area, the Penman formula is used to calculate the crop evapotranspiration in each period of the future water distribution cycle based on the meteorological forecast data and the soil moisture forecast data, and the crop evapotranspiration is used as the initial water depth required for the crop. The Penman formula is as follows:

[0036] ,

[0037] Where: ET0 is the evapotranspiration of crops, in mm / d; R n is the net radiation on the crop surface, in MJ / (m 2 ·d); G is the soil heat flux density, in MJ / (m 2 d); is the average temperature in °C; u2 is the wind speed in m / s; e s is the saturated water vapor pressure, in kPa; e a is the actual water vapor pressure, in kPa; is the slope of the saturated water vapor temperature curve, in kPa / ℃; is the hygrometer constant, in kPa / ℃;

[0038] S1033. For each sub-area, multiply the crop coefficient of the planted crop by the initial water depth required for planting the crop to obtain the actual water depth required for planting the crop; wherein the crop coefficient is an empirical value;

[0039] S1034. For each sub-area, the actual water depth required by the planted crops in each period of the future water distribution cycle is multiplied by the planting area of ​​the planted crops to obtain the irrigation water demand of the sub-area in each period of the future water distribution cycle.

[0040] For example, well irrigation area O x The irrigation water demand for each period in the future water distribution cycle is:

[0041] ,

[0042] In the formula, represents the irrigation water demand in period t in the future water distribution cycle, It indicates the actual water depth required by the crops in time period t in the future water distribution cycle. The water distribution cycle can be divided into multiple time periods according to the day, that is, one time period is one day.

[0043] S104. Build and train a real-time water distribution model for well-canal combined irrigation areas based on deep reinforcement learning.

[0044] Among them, the real-time water distribution model of the well-channel combined irrigation area takes the influencing factors that affect the real-time water distribution of the well-channel combined irrigation sub-area as the state, the start and stop of each well in the well-channel combined irrigation sub-area as the action, the quantification of the difference between the groundwater level and the exploitable water level after the action is executed as the immediate reward, the quantification of the influencing factors that exceed the preset irrigation constraints after the action is executed as the penalty, and the set of actions that will obtain the maximum cumulative reward in the future water distribution cycle is used as the optimal strategy output.

[0045] Specifically, the state of the real-time water distribution model of the well-channel combined irrigation area is:

[0046] ,

[0047] In the formula, s t represents the state of time period t, T represents the total number of time periods in the future water distribution cycle, They represent the rainfall, irrigation water demand, total irrigation water demand during the water distribution cycle, canal water supply, and pumped well water supply in the well-canal combined irrigation sub-area during period t;

[0048] The actions of the real-time water distribution model of well-channel combined irrigation area are:

[0049] ,

[0050] In the formula, a t represents the action during time period t, They represent the start / stop indicator variables of the 1st, …, jth, …, Jth pumped wells in the well-canal combined irrigation sub-area, respectively. The specific values ​​are 0 / 1. When the value is 1, it means it is started, and when the value is 0, it means it is stopped. J is the number of pumped wells in the well-canal combined irrigation sub-area.

[0051] The instant reward of the real-time water distribution model of well-canal combined irrigation area is:

[0052] ,

[0053] In the formula, r t represents the immediate reward in period t, Z t is the groundwater level during period t, Z red The exploitable water level of groundwater. are the quantitative coefficients respectively. In the immediate reward function, when the groundwater level is much higher than the exploitable water level, that is, the groundwater level is shallower than the exploitable water level, then Close to 0, the reward value r t Smaller; when the groundwater level is above but close to the exploitable water level, r t Larger; when the groundwater level is deeper than the exploitable water level, that is, exceeds the exploitable amount, then Z red -Z t is a negative value, rt By quantifying the objective function into an immediate reward value, when the groundwater volume is high, the use of well irrigation water can be increased to avoid excessive groundwater uplift and land salinization, and when the groundwater volume is low, the use of canal irrigation water can be increased.

[0054] The state transfer function of the real-time water distribution model of the well-channel combined irrigation area is:

[0055] ,

[0056] In the formula, s t+1 Indicates the state of the t+1 period, Indicates that from state s t Execute an action a t Then transfer to state s t+1 , They represent the rainfall, irrigation water demand, total irrigation water demand during the water distribution cycle, canal water supply, and pumped well water supply in the well-canal combined irrigation sub-area during the t+1 period; They represent the canal water diversion and pumped water volume of the well-canal combined irrigation sub-area in period t, represents the canal water utilization coefficient, represents the maximum flow rate of the channel, Indicates the time interval. represents the rated water extraction flow of pumped well j, They represent the recharge of groundwater by channel water diversion, pumped water from wells, and rainfall, respectively. is the groundwater exploitability coefficient, They represent the recharge coefficients of channel water diversion, pumped water from wells, and rainfall on groundwater. is the field water utilization coefficient, Area is the area of ​​the well-canal combined irrigation sub-area, and J is the number of wells in the well-canal combined irrigation sub-area.

[0057] The preset irrigation constraints include:

[0058] A. Channel water diversion constraints: ;

[0059] B. Constraints on pumping water from pumped wells: ;

[0060] C. Groundwater level fluctuation constraints: ;

[0061] D. Water demand constraints: ;

[0062] In the formula, represents the canal water diversion volume of the well-canal combined irrigation sub-area in period t, They represent the canal water supply and pumped well water supply of the well-canal combined irrigation sub-area in period t, They represent the minimum and maximum flow rates of the channel respectively. is the water extraction volume of well j in period t, represents the rated water extraction flow of well j, represents the canal water utilization coefficient, are the maximum and minimum water levels of well j in period t, is the water level of well j in period t, t represents time, and J is the number of wells in the well-channel combined irrigation sub-area.

[0063] The penalty of the real-time water distribution model of the well-canal combined irrigation area is:

[0064] ,

[0065] In the formula, c t represents the penalty in time period t, c1, c2, c3, c4, c5 represent the penalty values ​​of type 1, 2, 3, 4, 5 respectively, These are the conversion coefficients for the 1st, 2nd, 3rd, 4th, and 5th type penalties, representing the time interval.

[0066] The specific steps of the training process of the real-time water distribution model of the well-canal combined irrigation area are as follows:

[0067] 1) Initialize the experience pool with random weights and random weights Generate the evaluation network and the target network, set the total number of training rounds to EPISODE, the total number of water distribution cycle periods to T, the target network update step to C, the number of samples to N, the counter n=0, and the probability of randomly selecting an action to P k ;

[0068] 2) Set the training round e=1 and initialize the state s0;

[0069] 3) Set the initial state to s0;

[0070] 4) Generate a random number p in [0,1]. If p <P k Then randomly select an action as the action a in the current time period t Otherwise, select the action with the maximum value function value as the current action, and the maximum value function value is the maximum value of the accumulated reward;

[0071] 5) Execute action a t , calculate the immediate reward r according to the action t and the state s of the next time period t+1 , save the current experience (s t , a t , rt ,s t+1 ) in the experience pool;

[0072] 6) Set the count n=n+1; if n<=N, return to step 4); if n>N, randomly sample from the experience pool to obtain sampling experience, and perform the gradient descent algorithm to calculate the loss based on the sampling experience ;

[0073] 7) Set t=t+1 and return to step 4) until t>T, then go to step 8);

[0074] 8) Set e=e+1, and return to step 3), and synchronize the evaluation network to the target network every C rounds , when the training round e reaches EPISODE, the training is completed.

[0075] S105. For each well irrigation sub-area and canal irrigation sub-area, the water distribution amount is calculated according to the irrigation water demand in each period in the future water distribution cycle. For each well-canal combined irrigation sub-area, the optimal strategy is obtained according to the trained well-canal combined irrigation area real-time water distribution model, and the water distribution amount is calculated according to the optimal strategy. Finally, the water distribution amount of all sub-areas is combined as the optimal water distribution plan.

[0076] In specific implementation, the steps include:

[0077] S1051. For each well irrigation sub-area, the water allocation is calculated according to the following formula based on the irrigation water demand in each period of the future water allocation cycle:

[0078] ,

[0079] In the formula, Respectively represent the x-th well irrigation sub-area O x The water distribution, irrigation water demand, pumped well water supply, and pumped well rated water flow rate in period t. Indicates the time interval;

[0080] S1052. For each canal irrigation sub-area, the water allocation is calculated according to the following formula based on the irrigation water demand in each period of the future water allocation cycle:

[0081] ,

[0082] In the formula, Respectively represent the yth channel irrigation sub-area V y Water allocation, irrigation water demand, channel water supply, and maximum channel flow in period t;

[0083] S1053. For each well-canal combined irrigation sub-area, an optimal strategy is obtained according to the trained real-time water distribution model of the well-canal combined irrigation area, and the water distribution amount of each well-canal combined irrigation sub-area is calculated according to the following formula based on the optimal strategy;

[0084] ,

[0085] In the formula, Respectively represent the zth well-canal combined irrigation sub-area OV z The channel water supply and pumped well water supply in period t, Respectively represent the zth well-canal combined irrigation sub-area OV z Irrigation water demand, pumped water from wells, and canal water supply in period t, represents the maximum flow rate of the channel, Indicates the zth well-canal combined irrigation sub-area OV z The rated water extraction flow rate of the well j is represents the optimal strategy, express The optimal action in time period t is express The start and stop indicator variable of well j in;

[0086] S1054. The water distribution of all sub-areas is comprehensively output as the optimal water distribution plan.

[0087] This embodiment also provides a well-channel combined irrigation area water distribution optimization system based on deep reinforcement learning. The system can be implemented in software and / or hardware and can be configured in a terminal device, including:

[0088] Data acquisition module, used to obtain geographic satellite data, well information, channel information, soil moisture forecast data, and meteorological forecast data of the target area;

[0089] The regional division module is used to combine the geographic satellite data of the target area and divide the target area into several well irrigation sub-areas, several canal irrigation sub-areas and several well-canal combined irrigation sub-areas according to the well information and channel information;

[0090] The water demand forecasting module is used to combine soil moisture forecast data and meteorological forecast data to forecast the irrigation water demand of each sub-area in each period of the future water distribution cycle;

[0091] A model building and training module is used to build and train a real-time water distribution model for well-canal combined irrigation areas based on deep reinforcement learning, wherein the real-time water distribution model for well-canal combined irrigation areas takes the influencing factors that affect the real-time water distribution of the well-canal combined irrigation sub-area as the state, takes the start and stop of each well in the well-canal combined irrigation sub-area as the action, takes the quantification of the difference between the groundwater level and the exploitable water level after the action is executed as the immediate reward, takes the quantification of the influencing factors that exceed the preset irrigation constraints after the action is executed as the penalty, and outputs the set of actions that maximize the cumulative reward in the future water distribution cycle as the optimal strategy;

[0092] The water distribution optimization module is used to calculate the water distribution amount for each well irrigation sub-area and canal irrigation sub-area based on the irrigation water demand in each period in the future water distribution cycle. For each well-canal combined irrigation sub-area, the optimal strategy is obtained based on the trained well-canal combined irrigation area real-time water distribution model, and the water distribution amount is calculated based on the optimal strategy. Finally, the water distribution amount of all sub-areas is integrated as the optimal water distribution plan.

[0093] The system provided in the embodiment of the present invention can be used to execute the method provided in the embodiment of the present invention, and has the corresponding functions and beneficial effects of executing the method. For details, please refer to the method embodiment, which will not be repeated here.

[0094] It is worth noting that in the embodiment of the above system, the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0095] The embodiments described above are only illustrative, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, i.e., they may be located in one place, or they may be distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, or of course, it can be implemented only by hardware, as long as the function or effect can be achieved.

[0096] Finally, the present invention is simulated and verified. The simulation verification takes the typical area of ​​Doufuwo Irrigation District in Qihe, Shandong as the target area. Doufuwo Irrigation District is located in the southeast of Qihe County, Dezhou City. It governs Zhu'a Town and parts of Yancheng, Jiaomiao and Huadian Town. It is a medium-sized irrigation area for diverting the Yellow River in the lower reaches of the Yellow River. It realizes irrigation through pumping water from machine wells and diverting water from the Yellow River Gate, mainly planting grain crops. The region belongs to tropical subtropical marine climate and semi-humid climate, with an annual rainfall of 1500-2000 mm, an average annual temperature of 19℃-23℃, and an average annual sunshine hours of 1750-2200 hours. A total of 2 water diversion main canals and 6 branch canals have been built in the irrigation area, with 40 well irrigations, mainly concentrated in the west of Zhu'a Town, 9 groundwater level monitoring points and 5 soil moisture monitoring points. Among them, the typical area includes 15 well irrigations, 1 canal irrigation, 3 well-canal combination, 1 groundwater monitoring, and 1 soil moisture monitoring. It is a typical area for well-canal combined irrigation. Figure 2 The Doufuwo Irrigation Area is divided into regions and project distribution, where A, B, and C are well-canal combined irrigation areas, canal irrigation areas, and well-irrigation areas, respectively. Area A includes well-canal combined irrigation sub-areas {OV1, OV2, OV3}, area B includes canal irrigation sub-areas {V1} and 1 groundwater level monitoring point, and area C includes well irrigation sub-areas {O1, O2, ..., OV3}. 15 The main crop planted in Doufuwo Irrigation Area is winter wheat, which is mainly concentrated in the west of the irrigation area. The areas of A, B and C irrigation areas are 6.26 million m 2 , 5.8 million m 2 , 9.21 million m 2 According to the experimental research data of winter wheat, the crop coefficient kc values ​​at different growth stages are shown in Table 1. This example is calculated for the flowering-filling period of winter wheat, and the kc value is 1.24.

[0097] Table 1 Crop coefficient (kc) values ​​for winter wheat at different growth stages

[0098] Crop growth stage Seedling emergence-overwintering Wintering-Getting up Stand up-joint Jointing-heading Heading-Flowering Flowering-filling Grouting - Mature Whole reproductive period Crop coefficient - 0.54 0.76 0.99 1.15 1.24 1.18 0.9

[0099] The daily water demand process of each area in the next 7 days is shown in Table 2. It can be seen from Table 2 that the sub-areas {OV1, OV2, OV3}, {V1} and {O1, O2, ..., O 15 The water demand of the area is related to the regional area, and the daily water demand shows a trend of first decreasing and then increasing.

[0100] Table 2 Water demand process table of target area

[0101]

[0102] In this example, the groundwater monitoring point in area B is selected, and the exploitable water level is 7.9m. =0.75, The water demand is predicted on a daily basis, and the start and stop of the pumped wells are optimized on an hourly basis. Since there is no alternative water source in the pure well irrigation area, according to the irrigation habit, the pumped wells are turned on as long as water is needed. Therefore, in principle, the pumped wells in area C are turned on every day until the water demand of the area is met. Table 3 shows the daily start-up time of the pumped wells in the target area, where the area {O1, O2, ..., O 15 The running time of the wells in {OV1, OV2, OV3} is calculated by dividing the daily water demand of the region by the rated flow rate of the wells. It can be seen that the water extraction of the wells in this region meets the water demand requirements, and the daily running time of each well is less than 24 hours. The start-up and shutdown process of the wells in the region {OV1, OV2, OV3} is optimized by the real-time water distribution model of deep reinforcement learning. Since the optimization is performed hour by hour, the start-up time is rounded to the nearest hour. The water extraction of the wells in the well-channel combined area does not meet the regional water demand, and the remaining water demand is supplied by the channel. The start-up and shutdown process of the wells is shown in Tables 4 and 5.

[0103] Table 3. Pumped well startup time in target area

[0104]

[0105] Table 4 Start-up and shutdown process of wells in the well-channel combined irrigation sub-area from the 1st to the 12th hour

[0106]

[0107] Table 5 Start-up and shutdown process of wells in the well-channel combined irrigation sub-area from 13th to 24th hour

[0108]

[0109] The optimization results (abbreviated as optimization in the figure) of the scheme of the present invention are compared with the three operation strategies of fully open, fully closed and half open wells in the well-channel combined irrigation sub-area. The groundwater depth process of area B in the four cases is shown as follows Figure 3 As shown in the figure. When the wells are fully open, the groundwater depth decreases over time. At the end of the scheduling period, the groundwater depth is much greater than the exploitable water level. When the wells are fully closed, area B is supplied with water from the channel when the water supply allows. The channel irrigation and the wells in other areas are used to replenish the groundwater. The groundwater depth continues to increase, and the groundwater is not fully utilized. When the wells are half open, the water supply to area B is half from the wells and half from the channels. The groundwater depth exceeds the exploitable water level. Combining Tables 4 and 5, it can be seen that the wells in this optimization scheme are operated in the second half of the day, and the wells in area C are staggered for water extraction. During the entire scheduling period, the groundwater depth is less than the exploitable water level, and groundwater resources are used to a large extent. In all four cases, the groundwater depth shows a certain fluctuation pattern. This is caused by the concentrated water use of the wells in area C every day. The scheme of the present invention makes better use of the irrigation time difference between the two areas through deep reinforcement learning, and realizes the rational use of surface and groundwater in the well-channel combined area. Figure 4 The daily water supply of each area in the irrigation district is given. It can be seen that the total water extraction of the wells in the combined well-channel irrigation sub-area is less than the water diversion of the channels. Only on the 3rd and 5th days, the water extraction of the wells is higher than the water diversion of the channels. The water supply ratio of regional well irrigation and channel irrigation shows a daily dynamic change.

[0110] It should be understood that the above embodiments and descriptions only describe the principles, main features and advantages of the present invention. Without departing from the spirit and scope of the present invention, the present invention may be subject to various changes and improvements, and these changes and improvements all fall within the scope of protection of the present invention.

Claims

1. A method for optimizing water distribution in well-canal combined irrigation areas based on deep reinforcement learning, characterized in that: The following steps are involved: (1) Obtain geographic satellite data, well information, channel information, soil moisture forecast data, and meteorological forecast data of the target area; (2) In combination with the geographic satellite data of the target area, the target area is divided into several well irrigation sub-areas, several canal irrigation sub-areas, and several well-canal combined irrigation sub-areas according to the well information and canal information; (3) Combine soil moisture forecast data and meteorological forecast data to predict the irrigation water demand of each sub-area in each period of the future water distribution cycle; (4) Constructing and training a real-time water distribution model for well-canal irrigation areas based on deep reinforcement learning, wherein the real-time water distribution model for well-canal irrigation areas takes the influencing factors that affect the real-time water distribution of the well-canal irrigation sub-area as the state, takes the start and stop of each well in the well-canal irrigation sub-area as the action, takes the quantification of the difference between the groundwater level and the exploitable water level after the action is executed as the immediate reward, takes the quantification of the influencing factors that exceed the preset irrigation constraints after the action is executed as the penalty, and outputs the set of actions that maximize the cumulative reward in the future water distribution cycle as the optimal strategy; (5) For each well irrigation sub-area and canal irrigation sub-area, the water allocation is calculated based on the irrigation water demand in each period of the future water allocation cycle. For each well-canal combined irrigation sub-area, the optimal strategy is obtained based on the trained well-canal combined irrigation area real-time water allocation model, and the water allocation is calculated based on the optimal strategy. Finally, the water allocation of all sub-areas is combined as the optimal water allocation plan; Wherein, step (3) specifically includes: (3.1) For each sub-area, obtain the planting area of ​​the crops in the sub-area; (3.2) For each sub-area, the Penman formula is used to calculate the crop evapotranspiration in each period of the future water distribution cycle based on the meteorological forecast data and soil moisture forecast data, and the crop evapotranspiration is used as the initial water depth required for the crop. (3.3) For each sub-area, multiply the crop coefficient of the planted crop by the initial water depth required for the planted crop to obtain the actual water depth required for the planted crop; (3.4) For each sub-area, multiply the actual water depth required by the planted crops in each period of the future water distribution cycle by the planting area of ​​the planted crops to obtain the irrigation water demand of the sub-area in each period of the future water distribution cycle; The state of the real-time water distribution model of the well-channel combined irrigation area is: ， In the formula, s t represents the state of time period t, T represents the total number of time periods in the future water distribution cycle, They represent the rainfall, irrigation water demand, total irrigation water demand during the water distribution cycle, canal water supply, and pumped well water supply in the well-canal combined irrigation sub-area during period t; The actions of the real-time water distribution model of well-channel combined irrigation area are: , In the formula, a t represents the action during time period t, They represent the start / stop indicator variables of the 1st, …, jth, …, Jth pumped wells in the well-canal combined irrigation sub-area, with specific values ​​of 0 / 1. When the value is 1, it means it is on, and when the value is 0, it means it is off. J is the number of pumped wells in the well-canal combined irrigation sub-area; Step (5) specifically includes: (5.1) For each well irrigation sub-area, the water allocation is calculated according to the following formula based on the irrigation water demand in each period of the future water allocation cycle: , In the formula, Respectively represent the x-th well irrigation sub-area O x The water distribution, irrigation water demand, pumped well water supply, and pumped well rated water flow rate in period t. Indicates the time interval; (5.2) For each canal irrigation sub-area, the water allocation is calculated according to the following formula based on the irrigation water demand in each period of the future water allocation cycle: ， In the formula, Respectively represent the yth channel irrigation sub-area V y Water distribution, irrigation water demand, channel water supply, and maximum channel flow in period t, represents the canal water utilization coefficient; (5.3) For each well-canal combined irrigation sub-area, the optimal strategy is obtained according to the trained real-time water distribution model of the well-canal combined irrigation area, and the water distribution amount of each well-canal combined irrigation sub-area is calculated according to the following formula based on the optimal strategy; , In the formula, Respectively represent the zth well-canal combined irrigation sub-area OV z The channel water supply and pumped well water supply in period t, Respectively represent the zth well-canal combined irrigation sub-area OV z Irrigation water demand, pumped water from wells, and canal water supply in period t, represents the maximum flow rate of the channel, Indicates the zth well-canal combined irrigation sub-area OV z The rated water extraction flow rate of the well j is represents the optimal strategy, express The optimal action in time period t is Respectively The start and stop indicator variables of the wells 1, ..., j, ..., J in the well-canal combined irrigation sub-area, and T represents the total number of time periods in the future water distribution cycle; (5.4) The water allocation of all sub-areas is output as the optimal water allocation plan.

2. The method for optimizing water distribution in irrigation areas by combining wells and canals based on deep reinforcement learning according to claim 1, characterized in that: Step (2) specifically includes: (2.1) Based on the information of wells and canals, the target area is divided into well irrigation area, canal irrigation area and well-canal combined irrigation area according to the irrigation water source; (2.2) The area within the irrigation range of each pumped well in the well irrigation area is divided into a well irrigation sub-area, thereby dividing the well irrigation area into X well irrigation sub-areas [O1,…,O x ,…,O X ], where X is a positive integer greater than 0, x represents the xth well irrigation sub-area; (2.3) The irrigation area of ​​each canal in the canal irrigation area is divided into a canal irrigation sub-area, thereby dividing the canal irrigation area into Y canal irrigation sub-areas [V1,…,V y ,…,V Y ], where Y is a positive integer greater than 0, V y represents the yth canal irrigation sub-area; (2.4) The area of ​​irrigation of each canal in the well-canal combined irrigation area is divided into a well-canal combined irrigation area sub-area, thereby dividing the well-canal combined irrigation area into Z canal irrigation sub-areas [OV1,…,OV z ,…,OV Z ], where Z is a positive integer greater than 0, OV z It represents the zth well-channel combined irrigation sub-area, and each well-channel combined irrigation sub-area is the irrigation range of a single or multiple wells.

3. The method for optimizing water distribution in irrigation areas by combining wells and canals based on deep reinforcement learning according to claim 1, characterized in that: The instant reward of the real-time water distribution model of well-canal combined irrigation area is: , In the formula, r t represents the immediate reward in period t, Z t is the groundwater level in period t, Z red The exploitable water level of groundwater. are quantification coefficients respectively, and T represents the total number of time periods in the future water distribution cycle.

4. The method for optimizing water distribution in irrigation areas by combining wells and canals based on deep reinforcement learning according to claim 1, characterized in that: The state transfer function of the real-time water distribution model of the well-channel combined irrigation area is: , In the formula, s t+1 Indicates the state of the t+1 period, Indicates that from state s t Execute an action a t Then transfer to state s t+1 , They represent the rainfall, irrigation water demand, total irrigation water demand during the water distribution cycle, canal water supply, and pumped well water supply in the well-canal combined irrigation sub-area during the t+1 period; They represent the canal water diversion and pumped water volume of the well-canal combined irrigation sub-area in period t, represents the canal water utilization coefficient, represents the maximum flow rate of the channel, Indicates the time interval. represents the rated water extraction flow of well j, They represent the recharge of groundwater by channel water diversion, pumped water from wells, and rainfall, respectively. is the groundwater exploitability coefficient, They represent the recharge coefficients of channel water diversion, pumped water from wells, and rainfall on groundwater. is the field water utilization coefficient, Area is the area of ​​the well-canal combined irrigation sub-area, and J is the number of wells in the well-canal combined irrigation sub-area.

5. The method for optimizing water distribution in irrigation areas by combining wells and canals based on deep reinforcement learning according to claim 1, characterized in that: The preset irrigation constraints in step (4) include: A. Channel water diversion constraints: ; B. Constraints on pumping water from pumped wells: ; C. Groundwater level fluctuation constraints: ; D. Water demand constraints: ; In the formula, represents the canal water diversion volume of the well-canal combined irrigation sub-area in period t, They represent the canal water supply and pumped well water supply of the well-canal combined irrigation sub-area in period t, They represent the minimum and maximum flow rates of the channel respectively. is the water extraction volume of well j in period t, represents the rated water extraction flow of well j, represents the canal water utilization coefficient, are the maximum and minimum water levels of well j in period t, is the water level of well j in period t, t represents time, and J is the number of wells in the well-channel combined irrigation sub-area.

6. The method for optimizing water distribution in well-canal combined irrigation areas based on deep reinforcement learning according to claim 5 is characterized in that: The penalty of the real-time water distribution model of the well-canal combined irrigation area is: , In the formula, c t represents the penalty in time period t, c1, c2, c3, c4, c5 represent the penalty values ​​of type 1, 2, 3, 4, 5 respectively, are the conversion coefficients for the 1st, 2nd, 3rd, 4th, and 5th type penalties, Indicates the time interval.

7. A well-canal combined irrigation area water distribution optimization system based on deep reinforcement learning, characterized in that: The system is implemented by the well-canal combined irrigation area water distribution optimization method based on deep reinforcement learning as described in claim 1, including: Data acquisition module, used to obtain geographic satellite data, well information, channel information, soil moisture forecast data, and meteorological forecast data of the target area; The regional division module is used to combine the geographic satellite data of the target area and divide the target area into several well irrigation sub-areas, several canal irrigation sub-areas and several well-canal combined irrigation sub-areas according to the well information and channel information; The water demand forecasting module is used to combine soil moisture forecast data and meteorological forecast data to forecast the irrigation water demand of each sub-area in each period of the future water distribution cycle; A model building and training module is used to build and train a real-time water distribution model for well-canal combined irrigation areas based on deep reinforcement learning, wherein the real-time water distribution model for well-canal combined irrigation areas takes the influencing factors that affect the real-time water distribution of the well-canal combined irrigation sub-area as the state, takes the start and stop of each well in the well-canal combined irrigation sub-area as the action, takes the quantification of the difference between the groundwater level and the exploitable water level after the action is executed as the immediate reward, takes the quantification of the influencing factors that exceed the preset irrigation constraints after the action is executed as the penalty, and outputs the set of actions that maximize the cumulative reward in the future water distribution cycle as the optimal strategy; The water distribution optimization module is used to calculate the water distribution amount for each well irrigation sub-area and canal irrigation sub-area based on the irrigation water demand in each period in the future water distribution cycle. For each well-canal combined irrigation sub-area, the optimal strategy is obtained based on the trained well-canal combined irrigation area real-time water distribution model, and the water distribution amount is calculated based on the optimal strategy. Finally, the water distribution amount of all sub-areas is integrated as the optimal water distribution plan.

Images