Cement strength prediction method and system based on explicit mapping of clinker chemical mechanism and edge feature coding

Abstract

The present application relates to a kind of cement strength prediction method and system based on explicit mapping of clinker chemistry mechanism and edge feature coding, belong to cement material performance prediction and intelligent modeling technical field, including: obtaining clinker chemical composition data and target age parameter;Calculate mineral phase information, based on clinker chemical composition data, mineral phase information and target age parameter to build clinker chemistry mechanism diagram;Edge feature coding is carried out to edge connection, and external chemical edge feature is generated;Clinker chemistry mechanism diagram and external chemical edge feature are input into trained cement strength prediction model, and the cement strength prediction result under target age is acquired.The present application is by explicit construction clinker chemistry mechanism diagram and introduces edge feature coding, realize the effective expression of the structured relationship between clinker chemistry composition, mineral phase formation and strength development, enhance the modeling capacity of model to clinker chemistry mechanism relationship, to improve the accuracy and explainability of cement strength prediction.

Description

Technical Field

[0001] This invention relates to a method and system for predicting cement strength based on explicit graph construction and edge feature encoding of clinker chemical mechanism, belonging to the field of cement material performance prediction and intelligent modeling technology. Background Technology

[0002] Cement, as the most important cementitious material in modern infrastructure construction, directly affects the load-bearing capacity, structural safety, and service durability of engineering structures. Cement strength is not only a core indicator for evaluating cement performance and quality, but also a key basis for reflecting clinker quality, batching rationality, and production process stability. Accurate prediction of cement strength can help researchers and production personnel predict cement performance at target ages during cement production, quality control, and batching optimization, thus providing support for clinker quality evaluation, process control, and production decisions. Especially against the backdrop of the cement industry's continuous development towards high quality, low energy consumption, and refined management, conducting research on cement strength prediction has significant engineering application value and practical significance.

[0003] In existing cement production and performance testing processes, cement strength is typically determined through specimen preparation, standard curing, and compressive strength testing. While these methods can accurately reflect strength levels at target ages, they generally suffer from long testing cycles, high experimental costs, high levels of human intervention, and delayed feedback, making it difficult to meet the application requirements of rapid prediction, real-time monitoring, and timely control in modern cement production processes. This is especially true in scenarios with frequent fluctuations in clinker calcination quality, adjustments to cement batching, and changes in production conditions, hindering the early identification of quality risks and the control of product stability.

[0004] Cement strength prediction replaces some of the traditional testing processes that rely on long-term curing and mechanical testing by establishing a prediction model, thus enabling the acquisition of strength results at the target age at an earlier stage. Current cement strength prediction methods typically employ models such as empirical formulas, linear regression, support vector machines (SVM), random forests (RF), and neural networks, directly inputting clinker chemical composition, rate parameters, and related process variables as planar numerical features into the prediction model.

[0005] However, the formation of cement strength is not the result of a single independent factor, but rather a comprehensive manifestation influenced by the clinker's chemical composition, mineral phase composition, and age evolution. (The last sentence appears to be incomplete and unrelated to the preceding text.) , , , , , , , and There are complex coupling relationships among the oxide components, and different oxides not only participate in , , and The formation of various mineral phases also affects their stability, hydration activity, and subsequent strength growth. Furthermore, the contribution of different mineral phases to cement strength varies at different ages, and age further influences the role of each mineral phase. Therefore, cement strength prediction involves not only multi-source chemical composition information but also multi-level relationships with clear physicochemical significance between oxides, mineral phases, and age-dependent strength. The prediction process exhibits strong coupling, structural dependence, and mechanism dependence. Based on this, while existing prediction methods can establish a mapping relationship between input variables and strength results to some extent, their modeling methods typically remain at the numerical statistical fitting level, focusing more on the direct correspondence between input variables and strength results. They fail to explicitly characterize the mechanistic connection between clinker chemical composition and mineral phase formation, and struggle to depict the structured coupling relationships between different oxides, mineral phases, and age factors. Therefore, these methods still have shortcomings in terms of relationship expression ability, mechanism consistency, and interpretability of prediction results.

[0006] In recent years, graph neural networks, as a deep learning method for graph-structured data, have shown great application potential in data modeling tasks with complex relational structures. By organizing nodes and their relationships into a graph structure, graph models can enhance the ability to model complex relational information to a certain extent. However, most existing graph construction methods for tabular data rely on sample similarity, proximity relationships, or the K-nearest neighbor rule to establish edge connections. This type of graph construction mainly reflects the proximity or statistical correlation of data distribution and is difficult to reflect the action paths with clear physicochemical meaning in the clinker system. For cement strength prediction tasks, if the graph is constructed solely based on sample similarity relationships, it is often difficult to truly reflect the mechanistic influence of oxide composition, mineral phase formation, and age changes on the strength development process, thus limiting the application effect of graph structure modeling methods in this task.

[0007] Meanwhile, in existing graph models or other prediction models, the edge relationships between nodes are usually used only as connectivity information, with little explicit encoding of information such as edge relationship types, interaction strength, node activity, strength correlation, and risk attributes. This makes it difficult to effectively embed important relationship information in clinker chemistry mechanisms into the model. Especially in systems like cement materials with well-defined component relationships, mineral phase formation pathways, and performance transfer mechanisms, if edge relationships cannot be structurally modeled, it is difficult to fully utilize knowledge from the clinker chemistry field to improve prediction performance, and it is also detrimental to enhancing the model's mechanistic interpretability and engineering application credibility.

[0008] Therefore, how to construct a clinker chemical mechanism diagram with clear physicochemical meaning based on the influence of clinker chemical composition, mineral phase formation relationship and mineral relative strength development, and further realize the explicit expression of the relationship between nodes through edge feature encoding, thereby improving the accuracy and interpretability of cement strength prediction, has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0009] To address at least one of the technical problems in the background art, this invention constructs a clinker chemical mechanism diagram with clear physicochemical meaning based on the influence relationship between clinker chemical composition, mineral phase formation, and relative mineral strength development. By explicitly encoding the edge features of the edge connections between nodes, the invention enables the prediction of cement strength, thereby improving the accuracy and interpretability of the prediction results and providing support for clinker quality evaluation, batching optimization, and production process control.

[0010] The present invention adopts the following technical solution:

[0011] In a first aspect, the present invention provides a method for predicting cement strength based on explicit graph construction and edge feature encoding of clinker chemical mechanism, comprising:

[0012] Obtain clinker chemical composition data and target age parameters, whereby the clinker chemical composition data includes calcium oxide ( ), silicon dioxide ( ), aluminum oxide ( ), ferric oxide ( ), magnesium oxide ( ), sulfur trioxide Sodium oxide () ), potassium oxide ( ) and free calcium oxide ( );

[0013] The target age parameter can be determined manually according to needs, and is generally at least one of the 3rd day, the 7th day, and the 28th day.

[0014] Mineral phase information is calculated based on clinker chemical composition data. A clinker chemical mechanism diagram is constructed based on clinker chemical composition data, mineral phase information and target age parameters. The clinker chemical mechanism diagram includes oxide nodes, mineral phase nodes and age nodes, and edge connections are explicitly established between oxide nodes, mineral phase nodes and age nodes.

[0015] Edge features are encoded to generate external chemical edge features;

[0016] The clinker chemical mechanism diagram and external chemical edge features are input into the trained cement strength prediction model (DynamicGNN Transformer Predictor, or DGT Predictor for short) to obtain the cement strength prediction results at the target age. The cement strength prediction model extracts local structural features from the clinker chemical mechanism diagram through the local relation modeling module, performs global relation modeling by combining external chemical edge features through the global dependency modeling module, and outputs the cement strength prediction results at the target age through the feature fusion module and the graph-level regression module.

[0017] Furthermore, the process of calculating mineral phase information based on clinker chemical composition data is as follows:

[0018] Based on the Bogue formula and the chemical composition data of clinker, tricalcium silicate was calculated (…). ), dicalcium silicate ( Tricalcium aluminate ( ) and tetracalcium aluminoferrite ( ), generating mineral phase information;

[0019] Mineral phase information is used as an intermediate mechanism expression layer in the construction of clinker chemical mechanism diagram.

[0020] Furthermore, in the clinker chemical mechanism diagram, there are 9 oxide nodes, 4 mineral phase nodes, and 1 age node;

[0021] The types of edge connections include forming edges, regulating edges, synergistic edges, stability-influencing edges, age-related edges, and strength-developing edges; edge connections are explicitly established based on the chemical composition relationship of clinker, the mineral phase formation relationship, the material regulating relationship, and the strength-influencing relationship.

[0022] Furthermore, the process of encoding edge features for edge connections to generate external chemical edge features is as follows:

[0023] For each edge of the clinker chemical mechanism diagram, read the corresponding source node, target node, and edge relationship type;

[0024] Generate relation type codes based on edge relation types, and calculate the action intensity based on the numerical information and reaction activity information corresponding to the source node and target node;

[0025] Extract source node reactivity, target node reactivity, intensity-related markers, and risk markers;

[0026] The relation type encoding, action strength, source node reactivity, target node reactivity, strength correlation label and risk label are concatenated to generate the edge feature vector corresponding to each edge.

[0027] The edge feature vectors of each edge are organized into an edge feature matrix or an edge feature tensor to generate external chemical edge features.

[0028] Furthermore, the cement strength prediction model includes a local relation modeling module, a global dependency modeling module, a gated fusion module, and a graph-level regression module connected in sequence. The local relation modeling module includes a graph neural network encoding unit (i.e., a GNN encoding unit), the global dependency modeling module includes an edge-enhanced attention Transformer encoding unit, the gated fusion module includes a gated fusion unit, and the graph-level regression module includes a pooling unit, a graph representation transformation unit, and a fully connected regression unit.

[0029] Furthermore, the pooling unit simultaneously employs mean pooling and max pooling to aggregate the fused node representations, and then concatenates the aggregation results before inputting them into the graph representation transformation unit.

[0030] The predicted cement strength at the target age includes at least one of the following: 3-day compressive strength, 7-day compressive strength, and 28-day compressive strength.

[0031] Furthermore, the graph neural network encoding unit is used to extract the local structure representation of nodes based on the clinker chemical mechanism diagram. And generate gated vectors ;

[0032] The edge-enhanced attention Transformer encoding unit is used to perform feature mapping on external chemical edge features and utilize the mapped edge features to participate in attention relationship modeling between nodes, so as to obtain a global dependency representation of nodes. ;

[0033] The gate fusion module is used to utilize gate vectors Global dependency representation of nodes Gated modulation is performed, and the modulated global dependency representation of the node is fused with the local structure representation of the node to generate a fused node representation. , where ⊙ denotes element-wise multiplication;

[0034] The graph-level regression module is used to represent the fused nodes. Pooling aggregation and graphical representation transformation are performed, and the cement strength prediction results at the target age are output through fully connected regression units.

[0035] Secondly, the present invention provides a cement strength prediction system based on explicit graph construction and edge feature encoding of clinker chemical mechanism, comprising:

[0036] The data acquisition module is used to acquire clinker chemical composition data and target age parameters;

[0037] The mineral phase determination module is used to determine mineral phase information based on clinker chemical composition data;

[0038] The clinker chemical mechanism diagram construction module is used to construct a clinker chemical mechanism diagram based on clinker chemical composition data, mineral phase information and target age parameters, and to explicitly establish edge connections between oxide nodes, mineral phase nodes and age nodes.

[0039] The edge feature encoding module is used to encode the edge features of edge connections to generate external chemical edge features;

[0040] The strength prediction module is used to input the clinker chemical mechanism diagram and external chemical edge features into the trained cement strength prediction model to obtain the cement strength prediction results at the target age.

[0041] Thirdly, the present invention provides an electronic device, including a memory, a processor, and a computer program. The processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to implement the steps of the above-mentioned method for predicting cement strength based on explicit mapping and edge feature encoding of clinker chemical mechanism.

[0042] Fourthly, the present invention provides a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the steps of the cement strength prediction method based on explicit graphing and edge feature encoding of clinker chemical mechanism as described above.

[0043] Fifthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method for predicting cement strength based on explicit graphing and edge feature encoding of clinker chemical mechanisms.

[0044] For any details not covered in this invention, please refer to the prior art.

[0045] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0046] (1) This invention uses explicit mapping based on clinker chemical mechanisms. By constructing a clinker chemical mechanism diagram with clear physicochemical meaning by considering the influence of clinker chemical composition, mineral phase formation relationships, and relative mineral strength development, it replaces the traditional modeling method in some existing technologies that directly inputs clinker chemical composition and related process parameters as planar numerical features into the model. This achieves a structured expression of the clinker system, so that the prediction model no longer relies solely on statistical fitting at the numerical level, but combines the internal mechanism relationships of the material for modeling. This improves the consistency between the cement strength prediction process and the material mechanism, enhances the model's relational expression ability and the interpretability of the prediction results.

[0047] (2) This invention designs a cement strength prediction model (DGTPredictor) based on clinker chemical mechanism diagrams. It extracts local structural relationships from the clinker chemical mechanism diagram using graph neural network encoding units, models global dependencies between nodes using a global dependency modeling module based on edge-enhanced attention Transformer encoding units combined with external chemical edge features, and outputs the cement strength prediction results at the target age through a gating fusion module and a graph-level regression module. This technical solution can simultaneously utilize local interactions and global coupling relationships in the clinker system, overcoming the problems of insufficient modeling of complex material relationships and difficulty in balancing local structural information and global correlation information in existing methods. This improves the model's ability to represent complex clinker system relationships and is beneficial for enhancing the accuracy and stability of cement strength prediction.

[0048] (3) To address the problem that edge relationships in existing graph models are only used as connectivity information and cannot reflect the true chemical mechanisms, this invention explicitly encodes edge relationship types, interaction strengths, node activities, strength-related markers, and risk markers, and inputs the encoded external chemical edge features into the prediction model to participate in relationship modeling. In this way, edge relationships not only have connectivity meaning but also have clear mechanistic semantics, which is beneficial to improving the model's utilization of clinker chemical mechanism information and enhancing the interpretability, robustness, and stability of the prediction results.

[0049] (4) The technical solution provided by this invention can be applied to scenarios such as clinker quality evaluation, batching optimization, production process control, and strength trend prediction. By predicting the cement strength at the target age based on the chemical composition, mineral phase information, and structured relationship of clinker, the problem of result lag caused by relying solely on long-cycle mechanical tests for strength evaluation can be reduced to a certain extent. This allows production personnel to obtain information on strength development trends earlier, thereby improving the timeliness of prediction, control efficiency, and intelligence level in cement production and quality control processes. Attached Figure Description

[0050] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an undue limitation of this application.

[0051] Figure 1 This is a schematic diagram of the cement strength prediction method based on explicit graph construction and edge feature encoding of clinker chemical mechanism provided in Embodiment 1 of the present invention.

[0052] Figure 2 This is a schematic diagram of the construction of the clinker chemical mechanism diagram provided in Embodiment 1 of the present invention;

[0053] Figure 3This is a schematic diagram of the external chemical edge feature encoding provided in Embodiment 1 of the present invention;

[0054] Figure 4 This is a schematic diagram of the cement strength prediction model (DGT Predictor) provided in Embodiment 1 of the present invention;

[0055] Figure 5 This is a schematic diagram of the cement strength prediction model training process provided in Embodiment 1 of the present invention;

[0056] Figure 6 This is a schematic diagram of the cement strength prediction system provided in Embodiment 2 of the present invention. Detailed Implementation

[0057] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0058] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments of the invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form as well. Furthermore, it should be understood that the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0059] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0060] All data acquisition in this embodiment is carried out in accordance with laws and regulations and with user consent, and the data is used legally.

[0061] Example 1

[0062] A method for predicting cement strength based on explicit graph construction and edge feature encoding of clinker chemical mechanism, such as... Figure 1 As shown, it includes:

[0063] Obtain clinker chemical composition data and target age parameters, whereby the clinker chemical composition data includes calcium oxide ( ), silicon dioxide ( ), aluminum oxide ( ), ferric oxide ( ), magnesium oxide ( ), sulfur trioxide Sodium oxide () ), potassium oxide ( ) and free calcium oxide ( );

[0064] The target age parameter can be determined manually according to needs, and is generally at least one of the 3rd day, the 7th day, and the 28th day.

[0065] Mineral phase information is calculated based on clinker chemical composition data. In this embodiment, tricalcium silicate is calculated based on the Bogue formula and clinker chemical composition data. ), dicalcium silicate ( Tricalcium aluminate ( ) and tetracalcium aluminoferrite ( ), generating mineral phase information;

[0066] A clinker chemical mechanism diagram is constructed based on clinker chemical composition data, mineral phase information, and target age parameters. The clinker chemical mechanism diagram includes oxide nodes, mineral phase nodes, and age nodes, and edge connections are explicitly established between oxide nodes, mineral phase nodes, and age nodes.

[0067] Edge features are encoded to generate external chemical edge features;

[0068] The clinker chemical mechanism diagram and external chemical edge features are input into the trained cement strength prediction model (DynamicGNN Transformer Predictor, or DGT Predictor for short) to obtain the cement strength prediction results at the target age.

[0069] This embodiment constructs and applies a prediction method consisting of a clinker chemical mechanism diagram, external chemical edge feature encoding, and a cement strength prediction model. This method addresses the issue that there is a significant mechanistic correlation between clinker chemical composition and cement strength, and that traditional planar feature input methods are insufficient to fully express the interaction paths between components. It organizes the oxide composition, mineral phase formation relationships, and age-related influences in the clinker system into a clinker chemical mechanism diagram with clear physicochemical meaning. Furthermore, it incorporates information such as edge relationship types, interaction strength, and node activity into the prediction model through edge feature encoding, achieving collaborative modeling of local structural relationships and global dependencies, thereby improving the accuracy and interpretability of cement strength prediction.

[0070] Specifically, in this embodiment, the operation of calculating mineral phase information based on clinker chemical composition data includes:

[0071] According to clinker , , , and The content of mineral phases is calculated based on the Bogue formula; in one example, the mineral phase composition can be calculated using the following formula:

[0072] =4.071× -7.60× -6.718× -1.430× -2.852×

[0073] =2.867× -0.7544×

[0074] =2.650× -1.692×

[0075] =3.043×

[0076] in, , , , and These represent the mass percentage content of the corresponding oxides. , , and This indicates the calculated percentage content of mineral phases by mass. To ensure that the calculation results meet the reasonable range for engineering applications, the mineral phase results can be further limited to a preset range, such as between 0 and 100%.

[0077] The above formula is the standard Bogue formula, where each oxide is represented as a mass percentage (%). The numbers preceding the oxides are stoichiometric coefficients, used to calculate the relative content of the four main mineral phases based on the content of each major oxide in the clinker. These stoichiometric coefficients are derived from the theoretical chemical reaction stoichiometric relationships in cement clinker mineral formation. For example, coefficients 4.071, 7.60, 6.718, 1.430, and 2.852 are based on... During formation , , , and Empirical coefficients derived from stoichiometric requirements and interference corrections, , and The stoichiometric coefficients in [the original text] are obtained similarly.

[0078] In this way, the original chemical composition of clinker can be further transformed into an intermediate expression layer that is more mechanistic to the formation of strength, thereby establishing a transmission path that is more in line with the material mechanism between clinker composition and strength results.

[0079] Specifically, such as Figure 2 As shown, the clinker chemical mechanism diagram includes oxide node layers, mineral phase node layers, and age node layers. The oxide node layers include... , , , , , , , and Oxide nodes; mineral phase node layers include , , and Mineral phase nodes; the age node layer includes age nodes, which are used to represent target age information.

[0080] In this embodiment, the edge connections in the clinker chemical mechanism diagram include formation edges, regulation edges, synergistic edges, stability influence edges, age-related effects edges, and strength development edges.

[0081] Among them, the formation edge is used to characterize the relationship between oxide components and mineral phase formation, for example and , , The formation of relationships between them and , The formation of relationships between them and The formation relationship between them, and and The formation relationship between them;

[0082] Adjustment edges or stability-affecting edges are used to characterize , , , and The moderating effect and risk impact of components on mineral phase formation or stability;

[0083] Synergistic edges are used to characterize synergistic relationships between mineral phases, for example... and The synergistic relationship between silicates, and and Synergistic relationship of aluminates between them;

[0084] The age-dependent boundary is used to characterize the influence of age factors on the hydration behavior of mineral phases;

[0085] Intensity development edge is used to characterize the contribution relationship of minerals to intensity development at different ages.

[0086] In one specific implementation, the edge connections in the clinker chemical mechanism diagram can be represented as follows:

[0087]

[0088] in, Represents a set of nodes. Represents the set of edges; set of nodes At least includes a set of oxide nodes Mineral phase node set Age Node Set ,Right now:

[0089] = ∪ ∪

[0090] edge set This includes various edge connection types explicitly established based on the chemical composition relationship, mineral phase formation relationship, and intensity influence relationship of clinker. Through the above-described mapping method, the interaction path between "chemical composition-mineral phase-age intensity" in the clinker system can be expressed in a structured manner. Specifically, in this embodiment, to improve the model's ability to express edge relationships in the graph, edge feature encoding is performed on the edge connections in the clinker chemical mechanism graph to generate external chemical edge features.

[0091] like Figure 3 As shown, this relates to a single edge in the clinker chemical mechanism diagram. First, read the source node, target node, and edge relationship type corresponding to the edge. In a directed graph, each edge has a clear direction. For an edge, the source node refers to the starting point of the directed edge, that is, the edge originates from this node; the target node refers to the ending point of the directed edge, that is, the edge points to this node.

[0092] Generate relation type codes based on edge relation types, and calculate the action intensity based on the numerical information and reaction activity information corresponding to the source node and target node;

[0093] Extract source node reactivity, target node reactivity, intensity-related markers, and risk markers;

[0094] The relation type encoding, action strength, source node reactivity, target node reactivity, strength correlation label and risk label are concatenated to generate the edge feature vector corresponding to each edge.

[0095] The edge feature vectors of each edge are organized into an edge feature matrix or an edge feature tensor to generate external chemical edge features.

[0096] In one example, a single edge The edge feature vector can be represented as:

[0097] =[ , , , , , ]

[0098] in, Indicates the edge relationship type encoding. Indicates the strength of the edge effect. Indicates the reaction activity of the source node. Indicates the reactive activity of the target node. Indicates intensity-related markers, Indicates a risk marker.

[0099] In one specific implementation, the numerical information of the source node and the target node is obtained through real-time calculation and data extraction. After reading the measured oxide content from the external input, the percentage content of the four key mineral phases is calculated in real time based on these oxides using the built-in Bogue correction formula. According to the connection relationship defined in the figure, the content or age values ​​corresponding to the source node and the target node are dynamically obtained from the merged composition list.

[0100] The reactivity of source and target nodes can be obtained through an expert knowledge base index. During the initialization phase, a preset value representing the physicochemical activity of each chemical component is assigned. During graph construction, the corresponding activity constant is retrieved from this static attribute table based on the component name as an index.

[0101] The strength of the edge effect The calculation can be performed based on the numerical information of the source and target nodes, as well as the reactivity information. For non-age edges, the calculation can be performed in the following form:

[0102]

[0103] in, and These represent the numerical information corresponding to the source node and the target node, respectively. and These represent the reactive activity of the source node and the target node, respectively.

[0104] For edges related to age nodes, normalization can be performed based on the age value, for example:

[0105]

[0106] in, Indicates the current target age. This represents the preset maximum age. By applying logarithmic normalization to age-related edges, the influence of edge relationships on intensity development under different age conditions can be reflected more stably.

[0107] Edge strength related markers With risk marker It is generated using hard-coded rules based on prior knowledge of cement chemistry. Strength-related markers are obtained by identifying specific hydration reaction pathways. For example, when encountering an edge connecting "tricalcium silicate and age," the edge is identified as a strength-contributing type and its characteristic value is set to 1.0, while other edges not directly related to strength are set to 0.5 by default. Risk markers are obtained by identifying stability hazards. For example, when searching for a specific edge connecting "free calcium oxide" and "tricalcium silicate," it is marked as a stability risk type and its risk characteristic value is set to 1.0, while all other edges remain at 0.0.

[0108] After organizing the edge feature vectors of all edges according to the connection relationships between nodes, we can obtain the external chemical edge feature tensor:

[0109]

[0110] in, This indicates the number of nodes in the graph. This represents the dimension of the edge features. In this embodiment, A value of 6 is acceptable. Through the above edge feature encoding, edge relationships no longer only have a connecting function, but can also carry clear mechanistic semantic information, thereby enhancing the model's ability to express the relationship information between nodes.

[0111] like Figure 4 As shown, the cement strength prediction model includes a local relationship modeling module, a global dependency modeling module, a gating fusion module, and a graph-level regression module connected in sequence.

[0112] The local relation modeling module includes a graph neural network encoding unit (GNN encoding unit), which is used to extract the local structure representation of nodes based on the clinker chemical mechanism diagram and generate gating vectors; the global dependency modeling module includes an edge-enhanced attention Transformer encoding unit, which is used to perform feature mapping on external chemical edge features and use the mapped edge features to participate in the attention relationship modeling between nodes in order to obtain the global dependency representation of nodes.

[0113] The gated fusion module includes a gated fusion unit, which uses a gated vector to perform gated modulation on the global dependency representation of nodes and fuses the modulated global dependency representation of nodes with the local structure representation of nodes. The graph-level regression module includes a pooling unit, a graph representation transformation unit, and a fully connected regression unit, which are used to perform pooling aggregation, graph representation transformation, and regression calculation on the fused node representation to output the cement strength prediction results at the target age.

[0114] In one implementation, let the node feature matrix be:

[0115]

[0116] in, For the number of nodes, This represents the node feature dimension. The neural network encoding unit receives the node feature matrix from the clinker chemical mechanism diagram. Based on graph edge connection information, output a local structure representation. and gate vector :

[0117]

[0118] in, ∈ , ∈ ; The dimension representing the hidden layer features. Functions representing the local structure modeling encoding module. Gating vector. The gating strength for different nodes can be represented by the Sigmoid activation function, which can be constrained to between 0 and 1.

[0119] Subsequently, external chemical edge features Feature mapping is performed to obtain edge-enhanced representations, which are then combined with local structure representations. Input edge-enhanced attention Transformer encoding unit to obtain global dependency representation of nodes. :

[0120]

[0121] A function representing the global dependency modeling and coding module;

[0122] In the gating fusion phase, the fused node representation It can be represented as:

[0123]

[0124] Here, ⊙ represents element-wise multiplication. This gated fusion method allows for the modulation of the global dependency representation of the Transformer using local structural information extracted from the graph neural network, thereby achieving collaborative modeling of local structural relationships and global coupling relationships.

[0125] In the graph-level regression stage, the fused node representation Perform mean pooling and max pooling respectively to obtain the graph-level representation:

[0126]

[0127] This represents the average pooling function; This represents the max pooling function.

[0128] Subsequently, the graph representation is mapped using a graph representation transformation unit and a fully connected regression unit to obtain the final intensity prediction result:

[0129]

[0130] in, This represents the predicted cement strength at the target age. This indicates that the multilayer perceptron is used as a function to perform a nonlinear mapping on the input data.

[0131] In this embodiment, the graph representation transformation unit includes mean pooling and max pooling. It constructs a mapping by calculating the average of all 14 component feature vectors and extracting the strongest feature value in each dimension, and then concatenates the two pooled vectors end to end. The fully connected regression unit includes a multilayer perceptron (MLP), which performs mapping by abstracting nonlinear features layer by layer, and the linear layers concatenate the concatenated vectors. We perform weighted combinations, then simulate the threshold effect in chemical reactions using the nonlinear activation function ReLU, and finally compress the abstract high-dimensional features into a one-dimensional scalar result through regression mapping.

[0132] Through the above model structure, this embodiment realizes end-to-end intensity prediction based on clinker chemical mechanism graph and external chemical edge features. Specifically, the Generative Neural Network (GNN) encoding unit focuses on extracting local structural relationships in the graph, the edge-enhancing Transformer encoding unit focuses on modeling long-range dependencies and global coupling relationships in the graph, the gating fusion module achieves effective coordination of local structural information and global dependency information, and the graph-level regression module aggregates node-level representations into graph-level representations and outputs the final intensity prediction value.

[0133] Specifically, such as Figure 5 As shown, the training process of the cement strength prediction model includes:

[0134] Obtain historical clinker chemical composition data, age data, and their corresponding strength labels;

[0135] Mineral phase information is calculated based on clinker chemical composition data, a clinker chemical mechanism diagram is constructed, and edge features are encoded for edge connections to generate external chemical edge features;

[0136] Input the graph data, external chemical edge features and strength labels into the cement strength prediction model for forward calculation, and output the predicted strength value;

[0137] Calculate the loss function based on the predicted values ​​and the actual intensity labels;

[0138] Based on the loss function, the model parameters are updated using the backpropagation algorithm until the training termination condition is met. In one implementation, the loss function can be the mean squared error loss, and its calculation formula is as follows:

[0139]

[0140] in, Indicates the number of samples. This represents the predicted intensity value of the k-th sample. This represents the true strength label corresponding to the k-th sample. By minimizing the above loss function, the model parameters can be progressively optimized, enabling the model to learn the mapping relationship between clinker chemical composition, mineral phase information, age conditions, and cement strength.

[0141] In one specific embodiment, when constructing the clinker chemical mechanism diagram, one can select... , , , , , , , , , , , , The graph structure consists of 14 nodes, including the Age node. The first 9 nodes are measured oxide nodes, the middle 4 nodes are mineral phase nodes calculated using the Bogue formula, and the Age node represents the target age information. For each node, a node feature vector is constructed, including normalized values, molar mass, reactivity, importance markers, node type markers, and logarithmic transformation terms, thus forming a complete node feature matrix.

[0142] For edge feature construction, fixed-dimensional external chemical edge feature vectors can be generated based on edge relationship type, node activity, and influence strength, and organized into an edge feature tensor input model. During the training phase, the clinker chemical mechanism diagram, external chemical edge features, and true strength labels corresponding to each sample constitute the training dataset, which is then trained and evaluated using a training set, validation set, and test set partitioning method. After model training is complete, it can output at least one predicted result among 3-day compressive strength, 7-day compressive strength, and 28-day compressive strength.

[0143] Example 2

[0144] A cement strength prediction system based on explicit graph construction and edge feature encoding of clinker chemical mechanism, such as Figure 6 As shown, it includes:

[0145] The data acquisition module is used to acquire clinker chemical composition data and target age parameters;

[0146] The mineral phase determination module is used to determine mineral phase information based on clinker chemical composition data;

[0147] The clinker chemical mechanism diagram construction module is used to construct a clinker chemical mechanism diagram based on clinker chemical composition data, mineral phase information and target age parameters, and to explicitly establish edge connections between oxide nodes, mineral phase nodes and age nodes.

[0148] The edge feature encoding module is used to encode the edge features of edge connections to generate external chemical edge features;

[0149] The strength prediction module is used to input the clinker chemical mechanism diagram and external chemical edge features into the trained cement strength prediction model to obtain the cement strength prediction results at the target age.

[0150] Among them, the mineral phase information can be determined based on the Bogue formula to determine at least one mineral phase among tricalcium silicate, dicalcium silicate, tricalcium aluminate, and tetracalcium aluminoferrite.

[0151] The cement strength prediction model called by the strength prediction module can be the cement strength prediction model of Example 1, which is used to output the cement strength prediction results at the target age based on the clinker chemical mechanism diagram and external chemical edge features.

[0152] The cement strength prediction system provided in this embodiment can structurally express the chemical composition of clinker, the formation law of mineral phases and the influence of age, and combine edge feature encoding and graph model prediction to achieve effective prediction of cement strength at the target age, thereby improving the accuracy and interpretability of the prediction results.

[0153] Example 3

[0154] An electronic device includes a memory, a processor, and a computer program. The processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to implement the steps of the cement strength prediction method based on clinker chemical mechanism explicit mapping and edge feature encoding in Embodiment 1.

[0155] The detailed steps are the same as those of the cement strength prediction method based on clinker chemical mechanism explicit mapping and edge feature encoding provided in Example 1, and will not be repeated here.

[0156] Example 4

[0157] A computer-readable storage medium storing computer instructions, which, when executed by a processor, implement the steps of the cement strength prediction method based on explicit graphing and edge feature encoding of clinker chemical mechanism as described in Embodiment 1 above. The detailed steps are the same as those in Embodiment 1, and will not be repeated here.

[0158] Example 5

[0159] A computer program product includes a computer program that, when executed by a processor, implements the steps of the cement strength prediction method based on explicit graphing and edge feature encoding of clinker chemical mechanism as described in Example 1. The detailed steps are the same as those of the cement strength prediction method based on explicit graphing and edge feature encoding of clinker chemical mechanism provided in Example 1, and will not be repeated here.

[0160] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0161] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0162] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0163] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0164] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0165] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

[0166] The above description is merely a preferred embodiment of this practice and is not intended to limit the scope of this practice. Various modifications and variations can be made to this practice by those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this practice should be included within the protection scope of this practice.

Claims

1. A method for predicting cement strength based on explicit graph construction and edge feature encoding of clinker chemical mechanism, characterized in that, include: Obtain clinker chemical composition data and target age parameters, including clinker chemical composition data such as calcium oxide, silicon dioxide, aluminum oxide, ferric oxide, magnesium oxide, sulfur trioxide, sodium oxide, potassium oxide and free calcium oxide; Mineral phase information is calculated based on clinker chemical composition data. A clinker chemical mechanism diagram is constructed based on clinker chemical composition data, mineral phase information and target age parameters. The clinker chemical mechanism diagram includes oxide nodes, mineral phase nodes and age nodes, and edge connections are explicitly established between oxide nodes, mineral phase nodes and age nodes. Edge features are encoded to generate external chemical edge features; Input the clinker chemical mechanism diagram and external chemical edge features into the trained cement strength prediction model to obtain the cement strength prediction results at the target age. The cement strength prediction model includes a local relation modeling module, a global dependency modeling module, a gated fusion module, and a graph-level regression module connected in sequence. The local relation modeling module includes a graph neural network encoding unit, the global dependency modeling module includes an edge-enhanced attention Transformer encoding unit, the gated fusion module includes a gated fusion unit, and the graph-level regression module includes a pooling unit, a graph representation transformation unit, and a fully connected regression unit. The graph neural network encoding unit is used to extract the local structure representation of nodes based on the clinker chemical mechanism diagram. And generate gated vectors ; The edge-enhanced attention Transformer encoding unit is used to perform feature mapping on external chemical edge features and utilize the mapped edge features to participate in attention relationship modeling between nodes, so as to obtain a global dependency representation of nodes. ; The gate fusion module is used to utilize gate vectors Global dependency representation of nodes Gated modulation is performed, and the modulated global dependency representation of the node is fused with the local structure representation of the node to generate a fused node representation. , where ⊙ denotes element-wise multiplication; The graph-level regression module is used to represent the fused nodes. Pooling aggregation and graphical representation transformation are performed, and the cement strength prediction results at the target age are output through fully connected regression units.

2. The cement strength prediction method based on explicit graph construction and edge feature encoding of clinker chemical mechanism according to claim 1, characterized in that, The process of calculating mineral phase information based on clinker chemical composition data is as follows: Based on the Bogue formula, tricalcium silicate, dicalcium silicate, tricalcium aluminate, and tetracalcium aluminoferrite are calculated from the chemical composition data of clinker to generate mineral phase information; Mineral phase information is used as an intermediate mechanism expression layer in the construction of clinker chemical mechanism diagram.

3. The cement strength prediction method based on explicit graph construction and edge feature encoding of clinker chemical mechanism according to claim 2, characterized in that, In the clinker chemical mechanism diagram, there are 9 oxide nodes, 4 mineral phase nodes, and 1 age node; The types of edge connections include forming edges, regulating edges, synergistic edges, stability-influencing edges, age-related edges, and strength-developing edges; edge connections are explicitly established based on the chemical composition relationship of clinker, the mineral phase formation relationship, the material regulating relationship, and the strength-influencing relationship.

4. The cement strength prediction method based on explicit graph construction and edge feature encoding of clinker chemical mechanism according to claim 3, characterized in that, The process of encoding edge features to generate external chemical edge features is as follows: For each edge of the clinker chemical mechanism diagram, read the corresponding source node, target node, and edge relationship type; Generate relation type codes based on edge relation types, and calculate the action intensity based on the numerical information and reaction activity information corresponding to the source node and target node; Extract source node reactivity, target node reactivity, intensity-related markers, and risk markers; The relation type encoding, action strength, source node reactivity, target node reactivity, strength correlation label and risk label are concatenated to generate the edge feature vector corresponding to each edge. The edge feature vectors of each edge are organized into an edge feature matrix or an edge feature tensor to generate external chemical edge features.

5. The cement strength prediction method based on explicit graph construction and edge feature encoding of clinker chemical mechanism according to claim 4, characterized in that, The predicted cement strength at the target age includes at least one of the following: 3-day compressive strength, 7-day compressive strength, and 28-day compressive strength.

6. A cement strength prediction system based on explicit graph construction and edge feature encoding of clinker chemical mechanism, used to implement the cement strength prediction method based on explicit graph construction and edge feature encoding of clinker chemical mechanism as described in any one of claims 1 to 5, characterized in that, include: The data acquisition module is used to acquire clinker chemical composition data and target age parameters; The mineral phase determination module is used to determine mineral phase information based on clinker chemical composition data; The clinker chemical mechanism diagram construction module is used to construct a clinker chemical mechanism diagram based on clinker chemical composition data, mineral phase information and target age parameters, and to explicitly establish edge connections between oxide nodes, mineral phase nodes and age nodes. The edge feature encoding module is used to encode the edge features of edge connections to generate external chemical edge features; The strength prediction module is used to input the clinker chemical mechanism diagram and external chemical edge features into the trained cement strength prediction model to obtain the cement strength prediction results at the target age.

7. An electronic device comprising a memory, a processor, and a computer program, characterized in that, The processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to implement the steps of the cement strength prediction method based on clinker chemical mechanism explicit mapping and edge feature encoding as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing computer instructions thereon, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the cement strength prediction method based on explicit graphing and edge feature encoding of clinker chemical mechanism as described in any one of claims 1 to 5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the cement strength prediction method based on explicit graphing and edge feature encoding of clinker chemical mechanism as described in any one of claims 1 to 5.

Images