A concrete mix proportion design method for the plateau of the Tibetan region

Abstract

The application relates to the fields of building engineering concrete design and deep learning, and discloses a concrete mix proportion design method for the Tibetan plateau, which comprises the following steps: obtaining design requirements, the contents of the design requirements including environmental parameters, raw material parameters and target performance; preprocessing the design requirements to generate requirement data; loading a concrete mix proportion design model, generating a preliminary mix proportion according to the requirement data; based on the preliminary mix proportion, carrying out triple optimization aiming at the altitude, low-temperature environment and construction environment of the Tibetan plateau to generate a fourth mix proportion; the triple optimization includes high-altitude adaptability optimization, low-temperature adaptability optimization and local adaptability optimization; and carrying out performance verification and closed-loop feedback on the fourth mix proportion. According to the above technical scheme, the concrete can meet the extreme requirements of the Tibetan area, the engineering quality standard rate can be effectively improved, the design cycle can be reduced, the local raw materials of the Tibetan area can be adapted, and the construction window period can be shortened to adapt to the engineering environment of the engineering practice.

Description

Technical Field

[0001] This invention relates to the fields of concrete design and deep learning in building engineering, and more specifically, to a concrete mix design method for the Tibetan Plateau. Background Technology

[0002] Concrete is an indispensable material in modern construction engineering, widely used in civil engineering, water conservancy and hydropower projects, and other fields. The concrete mix proportion is a crucial factor in ensuring project quality. A suitable concrete mix proportion needs to be determined during the design phase to match the construction environment and usage conditions, avoiding problems such as corrosion, leakage, and cracking. The high-altitude characteristics of the Tibetan plateau lead to increasingly higher quality requirements and construction difficulties in water conservancy and hydropower projects, placing higher demands on concrete mix design: 1) In terms of environment, the construction areas in the Tibetan plateau are characterized by high altitude, extreme cold, and low air pressure. Designs based on traditional concrete design concepts and requirements are difficult to adapt to the extreme environment of the plateau. Therefore, concrete mix design usually relies on experience and extensive testing, which is not only time-consuming and labor-intensive but also difficult to meet the performance requirements of modern engineering projects; 2) In terms of construction, transportation is difficult and conditions are harsh, while the demand for concrete is large, thus requiring local sourcing of materials; 3) Regarding raw materials, to reduce transportation pressure, locally sourced mineral admixtures have different chemical compositions. If the mix proportion is not dynamically adjusted according to the actual performance of the raw materials, it will affect the final performance of the concrete.

[0003] In practical engineering implementation, the high altitude, extreme low oxygen, extreme temperature differences, and permafrost environment of the Tibetan Plateau have a significant impact on the hydration reaction of concrete. Mix proportions designed based on empirical formulas often suffer from delayed strength development and substandard frost and crack resistance. Although existing technologies have developed methods to assist concrete mix proportion design using intelligent predictive models, they do not address the multi-factor coupling of plateau scenarios and extreme environments. They lack improvements tailored to the plateau environment and local raw material characteristics, resulting in poor model generalization ability. In other words, existing technologies do not account for the simultaneous coupling of multiple factors affecting concrete performance, including the plateau environment (altitude, temperature difference, pressure), local raw material characteristics (uniform aggregate lithology, large gradation fluctuations, and cementitious material activity affected by low oxygen), and mix proportion parameters (water-cement ratio, admixture ratio, etc.). They fail to establish a complete "environment-materials-mix proportion-performance" mapping, thus failing to guarantee the reliability of mix proportion design results.

[0004] Therefore, there is an urgent need for a concrete mix design method that can comprehensively consider factors such as raw material characteristics, engineering requirements, construction technology and environmental conditions in the Tibetan Plateau region, so as to improve the performance and design efficiency of concrete. Summary of the Invention

[0005] To achieve the above objectives, this application provides a concrete mix design method for the Tibetan Plateau region, comprising the following steps:

[0006] Obtain design requirements, which include environmental parameters, raw material parameters, and target performance.

[0007] Preprocess design requirements to generate requirement data;

[0008] Load the concrete mix design model and generate a preliminary mix proportion based on the required data;

[0009] Based on the initial mix design, a fourth mix design was generated by triple optimization considering the altitude, low temperature environment, and construction environment of the Tibetan Plateau. The triple optimization includes: high altitude adaptability optimization, low temperature adaptability optimization, and local adaptability optimization.

[0010] Performance verification and closed-loop feedback were conducted for the fourth mixing ratio.

[0011] Before loading the concrete mix design model, the concrete mix design model is constructed, including the following steps:

[0012] Obtain raw data of historical engineering projects on the Tibetan Plateau; the raw data includes extreme environmental parameters of the Tibetan region, local raw material parameters, engineering indicators, mix proportion data and measured performance data; the raw data includes the final performance data of hydraulic concrete mix proportion tests in various Tibetan regions, reflecting the mapping relationship between "environment-materials-mix proportion-performance";

[0013] Preprocess the raw data, determine the dimensions of the mix design, and generate the design dataset;

[0014] Define the network model; the network model is used to construct a mapping relationship that supports "environment-materials-mix proportions-performance" to calculate the concrete mix proportions;

[0015] By designing a dataset, the network model is trained to construct a concrete mix design model.

[0016] Preprocessing includes:

[0017] Remove anomalous data specific to the Tibetan region and use the environmental clustering-K-nearest neighbor algorithm to fill in missing values;

[0018] Standardization processes include: normalizing continuous parameters; using unique thermal coding for categorical parameters; and quantifying and classifying unique parameters. Continuous parameters include: altitude and extreme temperature; categorical parameters include: permafrost type and aggregate lithology; and unique parameters include: low-oxygen and low-temperature hydration activity and freeze-thaw stability.

[0019] By using gradient boosting tree (XGBoost) and mutual information method, core features are selected from the standardized data, retaining features that are strongly correlated with the performance of the Tibetan plateau.

[0020] The structure of the network model includes the following parts:

[0021] The input layer is used to receive core feature parameters. The types of core feature parameters include Tibetan extreme environment parameters, Tibetan local raw material parameters, mixing ratio parameters, performance requirements, and measured performance data of the mixing ratio.

[0022] The CNN module uses a 3-layer convolutional network to extract the local coupling features of the "material-environment-performance" feature dimension of the input data.

[0023] The Transformer module employs a two-layer multi-head attention mechanism to capture global dependencies and establish a three-way mapping feature of "material-environment-mixing ratio-performance".

[0024] A correction module specifically designed for the Tibetan Plateau is used to address the unique characteristics of the Tibetan Plateau and achieve dual corrections tailored to the environmental features of the Tibetan region.

[0025] The fully connected layer uses a 3-layer network to achieve feature mapping of additive proportion parameters adapted to Tibetan areas;

[0026] The output layer is used to output the preliminary mix proportions and performance predictions.

[0027] The Tibetan Plateau-specific correction module includes a dual-branch fully connected layer, comprising:

[0028] The extreme environment correction branch is used to input influencing factors and outputs the hydration rate correction coefficient and frost heave resistance adjustment factor based on the existing dataset model. The influencing factors include altitude, extreme minimum temperature, permafrost type and sunshine duration.

[0029] The local material adaptation branch is used to input local aggregate characteristics and local cementitious material properties, and output material usage optimization coefficients.

[0030] Furthermore, training the network model includes:

[0031] The weighted mean square error is used as the loss function, highlighting the key performance requirements of projects in Tibetan areas.

[0032] For training samples at altitudes above 4,000 meters or located in permafrost regions, the sample weights are adjusted to make the model focus more on prediction accuracy under extreme high-altitude and cold conditions.

[0033] Introduce a dual regularization mechanism into the network structure;

[0034] The model is divided into three elevation ranges and two permafrost types, forming six sub-regions with clear physical meanings. For each sub-region, the global model is fine-tuned using a design dataset.

[0035] Furthermore, high-altitude adaptability optimization refers to: adjusting the initial mix proportion to generate a second mix proportion for high altitudes; including increasing the total amount of cementitious materials in the initial mix proportion and adding highly active local admixtures to the initial mix proportion.

[0036] Low-temperature adaptability optimization refers to: optimizing the second formula for resistance to frost heave in low-temperature environments to generate a third formula, including: reducing the water-cement ratio, increasing the amount of air-entraining agent, and controlling the bubble spacing coefficient to be less than a specified threshold.

[0037] Local adaptation optimization refers to increasing the amount of local admixtures in the third mix proportion to generate the fourth mix proportion.

[0038] This invention establishes a precise mapping relationship between "environment-materials-mix ratio-performance" in Tibetan areas through "exclusive data for Tibetan regions + targeted models + targeted optimization," enabling concrete mix ratios to meet the extreme requirements of Tibetan areas, effectively improving the project quality compliance rate, providing an economical and feasible technical solution for water conservancy and hydropower projects in Tibetan areas, and reducing the design cycle. In terms of project implementation, it adapts to local raw materials in Tibetan areas and shortens the construction window to suit the actual engineering environment. This invention can be applied to various hydraulic structures in Tibetan areas, providing technical support for the construction of water conservancy infrastructure in Tibetan areas. Attached Figure Description

[0039] Figure 1 This is a step diagram of the method for constructing a concrete mix design model for the Tibetan Plateau according to an embodiment of the present invention;

[0040] Figure 2 This is a schematic diagram of the construction architecture of a concrete mix design model for the Tibetan Plateau, provided by an embodiment of the present invention. Detailed Implementation

[0041] To address the effectiveness of concrete mix design in the Tibetan plateau environment, this invention proposes a technical solution of "dedicated big data construction + extreme environment coupled deep learning model + integrated design process," the architecture of which is as follows: Figure 2 As shown, a concrete mix design model is constructed through a data layer and a model layer. The preliminary mix proportion is obtained by using the concrete mix design model in the design layer, and the mix proportion is optimized for adaptability to the Tibetan plateau environment. Based on the combination of the Tibetan plateau environment, raw material parameters and performance requirements, the entire process of precise mix design is realized.

[0042] The specific implementation of the present invention will now be described in detail with reference to the accompanying drawings.

[0043] The method for constructing a concrete mix design model for the Tibetan Plateau provided by this invention is as follows: Figure 1 As shown, it includes the following steps:

[0044] Step S100: Obtain design requirements;

[0045] The design requirements specifically include:

[0046] Environmental parameters, including altitude, atmospheric pressure, extreme temperature, permafrost type, diurnal temperature range, and usage environment (underwater / exposed / freeze-thaw zone), etc.

[0047] Raw material parameters, including aggregate origin, lithology, particle size distribution, freeze-thaw stability, cementitious material type, manufacturer, low-oxygen hydration activity, and other information, as well as admixture and additive manufacturer, type and model information;

[0048] Target performance includes requirements for compressive strength grade (C15~C50), freeze resistance grade (F300~F400), crack resistance grade (≥LI), and low-temperature workability.

[0049] Step S110: Preprocess the design requirement data to generate requirement data;

[0050] Specifically, the preprocessing process includes: normalizing continuous parameters such as altitude, atmospheric pressure, and extreme temperature; using unique thermal coding for classification parameters such as permafrost type and aggregate lithology; quantitatively classifying parameters unique to the Tibetan region, such as low-oxygen and low-temperature hydration activity and freeze-thaw stability, into levels 1 to 5; and quantitatively classifying the transportation costs of the manufacturer information involved in the raw material parameters.

[0051] Step S120: Load the concrete mix design model and generate a preliminary mix proportion based on the required data.

[0052] Before loading the concrete mix design model, the concrete mix design model is constructed, including the following steps:

[0053] Step S121: Obtain the raw data of historical engineering projects in the Tibetan Plateau; the raw data includes extreme environmental parameters in the Tibetan region, local raw material parameters, engineering indicators, mix proportion data and measured performance data;

[0054] In this step, historical engineering data from dozens of hydraulic structures in the Tibetan region are collected as raw data. The raw data covers typical scenarios such as high-altitude reservoirs and water conveyance projects in permafrost areas. The data selection highlights the characteristics of the Tibetan plateau and specifically includes the following types:

[0055] 1) Extreme environmental parameters in Tibet: The extreme environment in Tibet is characterized by atmospheric pressure, extreme temperature, permafrost type, frost heave level, ultraviolet radiation intensity, and glacial meltwater pH value at characteristic altitudes above 4000m. Therefore, the raw data obtained include altitude (2800~5500m), atmospheric pressure, average annual temperature, extreme temperature, diurnal temperature range, duration of cold season, ultraviolet radiation intensity, permafrost type (including: seasonal permafrost and permafrost), frost heave level (I~V), and glacial meltwater pH value (4.5~7.0).

[0056] 2) Parameters of local raw materials in Tibet: The characteristics of local raw materials in Tibet are reflected in: the freeze-thaw cycle stability of aggregates, the low-oxygen and low-temperature hydration activity of cementitious materials, the UV resistance of admixtures, and the low-temperature workability of additives; therefore, the raw data obtained in this step includes:

[0057] Aggregate characteristics from local / adjacent areas in Tibet include: lithology (basalt / granite / limestone), particle size distribution, moisture content, apparent density, freeze-thaw cycle stability (mass loss rate after 50 freeze-thaw cycles), and salinity.

[0058] Characteristics of cementitious materials: model, strength grade, low-oxygen and low-temperature hydration activity (hydration degree at 5℃ for 7 days), specific surface area;

[0059] Characteristics of local admixtures: type (fly ash / slag powder / volcanic ash), grade, activity index, antifreeze modification effect, and UV aging resistance;

[0060] Additive characteristics: type (Tibet-adapted air-entraining water-reducing agent / antifreeze agent), model, water reduction rate, air entrainment rate, and low-temperature (-25℃~-5℃) workability;

[0061] 3) Mix proportion parameters: Commonly used mix proportion parameters in Tibetan areas are characterized by: air-entraining agent dosage ≥ 5 / 10,000, antifreeze dosage considered, and total amount of cementitious materials ≥ 400 kg / m³. 3 Therefore, the raw data obtained in this step includes: cement dosage, local admixture dosage, aggregate dosage (coarse / fine), admixture dosage, air-entraining agent dosage, antifreeze agent dosage, cementitious materials, water-cement ratio, sand ratio, and total amount of cementitious materials.

[0062] 4) Performance Requirements: The performance requirements for use in Tibetan areas include: frost resistance grade F300~F400, frost heave coefficient, strength retention rate after UV aging, low-temperature workability, and UV aging strength retention rate. Therefore, the raw data obtained in this step include: compressive strength (7d, 28d, 90d), frost resistance grade (F300~F400), crack resistance grade (L-Ⅰ~L-Ⅲ), frost heave coefficient, workability (slump at low temperature), peak heat of hydration (≤50℃), and strength retention rate after UV aging (≥85%).

[0063] 5) Actual performance data of mix proportion: Based on the requirements of the project in the Tibetan area, the performance indicators that need to be paid attention to include: environment (altitude, air pressure, sunshine duration, ultraviolet intensity, etc.), materials, mix proportion parameters, and performance (thermal properties, durability, mechanical properties).

[0064] The raw data obtained in this step includes the final performance data of hydraulic concrete mix proportion tests in various Tibetan areas. Each set of mix proportion data can form a mapping relationship of "environment-material-mix proportion-performance".

[0065] Step S122: Preprocess the raw data, determine the mix design dimensions, and generate the design dataset;

[0066] Specifically, preprocessing of different types of data in the original data includes:

[0067] 1) Data cleaning: Remove abnormal data unique to the Tibetan area, and then use environmental clustering-K nearest neighbor algorithm to fill in missing values. For example, remove data with abnormally high intensity under low oxygen conditions and data with excessive frost heave coefficient in permafrost areas, and then fill in missing values ​​by deriving frost heave level through altitude and permafrost type.

[0068] 2) Standardization processing; including normalization of continuous parameters such as altitude and extreme temperature; use of unique thermal coding for classification parameters such as permafrost type and aggregate lithology; and quantitative classification (level 1 to 5) of parameters unique to the Tibetan region such as low-oxygen and low-temperature hydration activity and freeze-thaw stability.

[0069] 3) Feature engineering processing: Core features are selected from the standardized data using gradient boosting tree XGBoost and mutual information method, with a focus on retaining features that are strongly correlated with the performance of the Tibetan plateau, reflecting the mix design dimension. The mix design dimension includes: altitude-water-cement ratio coupling term, diurnal temperature difference-cracking resistance grade, aggregate freeze-thaw stability-frost heave coefficient, and low oxygen activity of cementitious materials-strength and other related features.

[0070] The above preprocessing steps generate the design dataset.

[0071] like Figure 2 As shown, this step constructs the data layer of the concrete mix design model based on the design dataset. In the data layer, the dataset is divided into training set, validation set and test set in a 7:2:1 ratio to ensure that the training set covers different altitudes (2800~5000m) and different lithologies (basalt, granite, dacite, limestone, sandstone and other common plateau aggregates), thereby improving the model's plateau generalization ability.

[0072] The present invention provides an embodiment in which the effective sample size in the dataset is ≥6000 groups, covering more than 50 counties in the Tibetan region and the entire range of altitude from 2800 to 5500m, including ≥2 types of permafrost and ≥80 local raw material manufacturers, to ensure that the environmental and raw material characteristics of concrete cover the Tibetan region.

[0073] Step S123: Define the network model. The network model structure includes a CNN module, a Transformer module, and a correction module. Based on the design dataset, it can construct a multi-dimensional mapping relationship that supports "environment-material-mix ratio-performance" to calculate the concrete mix ratio. Among them, the correction module adopts a double-branch fully connected layer to realize the correction of the material dimension and the environment dimension.

[0074] The network model defined in this step is as follows: Figure 2 The model layers shown specifically include:

[0075] 1) Input layer, used to receive core feature parameters. Core feature parameters can reflect the multi-dimensional mapping relationship of "environment-material-mixing ratio-performance". The types include extreme environment parameters in Tibet, local raw material parameters in Tibet, mixing ratio parameters, performance requirements, and measured performance data of mixing ratio.

[0076] 2) The CNN module uses a 3-layer convolutional network to extract the local coupling features of the "material-environment-performance" feature dimension of the input data, such as the correlation features of aggregate freeze-thaw stability-frost heave coefficient and cementitious material low oxygen activity-strength.

[0077] 3) The Transformer module uses a two-layer multi-head attention mechanism to capture global dependencies and establish a three-way mapping feature of "material-environment-mix ratio-performance", such as: altitude-water-cement ratio-early strength, frozen soil type-air-entraining agent content-frost heave resistance and other three-way mapping features;

[0078] 4) Tibetan Plateau-Specific Correction Module: Addressing the unique characteristics of the Tibetan Plateau, including high altitude, frigid conditions, low air pressure, and strong radiation, a new dual-branch fully connected layer has been added to achieve dual corrections tailored to the Tibetan environment, including:

[0079] Extreme Environment Correction Branch: Since the strength and durability will change at high altitudes, such as above 4000m, the extreme environment correction branch increases the hydration rate and frost heave resistance to adapt to the actual extreme environment of the Tibetan Plateau. The extreme environment correction branch takes into account factors such as altitude, extreme minimum temperature, permafrost type and sunshine duration, and outputs the hydration rate correction coefficient and frost heave resistance adjustment factor based on the pre-trained learning model.

[0080] Local Material Adaptation Branch: Materials native to the Tibetan Plateau typically possess unique properties and requirements compared to those in inland areas. For instance, aggregates exhibit higher freeze-thaw cycle stability, and cementitious materials demonstrate greater resistance to UV aging. Therefore, the local material adaptation branch inputs local aggregate characteristics and local cementitious material properties. Based on a pre-trained learning model, it outputs material usage optimization coefficients and appropriately prioritizes increasing the proportion of local admixtures to improve the economic efficiency of the mix design.

[0081] 5) The fully connected layer adopts a 3-layer network to realize the feature mapping of the admixture ratio parameters adapted to Tibetan areas, such as the precise dosage of air-entraining agent and antifreeze agent.

[0082] 6) The output layer is used to output the preliminary mix proportions and performance prediction values, such as 7d / 28d / 90d compressive strength, frost resistance grade, crack resistance grade, and frost heave resistance grade.

[0083] Step S124: Train the network model by designing the dataset to construct a concrete mix design model.

[0084] The training process for a network model includes the following:

[0085] 1) In terms of loss function design, weighted mean square error (MSE) is used as the loss function to highlight the high requirements of key performance indicators for Tibetan engineering. The three core indicators of 7-day compressive strength, frost heave coefficient and frost resistance grade are assigned a weight of 2.0 to 2.5 times to enhance the model's fitting accuracy on these key outputs and reflect the priority of performance under the special environment of the plateau.

[0086] 2) Regarding the optimizer and learning rate scheduling strategy, the AdamW optimizer was selected, with an initial learning rate set to 0.001. Considering the unique characteristics of the Tibetan plateau environment, a dedicated sample weighting and learning scheduling mechanism was designed: for training samples at altitudes above 4000 meters or located in permafrost regions, the sample weights were adjusted (e.g., the sample weight was set to multiply the loss contribution by 1.3), making the model more focused on prediction accuracy under extreme high-altitude and cold conditions.

[0087] 3) Regarding the regularization strategy, in order to improve the generalization ability of the model and prevent overfitting, a dual regularization mechanism is introduced into the network structure: a Dropout layer is added after the fully connected layer, with a dropout rate of 0.3; at the same time, L2 weight decay regularization is adopted, with a regularization coefficient λ = 0.0001.

[0088] 4) In terms of transfer learning, based on geographical and permafrost characteristics, the Tibetan region is divided into three altitude ranges (3000–4000m, 4000–5000m, 5000–5500m) and two permafrost types (seasonal permafrost and perennial permafrost), forming six sub-regions with clear physical meanings. For each sub-region, the global model is fine-tuned using a designed dataset to construct six dedicated sub-models, which significantly improves the predictive relevance and accuracy under different plateau-permafrost coupling environments.

[0089] Through the above process, a concrete mix design model is generated.

[0090] In this step, input the required data into the concrete mix design model to output the preliminary mix proportion. The preliminary mix proportion includes data such as the dosage of cement, local admixtures, aggregates, admixtures, air-entraining agent / antifreeze agent, water-cement ratio, and sand ratio. Among them, cement and local admixtures are cementing materials. Local admixtures also include high-activity local admixtures and ordinary local admixtures.

[0091] Step S130: Based on the preliminary mix proportions, a triple optimization is performed considering the altitude, low temperature environment, and construction environment of the Tibetan Plateau, including the following steps:

[0092] 1) High-altitude adaptability optimization, that is, adjusting the initial mix proportion for high altitude to generate a second mix proportion; for example, to adapt to the environment at an altitude >4000m and ensure that the 7d strength meets the standard, the total amount of cementitious materials is increased by 10%~15%, and highly active local admixtures are added.

[0093] 2) Low temperature adaptability optimization, that is, to optimize the second formula for low temperature environment to generate a third formula, including: reducing the water-cement ratio, increasing the amount of air-entraining agent, and controlling the bubble spacing coefficient to be less than a specified threshold (e.g., ≤200μm).

[0094] 3) Local adaptation optimization: In order to achieve localization of raw materials and economic optimization, and reduce transportation costs, raw materials from local / nearby areas of Tibet are selected, and the amount of local admixtures is increased in the third mix proportion to generate the fourth mix proportion.

[0095] Step S140: Performance Verification and Closed-Loop Feedback

[0096] In this step, a trial mix of one set of samples was conducted in the laboratory for the fourth mix proportion, simulating the low-temperature and low-oxygen environment of the Tibetan region for curing, and the strength, frost resistance, crack resistance, and frost heave resistance were tested at 7 days, 28 days, and 90 days. If the deviation is >8%, the verification data (including measured environmental parameters in the Tibetan region) will be fed back to the big data set, the model will be readjusted, iterative optimization will be carried out, and the generated concrete mix proportion design model will be improved.

[0097] Verification has shown that the concrete mix design model constructed in this invention outputs mix proportions with component dosage errors ≤5%, 7-day strength prediction errors ≤10%, 28-day / 90-day strength errors ≤8%, frost resistance / crack resistance prediction accuracy ≥97%, and frost heave coefficient prediction errors ≤0.0002. Therefore, the concrete mix design model is suitable for Tibetan areas at altitudes of 2800-5500m, requiring concrete design standards of compressive strength C15-C50, frost resistance F300-F400, crack resistance ≥LI, and frost heave coefficient ≤0.0015. It is compatible with pumping, compaction, and pouring construction techniques and is compatible with mainstream local raw material types in Tibetan areas.

[0098] This invention effectively overcomes many limitations of traditional concrete mix design methods, such as cumbersome processes, time-consuming and labor-intensive methods, lagging strength development, and difficulty in meeting durability standards. Simultaneously, it also addresses the shortcomings of existing intelligent design models, such as poor adaptability to the Tibetan plateau environment, insufficient coupling with local climate and geological characteristics, and insufficient generalization ability. Based on the construction of a design dataset and the completion of deep learning model training, this invention can generate hydraulic concrete mix proportions suitable for the Tibetan plateau environment according to the design requirements of specific projects, and supports multiple rounds of performance verification and feedback optimization: by feeding back experimental or simulation verification results to the model, it drives dynamic parameter adjustment and scheme iteration, ultimately outputting a high-performance, highly adaptable concrete mix proportion scheme that highly matches the design objectives and local extreme environmental conditions (such as high altitude, low oxygen, and strong freeze-thaw cycles).

[0099] The above-disclosed embodiments are merely a few specific examples of the present invention. However, the present invention is not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A method for concrete mix design in the Tibetan Plateau region, characterized in that, Includes the following steps: Obtain design requirements, which include environmental parameters, raw material parameters, and target performance. The design requirements are preprocessed to generate requirement data; A concrete mix design model is loaded, and a preliminary mix proportion is generated based on the required data. The structure of the concrete mix design model includes an input layer for receiving core feature parameters, the types of which include extreme environmental parameters of the Tibetan region, local raw material parameters of the Tibetan region, mix proportion parameters, performance requirements, and measured performance data of the mix proportion. A CNN module is used to extract the local coupling features of the "material-environment-performance" feature dimension of the input data of the input layer. The Transformer module employs a two-layer multi-head attention mechanism to capture global dependencies and establish a three-way mapping feature of "material-environment-mix ratio-performance". The fully connected layer uses a three-layer network to realize the feature mapping of the Tibetan-adaptive admixture mix ratio parameters. The output layer outputs the preliminary mix ratio and performance prediction values. The Tibetan Plateau-specific correction module is used to achieve dual corrections tailored to the unique characteristics of the Tibetan Plateau environment. This module includes a dual-branch fully connected layer: an extreme environment correction branch, which inputs influencing factors and outputs a hydration rate correction coefficient and a frost heave adjustment factor; these influencing factors include altitude, extreme minimum temperature, permafrost type, and sunshine duration; and a local material adaptation branch, which inputs local aggregate characteristics and local cementitious material properties and outputs material dosage optimization coefficients. Based on the initial mix proportion, a fourth mix proportion is generated through triple optimization considering the altitude, low temperature environment, and construction environment of the Tibetan Plateau. This triple optimization includes: high-altitude adaptability optimization, low-temperature adaptability optimization, and local adaptability optimization. High-altitude adaptability optimization refers to adjusting the initial mix proportion to generate a second mix proportion for high altitude conditions, including increasing the total amount of cementitious materials in the initial mix proportion and adding highly active local admixtures. Low-temperature adaptability optimization refers to optimizing the second mix proportion for frost heave resistance in low-temperature environments to generate a third mix proportion, including reducing the water-cement ratio and increasing the amount of air-entraining agent. Local adaptability optimization refers to increasing the amount of local admixtures in the third mix proportion to generate a fourth mix proportion. Performance verification and closed-loop feedback were conducted for the fourth mixing ratio.

2. The concrete mix design method for the Tibetan Plateau region according to claim 1, characterized in that, Before loading the concrete mix design model, the concrete mix design model is constructed, including the following steps: Obtain raw data of historical engineering projects in the Tibetan Plateau; the raw data includes extreme environmental parameters of the Tibetan region, local raw material parameters, engineering indicators, mix proportion data and measured performance data; the raw data includes the final performance data of hydraulic concrete mix proportion tests in various Tibetan regions, reflecting the mapping relationship between "environment-materials-mix proportion-performance"; The raw data is preprocessed to determine the dimensions of the mix design and generate a design dataset. Define a network model; the network model is used to construct a mapping relationship that supports "environment-materials-mix proportions-performance" to calculate concrete mix proportions; By designing a dataset, the network model is trained to construct a concrete mix design model.

3. The concrete mix design method for the Tibetan Plateau region according to claim 2, characterized in that, The preprocessing includes: Remove anomalous data specific to the Tibetan region and use the environmental clustering-K-nearest neighbor algorithm to fill in missing values; Standardization processing includes: normalizing continuous parameters; using unique thermal coding for categorical parameters; and quantifying and classifying unique parameters. The continuous parameters include: altitude and extreme temperature; the categorical parameters include: permafrost type and aggregate lithology; and the unique parameters include: low-oxygen, low-temperature hydration activity and freeze-thaw stability. By using gradient boosting tree (XGBoost) and mutual information method, core features are selected from the standardized data, retaining features that are strongly correlated with the performance of the Tibetan plateau.

4. The concrete mix design method for the Tibetan Plateau region according to claim 2, characterized in that, Training the network model includes: The weighted mean square error is used as the loss function, highlighting the key performance requirements of projects in Tibetan areas. For training samples at altitudes above 4000 meters or located in permafrost regions, the sample weights are adjusted to make the network model focus on prediction accuracy under extreme high-altitude and cold conditions. Introduce a dual regularization mechanism into the network structure; The model is divided into three elevation ranges and two permafrost types, forming six sub-regions with clear physical meanings. For each sub-region, the global model is fine-tuned using a design dataset.

Images