A physical constraint multi-source feature fusion risk assessment method and system for mass concrete temperature control

Abstract

The present application belongs to the technical field of mass concrete temperature monitoring and structural health monitoring, aiming at the problems of single function of existing monitoring system, lack of physical constraints in algorithm and difficulty in multi-source data fusion, a mass concrete temperature control physical constraint multi-source feature fusion risk assessment method and system are disclosed. The method inputs multi-source engineering feature parameters, pre-processes, double-path parallel processes and feature weighted fusion, outputs future temperature prediction sequence, and realizes dynamic assessment of grade I-V cracking risk and optimization of cooling scheme. The system includes temperature automatic sensing and transmission, data processing and machine learning risk assessment, remote monitoring module, and realizes intelligent control of the whole process by embedding the above method. The present application forms a closed loop of sensing-analysis-decision-execution, significantly improves the accuracy of temperature control and the efficiency of engineering management and control.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary technical field of temperature monitoring and structural health monitoring of large-volume concrete, specifically a method and system for risk assessment of temperature control of large-volume concrete based on the fusion of multiple source features under physical constraints. Background Technology

[0002] In recent years, my country's transportation infrastructure construction has involved a large number of large-volume concrete pouring operations, such as concrete foundations, abutments, and bridges. Due to concrete's low tensile strength and poor thermal conductivity, the large amount of heat generated by the cement hydration reaction easily accumulates in the core, creating significant temperature differences and inducing temperature stress, ultimately leading to structural cracking and severely affecting the concrete's mechanical properties and service life. Therefore, temperature monitoring and temperature control optimization for large-volume concrete has become a core requirement in engineering construction.

[0003] With the development of sensing and information technologies, automated temperature monitoring systems have been gradually applied in the field of civil engineering. These systems, through pre-embedded temperature sensors, data acquisition instruments, and network transmission modules, achieve continuous acquisition and centralized display of temperature data, thus replacing, to some extent, the traditional intermittent manual monitoring method. However, existing technologies still have many shortcomings that urgently need to be addressed:

[0004] (1) Most existing monitoring systems only have temperature data display, storage and simple threshold alarm functions. They are essentially data loggers and cannot perform in-depth processing of monitoring data. The system can only trace back past temperature changes. It cannot predict the cracking risk in the future hydration heat stage, nor can it generate targeted cooling optimization schemes. It has failed to form an intelligent service closed loop of perception-analysis-decision-execution.

[0005] (2) The analysis modules of the existing system are mostly based on traditional statistical methods or deterministic heat conduction formulas, which are difficult to cope with the nonlinearity and time-varying nature of the heat release rate of concrete hydration, as well as the complex randomness of multi-dimensional boundary conditions such as environment, materials, and construction. This results in a large deviation between temperature prediction and risk assessment results, and cannot provide reliable support for engineering decision-making.

[0006] (3) Some studies have attempted to apply general machine learning models (such as SVR, RF, standard LSTM / GRU) to concrete temperature control, but direct application has significant limitations: First, physical consistency is lacking. The model only relies on the statistical regularity of historical data and lacks understanding of physical principles such as the law of heat conduction and energy conservation. Under new working conditions, it may output prediction results that violate common sense. Second, it is difficult to fuse multi-source heterogeneous data and cannot effectively balance the weights of different types of parameters such as materials, environment, and structure, making it difficult to capture their coupling effect.

[0007] In summary, existing technologies cannot meet the real-time, accurate, predictable, and adaptive requirements of temperature control for large-volume concrete in modern major engineering projects. There is an urgent need for a technical solution that combines the advantages of physical constraints and data-driven approaches, and can achieve risk prediction and solution optimization. Summary of the Invention

[0008] To address the shortcomings of existing monitoring systems, such as limited functionality, insufficient algorithm prediction accuracy, lack of physical consistency in general machine learning algorithms, and difficulties in fusing multi-source heterogeneous data, this invention aims to provide a physical constraint multi-source feature fusion risk assessment method and system for temperature control of large-volume concrete. Through collaborative design involving multi-source engineering feature grouping and embedding with dynamic weight adaptation, parallel processing of data-driven and physical constraint paths, feature weighted fusion, dynamic multi-dimensional risk assessment, and intelligent optimization of cooling schemes, this invention achieves high-precision prediction of the temperature field and heat of hydration of large-volume concrete over the next 1-28 days, real-time dynamic assessment of cracking risks (levels I-V), and adaptive optimization of cooling parameter combinations when risks exceed thresholds. Ultimately, this forms a closed-loop process encompassing perception, analysis, decision-making, and execution, meeting the modern engineering requirements for real-time, accurate, predictable, and adaptive temperature control of large-volume concrete.

[0009] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0010] On the one hand, a risk assessment method for physical constraint multi-source feature fusion in temperature control of large-volume concrete includes the following steps:

[0011] Obtain multi-source engineering characteristic parameters of concrete structures and perform preprocessing;

[0012] Based on adaptive attention, the preprocessed multi-source engineering feature parameters are grouped to obtain normalized features;

[0013] A feature extraction network consisting of a spatiotemporal feature encoder and a physical information constraint subnetwork is constructed. The feature extraction network extracts data features and physical features of the concrete structure from the normalized features, and performs feature fusion to obtain temperature prediction sequences for different parts of the concrete structure.

[0014] The risk level of cracking in concrete structures is classified based on temperature prediction sequences.

[0015] Furthermore, the multi-source engineering characteristic parameters include: time-series monitoring data, concrete materials and mix proportions, structural geometric parameters, environmental time-series data, and cooling measure parameters.

[0016] Furthermore, the feature extraction network specifically includes:

[0017] The spatiotemporal feature encoder learns the temporal law of temperature evolution of concrete structures from normalized features and outputs the data features F of the concrete structures.data ;

[0018] The physical information constraint subnetwork, through its built-in concrete heat conduction control equations containing time-varying internal heat sources and Newtonian cooling boundary conditions, outputs the physical characteristics F of the concrete structure. physics ;

[0019] The fusion layer, for data features F data and physical characteristics F physics Perform fusion and output fused features F fusion ;

[0020] Spatiotemporal feature decoder, based on fused feature F fusion Extract temperature prediction sequences from different parts of the concrete structure.

[0021] Furthermore, the expression for the governing equation of concrete heat conduction for the time-varying internal heat source is:

[0022]

[0023] Where T is the concrete temperature, t is time, and x is the spatial coordinate. (t) is the hydration heat release rate function, and ρ is the concrete density. denoted as , where is the specific heat capacity of concrete, and k is the thermal conductivity.

[0024] Furthermore, the expression for the Newtonian cooling boundary condition is:

[0025]

[0026] Where h is the overall heat transfer coefficient of the concrete surface, T a The ambient temperature.

[0027] Furthermore, the fusion feature F fusion The expression is:

[0028]

[0029] Where α is the weight.

[0030] Furthermore, based on the fusion feature F fusion The temperature prediction sequences for different parts of the concrete structure are extracted, specifically as follows:

[0031] The spatiotemporal feature decoder will fuse feature F fusion The temperature prediction tensor Y is mapped to the next 1-28 days, and the specific mapping expression is as follows:

[0032]

[0033] Where Decoder is the predicted temperature sequence, Y[t, i] represents the predicted temperature of the i-th concrete section at time t, while Y[t, i] represents the predicted temperature of the i-th concrete section on day t.

[0034] Furthermore, the risk value R(t) for cracking risk in concrete structures is calculated as follows:

[0035]

[0036] in, Let t be the predicted temperature difference of the concrete. Let t be the actual measured temperature of the concrete. Let t be the actual rate of change of concrete temperature. The age of the concrete at time t is used for evaluation. Let f1 be the risk mapping based on predicted temperature difference, f2 be the risk mapping based on real-time temperature difference, f3 be the risk mapping based on the measured temperature change rate of the concrete structure, f4 be the material sensitivity mapping based on age, and f5 be the risk mapping based on the hydration heat release rate. They are respectively , , , , .

[0037] Furthermore, the classification method for the cracking risk level of concrete structures is as follows: R(t)∈[0,0.2], is Level I; R(t)∈(0.2,0.4], is Level II; R(t)∈(0.4,0.6], is Level III; R(t)∈(0.6,0.8], is Level IV; R(t)>0.8, is Level V.

[0038] On the other hand, a multi-source feature fusion risk assessment system for physical constraints in large-volume concrete temperature control is implemented based on the aforementioned multi-source feature fusion risk assessment method for physical constraints in large-volume concrete temperature control, including:

[0039] The multi-source parameter acquisition and preprocessing module is used to acquire multi-source engineering characteristic parameters of concrete structures and perform standardized preprocessing.

[0040] The feature processing and temperature prediction module is used to group the preprocessed parameters based on adaptive attention to obtain normalized features. It extracts data features and physical features through a feature extraction network containing a spatiotemporal feature encoder and a physical information constraint subnetwork, and then weights and fuses them to output temperature prediction sequences for different parts of the concrete structure.

[0041] The crack risk level classification module is used to calculate the dynamic risk value R(t) based on the temperature prediction sequence and classify the crack risk level according to the range of R(t) values.

[0042] The beneficial effects of this invention are:

[0043] 1. This invention uses a Physical Information Constrained Subnetwork (PINN) with an embedded heat conduction equation to ensure that the prediction results conform to physical laws such as Fourier's law and energy conservation, thus solving the problem of lack of physical consistency in general machine learning algorithms and avoiding prediction biases that violate common sense under new working conditions. At the same time, through multi-source feature grouping embedding and dynamic weight adaptation mechanisms, it captures the coupling effect of heterogeneous parameters such as materials, environment, and structure, effectively solving the problem of multi-source data fusion, and significantly improving the accuracy of temperature prediction and hydration heat inversion.

[0044] 2. This invention constructs a multi-dimensional dynamic risk assessment model, integrating factors such as predicted temperature difference, real-time temperature difference, temperature change rate, age sensitivity, and hydration heat release rate. Through risk mapping functions and weighted calculations, it achieves risk level classification and real-time output from I to V. Combined with clear response times and triggering conditions, it upgrades risk prevention and control from "passive alarm" to proactive prediction and graded intervention, reducing the probability of concrete cracking.

[0045] 3. This invention forms a closed-loop chain of data perception, feature processing, temperature prediction, risk assessment, and scheme optimization. The system can automatically output the temperature prediction sequence for the next 1 to 28 days, the heat of hydration release rate function, and the optimized cooling parameter combination. It can complete the entire process from analysis to decision-making without manual intervention, significantly reducing the workload of on-site management personnel and improving the execution efficiency and scientific nature of temperature control schemes. Attached Figure Description

[0046] Figure 1 This is a flowchart of the risk assessment method of the present invention. Detailed Implementation

[0047] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0048] Example 1:

[0049] See attached document Figure 1 The method for risk assessment of physical constraints and multi-source feature fusion for temperature control of large-volume concrete, as shown, specifically includes the following steps:

[0050] S1. Obtain multi-source engineering characteristic parameters of the concrete structure and perform preprocessing;

[0051] Specifically, the detailed engineering characteristic parameters of multi-source temperature control for large-volume concrete are shown in Table 1 below:

[0052] Table 1. Specific Composition of Multi-Source Engineering Characteristic Parameters for Temperature Control of Mass Concrete

[0053]

[0054] The collected engineering characteristic parameters are standardized in format. Discrete parameters (such as cement type and aggregate type) are converted into a unified coding format, while continuous parameters (such as temperature and size) retain their original numerical units. For missing scattered data, the average value of similar working conditions in similar projects is used to fill in the gaps, ensuring that no key information is missing from the multi-source parameters, thus providing complete and standardized raw data for subsequent grouping and processing.

[0055] S2. Based on adaptive attention, the preprocessed multi-source engineering feature parameters are grouped to obtain normalized features;

[0056] Specifically, this step includes the following sub-steps:

[0057] S201, group the multi-source engineering characteristic parameters according to their attributes, specifically dividing them into time-series monitoring group G. t (T, ΔT), where T represents the historical temperature sequence data included in this group, i.e., the temperature time series of the concrete core and surface collected from the initial monitoring to the current moment, and ΔT represents the historical temperature difference sequence data between the inner and outer surfaces included in this group. The material group Gm, the environmental group Ge, the cooling group Gc, and the structural group Gs are each mapped to a feature vector of a unified dimension through an independent embedding layer (fully connected network). The mapping expression is:

[0058]

[0059] in, Let e ​​be the initial eigenvector of the g-th group. g is the mapped feature vector, Embedding is the fully connected neural network layer, and g is the index of the specific feature group whose weights are currently being calculated.

[0060] S202 uses the adaptive feature attention module SA to dynamically calculate the importance weights of each group of features at different construction stages. It adjusts the weight allocation based on the differences in influencing factors at different stages of hydration heat, and then obtains the normalized feature e through weighted integration. fused .

[0061] Specifically, the process of dynamically calculating the importance weights of each group of characteristics at different construction stages and adjusting the weight allocation based on the differences in influencing factors at different stages of hydration heat is as follows:

[0062] First, the working condition adaptive feature attention module SA maintains a learnable context row vector V. T This describes the aspects that need to be emphasized in the current construction phase, and is achieved by calculating the row vector V. T The dot product of the feature column vectors yields the "focus score" u.g :

[0063]

[0064] Among them: W a Let b be the weight matrix of the fully connected layer. a For bias vectors, It is the hyperbolic tangent activation function.

[0065] Continue to calculate the "focus score" u g Substituting the dynamic weight allocation formula, we obtain the dynamic weight allocation for each feature group.

[0066]

[0067] Where, α g The "focus weight" of group g is given, and satisfies the following conditions: 0≤α g ≤1, k is the summation index, that is, it includes the above feature group categories {t,m,e,c,s}, u k Let $\mathbf{k}$ be the "focus score" of any feature group containing index $k$ in the summation notation.

[0068] Finally, focus on weight α g With eigenvector e g Weighted fusion is performed to obtain the normalized feature e. fused :

[0069]

[0070] S3. Construct a feature extraction network that includes a spatiotemporal feature encoder and a physical information constraint sub-network. Extract data features and physical features of the concrete structure from the normalized features through the feature extraction network, and perform feature fusion to obtain temperature prediction sequences for different parts of the concrete structure.

[0071] Specifically, the feature extraction network includes:

[0072] Spatiotemporal feature encoder, from normalized feature e fused The study investigates the temporal evolution of temperature in concrete structures and outputs the data characteristics F of the concrete structures. data That is, F data It is e fused The result after mapping by the spatiotemporal feature encoder function Encoder has the following function mapping relationship:

[0073]

[0074] The spatiotemporal feature encoder uses the aforementioned GRU unit as its core data feature F. data, is defined as the final hidden state vector h of the GRU network after processing the entire input sequence. T Therefore, the three have the following equivalence relationship.

[0075]

[0076] in, h is the total number of time steps in the input sequence. T It is calculated from the final state formula of GRU at t=T.

[0077] The physical information constraint subnetwork, through its built-in concrete heat conduction control equations containing time-varying internal heat sources and Newtonian cooling boundary conditions, outputs the physical characteristics F of the concrete structure. physics ;

[0078] The fusion layer, for data features F data and physical characteristics F physics Perform fusion and output fused features F fusion ;

[0079] Spatiotemporal feature decoder, based on fused feature F fusion Extract temperature prediction sequences from different parts of the concrete structure.

[0080] Furthermore, specific methods for constructing feature extraction networks include:

[0081] A feature extraction network was constructed, comprising a spatiotemporal feature encoder, a physical information constraint sub-network, a fusion layer, and a spatiotemporal feature decoder. The feature extraction network was trained by first pre-training on multiple historical engineering project datasets to learn the common patterns of temperature control in large-volume concrete; then, using small sample monitoring data from 3 to 7 days after the pouring of new projects, rapid adaptive fine-tuning was performed to complete customized adaptation for the projects, and finally, network parameters were optimized to ensure that temperature prediction has both data accuracy and physical compliance.

[0082] (1) Spatiotemporal feature encoder

[0083] This invention employs a modified gated recurrent unit (GRU) as a spatiotemporal feature encoder for path processing based on data temporal patterns. The concrete temperature change rate ΔT / Δt from the previous moment is used as an additional input, forming a joint input of historical features and temperature change trends. This GRU processes the fused features of the input sequentially, step-by-step, with its reset gate γ... t Update Gate Z t and candidate status The specific expression is as follows:

[0084]

[0085] in, For normalized feature e fusion The input feature vector at time t, where δ is the difference operator. This represents the actual measured value of the concrete core temperature at time t-1. Let be the hidden state vector at time t. Let be the hidden state vector at time t−1. b z b h These are the weight matrices for the reset gate, update gate, and candidate state, respectively. W is the rate of change of concrete temperature at time t-1 (°C / h). dr W dz They are respectively the rate of temperature change To reset the gate and update the gate's weight matrix, , , These are the weight matrices input to the reset gate, update gate, and candidate state, respectively. The purpose of this is the Sigmoid function, which is to compress the output range to (0, 1).

[0086] (2) Physical information constrained subnetwork

[0087] This invention employs a Physical Information Constrained Subnetwork (PINN) based on the physical laws of heat conduction in large-volume concrete. This subnetwork incorporates the governing equations for heat conduction in concrete containing time-varying internal heat sources and Newtonian cooling boundary conditions. Standardized fusion features are screened through physical rationality verification, and physical features F conforming to the laws of heat conduction are extracted. physics And output;

[0088] Specifically, the expression for the governing equation of heat conduction in concrete containing time-varying internal heat sources is:

[0089]

[0090] Where T is the concrete temperature, t is time, and x is the spatial coordinate. (t) represents the heat of hydration release rate, and ρ represents the density of concrete. Here, k represents the specific heat capacity of concrete, and k is the thermal conductivity.

[0091] The expression for the Newtonian cooling boundary condition is:

[0092]

[0093] Where h is the overall heat transfer coefficient of the concrete surface, T a The ambient temperature;

[0094] Construct a fully connected physical information neural network N physics(x,t,θ p To approximate the solution T(x,t), its loss function L PINN Including the physical equation residual term L PDE Boundary condition residual term L BC and data fitting term L Data The specific expression is as follows:

[0095]

[0096] in, The boundary condition loss weighting coefficients, The loss weight coefficients are used to fit the data.

[0097] Among them, the residual term L of the physical equation PDE This is to ensure that the network satisfies the control equation, which is expressed as:

[0098]

[0099] The boundary condition residual term L BC To ensure that the network output satisfies boundary conditions (such as Newton's law of cooling), the expression is:

[0100]

[0101] L Date As a data fitting term, it aims to ensure that the network output closely approximates the actual measured temperature of concrete. Its expression is:

[0102]

[0103] Where, N f N represents the total number of configuration points randomly sampled and integrated within the solution domain (the spatial and temporal range within the concrete structure). b N represents the total number of boundary points generated by sampling on the solution domain boundaries (spatial boundary and initial time boundary t=0). d T represents the total number of measured temperature data points. meas (x k ,t k ) is at position x k Locations (such as the core and surface of the concrete) and t k Actual temperature measurement at time N physics (x k ,t k ) represents the temperature prediction output of the neural network in the corresponding coordinates. i ,t i} represents the collocation points within the domain (generated via Latin hypercube sampling), {x j , t j} represents the boundary / initial condition point, {x k , t k} represents the measured data points of concrete temperature, and BC represents the boundary condition operators.

[0104] Physical law characteristics F physics Defined as the activation value vector of the last hidden layer of the PINN subnetwork, its function expression is:

[0105]

[0106] Where x is the spatial coordinate and t is the time variable. This is the set of all trainable parameters (including biases and weights) for the PINN subnetwork.

[0107] H physics (x,t,θ) is the output function of the last hidden layer in the Physical Information Constraint Subnetwork PINN, and its expression is as follows:

[0108]

[0109] Among them, W out and b out For the weights and biases of the output layer, The parameters of the feature part, their magnitude ;

[0110] physical laws characteristics As the activation vector of this hidden layer, it is processed through the physical equation residual term L during the learning process. PDE Boundary condition residual term L BC and data fitting term L Data The common constraints encode the intrinsic laws of the entire process of concrete hydration heat, thus serving as an effective feature of physical consistency, and are combined with data-driven features. Integration.

[0111] (3) Fusion layer

[0112] A weighted summation method is used to achieve deep integration of data features and physical features. The fusion expression is as follows:

[0113]

[0114] Among them, F data For data features, F physics α represents the physical characteristics, and α is a learnable weight parameter.

[0115] Specifically, this step combines the advantages of temporal regularity of data features with the advantages of rationality of physical features through weighted fusion, and outputs the fused feature F. fusionIt combines data accuracy and physical compliance, allowing direct input into the next step for temperature prediction sequence mapping, achieving a tight connection between feature fusion and prediction mapping. physics As a correction signal, it provides physical constraints and supplements to the purely data-driven features to ensure the fused F fusion It encompasses complex patterns in the data while also conforming to basic physical characteristics and laws.

[0116] (4) Spatiotemporal feature decoder

[0117] In this invention, the spatiotemporal feature decoder is used to fuse the feature F. fusion The mapping process is performed to generate a temperature prediction sequence for the next 1-28 days. The mapping expression is as follows:

[0118]

[0119] Wherein, Decoder is the neural network mapping function of the spatiotemporal feature decoder, which learns to fuse features F fusion The mapping, Y[t, i], represents the predicted temperature value of the i-th concrete part on day t (when i=0 represents the core concrete part, and i=1 represents the concrete surface layer).

[0120] This temperature prediction sequence covers daily temperature data for the core and surface of large-volume concrete, fully reflecting the temperature evolution trend of different parts of the concrete, and providing core data support for subsequent hydration heat inversion and cracking risk assessment.

[0121] S4. Classify the cracking risk level of concrete structures based on the temperature prediction sequence.

[0122] Specifically, this step includes the following sub-steps:

[0123] S401, based on temperature prediction sequence and hydration heat release rate function The dynamic risk value R(t) is calculated based on real-time monitoring data, and the weighted risk scoring model is called. The expression is:

[0124]

[0125] in, Let t be the predicted temperature difference of the concrete. Let t be the actual measured temperature of the concrete. Let t be the actual rate of change of concrete temperature. The age of the concrete at time t is used for evaluation. This is a function of the heat release rate of concrete hydration.

[0126] Specifically The predicted temperature difference for day t is calculated from the temperature prediction sequence Y, and the expression is as follows:

[0127]

[0128] In this context, i=0 represents the concrete core, while i=1 represents the concrete surface layer.

[0129] f1~f5 are risk mapping functions. The weights are specified in the table below:

[0130] Table 2 Weight Assignment Table

[0131]

[0132] Among them, f1 (predicted temperature difference risk): the risk prediction of the future hydration heat stage of the measured concrete is a proactive measure to prevent its cracking and damage, so its weight value should be second only to f2 (real-time temperature difference risk).

[0133] f2 (Real-time temperature difference risk): Since the measured temperature difference is direct data, it has certainty and directness, and therefore can be given the highest weight.

[0134] f3 (Risk of Temperature Change Rate): If the temperature stress generated by a large temperature change rate is higher than the tensile strength of concrete, it will cause cracking of the concrete structure. In addition, large shrinkage and expansion deformations will also easily lead to cracking of the concrete structure.

[0135] f4 (Hydration heat age sensitivity): Concrete has extremely low tensile strength during the early hardening process, and is therefore highly sensitive to temperature stress.

[0136] f5 (Risk of heat release rate of hydration): During the hardening process of concrete, the heat release of cement hydration is the fundamental reason for the temperature rise of concrete, but its impact is already partially reflected in the temperature difference, so it is assigned the lowest value.

[0137] Specifically, the risk mapping functions are as follows:

[0138] f1: Risk mapping based on predicted temperature difference, using the Sigmoid function, with the expression:

[0139]

[0140] Where k1 is the kurtosis factor (0.3~0.5), ΔT pred To predict temperature difference, T th_pred To predict the temperature difference risk threshold (25℃); when the maximum temperature difference is close to or greater than the threshold, the output approaches 1 (higher risk); otherwise it approaches 0 (lower risk).

[0141] f2: Risk mapping based on real-time temperature difference, using the Sigmoid function, with the expression:

[0142]

[0143] Where k2 is the kurtosis factor (0.8~1.2), ΔT real To measure the temperature difference in real time, T th-real The threshold for real-time temperature difference warning (below 25℃) emphasizes the urgency of immediate intervention.

[0144] f3: Risk mapping based on the rate of temperature change, using the absolute value of the hyperbolic tangent function, expressed as:

[0145]

[0146] Where, r th The reference threshold for the rate of temperature change is 0~15℃ / day. The instantaneous rate of change of the current core temperature (°C / h) is used to symmetrically assess the risk of heating and cooling rates.

[0147] f4: Age-based material sensitivity mapping, using a Gaussian function, expressed as:

[0148]

[0149] Among them, t peak The peak risk age is 3 days, and σ is the standard deviation, assigning a higher risk weight to the 1-3 days after pouring.

[0150] f5: Risk mapping based on hydration heat release rate, using a linear normalization function, expressed as:

[0151]

[0152] Among them, Q max For the maximum reference heat of hydration release rate, the output value changes linearly within the range of 0 to 1, and when ≥ When the value is 1, the value is taken as 1.

[0153] S402. Based on the calculated dynamic risk value R(t), it is mapped to the cracking risk levels of I to V. The specific classification criteria are as follows:

[0154] Table 3 Crack Risk Levels I to V

[0155]

[0156] Note: ΔT represents the temperature difference between the inner and outer surfaces, and the temperature difference between the surface of the concrete structure and the environment.

[0157] Finally, when the risk level reaches a preset threshold (Level III and above, i.e., R(t) ≥ 0.4), the optimized cooling parameter combination is output with the goal of minimizing the predicted temperature difference and cooling energy consumption. This optimizer uses the current cooling parameters as the initial point and aims to minimize the predicted maximum temperature difference max(T) over the next 24 hours. pred The optimization of the cooling system energy consumption is a multi-objective problem. A heuristic approach is used to search for the optimal control parameters, including adjusting the cooling water flow rate, changing the inlet and outlet water direction (which can be adjusted to be changed every 2 hours), adjusting the inlet water temperature, and adjusting the thickness of the concrete insulation layer.

[0158] Example 2:

[0159] This embodiment provides a feasible implementation scheme for a temperature control risk assessment system for large-volume concrete, which consists of a multi-source parameter acquisition and preprocessing module, a feature processing and temperature prediction module, and a cracking risk level classification module.

[0160] The multi-source parameter acquisition and preprocessing module collects time-series temperature data through temperature sensors and inputs concrete material and mix proportions, structural geometric parameters, environmental time-series data, and cooling measure parameters via a touch terminal. It automatically converts discrete parameters into a unified code, retains the original units for continuous parameters, fills in missing data with the average values ​​of similar working conditions from comparable projects, and outputs standardized preprocessed parameters.

[0161] The feature processing and temperature prediction module groups preprocessing parameters based on an adaptive attention mechanism and outputs normalized features. A feature extraction network containing a spatiotemporal feature encoder and a physical information constraint subnetwork extracts data features F. data and physical characteristics F physics The weighted fusion yields the fusion feature F. fusion ;

[0162] The crack risk level classification module calculates the dynamic risk value R(t) based on the temperature prediction sequence, hydration heat release rate curve, and real-time data. R(t) is classified into crack risk levels I to V. When the risk level is ≥ III, the module automatically outputs optimized cooling parameter combinations (such as cooling water flow rate and insulation layer thickness adjustment suggestions).

[0163] The system in this invention has two service items: a "standard engineering application function panel" and a "machine learning risk assessment panel".

[0164] The "Engineering Applications General Function Panel" contains 7 functions:

[0165] F1: Real-time temperature display: Clicking this button will automatically draw and display the real-time temperature change curves of the core, surface, and ambient temperature of the concrete structure from midnight to the current time.

[0166] F2: Temperature difference data calculation: After clicking, the system automatically analyzes and calculates the temperature difference between the inner and outer surfaces of the concrete and the temperature difference between the surface and the environment based on the real-time collected temperature data, and generates a dynamic temperature difference curve simultaneously.

[0167] F3: Temperature warning settings: After clicking, you can set the peak temperature (core temperature), inner and outer surface temperature difference, and concrete surface temperature difference with the environment for large-volume concrete. When the monitored temperature reaches or exceeds the warning value, the host and the automated temperature data acquisition box will sound simultaneously.

[0168] F4: Temperature Data Transmission: After clicking, enter the system number you want to transmit the data to to the target remote control host, thus achieving temperature data sharing;

[0169] F5: Historical Temperature Data: Click to view all temperature data from the initial monitoring time to the present, supporting filtering and trend backtracking by time, measurement point, and other conditions;

[0170] F6: Temperature monitoring interval: Clicking this allows you to set the temperature monitoring time interval (e.g., 10 min, 1 h, 1 d, etc.) for the temperature measurement system in the "Automatic Temperature Sensing and Transmission Module".

[0171] F7: Change Account Password: Click to change your account and password to ensure data security;

[0172] In the "Machine Learning Risk Assessment Panel," users can input parameters for the concrete structure to be monitored and the layout of on-site cooling measures in the actual project. These parameters include the dimensions of the concrete structure, the type of concrete raw materials, the mix proportion and strength grade, the initial temperature of the concrete during pouring, and the thickness of the concrete insulation layer. If cooling pipes are used, the diameter of the cooling pipes, the number of layers, the spacing between layers, the spacing between pipes in the same layer, the cooling water flow rate, and its inlet temperature can also be set. After the parameters are input, the system automatically calculates the heat of hydration of cement, the specific heat capacity and thermal conductivity of concrete, and the thermal conductivity of the insulation material using built-in methods. It automatically generates the predicted curves of the heat of hydration of the measured concrete at various ages (1~28 days). Users can view these curves by clicking the "Predicted Curves of Heat of Hydration of Concrete at Various Ages" function. After loading the PIMSF-Net basic model based on meta-learning pre-training, the system automatically performs rapid adaptive fine-tuning (approximately 10 iterations, taking <1 minute) in the background using the temperature data (approximately 144 data points) automatically collected in the first 24 hours after the project's pouring. Once the fine-tuning is complete, the model is customized for this specific project.

[0173] Remote temperature data monitoring module: This includes a remote monitoring host and a router. After connecting to the network, click the "Temperature Data Transmission" function, enter the system number (sender) of the "Automatic Temperature Monitoring System for Large-Volume Concrete" in the on-site monitoring host, and you can obtain the latest temperature data on site. The data will be continuously updated as the network connection is established, eventually making the on-site detection data completely consistent with the remote monitoring data, thus realizing the sharing of temperature data.

[0174] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for risk assessment of physical constraints and multi-source feature fusion for temperature control of large-volume concrete, characterized in that, Includes the following steps: Obtain multi-source engineering characteristic parameters of concrete structures and perform preprocessing; Multi-source engineering characteristic parameters include: time-series monitoring data, concrete materials and mix proportions, structural geometric parameters, environmental time-series data, and cooling measure parameters; Based on adaptive attention, the preprocessed multi-source engineering feature parameters are grouped to obtain normalized features; A feature extraction network consisting of a spatiotemporal feature encoder and a physical information constraint subnetwork is constructed. The feature extraction network extracts data features and physical features of the concrete structure from the normalized features, and performs feature fusion to obtain temperature prediction sequences for different parts of the concrete structure. The dynamic risk value R(t) is calculated based on the temperature prediction sequence and real-time monitoring data, and the cracking risk of concrete structures is divided into levels I to V based on the numerical range of the dynamic risk value R(t). The calculation method for the risk value R(t) of concrete structure cracking risk is as follows: ； in, Let t be the predicted temperature difference of the concrete. Let t be the actual measured temperature of the concrete. Let t be the actual rate of change of concrete temperature. The age of the concrete at time t is used for evaluation. Let f1 be the risk mapping based on the predicted temperature difference, f2 be the risk mapping based on the real-time temperature difference, f3 be the risk mapping based on the measured temperature change rate of the concrete structure, f4 be the material sensitivity mapping based on the age, and f5 be the risk mapping based on the hydration heat release rate; ω1~ω5 are ω1=0.25, ω2=0.35, ω3=0.15, ω4=0.15, and ω5=0.10, respectively.

2. The method for risk assessment of physical constraints and multi-source feature fusion for temperature control of large-volume concrete according to claim 1, characterized in that, Feature extraction networks specifically include: The spatiotemporal feature encoder learns the temporal law of temperature evolution of concrete structures from normalized features and outputs the data features F of the concrete structures. data ; The physical information constraint subnetwork, through its built-in concrete heat conduction control equations containing time-varying internal heat sources and Newtonian cooling boundary conditions, outputs the physical characteristics F of the concrete structure. physics ; The fusion layer, for data features F data and physical characteristics F physics Perform fusion and output fused features F fusion ; Spatiotemporal feature decoder, based on fused feature F fusion Extract temperature prediction sequences from different parts of the concrete structure.

3. The method for risk assessment of physical constraints and multi-source feature fusion for temperature control of large-volume concrete according to claim 2, characterized in that, The expression for the governing equation of concrete heat conduction for a time-varying internal heat source is: ； Where T is the concrete temperature, t is time, and x is the spatial coordinate. (t) is the hydration heat release rate function, ρ is the concrete density, and c p denoted as , where is the specific heat capacity of concrete, and k is the thermal conductivity.

4. The method for risk assessment of physical constraints and multi-source feature fusion for temperature control of large-volume concrete according to claim 2, characterized in that, The expression for the Newtonian cooling boundary condition is: ； Where h is the overall heat transfer coefficient of the concrete surface, T a Let t be the ambient temperature, k be the thermal conductivity, and T(0,t) be the temperature of the concrete surface at time t.

5. The method for risk assessment of physical constraints and multi-source feature fusion for temperature control of large-volume concrete according to claim 2, characterized in that, Fusion feature F fusion The expression is: ； Where α is the weight.

6. The method for risk assessment of physical constraints and multi-source feature fusion for temperature control of large-volume concrete according to claim 2, characterized in that, According to fusion feature F fusion The temperature prediction sequences for different parts of the concrete structure are extracted, specifically as follows: The spatiotemporal feature decoder will fuse feature F fusion The temperature prediction tensor Y is mapped to the next 1-28 days, and the specific mapping expression is as follows: ； Where Decoder is the predicted temperature sequence, Y[t, i] represents the predicted temperature of the i-th concrete section at time t, while Y[t, i] represents the predicted temperature of the i-th concrete section on day t.

7. The method for risk assessment of physical constraints and multi-source feature fusion for temperature control of large-volume concrete according to claim 1, characterized in that, The classification method for crack risk levels of concrete structures is as follows: R(t)∈[0,0.2], is Level I; R(t)∈(0.2,0.4], is Level II; R(t)∈(0.4,0.6], is Level III; R(t)∈(0.6,0.8], is Level IV; R(t)>0.8, is Level V.

8. A risk assessment system for physical constraint multi-source feature fusion for temperature control of large-volume concrete, implemented based on the risk assessment method for physical constraint multi-source feature fusion for temperature control of large-volume concrete as described in any one of claims 1 to 7, characterized in that, include: The multi-source parameter acquisition and preprocessing module is used to acquire multi-source engineering characteristic parameters of concrete structures and perform standardized preprocessing. Multi-source engineering characteristic parameters include: time-series monitoring data, concrete materials and mix proportions, structural geometric parameters, environmental time-series data, and cooling measure parameters; The feature processing and temperature prediction module is used to group the preprocessed parameters based on adaptive attention to obtain normalized features. It extracts data features and physical features through a feature extraction network containing a spatiotemporal feature encoder and a physical information constraint subnetwork, and then weights and fuses them to output temperature prediction sequences for different parts of the concrete structure. The cracking risk level classification module is used to calculate the dynamic risk value R(t) based on the temperature prediction sequence and real-time monitoring data, and classify the cracking risk of concrete structures into levels I to V based on the numerical range of the dynamic risk value R(t). The calculation method for the risk value R(t) of concrete structure cracking risk is as follows: ； in, Let t be the predicted temperature difference of the concrete. Let t be the actual measured temperature of the concrete. Let t be the actual rate of change of concrete temperature. The age of the concrete at time t is used for evaluation. Let f1 be the risk mapping based on the predicted temperature difference, f2 be the risk mapping based on the real-time temperature difference, f3 be the risk mapping based on the measured temperature change rate of the concrete structure, f4 be the material sensitivity mapping based on the age, and f5 be the risk mapping based on the hydration heat release rate; ω1~ω5 are ω1=0.25, ω2=0.35, ω3=0.15, ω4=0.15, and ω5=0.10, respectively.

Images