Method and equipment for predicting compressive strength of fiber-reinforced flow-state cement fly ash

Abstract

The invention provides a method and equipment for predicting compressive strength of fiber-reinforced flow-state cement fly ash, and relates to the technical field of performance prediction of volcanic ash inorganic binders. The method comprises the following steps: acquiring a compressive strength test parameter set of to-be-predicted fiber-reinforced flow-state cement fly ash; inputting the compressive strength test parameter set into a pre-trained compressive strength prediction model to obtain a predicted value of the compressive strength of the to-be-predicted fiber-reinforced flow-state cement fly ash; wherein the compressive strength prediction model is constructed on the basis of an LSTM model, hyper-parameters in the compressive strength prediction model are determined on the basis of a dream optimization algorithm, and the hyper-parameters comprise the number of neurons, the random discarding rate and the optimal numerical value of the batch size; the fitness function of the compressive strength prediction model comprises an activity constraint function constructed based on the content ratio of SiO2 to Al2O3 in the fly ash. According to the method, the compressive strength of the fiber-reinforced flow-state cement fly ash can be accurately predicted.

Description

Technical Field

[0001] This invention relates to the field of performance prediction technology for volcanic ash inorganic binders, and in particular to a method and equipment for predicting the compressive strength of fiber-reinforced fluid cement fly ash. Background Technology

[0002] Fiber-reinforced fluidized cement fly ash is a new type of lightweight roadbed filler used to solve problems such as uneven settlement of roadbeds and bridge approach slab settlement. Its mechanical properties can meet the requirements of roadbed strength, reduce the self-weight of embankments, reduce additional stress on the foundation, and reduce post-construction settlement of the roadbed. It can maximize the stability of the roadbed and effectively save natural sand and gravel resources, thus having significant economic and social benefits.

[0003] The mix proportion of fiber-reinforced fluidized cement fly ash is related to the compressive strength of the roadbed. Currently, besides determining the compressive strength through manual experiments, neural network models are also used to predict the compressive strength of fiber-reinforced fluidized cement fly ash with different mix proportions. Among various neural network models, Long Short-Term Memory (LSTM) is a special type of recurrent neural network specifically designed to solve the gradient vanishing or exploding problems that traditional recurrent neural networks encounter when processing long sequence data. LSTM captures long-term dependencies by introducing a "gating mechanism" and "cell state".

[0004] However, LSTM models are quite sensitive to hyperparameters. If the hyperparameters are not chosen properly, the constructed prediction model will be difficult to accurately predict the compressive strength of fiber-reinforced fluidized cement fly ash. Summary of the Invention

[0005] This invention provides a method and device for predicting the compressive strength of fiber-reinforced fluidized cement fly ash, in order to solve the problem that the hyperparameters of LSTM cannot be accurately determined when predicting the compressive strength of fiber-reinforced fluidized cement fly ash.

[0006] In a first aspect, embodiments of the present invention provide a method for predicting the compressive strength of fiber-reinforced fluidized cement fly ash, comprising:

[0007] Obtain the set of test parameters for the compressive strength of the fiber-reinforced fluid cement fly ash to be predicted;

[0008] The set of compressive strength test parameters is input into the pre-trained compressive strength prediction model to obtain the predicted value of the compressive strength of the fiber-reinforced fluid cement fly ash to be predicted.

[0009] Among them, the compressive strength prediction model is constructed based on the LSTM model. The hyperparameters in the compressive strength prediction model are determined based on the dream optimization algorithm. The hyperparameters include the optimal values ​​of the number of neurons, the random dropout rate, and the batch size. The fitness function of the compressive strength prediction model includes an activity constraint function constructed based on the content ratio of SiO2 to Al2O3 in fly ash.

[0010] In one possible implementation, the set of compressive strength test parameters includes the mass ratio of cement to fly ash, the mass ratio of water-reducing agent to cement, the mass ratio of reinforcing agent to the first mixture, the mass ratio of foaming agent to the first mixture, the mass ratio of water to the first mixture, the mass ratio of fiber to the first mixture, the curing time, and the content ratio of SiO2 to Al2O3 in fly ash; wherein the first mixture is cement and fly ash.

[0011] In one possible implementation, the fitness function of the compressive strength prediction model includes an accuracy function, a generalization function, and an activity constraint function;

[0012] The precision function is constructed based on the root mean square error of the predicted intensity and the true intensity on the validation set; the generalization function is constructed based on the absolute value of the error on the training set and the error on the validation set.

[0013] The activity constraint function is constructed based on the Sigmoid function, and the input of the Sigmoid function is constructed based on the content ratio of SiO2 to Al2O3 in fly ash in the training set.

[0014] In one possible implementation, the input to the Sigmoid function is determined based on the difference between a first content and a standard content, where the first content is the content of SiO2 and Al2O3 in the fly ash of the training set.

[0015] In one possible implementation, the fitness function of the compressive strength prediction model is obtained by a weighted sum of the accuracy function, the generalization function, and the activity constraint function, with the sum of the weights of the three functions being 1.

[0016] In one possible implementation, the hyperparameters of the compressive strength prediction model are determined based on the convergence of the fitness function and the stability of the global optimal solution;

[0017] Fitness function convergence means that the rate of change of the fitness function is less than a preset value in a preset number of iterations, and global optimal solution stability means that the position of the global optimal solution remains unchanged in multiple iterations.

[0018] In one possible implementation, the compressive strength prediction model also includes an incremental learning module;

[0019] After the compressive strength prediction model is trained, the training set of the compressive strength prediction model is continuously updated. When the mean square error between the predicted value and the true value is greater than the first error threshold, the compressive strength prediction model is updated.

[0020] In one possible implementation, when the error between the predicted value and the true value is greater than a first error threshold, the weights of the anti-compression intensity prediction model are updated based on the gradient descent algorithm.

[0021] In one possible implementation, the set of compressive strength test parameters is input into a pre-trained compressive strength prediction model, including:

[0022] The compressive strength test parameter set is normalized to obtain the processed compressive strength test parameter set.

[0023] The processed set of compressive strength test parameters is input into the pre-trained compressive strength prediction model.

[0024] In a second aspect, embodiments of the present invention provide a fiber-reinforced fluidized cement fly ash compressive strength prediction device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect or any possible implementation of the first aspect.

[0025] In this embodiment of the invention, by inputting the obtained set of compressive strength test parameters into a pre-trained compressive strength prediction model, the compressive strength of the fiber-reinforced fluidized cement fly ash to be predicted can be calculated. To accurately predict compressive strength, the optimal values ​​for the number of neurons, random dropout rate, and batch size in the compressive strength prediction model of this invention are determined using a dream optimization algorithm. The dream optimization algorithm effectively avoids model parameters getting trapped in local optima, achieving global balance and ensuring that all three parameters can obtain optimal solutions, thereby improving the model's prediction performance under complex working conditions and increasing prediction accuracy. Furthermore, by introducing the SiO2 to Al2O3 content ratio in the fly ash into the fitness function, the SiO2 to Al2O3 content ratio can reflect the activity index of the fly ash, allowing the compressive strength prediction model to achieve a balance between pursuing prediction accuracy and adhering to the chemical properties of the material, thus enabling more accurate prediction of compressive strength. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating the implementation of the fiber-reinforced fluidized cement fly ash compressive strength prediction method provided in this embodiment of the invention.

[0027] Figure 2 This is a flowchart illustrating the implementation of determining hyperparameters based on the dream optimization algorithm provided in this embodiment of the invention.

[0028] Figure 3 This is a schematic diagram of the structure of the fiber-reinforced fluidized cement fly ash compressive strength prediction device provided in an embodiment of the present invention. Detailed Implementation

[0029] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0030] As described in the background section, although common grid search methods can optimize the hyperparameters of LSTM to some extent, these methods are inefficient and cannot find the optimal solution, thus failing to adequately meet the optimization requirements of complex engineering problems. For the compressive strength of fiber-reinforced fluidized cement fly ash, the hyperparameters of LSTM are even more difficult to determine due to the complex working conditions and numerous influencing factors. Therefore, the method provided in this invention is proposed to quickly and accurately determine the hyperparameters of LSTM.

[0031] See Figure 1 The document illustrates a flowchart of the method for predicting the compressive strength of fiber-reinforced fluid cement fly ash provided in this embodiment of the invention, detailed below:

[0032] S110. Obtain the set of test parameters for the compressive strength of the fiber-reinforced fluidized cement fly ash to be predicted.

[0033] The compressive strength test parameter set includes the mass ratio of cement to fly ash, the mass ratio of water-reducing agent to cement, the mass ratio of reinforcing agent to the first mixture, the mass ratio of foaming agent to the first mixture, the mass ratio of water to the first mixture, the mass ratio of fiber to the first mixture, curing time, and the SiO2 to Al2O3 content ratio in fly ash. The SiO2 to Al2O3 content ratio in fly ash is used to characterize the activity index of fly ash. The first mixture consists of cement and fly ash.

[0034] S120. Input the set of compressive strength test parameters into the pre-trained compressive strength prediction model to obtain the predicted value of the compressive strength of the fiber-reinforced fluid cement fly ash to be predicted.

[0035] The compressive strength prediction model is built based on the LSTM model. The hyperparameters in the compressive strength prediction model are determined based on the dream optimization algorithm, and the hyperparameters include the number of neurons, the dropout rate, and the optimal batch size.

[0036] The fitness function of the compressive strength prediction model includes an activity constraint function based on the content ratio of SiO2 to Al2O3 in fly ash.

[0037] In some embodiments, the fitness function of the compressive strength prediction model includes an accuracy function, a generalization function, and an activity constraint function.

[0038] The accuracy function is constructed based on the root mean square error between the predicted and actual intensities on the validation set. The generalization function is constructed based on the absolute values ​​of the errors on the training and validation sets. The activity constraint function is constructed based on the sigmoid function, whose input is based on the SiO2 to Al2O3 content ratio in the fly ash of the training set.

[0039] In this embodiment, the input to the Sigmoid function is determined based on the difference between a first content and a standard content, where the first content is the content of SiO2 and Al2O3 in the fly ash of the training set.

[0040] In this embodiment, the fitness function of the compressive strength prediction model is obtained by weighted summation of the accuracy function, generalization function and activity constraint function, and the sum of the weights of the three functions is 1.

[0041] Specifically, the fitness function is Fitness, the accuracy function is RMSE, the generalization function is ΔRMSE, and the activity constraint function is... The weights are 6:3:1.

[0042]

[0043] ΔRMSE=|RMSE train -RMSE val |;

[0044]

[0045] y i To determine the prediction strength of the validation set, The true strength of the validation set is given by N, where N is the number of samples in the validation set, and RMSE is given by N. train RMSE is the error of the training set. val The error of the validation set is given by k = 5.0. α0 represents the ideal ratio of the standard content to 2.2, corresponding to the optimal reactivity.

[0046] In some embodiments, since the data in the compressive strength test parameter set are of different types, it is necessary to first normalize the compressive strength test parameter set to facilitate calculation, thereby obtaining a processed compressive strength test parameter set. Then, the processed compressive strength test parameter set is input into a pre-trained compressive strength prediction model.

[0047] In this embodiment, the LSTM neural network structure includes multiple recurrent units (memory units), and uses input gates, forget gates, and output gates to control the flow of information and the updating of memory, ensuring effective handling of long-term dependencies in long sequence data. To reduce the impact of the magnitude difference between different input gates on the prediction results, maximum and minimum values ​​are selected for normalization, transforming all data into values ​​between [0,1].

[0048] The formula for normalizing the compressive strength test parameter set is:

[0049]

[0050] In the formula: X ′ The values ​​represent the normalized data values, and min(X) and max(X) represent the minimum and maximum values ​​of the indicator in the dataset, respectively.

[0051] The LSTM model in this invention is built using the Python language.

[0052] In some embodiments, the training process of the compressive strength prediction model is as follows:

[0053] First, an initial LSTM model needs to be established. The LSTM model consists of multiple circularly connected "memory units." Each unit consists of three multiplication gates connected together, i.e., input gate i. t Forgotten Gate t and output gate o t It can also be written as write, reset, and read operations. Gates in the memory cells facilitate long-term retention and access to the internal cell state. The LSTM output will depend on all previous inputs. Previous information is neither completely discarded nor completely transferred to the current state. Instead, the effect of previous information on the current state is carefully controlled through gate signals.

[0054] The gate calculation formula for an LSTM recurrent neural network is:

[0055] f t =σ(W f (h t-1 ,X t )+b f );

[0056] i t =σ(W i (h t-1 ,X t )+b i );

[0057] g t =tanh(W g (h t-1 ,Xt )+b g );

[0058] c t =f t c t-1 +i t g t ;

[0059] o t =σ(W o (h t-1 ,X t )+b o );

[0060] h t =o t tanh(c t );

[0061] In the formula: f t i t g t and o t These represent the values ​​of the forget gate, input gate, update gate, and output gate, respectively; W f W i W g and W o These are the weight vectors for the four gates; b represents the bias vector; σ and c are... t These are the activation functions for the input and output gates. Typically, the sigmoid function is used as the input activation function, and the tanh function is used as the output activation function.

[0062] The Sigmoid function is:

[0063]

[0064] The Tanh function is:

[0065]

[0066] After constructing the initial LSTM model, the hyperparameters of the LSTM model can be tuned based on the Dream Optimization Algorithm (DOA), and an activity constraint function can be introduced into the fitness function.

[0067] The Dream Optimization Algorithm (DOA) is a metaheuristic optimization algorithm inspired by the characteristics of human dreams. It simulates features such as memory retention, partial forgetting, and self-organization in dreams, balancing the exploration (broad search of the solution space) and development (refined search based on existing solutions) phases of the optimization process. The algorithm initializes a random population, introduces forgetting and self-organization strategies during the exploration phase to update solutions in certain dimensions, and allows individuals to share dream information, enhancing their ability to escape local optima. In the development phase, individuals update based on the globally optimal solution, thereby improving the algorithm's global search capability and local optimization capability.

[0068] Therefore, DOA can be used to find the optimal values ​​for the number of neurons, random dropout rate, and batch size in an LSTM network, as follows: Figure 2 As shown,

[0069] First, initialization. Set the population size N = 50 and the hyperparameter dimension Dim = 3. Mapping the actual range of hyperparameters: Set the optimal values ​​for the number of neurons, random dropout rate, and batch size within the actual range of these three hyperparameters.

[0070] The fitness function is Fitness, the accuracy function is RMSE, the generalization function is ΔRMSE, and the activity constraint function is... The weights are 6:3:1.

[0071]

[0072] ΔRMSE=|RMSE train -RMSE val |;

[0073]

[0074] y i To determine the prediction strength of the validation set, The true strength of the validation set is given by N, where N is the number of samples in the validation set, and RMSE is given by N. train RMSE is the error of the training set. val The error of the validation set is given by k = 5.0. α0 represents the ideal ratio of the standard content to 2.2, corresponding to the optimal reactivity.

[0075] Second, DOA randomly generates a set of initial hyperparameter combinations. These hyperparameter combinations represent different configurations of the LSTM model. The LSTM model built based on these initial hyperparameter combinations is trained, and the Fitness value corresponding to each hyperparameter is calculated.

[0076] Then, the exploration phase begins. The population is divided into 5 groups according to memory capacity, with each group having a forgetting dimension K.q Unlike other groups, the high-forgetting-rate group tends to use a global search. Three strategies are executed alternately:

[0077] Memory strategy: Individuals are reset to the position of the historical best solution within the group.

[0078]

[0079] Let i be the position of the i-th individual in iteration t+1. The best individual for group q in iteration t.

[0080] Forget-replenishment strategy (probability 0.9): In a randomly selected K... q In each dimension, the cosine annealing term is used to control the range of randomness (large-scale exploration in the early stage, convergence in the later stage).

[0081]

[0082] Where, x l,j and x u,j These are the minimum and maximum values ​​for that dimension, respectively. `rand` is a random number. For the position of the i-th individual in the j-th dimension at iteration t+1, Let T be the position of the best individual in group q in the j-th dimension during iteration t; t is the current iteration number, and T is the position of the best individual in group q in iteration t. max T represents the maximum number of iterations. d This represents the maximum number of iterations during the exploration phase.

[0083] Dream sharing strategy (probability 0.1):

[0084]

[0085] m is a natural number randomly selected from [1, N] and used for updates in each dimension. Values ​​from other individuals are copied in the forget dimension to enhance population diversity; m is a random individual.

[0086] Third, in the development phase, grouping is removed, and all groups use the same forgetting dimension K. r A dual-strategy approach will be adopted:

[0087] Memory strategy: Individuals are reset to the position of the global historical best solution.

[0088]

[0089] Forget-replenish strategy: cosine phase change to enhance local fine-grained search.

[0090]

[0091] Fourth, fitness evaluation: After each iteration, the hyperparameter combination of each individual needs to be evaluated. This is done by calculating their fitness value.

[0092] Fifth, optimal solution update: DOA updates the optimal solution for each individual and the global optimal solution based on the fitness of each individual. It also determines whether to move from the exploration phase to the development phase based on whether the fitness change rate reaches a preset threshold.

[0093] Sixth, the termination condition is determined based on the convergence of the fitness function and the stability of the global optimal solution.

[0094] Fitness function convergence means that the fitness function does not change within a preset number of iterations, or its rate of change is less than a preset value. In this case, the optimization process can be considered converged, reaching the global optimum or close to the optimum. Global optimum stability means that the position of the global optimum remains unchanged across multiple iterations, indicating that the algorithm has found the optimal solution and cannot be further improved.

[0095] After the above process, the values ​​of the hyperparameters in the LSTM can be determined. In addition, the LSTM needs to be trained using training and validation sets. The trained model is the compressive strength prediction model, which can be used for subsequent compressive strength prediction.

[0096] Both the training set and the validation set include multiple samples, and each sample includes the compressive strength of fiber-reinforced fluidized cement fly ash and the set of compressive strength test parameters corresponding to that compressive strength.

[0097] In some embodiments, in order to ensure the accuracy of the compressive strength prediction model, the compressive strength prediction model also includes an incremental learning module. After the compressive strength prediction model is trained, the training set of the compressive strength prediction model needs to be continuously updated. When the mean square error between the predicted value and the true value is greater than the first error threshold, the compressive strength prediction model is updated.

[0098] In this embodiment, when the error between the predicted value and the true value is greater than a first error threshold, the weights of the anti-compression intensity prediction model are updated based on the gradient descent algorithm.

[0099] Specifically, the goal of incremental learning is to fine-tune an existing model based on new data, rather than completely retraining it. This saves time and computational resources while ensuring the model continuously improves with the addition of new data. For LSTM models, the core of incremental learning is to make small updates to the model's parameters, mainly adjusting the weights of the last fully connected layer, or performing local optimizations on the entire network when the error is large.

[0100] A prediction error threshold is set so that incremental learning is triggered when the model's prediction error exceeds this threshold. The prediction error is calculated as mean squared error.

[0101]

[0102] Where N is the number of samples, y i It is the true value of the i-th sample. θ is the predicted value of the i-th sample, and θ is the parameter (including weights) of the LSTM model.

[0103] Gradient calculation involves using backpropagation to compute the gradient of the loss function for each model parameter. This gradient represents the rate of change of the loss function with respect to each parameter. The gradient descent algorithm uses this information to update the model's weights.

[0104] Gradient descent is used to update the model's weights. The update formula for incremental learning is:

[0105]

[0106] Where, θ t These are the model parameters, i.e., the weights, at the current time step. η is the learning rate, which controls the magnitude of each weight update.

[0107] The prediction method provided by this invention predicts the compressive strength of fiber-reinforced fluidized cement fly ash by inputting the acquired compressive strength test parameter set into a pre-trained compressive strength prediction model. To accurately predict compressive strength, the compressive strength test parameter set in this invention includes various factors that influence compressive strength, allowing for a comprehensive evaluation of fiber-reinforced fluidized cement fly ash. Furthermore, the optimal values ​​for the number of neurons, random dropout rate, and batch size in the compressive strength prediction model of this invention are determined using a dream optimization algorithm. This algorithm effectively avoids model parameters getting trapped in local optima, achieving global balance and ensuring that all three parameters obtain optimal solutions, thereby improving the model's prediction performance under complex working conditions and increasing prediction accuracy. In addition, by introducing the SiO2 to Al2O3 content ratio in the fly ash into the fitness function, the SiO2 to Al2O3 content ratio reflects the activity index of the fly ash, allowing the compressive strength prediction model to achieve a balance between pursuing prediction accuracy and adhering to the chemical properties of the material, thus enabling more accurate prediction of compressive strength.

[0108] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0109] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0110] Figure 3 A schematic diagram of the fiber-reinforced fluidized cement fly ash compressive strength prediction device provided in an embodiment of the present invention is shown. For ease of explanation, only the parts relevant to the embodiment of the present invention are shown, and are described in detail below:

[0111] like Figure 3 As shown, the fiber-reinforced fluidized cement fly ash compressive strength prediction device 300 includes:

[0112] The data acquisition module 310 is used to acquire the set of test parameters for the compressive strength of the fiber-reinforced fluid cement fly ash to be predicted.

[0113] The strength prediction module 320 is used to input the set of compressive strength test parameters into the pre-trained compressive strength prediction model to obtain the predicted value of the compressive strength of the fiber-reinforced fluid cement fly ash to be predicted.

[0114] Among them, the compressive strength prediction model is constructed based on the LSTM model. The hyperparameters in the compressive strength prediction model are determined based on the dream optimization algorithm. The hyperparameters include the optimal values ​​of the number of neurons, the random dropout rate, and the batch size. The fitness function of the compressive strength prediction model includes an activity constraint function constructed based on the content ratio of SiO2 to Al2O3 in fly ash.

[0115] In one possible implementation, the set of compressive strength test parameters includes the mass ratio of cement to fly ash, the mass ratio of water-reducing agent to cement, the mass ratio of reinforcing agent to the first mixture, the mass ratio of foaming agent to the first mixture, the mass ratio of water to the first mixture, the mass ratio of fiber to the first mixture, the curing time, and the content ratio of SiO2 to Al2O3 in fly ash; wherein the first mixture is cement and fly ash.

[0116] In one possible implementation, the fitness function of the compressive strength prediction model includes an accuracy function, a generalization function, and an activity constraint function;

[0117] The precision function is constructed based on the root mean square error of the predicted intensity and the true intensity on the validation set; the generalization function is constructed based on the absolute value of the error on the training set and the error on the validation set.

[0118] The activity constraint function is constructed based on the Sigmoid function, and the input of the Sigmoid function is constructed based on the content ratio of SiO2 to Al2O3 in fly ash in the training set.

[0119] In one possible implementation, the input to the Sigmoid function is determined based on the difference between a first content and a standard content, where the first content is the content of SiO2 and Al2O3 in the fly ash of the training set.

[0120] In one possible implementation, the fitness function of the compressive strength prediction model is obtained by a weighted sum of the accuracy function, the generalization function, and the activity constraint function, with the sum of the weights of the three functions being 1.

[0121] In one possible implementation, the hyperparameters of the compressive strength prediction model are determined based on the convergence of the fitness function and the stability of the global optimal solution;

[0122] Fitness function convergence means that the rate of change of the fitness function is less than a preset value in a preset number of iterations, and global optimal solution stability means that the position of the global optimal solution remains unchanged in multiple iterations.

[0123] In one possible implementation, the compressive strength prediction model also includes an incremental learning module;

[0124] After the compressive strength prediction model is trained, the training set of the compressive strength prediction model is continuously updated. When the mean square error between the predicted value and the true value is greater than the first error threshold, the compressive strength prediction model is updated.

[0125] In one possible implementation, when the error between the predicted value and the true value is greater than a first error threshold, the weights of the anti-compression intensity prediction model are updated based on the gradient descent algorithm.

[0126] In one possible implementation, the prediction strength module 320 is used to normalize the compressive strength test parameter set to obtain the processed compressive strength test parameter set.

[0127] The processed set of compressive strength test parameters is input into the pre-trained compressive strength prediction model.

[0128] The prediction device provided by this invention can predict the compressive strength of fiber-reinforced fluidized cement fly ash by inputting the acquired set of compressive strength test parameters into a pre-trained compressive strength prediction model. To accurately predict compressive strength, the optimal values ​​for the number of neurons, random dropout rate, and batch size in the compressive strength prediction model of this invention are determined using a dream optimization algorithm. The dream optimization algorithm effectively avoids model parameters getting trapped in local optima, achieving global balance and ensuring that all three parameters can obtain optimal solutions, thereby improving the model's prediction performance under complex working conditions and increasing prediction accuracy. Furthermore, by introducing the SiO2 to Al2O3 content ratio in the fly ash into the fitness function, the activity index of the fly ash can be reflected through the SiO2 to Al2O3 content ratio, thus enabling more accurate prediction of compressive strength.

[0129] This invention also provides a device for predicting the compressive strength of fiber-reinforced fluidized cement fly ash, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described in the above embodiments.

[0130] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0131] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for predicting the compressive strength of fiber-reinforced fluidized cement fly ash, characterized in that, include: Obtain the set of test parameters for the compressive strength of fiber-reinforced fluid cement fly ash to be predicted; The set of compressive strength test parameters is input into a pre-trained compressive strength prediction model to obtain the predicted value of the compressive strength of the fiber-reinforced fluid cement fly ash to be predicted. The compressive strength prediction model is constructed based on the LSTM model. The hyperparameters in the compressive strength prediction model are determined based on the dream optimization algorithm. The hyperparameters include the optimal values ​​of the number of neurons, the random dropout rate, and the batch size. The fitness function of the compressive strength prediction model includes an activity constraint function constructed based on the content ratio of SiO2 to Al2O3 in fly ash.

2. The method for predicting the compressive strength of fiber-reinforced fluidized cement fly ash according to claim 1, characterized in that, The set of compressive strength test parameters includes the mass ratio of cement to fly ash, the mass ratio of water-reducing agent to cement, the mass ratio of reinforcing agent to the first mixture, the mass ratio of foaming agent to the first mixture, the mass ratio of water to the first mixture, the mass ratio of fiber to the first mixture, the curing time, and the content ratio of SiO2 to Al2O3 in fly ash; wherein, the first mixture is cement and fly ash.

3. The method for predicting the compressive strength of fiber-reinforced fluidized cement fly ash according to claim 1, characterized in that, The fitness function of the compressive strength prediction model includes an accuracy function, a generalization function, and an activity constraint function; The accuracy function is constructed based on the root mean square error of the predicted intensity and the true intensity of the validation set; the generalization function is constructed based on the absolute value of the error of the training set and the error of the validation set. The activity constraint function is constructed based on the Sigmoid function, and the input of the Sigmoid function is constructed based on the content ratio of SiO2 to Al2O3 in fly ash in the training set.

4. The method for predicting the compressive strength of fiber-reinforced fluidized cement fly ash according to claim 3, characterized in that, The input to the Sigmoid function is determined based on the difference between a first content and a standard content, where the first content is the content of SiO2 and Al2O3 in the fly ash of the training set.

5. The method for predicting the compressive strength of fiber-reinforced fluidized cement fly ash according to claim 3, characterized in that, The fitness function of the compressive strength prediction model is obtained by weighted summation of the accuracy function, generalization function and activity constraint function, with the sum of the weights of the three functions being 1.

6. The method for predicting the compressive strength of fiber-reinforced fluidized cement fly ash according to claim 1, characterized in that, The hyperparameters of the compressive strength prediction model are determined based on the convergence of the fitness function and the stability of the global optimal solution; The convergence of the fitness function means that the rate of change of the fitness function is less than a preset value in a preset number of iterations, and the stability of the global optimal solution means that the position of the global optimal solution remains unchanged in multiple iterations.

7. The method for predicting the compressive strength of fiber-reinforced fluidized cement fly ash according to claim 1, characterized in that, The compressive strength prediction model also includes an incremental learning module; After the compressive strength prediction model is trained, the training set of the compressive strength prediction model is continuously updated. When the mean square error between the predicted value and the true value is greater than the first error threshold, the parameters in the compressive strength prediction model are updated.

8. The method for predicting the compressive strength of fiber-reinforced fluidized cement fly ash according to claim 7, characterized in that, When the error between the predicted value and the true value is greater than the first error threshold, the weights of the compressive strength prediction model are updated based on the gradient descent algorithm.

9. The method for predicting the compressive strength of fiber-reinforced fluidized cement fly ash according to any one of claims 1-8, characterized in that, The step of inputting the set of compressive strength test parameters into a pre-trained compressive strength prediction model includes: The set of compressive strength test parameters is normalized to obtain the processed set of compressive strength test parameters; The processed set of compressive strength test parameters is input into a pre-trained compressive strength prediction model.

10. A device for predicting the compressive strength of fiber-reinforced fluidized cement fly ash, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1 to 9.

Images