Concrete compressive strength prediction method, device and equipment and storage medium

Abstract

The invention relates to a concrete compressive strength prediction method and device, equipment and a storage medium. The method comprises the steps of obtaining multi-dimensional feature data; preprocessing the feature data to obtain normalized data; calculating a dynamic weight based on the normalized data; obtaining a hybrid model architecture; and performing prediction based on the dynamic weight and the hybrid model architecture, and outputting a prediction result. The method has the effect of quickly and accurately predicting the concrete compressive strength.

Description

Technical Field

[0001] This application relates to the technical field of big data intelligence, and in particular to a method, apparatus, equipment and storage medium for predicting the compressive strength of concrete. Background Technology

[0002] Concrete, as one of the most widely used materials in modern construction engineering, uses cement as the main gelling material, along with aggregates, water, and, when necessary, chemical admixtures and mineral admixtures, and then mixes them in an appropriate proportion, and hardens and takes shape through uniform pouring, mixing, compaction and curing.

[0003] Compressive strength is a key indicator for evaluating concrete performance, and compressive strength tests are typically conducted after concrete pouring. These tests require a waiting period of several days, which may delay construction progress. However, neglecting the test could lead to serious quality problems and even severe collapse accidents. Therefore, how to quickly and accurately predict the compressive strength of concrete has become an urgent problem to be solved. Summary of the Invention

[0004] In order to quickly and accurately predict the compressive strength of concrete, this application provides a method, apparatus, equipment and storage medium for predicting the compressive strength of concrete.

[0005] Firstly, the method for predicting the compressive strength of concrete provided in this application adopts the following technical solution: A method for predicting the compressive strength of concrete includes: Obtain multi-dimensional feature data; The feature data is preprocessed to obtain normalized data; Calculate dynamic weights based on the normalized data; Obtain the hybrid model architecture; The prediction is made and the prediction results are output based on the dynamic weights and hybrid model architecture.

[0006] By adopting the above technical solution, the electronic device acquires multi-dimensional feature data, covering various factors affecting the compressive strength of concrete. This comprehensive approach avoids the limitations of relying on single or limited features. The feature data is normalized to eliminate the influence of different dimensions and orders of magnitude for subsequent calculations. Dynamic weights are then calculated based on the normalized data, allowing for dynamic adjustments based on the input feature data. This ensures the dynamic weights better reflect reality and improves the accuracy of subsequent predictions. A hybrid model architecture integrates the advantages of different models, achieving complementary strengths and better handling complex prediction scenarios. Finally, the electronic device performs predictions based on the dynamic weights and the hybrid model architecture, outputting the prediction results. This not only enables the prediction of concrete compressive strength but also improves the accuracy of the predictions.

[0007] Optionally, the feature data may be preprocessed to obtain normalized data, including: Sub-features are extracted based on the aforementioned feature data; The sub-features are normalized to obtain normalized data; The formula for calculating the normalized data is: ; in, The minimum value in the dataset of the sub-feature; The maximum value in the dataset for the sub-feature; The current collected value of the sub-feature; For normalized data; Before normalizing the sub-features and obtaining normalized data, the process also includes: All of the aforementioned sub-features are removed. The rejection process includes: Extract any two sub-features from all the sub-features, and use the two extracted sub-features as the first sub-feature and the second sub-feature, respectively. Calculate the correlation coefficient between the first sub-feature and the second sub-feature; Determine whether the correlation coefficient is greater than a first preset threshold; If the correlation coefficient is greater than the first preset threshold, then the first sub-feature or the second sub-feature is removed; The formula for calculating the correlation coefficient is: ; in, The correlation coefficient is mentioned above; For the first sub-feature A number; This is the average value of the first sub-feature; The second sub-feature A number; This is the average value of the second sub-feature.

[0008] By employing the above technical solution, sub-feature normalization can eliminate issues related to dimensional differences and orders of magnitude. Furthermore, before normalization, sub-features undergo a rejection process to remove highly linearly correlated or repetitive sub-features, preventing interference from redundant sub-features in subsequent calculations, improving prediction accuracy, reducing redundant sub-features to improve feature quality, and minimizing unnecessary feature dimensions. This improves both prediction accuracy and efficiency. During the rejection process, a correlation coefficient is calculated to provide a quantitative indicator for determining whether a sub-feature is redundant. If the correlation coefficient exceeds a first preset threshold, two sub-features are considered redundant, and one of them—either the first or second sub-feature—must be removed.

[0009] Optionally, calculating the dynamic weights based on the normalized data includes: Calculate the marginal contribution value of the sub-feature; Calculate the initial weights based on the marginal contribution values; Obtain the adjustment coefficient input by the user; The dynamic weights are calculated based on the initial weights and the adjustment coefficients. The formula for calculating the marginal contribution value is as follows: ; in, The set of all the aforementioned sub-features; To not include the sub-feature A subset of; For input subset only The response output value at that time; For the sub-feature The marginal contribution value; The formula for calculating the initial weights is: ; in, This represents the normalization function, used to normalize... Mapped to values ​​within a reasonable range; The formula for calculating the dynamic weight is: ; in, The adjustment coefficient input by the user; The dynamic weight is referred to here.

[0010] By employing the above technical solution, the marginal contribution value of each sub-feature is calculated, quantifying its importance to assess its significance. Then, the initial weight is calculated using this marginal contribution value. Different initial weights correspond to different importance of sub-features. Setting the initial weights based on the sub-feature's importance avoids situations where user-defined weights lead to significant deviations in the final prediction results. This allows for weight setting based on the actual situation of the sub-features, reducing computation and improving efficiency. Furthermore, setting weights based on the actual situation of the sub-features also improves the accuracy of the final prediction results. Simultaneously, the calculation of dynamic weights also requires an adjustment coefficient. It is calculated using both the dynamic initial weights and the user-input dynamic adjustment coefficient, ensuring that the dynamic weights not only adapt dynamically but also improve the accuracy of the prediction results. The initial weights are data-driven, while the adjustment coefficients are experience-driven; combining the two further enhances the rationality of the dynamic weights.

[0011] Optionally, the hybrid model architecture includes an association sub-model and an environmental dynamic response sub-model; The associated sub-model is: ; in, For nodes The embedding vector, using the sub-features as nodes. ; Represents a node; Indicates the first Layer stacking; For nodes The set of neighboring nodes; Indicate neighboring nodes; Representing neighboring nodes exist Layer embedding vector; Indicates the first The first learned weight matrix of the layer has known values. Used to normalize the contributions of different neighboring nodes; For activation functions; Indicates a node All neighboring nodes Perform summation.

[0012] By adopting the above technical solution, sub-features are viewed as nodes in a graph, and the relationships between sub-features are represented by edges or adjacencies between nodes, forming a sub-feature association graph. Through a graph neural network, the interaction relationships between features are explicitly modeled, rather than treating all features as independent inputs. Furthermore, by stacking multiple layers, a progression from local to global is achieved, thereby capturing the complex relationships between sub-features. As part of the normalization process, the influence of different neighboring nodes on the central node is balanced, preventing highly connected nodes from dominating information aggregation. The introduction of the activation function introduces non-linear transformation capabilities to graph convolution operations, enabling the associated sub-model to go beyond linear superposition and fit complex, non-linear interactions between sub-features.

[0013] Optionally, the method further includes: Extract time-series data of temperature and humidity during the maintenance period based on all the aforementioned sub-features; The temperature and humidity time-series data includes temperature and humidity ; The environmental dynamic response sub-model is as follows: ; Extracting time-series feature vectors based on the aforementioned environmental dynamic response sub-model ; in, Indicates time; express Temperature at any moment; express Humidity at any given moment; express Moving forward in time Hidden state; express Backwards in time Hidden state; express Moment Hidden state; Indicates forward unit; Indicates backward unit; express The temperature and humidity vectors input at any time are concatenated into a single vector and used as the input vector. express Moving forward in time Hidden state; express Backwards in time Hidden state; This indicates vector concatenation.

[0014] By adopting the above technical solution, the environmental dynamic response sub-model is based on bidirectional... The network models the time-series data of temperature and humidity during concrete curing and extracts time-series feature vectors. This is used to characterize the dynamic impact of environmental conditions changing over time. Compared to raw temperature and humidity values ​​or simple statistics, time-series feature vectors... Including richer and more intelligent dynamic environmental information enables subsequent prediction processes to be more accurate. Furthermore, in actual engineering, the concrete curing environment is difficult to maintain in an ideal state at all times. Therefore, it is necessary to calculate the time-series data of the curing environment and integrate it with the output results of the associated sub-model. The two complement each other to form a more comprehensive prediction system, thereby improving the accuracy of the prediction results.

[0015] Optionally, the step of making predictions based on the dynamic weights and hybrid model architecture and outputting the prediction results includes: embed vector and time series feature vectors All are used as feature vectors ; Each feature sub-vector is weighted based on the dynamic weights to obtain a weighted sub-vector; Calculate the global weight based on all the weighted sub-vectors mentioned above; The fusion subvector is calculated based on the global weights and the weighted subvectors. The prediction result is calculated based on the fused sub-vectors.

[0016] By employing the above technical solution, dynamic weights and feature sub-vectors are weighted to obtain weighted sub-vectors. The calculation of global weights coordinates the contribution relationships between different feature sub-vectors from a global perspective, avoiding local overweighting or information imbalance, and uncovering synergistic effects or compensatory relationships between different feature sub-vectors, thus improving the comprehensiveness, balance, and rationality of the prediction results. The sub-vectors obtained after dynamic weighting and global weight fusion are further combined to generate fused sub-vectors. These fused sub-vectors represent a condensed and unified representation of all key information. The electronic device then calculates the prediction results based on these fused sub-vectors, achieving multi-dimensional, multi-source information fusion for predicting the compressive strength of concrete, with high accuracy and reliability.

[0017] Optionally, the method further includes: The formula for calculating the weighted sub-vector is: ; in, These are weighted subvectors; Dynamic weights; The formula for calculating the global weight is: ; in, Sub-features Attention score; This represents summing the attention scores over all sub-features; The formula for calculating the attention score is as follows: ; in, express transpose; A learnable parameter vector; Preset values ​​entered by the user; Marginal contribution value The result after normalization; The formula for calculating the fused sub-vector is: ; The formula for calculating the prediction result is as follows: ; in, The output weight matrix is ​​learnable; This is a bias term.

[0018] By adopting the above technical solution, and utilizing a joint fusion strategy of dynamic weights, attention mechanisms, marginal contribution guidance, and learnable parameters, refined, adaptive, and intelligent fusion of multi-source heterogeneous features is achieved. Furthermore, the calculation of dynamic weights and attention scores is influenced by marginal contribution values, thus realizing a multi-stage dynamic adjustment mechanism. This multi-stage dynamic adjustment can adapt to the influence degree of sub-features and, using user-input preset values, can also be adjusted based on actual user needs. The calculation of dynamic weights and global weights both utilize marginal contribution values, thus intervening in feature-level weighted calculations and feature fusion stage calculations. Marginal contribution values ​​play an important reference role in weighting mechanisms at different levels and with different purposes, enabling multiple, layered, and systematic adjustments to feature influence. Moreover, dynamic weights are calculated at the feature level for individual weighting, representing local adjustments, while global weights are calculated at the feature fusion level, representing global adjustments. This considers both the importance of individual features and the relative contributions and synergistic effects among multiple features.

[0019] Secondly, the concrete compressive strength prediction device provided in this application adopts the following technical solution: A device for predicting the compressive strength of concrete, comprising: The first acquisition module is used to acquire multi-dimensional feature data; The first processing module is used to preprocess the feature data to obtain normalized data; The first calculation module is used to calculate dynamic weights based on the normalized data; The second acquisition module is used to acquire the hybrid model architecture; The prediction output module is used to make predictions based on the dynamic weights and hybrid model architecture and output the prediction results.

[0020] Thirdly, the electronic device provided in this application adopts the following technical solution: An electronic device includes a processor coupled to a memory; the processor is configured to execute a computer program stored in the memory such that the electronic device performs the method as described in the first aspect.

[0021] Fourthly, the computer-readable storage medium provided in this application adopts the following technical solution: A computer-readable storage medium includes a computer program or instructions that, when executed on a computer, cause the computer to perform the method as described in the first aspect.

[0022] In summary, this application includes at least one of the following beneficial technical effects: The electronic device acquires multi-dimensional feature data, covering various factors affecting the compressive strength of concrete. This comprehensive approach avoids the limitations of relying on single or limited features. The feature data is normalized to eliminate the influence of different dimensions and orders of magnitude for subsequent calculations. Dynamic weights are then calculated based on the normalized data, allowing for dynamic adjustments to better reflect the actual situation and improve the accuracy of subsequent predictions. A hybrid model architecture integrates the advantages of different models, achieving complementary strengths and better handling complex prediction scenarios. Finally, the electronic device performs predictions based on the dynamic weights and the hybrid model architecture, outputting the results. This not only predicts the compressive strength of concrete but also improves the accuracy of the predictions. The environmental dynamic response sub-model is based on a bidirectional LSTM network, which models the time-series data of temperature and humidity during concrete curing and extracts time-series feature vectors. This is used to characterize the dynamic impact of environmental conditions changing over time. Compared to raw temperature and humidity values ​​or simple statistics, time-series feature vectors... Including richer and more intelligent dynamic environmental information enables subsequent prediction processes to be more accurate. Furthermore, in actual engineering, the concrete curing environment is difficult to maintain in an ideal state at all times. Therefore, it is necessary to calculate the time-series data of the curing environment to fuse with the output results of the associated sub-model. The two complement each other to form a more comprehensive prediction system, thereby improving the accuracy of the prediction results. The dynamic weights and feature sub-vectors are weighted to obtain a weighted sub-vector. The calculation of global weights coordinates the contribution relationships between different feature sub-vectors from a global perspective, avoiding local overweighting or information imbalance, and uncovering synergistic effects or compensatory relationships between different feature sub-vectors, thus improving the comprehensiveness, balance, and rationality of the prediction results. The sub-vectors obtained after dynamic weighting and global weight fusion are further combined to generate a fused sub-vector. The fused sub-vector is a condensed and unified representation of all key information. The electronic device then calculates the prediction result based on the fused sub-vector, achieving multi-dimensional, multi-source information fusion for predicting the compressive strength of concrete, with high accuracy and reliability. The calculation of dynamic weights and attention scores are both influenced by marginal contribution values, thus implementing a multi-dynamic adjustment mechanism. This multi-dynamic adjustment can adapt to the influence degree of sub-features, and can also be adjusted based on actual user needs using preset values ​​input by the user. The calculation of dynamic weights and global weights both utilize marginal contribution values, thus intervening in feature-level weighted calculations and feature fusion stage calculations. Marginal contribution values ​​play an important reference role in weighting mechanisms that are not hierarchical or purposeful, enabling multiple, hierarchical, and systematic adjustments to the influence of features. Attached Figure Description

[0023] Figure 1 This is a flowchart of the concrete compressive strength prediction method according to an embodiment of this application.

[0024] Figure 2 This is a block diagram of the concrete compressive strength prediction device according to an embodiment of this application.

[0025] Figure 3 This is a block diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0026] This specific embodiment is merely an explanation of this application and is not intended to limit it. Users skilled in the art can make modifications to this embodiment without contributing any inventive step after reading this specification, but such modifications are protected by patent law as long as they fall within the scope of the claims of this application.

[0027] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by users of ordinary skills in the art without creative effort are within the scope of protection of this application.

[0028] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0029] This application discloses a method for predicting the compressive strength of concrete. This method can be executed by an electronic device. The electronic device can be a server or a terminal device. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The terminal device can be a smartphone, tablet, desktop computer, etc., but is not limited to these.

[0030] This application discloses a method for predicting the compressive strength of concrete. (Refer to...) Figure 1 The main process of a method for predicting the compressive strength of concrete is described below (S100~S500): Step S100: Obtain multi-dimensional feature data; Step S200: Preprocess the feature data to obtain normalized data; Step S300: Calculate dynamic weights based on normalized data; Step S400: Obtain the hybrid model architecture; Step S500: Make predictions based on dynamic weights and a hybrid model architecture, and output the prediction results.

[0031] Electronic devices acquire multi-dimensional feature data, including cement, aggregates, admixtures, etc., and then preprocess the multi-dimensional feature data to obtain normalized data, enabling subsequent calculations. The electronic devices calculate dynamic weights based on the normalized data, and the dynamic weights can be automatically and dynamically adjusted according to the actual situation to make subsequent calculations more consistent with the actual situation. The electronic devices make predictions based on dynamic weights and a hybrid model architecture and output prediction results. The hybrid model architecture can integrate different variables to achieve more comprehensive data and scenario coverage, thereby obtaining more accurate prediction results.

[0032] Specifically, the feature data is preprocessed to obtain normalized data, including: extracting sub-features based on the feature data; normalizing the sub-features to obtain normalized data; The formula for calculating normalized data is: ; in, The minimum value in the dataset of sub-features; The maximum value in the dataset for the sub-feature; This represents the current collected value of the sub-feature; For normalized data; Before normalizing the sub-features and obtaining normalized data, the process further includes: removing all sub-features; the removal process includes: extracting any two sub-features from all sub-features and using the two extracted sub-features as the first sub-feature and the second sub-feature, respectively; calculating the correlation coefficient between the first sub-feature and the second sub-feature; determining whether the correlation coefficient is greater than a first preset threshold; if the correlation coefficient is greater than the first preset threshold, then removing either the first sub-feature or the second sub-feature. The formula for calculating the correlation coefficient is: ; in, The correlation coefficient; The first sub-feature A number; The average value of the first sub-feature; The second sub-feature A number; This is the average value of the second sub-feature.

[0033] Electronic equipment preprocesses feature data, specifically including extracting sub-features based on feature data. Sub-features of cement include parameters such as water-cement ratio and mineral phase composition. Sub-features of aggregates include particle size distribution, crushing value, and surface water absorption rate. Sub-features of admixtures include water-reducing agent type and dosage. The above sub-features are for illustrative purposes only, and are not limited to the above parameters in actual calculation and prediction.

[0034] Sub-features are normalized using a formula for normalizing data. Before normalization, sub-features need to be eliminated, as excessive sub-features may be redundant. Redundancy is determined by calculating the correlation coefficient between any two sub-features using a formula. The electronic device then checks if the correlation coefficient exceeds a first preset threshold. If it does, the two sub-features are considered overly correlated, and one of them is eliminated. Eliminating redundant features reduces unnecessary calculations and the impact of redundancy on prediction results, thus improving the accuracy of predictions.

[0035] Specifically, the dynamic weights are calculated based on normalized data, including: calculating the marginal contribution value of sub-features; calculating initial weights based on the marginal contribution value; obtaining the adjustment coefficient input by the user; and calculating the dynamic weights based on the initial weights and the adjustment coefficient. The formula for calculating the marginal contribution value is: ; in, It is the set of all sub-features; For features that do not include sub-features A subset of; For input subset only The response output value at that time; For sub-features The marginal contribution value; The formula for calculating the initial weights is: ; in, This represents the normalization function, used to normalize... Mapped to values ​​within a reasonable range; The formula for calculating dynamic weights is: ; in, The adjustment coefficient input by the user; For dynamic weights.

[0036] In calculating dynamic weights, the initial weights are obtained using marginal contribution values, which reflect the importance of sub-features. Therefore, using these values ​​to calculate the initial weights is more consistent with reality. Furthermore, initial weights can be automatically calculated for sub-features of different importance, meaning the initial weights dynamically change based on these sub-features. In addition to being related to the initial features, the dynamic weights are also adjusted based on user-inputted adjustment coefficients. This dual adjustment, not only for different sub-features but also based on the user's actual processing, makes the prediction results more accurate in subsequent calculations and predictions.

[0037] Specifically, the hybrid model architecture includes a relational sub-model and an environmental dynamic response sub-model; The associated sub-model is: ; in, For nodes The embedding vector, with sub-features as nodes ; Represents a node; Indicates the first Layer stacking; For nodes The set of neighboring nodes; Indicate neighboring nodes; Representing neighboring nodes exist Layer embedding vector; Indicates the first The first learned weight matrix of the layer has known values. Used to normalize the contributions of different neighboring nodes; For activation functions; Indicates a node All neighboring nodes Perform summation.

[0038] For activation function ReLU, Sigmoid, Tanh, etc. can be used; ReLU is used in this embodiment. In this embodiment, it represents the reciprocal of the number of neighboring nodes, that is... The number of neighboring nodes is denoted as . The initial state of the first learning weight is randomly initialized, and it is continuously updated iteratively through backpropagation and optimization algorithms to complete the learning process. The backpropagation and optimization algorithms will not be described in detail in this embodiment, as existing technologies can be used, and this embodiment is not limited to them.

[0039] Graph neural networks (GNNs) represent sub-features with nodes and analyze the interactions between nodes. They can effectively capture microscopic structural information. Through message passing mechanisms, each node can aggregate information from its neighbors, combining this information with the current node's data. This allows for a comprehensive consideration of the local environment's influence on the nodes, more accurately describing the material's microscopic properties and linking microscopic structural information with macroscopic performance. After multi-layer GNN computation and updates, the resulting embedding vectors contain rich microscopic structural information, which can be correlated with the material's macroscopic performance, providing a foundation for predicting macroscopic properties. Furthermore, the calculation formula utilizes a weighted summation transformation of neighboring nodes using the first learned weight matrix, automatically extracting important features and improving the efficiency and accuracy of feature extraction.

[0040] Specifically, this also includes: extracting time-series temperature and humidity data during the maintenance period based on all sub-features; the time-series temperature and humidity data includes temperature and humidity ; The environmental dynamic response sub-model is as follows: ; Extracting time-series feature vectors based on environmental dynamic response sub-model ; in, Indicates time; express Temperature at any moment; express Humidity at any given moment; express Moving forward in time Hidden state; express Backwards in time Hidden state; express Moment Hidden state; Indicates forward unit; Indicates backward unit; express The temperature and humidity vectors input at any time are concatenated into a single vector and used as the input vector. express Moving forward in time Hidden state; express Backwards in time Hidden state; This indicates vector concatenation.

[0041] Temperature and humidity are concatenated to obtain a temperature and humidity vector, which serves as the input to the environmental dynamic response sub-model.

[0042] During maintenance and use, environmental factors such as temperature and humidity change over time. Furthermore, the current environmental conditions are not only related to historical conditions but may also be influenced by future trends. Therefore, by looking forward... Unit and backward A single unit can simultaneously consider information from both preceding and following time series, capturing more comprehensively the dynamic changes of environmental factors. Then, the vectors corresponding to two units are concatenated to obtain a comprehensive result. Hidden states fully utilize the preceding and following information in the time series, improving the ability to model dynamic environmental changes. Subsequently, electronic devices extract time-series feature vectors based on the environmental dynamic response sub-model. , used with embedding vector Perform fusion computing.

[0043] Specifically, predictions are made and output based on dynamic weights and a hybrid model architecture, including: embedding vectors... and time series feature vectors All are used as feature vectors The system calculates a weighted subvector by weighting each feature subvector based on dynamic weights; calculates the global weight based on all weighted subvectors; calculates the fused subvector based on the global weight and the weighted subvectors; and calculates the prediction result based on the fused subvector.

[0044] Specifically, it also includes: The formula for calculating the weighted subvector is: ; in, These are weighted subvectors; Dynamic weights; The formula for calculating the global weight is: ; in, Sub-features Attention score; This represents summing the attention scores over all sub-features; The formula for calculating attention score is: ; in, express transpose; A learnable parameter vector; Preset values ​​entered by the user; Marginal contribution value The result after normalization; The formula for calculating the fused sub-vector is: ; The formula for calculating the prediction result is: ; in, The output weight matrix is ​​learnable; This is a bias term.

[0045] Learnable output weight matrix Initially, the values ​​are random and require training and optimization to continuously adjust them, thereby minimizing the error of the prediction results. This is achieved through a learnable output weight matrix. The training optimization will not be elaborated here. This embodiment does not apply to the learnable output weight matrix. With limitations, existing technologies can be used. The learnable output weight matrix... Alternatively, a user-preset method can be used. Even with a fixed weight matrix, the prediction result can still be calculated.

[0046] Dynamic weights are used to weight the feature vectors, resulting in weighted vectors that better reflect reality and improve prediction accuracy. Next, global weights are calculated, and a fused vector is obtained using these global weights and the weighted vectors. Finally, the prediction result is calculated using the prediction formula and output. Before applying global weights, an attention score is calculated. This attention score calculation incorporates marginal contribution values ​​in addition to the weighted vectors. The normalized values ​​and preset values, that is, the calculation process of the prediction results, adopt multiple dynamic adjustment methods, fully consider the influence of highly important sub-features, and also take into account the user's active adjustment. It can not only predict the compressive strength of concrete, but also greatly improve the accuracy of the prediction results.

[0047] Figure 2A structural block diagram of a concrete compressive strength prediction device 600 provided in this application embodiment is shown below. Figure 2 As shown, the concrete compressive strength prediction device 600 includes: The first acquisition module 601 is used to acquire multi-dimensional feature data; The first processing module 602 is used to preprocess the feature data to obtain normalized data; The first calculation module 603 is used to calculate dynamic weights based on normalized data; The second acquisition module 604 is used to acquire the hybrid model architecture; The prediction output module 605 is used to make predictions based on dynamic weights and a hybrid model architecture and output the prediction results.

[0048] Specifically, the first processing module 602 includes: The first extraction submodule is used to extract sub-features based on feature data; The first processing submodule is used to normalize the sub-features and obtain normalized data; the formula for calculating the normalized data is: ;in, The minimum value in the dataset of sub-features; The maximum value in the dataset for the sub-feature; This represents the current collected value of the sub-feature; For normalized data; The first elimination submodule is used to eliminate all sub-features before normalizing the sub-features and obtaining normalized data. The first elimination submodule includes: The second extraction submodule a is used to extract any two sub-features from all sub-features, and to use the two extracted sub-features as the first sub-feature and the second sub-feature, respectively. The second calculation submodule a is used to calculate the correlation coefficient between the first sub-feature and the second sub-feature; The first judgment submodule a is used to determine whether the correlation coefficient is greater than the first preset threshold; if the correlation coefficient is greater than the first preset threshold, the first sub-feature or the second sub-feature is removed. The formula for calculating the correlation coefficient is: ; in, The correlation coefficient; The first sub-feature A number; The average value of the first sub-feature; The second sub-feature A number; This is the average value of the second sub-feature.

[0049] Specifically, the first computing module 603 includes: The third calculation submodule b is used to calculate the marginal contribution value of the sub-feature; The fourth calculation submodule b is used to calculate the initial weights based on the marginal contribution value; The third acquisition submodule b is used to acquire the adjustment coefficient input by the user; The fifth calculation submodule b is used to calculate the dynamic weights based on the initial weights and adjustment coefficients; The formula for calculating the marginal contribution value is: ; in, It is the set of all sub-features; For features that do not include sub-features A subset of; For input subset only The response output value at that time; For sub-features The marginal contribution value; The formula for calculating the initial weights is: ; in, This represents the normalization function, used to normalize... Mapped to values ​​within a reasonable range; The formula for calculating dynamic weights is: ; in, The adjustment coefficient input by the user; For dynamic weights.

[0050] Specifically, the second acquisition module 604 includes: The hybrid model architecture includes a relational sub-model and an environmental dynamic response sub-model; The associated sub-model is: ; in, For nodes The embedding vector, with sub-features as nodes ; Represents a node; Indicates the first Layer stacking; For nodes The set of neighboring nodes; Indicate neighboring nodes; Representing neighboring nodes exist Layer embedding vector; Indicates the first The first learned weight matrix of the layer has known values. Used to normalize the contributions of different neighboring nodes; For activation functions; Indicates a node All neighboring nodes Perform summation.

[0051] In this optional embodiment, the concrete compressive strength prediction device 600 further includes: The third extraction submodule is used to extract time-series data of temperature and humidity during the maintenance period based on all sub-features; the time-series data of temperature and humidity includes temperature and humidity ; The environmental dynamic response sub-model is as follows: ; Extracting time-series feature vectors based on environmental dynamic response sub-model ; in, Indicates time; express Temperature at any moment; express Humidity at any given moment; express Moving forward in time Hidden state; express Backwards in time Hidden state; express Moment Hidden state; Indicates forward unit; Indicates backward unit; express The temperature and humidity vectors input at any time are concatenated into a single vector and used as the input vector. express Moving forward in time Hidden state; express Backwards in time Hidden state; This indicates vector concatenation.

[0052] Specifically, the prediction output module 605 includes: The sixth computational submodule is used to process the embedding vector. and time series feature vectors All are used as feature vectors Each feature sub-vector is weighted based on dynamic weights to obtain a weighted sub-vector. The seventh calculation submodule is used to calculate the global weight based on all weighted subvectors; The eighth calculation submodule is used to calculate the fused subvector based on the global weights and weighted subvectors; The ninth calculation submodule is used to calculate the prediction results based on the fused subvectors.

[0053] In this optional embodiment, the concrete compressive strength prediction device 600 further includes: The formula for calculating the weighted subvector is: ; in, These are weighted subvectors; Dynamic weights; The formula for calculating the global weight is: ; in, Sub-features Attention score; This represents summing the attention scores over all sub-features; The formula for calculating attention score is: ; in, express transpose; A learnable parameter vector; Preset values ​​entered by the user; Marginal contribution value The result after normalization; The formula for calculating the fused sub-vector is: ; The formula for calculating the prediction result is: ; in, The output weight matrix is ​​learnable; This is a bias term.

[0054] Figure 3 This is a structural block diagram of an electronic device 700 provided in an embodiment of this application. The electronic device 700 can be a mobile phone, tablet computer, PC, server, or other similar device. Figure 3 As shown, the electronic device 700 includes a memory 701, a processor 702, and a communication bus 703; the memory and the processor 702 are connected via the communication bus 703. The memory 701 stores a computer program that can be loaded by the processor 702 and executed as described in the above embodiments for predicting the compressive strength of concrete.

[0055] The memory 701 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 701 may include a stored program area and a stored managed data area. The stored program area may store instructions for implementing an operating system, instructions for at least one function, and instructions for implementing the concrete compressive strength prediction method provided in the above embodiments, etc. The stored managed data area may store managed data involved in the concrete compressive strength prediction method provided in the above embodiments, etc.

[0056] Processor 702 may include one or more processing cores. Processor 702 executes instructions, programs, code sets, or instruction sets stored in memory 701, and calls managed data stored in memory 701 to perform various functions of this application and process managed data. Processor 702 may be at least one of the following: Application-Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), controller, microcontroller, and microprocessor. It is understood that, for different devices, the electronic devices used to implement the functions of processor 702 may also be other types, and this application embodiment does not specifically limit the specific devices used.

[0057] The communication bus 703 may include a path for transmitting information between the aforementioned components. The communication bus 703 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The communication bus 703 can be divided into an address bus, a managed data bus, a control bus, etc. For ease of representation, Figure 3 The symbol is represented by a single double arrow, but this does not mean that there is only one bus or one type of bus.

[0058] This application provides a computer storage medium storing a computer program that can be loaded by a processor and executed as described in the above embodiments for predicting the compressive strength of concrete.

[0059] In this embodiment, the computer storage medium can be a tangible device that holds and stores instructions used by the instruction execution device. The computer storage medium can be, but is not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof. Specifically, the computer storage medium can be a portable computer disk, a hard disk, a USB flash drive, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), speaker random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory stick, floppy disk, optical disk, magnetic disk, mechanical encoding device, or any combination thereof.

[0060] The terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

Claims

1. A method for predicting the compressive strength of concrete, characterized in that, include: Obtain multi-dimensional feature data; The feature data is preprocessed to obtain normalized data; Calculate dynamic weights based on the normalized data; Obtain the hybrid model architecture; The prediction is made and the prediction results are output based on the dynamic weights and hybrid model architecture.

2. The method for predicting the compressive strength of concrete according to claim 1, characterized in that, The feature data is preprocessed to obtain normalized data, including: Sub-features are extracted based on the aforementioned feature data; The sub-features are normalized to obtain normalized data; The formula for calculating the normalized data is: ; in, The minimum value in the dataset of the sub-feature; The maximum value in the dataset for the sub-feature; The current collected value of the sub-feature; For normalized data; Before normalizing the sub-features and obtaining normalized data, the process also includes: All of the aforementioned sub-features are removed. The rejection process includes: Extract any two sub-features from all the sub-features, and use the two extracted sub-features as the first sub-feature and the second sub-feature, respectively. Calculate the correlation coefficient between the first sub-feature and the second sub-feature; Determine whether the correlation coefficient is greater than a first preset threshold; If the correlation coefficient is greater than the first preset threshold, then the first sub-feature or the second sub-feature is removed; The formula for calculating the correlation coefficient is: ; in, The correlation coefficient is mentioned above; For the first sub-feature A number; This is the average value of the first sub-feature; The second sub-feature A number; This is the average value of the second sub-feature.

3. The method for predicting the compressive strength of concrete according to claim 2, characterized in that, The calculation of dynamic weights based on the normalized data includes: Calculate the marginal contribution value of the sub-feature; Calculate the initial weights based on the marginal contribution values; Obtain the adjustment coefficient input by the user; The dynamic weights are calculated based on the initial weights and the adjustment coefficients. The formula for calculating the marginal contribution value is as follows: ; in, The set of all the aforementioned sub-features; To not include the sub-feature A subset of; For input subset only The response output value at that time; For the sub-feature The marginal contribution value; The formula for calculating the initial weights is: ; in, This represents the normalization function, used to normalize... Mapped to values ​​within a reasonable range; The formula for calculating the dynamic weight is: ; in, The adjustment coefficient input by the user; The dynamic weight is referred to here.

4. A method for predicting the compressive strength of concrete according to claim 1 or 3, characterized in that, The hybrid model architecture includes a correlation sub-model and an environmental dynamic response sub-model; The associated sub-model is: ; in, For nodes The embedding vector, using the sub-features as nodes. ; Represents a node; Indicates the first Layer stacking; For nodes The set of neighboring nodes; Indicate neighboring nodes; Representing neighboring nodes exist Layer embedding vector; Indicates the first The first learned weight matrix of the layer has known values. Used to normalize the contributions of different neighboring nodes; For activation functions; Indicates a node All neighboring nodes Perform summation.

5. The method for predicting the compressive strength of concrete according to claim 4, characterized in that, The method further includes: Extract time-series data of temperature and humidity during the maintenance period based on all the aforementioned sub-features; The temperature and humidity time-series data includes temperature and humidity ; The environmental dynamic response sub-model is as follows: ; Extracting time-series feature vectors based on the aforementioned environmental dynamic response sub-model ; in, Indicates time; express Temperature at any moment; express Humidity at any given moment; express Moving forward in time Hidden state; express Backwards in time Hidden state; express Moment Hidden state; Indicates forward unit; Indicates backward unit; express The temperature and humidity vectors input at any time are concatenated into a single vector and used as the input vector. express Moving forward in time Hidden state; express Backwards in time Hidden state; This indicates vector concatenation.

6. The method for predicting the compressive strength of concrete according to claim 5, characterized in that, The prediction based on the dynamic weights and hybrid model architecture, and the output of the prediction results, include: embed vector and time series feature vectors All are used as feature vectors ; Each feature sub-vector is weighted based on the dynamic weights to obtain a weighted sub-vector; Calculate the global weight based on all the weighted sub-vectors mentioned above; The fusion subvector is calculated based on the global weights and the weighted subvectors. The prediction result is calculated based on the fused sub-vectors.

7. The method for predicting the compressive strength of concrete according to claim 6, characterized in that, The method further includes: The formula for calculating the weighted sub-vector is: ; in, These are weighted subvectors; Dynamic weights; The formula for calculating the global weight is: ; in, Sub-features Attention score; This represents summing the attention scores over all sub-features; The formula for calculating the attention score is as follows: ; in, express transpose; A learnable parameter vector; Preset values ​​entered by the user; Marginal contribution value The result after normalization; The formula for calculating the fused sub-vector is: ; The formula for calculating the prediction result is as follows: ; in, The output weight matrix is ​​learnable; This is a bias term.

8. A device for predicting the compressive strength of concrete, characterized in that, include: The first acquisition module is used to acquire multi-dimensional feature data; The first processing module is used to preprocess the feature data to obtain normalized data; The first calculation module is used to calculate dynamic weights based on the normalized data; The second acquisition module is used to acquire the hybrid model architecture; The prediction output module is used to make predictions based on the dynamic weights and hybrid model architecture and output the prediction results.

9. An electronic device, characterized in that, The device includes a processor coupled to a memory; the processor is configured to execute a computer program stored in the memory to cause the electronic device to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It includes a computer program or instructions that, when run on a computer, cause the computer to perform the method as described in any one of claims 1 to 7.

Images