Method for constructing a knowledge graph model for water resource allocation in shale gas exploitation

Abstract

A method for constructing a knowledge graph model for water resource allocation in shale gas exploitation, comprising the following steps: 1) The engineering data of water resource allocation in shale gas exploitation comes from Chinese databases, patent databases and research reports, and the text data is preprocessed; 2) Obtain knowledge nodes and relationship edges; 3) According to the defined knowledge nodes and relationship edges, list the schematic diagram of the knowledge graph model, and expand the knowledge graph semantic network diagram to several knowledge triples according to the number of knowledge triples; 4) According to the knowledge graph semantic network, relevant knowledge reasoning projects can be carried out, such as predicting the next entity from one entity and finding the relationship between them, thereby constructing a knowledge graph intelligent decision-making model.

Description

Technical Field

[0001] This application belongs to the technical field of shale gas extraction and its affiliated technical fields. Specifically, it relates to an industrial waste liquid recycling technology, and particularly to a method for constructing a knowledge graph model for water resource allocation in shale gas extraction. This invention is a divisional application of the invention patent with the application number 2021103519528. Background Art

[0002] Shale gas is an unconventional natural gas that exists in the form of free state, adsorbed state or dissolved state in dark-colored organic-rich and extremely low porosity and permeability shale and muddy siltstone. It has the characteristics of self-generation and self-storage, and continuous accumulation. With its large resource volume and low carbon emissions, it has become a new type of energy that has attracted much attention. In today's context where the country calls for energy conservation, emission reduction, and low-carbon development, it shows great advantages. Shale gas may become the most reliable energy replacement type in the future of China. With the continuous development of shale gas extraction, the impact on the regional water resource system and its ecological system during the extraction process has become a key concern in the industry.

[0003] The main methods for shale gas extraction are hydraulic fracturing and horizontal wells, etc. These extraction technologies are widely used in the current shale gas extraction process. However, there are problems such as the consumption of fresh water resources, the pressure on regional water resource supply, and the pollution of surface water and groundwater. This is mainly because the water used in current shale gas extraction is mainly fresh water resources such as the surrounding areas and groundwater. The hydraulic fracturing process takes several weeks, during which a large amount of local surface water or groundwater resources are consumed, which is very likely to affect local production, living, and ecological water use. In addition, a large amount of waste will be generated during the shale gas extraction process, leading to ecological and environmental problems such as soil pollution, water pollution, and heavy metal accumulation. It can be seen that during the shale gas extraction process, there are problems of water use contradictions that require a large amount of water resources and waste liquid treatment problems that generate a large amount of waste liquid. How to reasonably allocate and dispatch water resources and how to recycle the waste liquid generated during the extraction process are urgent problems to be solved in the shale gas development process.

[0004] In view of the problems in the existing shale gas extraction process, there is an urgent need to form a reasonable, efficient, and intelligent intelligent system for water resource recycling in shale gas extraction. Summary of the Invention

[0005] The present invention mainly aims at the technical problems in shale gas extraction that a large amount of water resources are consumed and a large amount of waste liquid is generated, and it is difficult to stably, reliably, and sustainably supply a large amount of water resources and it is difficult to treat a large amount of waste liquid. It provides an intelligent system for cascaded water cycle utilization in shale gas extraction that can reasonably allocate and dispatch water resources during extraction and can effectively and reasonably treat the waste liquid generated during extraction.

[0006] An intelligent system for cascaded water cycle utilization in shale gas extraction, which includes a cascaded water cycle utilization system. The data output end of the cascaded water cycle utilization system is connected to the data input end of the information system. The data output end of the information system is connected to the data input end of the decision-making system. The data output end of the decision-making system is connected to the regulator;

[0007] The cascaded water cycle utilization system includes at least one water volume judgment. When the water volume does not exceed the water safety threshold, the current shale gas extraction intensity is continued. When the water volume exceeds the water safety threshold, the monitoring data is transmitted to the information system;

[0008] The information system is also used to collect the gas production volume and water shortage value at the current stage of shale gas extraction, and transmit the collected data to the decision-making system. Based on the data collected by the information system, the best gas production volume, the required water volume of various water sources, and the water safety threshold in the next stage are obtained in the decision-making system. The decision-making system is used to predict the best gas production volume in the next stage;

[0009] The regulator is used to regulate the extraction intensity of shale gas at the current stage.

[0010] Before the first water volume judgment, the shale gas extraction waste liquid is treated by water treatment and then mixed with the rainwater treated in the rainwater treatment pool to be used as the type I water source for the water cycle utilization in shale gas extraction. The recycled water volume of this type of water can be predicted by a water volume predictor. After determining the recyclable water volume, the first judgment on whether the water quality meets the standard is carried out. If the water quality does not meet the standard, the mixed liquid is treated by water treatment again, and this cycle continues until the water quality meets the standard. After the water quality meets the standard, the first judgment on whether the water volume is sufficient is carried out.

[0011] If the first water volume is sufficient, the judgment on whether the first water volume is fully utilized is carried out. If the water volume can be fully utilized, this type of water is used for shale gas extraction. If there is surplus water after utilization, the excess water is reinjected into the groundwater. If the water volume is too small and insufficient, other water sources need to be introduced.

[0012] If the water volume is too small and insufficient, the type II water source needs to be introduced. The type II water source includes purchased water. The insufficient type I water source and the type II water source are mixed, and then the second judgment on whether the water volume is sufficient is carried out. If the water volume is sufficient, the judgment on whether the second water volume is fully utilized is carried out. If the water volume can be fully utilized, this type of water is used for shale gas extraction. If there is surplus water after utilization, the excess water is reinjected into the groundwater. If the water volume is too small and insufficient, other water sources need to be introduced.

[0013] If the water volume is too small and insufficient, it is necessary to introduce Class III water sources. The insufficient Class I and Class II water sources need to be mixed with Class III water sources, and then a third judgment on whether the water volume is sufficient is carried out. Class III water sources come from surface water and / or groundwater, and the available water volume is predicted by a water volume predictor. If the water volume is sufficient, a third judgment on whether the water volume is fully utilized is carried out. If the water volume can be fully utilized, this type of water is used for shale gas extraction. If there is still surplus water after utilization, the excess water is re-injected into the groundwater. If the water volume is too small and insufficient, a water safety threshold judgment is carried out.

[0014] After mixing Class I, Class II, and Class III water sources, a judgment is made on whether the water volume exceeds the safety threshold. If the water volume does not exceed the safety threshold, the current shale gas extraction intensity is continued; if the water volume exceeds the safety threshold, during the current shale gas extraction process, the current gas production volume, the proportion of Class I water sources, the proportion of Class II water sources, and the water shortage value are collected through an information system, and these data are input into the information system and the decision-making system.

[0015] The decision-making system is an intelligent decision-making system for literature mining based on a knowledge graph, which is used to predict the gas production volume in the next stage of shale gas extraction and the optimal allocation of water requirements for various water sources.

[0016] The present invention also includes a method for constructing a knowledge graph model for water resource allocation in shale gas extraction, including the following steps:

[0017] 1) The engineering data for water resource allocation in shale gas extraction comes from Chinese databases, patent databases, and research reports, which are characterized by wide sources and poor structure. Therefore, after preprocessing the text data, knowledge nodes and relationship edges can be constructed through knowledge graph construction technology, thereby constructing a knowledge graph intelligent decision-making model, which aims to discover the relationships between different knowledge nodes from the text data.

[0018] 2) From the perspective of water resource allocation, the knowledge nodes of the knowledge graph model can be selected as the project name (a certain shale gas extraction area), the allocation time, and the water resource categories (Class I, Class II, and Class III water); at the same time, the water requirements of a certain type of water and the shale gas extraction intensity can also be used as knowledge nodes; finally, "time is", "water requirement is", and "contains" can be used as the relationship edges between the nodes. The knowledge graph intelligent decision-making model constructed in this way can realize the decision-making function of water requirements, etc.

[0019] 3) According to the defined knowledge nodes and relationship edges, a schematic diagram of the knowledge graph model can be listed. According to the number of knowledge triples, this knowledge graph semantic network diagram can be extended to thousands or even millions of knowledge triples.

[0020] 4) Corresponding knowledge reasoning engineering can be carried out according to the semantic network of the knowledge graph, inferring the next entity from one entity and finding the relationship between them, etc., thereby constructing a knowledge graph intelligent decision-making model.

[0021] When carrying out knowledge reasoning, the following steps are adopted:

[0022] Define a knowledge triple, where the knowledge head node is h, the relationship edge is r, and the knowledge tail node is t, thereby determining a knowledge triple (h, r, t);

[0023] Using the method of supervised random walk, first generate some path features. The path consists of a series of knowledge nodes and relationship edges, which is:

[0024]

[0025] In the formula, T n is the scope of the relationship edge r n and the domain of the relationship edge r n-1 , that is, T n = range(r n ) = domain(r n-1 ). Thus, a distribution of relationship edges and knowledge nodes is defined, and the value obtained based on the distribution is the feature value X h，p(t) of each walking path. X h，p(t) can be understood as the probability of reaching the knowledge tail node t starting from the knowledge head node h along a certain path p. The update rule of X h，p(t) is:

[0026]

[0027] Among them, if e = S (the knowledge node in path p), then X h，p(e) = 1, otherwise X h，p(e) = 0. represents the probability of reaching the knowledge node e starting from the knowledge node e' along the relationship edge r1. r1(e', e) represents whether there is a path of relationship type r1 between the knowledge nodes e' and e. If it exists, its value is 1; if not, the value is 0. |r l (e′, ·)| represents the number of knowledge nodes that can be reached by the relationship edge of the specified path starting from the node e';

[0028] If you want to determine the relationship edge r between two certain knowledge nodes, a set of feature paths P r = (P1,..., P n ) needs to be obtained through supervised random walk. Supervised random walk is based on random walk and adds a supervised method to guide the walking knowledge nodes to walk, making the walk more purposeful;

[0029] Purposefully search for knowledge nodes and relationship edges related to water demand through supervised random walks. Then, use these characteristic paths to train a ranking model for predicting entities. This model can be modeled using a linear model method:

[0030]

[0031] In the formula, f(h, r i , t) represents the probability S i that there is a relationship r between knowledge node h and knowledge node t i , and θ p represents the weight factor of the corresponding characteristic path to P r . Through training, the value of θ p can be obtained.

[0032] Among them, y i = {0, 1} can be used to represent the value of a certain training sample. If it is 1, it means that the relationship edge r between the two knowledge nodes exists; if it is 0, it means that the relationship edge r does not exist. Usually, the sigmoid function can be used to map the prediction result to the interval [0, 1]. The specific form is as follows:

[0033]

[0034] For the weight factor θ p , the loss function can be designed through the following linear transformation plus maximum likelihood estimation:

[0035]

[0036]

[0037] Finally, the intelligent decision-making model based on the knowledge graph semantic web can be converted into an optimization objective function for the optimal weight factor θ p .

[0038] After the initial construction of the knowledge graph intelligent decision-making model, evaluate the knowledge graph intelligent decision-making model. If the evaluation result shows that the decision-making model can be used for decision-making, the output value such as water demand can be obtained through input from the information system; if the evaluation result shows that the decision-making model cannot be used for decision-making, then update the extraction of knowledge triples to update the intelligent decision-making model.

[0039] Among them, the input of the knowledge graph intelligent decision-making model is the prediction of the water demand of various water sources, and this prediction is the basis for supervised random walks. According to this prediction, search and decision-making can be carried out in the knowledge graph intelligent decision-making model to seek the most optimal water demand output result. The following lists a non-limiting example of the water demand prediction at the input end.

[0040] It is assumed that the water demands of a certain type of water source from time 1 to time t within a certain period are Q1, Q2... Q respectively t , and the water demand Qt+1 of this type of water source at time t + 1 can be predicted through a water demand prediction model, and the following prediction model can be adopted. t+1 For the prediction, the following prediction model can be used.

[0041] Change value of water demand at time t:

[0042] ΔQ t = Qt t - Qt - 1 t-1 (7)

[0043] Second-order difference change value of water demand at time t:

[0044] Δ 2 2Qt t = ΔQt t - ΔQt - 1 t-1 (8)

[0045] Predicted value of second-order difference change of water demand at time t:

[0046]

[0047] Therefore, the predicted value of water demand at time t + 1:

[0048]

[0049] Input the value, search through the already constructed knowledge graph intelligent decision-making model, and the corresponding water demand value can be output.

[0050] Compared with the prior art, the present invention has the following technical effects:

[0051] 1) In the process of shale gas extraction, fracturing fluid is usually injected for shale gas extraction, and a large amount of waste liquid is brought out by the produced backflow fluid. The present invention can effectively and efficiently utilize, plan, and treat the waste liquid, not only reasonably allocate and dispatch the water resources in the extraction process, but also realize the recycling of the waste liquid in the whole extraction process, and well solve the technical defects that the water resources cannot be stably and sustainably supplied and a large amount of waste liquid is difficult to treat and the treatment cost is high in the process of shale gas extraction;

[0052] 2) By introducing a decision-making system, and further, the decision-making system is an intelligent decision-making system for literature mining based on a knowledge graph, which can construct the relationships between various knowledge nodes regarding shale gas extraction industry. It has the characteristics of a large data scale, can visualize the knowledge of the shale gas extraction industry and enhance the coherence of knowledge, which is beneficial for staff management. Brief Description of the Drawings

[0053] The present invention will be further described below in conjunction with the drawings and embodiments:

[0054] Figure 1 is the system block diagram of the present invention;

[0055] Figure 2 is the schematic diagram of the allocation model of the optimization allocation system;

[0056] Figure 3 is the structural block diagram of the decision-making system;

[0057] Figure 4 is the partial schematic diagram of the semantic network of the knowledge graph. Detailed Embodiment

[0058] As Figure 1 shown, an intelligent system for cascaded water cycle utilization in shale gas extraction includes a cascaded water cycle utilization system 1. The data output end of the cascaded water cycle utilization system 1 is connected to the data input end of the information system 2. The data output end of the information system 2 is connected to the data input end of the decision-making system 3. The data output end of the decision-making system 3 is connected to the regulator 4;

[0059] The cascaded water cycle utilization system 1 includes at least one water volume judgment. When the water volume does not exceed the water safety threshold, the current shale gas extraction intensity is continued; when the water volume exceeds the water safety threshold, the monitoring data is transmitted to the information system 2;

[0060] The information system 2 is also used to collect the gas production volume and water shortage value at the current stage of shale gas extraction, and transmit the collected data to the decision-making system 3. Based on the data collected by the information system 2, the best gas production volume in the next stage, the required water volume of various water sources in the next stage, and the water safety threshold in the next stage are obtained in the decision-making system 3. The decision-making system 3 is used to predict the best gas production volume in the next stage;

[0061] The regulator 4 is used to regulate the current shale gas extraction intensity

[0062] Among them, the cascaded water cycle utilization system includes a primary water quality judgment, three judgments on whether the water volume is sufficient, a judgment on whether the water volume is fully used for shale gas extraction, and a judgment on whether the water volume exceeds the water safety threshold.

[0063] As Figure 1As shown, specifically, the flowback liquid waste from a certain shale gas extraction can be purified through the physical, chemical, and biological evolution process methods of traditional waste liquid. Meanwhile, in the shale gas extraction area, natural rainfall can be collected through a rainwater collection system, and the collected rainwater can be treated for water quality in a rainwater treatment pond. The shale gas extraction waste liquid is mixed with the rainwater treated in the rainwater treatment pond after water treatment as the Class I water source for the recycling of shale gas extraction water. The recovered water volume of this type of water can be predicted by a water volume predictor. After determining the recoverable water volume, a judgment is made on whether the water quality meets the standard for the first time. If the water quality does not meet the standard, the mixed liquid is treated again for water treatment; if the water quality meets the standard, a judgment is made on whether the water volume is sufficient for the first time.

[0064] Among them, the water volume predictor and the intelligent regulator are based on the industrial production standards of shale gas extraction to conduct water volume prediction and regulation. The water volume predictor and the intelligent regulator can select a programmable logic controller PLC, which is a digital operation and control electronic system designed specifically for application in industrial environments. It has functions such as sequential control, timing, and counting, and can control various machinery and production processes. Its model can be selected as Siemens PLC S7-200.

[0065] The Class I water source with qualified water quality needs to be judged on whether the water volume is sufficient for the first time. If the water volume is sufficient, a judgment is made on whether the water volume is fully utilized for the first time. If the water volume is fully utilized, this type of water is used for shale gas extraction. If the water volume is not fully utilized, the excess water is reinjected into the groundwater; if the water volume is not sufficient, Class II water source needs to be introduced.

[0066] The Class I water source with insufficient water volume needs to be mixed with the Class II water source and then judged on whether the water volume is sufficient for the second time. Among them, the Class II water source comes from purchased water, and the water volume of the purchased water is predicted by a purchasable water volume predictor. If the water volume is sufficient, a judgment is made on whether the water volume is fully utilized for the second time. If the water volume is fully utilized, this type of water is used for shale gas extraction. If the water volume is not fully utilized, the excess water is reinjected into the groundwater; if the water volume is not sufficient, Class III water source needs to be introduced.

[0067] The Class I and Class II water sources with insufficient water volume need to be mixed with the Class III water source and then judged on whether the water volume is sufficient for the third time. Among them, the Class III water source comes from surface water and groundwater, and the available water volume is predicted by an available water volume predictor. If the water volume is sufficient, a judgment is made on whether the water volume is fully utilized for the third time. If the water volume is fully utilized, this type of water is used for shale gas extraction. If the water volume is not fully utilized, the excess water is reinjected into the groundwater; if the water volume is not sufficient, a water safety threshold judgment needs to be made.

[0068] Among them, the respective extraction amounts of surface water and groundwater in the Class III water source are optimized and distributed through an optimization distribution system, such asFigure 2 As shown in the figure, the allocation objects of the optimization allocation system are the groundwater water consumption and the surface water water consumption. The strategy adopted for allocation is the available amount of surface water, the available amount of groundwater, and the water demand for shale gas extraction. Then, the parameters of the game payment function are determined according to the allocation strategy, and finally, the optimization allocation model is solved. If the solved model meets the allocation target, the optimization allocation system is introduced for the optimization allocation of Class III water sources; if the solved optimization allocation model does not meet the allocation target, the parameters are reset and solved again.

[0069] As Figure 1 shown in the figure, after the mixing of Class I, Class II, and Class III water sources, it is judged whether the water volume exceeds the safety threshold. If the water volume does not exceed the safety threshold, the current shale gas extraction intensity is continued; if the water volume exceeds the safety threshold, the current gas production volume, the proportion of Class I water sources, the proportion of Class II water sources, and the water shortage value during the current shale gas extraction process are collected through the information system, and these data are input into the decision-making system.

[0070] Among them, the information system can store data efficiently and securely, and needs to adapt to massive data scenarios such as the Internet of Things and big data. Here, Tencent Cloud Time Series Database - TencentDB for CTSDB can be selected.

[0071] Preferably, the decision-making system is a literature mining intelligent decision-making system based on a knowledge graph.

[0072] The literature mining intelligent decision-making system based on a knowledge graph constructs a large-scale semantic network to make the semantic network the carrier of big data, and then conducts knowledge reasoning through a knowledge reasoning algorithm to find the relationships between knowledge nodes, which is a characteristic that other intelligent algorithms do not have. The knowledge graph can make the intelligence of this decision-making system superior to that of traditional intelligent systems.

[0073] The literature mining intelligent decision-making system based on a knowledge graph can update the knowledge triple extraction method through the feedback in the industrial production process of shale gas extraction, thereby updating the knowledge graph intelligent decision-making model, which has the characteristics of real-time and variability, thus increasing its scope of application.

[0074] As Figure 1 shown in the figure, based on the data collected by the information system, the optimal gas production volume in the next stage, the water demand for various water sources in the next stage, and the water safety threshold in the next stage can be obtained in the literature mining intelligent decision-making system based on a knowledge graph.

[0075] Among them, the specific process of the literature mining intelligent decision-making system based on a knowledge graph is as Figure 3 shown in the figure.

[0076] Figure 3 The schematic diagram shows a non-limiting example of the literature mining intelligent decision-making system based on a knowledge graph according to the present invention.

[0077] The information input by the information system, including the current gas production volume, the proportion of various water sources, and the water shortage value, is input into the knowledge graph decision-making model that has been constructed in the decision-making system. By searching through the semantic network in the decision-making model, the water demand of type I water sources, the water demand of type II water sources, the water demand of type III water sources, the water safety threshold, and the gas production volume in the next stage can be output. After obtaining the output results, the first judgment is made. If the water demand obtained by the decision meets the current water shortage value, it is used for the allocation of various types of water in the next stage; if the water demand obtained by the decision does not meet the current water shortage value, the decision-making model is used again to determine the water demand of various types of water. Then, the second judgment is made. If the water volume is sufficient during the actual allocation process, the decision-making model of the previous stage is continued to be used; if the water volume is insufficient, knowledge triple extraction and update are carried out to update the decision-making model.

[0078] Among them, the literature for constructing the knowledge graph comes from Chinese journal databases, patent databases, and research reports.

[0079] Among them, the knowledge triple extraction method includes any n extraction methods. For the structured and unstructured data in the literature, multiple non-limiting examples can be listed.

[0080] Extraction method 1 can be used to extract knowledge triples related to the water demand of type I water sources; extraction method 2 can be used to extract knowledge triples related to the water demand of type II water sources; and so on. Extraction method n can be used to extract knowledge triples related to the relationship between water demand and water safety threshold.

[0081] Among them, the knowledge graph intelligent decision-making model includes any n prediction models, which are constructed from different knowledge extraction results to obtain the knowledge graph. Multiple non-limiting examples can be listed.

[0082] Extraction result 1 is used to construct the knowledge graph intelligent decision-making model 1, extraction result 2 is used to construct the knowledge graph intelligent decision-making model 2, and so on. Extraction result n is used to construct the knowledge graph intelligent decision-making model n.

[0083] Among them, when constructing the knowledge graph intelligent decision-making model, the following steps are adopted:

[0084] 1) The engineering data for shale gas exploitation water resource allocation comes from Chinese databases, patent databases, and research reports, which are characterized by wide sources and poor structure. Therefore, the text data can be preprocessed and then knowledge nodes and relationship edges can be constructed through knowledge graph construction technology to construct the knowledge graph intelligent decision-making model. This model aims to discover the relationships between different knowledge nodes from the text data;

[0085] 2) From the perspective of water resource allocation, the knowledge nodes of the knowledge graph model can be selected as the project name (a shale gas mining area), the allocation time, and the water resource categories (Type I, II, and III water); at the same time, the water demand of a certain type of water and the shale gas mining intensity can also be used as knowledge nodes; finally, "time is", "water demand is", and "contains" can be used as the relationship edges between the nodes. The knowledge graph intelligent decision-making model constructed in this way can realize the decision-making function of water demand and so on.

[0086] 3) According to the defined knowledge nodes and relationship edges, the schematic diagram of the knowledge graph model can be listed, and the partial schematic diagram is as Figure 4 shown. According to the number of knowledge triples, this knowledge graph semantic network diagram can be extended to thousands or even millions of knowledge triples.

[0087] 4) According to the knowledge graph semantic network, relevant knowledge reasoning projects can be carried out, such as predicting and inferring the next entity from one entity and finding the relationship between them, thereby constructing a knowledge graph intelligent decision-making model.

[0088] When carrying out knowledge reasoning, the reasoning algorithm Path Ranking Algorithm based on the graph structure of the knowledge graph can be adopted, which uses the relationship edges between knowledge nodes as features to perform link prediction reasoning.

[0089] Among them, define the knowledge triple, the knowledge head node is h, the relationship edge is r, and the knowledge tail node is t, so a knowledge triple (h, r, t) can be determined.

[0090] In the PRA algorithm, by using the way of supervised random walk, some path features are generated first. The path consists of a series of knowledge nodes and relationship edges, which is:

[0091]

[0092] In the formula, T n is the scope and range of the relationship edge r n , that is, T n-1 = range(r n ) = domain(r n ). This algorithm defines a distribution of relationship edges and knowledge nodes, and the value obtained based on the distribution is the feature value X n-1 of each walking path. X h，p(t) can be understood as the probability of reaching the knowledge tail node t starting from the knowledge head node h along a certain path p. The update rule of X h，p(t) is: h，p(t)

[0093]

[0094] Among them, if e = S (knowledge nodes in path p), then X h，p(e) = 1, otherwise X h，p(e) = 0. represents the probability of reaching knowledge node e along the relational edge r1 starting from knowledge node e'. r1(e', e) indicates whether there is a path of relational type r1 between knowledge nodes e' and e. If it exists, its value is 1; if not, the value is 0. |r l (e′, ·)| represents the number of knowledge nodes that can be reached by the relational edges of the specified path starting from node e'.

[0095] If you want to determine the relational edge r between two certain knowledge nodes, a set of characteristic paths P r = (P1,..., P n ) needs to be obtained through supervised random walk. Supervised random walk is based on random walk and adds a supervised method to guide the wandering knowledge nodes to wander, making the wandering more purposeful. In this example, the knowledge nodes and relational edges related to water demand can be purposefully searched through the method of supervised random walk. Then, a ranking model for predicting entities is trained using these characteristic paths. This model can be modeled using the method of a linear model:

[0096]

[0097] In the formula, f(h, r i , t) represents the likelihood S i that there is a relationship r between knowledge node h and knowledge node t i , and θ p represents the weight factor corresponding to P r . Through training, the value of θ p can be obtained.

[0098] Among them, y i = {0, 1} can be used to represent the value of a certain training sample. If it is 1, it means that the relational edge r between the two knowledge nodes exists; if it is 0, it means that the relational edge r does not exist. Usually, the sigmoid function can be used to map the prediction result to the interval [0, 1]. The specific form is as follows:

[0099]

[0100] For the weight factor θ p , the loss function can be designed through the following linear transformation plus maximum likelihood estimation:

[0101]

[0102]

[0103] Finally, the intelligent decision-making model based on the knowledge graph semantic web can be converted into an optimization objective function regarding the optimal weight factor θ p .

[0104] After the initial construction of the knowledge graph intelligent decision-making model, evaluate the knowledge graph intelligent decision-making model. If the evaluation result shows that the decision-making model can be used for decision-making, the output values such as water demand can be obtained through input from the information system;

[0105] If the evaluation result shows that the decision-making model cannot be used for decision-making, perform knowledge triple update extraction to update the intelligent decision-making model.

[0106] Among them, the input of the knowledge graph intelligent decision-making model is the prediction of the water demand of various water sources, and this prediction is the basis for supervised random walk. According to this prediction, search and decision-making can be carried out in the knowledge graph intelligent decision-making model to seek the optimal water demand output result. The following lists a non-limiting example of the water demand prediction at the input end.

[0107] Suppose it is known that within a certain period of time, the water demands of a certain type of water source from time 1 to time t are Q1, Q2...

[0108] Q t , the water demand Q t+1 at time t + 1 of this type of water source can be predicted through the water demand prediction model, and the following prediction model can be used.

[0109] The change value of the water demand at time t:

[0110] ΔQ t = Q t - Q t-1 (7)

[0111] The second-order difference change value of the water demand at time t:

[0112] Δ 2 Q t = ΔQ t - ΔQ t-1 (8)

[0113] The predicted value of the second-order difference change of the water demand at time t:

[0114]

[0115] Therefore, the predicted value of the water demand at time t + 1:

[0116]

[0117] Input By searching through the intelligent decision-making model of the already constructed knowledge graph, the corresponding water demand value can be output.

[0118] As Figure 1 shown, the regulation of the regulator is based on the output of the decision-making system for the gas production volume, water demand, and water safety threshold in the next stage. Through the specific output results of the decision-making system, the current shale gas extraction intensity and water demand can be adjusted by the regulator, which can be achieved by reducing the waste liquid production volume and increasing the rainwater usage.

Claims

1. A method for constructing a knowledge graph model for water resource allocation in shale gas exploitation, characterized in that It includes the following steps: In step 1), the engineering data for shale gas extraction water resource allocation comes from Chinese databases, patent databases, and research reports, and the said engineering data is preprocessed; In step 2), from the perspective of water resource allocation, select the project name, allocation time, and water resource categories to be allocated as the knowledge nodes of the knowledge graph model; select the water demand of the target type of water and the shale gas extraction intensity as the knowledge nodes; finally, select "time is", "water demand is", and "contains" as the relationship edges between the nodes; In step 3), according to the defined knowledge nodes and relationship edges, list the schematic diagram of the knowledge graph model, and expand the knowledge graph semantic network diagram to several knowledge triples according to the number of knowledge triples; In step 4), conduct relevant knowledge reasoning engineering according to the knowledge graph semantic network, predict the next entity from one entity and find the relationship between them, thereby constructing a knowledge graph intelligent decision-making model; In step 4), conduct relevant knowledge reasoning engineering, predict the next entity from one entity and find the relationship between them, thereby constructing a knowledge graph intelligent decision-making model. Specifically, the following steps are adopted: Define a knowledge triple, where the knowledge head node is h, the relationship edge is r, and the knowledge tail node is t, thereby determining the knowledge triple (h, r, t); Using the method of supervised random walk, first generate some paths. The paths are composed of a series of knowledge nodes and relationship edges. The specific expression is: Where, T n is the scope range of the relationship edge r n and the domain range of the relationship edge r n-1 , that is, T n = range(r n ) = domain(r n-1 ), thus defining a distribution of relationship edges and knowledge nodes. The value obtained based on the distribution is the eigenvalue X h，p(t) of each random walk path. X h，p(t) is understood as the probability of reaching the knowledge tail node t starting from the knowledge head node h along a path p. The update rule of X h，p(t) is as follows: Denote the probability of reaching knowledge node e along relationship edge r starting from knowledge node e'. l ; r l (e’, e) indicates whether there exists a path of relationship type r between knowledge nodes e’ and e. If it exists, its value is 1; otherwise, the value is 0. |r l (e', ·)| represents the number of knowledge nodes that can be reached from node e' through the relationship edges of the specified path; l (e', ·)| represents the number of knowledge nodes that can be reached from node e' through the relationship edges of the specified path; If you want to determine the relationship edge r between two knowledge nodes, you need to obtain a set of paths P through supervised random walk r =(P1,..., P n ), and supervised random walk is based on random walk, adding a supervised method to guide the knowledge nodes during the walk, making the walk more purposeful; Through the method of supervised random walk, purposefully search for knowledge nodes and relationship edges related to water demand, and then use these paths to train a ranking model for predicting entities. This model is modeled using the method of a linear model: where, f(h, r i , t) represents the probability S i that there is a relationship r i between knowledge node h and knowledge node t p , θ r represents the weight factor of the feature path connected to P p , and through training, the value of θ where y i = {0, 1} represents the value of the target training sample. If it is 1, it means that the relationship edge r between the two knowledge nodes exists. If it is 0, it means that the relationship edge r does not exist. Usually, the sigmoid function is used to map the prediction result to the interval [0, 1]. The specific form is as follows: For the weight factor θ p , the loss function is designed through the following linear transformation and maximum likelihood estimation: Finally, the intelligent decision-making model based on the knowledge graph semantic web is converted into an optimization objective function for the optimal weight factor θ p ; The input of the constructed knowledge graph intelligent decision-making model is the prediction of the water demand of various water sources. This prediction is the basis for the supervised random walk. According to this prediction, search and make decisions in the knowledge graph intelligent decision-making model to seek the optimal water demand output result. The optimal water demand output result is the output of the constructed knowledge graph intelligent decision-making model.

2. The method according to claim 1, wherein: After the knowledge graph intelligent decision-making model is initially constructed, evaluate the knowledge graph intelligent decision-making model. If the evaluation result shows that the decision-making model can be used for decision-making, input through the information system to obtain the water demand output value; If the evaluation result shows that the decision-making model cannot be used for decision-making, then update and extract the knowledge triples, thereby updating the intelligent decision-making model.

Images