Intelligent Prediction Method and System for Recycled Aggregate Concrete Integrating Machine Learning

Abstract

This invention discloses an intelligent prediction method and system for the strength of recycled aggregate concrete integrating machine learning, belonging to the field of concrete strength prediction technology. The method includes: using recycled aggregate concrete as a guide, collecting a sample training set to train a machine learning model and constructing a concrete strength prediction channel; during the mix proportion optimization process of recycled aggregate concrete, reading the initial mix parameter set and the current optimization progress of the current optimization stage; performing strength prediction complexity analysis based on the sample training set and the initial mix parameter set to output a first prediction complexity coefficient; performing strength prediction complexity analysis based on the current optimization progress to output a second prediction complexity coefficient; formulating an adaptive strength prediction mechanism based on the first and second prediction complexity coefficients, activating the concrete strength prediction channel, performing strength prediction based on the initial mix parameter set, and outputting the concrete strength prediction result. This invention effectively improves the prediction accuracy of the strength of recycled aggregate concrete.

Description

Technical Field

[0001] This invention relates to the field of concrete strength prediction technology, specifically to an intelligent prediction method and system for the strength of recycled aggregate concrete that integrates machine learning. Background Technology

[0002] With the accelerated recycling of construction waste and the booming development of the green building industry, recycled aggregate concrete, with its advantages of energy conservation, emission reduction, and resource recycling, has become a key material promoted in the construction engineering field. As a critical performance indicator, the accurate prediction of concrete strength is crucial for project quality control. Existing technologies are gradually adopting machine learning models combined with sample data to predict concrete strength, replacing traditional experimental methods with data-driven models to achieve rapid strength estimation.

[0003] However, existing machine learning prediction methods mostly adopt fixed prediction mechanisms and lack targeted analysis of prediction complexity. This results in the prediction strategy being unable to be dynamically adjusted according to the actual scenario, leading to problems such as low prediction accuracy and insufficient scenario adaptability. Consequently, it is difficult to meet the dynamic and accurate requirements for intensity prediction in the process of mix ratio optimization. Summary of the Invention

[0004] This invention provides a method and system for intelligent prediction of the strength of recycled aggregate concrete that integrates machine learning, aiming to solve the technical problem of insufficient accuracy in the prediction of the strength of recycled aggregate concrete in the prior art.

[0005] In view of the above problems, the present invention provides a method and system for intelligent prediction of the strength of recycled aggregate concrete that integrates machine learning.

[0006] In a first aspect, the present invention provides an intelligent prediction method for the strength of recycled aggregate concrete that integrates machine learning, including:

[0007] Using recycled aggregate concrete as a guide, a sample training set is collected to train a machine learning model and construct a concrete strength prediction channel.

[0008] During the mix proportion optimization process of recycled aggregate concrete, the initial mix parameter set and the current optimization process of the current optimization stage are read.

[0009] Based on the sample training set and the initial coordination parameter set, an intensity prediction complexity analysis is performed to output the first prediction complexity coefficient;

[0010] Based on the current optimization process, perform intensity prediction complexity analysis and output a second prediction complexity coefficient;

[0011] An adaptive strength prediction mechanism is established based on the first and second prediction complexity coefficients, and the concrete strength prediction channel is activated. Strength prediction is performed based on the initial set of matching parameters, and the concrete strength prediction result is output.

[0012] Secondly, this invention provides an intelligent prediction system for the strength of recycled aggregate concrete that integrates machine learning, comprising:

[0013] The prediction channel construction module is used to collect a sample training set to train a machine learning model and construct a concrete strength prediction channel, guided by recycled aggregate concrete.

[0014] The optimization parameter reading module is used to read the initial mix parameter set and the current optimization process in the current optimization stage during the mix proportion optimization process of recycled aggregate concrete.

[0015] The first coefficient analysis module is used to perform intensity prediction complexity analysis based on the sample training set and the initial matching parameter set, and output the first prediction complexity coefficient.

[0016] The second coefficient analysis module is used to perform intensity prediction complexity analysis based on the current optimization process and output a second prediction complexity coefficient.

[0017] The strength prediction output module is used to formulate an adaptive strength prediction mechanism based on the first prediction complexity coefficient and the second prediction complexity coefficient, activate the concrete strength prediction channel, perform strength prediction based on the initial mix parameter group, and output the concrete strength prediction result.

[0018] One or more technical solutions provided in this invention have at least the following technical effects or advantages:

[0019] This invention provides a method and system for intelligent prediction of the strength of recycled aggregate concrete that integrates machine learning. First, by constructing a machine learning prediction channel, a foundation for high-precision intelligent prediction is laid. Second, during the optimization process, the mix design parameters and optimization progress are read in real time, ensuring close synchronization between the prediction and the current design state. Furthermore, by performing complexity analysis on data characteristics and optimization stages respectively, a dynamic quantitative assessment of the difficulty of the prediction task is achieved. Based on this, an adaptive prediction mechanism is formulated according to dual coefficients, enabling the system to dynamically select or adjust the prediction model strategy according to specific scenarios, realizing intelligent allocation of prediction resources. Finally, the prediction channel output results are activated, thereby dynamically balancing the prediction speed and accuracy in an adaptive manner throughout the entire process of recycled aggregate concrete mix design optimization, effectively improving the reliability, stability, and overall optimization efficiency of strength prediction. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart illustrating the intelligent prediction method for the strength of recycled aggregate concrete that integrates machine learning, provided in an embodiment of the present invention.

[0022] Figure 2 A schematic diagram of the structure of the intelligent prediction system for the strength of recycled aggregate concrete that integrates machine learning, provided in an embodiment of the present invention.

[0023] The components represented by each number in the attached diagram are explained below:

[0024] The module includes a prediction channel construction module 11, an optimization parameter reading module 12, a first coefficient analysis module 13, a second coefficient analysis module 14, and an intensity prediction output module 15. Detailed Implementation

[0025] This invention provides a method and system for intelligent prediction of the strength of recycled aggregate concrete that integrates machine learning, in order to address the technical problem of insufficient accuracy in the prediction of the strength of recycled aggregate concrete in the prior art.

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0027] It should be noted that the terms "comprising" and "having" are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to these processes, methods, products, or devices.

[0028] Example 1, as Figure 1 As shown, this invention provides an intelligent prediction method for the strength of recycled aggregate concrete that integrates machine learning. The method includes:

[0029] S100: Guided by recycled aggregate concrete, a sample training set is collected to train a machine learning model and construct a concrete strength prediction channel.

[0030] In this embodiment of the invention, recycled aggregate concrete is used as a guide to collect a sample training set to train a machine learning model and construct a concrete strength prediction channel. The strength of recycled aggregate concrete is affected by multiple interacting factors, and its mix parameters and final strength exhibit a highly nonlinear and coupled relationship. Traditional empirical formulas or single models are difficult to stably and accurately characterize this complex mapping. To ensure the robustness and generalization ability of the prediction model and avoid overfitting or underfitting due to the randomness of training data partitioning, this step adopts the idea of ​​ensemble learning. Multiple base models are trained through cross-validation to construct an aggregate prediction channel, thereby learning the inherent laws of the data more comprehensively and improving the stability and accuracy of the prediction.

[0031] Step S100 in the method provided in this embodiment of the invention includes:

[0032] Guided by recycled aggregate concrete, information retrieval is carried out based on building big data, a sample mix parameter scheme set is collected, and the historical concrete test strength of different sample mix parameter schemes is obtained as the sample concrete strength, thus obtaining the sample concrete strength set.

[0033] The sample combination parameter scheme set and the sample concrete strength set are used as the sample training set, and the sample training set is divided into J-fold cross partitions to obtain J sample training data groups, where J is an integer greater than or equal to 10.

[0034] Using the sample matching parameter scheme as input data and the sample concrete strength as supervision data, feedforward neural networks are trained to convergence using the J sample training data groups to generate J concrete strength prediction units, which are then combined to obtain a concrete strength prediction channel.

[0035] The sample mix design includes several mix parameters, which include at least the following: cement dosage, type of mineral admixture, dosage of mineral admixture, replacement rate of recycled coarse aggregate, replacement rate of recycled fine aggregate, total aggregate volume, key properties of recycled aggregate, water-cement ratio, type of admixture, and dosage of admixture.

[0036] First, guided by recycled aggregate concrete, information retrieval is conducted based on construction big data to collect a set of sample mix design parameters. Historical concrete test strengths for different sample mix design parameters are then obtained as the sample concrete strengths, resulting in a set of sample concrete strengths. Each sample mix design parameter includes several mix parameters, which at least include cement content, mineral admixture type, mineral admixture content, recycled coarse aggregate replacement rate, recycled fine aggregate replacement rate, total aggregate volume, key properties of recycled aggregate, water-cement ratio, admixture type, and admixture dosage. Construction big data refers to datasets encompassing multi-dimensional information such as material formulations, test data, and engineering application records in construction projects. The sample mix design parameter set is a collection of multiple sets of recycled aggregate concrete mix parameters, each set corresponding to a specific concrete formulation. In the mix design parameters, cement content refers to the mass of cement in a single cubic meter of concrete; mineral admixture type and dosage refer to the types of materials such as fly ash and slag powder, and their proportion in the cementitious materials; recycled coarse aggregate replacement rate refers to the percentage of recycled coarse aggregate in the total mass of coarse aggregate; recycled fine aggregate replacement rate refers to the percentage of recycled fine aggregate in the total mass of fine aggregate; total aggregate volume refers to the absolute volume of aggregate in concrete; key properties of recycled aggregate include apparent density, water absorption rate, crushing index, and other core characteristic parameters that distinguish it from natural aggregate; water-cement ratio refers to the mass ratio of water to the encapsulated material (cement + admixtures), which is one of the most critical parameters affecting strength; admixture type and dosage refer to materials such as water-reducing agents and air-entraining agents, and their dosage, usually expressed as a percentage of the mass of cementitious materials. The sample concrete strength set refers to the concrete cube compressive strength data measured by physical tests, corresponding one-to-one with the sample mix design parameter scheme.

[0037] For example, a search of a construction engineering database yielded one sample set of parameters: cement usage: 300 Mineral admixtures: 20% fly ash; Recycled coarse aggregate replacement rate: 50%; Recycled fine aggregate replacement rate: 0%; Total aggregate volume: 0.68 Key properties of recycled aggregate: water absorption rate 3.5%; water-cement ratio: 0.45; admixture: 1% water-reducing agent. Historical test strength of this scheme was also extracted: 40 MPa. This process was repeated to collect 1000 sets of similar schemes and corresponding strengths, forming a sample mix parameter scheme set and a sample concrete strength set.

[0038] Secondly, the sample mix design set and the sample concrete strength set are used as the sample training set. The sample training set is then subjected to J-fold cross-validation to obtain J sample training data groups, where J is an integer greater than or equal to 10. J-fold cross-validation is a statistical method for evaluating the performance of machine learning models. It aims to reduce the impact of the randomness of dataset partitioning on model evaluation by training and testing the model multiple times. The core idea is to divide the dataset into J subsets, then use J-1 subsets for training, and the remaining subset for testing, repeating this process J times. Finally, all test results are combined to evaluate the model's performance. The sample mix design set and the sample concrete strength set are combined into a sample training set, and then randomly divided into J mutually exclusive and approximately similar subsets using J-fold cross-validation to obtain J sample training data groups. For example, taking J=10, for the aforementioned 1000 sample training sets, they are randomly divided into 10 sample training data groups, each containing 100 mix design schemes and their corresponding strengths.

[0039] Further, using the sample mix design parameters as input data and the sample concrete strength as supervisory data, the feedforward neural network is trained to convergence using the J sample training data sets, generating J concrete strength prediction units, which are then combined to obtain a concrete strength prediction channel. A feedforward neural network (FNN) is an artificial neural network with neurons arranged in layers. Information is transmitted unidirectionally from the input layer to the output layer, passing through several hidden layers. It can automatically learn the relationships between high-dimensional complex features through multiple nonlinear transformations, making it suitable for fitting the nonlinear mapping between concrete mix proportions and strength. Convergence means that the model loss function drops to a preset small threshold, indicating that the prediction deviation meets the requirements. The concrete strength prediction channel is a set of J independent prediction units, which can be called for prediction as needed. Using the mix design parameters of each sample training data set as input and the corresponding strength as supervisory data, the feedforward neural network is trained separately until the loss function is lower than a preset threshold, such as a mean squared error ≤ 0.001. At this point, convergence is determined, resulting in J concrete strength prediction units; these J units are then combined to form the concrete strength prediction channel.

[0040] For example, 10 feedforward neural networks are constructed. The number of neurons in the input layer equals the number of matching parameters for a single sample, corresponding to 8 parameters, with each neuron corresponding to one specific matching parameter. The output layer has only one neuron, corresponding to the concrete test strength of the sample. Taking the first set of training data as an example: Input: 100 matching parameters for the first set; Supervision data: 100 strengths; Training: Set 2 hidden layers, each with 16 neurons, and train the feedforward neural network until the mean squared error drops to 0.0008, satisfying the convergence threshold, to obtain prediction unit 1. Similarly, use the data from sets 2-10 to train the feedforward neural network, obtaining prediction units 2-10. Combining these 10 prediction units yields the concrete strength prediction channel.

[0041] In this embodiment of the invention, the diversity and coverage of the training set are ensured by collecting samples through building big data; J-fold cross-partitioning and multi-unit training improve the generalization ability of the model, and the prediction channel composed of multiple units enhances the reliability of subsequent predictions; the prediction channel finally constructed provides an accurate and efficient tool for intensity prediction in mix ratio optimization, replacing traditional physical experiments and effectively reducing time and economic costs.

[0042] S200: During the mix proportion optimization process of recycled aggregate concrete, read the initial mix parameter set and the current optimization process of the current optimization stage.

[0043] In this embodiment of the invention, during the mix proportion optimization process of recycled aggregate concrete, the initial mix parameter set and the current optimization progress of the current optimization stage are read. Mix proportion optimization is a dynamic process of iteratively adjusting the parameters of recycled aggregate concrete to obtain the optimal strength. The differences in initial parameters and the speed of optimization progress at different optimization stages will affect the accuracy requirements and strategy adaptability of subsequent strength prediction. If the parameter state and process characteristics of the current optimization are ignored, subsequent predictions are prone to overfitting early parameters or lagging adaptation. Therefore, it is necessary to first read the initial mix parameters and progress indicators of the current stage to provide a basis for the subsequent dynamic prediction mechanism.

[0044] Step S200 in the method provided in this embodiment of the invention includes:

[0045] Specifically, the current optimization process is defined as the ratio of the current iteration optimization count to the average predicted intensity improvement within the sliding time window. The current optimization count ratio is the ratio of the current iteration optimization count to the preset total iteration optimization count. The sliding time window is the most recent K consecutive iterations of optimization, where K is an integer greater than or equal to 5 and less than 30.

[0046] First, during the mix design optimization of recycled aggregate concrete, the initial mix design parameter set for the current optimization stage is retrieved. The mix design optimization process refers to iteratively adjusting the mix design parameters of recycled aggregate concrete through an algorithm to achieve optimal strength, cost, and other objectives. The current optimization stage refers to a single, independent iteration cycle of the mix design optimization algorithm. The initial mix design parameter set refers to a complete set of recycled aggregate concrete mix design parameters used at the start of the current optimization stage. Within the current iteration cycle of the mix design optimization algorithm, the combination of recycled aggregate concrete mix design parameters at the start of that cycle is extracted. For example, assuming a particle swarm optimization algorithm is used for mix design optimization, and the current optimization stage is the 15th iteration, the initial mix design parameter set for this stage is: Cement content: 310... Mineral admixtures: fly ash 15%; Recycled coarse aggregate replacement rate: 60%; Recycled fine aggregate replacement rate: 10%; Total aggregate volume: 0.67 Key properties of recycled aggregate: water absorption rate 4%; water-cement ratio: 0.43; admixtures: water-reducing agent 1.2%.

[0047] Secondly, the current iteration optimization count percentage and the average predicted intensity improvement within the sliding time window are read as indicators of the current optimization progress. The current iteration optimization count percentage is the ratio of the current iteration optimization count to the preset total number of iteration optimizations. The sliding time window is the most recent K consecutive iterations, where K is an integer greater than or equal to 5 and less than 30. The current iteration optimization count percentage = current iteration optimization count / preset total number of iteration optimizations, reflecting the optimization progress. The sliding time window refers to the most recent K consecutive iterations, used to capture recent dynamic changes in optimization, where K ≥ 5 and < 30. The average predicted intensity improvement is the average difference between the predicted intensity of each generation and the previous generation within the sliding window, reflecting the speed of recent intensity optimization. First, the ratio of the current iteration count to the preset total number of iterations is calculated. Then, the most recent K consecutive iterations are selected, and the average predicted intensity improvement during the iterations is calculated. These two are used as indicators of the current optimization progress.

[0048] For example, the preset total number of iterations for optimization is 100, and the current iteration is 15. The current iteration's optimization percentage is 15 / 100 = 0.15, meaning the optimization progress is 15%. Choosing K=10, since this is the 15th iteration, the sliding window is the most recent 10 generations, i.e., generations 6 to 15. The prediction intensities for each generation are: 31.2 MPa, 31.5 MPa, 31.8 MPa, 32.0 MPa, 32.2 MPa, 32.4 MPa, 32.5 MPa, 32.7 MPa, 32.8 MPa, and 33.0 MPa. The improvement in prediction intensity for each generation is calculated as: 31.5 - 31.2 = 0.3 MPa, 31.8 - 31.5 = 0.3 MPa. 32.0-31.8=0.2MPa, 32.2-32.0=0.2MPa, 32.4-32.2=0.2MPa, 32.5-32.4=0.1MPa, 32.7-32.5=0.2MPa, 32.8-32.7=0.1MPa, 33.0-32.8=0.2MPa; the average improvement in predicted intensity is (0.3+0.3+0.2+0.2+0.2+0.1+0.2+0.1+0.2) / 9=0.2MPa. That is, the current optimization process is: the current iteration count accounts for 15%, and the average improvement in predicted intensity is 0.2MPa.

[0049] In this embodiment of the invention, by reading the initial parameter set and process indicators of the current optimization, the parameter objects for subsequent prediction are clarified, and the progress of optimization and recent optimization speed are captured. This provides accurate scene feature basis for subsequent prediction complexity analysis, avoids the disconnect between subsequent prediction and the current optimization stage, and ensures the dynamic adaptability of the prediction strategy.

[0050] S300: Based on the sample training set and the initial matching parameter set, perform intensity prediction complexity analysis and output the first prediction complexity coefficient.

[0051] In this embodiment of the invention, a first prediction complexity coefficient is output based on the strength prediction complexity analysis performed on the sample training set and the initial mix design parameter set. The difficulty of predicting the strength of recycled aggregate concrete is closely related to the degree of matching and parameter completeness between the current initial mix design parameter set and the sample training set. If the initial parameters have high similarity to the samples and the parameters are complete, the prediction difficulty is low; if the similarity is uneven and parameters are missing, the prediction difficulty will increase significantly. If a uniform prediction strategy is directly adopted while ignoring the differences, it is easy to encounter problems such as prediction redundancy when the similarity is high or prediction deviation is large when the similarity is low. Therefore, it is necessary to analyze indicators such as parameter similarity and completeness to output a first prediction complexity coefficient, providing an accurate basis for subsequent adaptation of the prediction mechanism.

[0052] Step S300 in the method provided in this embodiment of the invention includes:

[0053] The initial matching parameter set is compared with multiple matching parameter schemes in the sample training set to obtain the similarity of multiple parameters.

[0054] The average similarity of the multiple parameters is calculated to obtain the mean parameter similarity.

[0055] The similarity fluctuation of the multiple parameters is calculated to obtain the parameter similarity variation coefficient;

[0056] An intensity prediction complexity analysis is performed based on the mean of parameter similarity and the coefficient of variation of parameter similarity to evaluate and determine the first prediction complexity coefficient.

[0057] First, the initial matching parameter set is compared with multiple matching parameter schemes in the sample training set to obtain multiple parameter similarities. Parameter similarity is used to quantify the comprehensive similarity of two matching schemes across all parameter dimensions, with a value range of 0-1. The initial matching parameter set is compared with each matching parameter scheme in the sample training set one by one. The comparison needs to consider the dimensions and importance of different parameters. Usually, all parameters are first normalized, with numerical parameters mapped to the [0,1] interval, and categorical parameters converted into numerical vectors using one-hot encoding. Then, a suitable similarity measurement algorithm, such as cosine similarity, is used to calculate multiple parameter similarity values, with a value range of 0-1, where 1 indicates complete similarity.

[0058] For example, the initial mixing parameter set is: cement dosage: 310 Mineral admixtures: fly ash 15%; Recycled coarse aggregate replacement rate: 60%; Recycled fine aggregate replacement rate: 10%; Total aggregate volume: 0.67 Key properties of recycled aggregate: water absorption 4%; water-cement ratio: 0.43; admixtures: water-reducing agent 1.2%. The mixing parameters for a specific sample scheme in the training set are: cement dosage: 300... Mineral admixtures: 20% fly ash; Recycled coarse aggregate replacement rate: 50%; Recycled fine aggregate replacement rate: 0%; Total aggregate volume: 0.68 Key properties of recycled aggregate: water absorption 3.5%; water-cement ratio: 0.45; admixtures: water-reducing agent 1%. First, all parameters were normalized; for example, the cement dosage was set to the maximum value of 400 mg / L in the sample set. Minimum value 250 Normalization is performed. The normalized value of the initial group's cement usage is (310-250) / (400-250) = 0.4, and the normalized value of the sample scheme is (300-250) / (400-250) ≈ 0.33. After normalizing the remaining parameters using the same logic, the cosine similarity between the normalized vectors of the two schemes is calculated, assuming a similarity value of 0.85. This process is repeated, comparing the current parameter group with all 1000 historical schemes in the sample training set, ultimately obtaining 1000 parameter similarity values.

[0059] Next, the average of the multiple parameter similarities is calculated to obtain the mean parameter similarity. The mean parameter similarity is an indicator representing the overall similarity between the current initial matching parameter set and all schemes in the sample training set; a higher value indicates a higher overall matching degree. All parameter similarity values ​​are summed, and the sum is then divided by the total number of schemes compared in the sample training set to obtain the final mean parameter similarity. For example, if the total similarity is 780, then the mean parameter similarity = 780 / 1000 = 0.78.

[0060] Furthermore, the similarity fluctuation of the multiple parameter similarities is calculated to obtain the parameter similarity coefficient of variation. The parameter similarity coefficient of variation is an indicator characterizing the degree of similarity difference between the current initial parameter set and different sample schemes; the larger the value, the higher the degree of unevenness in similarity. Parameter similarity coefficient of variation = parameter similarity standard deviation / parameter similarity mean. First, the standard deviation of all parameter similarity values ​​is calculated to reflect the degree of dispersion and fluctuation of similarity values; then, the standard deviation is divided by the parameter similarity mean, and the result is the parameter similarity coefficient of variation. For example, for the obtained 1000 parameter similarity values, statistical calculation shows that its standard deviation is 0.13, the parameter similarity mean is 0.78, and the parameter similarity coefficient of variation = 0.13 / 0.78 ≈ 0.167.

[0061] Subsequently, the intensity prediction complexity analysis is performed based on the mean parameter similarity and the coefficient of variation of parameter similarity to evaluate and determine the first prediction complexity coefficient.

[0062] The process includes analyzing the complexity of intensity prediction based on the mean and coefficient of variation of parameter similarity, and evaluating and determining a first prediction complexity coefficient, including:

[0063] A parameter integrity analysis is performed on the initial mating parameter group to determine the parameter integrity, wherein the parameter integrity is the ratio of the number of parameters in the initial mating parameter group to the total number of parameters in the preset mating parameter table;

[0064] An intensity prediction complexity analysis is performed based on the mean parameter similarity, the coefficient of variation of parameter similarity, and the parameter completeness to evaluate and determine a first prediction complexity coefficient. The first prediction complexity coefficient is negatively correlated with the mean parameter similarity and the parameter completeness, and positively correlated with the coefficient of variation of parameter similarity.

[0065] First, a parameter completeness analysis is performed on the initial mix design parameter group to determine the parameter completeness. The parameter completeness is the ratio of the number of parameters in the initial mix design parameter group to the total number of parameters in the preset mix design parameter table. Parameter completeness is an indicator of the coverage of parameters in the current initial mix design parameter group; a higher value indicates more detailed and comprehensive input indicators involved in the prediction. The parameter completeness is calculated by counting the actual number of parameters included in the current initial mix design parameter group and then dividing that number by the total number of parameters in the preset mix design parameter table. For example, if the preset mix design parameter table contains a total of 8 parameters—cement dosage, mineral admixture ratio, recycled coarse aggregate replacement rate, recycled fine aggregate replacement rate, total aggregate volume, recycled aggregate water absorption rate, water-cement ratio, and admixture dosage—and the current initial mix design parameter group fully includes all 8 parameters, then the parameter completeness = 8 / 8 = 1.

[0066] Secondly, an intensity prediction complexity analysis is performed based on the mean parameter similarity, coefficient of variation of parameter similarity, and parameter completeness to evaluate and determine a first prediction complexity coefficient. This first prediction complexity coefficient is negatively correlated with the mean parameter similarity and parameter completeness, and positively correlated with the coefficient of variation of parameter similarity. The first prediction complexity coefficient is a prediction difficulty index quantified based on the matching and parameter completeness of the initial parameter set with the sample training set, and is one of the core bases for formulating subsequent adaptive prediction mechanisms. A larger mean similarity indicates a higher similarity between the representation and the sample data, resulting in a smaller prediction complexity coefficient; a higher parameter completeness indicates more detailed input indicators during representation prediction, resulting in a smaller prediction complexity coefficient; a higher coefficient of variation of parameter similarity indicates uneven overall similarity of the representation, resulting in higher prediction complexity. Based on the negative correlation between the first prediction complexity coefficient and the mean parameter similarity and parameter completeness, and the positive correlation with the coefficient of variation of parameter similarity, the formula is used to calculate: First prediction complexity coefficient = Coefficient of variation of parameter similarity / (Mean parameter similarity × Parameter completeness). The larger the calculated result, the higher the intensity prediction complexity corresponding to the current initial parameter set. For example, if the mean parameter similarity is 0.78, the coefficient of variation of parameter similarity is 0.167, and the parameter completeness is 1, substituting these values ​​into the formula, the first prediction complexity coefficient is 0.167 / (0.78×1)≈0.214.

[0067] In this embodiment of the invention, a scientific prediction difficulty assessment index is constructed by quantifying the overall matching degree, similarity dispersion, and parameter completeness between the initial parameter set and the sample training set. By calculating the first prediction complexity coefficient, the complexity of intensity prediction is accurately quantified. The first prediction complexity coefficient clearly reflects the difficulty of the prediction scenario corresponding to the current initial parameter set, providing a reliable basis for subsequently developing differentiated adaptive prediction mechanisms and effectively avoiding the accuracy loss problem caused by a uniform prediction strategy.

[0068] S400: Based on the current optimization process, perform intensity prediction complexity analysis and output a second prediction complexity coefficient.

[0069] In this embodiment of the invention, a second prediction complexity coefficient is output based on the complexity analysis of the current optimization process. The requirements for accuracy and efficiency in strength prediction differ significantly at different stages of iterative optimization of the recycled aggregate concrete mix proportion: in the early stages of optimization, the number of iterations is small and the strength improvement is large, requiring a focus on efficient traversal of the parameter space without the need for high-complexity prediction; in the later stages, the number of iterations is large and the strength improvement is small, requiring a focus on finding high-quality parameters to accurately lock in the optimal parameters, necessitating increased prediction complexity. Using a uniform prediction complexity coefficient would lead to wasted resources in the early stages and insufficient accuracy in the later stages. Therefore, a second prediction complexity coefficient needs to be analyzed and output based on the current percentage of iterations and the average strength improvement, providing a phased basis for subsequent adaptation of the prediction mechanism.

[0070] Step S400 in the method provided in this embodiment of the invention includes:

[0071] The second prediction complexity coefficient is determined based on the proportion of the current iteration optimization count and the average improvement magnitude of the prediction intensity. The second prediction complexity coefficient is positively correlated with the proportion of the current iteration optimization count and negatively correlated with the average improvement magnitude of the prediction intensity.

[0072] In this embodiment of the invention, a second prediction complexity coefficient is determined based on the proportion of the current iteration optimization count and the average improvement magnitude of the prediction intensity. The second prediction complexity coefficient is positively correlated with the proportion of the current iteration optimization count and negatively correlated with the average improvement magnitude of the prediction intensity. A smaller proportion of iterations indicates a larger improvement magnitude, signifying a rapid optimization phase, thus emphasizing itinerary efficiency; therefore, a lower prediction complexity coefficient can be set. Conversely, a larger proportion of iterations indicates a smaller improvement magnitude, signifying an imminent convergence phase, requiring higher evaluation accuracy and emphasizing quality optimization. First, the proportion of the current iteration optimization count and the average improvement magnitude of the prediction intensity are normalized. Since the proportion of the current iteration optimization count is between 0 and 1, no normalization is needed. The theoretical or expected range of the average improvement magnitude is set to [0 MPa, 0.5 MPa], and Min-Max normalization is performed based on this range. The weight of the proportion of the current iteration optimization count is then set. and the weight of the mean improvement after normalization , Combining the correlation rule, the second prediction complexity coefficient = × Percentage of current iteration optimization counts + × (1 - mean improvement after normalization).

[0073] For example, setting , Given that the average improvement magnitude is 0.2 MPa, the normalized average improvement magnitude is 0.2 / 0.5 = 0.4, and the current iteration optimization count ratio is 0.15, substituting into the formula, we get: the second prediction complexity coefficient = 0.6 × 0.15 + 0.4 × (1 - 0.4) = 0.33.

[0074] In this embodiment of the invention, by quantitatively analyzing the progress and improvement of the optimization algorithm, a second prediction complexity coefficient representing the prediction complexity on the process side is generated. This achieves a dynamic and sensitive response to the optimization process. It automatically reduces the prediction complexity requirement in the rapid optimization stage where high efficiency is required, and automatically increases the prediction complexity requirement in the convergence stage where high accuracy is required. It can intelligently adjust the focus of work according to the actual progress of the optimization task, and provides core decision parameters for achieving dynamic and reasonable allocation of prediction resources and computing costs throughout the entire optimization lifecycle.

[0075] S500: Based on the first prediction complexity coefficient and the second prediction complexity coefficient, formulate an adaptive strength prediction mechanism, activate the concrete strength prediction channel, perform strength prediction based on the initial mix parameter group, and output the concrete strength prediction result.

[0076] In this embodiment of the invention, an adaptive strength prediction mechanism is established based on the first and second prediction complexity coefficients, and the concrete strength prediction channel is activated. Strength prediction is then performed based on the initial set of parameters, and the concrete strength prediction result is output. The first prediction complexity coefficient output in S300 quantifies the matching difficulty between the initial parameter set and the sample training set, and the second prediction complexity coefficient output in S400 quantifies the difference in prediction requirements at the current optimization stage. These two coefficients correspond to the two core dimensions of parameter adaptability and optimization stage, respectively. If the prediction mechanism is based solely on a single coefficient, it can easily lead to an imbalance between accuracy and efficiency. Therefore, it is necessary to comprehensively consider both the first and second prediction complexity coefficients to establish an adaptive strength prediction mechanism. After activating the prediction channel, the prediction unit is accurately selected to achieve a balance between prediction accuracy and optimization efficiency, outputting reliable strength prediction results.

[0077] Step S500 in the method provided in this embodiment of the invention includes:

[0078] The adaptation strength prediction mechanism is formulated based on the first prediction complexity coefficient and the second prediction complexity coefficient, including:

[0079] The ratio of the first prediction complexity coefficient to the preset first standard complexity coefficient is used as the compensation index for the first unit selection.

[0080] The ratio of the second prediction complexity coefficient to the preset second standard complexity coefficient is used as the compensation index for the second unit selection.

[0081] The comprehensive unit selection compensation index is obtained by weighting the compensation index selected by the first unit and the compensation index selected by the second unit.

[0082] The product of the comprehensive unit selection compensation index and J is rounded down to obtain the number of adaptation units selected, K, which serves as the adaptation strength prediction mechanism. Here, K is greater than or equal to 2 and less than or equal to J.

[0083] Specifically, the concrete strength prediction channel is activated, and K prediction units are randomly selected from the J concrete strength prediction units in the concrete strength prediction channel to predict the strength of the initial mix parameter group. The average of the K predicted strengths is then used as the final concrete strength prediction result.

[0084] First, the ratio of the first prediction complexity coefficient to the preset first standard complexity coefficient is used as the first unit selection compensation index. The first unit selection compensation index is an indicator characterizing the degree of deviation of the parameter matching difficulty from the standard value, used to quantify the requirement for the number of prediction units for the parameter dimension. The preset first standard complexity coefficient is a standard threshold for parameter matching difficulty determined based on a large number of recycled aggregate concrete samples, ranging from 0 to 1, used to calibrate the influence of the first prediction complexity coefficient. The preset first standard complexity coefficient is set, and the first unit selection compensation index = first prediction complexity coefficient / preset first standard complexity coefficient. A ratio greater than 1 indicates that the parameter matching difficulty is higher than the standard value, requiring more prediction units to improve accuracy; a ratio less than 1 indicates that the difficulty is lower than the standard value, allowing for a reduction in prediction units to improve efficiency.

[0085] For example, the current first prediction complexity coefficient is 0.214; based on the experience of recycled aggregate concrete engineering, the preset first standard complexity coefficient is set to 0.3; the first unit selection compensation index is 0.214 / 0.3≈0.71<1, indicating that the matching difficulty between the current initial parameter set and the sample is lower than the standard value, and the prediction units need to be appropriately reduced to improve efficiency.

[0086] Secondly, the ratio of the second prediction complexity coefficient to the preset second standard complexity coefficient is used as the second unit selection compensation index. The second unit selection compensation index is an indicator characterizing the degree of deviation of the prediction demand in the optimization stage from the standard value, used to quantify the demand for prediction units in the optimization dimension. The preset second standard complexity coefficient is a standard threshold for the optimization stage demand determined based on historical optimization data statistics, with a value range of 0-1, used to calibrate the influence of the second prediction complexity coefficient. The preset second standard complexity coefficient is set, and the second unit selection compensation index = second prediction complexity coefficient / preset second standard complexity coefficient. A ratio greater than 1 indicates that the demand for prediction accuracy in the optimization stage is higher than the standard value, requiring more prediction units; a ratio less than 1 indicates that the demand is lower than the standard value, allowing for a reduction in prediction units to adapt to efficiency requirements.

[0087] For example, the current second prediction complexity coefficient is 0.33; based on the experience of optimizing recycled aggregate concrete, the preset second standard complexity coefficient is set to 0.3; the compensation index for the second unit is 0.33 / 0.3 = 1.1 > 1, indicating that the current 15th iteration stage requires slightly higher prediction accuracy than the standard value, and the prediction units need to be appropriately increased to match the quality requirements of the search.

[0088] Furthermore, a comprehensive unit selection compensation index is calculated by weighting the first unit selection compensation index and the second unit selection compensation index. The comprehensive unit selection compensation index is a comprehensive indicator that integrates the two compensation indices, taking into account both parameter matching difficulty and optimization stage requirements, and is used to accurately calculate the number of suitable prediction units. Weighting coefficients are set for the first and second unit selection compensation indices, and the weight allocation needs to balance the impact of parameter adaptability and the optimization stage. A weighted summation formula is used to multiply the two compensation indices by their corresponding weights and then sum them to obtain the comprehensive unit selection compensation index. For example, setting the weighting coefficients as follows: first unit selection compensation index weight = 0.5, second unit selection compensation index weight = 0.5, comprehensive unit selection compensation index = 0.5 × 0.71 + 0.5 × 1.1 = 0.905 < 1, indicating that considering the dual-dimensional requirements, the complexity of the current overall prediction scenario is lower than the preset benchmark level, and fewer prediction units than the benchmark need to be selected.

[0089] Subsequently, the product of the comprehensive unit selection compensation index and J is rounded to obtain the number of adaptive units selected, K, which serves as the adaptive strength prediction mechanism. Here, K is greater than or equal to 2 and less than or equal to J. The number of adaptive units selected, K, refers to the number of concrete strength prediction units participating in this strength prediction and is the core of the adaptive strength prediction mechanism. The K value controls the balance between prediction accuracy and efficiency. The total number of prediction units, J, refers to the total number of concrete strength prediction units obtained from S100 through 10-fold cross-training. The comprehensive unit selection compensation index is multiplied by the total number of prediction units J obtained from S100, and the result is rounded to the nearest integer. After rounding, it is necessary to verify whether the result satisfies 2≤K≤J. If K<2, it is adjusted to 2; if K>J, it is adjusted to J. The final determined K value is the core parameter of the adaptive strength prediction mechanism. For example, given that the total number of prediction units J=10 and the comprehensive unit selection compensation index=0.905; the number of adaptive units selected, K=0.905×10=9.05≈9, means that 9 prediction units need to be selected for this strength prediction.

[0090] Finally, the concrete strength prediction channel is activated, and K prediction units are randomly selected from the J concrete strength prediction units in the channel to predict the strength of the initial mix design parameters. The average of the K predicted strengths is taken as the final concrete strength prediction result. The concrete strength prediction channel refers to the prediction set in S100 composed of J independently trained and converged feedforward neural networks, which can be activated and selected as needed to participate in the prediction. The final concrete strength prediction result is the average of the output values ​​of the K prediction units, used to reduce the prediction bias of a single unit and improve the stability and reliability of the prediction result. First, the concrete strength prediction channel constructed in S100 is activated to make it operational; then, K units are randomly selected from the J prediction units; the initial mix design parameters are input into the selected K prediction units, and each unit independently outputs a predicted strength value; finally, the arithmetic mean of the K predicted strength values ​​is calculated, and this average is the final concrete strength prediction result, which is output to the mix design optimization system.

[0091] For example, the prediction function is started by calling 10 concrete strength prediction units constructed by S100; prediction units are selected: since K=9, 9 prediction units are randomly selected from the 10 concrete strength prediction units to participate in the prediction; input parameters are predicted: the initial mix design parameters of the 15th generation are set as follows: cement content: 310 Mineral admixtures: fly ash 15%; Recycled coarse aggregate replacement rate: 60%; Recycled fine aggregate replacement rate: 10%; Total aggregate volume: 0.67 Key properties of recycled aggregate: water absorption rate 4%; water-cement ratio: 0.43; admixture: water-reducing agent 1.2%. These are input into 9 prediction units, and the predicted strengths output by each unit are: 41.5, 41.8, 41.2, 42.0, 41.6, 41.3, 41.9, 41.4, and 41.7. The calculated and output results are: mean predicted strength = (41.5 + 41.8 + 41.2 + 42.0 + 41.6 + 41.3 + 41.9 + 41.4 + 41.7) / 9 = 374.4 / 9 ≈ 41.6 MPa. 41.6 MPa is taken as the final prediction result.

[0092] In this embodiment of the invention, an adaptive strength prediction mechanism is formulated by comprehensively considering the first and second prediction complexity coefficients, and the number K of adaptive units is accurately determined. This balances the requirements for prediction accuracy due to the difficulty of parameter matching with the balance requirements for efficiency and quality in the optimization stage, avoiding the limitations of single-dimensional decision-making. By activating prediction channels, selecting an appropriate number of prediction units, and taking the average of the output results, the bias risk of a single prediction unit is effectively reduced, improving the stability, accuracy, and reliability of the strength prediction of recycled aggregate concrete. A complete technical chain is formed, consisting of model construction, process analysis, difficulty quantification, mechanism adaptation, and prediction output, effectively improving the prediction accuracy of recycled aggregate concrete strength and providing efficient and reliable strength data support for the optimization of recycled aggregate concrete mix proportions.

[0093] Through the specific implementation methods described above, the embodiments of the present invention achieve the following technical effects:

[0094] This invention provides an intelligent prediction method and system for the strength of recycled aggregate concrete that integrates machine learning. First, it constructs a multi-unit prediction channel based on building big data and J-fold cross-validation, providing a stable and reliable model foundation for prediction. Then, by reading parameter sets and progress indicators from the mix design optimization stage, it quantifies the prediction complexity coefficients from two dimensions: the matching degree between initial parameters and samples, and the efficiency and progress of the optimization stage, breaking the limitations of traditional fixed prediction strategies. Finally, it determines the appropriate number of prediction units based on dual-coefficient weighted calculation, and effectively reduces the prediction bias of a single model by averaging the prediction results from multiple units, achieving a dynamic balance between prediction accuracy and optimization efficiency. This invention effectively improves the accuracy and stability of recycled aggregate concrete strength prediction, provides efficient data support for mix design optimization, and reduces the cost and cycle of traditional physical experiments.

[0095] Example 2, as Figure 2 As shown, this invention provides an intelligent prediction system for the strength of recycled aggregate concrete that integrates machine learning. The system includes:

[0096] Prediction channel construction module 11 is used to collect a sample training set to train a machine learning model and construct a concrete strength prediction channel, guided by recycled aggregate concrete.

[0097] The optimization parameter reading module 12 is used to read the initial mix parameter set and the current optimization process of the current optimization stage during the mix proportion optimization process of recycled aggregate concrete.

[0098] The first coefficient analysis module 13 is used to perform intensity prediction complexity analysis based on the sample training set and the initial matching parameter set, and output the first prediction complexity coefficient.

[0099] The second coefficient analysis module 14 is used to perform intensity prediction complexity analysis based on the current optimization process and output a second prediction complexity coefficient.

[0100] The strength prediction output module 15 is used to formulate an adaptive strength prediction mechanism based on the first prediction complexity coefficient and the second prediction complexity coefficient, activate the concrete strength prediction channel, perform strength prediction based on the initial mix parameter group, and output the concrete strength prediction result.

[0101] In one embodiment, the prediction channel building module 11 is further configured to:

[0102] Guided by recycled aggregate concrete, information retrieval is carried out based on building big data, a sample mix parameter scheme set is collected, and the historical concrete test strength of different sample mix parameter schemes is obtained as the sample concrete strength, thus obtaining the sample concrete strength set.

[0103] The sample combination parameter scheme set and the sample concrete strength set are used as the sample training set, and the sample training set is divided into J-fold cross partitions to obtain J sample training data groups, where J is an integer greater than or equal to 10.

[0104] Using the sample matching parameter scheme as input data and the sample concrete strength as supervision data, feedforward neural networks are trained to convergence using the J sample training data groups to generate J concrete strength prediction units, which are then combined to obtain a concrete strength prediction channel.

[0105] The sample mix design includes several mix parameters, which include at least the following: cement dosage, type of mineral admixture, dosage of mineral admixture, replacement rate of recycled coarse aggregate, replacement rate of recycled fine aggregate, total aggregate volume, key properties of recycled aggregate, water-cement ratio, type of admixture, and dosage of admixture.

[0106] In one embodiment, the optimization parameter reading module 12 is further configured to:

[0107] Specifically, the current optimization process is defined as the ratio of the current iteration optimization count to the average predicted intensity improvement within the sliding time window. The current optimization count ratio is the ratio of the current iteration optimization count to the preset total iteration optimization count. The sliding time window is the most recent K consecutive iterations of optimization, where K is an integer greater than or equal to 5 and less than 30.

[0108] In one embodiment, the first coefficient analysis module 13 is further configured to:

[0109] The initial matching parameter set is compared with multiple matching parameter schemes in the sample training set to obtain the similarity of multiple parameters.

[0110] The average similarity of the multiple parameters is calculated to obtain the mean parameter similarity.

[0111] The similarity fluctuation of the multiple parameters is calculated to obtain the parameter similarity variation coefficient;

[0112] An intensity prediction complexity analysis is performed based on the mean of parameter similarity and the coefficient of variation of parameter similarity to evaluate and determine the first prediction complexity coefficient.

[0113] The process includes analyzing the complexity of intensity prediction based on the mean and coefficient of variation of parameter similarity, and evaluating and determining a first prediction complexity coefficient, including:

[0114] A parameter integrity analysis is performed on the initial mating parameter group to determine the parameter integrity, wherein the parameter integrity is the ratio of the number of parameters in the initial mating parameter group to the total number of parameters in the preset mating parameter table;

[0115] An intensity prediction complexity analysis is performed based on the mean parameter similarity, the coefficient of variation of parameter similarity, and the parameter completeness to evaluate and determine a first prediction complexity coefficient. The first prediction complexity coefficient is negatively correlated with the mean parameter similarity and the parameter completeness, and positively correlated with the coefficient of variation of parameter similarity.

[0116] In one embodiment, the second coefficient analysis module 14 is further configured to:

[0117] The second prediction complexity coefficient is determined based on the proportion of the current iteration optimization count and the average improvement magnitude of the prediction intensity. The second prediction complexity coefficient is positively correlated with the proportion of the current iteration optimization count and negatively correlated with the average improvement magnitude of the prediction intensity.

[0118] In one embodiment, the intensity prediction output module 15 is further configured to:

[0119] The adaptation strength prediction mechanism is formulated based on the first prediction complexity coefficient and the second prediction complexity coefficient, including:

[0120] The ratio of the first prediction complexity coefficient to the preset first standard complexity coefficient is used as the compensation index for the first unit selection.

[0121] The ratio of the second prediction complexity coefficient to the preset second standard complexity coefficient is used as the compensation index for the second unit selection.

[0122] The comprehensive unit selection compensation index is obtained by weighting the compensation index selected by the first unit and the compensation index selected by the second unit.

[0123] The product of the comprehensive unit selection compensation index and J is rounded down to obtain the number of adaptation units selected, K, which serves as the adaptation strength prediction mechanism. Here, K is greater than or equal to 2 and less than or equal to J.

[0124] Specifically, the concrete strength prediction channel is activated, and K prediction units are randomly selected from the J concrete strength prediction units in the concrete strength prediction channel to predict the strength of the initial mix parameter group. The average of the K predicted strengths is then used as the final concrete strength prediction result.

[0125] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0126] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0127] This specification and accompanying drawings are merely illustrative examples of the invention and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its scope. Therefore, if such modifications and modifications fall within the scope of the invention and its equivalents, the invention is intended to include these modifications and modifications.

Claims

1. A method for intelligent prediction of the strength of recycled aggregate concrete integrating machine learning, characterized in that, The methods include: Guided by recycled aggregate concrete, a sample training set is collected to train a machine learning model and construct a concrete strength prediction channel, wherein the concrete strength prediction channel includes J concrete strength prediction units. In the process of optimizing the mix proportion of recycled aggregate concrete, the initial mix parameter set and the current optimization process of the current optimization stage are read. The current optimization process includes the proportion of the current iteration optimization number in the current optimization stage and the average value of the predicted strength improvement within the sliding time window. The proportion of the current iteration optimization number is the ratio of the current iteration optimization number to the preset total iteration optimization number. The sliding time window is the most recent K consecutive iterations of optimization, where K is an integer greater than or equal to 5 and less than 30. The initial matching parameter set is compared with multiple matching parameter schemes in the sample training set to obtain the similarity of multiple parameters. The average similarity of the multiple parameters is calculated to obtain the mean parameter similarity. The similarity fluctuation of the multiple parameters is calculated to obtain the parameter similarity variation coefficient; A parameter integrity analysis is performed on the initial mating parameter group to determine the parameter integrity, wherein the parameter integrity is the ratio of the number of parameters in the initial mating parameter group to the total number of parameters in the preset mating parameter table; An intensity prediction complexity analysis is performed based on the mean parameter similarity, coefficient of variation of parameter similarity, and parameter completeness to evaluate and determine a first prediction complexity coefficient. This first prediction complexity coefficient is negatively correlated with the mean parameter similarity and parameter completeness, and positively correlated with the coefficient of variation of parameter similarity. A second prediction complexity coefficient is then output based on the current optimization process. The number of randomly selected prediction units is determined based on the first prediction complexity coefficient and the second prediction complexity coefficient, and the concrete strength prediction channel is activated. Strength prediction is performed based on the initial mix parameter group and the number of randomly selected prediction units, and the concrete strength prediction result is output.

2. The intelligent prediction method for the strength of recycled aggregate concrete based on machine learning as described in claim 1, characterized in that, Guided by recycled aggregate concrete, a sample training set is collected to train a machine learning model and construct a concrete strength prediction channel, including: Guided by recycled aggregate concrete, information retrieval is carried out based on building big data, a sample mix parameter scheme set is collected, and the historical concrete test strength of different sample mix parameter schemes is obtained as the sample concrete strength, thus obtaining the sample concrete strength set. The sample combination parameter scheme set and the sample concrete strength set are used as the sample training set, and the sample training set is divided into J-fold cross partitions to obtain J sample training data groups, where J is an integer greater than or equal to 10. Using the sample matching parameter scheme as input data and the sample concrete strength as supervision data, feedforward neural networks are trained to convergence using the J sample training data groups to generate J concrete strength prediction units, which are then combined to obtain a concrete strength prediction channel.

3. The intelligent prediction method for the strength of recycled aggregate concrete based on machine learning according to claim 2, characterized in that, The sample mix design includes several mix parameters, among which the mix parameters include at least the cement content, mineral admixture type, mineral admixture content, recycled coarse aggregate replacement rate, recycled fine aggregate replacement rate, total aggregate volume, key properties of recycled aggregate, water-cement ratio, admixture type, and admixture dosage.

4. The intelligent prediction method for the strength of recycled aggregate concrete based on machine learning as described in claim 1, characterized in that, The second prediction complexity coefficient is determined based on the proportion of the current iteration optimization count and the average improvement magnitude of the prediction intensity. The second prediction complexity coefficient is positively correlated with the proportion of the current iteration optimization count and negatively correlated with the average improvement magnitude of the prediction intensity.

5. The intelligent prediction method for the strength of recycled aggregate concrete based on machine learning according to claim 2, characterized in that, An adaptation strength prediction mechanism is formulated based on the first prediction complexity coefficient and the second prediction complexity coefficient, including: The ratio of the first prediction complexity coefficient to the preset first standard complexity coefficient is used as the compensation index for the first unit selection. The ratio of the second prediction complexity coefficient to the preset second standard complexity coefficient is used as the compensation index for the second unit selection. The comprehensive unit selection compensation index is obtained by weighting the compensation index selected by the first unit and the compensation index selected by the second unit. The product of the comprehensive unit selection compensation index and J is rounded down to obtain the number of adaptation units selected, K, which serves as the adaptation strength prediction mechanism. Here, K is greater than or equal to 2 and less than or equal to J.

6. The intelligent prediction method for the strength of recycled aggregate concrete based on machine learning according to claim 5, characterized in that, The concrete strength prediction channel is activated, and K prediction units are randomly selected from the J concrete strength prediction units in the concrete strength prediction channel to predict the strength of the initial mix parameter group. The average of the K predicted strengths is taken as the final concrete strength prediction result.

7. A smart prediction system for the strength of recycled aggregate concrete integrating machine learning, characterized in that, The system for implementing the intelligent prediction method for the strength of recycled aggregate concrete fused with machine learning as described in any one of claims 1-6, the system comprising: The prediction channel construction module is used to collect a sample training set to train a machine learning model and construct a concrete strength prediction channel, guided by recycled aggregate concrete. The optimization parameter reading module is used to read the initial mix parameter set and the current optimization process in the current optimization stage during the mix proportion optimization process of recycled aggregate concrete. The first coefficient analysis module is used to perform intensity prediction complexity analysis based on the sample training set and the initial matching parameter set, and output the first prediction complexity coefficient. The second coefficient analysis module is used to perform intensity prediction complexity analysis based on the current optimization process and output a second prediction complexity coefficient. The strength prediction output module is used to formulate an adaptive strength prediction mechanism based on the first prediction complexity coefficient and the second prediction complexity coefficient, activate the concrete strength prediction channel, perform strength prediction based on the initial mix parameter group, and output the concrete strength prediction result.

Images