Novel cement clinker system design method and device based on data driving

Abstract

The invention relates to the technical field of cement clinker, and discloses a novel cement clinker system design method and device based on data driving, and the method comprises the steps: constructing a cement clinker test data set, carrying out the augmentation of the cement clinker test data set, training a preset performance prediction model through the augmentation data set, and carrying out the prediction of the preset performance. Obtaining a trained performance prediction model; generating a plurality of cement clinker systems randomly or by adopting a genetic algorithm, and inputting the cement clinker systems into the performance prediction model to obtain a performance index prediction value of each cement clinker system; and screening out a novel cement clinker system meeting a preset requirement according to the performance index predicted value of each cement clinker system. According to the method, the hidden space of the clinker system possibly having excellent performance is expanded by widening the cement clinker data, the limitation of few data samples on the clinker composition system is broken through, the cement clinker system is designed according to performance requirements, and the research and development efficiency of the cement clinker is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of cement clinker, and in particular to a data-driven novel cement clinker system design method and device. Background Art

[0002] The development of low-carbon technology in the cement industry is a key issue of global concern and an inevitable trend in the development of the cement industry. In the cement production process, 50%-60% of CO2 is produced by the decomposition of raw materials. Therefore, reducing the high-carbon emission mineral phases in the clinker (such as tricalcium silicate) is an important way to effectively reduce the carbon emissions of clinker. In addition to traditional silicate cement clinker, high-belite silicate cement clinker, sulphoaluminate cement clinker, ferroaluminate cement clinker, etc. have been developed, but these clinkers are only suitable for special application scenarios. Due to the reduction of the content of high-carbon emission mineral phases, CO2 emissions in the cement production process can be reduced by more than 10%. At present, research on low-carbon clinker at home and abroad is limited to the influence of low-carbon minerals such as dicalcium silicate and calcium sulphosilicate on clinker performance, and most research methods are based on the experience and knowledge of researchers and expensive experiments. How to efficiently screen suitable clinker systems and composition ratios is a major technical problem.

[0003] The emergence of machine learning can solve the above problems to a certain extent. Machine learning is a data-driven design theory, which obtains prior knowledge and predictive models from known material data sets, and then predicts complex material properties based on these models. In recent years, machine learning has successfully predicted various properties of cement and concrete, and reversely designed the proportions of concrete components according to target properties, but there has been no development in the research and development of new cement clinker systems. The reason for limiting the research and development of clinker systems is the lack of high-quality and large-capacity data, and there are great obstacles to obtaining experimental data ("real" or empirical data, used to build extensive and high-quality training data sets). Summary of the invention

[0004] In view of this, the present invention provides a new cement clinker system design method and device based on data-driven, so as to solve the problem that the cement clinker system cannot be quickly designed due to the small amount and low quality of new cement clinker data.

[0005] In a first aspect, the present invention provides a novel cement clinker system design method based on data-driven, the method comprising:

[0006] Construct a cement clinker test data set, and augment the cement clinker test data set, and use the augmented data set to train a preset performance prediction model to obtain a trained performance prediction model;

[0007] Generate multiple cement clinker systems randomly or by genetic algorithm, and use the trained performance prediction model to obtain the predicted value of the performance index of each cement clinker system;

[0008] According to the predicted values ​​of performance indicators of each cement clinker system, a new cement clinker system that meets the preset requirements is screened out.

[0009] The novel data-driven cement clinker system design method provided by the present invention expands the latent space of clinker systems that may have excellent performance by augmenting cement clinker data, breaks through the limitation of small data samples on clinker composition systems, and predicts the performance of each augmented cement clinker composition system through a performance prediction model, thereby achieving the purpose of designing a clinker composition system according to performance requirements, shortening the research and development cycle of cement clinker, and reducing research and development costs.

[0010] In an optional implementation, the cement clinker test data set is augmented, and the augmented data set is used to train a preset performance prediction model to obtain a trained performance prediction model, including:

[0011] The adversarial neural network model is trained according to the cement clinker test data set to obtain the data augmentation model;

[0012] Input the cement clinker test data set into the data augmentation model to obtain the cement clinker augmented data set;

[0013] The preset performance prediction model is trained using the cement clinker augmented data set to obtain a trained performance prediction model.

[0014] The novel data-driven cement clinker system design method provided by the present invention utilizes a data augmentation model to expand the data volume of the cement clinker data set, breaking through the limitation of cement clinker system design under the condition of a small number of data samples, increasing the number of optional cement clinker systems, and facilitating the training of a more accurate performance prediction model.

[0015] In an optional implementation, the adversarial neural network model includes: a generator and a discriminator, and a data augmentation model is obtained by training a preset data augmentation model according to a cement clinker test data set, including:

[0016] Randomly initialize the parameters of the generator and discriminator;

[0017] The discriminator is trained according to the cement clinker test data set so that the discriminator can identify real data, where the real data is the data in the cement clinker test data set;

[0018] According to the cement clinker test data set, the generator is used to generate synthetic data, and the discriminator is used to distinguish the synthetic data from the real data to obtain the distinction result;

[0019] Use the distinction results to calculate the loss function of the discriminator and the loss function of the generator;

[0020] Repeat the above process of using the generator to generate synthetic data and using the discriminator to distinguish the synthetic data from the real data until the difference between the loss function of the discriminator and the preset discrimination threshold is less than the first preset difference threshold, and the difference between the loss function of the generator and the preset generation threshold is less than the second preset difference threshold, to obtain the trained generator and discriminator to form a data augmentation model.

[0021] The novel data-driven cement clinker system design method provided by the present invention utilizes the continuous confrontation and optimization of the generator and the discriminator in the adversarial neural network to make the synthetic data closer and closer to the real cement clinker data. The data augmentation model obtained by multiple iterative training can generate more realistic augmented data based on the cement clinker test data set.

[0022] In an optional implementation, the parameters of cement clinker augmented data in the cement clinker augmented data set include: cement clinker system and performance index;

[0023] The preset performance prediction model is trained using the cement clinker augmented data set to obtain a trained performance prediction model, including:

[0024] The cement clinker system in the cement clinker augmented data is used as the input of the preset performance prediction model, and the performance index in the cement clinker augmented data is used as the output of the preset performance prediction model to form a sample data set;

[0025] The sample data set is divided into a training set and a test set according to a preset ratio, and a performance prediction model is obtained through training. The sample data in the test set are all cement clinker test data.

[0026] The novel data-driven cement clinker system design method provided by the present invention utilizes the cement clinker augmented data set to train a performance prediction model, thereby expanding the number of sample data for model training and improving the performance and generalization ability of the cement clinker performance prediction model.

[0027] In an optional implementation, the parameters of the cement clinker test data include: a cement clinker system and performance indicators; augmenting the cement clinker test data set, using the augmented data set to train a preset performance prediction model to obtain a trained performance prediction model, and further comprising:

[0028] Use the cement clinker test data set to train the teacher model and obtain a trained teacher model;

[0029] Generate multiple cement clinker systems randomly or by genetic algorithm, and input the trained teacher model to obtain corresponding performance indicators. Multiple cement clinker systems and corresponding performance indicators constitute synthetic data;

[0030] The cement clinker dataset and synthetic data are combined into an augmented dataset to jointly train the student model, and a trained student model is obtained. The model structure of the student model is the same as that of the teacher model.

[0031] The trained student model is used as the teacher model, and the process of randomly or genetically generating multiple cement clinker systems is repeated, and the loss function of the corresponding student model is calculated until the loss function of the student model converges. The student model is used as the preset performance prediction model.

[0032] The novel data-driven cement clinker system design method provided by the present invention adopts a self-training algorithm, uses a teacher model to generate performance indicators of each cement clinker system, expands the cement clinker data, and uses the expanded data to train a student model. The above process is repeated until the student model converges, thereby improving the generalization ability and learning efficiency of the model.

[0033] In an optional embodiment, a plurality of cement clinker systems are generated using a genetic algorithm, including:

[0034] Randomly generating an initial population including a first number of cement clinker systems, and inputting the first number of cement clinker systems into a performance prediction model to obtain performance index prediction values ​​of each cement clinker system;

[0035] According to the predicted values ​​of the performance indicators of each cement clinker system, the fitness of each individual in the initial population is calculated using a preset fitness function;

[0036] According to the fitness of each individual in the initial population, each individual is selected, crossed, and mutated to obtain a new generation of population;

[0037] The process of selecting, crossing over and mutating each individual in the new generation population is repeated until the fitness of each individual in the population converges, and the second number of cement clinker systems in the population is used as multiple cement clinker systems generated by the genetic algorithm, and the second number is greater than the first number.

[0038] The novel data-driven cement clinker system design method provided by the present invention utilizes a genetic algorithm to process randomly acquired sample data to obtain individuals with higher fitness as the cement clinker system, and then generates a cement clinker system whose performance indicators are close to preset requirements. After being processed by a performance prediction model, it is convenient to quickly screen out the cement clinker system that meets the preset requirements from the prediction results, thereby improving the design efficiency.

[0039] In a second aspect, the present invention provides a novel cement clinker system design device based on data drive, the device comprising:

[0040] The performance prediction model training module is used to construct a cement clinker test data set, and to augment the cement clinker test data set, and to train a preset performance prediction model using the augmented data set to obtain a trained performance prediction model;

[0041] The performance index prediction module is used to generate multiple cement clinker systems randomly or by using a genetic algorithm, and obtain the performance index prediction value of each cement clinker system by using the trained performance prediction model;

[0042] The system screening module is used to screen out new cement clinker systems that meet the preset requirements based on the predicted values ​​of the performance indicators of each cement clinker system.

[0043] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the method of the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0044] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to cause a computer to execute the method of the first aspect or any corresponding embodiment thereof.

[0045] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions for causing a computer to execute the method of the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0047] Figure 1 is a flow chart of a novel cement clinker system design method based on data-driven according to an embodiment of the present invention;

[0048] Figure 2 is a flow chart of another novel cement clinker system design method based on data-driven according to an embodiment of the present invention;

[0049] Figure 3 is a flow chart of another novel cement clinker system design method based on data-driven according to an embodiment of the present invention;

[0050] Figure 4is a flow chart of a specific embodiment of a novel cement clinker system design method based on data-driven according to an embodiment of the present invention;

[0051] Figure 5 is a flow chart of another specific embodiment of the data-driven novel cement clinker system design method according to an embodiment of the present invention;

[0052] Figure 6 is a structural block diagram of a novel cement clinker system design device based on data drive according to an embodiment of the present invention;

[0053] Figure 7 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0055] The embodiment of the present invention provides a novel cement clinker system design method based on data-driven, which obtains more cement clinker systems by using a performance prediction model after data augmentation based on test data, so as to achieve the effect of designing a cement clinker system according to performance requirements.

[0056] According to an embodiment of the present invention, an embodiment of a novel cement clinker system design method based on data-driven is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and, although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0057] In this embodiment, a new cement clinker system design method based on data drive is provided, which can be used in the above-mentioned computer system. Figure 1 is a flow chart of a novel cement clinker system design method based on data drive according to an embodiment of the present invention, such as Figure 1 As shown, the process includes the following steps:

[0058] Step S101, constructing a cement clinker test data set, and augmenting the cement clinker test data set, using the augmented data set to train a preset performance prediction model to obtain a trained performance prediction model.

[0059] Specifically, the cement clinker test data set is the cement clinker related data obtained through experiments, and its parameters include: clinker mineral phase composition or chemical composition, fineness, 3d / 28d hydration product composition, carbon emissions, mechanical properties, durability, etc. Among them, the clinker mineral phase composition and chemical composition can be inferred from each other. The clinker mineral phases include: tricalcium silicate (C3S), dicalcium silicate (C2S), calcium silicate (CS), calcite (C3S2), tricalcium aluminate (C3A), monocalcium aluminate (CA), monocalcium dialuminate (CA2), dodecaaluminate (C3S2), calcium silicate (C3S3), calcium silicate (C3S4), calcium silicate (C3S5), calcium silicate (C3S6), calcium silicate (C3S7), calcium silicate (C3S8), calcium silicate (C3S9), calcium silicate (C3S10), calcium silicate (C3S2), calcium silicate (C3S2), calcium silicate (C3S2), calcium silicate (C3S2), calcium silicate (C3S3), calcium silicate (C3S4), calcium silicate (C3S5), calcium silicate (C3S6), calcium silicate (C3S7), calcium silicate (C3S8), calcium silicate (C3S10), calcium silicate (C3S11), calcium silicate (C3S12), calcium silicate (C3S13), calcium silicate (C3S14), calcium silicate (C3S15), calcium silicate (C3S16), calcium silicate (C3S17), calcium silicate (C3S18), calcium silicate (C3S19), calcium silicate (C3S19), calcium silicate (C3S11), calcium silicate (C3S12), calcium silicate (C3S13), calcium silicate (C3S19), calcium silicate (C3S11), calcium silicate (C3S13), calcium silica 12 A7), dicalcium ferrite (C2F), tetracalcium aluminoferrate (C4AF), hexacalcium aluminoferrate (C6AF2), hexacalcium ferroaluminate (C6A2F), calcium sulfoaluminate (C4A3$), calcium sulfosilicate (C5S2$), calcium barium sulfoaluminate (C3A3$·BaO), calcium strontium aluminate (C3A3$·SrO), calcium fluoroaluminate (C 11 A7·CaF2), calcium chloroaluminate (C 11 A7·CaCl2), tricalcium phosphate (C3P), tetracalcium phosphate (C4P), calcium aluminate phosphate (C8A4P), P phase (C6A4FS), Q phase (Ca 20 Al 32-2n Mg n Si n O 68 ), grosvenorite (C2AS), alinite (Ca 22 Si6Al2Cl2O 36 ), belinite (Ca3SiCl2O4), periclase (MgO), free gypsum (f-CaSO4) and free calcium oxide (f-CaO) are used as examples only, but are not limited to them. The mineral phase system is divided into binary system, ternary system, quaternary system and quinary system according to the number of mineral phase types that make up the clinker, and the proportion of mineral phase composition in a certain system is a random combination, and the sum of the proportions is 100%; the composition of 3d / 28d hydration products is an intermediate product in the manufacturing process of cement clinker. The types and proportions of hydration products are obtained according to thermodynamic calculations, which affect the mechanical properties and durability; the carbon emissions are calculated from the carbon emissions of a single mineral phase and its proportion. The calculation process is a mature existing technology and will not be repeated here; the mechanical properties are standard test blocks of cement mortar, including compressive strength, flexural strength, tensile strength and splitting tensile strength; the durability includes expansion and contraction properties, carbonization resistance, sulfate erosion resistance, chloride ion erosion resistance, etc.

[0060] For the cement clinker test data directly obtained, data cleaning is first performed on it to screen out valid data, and the cleaned data is used for data augmentation. The data cleaning process is a mature existing technology and will not be described in detail here.

[0061] Specifically, a preset performance prediction model of cement clinker is constructed using a deep neural network model, and the preset performance prediction model is trained using an expanded cement clinker augmented data set to obtain a trained performance prediction model.

[0062] Step S102, generating multiple cement clinker systems randomly or by using a genetic algorithm, and obtaining a predicted value of a performance index of each cement clinker system by using a trained performance prediction model.

[0063] Specifically, in the process of cement clinker research and development, multiple cement clinker systems can be randomly generated, and each cement clinker system can be input into the performance prediction model to obtain the corresponding performance index prediction value, so as to select the appropriate cement clinker system according to the actual production process and application scenario requirements, without the need for tedious experimental testing, saving research and development costs and improving research and development efficiency. The performance indicators include: carbon emissions, mechanical properties, and durability. The carbon emissions of the cement clinker system are calculated based on the carbon emissions of each mineral phase in the cement clinker system and the proportion of each mineral phase in the cement clinker system. Each cement clinker system is input into the performance prediction model to obtain the mechanical properties prediction value and durability prediction value of each cement clinker system. Since the parameters related to mechanical properties and durability performance need to be measured by experiments in traditional technologies, which is time-consuming and laborious, in this embodiment, each cement clinker system is input into the performance prediction model to obtain the mechanical properties prediction value and durability prediction value of each cement clinker system, thereby improving the research and development efficiency of cement clinker.

[0064] For example, the randomly generated clinker mineral composition is: 30% tricalcium silicate, 11% dicalcium silicate, 30% monocalcium aluminate, 26% dodecaluminate heptaaluminate, and the fineness is d 50 26 microns, 12% dihydrate gypsum is also added for performance testing. The composition of the hydration product after 28 days is: 30% hydrated calcium silicate, 7% hydrated calcium aluminum silicate, 39% ettringite, 12% strong calcium oxide. Calculate the carbon emissions: Total CO2emission = ∑w i ×d i =644kg / t. Among them, w i is the relative mass of the mineral phase, d i is the direct carbon emission of this mineral phase.

[0065] The above data are input into the performance prediction model to obtain the predicted values ​​of performance indicators, mainly including: compressive strength, flexural strength, tensile strength and splitting tensile strength in mechanical properties, expansion and contraction properties, carbonization resistance, sulfate corrosion resistance and chloride ion corrosion resistance in durability properties. This is only for example, but not limited to this.

[0066] In addition to randomly generating cement clinker systems, optimization algorithms such as genetic algorithms may also be used to generate cement clinker systems whose performance indicators are closer to preset requirements, which is only an example but not limited to this.

[0067] Step S103, selecting a new cement clinker system that meets preset requirements according to the predicted values ​​of the performance indicators of each cement clinker system.

[0068] Specifically, the predicted values ​​of the performance indicators of each cement clinker system are compared with the limits of the preset requirements to screen out new cement clinker systems that meet the preset requirements. The preset requirements do not necessarily limit all performance indicators. When low carbon is used as the preset requirement, a new low-carbon cement clinker system can be obtained. For example, with the carbon emission between 602-688kgCO2 / kg and the compressive strength of cement mortar strength>48.5MPa as the limit conditions, two of the cement clinker systems screened out are: the first one: 26.18% C3S, 1.29% C2S, 25.84% CA, 36.19% C 12 A7, 10.50% CaSO42H2O; The second: 1.43% C5S2$, 11.18% C2S, 36.43C3A3$·SrO, 36.61% C 11 A7·CaF2, 14.36% CaSO4.2H2O is only used as an example, but is not limited to this.

[0069] The novel data-driven cement clinker system design method provided in this embodiment utilizes a data augmentation model to augment cement clinker data, thereby expanding the latent space of clinker systems that may have excellent performance, breaking through the limitation of small data samples on clinker composition systems, and predicting the performance of each augmented cement clinker composition system through a performance prediction model, thereby achieving the purpose of designing a clinker composition system according to performance requirements, shortening the research and development cycle of cement clinker, and reducing research and development costs.

[0070] In this embodiment, a new cement clinker system design method based on data drive is provided, which can be used in the above-mentioned computer system. Figure 2 is a flow chart of a novel cement clinker system design method based on data drive according to an embodiment of the present invention, such as Figure 2 As shown, the process includes the following steps:

[0071] Step S201, constructing a cement clinker test data set, and augmenting the cement clinker test data set, using the augmented data set to train a preset performance prediction model to obtain a trained performance prediction model.

[0072] Specifically, the above step S201 includes:

[0073] Step S2011, training an adversarial neural network model based on the cement clinker test data set to obtain a data augmentation model.

[0074] Step S2012: input the cement clinker test data set into the data augmentation model to obtain the cement clinker augmented data set.

[0075] Specifically, the cement clinker test data set is input into the data augmentation model to obtain the synthetic data of cement clinker. The parameters of the synthetic data are the same as those of the cement clinker test data, but the values ​​of the parameters can be different. The cement clinker test data and the synthetic data together constitute the cement clinker augmented data set, which realizes data expansion and is used for the training of the subsequent performance prediction model.

[0076] In some optional implementations, the adversarial neural network model includes: a generator and a discriminator, and a data augmentation model is obtained by training a preset data augmentation model according to a cement clinker test data set, including:

[0077] Step a1, randomly initialize the parameters of the generator and discriminator.

[0078] Specifically, in this embodiment, the Generative Adversarial Networks (GAN) algorithm is used as the preset data augmentation model. The adversarial neural network mainly includes two parts: a generator and a discriminator. The cleaned cement clinker test data set is used as real data, and the generator is used to generate synthetic data similar to the real data. The discriminator is used to discriminate between real data and synthetic data.

[0079] The main parameters of the adversarial neural network algorithm include: noise dimension, number of neurons, learning rate, activation function, single training data volume, number of training rounds, number of random production samples, and optimizer. The noise dimension is used to enhance the diversity and complexity of synthetic data and is generally set to 50-200; the number of neurons is generally set to 256-2048; the learning rate in the generator and discriminator is used to control the speed and direction of the iteration of the GAN model parameters and is generally set to 10 -3 -10 -5 ; The activation function is used to enhance the ability of the GAN model to handle nonlinear problems. Generally, one or more of sigmoid, Tanh, ReLu, and LeakyReLu are selected. The amount of single training data and the number of training rounds are used to update the GAN model parameters. Due to the limited amount of data, they should not be set too high. Generally, they are set to 16-64 and 1000-2000 respectively. The number of random production samples is determined by the number of real data in the cement clinker test data set, and is generally set to 0.5-7 times the number of real data. The optimizer is used to improve the performance and generalization ability of the GAN model. Generally, Adam, SGD, or AdamW are selected.

[0080] Step a2, training a discriminator according to the cement clinker test data set so that the discriminator can recognize real data, where the real data is the data in the cement clinker test data set.

[0081] Step a3: Based on the cement clinker test data set, a generator is used to generate synthetic data, and a discriminator is used to distinguish the synthetic data from the real data to obtain a distinction result.

[0082] Specifically, the cement clinker test data in the cement clinker test dataset is taken as real data. Based on the real data, the generator is used to learn to generate synthetic data that is as close to the real data as possible, with the aim of deceiving the discriminator. The discriminator is used to distinguish between the cement clinker test data and the synthetic data, with the aim of determining whether the input data is real.

[0083] Step a4, using the distinction results to calculate the loss function of the discriminator and the loss function of the generator.

[0084] Specifically, the generator and discriminator are trained in a competitive manner, and the generator's loss function is used to evaluate the generator's model performance, while the discriminator's loss function is used to evaluate the discriminator's model performance.

[0085] Step a5, repeat the above process of using the generator to generate synthetic data, and using the discriminator to distinguish the synthetic data from the real data, until the difference between the loss function of the discriminator and the preset discrimination threshold is less than the first preset difference threshold, and the difference between the loss function of the generator and the preset generation threshold is less than the second preset difference threshold, and the trained generator and discriminator are obtained to form a data augmentation model.

[0086] Specifically, in each training, the generator generates synthetic data, and the discriminator distinguishes between synthetic data and real data. The discrimination results are fed back to the generator and the discriminator. Their performance is evaluated through the loss function, and the parameters are updated to gradually improve the capabilities of the generator and the discriminator. This makes the synthetic data generated by the generator gradually close to the real data, and the discriminator has difficulty distinguishing between the synthetic data generated by the generator and the real data. The trained adversarial neural network model is used as a data augmentation model.

[0087] Each time the generator generates synthetic data, the discriminator verifies whether the distribution of the synthetic data is similar to that of the test data through the KS test method, realizes the discrimination between real data and synthetic data, feeds back the learned rules to the generator, and repeats the process until the discriminator cannot recognize the synthetic data generated by the generator. The loss function is used to evaluate the GAN model during model training. The performance of the model in generation and discrimination tasks is quantified by calculating the loss function values ​​of the generator and the discriminator respectively. A smaller generator loss value indicates that the generated data is realistic, while a smaller discriminator loss value indicates a strong ability to distinguish data. The loss function serves as an optimization target to guide the training process of the generator and the discriminator. The model parameters are adjusted through back propagation to gradually improve the generation and discrimination capabilities of the model. The process of evaluating the model using the loss function is a mature existing technology and will not be repeated here.

[0088] The novel data-driven cement clinker system design method provided in this embodiment utilizes the continuous confrontation and optimization of the generator and the discriminator in the adversarial neural network to make the synthetic data closer and closer to the real cement clinker data. The data augmentation model obtained through multiple iterative training can generate more realistic augmented data based on the cement clinker test data set.

[0089] Step S2013, using the cement clinker augmented data set to train a preset performance prediction model to obtain a trained performance prediction model.

[0090] The novel data-driven cement clinker system design method provided in this embodiment utilizes a data augmentation model to expand the data volume of the cement clinker data set, thus breaking through the limitation on the design of the cement clinker system under the condition of a small number of data samples, increasing the number of optional cement clinker systems, and facilitating the training of a more accurate performance prediction model.

[0091] In some optional implementations, the parameters of cement clinker augmented data in the cement clinker augmented data set include: cement clinker system and performance index; the above step S2013 includes:

[0092] Step b1, taking the cement clinker system in the cement clinker augmented data as the input of the preset performance prediction model, and taking the performance index in the cement clinker augmented data as the output of the preset performance prediction model, to form a sample data set.

[0093] Specifically, a preset performance prediction model for cement clinker is built using a fully connected layer based on a neural network. The main parameters of the model are the number of neural layers, the number of neurons, the learning rate, the activation function, the amount of single training data, the number of training rounds, the optimizer, and the weight decay rate. The number of neural layers is generally set to 2-15; the number of neurons is generally set to 4-256; the learning rate is used to control the speed and direction of the iteration of the parameters of the cement clinker performance prediction model, and is generally set to 10 -2-10 -5 ; The activation function is used to enhance the ability of cement clinker performance prediction model to handle nonlinear problems. Generally, one or more of sigmoid and ReLu are selected. The amount of single training data and the number of training rounds are used to update the parameters of cement clinker performance prediction model. They are controlled by the amount of data and are generally set to 16-256 and 1000-2000 respectively. The optimizer is used to improve the performance and generalization ability of cement clinker performance prediction model. Generally, Adam, SGD or AdamW are selected. The weight decay rate is an important parameter to prevent the model from overfitting. By limiting the model parameters, the model has better generalization ability. It is generally set to 10 -3 -5*10 -1 .

[0094] Step b2, dividing the sample data set into a training set and a test set according to a preset ratio, training to obtain a performance prediction model, and the sample data in the test set are all cement clinker test data.

[0095] Specifically, during the training process, the cement clinker system in the cement clinker augmented data is used as the input of the preset performance prediction model, and the performance index in the cement clinker augmented data is used as the output of the preset performance prediction model to form a sample data set. The sample data set is divided into a training set and a test set, wherein 85% of the data is used as the training set and 15% of the data is used as the test set. The test set is all real data in the cement clinker test data set. The mean square error can be used as an evaluation index of the cement clinker performance prediction model, which is only an example, but not limited to this.

[0096] The novel data-driven cement clinker system design method provided in this embodiment uses the cement clinker augmented data set to train a performance prediction model, expands the number of sample data for model training, and improves the performance and generalization ability of the cement clinker performance prediction model.

[0097] Step S202, generating multiple cement clinker systems randomly or by using a genetic algorithm, and obtaining a predicted value of a performance index of each cement clinker system by using a trained performance prediction model.

[0098] Specifically, an optimization algorithm is added on the basis of the performance prediction model. The optimization algorithm can be a genetic algorithm. The main parameters include clinker gene length, population size, mutation and crossover rate, number of generations, etc. The gene length is 6-16 (including mineral types and composition, as well as hydration product types and composition), the initial population size is 100-300, the crossover probability is 0.4-0.8, the mutation probability is 0.05-0.3, and the number of generations is 100-500. The process of using genetic algorithms to generate multiple cement clinker systems specifically includes:

[0099] Step S2021, randomly generating an initial population including a first number of cement clinker systems, and inputting the first number of cement clinker systems into a performance prediction model to obtain performance index prediction values ​​of each cement clinker system.

[0100] Specifically, the gene code of each individual in the genetic algorithm represents a possible solution. In this embodiment, the gene code of the individual is the composition ratio of each mineral phase in the cement clinker system.

[0101] Assuming that clinker is mainly composed of the following five components: C3S, C2S, C3A, C4AF, and gypsum, the genes of each individual can be represented as the proportion of these five components. For example, using real number coding, each individual can be represented as a five-dimensional vector: individual = [C3S, C2S, C3A, C4AF, gypsum]. Each value in this five-dimensional vector represents the proportion of the component in the clinker, and it is assumed that the sum of these proportions is 100%.

[0102] Randomly generate multiple possible clinker mineral ratios as the initial population. Each individual is a clinker ratio scheme, and the population size can be set to 100-300 individuals to ensure diversity.

[0103] The mineral ratio of each individual is randomly generated within a certain range. For example: C3S: 30%-70%, C2S: 10%-50%, C3A: 5%-15%, C4AF: 5%-15%, gypsum: 3%-5%. By randomly generating a variety of ratios, we ensure that the initial population covers a wide range of search space.

[0104] The cement clinker system represented by each individual in the initial population is input into the performance prediction model to simulate and predict the actual performance of each individual in the initial population, such as compressive strength, carbon emissions, etc.

[0105] Step S2022, according to the predicted values ​​of the performance indicators of each cement clinker system, the fitness of each individual in the initial population is calculated using a preset fitness function.

[0106] Specifically, the fitness function can be expressed as a single-objective or multi-objective optimization problem. For example, if compressive strength and carbon emissions are used as the optimization targets in this embodiment, the fitness function can be expressed as the weighted sum of the two: fitness = α × compressive strength + β × carbon emissions, which is only used as an example, but not limited to this. The performance prediction value obtained by the performance prediction model is substituted into the fitness function to calculate the fitness value of each individual. The higher the individual fitness value, the closer the performance of the individual is to the design target.

[0107] Step S2023, according to the fitness of each individual in the initial population, each individual is selected, crossed, and mutated to obtain a new generation population.

[0108] Specifically, selection refers to selecting individuals with higher fitness from the current population and entering the next generation of reproduction. The purpose of selection is to give priority to retaining individuals with higher fitness so that they can inherit excellent gene combinations in the subsequent reproduction process.

[0109] Crossover is a process that simulates genetic recombination, which produces new offspring individuals by combining the genes of two parents. Crossover operations help explore new mineral combinations. Common crossover methods include:

[0110] (1) Single-point crossover: Select a crossover point in the parent gene, and take the genes before the crossover point from parent 1 and the genes after the crossover point from parent 2. For example:

[0111] Parent 1 is: 60%, 20%, 10%, 5%, 3%, 2%60\%, 20\%, 10\%, 5\%, 3\%, 2\%60%, 20%, 10%, 5%, 3%, 2%;

[0112] Parent 2 is: 50%, 30%, 10%, 4%, 4%, 2% 50\%, 30\%, 10\%, 4\%, 4\%, 2\% 50%, 30%, 10%, 4%, 4%, 2%;

[0113] The offspring produced after crossover may be: Offspring 1: 60%, 20%, 10%, 4%, 4%, 2% 60\%, 20\%, 10\%, 4\%, 4\%, 2\% 60%, 20%, 10%, 4%, 4%, 2%.

[0114] (2) Two-point crossover: Genes are exchanged between two points to generate more recombination possibilities. The crossover probability is generally set to 0.4-0.8, indicating that 40%-80% of individuals will undergo a crossover operation.

[0115] The mutation operation increases the diversity of the population and prevents the algorithm from falling into a local optimal solution by randomly modifying certain values ​​in the individual genes. For the mineral ratio of cement clinker, the mutation operation can be to randomly change the proportion of a certain mineral. For example: the C3S ratio in an individual is mutated from 60% to 58%. The mutation probability is generally set between 0.05-0.3. A lower mutation probability helps the stability of the algorithm.

[0116] A new generation of population is generated through selection, crossover and mutation operations. The individuals of each new generation of population are generated by the genes of the parent generation through crossover and mutation operations. These individuals inherit the excellent genes of the parent generation and also have a certain degree of diversity.

[0117] Step S2024, repeating the process of selecting, crossing over, and mutating each individual in the new generation population until the fitness of each individual in the population converges, and using a second number of cement clinker systems in the population as multiple cement clinker systems generated by the genetic algorithm, the second number being greater than the first number.

[0118] Specifically, the process of selecting, crossing over, and mutating each individual in the new generation population is repeated, the fitness value of each generation population is calculated, and individuals with better clinker ratios are screened out. Each generation of new populations will be continuously optimized through genetic operations and gradually approach the optimal solution. This process continues until the following termination conditions are met: the preset maximum number of generations (for example, 100-500 generations) is reached, or the fitness value in the population converges, indicating that the performance has approached the optimal solution.

[0119] After the algorithm terminates, the individual with the highest fitness value is selected as the final clinker ratio scheme, which should be able to meet the preset target performance, such as the best compressive strength and the lowest carbon emission. Genetic algorithms are used to process randomly acquired sample data to obtain individuals with higher fitness as cement clinker systems, and then generate cement clinker systems with performance indicators close to the preset requirements. After processing by the performance prediction model, it is easy to quickly select cement clinker systems that meet the preset requirements from the prediction results, thereby improving design efficiency.

[0120] The genetic algorithm is used as an optimization algorithm for the generation of cement clinker system as an example, but it is not limited to this. Genetic algorithm is one of the global search algorithms for reverse design of clinker with high strength and high durability. In addition, other optimization algorithms such as Bayesian optimization algorithm can also be used to achieve the purpose of designing low-carbon clinker.

[0121] Step S203, select a new cement clinker system that meets the preset requirements according to the predicted values ​​of the performance indicators of each cement clinker system. Figure 1 Step S103 of the illustrated embodiment will not be described in detail here.

[0122] The novel data-driven cement clinker system design method provided in this embodiment utilizes a data augmentation model to augment cement clinker data, thereby expanding the latent space of clinker systems that may have excellent performance, breaking through the limitation of small data samples on clinker composition systems, and predicting the performance of each augmented cement clinker composition system through a performance prediction model, thereby achieving the purpose of designing a clinker composition system according to performance requirements, shortening the research and development cycle of cement clinker, and reducing research and development costs.

[0123] In this embodiment, a new cement clinker system design method based on data drive is provided, which can be used in the above-mentioned computer system. Figure 3is a flow chart of a novel cement clinker system design method based on data drive according to an embodiment of the present invention, such as Figure 3 As shown, the process includes the following steps:

[0124] Step S301, constructing a cement clinker test data set, and augmenting the cement clinker test data set, using the augmented data set to train a preset performance prediction model to obtain a trained performance prediction model.

[0125] Specifically, the parameters of cement clinker test data include: cement clinker system and performance indexes. The above step S301 includes:

[0126] Step S3011, use the cement clinker test data set to train the teacher model to obtain a trained teacher model.

[0127] Specifically, a self-training algorithm is used, and the training process of the performance prediction model and the data augmentation process are performed alternately. The model involved in the data augmentation in the self-training algorithm is the teacher model, and the trained model is the student model. A teacher model is obtained by training a fully connected layer based on a neural network using the cement clinker test data set. The role of the teacher model is to input the cement clinker system and obtain the corresponding performance index value.

[0128] Step S3012, randomly or by using a genetic algorithm to generate multiple cement clinker systems, and input the trained teacher model to obtain corresponding performance indicators. Multiple cement clinker systems and corresponding performance indicators constitute synthetic data.

[0129] Specifically, multiple cement clinker systems are generated randomly or by using a genetic algorithm, and the multiple cement clinker systems are input into a trained teacher model to obtain corresponding performance indicators, and the multiple cement clinker systems and the corresponding performance indicators constitute synthetic data.

[0130] Step S3013, the cement clinker data set and the synthetic data are combined into an augmented data set to jointly train the student model, so as to obtain a trained student model, wherein the model structure of the student model is the same as that of the teacher model;

[0131] Specifically, multiple cement clinker systems and corresponding performance indicators are merged to obtain multiple complete synthetic data, which are then combined with the cement clinker test data set to form an augmented data set for joint training to obtain a student model. The student model has the same network structure as the teacher model.

[0132] Step S3014, using the trained student model as the teacher model, repeating the process of randomly or using a genetic algorithm to generate multiple cement clinker systems, and calculating the loss function of the corresponding student model until the loss function of the student model converges, ending the cycle, and using the student model as the preset performance prediction model.

[0133] Specifically, the trained student model is used as the new teacher model, and multiple randomly generated cement clinker systems are input into the teacher model to obtain the corresponding performance indicators. The multiple cement clinker systems and the corresponding performance indicators are respectively merged into multiple synthetic data, and the new student model is trained together with the cement clinker test dataset. The above process is repeated until the MSE loss error of the student model converges, and the cycle is terminated. The student model obtained for the last time is used as the performance prediction model.

[0134] The novel data-driven cement clinker system design method provided in this embodiment uses a teacher model to generate performance indicators of each cement clinker system, expands the cement clinker data, and uses the expanded data to train a student model. The above process is repeated until the student model converges, thereby improving the generalization ability and learning efficiency of the model.

[0135] Step S302, randomly or by using a genetic algorithm to generate multiple cement clinker systems, and use the trained performance prediction model to obtain the performance index prediction value of each cement clinker system. Figure 2 Step S202 of the illustrated embodiment will not be described in detail here.

[0136] Step S303, select a new cement clinker system that meets the preset requirements according to the predicted values ​​of the performance indicators of each cement clinker system. Figure 2 Step S203 of the illustrated embodiment will not be described in detail here.

[0137] In a specific embodiment, Figure 4 As shown in the figure, the process of data augmentation using adversarial neural network is shown. The important parameters in the data augmentation model include: noise dimension is 100, number of neurons is 512, learning rate is 0.001, activation function is Relu in the first layer of the generator, and the final output is Tanh. Each layer of the network in the discriminator is LeakyRelu, the amount of single training data is 32, the number of training rounds is 1500, the number of random production samples is 1000, and the optimizer uses Adam. The input feature values ​​are clinker mineral phase composition / chemical composition, fineness, 3d / 28d hydration product composition, and the output value is the 28d compressive strength of clinker. The generated data is verified by the KS test method to see if it is similar to the distribution of the test data. The result is that the p-value is 0.901 and the KS value is 0.039. The results show that the synthetic data is similar to the test data under the model parameters.

[0138] A cement clinker performance prediction model is built using a fully connected layer based on a neural network. 85% of the data is used as a training set, and 15% of the data is used as a test set to train the model. There will be no synthetic data in the test set, all of which are real data. The mean square error is used as the evaluation indicator of the model. The performance prediction model has 6 neural layers, 32 neurons, a learning rate of 0.0004490821, an activation function of LeakyRelu, 64 single training data, 2000 training rounds, an optimizer of AdamW, and a weight decay rate of 0.034629. At this time, the mean square error of the performance prediction model is 2.86, and the correlation coefficient R 2 The value is 0.898, indicating that the prediction result of the model is good. After the model is trained, the randomly generated data (10,000) are put into the model, and the carbon emission is between 602-688kg CO2 / kg and the compressive strength is cement mortar strength>48.5MPa as the restriction conditions to screen out the clinker mineral phase that meets the requirements. Table 1 lists the clinker systems that meet the requirements screened by this method (taking the ratio of the first three groups of output as an example).

[0139] Table 1

[0140]

[0141] In another specific embodiment, Figure 5 As shown in the figure, the process of using the self-training strategy to realize data augmentation is shown. The data augmentation model uses a deep neural network as the basic framework, 85% of the data is used as the training set, and 15% of the data is used as the test set to train the data augmentation model. In the teacher model of the first training, the number of neural layers is 3, the number of neurons is 22, the learning rate is 0.00032817, the activation function is ReLu, the amount of single training data is 32, the number of training rounds is 2000, the optimizer is SGD, and the mean square error MSE of the model is 48.37. Subsequently, the new data predicted by the teacher model (15% of the original data set) is added to the original data set to train the model again. After 7 iterations of this process, the MSE reaches the lowest, and after 8 iterations, the model MSE begins to rise. At this time, the number of neural layers of the data augmentation model is 7, the number of neurons is 62, the learning rate is 0.00020555, the activation function is ReLu, the amount of single training data is 128, the number of training rounds is 2000, the optimizer is Adam, and the MSE is 3.01.

[0142] A genetic algorithm is added after the last generation of student model code to screen the clinker system that meets the requirements. The gene length of the algorithm is 6-16 (including mineral types and compositions, as well as hydration product types and compositions, determined by the generated clinker ratio), the initial population size is 300, the crossover probability is 0.4, the mutation probability is 0.3, and the number of generations is 500. Each population is restricted by carbon emissions between 645-731kg CO2 / kg, sulfate corrosion resistance coefficient > 0.95 after 90 dry-wet cycles, and no peeling on the surface of the specimen. Table 2 lists the clinker systems that meet the requirements screened by this method (taking the ratio of the first 3 groups as an example).

[0143] Table 2

[0144]

[0145] In this embodiment, a new cement clinker system design device based on data drive is also provided, which is used to implement the above-mentioned embodiments and preferred embodiments, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware is also possible and conceived.

[0146] This embodiment provides a new cement clinker system design device based on data drive, such as Figure 6 As shown, including:

[0147] The performance prediction model training module 601 is used to construct a cement clinker test data set, and to augment the cement clinker test data set, and to train a preset performance prediction model using the augmented data set to obtain a trained performance prediction model.

[0148] The performance index prediction module 602 is used to generate multiple cement clinker systems randomly or by using a genetic algorithm, and obtain the performance index prediction value of each cement clinker system by using a trained performance prediction model.

[0149] The system screening module 603 is used to screen out a new cement clinker system that meets the preset requirements according to the predicted values ​​of the performance indicators of each cement clinker system.

[0150] In some optional implementations, the performance prediction model training module 601 includes:

[0151] The data augmentation model training unit is used to train the adversarial neural network model according to the cement clinker test data set to obtain the data augmentation model.

[0152] The data augmentation unit is used to input the cement clinker test data set into the data augmentation model to obtain the cement clinker augmented data set.

[0153] The performance prediction model training unit is used to train a preset performance prediction model using the cement clinker augmented data set to obtain a trained performance prediction model.

[0154] In some optional implementations, the data augmentation model training unit includes:

[0155] The parameter initialization subunit is used to randomly initialize the parameters of the generator and discriminator.

[0156] The discriminator training subunit is used to train the discriminator according to the cement clinker test data set, so that the discriminator can identify real data, and the real data is the data in the cement clinker test data set.

[0157] The generator training subunit is used to generate synthetic data using the generator according to the cement clinker test data set, and to distinguish the synthetic data from the real data using the discriminator to obtain the distinction result.

[0158] The loss function calculation subunit is used to calculate the loss function of the discriminator and the loss function of the generator using the differentiation results.

[0159] The iterative training subunit is used to repeat the above process of using the generator to generate synthetic data and using the discriminator to distinguish the synthetic data from the real data, until the difference between the loss function of the discriminator and the preset discrimination threshold is less than the first preset difference threshold, and the difference between the loss function of the generator and the preset generation threshold is less than the second preset difference threshold, so as to obtain the trained generator and discriminator to form a data augmentation model.

[0160] In some optional implementations, the performance prediction model training unit includes:

[0161] The sample determination subunit is used to take the cement clinker system in the cement clinker augmented data as the input of the preset performance prediction model, and take the performance indicators in the cement clinker augmented data as the output of the preset performance prediction model to form a sample data set.

[0162] The performance prediction model training subunit is used to divide the sample data set into a training set and a test set according to a preset ratio, and train the performance prediction model. The sample data in the test set are all cement clinker test data.

[0163] In some optional implementations, the performance prediction model training module 601 further includes:

[0164] The teacher model training unit is used to train the teacher model using the cement clinker test data set to obtain a trained teacher model.

[0165] The data generation unit is used to generate multiple cement clinker systems randomly or by using a genetic algorithm, and input the trained teacher model to obtain corresponding performance indicators. Multiple cement clinker systems and corresponding performance indicators constitute synthetic data.

[0166] The student model training unit is used to combine the cement clinker data set and the synthetic data into an augmented data set to jointly train the student model, so as to obtain a trained student model. The model structure of the student model is the same as that of the teacher model.

[0167] The model iteration training unit is used to use the trained student model as the teacher model, repeat the process of randomly or using a genetic algorithm to generate multiple cement clinker systems, and calculate the loss function of the corresponding student model until the loss function of the student model converges, and use the student model as the preset performance prediction model.

[0168] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0169] The data-driven new cement clinker system design device in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0170] The embodiment of the present invention also provides a computer device having the above Figure 6 The new cement clinker system design device based on data-driven is shown.

[0171] See also Figure 7 , Figure 7 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Figure 7 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 7A processor 10 is taken as an example.

[0172] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0173] The memory 20 stores instructions executable by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0174] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0175] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0176] The computer device further comprises a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0177] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0178] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0179] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A new cement clinker system design method based on data drive, characterized in that: The method comprises: Constructing a cement clinker test data set, and augmenting the cement clinker test data set, and using the augmented data set to train a preset performance prediction model to obtain a trained performance prediction model; Generating a plurality of cement clinker systems randomly or by using a genetic algorithm, and obtaining a predicted value of a performance index of each cement clinker system by using the trained performance prediction model; According to the predicted values ​​of performance indicators of each cement clinker system, a new cement clinker system that meets the preset requirements is screened out.

2. The method according to claim 1, characterized in that: The method of augmenting the cement clinker test data set and using the augmented data set to train a preset performance prediction model to obtain a trained performance prediction model includes: Training an adversarial neural network model according to the cement clinker test data set to obtain a data augmentation model; Inputting the cement clinker test data set into the data augmentation model to obtain a cement clinker augmented data set; The cement clinker augmented data set is used to train a preset performance prediction model to obtain a trained performance prediction model.

3. The method according to claim 2, characterized in that The adversarial neural network model includes: a generator and a discriminator. The data augmentation model is obtained by training a preset data augmentation model according to the cement clinker test data set, including: Randomly initialize the parameters of the generator and discriminator; The discriminator is trained according to the cement clinker test data set so that the discriminator can identify real data, where the real data is data in the cement clinker test data set; According to the cement clinker test data set, a generator is used to generate synthetic data, and a discriminator is used to distinguish the synthetic data from the real data to obtain a distinction result; Calculating the loss function of the discriminator and the loss function of the generator using the distinction result; Repeat the above process of using the generator to generate synthetic data and using the discriminator to distinguish the synthetic data from the real data, until the difference between the loss function of the discriminator and the preset discrimination threshold is less than the first preset difference threshold, and the difference between the loss function of the generator and the preset generation threshold is less than the second preset difference threshold, to obtain a trained generator and discriminator to form a data augmentation model.

4. The method according to claim 2 or 3, characterized in that: The parameters of cement clinker augmented data in the cement clinker augmented data set include: cement clinker system and performance index; The method of using the cement clinker augmented data set to train a preset performance prediction model to obtain a trained performance prediction model includes: The cement clinker system in the cement clinker augmented data is used as the input of the preset performance prediction model, and the performance index in the cement clinker augmented data is used as the output of the preset performance prediction model to form a sample data set; The sample data set is divided into a training set and a test set according to a preset ratio, and a performance prediction model is obtained through training. The sample data in the test set are all cement clinker test data.

5. The method according to claim 1, characterized in that The parameters of the cement clinker test data include: cement clinker system and performance indicators; the method of augmenting the cement clinker test data set, using the augmented data set to train a preset performance prediction model to obtain a trained performance prediction model, also includes: Using the cement clinker test data set to train a teacher model to obtain a trained teacher model; Generating a plurality of cement clinker systems randomly or by using a genetic algorithm, and inputting the trained teacher model to obtain corresponding performance indicators, wherein the plurality of cement clinker systems and the corresponding performance indicators constitute synthetic data; The cement clinker test data set and the synthetic data are combined into an augmented data set to jointly train a student model to obtain a trained student model, wherein the student model has the same model structure as the teacher model; The trained student model is used as the teacher model, and the process of randomly or genetically generating multiple cement clinker systems is repeated, and the loss function of the corresponding student model is calculated until the loss function of the student model converges, and the student model is used as the preset performance prediction model.

6. The method according to claim 1 or 5, characterized in that: The method of using a genetic algorithm to generate multiple cement clinker systems comprises: Randomly generating an initial population including a first number of cement clinker systems, and inputting the first number of cement clinker systems into a performance prediction model to obtain performance index prediction values ​​of each cement clinker system; According to the predicted values ​​of the performance indicators of each cement clinker system, the fitness of each individual in the initial population is calculated using a preset fitness function; According to the fitness of each individual in the initial population, each individual is selected, crossed, and mutated to obtain a new generation of population; The process of selecting, crossing over and mutating each individual in the new generation population is repeated until the fitness of each individual in the population converges, and a second number of cement clinker systems in the population is used as multiple cement clinker systems generated by the genetic algorithm, wherein the second number is greater than the first number.

7. A new cement clinker system design device based on data drive, characterized in that: The device comprises: A performance prediction model training module is used to construct a cement clinker test data set, and to augment the cement clinker test data set, and to train a preset performance prediction model using the augmented data set to obtain a trained performance prediction model; A performance index prediction module is used to generate multiple cement clinker systems randomly or by using a genetic algorithm, and obtain the performance index prediction value of each cement clinker system by using the trained performance prediction model; The system screening module is used to screen out new cement clinker systems that meet the preset requirements based on the predicted values ​​of the performance indicators of each cement clinker system.

8. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method according to any one of claims 1 to 6 by executing the computer instructions.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that The method comprises computer instructions for causing a computer to execute the method according to any one of claims 1 to 6.

Images