Resource utilization scheme generation method, device, equipment and medium of river sediment

Abstract

The invention provides a resource utilization scheme generation method, device, equipment and medium for river sediment, and the method comprises the steps: constructing a target node and a target adjacent matrix based on the component information of the river sediment, and enabling the component types corresponding to two target nodes connected with the node edge of the target adjacent matrix to achieve a synergistic effect; inputting each target node and the target adjacent matrix into a prediction model; graph attention coding and global attention pooling are carried out through a graph attention network to obtain a graph embedding vector, a first prediction head predicts a plurality of target materials based on the graph embedding vector, and a second prediction head predicts a target process corresponding to the target materials based on a target node; and generating a resource utilization scheme based on the plurality of target materials and the plurality of target processes. According to the technical scheme provided by the embodiment of the invention, various target materials and corresponding target processes can be respectively predicted by utilizing the prediction model, and a reliable resource utilization scheme is efficiently generated before the river sediment is removed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent environmental protection engineering, and in particular relates to a resource utilization scheme generation method, device, equipment and medium for river sediment. BACKGROUND

[0002] At present, the problem of river sedimentation is increasingly prominent. With the acceleration of urbanization, the hardening area of the ground surface is continuously expanding, and water and soil loss is intensifying. A large amount of sediment and pollutants flow into the river with rainfall runoff, causing the riverbed to continuously rise, the water section to shrink, and the flood control capacity to decline. Not only does this exacerbate the risk of urban waterlogging, but it also damages the stability and biodiversity of the water ecosystem. In traditional dredging projects, more than 90% of the removed sediment is directly transported to a landfill for simple disposal. This disposal method not only occupies a large amount of valuable land resources, but also poses a risk of secondary pollution of soil and groundwater due to the presence of heavy metals, organic pollutants, and nutrients in the sediment.

[0003] The resources such as organic matter, nutrients, and building materials that can be recycled and utilized in abandoned sediment can be resourcefully utilized according to the composition of the sediment. For example, it can be used to increase the soil nutrients of higher wetland plants and restore wetland ecology, and also used as engineering filler or building material. The sediment in different regions of the river contains different substances. The existing technology usually detects the composition of the sediment after removing it from the river, and manually determines the suitable use according to the composition. However, there are various ways to remove sediment, and different removal methods will cause changes in the composition type and proportion of the sediment. The existing technology cannot determine the resource utilization scheme based on the original state of the sediment, and the reliability of the resource utilization scheme cannot be guaranteed. SUMMARY

[0004] The present application aims to at least solve one of the technical problems existing in the prior art. To this end, the present application provides a resource utilization scheme generation method, device, equipment and medium for river sediment, which can efficiently generate a reliable resource utilization scheme.

[0005] In a first aspect, an embodiment of the present application provides a resource utilization scheme generation method for river sediment, comprising: obtaining a plurality of composition information obtained by detecting the composition of the river sediment in a target area, wherein the composition information comprises a composition type and a composition parameter; constructing a target node based on each of the composition information, and constructing a target adjacency matrix based on a plurality of target nodes, wherein the target adjacency matrix comprises a plurality of node edges, and the composition types corresponding to the two target nodes connected by the node edge can synergize; inputting each of the target nodes and the target adjacency matrix into a trained prediction model, wherein the prediction model comprises a graph attention network, a first prediction head, and a second prediction head, the first prediction head is preset with a plurality of preset materials capable of being made from riverbed mud, and the second prediction head is preset with a plurality of preset processes of the riverbed mud; performing graph attention coding on the target adjacency matrix and the target nodes by the graph attention network to obtain a plurality of coding nodes, performing global attention pooling on the plurality of coding nodes to obtain a graph embedding vector, the first prediction head predicts a plurality of target materials based on the graph embedding vector, and the second prediction head predicts a target process corresponding to each of the target materials based on the target nodes; generating a resource utilization scheme based on the plurality of target materials and the plurality of target processes.

[0006] According to some embodiments of the present application, a target node is constructed based on each of the component information, and a target adjacency matrix is constructed based on a plurality of target nodes, comprising: Based on any of the component information, the component parameters are normalized to determine the feature data of the corresponding target node, and the component type is determined as the node index of the corresponding target node; An environmental node is determined from the target nodes, and the remaining target nodes are determined as feature nodes, wherein the component type of the environmental node is water or pH value; The target adjacency matrix is constructed based on all the feature nodes.

[0007] According to some embodiments of the present application, the target adjacency matrix is constructed based on all the feature nodes, comprising: obtaining a preset prior knowledge template, wherein the prior knowledge template is used to record the synergistic effect between the component types; determining a plurality of node edges based on the prior knowledge template and the feature nodes, and constructing the target adjacency matrix based on the feature nodes and the node edges.

[0008] According to some embodiments of the present application, the graph attention coding on the target adjacency matrix and the target nodes by the graph attention network comprises: determining a plurality of key nodes and adjacency nodes of each of the key nodes based on the target adjacency matrix, wherein the key nodes are the feature nodes connected with the node edges, and the adjacency nodes are the feature nodes connected with the key nodes through the node edges; determine a first attention weight between each of the key nodes and the adjacent nodes based on any one of the key nodes, splice the key nodes and the adjacent nodes based on the first attention weight to obtain a synergy node, and form the encoding node based on a plurality of the synergy nodes, wherein the first attention weight is used to indicate a strength of the synergy, and each of the synergy nodes is a dimension of the encoding node.

[0009] According to some embodiments of the present application, global attention pooling is performed on the plurality of encoding nodes to obtain a graph embedding vector, including: determine a second attention weight between each of the encoding nodes and the environment nodes based on any one of the encoding nodes and the plurality of synergy nodes; perform weighted summation on the plurality of encoding nodes and the corresponding second attention weights to obtain the graph embedding vector.

[0010] According to some embodiments of the present application, the first prediction head includes a fully connected layer and a Sigmoid activation function, and the first prediction head predicts a plurality of target materials based on the graph embedding vector, including: pass the graph embedding vector through the fully connected layer and the Sigmoid activation function in sequence to obtain a target probability vector, wherein the target probability vector is used to indicate an application probability of each of the preset materials; determine the plurality of target materials based on the target probability vector and a preset probability threshold.

[0011] According to some embodiments of the present application, each of the preset processes is provided with a target list, the target list records a plurality of component types and corresponding preset parameters, and the second prediction head predicts a target process corresponding to each of the target materials based on the target node, including: input the plurality of target materials into the second prediction head, and determine a plurality of candidate processes from the plurality of preset processes through the second prediction head; based on any one of the candidate processes, when the component type of the target node is not recorded in the target list, and each of the component parameters satisfies the corresponding preset parameter, determine the corresponding candidate process as the target process; remove the target material corresponding to the candidate process that is not determined as the target process.

[0012] In a second aspect, an embodiment of the present application provides a river sediment resource utilization scheme generation device, comprising at least one control processor and a memory connected in communication with the at least one control processor; the memory stores instructions executable by the at least one control processor, and the instructions are executed by the at least one control processor to enable the at least one control processor to execute the river sediment resource utilization scheme generation method of the first aspect.

[0013] In a third aspect, an embodiment of the present application provides an electronic device comprising the river sediment resource utilization scheme generation device of the second aspect.

[0014] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium storing computer-executable instructions for executing the river sediment resource utilization scheme generation method of the first aspect.

[0015] The river sediment resource utilization scheme generation method according to the embodiment of the present application has at least the following beneficial effects: a plurality of component information obtained by component detection of river sediment in a target area is acquired, wherein the component information includes a component type and a component parameter; a target node is constructed based on each of the component information, and a target adjacency matrix is constructed based on a plurality of the target nodes, wherein the target adjacency matrix includes a plurality of node edges, and the component types corresponding to two target nodes connected by the node edge can cooperate; each of the target nodes and the target adjacency matrix are input into a trained prediction model, wherein the prediction model includes a graph attention network, a first prediction head, and a second prediction head, the first prediction head is preconfigured with a plurality of preset materials that can be made of river sediment, and the second prediction head is preconfigured with a plurality of preset processes of river sediment; a plurality of coding nodes are obtained by graph attention coding of the target adjacency matrix and the target nodes based on the graph attention network, a graph embedding vector is obtained by global attention pooling of the plurality of coding nodes, the first prediction head predicts a plurality of target materials based on the graph embedding vector, and the second prediction head predicts a target process corresponding to each of the target materials based on the target node; and a resource utilization scheme is generated based on the plurality of target materials and the plurality of target processes. According to the technical scheme of the embodiment of the present application, the component information can be collected before the river sediment is removed, the target adjacency matrix is used to represent the cooperation between different components, the prediction model with two prediction heads is used to predict a plurality of target materials and corresponding target processes, and a reliable resource utilization scheme can be efficiently generated before the river sediment is removed. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1is a flow chart of a river channel sediment resource utilization scheme generation method provided by an embodiment of the present application; Figure 2 is a complete flow chart of a river channel sediment resource utilization scheme generation method provided by another embodiment of the present application; Figure 3 is a structural diagram of a river channel sediment resource utilization scheme generation device provided by another embodiment of the present application. DETAILED DESCRIPTION

[0017] Embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference signs represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be understood as a limitation of the present application.

[0018] In the description of the present application, it should be understood that the orientation description, such as the orientation or position relationship indicated by up, down, front, back, left, right, etc. is based on the orientation or position relationship shown in the drawings, and is only for the convenience of describing the present application and simplifying the description, and does not indicate or imply that the device or element indicated must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation of the present application.

[0019] In the description of the present application, the meaning of several is one or more, and the meaning of multiple is more than two, greater than, less than, more than, etc. are understood as not including the number, and above, below, etc. are understood as including the number. If the first, second is described, it is only for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features or the order of indicated technical features.

[0020] In the description of the present application, unless otherwise explicitly limited, the words such as setting, installing, connecting, etc. should be broadly understood, and the person skilled in the art can reasonably determine the specific meaning of the above words in the present application in combination with the specific content of the technical solution.

[0021] The embodiment of the application provides a river channel sediment resource utilization scheme generation method, device, equipment and medium, wherein the river channel sediment resource utilization scheme generation method comprises the following steps: obtaining a plurality of component information obtained by performing component detection on river channel sediment in a target area, wherein the component information comprises a component type and a component parameter; constructing a target node based on each component information, constructing a target adjacency matrix based on a plurality of target nodes, wherein the target adjacency matrix comprises a plurality of node edges, and the component types corresponding to two target nodes connected by the node edge can cooperate; inputting each target node and the target adjacency matrix into a trained prediction model, wherein the prediction model comprises a graph attention network, a first prediction head and a second prediction head, the first prediction head is preconfigured with a plurality of preset materials that can be made from river channel sediment, and the second prediction head is preconfigured with a plurality of preset processes of river channel sediment; performing graph attention coding on the target adjacency matrix and the target node based on the graph attention network to obtain a plurality of coding nodes, performing global attention pooling on the plurality of coding nodes to obtain a graph embedding vector, the first prediction head predicts a plurality of target materials based on the graph embedding vector, and the second prediction head predicts a target process corresponding to each target material based on the target node; and generating a resource utilization scheme based on the plurality of target materials and the plurality of target processes. According to the technical scheme of the embodiment of the application, the component information can be collected before the river channel sediment is removed, the target adjacency matrix is used to represent the cooperation between different components, the prediction model with two prediction heads is used to predict a plurality of target materials and corresponding target processes, and a reliable resource utilization scheme can be efficiently generated before the river channel sediment is removed.

[0022] The technical scheme of the embodiment of the application is further described below based on the drawings.

[0023] Referring to Figure 1 , Figure 1 A flowchart of a river channel sediment resource utilization scheme generation method provided by the embodiment of the application is provided, and the river channel sediment resource utilization scheme generation method comprises the following steps but is not limited to the following steps: S10, obtaining a plurality of component information obtained by performing component detection on river channel sediment in a target area, wherein the component information comprises a component type and a component parameter.

[0024] It should be noted that the target area is an area where river channel sediment needs to be removed, and the embodiment performs data processing to obtain a resource utilization scheme before removing the river channel sediment. In order to obtain component information, sampling of the river channel sediment from the target area is required. The embodiment can perform in-situ and rapid semi-quantitative detection of heavy metals in the sediment in different areas of the river channel by using XRF (X-ray fluorescence spectroscopy), LIBS (laser-induced breakdown spectroscopy), and the like. According to the in-situ screening result, representative samples of the river channel sediment from the target area are collected and sent to a laboratory for accurate analysis to obtain multiple component information by using ICP-MS (inductively coupled plasma mass spectrometry), GC-MS (gas chromatography-mass spectrometry), and the like.

[0025] It should be noted that each component information corresponds to a component type, and the component type can include various heavy metals, organic matter, nutrient salts, and physical components commonly found in river channel sediment. The heavy metals can be chromium, mercury, arsenic, lead, copper, zinc, or nickel commonly found in river channel sediment. The organic matter can be persistent organic pollutants such as polycyclic aromatic hydrocarbons, polychlorinated biphenyls, and pesticides. The nutrient salt can be total nitrogen, total phosphorus, or organic matter content. The physical component can be particle size distribution, mineral composition (XRD), specific surface area, or pH value. Since the sediment components in different areas are different, the number of component information in the target area can be determined according to the actual detection result, and the concentration information can be determined according to the actual detection result. The embodiment retains the detected component information, for example, in the case where polycyclic aromatic hydrocarbons are not detected, the component information corresponding to polycyclic aromatic hydrocarbons is not included, and subsequent repetition is not described.

[0026] It should be noted that the embodiment stores the component information in the server, deploys a prediction model in the server, and inputs the multiple component information into the prediction model for automatic calculation. In the case where the number of component information is large and the calculation process is complex, the prediction model automatically and efficiently generates a resource utilization scheme with high reliability.

[0027] S20, based on each component information, a target node is constructed, and based on multiple target nodes, a target adjacency matrix is constructed, wherein the target adjacency matrix includes multiple node edges, and the component types corresponding to the two target nodes connected by the node edge can synergistically act.

[0028] It should be noted that each component information uniquely corresponds to a component type, and therefore each component information corresponds to a target node. The component parameter can be a related parameter such as concentration, depth, and flow rate, which is determined according to the parameter that can be detected for the component type. The embodiment combines the component type and the component parameter into an array as a target node, for example, node 1 is (chromium, 2.1 mg / kg), and node 2 is (silicon dioxide, 58.5%).

[0029] It should be noted that different component types can synergize in different material making, for example, silicon dioxide and aluminum oxide can synergize in river sediment for making cement raw materials, silicon, aluminum and iron can synergize in sintered bricks, and the synergism of different component types corresponding to each type of target material is known, so multiple node edges can be automatically generated according to prior knowledge, and the synergism of the component types corresponding to the two target nodes is represented by the node edges.

[0030] It should be noted that the river sediment may not have all component types corresponding to a target material, and the embodiment does not limit that the node edges must be generated in the case that all component types of the target material exist in the river sediment, but as long as the two component types can have synergism, the node edges are generated, and in the subsequent first prediction head, the target material lacking some component types will not be predicted, and the exhaustiveness of the node edges will not affect the reliability of the embodiment.

[0031] S30, input each target node and target adjacency matrix into the trained prediction model, wherein the prediction model includes a graph attention network, a first prediction head and a second prediction head, the first prediction head is preset with multiple preset materials that can be made from river sediment, and the second prediction head is preset with multiple preset processes of the river sediment.

[0032] It should be noted that the components of the sediment are not isolated, and multiple component types have synergism (such as the combination of heavy metals and clay, and the influence of organic matter on particle structure). Traditional fully connected neural networks are difficult to capture such complex relationships between components, so the embodiment introduces a graph attention network to learn the synergistic relationship between multiple components, ensuring that the input features used for subsequent prediction can accurately represent the composition of the river sediment.

[0033] It should be noted that the first prediction head and the second prediction head can use common prediction networks, the first prediction head is trained based on multiple preset materials, so that the first prediction head can predict the probability of each preset material according to the input features, and the preset materials can be materials that can be made from river sediment, such as sintered bricks, cement raw materials, ceramsite, geopolymer, microcrystalline glass, adsorbent or functional filler, etc. The second prediction head is trained based on multiple preset processes, so that the second prediction head can exclude the preset processes that cannot be used according to the input information. The preset processes not only include each step of removing the river sediment from the riverbed, but also include the processing process after obtaining the river sediment, for example, the removal process includes dredging, adding chemicals, etc., and the processing process includes washing, high-temperature sterilization, etc.

[0034] S40, obtaining a plurality of encoding nodes by performing graph attention coding on the target adjacency matrix and the target node based on the graph attention network, performing global attention pooling on the plurality of encoding nodes to obtain a graph embedding vector, and predicting a plurality of target materials based on the graph embedding vector by a first prediction head and predicting a target process corresponding to each target material based on the target node by a second prediction head.

[0035] It should be noted that the graph attention network in this embodiment obtains a plurality of encoding nodes by graph attention coding, and obtains a global graph embedding vector by global attention pooling. The graph embedding vector carries component information represented by all target nodes and synergies between each component information. The graph embedding vector is a digital fingerprint representing the component composition of the riverbed mud in the target area. The first prediction head can comprehensively consider all components of the riverbed mud when predicting based on the graph embedding vector.

[0036] It should be noted that the first prediction head can predict all target materials that can be manufactured from the riverbed mud in the target area according to the graph embedding vector. For example, the first prediction head outputs the applicability probability of each preset material. The higher the probability, the more suitable the riverbed mud is for manufacturing the preset material, thereby determining which preset materials the riverbed mud can be applied to.

[0037] It should be noted that in different component compositions and different target materials required for manufacturing, the second prediction head is used in this embodiment to perform exclusive prediction to determine the process that needs to be avoided when the riverbed mud is used to prepare each target material. For example, if the riverbed mud is used to prepare a food-grade adsorbent, chlorine ions cannot be introduced into the process for washing to avoid affecting the durability of the food-grade adsorbent. For another example, when the riverbed mud is used to prepare a biological fertilizer, the riverbed mud cannot be subjected to high-temperature sterilization after being removed from the riverbed to avoid loss of beneficial microorganisms.

[0038] S50, generating a resource utilization scheme based on the plurality of target materials and the plurality of target processes.

[0039] It should be noted that the second prediction head excludes the remaining target processes as processes that can be used to collect the riverbed mud, and each target process corresponds to a target material. Therefore, the prediction model can generate a plurality of mapping pairs according to the target material and the target process, each mapping pair including a target material and a target process. In this embodiment, all mapping pairs are combined into a resource utilization scheme. The resource utilization scheme records the target materials that can be manufactured from the riverbed mud in the target area and the target processes that can be used to prepare each target material. Through automatic identification and processing of the prediction model, the reliability and generation efficiency of the resource utilization scheme are improved.

[0040] In addition, in an embodiment, referring to Figure 2, step S20 specifically includes but is not limited to the following steps: S21, based on any component information, the component parameters are normalized to determine the feature data of the corresponding target node, and the component type is determined as the node index of the corresponding target node; S22, determine the environment node from the target node, and determine the remaining target node as the feature node, wherein the component type of the environment node is water or pH value; S23, based on all feature nodes, construct a target adjacency matrix.

[0041] It should be noted that, since the concentration units of different components are different, in order to ensure that the subsequent prediction model can perform unified dimension calculation, the component parameters are normalized in this embodiment, for example, the target node (chromium, 2.1 mg / kg) is normalized to (chromium, 0.21), and the target node (silicon dioxide, 58%) is normalized to (silicon dioxide, 0.58). Each component information is processed one by one.

[0042] It should be noted that the subsequent processing of this embodiment is based on the component type, so this embodiment determines the component type as the node index of the target node, ensuring that each component type corresponds to only one target node. The specific form can be referred to the above target node example.

[0043] It should be noted that water and pH value do not belong to the components of river sediment, but belong to the environment of river sediment. When determining the target process, the component parameters of the environment node have a certain influence, so this embodiment determines the target node corresponding to water and pH value as the environment node, and determines the remaining as the feature node. When constructing the target adjacency matrix, only the feature nodes are considered to ensure that the target material can be determined according to the components recorded in the feature nodes.

[0044] In addition, in an embodiment, referring to Figure 2 , step S23 specifically includes but is not limited to the following steps: S231, obtain a preset priori knowledge template, wherein the priori knowledge template is used to record the synergistic effect between component types; S232, based on the priori knowledge template and the feature nodes, determine a plurality of node edges, and based on the feature nodes and the node edges, construct a target adjacency matrix.

[0045] It should be noted that the priori knowledge template of the embodiment is an N*N matrix, N is the number of components existing in all known river channel sediments, and is not limited to the target area. The priori knowledge template can be pre-set according to chemical knowledge. For example, silicon dioxide and aluminum trioxide are the core of silicate and aluminate materials, so the synergistic effect of silicon dioxide and aluminum trioxide is constructed in the priori knowledge template. For example, various heavy metals usually coexist and affect ecological risk, so there is a certain synergistic effect between various heavy metals. For example, the relationship between pH and heavy metals is weak, and the synergistic effect of pH and heavy metals is not recorded in the priori knowledge template.

[0046] It should be noted that after obtaining a plurality of feature nodes, the feature nodes are queried in the priori knowledge template. If two feature nodes have a synergistic effect recorded in the priori knowledge template, a node edge is constructed for the two feature nodes in the target adjacency matrix, and the accuracy of the node edge is ensured by traversing each feature node.

[0047] In addition, in an embodiment, with reference to Figure 2 In step S40, a plurality of encoding nodes are obtained by a graph attention network based on the target adjacency matrix and the target node, including but not limited to the following steps: S411, determining a plurality of key nodes based on the target adjacency matrix, and the adjacency nodes of each key node, wherein the key node is a feature node connected by the node edge, and the adjacency node is a feature node connected with the key node through the node edge; S412, determining the first attention weight between any key node and each adjacency node based on the key node, splicing the key node and the adjacency node based on the first attention weight to obtain a synergistic node, and composing the encoding node based on a plurality of synergistic nodes, wherein the first attention weight is used to indicate the strength of the synergistic effect, and each synergistic node is a dimension of the encoding node.

[0048] It should be noted that the target adjacency matrix of the embodiment records a plurality of node edges, and the target node connected by the node edge is the key node of the embodiment, and the other end of the node edge is the adjacency node. At the same time, a small neural network is pre-set in the graph attention network, which is used to calculate the attention score between two nodes.

[0049] For example, taking the target node (chromium, 0.21) as the key node and the adjacency node (lead, 0.5) as an example, a first attention weight W1 is calculated by the small neural network. The greater the value of the first attention weight, the stronger the synergistic effect between the key node and the adjacency node, and the synergistic node obtained is (chromium-lead, W1x0.21||W1x0.5), wherein || represents the splicing operation.

[0050] It should be noted that each key node obtains a coordination node for each adjacent node, in the case of a key node having multiple adjacent nodes, multiple coordination nodes can be obtained, for example, taking the key node (chromium, 0.21) as an example, the obtained coordination nodes can be (chromium-lead, W1x0.21 || W1x0.5) and (chromium-arsenic, W2x0.21 || W2x0.6). All coordination nodes are combined into an encoding node after deduplication in this embodiment, for example, the encoding node corresponding to (chromium, 0.21) is (chromium-lead, W1x0.5 || W1x0.21, chromium-arsenic, W2x0.21 || W2x0.6, ···), the dimension number of the encoding node is greater than or equal to the number of coordination nodes, so that the encoding node can represent all other component types that have a synergistic effect with a component type, and subsequent material prediction pays more attention to the relevance between components.

[0051] It is worth noting that for the non-key nodes of the target adjacency matrix, that is, the feature nodes that do not have any node edges connected, the encoding node is itself, which will not be described here.

[0052] In addition, in an embodiment, with reference to Figure 2 In step S40, the plurality of encoding nodes are globally attention-pooled to obtain a graph embedding vector, specifically including but not limited to the following steps: S421, based on any encoding node, determining a second attention weight between the environment node and the plurality of coordination nodes; S422, performing weighted summation based on the plurality of encoding nodes and the corresponding second attention weights to obtain a graph embedding vector.

[0053] It should be noted that according to the description of the above embodiment, the environment node is the target node corresponding to water and pH value, and the number of nodes of the target adjacency matrix depends on the number of characteristic nodes, for example, there are 6 types of components in addition to water and pH value, and the number of nodes of the target adjacency matrix is 6. When determining the target material, the influence of water content and pH value needs to be introduced, so the embodiment can further determine the association between the encoding node and the environment node by using the small neural network described above, and input all the collaborative nodes and environment nodes into the small neural network to obtain a second attention weight. Each encoding node in the embodiment corresponds to a second attention weight, so that the second attention weight can represent the collaborative relationship between the environment and the multiple types of components that have a collaborative effect. For example, the encoding node obtained above is (chromium-lead, W1x0.5 || W1x0.21, chromium-arsenic, W2x0.21 || W2x0.6, ···), and the second attention weight calculated is taken as W3. The encoding node is updated to (chromium-lead, W1xW3x0.5 || W1xW3x0.21, chromium-arsenic, W2xW3x0.21 || W2xW3x0.6, ···).

[0054] It should be noted that after obtaining the updated encoding nodes, the embodiment splices the encoding nodes to obtain a graph embedding vector. For example, the encoding node is represented by (chromium-lead, W1xW3x0.5 || W1xW3x0.21, chromium-arsenic, W2xW3x0.21 || W2xW3x0.6, ···) as Node1, and the encoding node corresponding to the component type of polycyclic aromatic hydrocarbon is represented by Node2. The graph embedding vector obtained is (Node1, Node2, ···), so that the graph embedding vector can record all component types and the association relationship between components, and a graph embedding vector represents the component digital fingerprint of the river sediment in the target area.

[0055] In addition, in an embodiment, with reference to Figure 2 , the first prediction head includes a fully connected layer and a Sigmoid activation function, and in step S40, the first prediction head predicts a plurality of target materials based on the graph embedding vector, specifically including but not limited to the following steps: S431, the graph embedding vector is sequentially input into the fully connected layer and the Sigmoid activation function to obtain a target probability vector, wherein the target probability vector is used to indicate the application probability of each preset material; S432, determining a plurality of target materials based on the target probability vector and a preset probability threshold.

[0056] It should be noted that the fully connected layer and the sigmoid activation function are the structures for the prediction field scenario. In the case of a large number of component types of the riverbed sediment, the feature dimension of the graph embedding vector is large. The embodiment receives the highly abstract and condensed feature information recorded in the graph embedding vector through the fully connected layer, can comprehensively and linearly combine all the features, for example, determines the conclusions such as "high silicon aluminum content", "heavy metal Cd exceeds the standard", "moderate organic matter", and "moderate particle size" according to a plurality of encoding nodes in the graph embedding vector, calculates a raw and unscaled score for each possible preset material, and the score is the initial score of the application probability. For example, the output of the fully connected layer is [geopolymer: 8.5, ceramic particle: 2.1, adsorbent: -1.2, fertilizer: -5.0, fill material: 0.5], and the geopolymer and ceramic particle are preset materials.

[0057] It should be noted that after the raw score output by the fully connected layer is input into the sigmoid activation function, the sigmoid activation function can squeeze any real number input into a value between 0 and 1. In the embodiment, the value is determined as the application probability. Taking the geopolymer in the above example as an example, sigmoid(8.5)=0.9998 is calculated, so the application probability corresponding to the geopolymer is 0.999, and the other preset materials are calculated in the same way. For example, the target probability vector obtained by the fully connected layer in the above example is [geopolymer: 0.999, ceramic particle: 0.891, adsorbent: 0.232, fertilizer: 0.007, fill material: 0.623].

[0058] It should be noted that the embodiment aims to generate a resource utilization scheme, and therefore a probability threshold is set in advance. The application probability greater than the probability threshold is determined as the target material. For example, when the probability threshold is 0.5, the geopolymer, ceramic particle, and fill material in the above example can be determined as the target material.

[0059] In addition, in an embodiment, with reference to Figure 2 , each preset process is provided with a target list, and the target list records a plurality of component types and corresponding preset parameters. In step S40, the second prediction head predicts the target process corresponding to each target material based on the target node, which specifically includes but is not limited to the following steps: S441, inputting a plurality of target materials into the second prediction head, and determining a plurality of candidate processes from a plurality of preset processes through the second prediction head; S442, based on any candidate process, when the component type of the target node is not recorded in the target list, and each component parameter meets the corresponding preset parameter, the corresponding candidate process is determined as the target process; S443, removing the target material corresponding to the candidate process not determined as the target process.

[0060] It should be noted that the target material is determined based on graph embedding vectors, which already consider environmental nodes. Therefore, in this embodiment, the graph embedding vectors and the target material are directly input into the second prediction head. In the second prediction head, multiple candidate processes are activated based on the target material. The candidate processes only consider those that can produce the target material and do not exclude whether the riverbed sediment meets the collection conditions. Therefore, in this embodiment, a target list is set for each preset process. The target list records the types of components that need to be excluded and preset parameters. For example, for water, the preset parameter is water depth. When the water depth recorded by the environmental node is greater than the preset parameter, the corresponding candidate process cannot be executed and is excluded, and the corresponding target material will not be recorded in the resource utilization scheme.

[0061] like Figure 3 As shown, Figure 3 This is a structural diagram of a device for generating a resource utilization scheme for riverbed sediment according to an embodiment of the present invention. The present invention also provides a device for generating a resource utilization scheme for riverbed sediment, comprising: The processor 401 can be implemented using a general-purpose central processing unit (CPU), microprocessor, application specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 402 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 402 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 402 and is called and executed by the processor 401 to execute the method for generating a resource utilization scheme for riverbed sediment according to the embodiments of this application. Input / output interface 403 is used to implement information input and output; The communication interface 404 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 405 transmits information between various components of the device (e.g., processor 401, memory 402, input / output interface 403, and communication interface 404); The processor 401, the memory 402, the input / output interface 403 and the communication interface 404 are communicatively connected with each other through the bus 405.

[0062] The embodiment of the present application further provides an electronic device, which comprises the river channel sediment resource utilization scheme generation device.

[0063] The embodiment of the present application further provides a storage medium, which is a computer readable storage medium, and stores a computer program, and the computer program is executed by a processor to implement the river channel sediment resource utilization scheme generation method.

[0064] The memory is a non-transitory computer readable storage medium, and can be used to store non-transitory software programs and non-transitory computer executable programs. In addition, the memory can include a high-speed random access memory, and can also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state memory device. In some embodiments, the memory can optionally include a memory remotely arranged relative to the processor, and the remote memory can be connected to the processor through a network. Examples of the network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof. The above-described device embodiments are only schematic, and units described as separate units can or can not be physically separated, and can be implemented in one place or can be distributed on multiple network units. According to actual needs, part or all of the modules can be selected to achieve the purpose of the embodiment scheme.

[0065] As will be appreciated by one of ordinary skill in the art, all or some of the steps, systems, etc. in the above-disclosed methods can be embodied in software, firmware, hardware, and / or suitable combinations thereof. Some or all of the physical components can be implemented with software executed by a processor, such as a central processing unit, a digital signal processor, or microprocessor, or can be implemented as hardware, or as an integrated circuit, such as an application- specific integrated circuit. Such software can be distributed on computer readable media, which can comprise computer storage media (or non-transitory media), and communication media (or transitory media). As is well known to those of ordinary skill in the art, computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by a computer. Further, as is well known to those of ordinary skill in the art, communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal, such as carrier waves or other transport mechanisms, and includes any information delivery media.

[0066] The above description is that of the preferred embodiments of the application. Various modifications and changes can be made thereto without departing from the spirit and scope of the application, which is to be given the broadest interpretation of the laws.

Claims

1. A method for generating a resource utilization scheme of river channel sediment, characterized in that, The method comprises the following steps: obtaining a plurality of component information obtained by detecting the composition of the riverbed sediment in the target area, wherein the component information comprises a component type and a component parameter; constructing a target node based on each component information, and constructing a target adjacency matrix based on a plurality of target nodes, wherein the target adjacency matrix comprises a plurality of node edges, and the component types corresponding to the two target nodes connected by the node edges can cooperate; inputting each target node and the target adjacency matrix into a trained prediction model, wherein the prediction model comprises a graph attention network, a first prediction head and a second prediction head, the first prediction head is preconfigured with a plurality of preset materials that can be made from riverbed sediment, and the second prediction head is preconfigured with a plurality of preset processes of riverbed sediment; performing graph attention coding on the target adjacency matrix and the target node based on the graph attention network to obtain a plurality of coding nodes, performing global attention pooling on a plurality of coding nodes to obtain a graph embedding vector, the first prediction head predicts a plurality of target materials based on the graph embedding vector, and the second prediction head predicts a target process corresponding to each target material based on the target node; generating a resource utilization scheme based on a plurality of target materials and a plurality of target processes.

2. The river channel sediment resource utilization scheme generating method according to claim 1, characterized by, Constructing a target node based on each component information and constructing a target adjacency matrix based on a plurality of target nodes comprises: based on any component information, determining the component parameter as the feature data of the corresponding target node after normalization processing, and determining the component type as the node index of the corresponding target node; determining an environmental node from the target nodes, and determining the remaining target nodes as feature nodes, wherein the component type of the environmental node is water or pH value; constructing the target adjacency matrix based on all feature nodes.

3. The river channel sediment resource utilization scheme generating method according to claim 2, characterized by, Constructing the target adjacency matrix based on all feature nodes comprises: obtaining a preset prior knowledge template, wherein the prior knowledge template is used to record the synergies between the component types; determining a plurality of node edges based on the prior knowledge template and the feature nodes, and constructing the target adjacency matrix based on the feature nodes and the node edges.

4. The method for generating a river sediment resource utilization scheme according to claim 2, wherein, Performing graph attention coding on the target adjacency matrix and the target node based on the graph attention network to obtain a plurality of coding nodes comprises: determining a plurality of key nodes and the adjacency nodes of each key node based on the target adjacency matrix, wherein the key node is the feature node connected with the node edge, and the adjacency node is the feature node connected with the key node through the node edge; based on any key node, determining a first attention weight between each adjacency node, splicing the key node and the adjacency node based on the first attention weight to obtain a synergistic node, and grouping a plurality of synergistic nodes to obtain the coding node, wherein the first attention weight is used to indicate the strength of the synergy, and each synergistic node is a dimension of the coding node.

5. The method for generating a river sediment resource utilization scheme according to claim 4, wherein, The global attention pooling of the plurality of the encoding nodes obtains a graph embedding vector, comprising: Based on any of the encoding nodes, a second attention weight between the environment node and a plurality of the collaborative nodes is determined; Based on a plurality of the encoding nodes and the corresponding second attention weight, a weighted sum is obtained to obtain the graph embedding vector.

6. The river channel sediment resource utilization scheme generating method according to claim 5, wherein The first prediction head includes a fully connected layer and a Sigmoid activation function, and the first prediction head predicts a plurality of target materials based on the graph embedding vector, comprising: The graph embedding vector is sequentially input into the fully connected layer and the Sigmoid activation function to obtain a target probability vector, wherein the target probability vector is used to indicate the application probability of each of the preset materials; Based on the target probability vector and a preset probability threshold, a plurality of the target materials are determined.

7. The method for generating a river sediment resource utilization scheme according to claim 5, wherein, Each of the preset processes is preset with a target list, and the target list records a plurality of the component types and corresponding preset parameters, and the second prediction head predicts a target process corresponding to each of the target materials based on the target node, comprising: A plurality of the target materials are input into the second prediction head to determine a plurality of candidate processes from a plurality of the preset processes through the second prediction head; Based on any of the candidate processes, when the component type of the target node is not recorded in the target list, and each of the component parameters satisfies the corresponding preset parameter, the corresponding candidate process is determined as the target process; The target material corresponding to the candidate process which is not determined as the target process is removed.

8. A river channel sediment resource utilization scheme generation device characterized by comprising: a river channel sediment resource utilization scheme generation unit that generates a river channel sediment resource utilization scheme based on the river channel sediment resource utilization scheme generation conditions. The device comprises at least one control processor and a memory connected in communication with the at least one control processor; the memory stores instructions executable by the at least one control processor, and the instructions are executed by the at least one control processor to enable the at least one control processor to perform the river sediment resource utilization scheme generation method according to any one of claims 1 to 7.

9. An electronic device, comprising: The device for generating a river sediment resource utilization scheme according to claim 8 is provided.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer executable instructions for causing a computer to execute the river sediment resource utilization scheme generation method according to any one of claims 1 to 7.

Images