Method and system for constructing water consumption knowledge graph for irrigation area

Abstract

The invention discloses a construction method and system of a water consumption knowledge graph for an irrigation district, and relates to the technical field of irrigation district water management. A first manual scheduling machine learning model is trained based on a water consumption characteristic data set and scheduling flow data of the irrigation district, and a critical value of scheduling flow is determined based on the first manual scheduling machine learning model; determining target scheduling traffic data according to the critical value of the scheduling traffic; constructing a second manual scheduling machine learning model based on the target scheduling flow data; constructing an irrigation area flow scheduling scene template; scheduling data of manual historical experience is learned through a machine learning model, the manual scheduling experience is knowledge-based, meanwhile, the manual scheduling experience is optimized, target flow data are obtained through the machine learning model according to the optimized manual scheduling experience, and an irrigation area water flow knowledge system with the scheduling experience as the main body is formed.

Description

Technical Field

[0001] This invention relates to the field of irrigation district water management technology, specifically a method and system for constructing a knowledge graph of water consumption in irrigation districts. Background Technology

[0002] Irrigation water use refers to water resources used to meet the needs of crop growth. The rational utilization and management of irrigation water use is of great significance for ensuring agricultural production, promoting the sustainable use of water resources, and protecting the ecological environment. For irrigation districts that rely on the manual experience of dispatchers for water management, a decision-making network based on manual experience has been formed. However, in the process of irrigation water use, years of manual experience may not be reasonable decisions in themselves. Since the error between the flow dispatch value obtained by the decision-making network based on manual experience and the reasonable flow dispatch value does not exceed the critical value, the defects of historical experience itself are difficult to be discovered. Therefore, how to improve the effectiveness of irrigation water use decision-making has become an urgent problem to be solved. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for constructing a knowledge graph of water consumption in irrigation areas, so as to solve the problems raised in the prior art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a system for constructing a knowledge graph of water consumption in an irrigation district, comprising an experience analysis module, a data acquisition module, a data storage module, and a graph database module; the data acquisition module is used to acquire water consumption characteristic datasets and scheduled flow data of the irrigation district; the data storage module is used to store historical water consumption characteristic datasets and scheduled flow data of the irrigation district; the experience analysis module trains a first artificial scheduling machine learning model based on the water consumption characteristic datasets and scheduled flow data of the irrigation district, the first artificial scheduling machine learning model is used to determine the critical value of the scheduled flow, and the target scheduled flow data is determined based on the critical value of the scheduled flow; a second artificial scheduling machine learning model is constructed based on the target scheduled flow data; the graph database module is used to construct the storage structure of the knowledge graph.

[0005] The graph database module further includes a matching unit, an input unit, and an output unit; the matching unit is used to match the dispatcher's scenario data with the irrigation district flow dispatch scenario template; the input unit is used to input the dispatcher's scenario information; and the output unit is used to provide the dispatcher with the search results.

[0006] The experience analysis module further includes a water use feature classification unit, a machine learning unit, a target flow calculation unit, and a critical value analysis unit. The water use feature classification unit is used to obtain different types of water use feature data. The machine learning unit trains a first artificial machine learning model based on the water feature data and the scheduled flow data. The machine learning unit trains a second artificial machine learning model based on the water feature data and the target scheduled flow data. The target flow value calculation unit is used to determine the target flow data corresponding to the standard scheduled flow data. The critical value analysis unit is used to determine the critical value of the scheduled flow.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for constructing a knowledge graph of water consumption in irrigation areas, specifically including the following steps S1 to S5:

[0008] S1, acquire the water use characteristic dataset and the scheduled flow data of the irrigation area, wherein the water use characteristics are the factors affecting water use in the irrigation area; and the scheduled flow data is the water flow manually allocated to the irrigation area.

[0009] S2, classify the water use feature dataset according to the water use scenario to obtain different types of water use feature data;

[0010] S3. Train the first artificial scheduling machine learning model based on the water use characteristic dataset and scheduling flow data of the irrigation area. Determine the critical value of the scheduling flow based on the first artificial scheduling machine learning model. Determine the target scheduling flow data based on the critical value of the scheduling flow.

[0011] S4, Construct a second manual scheduling machine learning model based on target scheduling traffic data;

[0012] S5 constructs an irrigation district flow scheduling scenario template based on scheduling flow data, different types of water use characteristic data, and target scheduling flow data corresponding to water use characteristic data.

[0013] Specifically, in step S2, different types of water use characteristic data are obtained through the DBSCAN algorithm, including the following steps:

[0014] The input vector is obtained based on water use characteristic data of the same type, and the neighborhood radius rps and minimum number of points mpts are set for the DBSACN algorithm; different combinations of neighborhood radius and minimum number of points are tested using cross-validation.

[0015] Step 1: Select an input vector as the starting point;

[0016] Step 2: Using the selected input vector as the center, find other input vectors within the neighborhood radius;

[0017] Step 3: If the number of other input vectors found within the neighborhood radius is not less than the minimum number of points (mpts), the selected input vector is marked as a core point, and a cluster is formed with this core point; if the number of other input vectors found within the neighborhood radius (rps) is less than the minimum number of points (mpts), this input vector is marked as a noise point.

[0018] Step 4: For the core point, add all input vectors within the neighborhood radius rps centered on the core point to the cluster formed by the core point; for other input vectors added to the cluster of the core point, determine whether they are core points according to Step 3. If they are core points, continue to expand the cluster; if they are not core points, do not expand.

[0019] Repeat steps one through four until all input vectors have been accessed to obtain the final clusters; water use feature data in the same cluster are water use feature data of one type.

[0020] Specifically, in step S3, determining the target scheduling traffic data further includes the following steps:

[0021] Input feature data is formed from water use characteristic data and scheduled flow data of the same type. All input feature data are divided into training and testing sets. One input feature data is selected from the training set as a reference input feature data. The error [Lb-Li, Qb-Qi] between the other input feature data in the training set and the reference input feature data is calculated, where Li represents the i-th water use feature data vector in the training set, Q represents the i-th scheduled flow data in the training set, Lb represents the reference water use feature data vector, and Qb represents the reference scheduled flow data. The water use feature data vector is formed from the water use feature data. Using Lb-Li as input and Qb-Qi as output, a first artificial scheduling machine learning model is trained, and the first artificial scheduling machine learning model is validated using data from the testing set. After validation, all scheduled flow data of the same type are transformed onto the reference feature data, specifically including the following steps:

[0022] Let Lj and Qj represent the j-th water use feature data vector and the scheduled flow data of the same type. Input the error Lb-Lj into the first artificial scheduling machine learning model to obtain the output ΔQj. Add Qj to the output ΔQj to complete the transformation of the j-th scheduled flow data onto the reference feature data.

[0023] Specifically, determining the critical value of the scheduling traffic based on the first manual scheduling machine learning model further includes the following steps S10 to S40:

[0024] S10, based on the reference scheduling traffic data, obtain the interval [(1-k1)Qb, (1+k2)Qb] to which the target reference scheduling traffic data belongs, where k1 and k2 are positive constants; set the set W to [(1-k1)Qb, (1+k2)Qb];

[0025] S20, for j=1 to n, execute step S30, where n is the number of standard scheduling flow data of the same type; the standard scheduling flow data is the water flow manually allocated to the irrigation district and which does not produce obvious defects;

[0026] S30, based on the j-th standard scheduling traffic data, obtain the interval [(1-k1)Qj, (1+k2)Qj] to which the j-th reference scheduling traffic data belongs; take the union of set W and interval [(1-k1)Qj, (1+k2)Qj] to update the value of set W, W=W∪[(1-k1)Qj, (1+k2)Qj];

[0027] S40. Based on the upper and lower bounds of set U [Qdown, Qup], obtain the critical values ​​k3 and k4 for scheduling flow, k3 = 1 - Qdown / Qb, k4 = Qup / Qb - 1.

[0028] Specifically, determining the target scheduled traffic data based on the critical value of the scheduled traffic also includes the following steps:

[0029] Based on the critical value of the scheduled flow, the feasible interval [(1-k3)Qj, (1+k4)Qj] of the j-th standard scheduled flow data is obtained, and the quantile Fj of Qj is determined, Fj=Qj / [(1-k3)Qj+(1+k4)Qj]; the quantiles of all standard scheduled flow data of the same type are obtained;

[0030] Acquire comparative flow data and comparative water use characteristic data. The comparative flow data refers to the water use flow rate manually allocated to the irrigation district that results in obvious defects. The comparative water use characteristic data refers to the water use characteristic data corresponding to the comparative flow data. Obtain a comparative water use characteristic data vector based on the comparative water use characteristic data. Calculate the deviation between the reference water use characteristic data vector and the comparative water use characteristic data vector to obtain the comparative deviation. Input the comparative deviation into the first manual scheduling machine learning model to obtain the output Qout. Based on the manual adjustment results on the comparative flow data, obtain the deviation degree e of the comparative flow data, e=(Qt-Qc) / Qout, where Qt represents the manual adjustment result and Qc is the comparative flow data. Determine the deviation quantile DF of the comparative flow data based on the deviation degree, DF=e / [(1-k3)Qb+(1+k4)Qb]; Obtain the average value ADF of the deviation quantile DF based on all comparative flow data and comparative water use characteristic data.

[0031] Based on ADF, obtain the target traffic data Qjopt corresponding to the j-th standard scheduled traffic data, Qjopt=(Fj+ADF×ΔQj)×[(1-k3)Qj+(1+k4)Qj]; obtain the target traffic data corresponding to all standard scheduled traffic data;

[0032] The target flow data is used as the output, and the water usage characteristic data corresponding to the standard scheduling flow data is used as the input to train the second artificial scheduling machine learning model.

[0033] Specifically, constructing an irrigation district flow scheduling scenario template based on scheduled flow data, different types of water use characteristic data, and target scheduled flow data corresponding to the water use characteristic data also includes the following steps:

[0034] The scheduling flow data, different types of water use characteristic data, and water use characteristic data are stored as entities in the graph database Neo4j; a template for irrigation district flow scheduling scenario is constructed based on the graph database Neo4j;

[0035] Obtain the scenario data of the dispatchers, match the scenario data of the dispatchers with the irrigation district flow scheduling scenario template, and search the graph database Neo4j according to the irrigation district flow scheduling scenario template. The graph database Neo4j returns the search results.

[0036] Compared with the prior art, the beneficial effects of the present invention are: by learning from the scheduling data of human historical experience through machine learning models, the human scheduling experience is transformed into knowledge; at the same time, by optimizing the human scheduling experience, the target flow data is obtained by using machine learning models based on the optimized human scheduling experience, thus forming a knowledge system of irrigation area water flow based on scheduling experience. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the structure of a system for constructing a knowledge graph of water consumption in an irrigation area according to the present invention;

[0038] Figure 2 This is a flowchart illustrating a method for constructing a knowledge graph of water consumption in an irrigation area according to the present invention. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] Example: Figure 1As shown, a system for constructing a knowledge graph of water consumption in an irrigation district includes an experience analysis module, a data acquisition module, a data storage module, and a graph database module. The data acquisition module is used to acquire water consumption characteristic datasets and scheduled flow data of the irrigation district. The data storage module is used to store historical water consumption characteristic datasets and scheduled flow data of the irrigation district. The experience analysis module trains a first artificial scheduling machine learning model based on the water consumption characteristic datasets and scheduled flow data of the irrigation district. The first artificial scheduling machine learning model is used to determine the critical value of the scheduled flow, and the target scheduled flow data is determined based on the critical value of the scheduled flow. A second artificial scheduling machine learning model is constructed based on the target scheduled flow data. The graph database module is used to construct the storage structure of the knowledge graph.

[0041] The graph database module also includes a matching unit, an input unit, and an output unit. The matching unit matches the dispatcher's scenario data with the irrigation district flow scheduling scenario template. The input unit inputs the dispatcher's scenario information. The output unit provides feedback on the search results to the dispatcher. The experience analysis module also includes a water use feature classification unit, a machine learning unit, a target flow calculation unit, and a critical value analysis unit. The water use feature classification unit obtains different types of water use feature data. The machine learning unit trains a first artificial machine learning model based on the water feature data and the scheduled flow data. The machine learning unit trains a second artificial machine learning model based on the water feature data and the target scheduled flow data. The target flow value calculation unit determines the target flow data corresponding to the standard scheduled flow data. The critical value analysis unit determines the critical value of the scheduled flow.

[0042] In another embodiment of the present invention, the present invention provides a method for constructing a knowledge graph of water consumption in irrigation areas, comprising the following steps:

[0043] S1, acquire the water use characteristic dataset and the scheduled flow data of the irrigation area, wherein the water use characteristics are the factors affecting water use in the irrigation area; and the scheduled flow data is the water flow manually allocated to the irrigation area.

[0044] S2, classify the water use feature dataset according to the water use scenario to obtain different types of water use feature data;

[0045] S3. Train the first artificial scheduling machine learning model based on the water use characteristic dataset and scheduling flow data of the irrigation area. Determine the critical value of the scheduling flow based on the first artificial scheduling machine learning model. Determine the target scheduling flow data based on the critical value of the scheduling flow.

[0046] S4, Construct a second manual scheduling machine learning model based on target scheduling traffic data;

[0047] S5 constructs an irrigation district flow scheduling scenario template based on scheduling flow data, different types of water use characteristic data, and target scheduling flow data corresponding to water use characteristic data.

[0048] Factors affecting irrigation water use include, but are not limited to, temperature, humidity, wind speed, crop growth stage, irrigation period, and rainfall. Temperature, humidity, and wind speed affect soil moisture by influencing evaporation, while rainfall directly affects soil moisture. The crop growth stage and irrigation period determine water demand. Scenarios are categorized based on whether urban and ecological water replenishment is considered, and whether dispatch telegrams are received.

[0049] The DBSCAN algorithm is used to obtain water use characteristic data of different types, including the following steps:

[0050] The input vector is obtained based on water use characteristic data of the same type, and the neighborhood radius rps and minimum number of points mpts are set for the DBSACN algorithm; different combinations of neighborhood radius and minimum number of points are tested using cross-validation.

[0051] Step 1: Select an input vector as the starting point;

[0052] Step 2: Using the selected input vector as the center, find other input vectors within the neighborhood radius;

[0053] Step 3: If the number of other input vectors found within the neighborhood radius is not less than the minimum number of points (mpts), the selected input vector is marked as a core point, and a cluster is formed with this core point; if the number of other input vectors found within the neighborhood radius (rps) is less than the minimum number of points (mpts), this input vector is marked as a noise point.

[0054] Step 4: For the core point, add all input vectors within the neighborhood radius rps centered on the core point to the cluster formed by the core point; for other input vectors added to the cluster of the core point, determine whether they are core points according to Step 3. If they are core points, continue to expand the cluster; if they are not core points, do not expand.

[0055] Repeat steps one through four until all input vectors have been accessed to obtain the final clusters; water use feature data in the same cluster are water use feature data of one type.

[0056] Determining the target scheduling traffic data also includes the following steps:

[0057] Input feature data is formed from water use characteristic data and scheduled flow data of the same type. All input feature data are divided into training and testing sets. One input feature data is selected from the training set as a reference input feature data. The error [Lb-Li, Qb-Qi] between the other input feature data in the training set and the reference input feature data is calculated, where Li represents the i-th water use feature data vector in the training set, Q represents the i-th scheduled flow data in the training set, Lb represents the reference water use feature data vector, and Qb represents the reference scheduled flow data. The water use feature data vector is formed from the water use feature data. Using Lb-Li as input and Qb-Qi as output, a first artificial scheduling machine learning model is trained, and the first artificial scheduling machine learning model is validated using data from the testing set. After validation, all scheduled flow data of the same type are transformed onto the reference feature data, specifically including the following steps:

[0058] Let Lj and Qj represent the j-th water use feature data vector and the scheduled flow data of the same type. Input the error Lb-Lj into the first artificial scheduling machine learning model to obtain the output ΔQj. Add Qj to the output ΔQj to complete the transformation of the j-th scheduled flow data onto the reference feature data.

[0059] When the error between the flow scheduling value derived from a decision-making network based on historical experience and the reasonable value does not exceed a critical threshold, its defects are indeed easily masked. When the error of experience-based decisions is within a critical range, managers are likely to form a judgment based on experience as valid. As long as there are no significant problems with crop yield or water allocation, they will not actively question the rationality of the experience. For example, an irrigation district has long allocated flow based on experience, consuming 10% more water resources than the optimal value. However, because this does not exceed the allowable loss threshold, the excess water consumption goes unnoticed, and since the crops are growing normally, the experience is assumed to be sound, ignoring the hidden water-saving potential. The decision-making logic based on human experience is often based solely on whether a problem occurs, making it difficult to discover defects in the experience. It can yield feasible but not entirely reasonable decision-making logic.

[0060] The steps involved in determining the critical value of scheduling traffic based on the first manual scheduling machine learning model are as follows:

[0061] S10, based on the reference scheduling traffic data, obtain the interval [(1-k1)Qb, (1+k2)Qb] to which the target reference scheduling traffic data belongs, where k1 and k2 are positive constants; set the set W to [(1-k1)Qb, (1+k2)Qb];

[0062] S20, for j=1 to n, execute step S30, where n is the number of standard scheduling flow data of the same type; the standard scheduling flow data is the water flow manually allocated to the irrigation district and which does not produce obvious defects;

[0063] S30, based on the j-th standard scheduling traffic data, obtain the interval [(1-k1)Qj, (1+k2)Qj] to which the j-th reference scheduling traffic data belongs; take the union of set W and interval [(1-k1)Qj, (1+k2)Qj] to update the value of set W, W=W∪[(1-k1)Qj, (1+k2)Qj];

[0064] S40. Based on the upper and lower bounds of set U [Qdown, Qup], obtain the critical values ​​k3 and k4 for scheduling flow, k3 = 1 - Qdown / Qb, k4 = Qup / Qb - 1.

[0065] First, since standard scheduling flow data does not produce explicit defects, standard scheduling flow data based on human experience are all feasible decision-making logics; explicit defects refer to defects that humans can identify as having problems with flow scheduling, such as insufficient water resources for crops leading to poor crop growth, or excessive water resource scheduling affecting surrounding cities and ecological water use, etc.

[0066] The first manual scheduling machine learning model is trained based on scheduling flow data and water use characteristic data based on human experience. The input is the difference between water use characteristic data and the output is the difference between scheduling flow data. The machine learning model is used to establish the relationship between these two differences. For example, if the rainfall difference is 'a', the scheduling flow data difference is 'b', capturing the mapping relationship between 'a' and 'b'.

[0067] Transform the standard scheduling flow data under different water use characteristics into standard scheduling flow data under reference characteristics. Since none of the standard scheduling flow data produced any obvious defects, the standard scheduling flow data based on these human experience are all feasible decision-making logics.

[0068] Since the standard scheduled traffic data does not produce any explicit defects, it falls within a certain range of the target scheduled traffic data. The target scheduled traffic data is the most reasonable scheduled traffic data, and conversely, it falls within a certain range of the standard scheduled traffic data. Since none of the standard scheduled traffic data produces any explicit defects, the target scheduled traffic data also falls within a certain range of all the standard scheduled traffic data. This range is represented by positive constants k1 and k2, which can be set to smaller values. The interval W is obtained by taking the union of consecutive values. This interval W is the feasible range of the target scheduled traffic data, and all the standard scheduled traffic data falls within this interval W. Therefore, no explicit defects have occurred, which is consistent with the actual situation.

[0069] Determining the target scheduled traffic data based on the critical value of the scheduled traffic also includes the following steps:

[0070] Based on the critical value of the scheduled flow, the feasible interval [(1-k3)Qj, (1+k4)Qj] of the j-th standard scheduled flow data is obtained, and the quantile Fj of Qj is determined, Fj=Qj / [(1-k3)Qj+(1+k4)Qj]; the quantiles of all standard scheduled flow data of the same type are obtained;

[0071] Acquire comparative flow data and comparative water use characteristic data. The comparative flow data refers to the water use flow rate manually allocated to the irrigation district that results in obvious defects. The comparative water use characteristic data refers to the water use characteristic data corresponding to the comparative flow data. Obtain a comparative water use characteristic data vector based on the comparative water use characteristic data. Calculate the deviation between the reference water use characteristic data vector and the comparative water use characteristic data vector to obtain the comparative deviation. Input the comparative deviation into the first manual scheduling machine learning model to obtain the output Qout. Based on the manual adjustment results on the comparative flow data, obtain the deviation degree e of the comparative flow data, e=(Qt-Qc) / Qout, where Qt represents the manual adjustment result and Qc is the comparative flow data. Determine the deviation quantile DF of the comparative flow data based on the deviation degree, DF=e / [(1-k3)Qb+(1+k4)Qb]; Obtain the average value ADF of the deviation quantile DF based on all comparative flow data and comparative water use characteristic data.

[0072] Based on ADF, obtain the target traffic data Qjopt corresponding to the j-th standard scheduled traffic data, Qjopt=(Fj+ADF×ΔQj)×[(1-k3)Qj+(1+k4)Qj]; obtain the target traffic data corresponding to all standard scheduled traffic data;

[0073] The target flow data is used as the output, and the water usage characteristic data corresponding to the standard scheduling flow data is used as the input to train the second artificial scheduling machine learning model.

[0074] By comparing flow data, we can identify flaws in decision-making logic based on human experience. These flaws also exist in standard flow data, but they are not explicit and their impact is not obvious. Therefore, by comparing flow data, we can capture the flaws in human experience. The comparative flow data represents the water flow that produces the explicit flaws. Thus, when people discover problems in water flow allocation, they will reallocate the flow; this reallocated flow is Qt. Based on Qt and Qc, we determine the quantiles of the deviation and map these quantiles onto the standard flow data. We then correct the standard flow data according to the deviation quantiles to obtain the target flow data. To prevent the corrected target flow data from deviating from the feasible range, we can add a constraint to the target flow data: if the target flow data does not deviate from the feasible range, the target flow data remains unchanged; if the target flow data deviates from the feasible range, the target flow data is set to the endpoint of the nearest feasible range.

[0075] The second manual scheduling machine learning model is trained based on the corrected target traffic data. This improves upon human experience and is more reasonable than directly using human experience data output, thus improving the performance of the machine learning model. At the same time, constraints are imposed on the fitting target, i.e., the target traffic data, to ensure that the output of the second manual scheduling machine learning model is within a feasible range.

[0076] Constructing an irrigation district flow scheduling scenario template based on scheduled flow data, different types of water use characteristic data, and target scheduled flow data corresponding to water use characteristic data also includes the following steps:

[0077] The scheduling flow data, different types of water use characteristic data, and water use characteristic data are stored as entities in the graph database Neo4j; a template for irrigation district flow scheduling scenario is constructed based on the graph database Neo4j;

[0078] Obtain the scenario data of the dispatchers, match the scenario data of the dispatchers with the irrigation district flow scheduling scenario template, and search the graph database Neo4j according to the irrigation district flow scheduling scenario template. The graph database Neo4j returns the search results.

[0079] A template for irrigation district flow scheduling is built on the Neo4j graph database, which enables the knowledge-based management of irrigation district water use scheduling experience. This provides schedulers with a knowledge system that facilitates the judgment of scheduled flow. Schedulers can judge the scheduled flow based on the second manual scheduling machine learning model found in the search, and at the same time, improve existing historical manual experience based on the search results.

[0080] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for constructing a knowledge graph of water consumption in an irrigation district, characterized in that, Includes the following steps: S1, acquire the water use characteristic dataset and the scheduled flow data of the irrigation area, wherein the water use characteristics are the factors affecting water use in the irrigation area; and the scheduled flow data is the water flow manually allocated to the irrigation area. S2, classify the water use feature dataset according to the water use scenario to obtain different types of water use feature data; S3. Train the first artificial scheduling machine learning model based on the water use characteristic dataset and scheduling flow data of the irrigation area. Determine the critical value of the scheduling flow based on the first artificial scheduling machine learning model. Determine the target scheduling flow data based on the critical value of the scheduling flow. S4, Construct a second manual scheduling machine learning model based on target scheduling traffic data; S5 constructs an irrigation district flow scheduling scenario template based on scheduling flow data, different types of water use characteristic data, and target scheduling flow data corresponding to water use characteristic data.

2. The method for constructing a knowledge graph of water consumption in an irrigation area according to claim 1, characterized in that, In step S2, water use characteristic data of different types are obtained through the DBSCAN algorithm, including the following steps: The input vector is obtained based on water use characteristic data of the same type, and the neighborhood radius rps and minimum number of points mpts are set for the DBSACN algorithm; different combinations of neighborhood radius and minimum number of points are tested using cross-validation. Step 1: Select an input vector as the starting point; Step 2: Using the selected input vector as the center, find other input vectors within the neighborhood radius; Step 3: If the number of other input vectors found within the neighborhood radius is not less than the minimum number of points (mpts), the selected input vector is marked as a core point, and a cluster is formed with this core point; if the number of other input vectors found within the neighborhood radius (rps) is less than the minimum number of points (mpts), this input vector is marked as a noise point. Step 4: For the core point, add all input vectors within the neighborhood radius rps centered on the core point to the cluster formed by the core point; for other input vectors added to the cluster of the core point, determine whether they are core points according to Step 3. If they are core points, continue to expand the cluster; if they are not core points, do not expand. Repeat steps one through four until all input vectors have been accessed to obtain the final clusters; water use feature data in the same cluster are water use feature data of one type.

3. The method for constructing a knowledge graph of water consumption in an irrigation area according to claim 2, characterized in that, In step S3, determining the target scheduling traffic data further includes the following steps: Input feature data is formed from water use characteristic data and scheduled flow data of the same type. All input feature data are divided into training and testing sets. One input feature data is selected from the training set as a reference input feature data. The error [Lb-Li, Qb-Qi] between the other input feature data in the training set and the reference input feature data is calculated, where Li represents the i-th water use feature data vector in the training set, Q represents the i-th scheduled flow data in the training set, Lb represents the reference water use feature data vector, and Qb represents the reference scheduled flow data. The water use feature data vector is formed from the water use feature data. Using Lb-Li as input and Qb-Qi as output, a first artificial scheduling machine learning model is trained, and the first artificial scheduling machine learning model is validated using data from the testing set. After validation, all scheduled flow data of the same type are transformed onto the reference feature data, specifically including the following steps: Let Lj and Qj represent the j-th water use feature data vector and the scheduled flow data of the same type. Input the error Lb-Lj into the first artificial scheduling machine learning model to obtain the output ΔQj. Add Qj to the output ΔQj to complete the transformation of the j-th scheduled flow data onto the reference feature data.

4. The method for constructing a knowledge graph of water consumption in an irrigation area according to claim 3, characterized in that, The steps involved in determining the critical value of scheduling traffic based on the first manual scheduling machine learning model are as follows: S10, based on the reference scheduling traffic data, obtain the interval [(1-k1)Qb, (1+k2)Qb] to which the target reference scheduling traffic data belongs, where k1 and k2 are positive constants; set the set W to [(1-k1)Qb, (1+k2)Qb]; S20, for j=1 to n, execute step S30, where n is the number of standard scheduling flow data of the same type; the standard scheduling flow data is the water flow manually allocated to the irrigation district and which does not produce obvious defects; S30, based on the j-th standard scheduling traffic data, obtain the interval [(1-k1)Qj, (1+k2)Qj] to which the j-th reference scheduling traffic data belongs; take the union of set W and interval [(1-k1)Qj, (1+k2)Qj] to update the value of set W, W=W∪[(1-k1)Qj, (1+k2)Qj]; S40. Based on the upper and lower bounds of set U [Qdown, Qup], obtain the critical values ​​k3 and k4 for scheduling flow, k3 = 1 - Qdown / Qb, k4 = Qup / Qb - 1.

5. The method for constructing a knowledge graph of water consumption in an irrigation area according to claim 4, characterized in that, The process of determining the target scheduling traffic data based on the critical value of the scheduling traffic also includes the following steps: Based on the critical value of the scheduled flow, the feasible interval [(1-k3)Qj, (1+k4)Qj] of the j-th standard scheduled flow data is obtained, and the quantile Fj of Qj is determined, Fj=Qj / [(1-k3)Qj+(1+k4)Qj]; the quantiles of all standard scheduled flow data of the same type are obtained; Acquire comparative flow data and comparative water use characteristic data. The comparative flow data refers to the water use flow rate manually allocated to the irrigation district that results in obvious defects. The comparative water use characteristic data refers to the water use characteristic data corresponding to the comparative flow data. Obtain a comparative water use characteristic data vector based on the comparative water use characteristic data. Calculate the deviation between the reference water use characteristic data vector and the comparative water use characteristic data vector to obtain the comparative deviation. Input the comparative deviation into the first manual scheduling machine learning model to obtain the output Qout. Based on the manual adjustment results on the comparative flow data, obtain the deviation degree e of the comparative flow data, e=(Qt-Qc) / Qout, where Qt represents the manual adjustment result and Qc is the comparative flow data. Determine the deviation quantile DF of the comparative flow data based on the deviation degree, DF=e / [(1-k3)Qb+(1+k4)Qb]; Obtain the average value ADF of the deviation quantile DF based on all comparative flow data and comparative water use characteristic data. Based on ADF, obtain the target traffic data Qjopt corresponding to the j-th standard scheduled traffic data, Qjopt=(Fj+ADF×ΔQj)×[(1-k3)Qj+(1+k4)Qj]; obtain the target traffic data corresponding to all standard scheduled traffic data; The target flow data is used as the output, and the water usage characteristic data corresponding to the standard scheduling flow data is used as the input to train the second artificial scheduling machine learning model.

6. The method for constructing a knowledge graph of water consumption in an irrigation area according to claim 5, characterized in that, The process of constructing an irrigation district flow scheduling scenario template based on scheduled flow data, different types of water use characteristic data, and target scheduled flow data corresponding to water use characteristic data also includes the following steps: The scheduling flow data, different types of water use characteristic data, and water use characteristic data are stored as entities in the graph database Neo4j; a template for irrigation district flow scheduling scenario is constructed based on the graph database Neo4j; Obtain the scenario data of the dispatchers, match the scenario data of the dispatchers with the irrigation district flow scheduling scenario template, and search the graph database Neo4j according to the irrigation district flow scheduling scenario template. The graph database Neo4j returns the search results.

7. A system for constructing a knowledge graph of water consumption in an irrigation district, characterized in that, It includes an experience analysis module, a data acquisition module, a data storage module, and a graph database module; the data acquisition module is used to acquire water use characteristic datasets and scheduled flow data of the irrigation area; the data storage module is used to store historical water use characteristic datasets and scheduled flow data of the irrigation area; the experience analysis module trains a first artificial scheduling machine learning model based on the water use characteristic datasets and scheduled flow data of the irrigation area, and the first artificial scheduling machine learning model is used to determine the critical value of the scheduled flow, and determine the target scheduled flow data based on the critical value of the scheduled flow. A second manual scheduling machine learning model is constructed based on target scheduling traffic data; the graph database module is used to construct the storage structure of the knowledge graph.

8. The system for constructing a knowledge graph of water consumption in an irrigation area according to claim 7, characterized in that, The graph database module further includes a matching unit, an input unit, and an output unit; the matching unit is used to match the dispatcher's scenario data with the irrigation district flow dispatch scenario template; the input unit is used to input the dispatcher's scenario information; and the output unit is used to provide the dispatcher with the search results.

9. The system for constructing a knowledge graph of water consumption in an irrigation area according to claim 8, characterized in that, The experience analysis module further includes a water use feature classification unit, a machine learning unit, a target flow calculation unit, and a critical value analysis unit. The water use feature classification unit is used to obtain different types of water use feature data. The machine learning unit trains a first artificial machine learning model based on the water feature data and the scheduled flow data. The machine learning unit trains a second artificial machine learning model based on the water feature data and the target scheduled flow data. The target flow value calculation unit is used to determine the target flow data corresponding to the standard scheduled flow data. The critical value analysis unit is used to determine the critical value of the scheduled flow.

Images