Intelligent temperature control method and system for mass concrete based on three-dimensional temperature field

Abstract

The application discloses a mass concrete intelligent temperature control method and system based on a three-dimensional temperature field, which comprises the following steps: collecting multi-modal image data of the surface of a target mass concrete structure and constructing a continuous three-dimensional temperature field; identifying a temperature abnormal area based on the three-dimensional temperature field, and combining historical monitoring data, construction period environmental parameters and concrete mix proportion parameters to predict a temperature change trend in a future preset period; generating a cooling control instruction set according to the predicted temperature change trend and the current temperature state; executing cooling control and performing closed-loop adjustment based on real-time feedback data until the internal temperature reaches a target range. By constructing a high-precision, dynamically updated three-dimensional temperature field model, combining temperature prediction driven by deep learning and self-adaptive decision mechanism driven by reinforcement learning, the application realizes the transformation from passive response to active intervention, and significantly improves the accuracy, foresight and intelligent level of mass concrete structure temperature control.

Description

Technical Field

[0001] This application belongs to the field of intelligent construction and automatic control technology in civil engineering, specifically involving an intelligent temperature control method and system for large-volume concrete based on a three-dimensional temperature field. It is applicable to temperature monitoring, prediction and active control during the construction of various large-volume concrete structures such as bridge abutments, dams, nuclear power plant foundations, and super high-rise building slabs. Background Technology

[0002] During the hardening process of mass concrete, the hydration of cement generates a large amount of heat, causing a significant increase in the internal temperature of the concrete. When the difference between the highest internal temperature of the concrete and the ambient temperature or the surface temperature of the concrete exceeds a certain limit, thermal stress will be generated, leading to temperature cracks and seriously affecting the integrity, impermeability, durability, and safety of the concrete structure. According to the "Standard for Construction of Mass Concrete" (GB 50496-2018) and the "Technical Specification for Temperature Measurement and Control of Mass Concrete" (GB / T 51028-2015), continuous monitoring of the internal temperature of the concrete and effective temperature control measures are required during the construction of mass concrete.

[0003] However, existing technologies for temperature monitoring and control of large-volume concrete mainly have the following problems:

[0004] 1. Traditional monitoring systems adopt a passive response mode, which only starts cooling when the temperature sensor detects an anomaly. It takes an average of 2-3 hours from the detection of the anomaly to the cooling taking effect. By this time, the temperature has already risen by 5-8°C, missing the best time for regulation and failing to effectively curb the temperature peak.

[0005] 2. Traditional methods of pre-embedded temperature sensors require the installation of a large number of sensors before concrete pouring. However, the density of their arrangement is limited, making it difficult to obtain complete temperature field distribution information. This results in a lack of global perspective in temperature control decisions, with some areas potentially becoming too cold or too hot. Furthermore, once a sensor is damaged, it cannot be replaced, leading to discontinuous monitoring data.

[0006] 3. Existing infrared thermal imaging monitoring technologies mainly use fixed or handheld thermal imagers, which have limited monitoring range and cannot comprehensively and efficiently monitor large-volume concrete structures such as large bridge abutments. Furthermore, the monitoring results are greatly affected by the viewing angle and distance, making it difficult to achieve accurate three-dimensional temperature field reconstruction.

[0007] 4. Existing temperature control systems mostly use fixed PID parameters or simple threshold judgments, which cannot adaptively adjust according to the actual state of concrete and environmental changes, resulting in low control accuracy.

[0008] 5. Existing technologies cannot achieve real-time three-dimensional reconstruction of structures and precise superposition of temperature fields, making it difficult to intuitively display temperature distribution characteristics and change patterns, which is not conducive to timely detection of temperature anomalies and the implementation of proactive temperature control measures.

[0009] 6. Traditional monitoring methods have low data acquisition frequency and long monitoring cycles, which cannot meet the needs of rapid response to temperature changes during the construction of large-volume concrete.

[0010] Therefore, developing a method and system that can accurately monitor the hydration heat temperature of large-volume concrete structures such as bridge abutments, perform real-time three-dimensional reconstruction, intelligent prediction, and active control is of great significance for improving the construction quality of large-volume concrete and preventing temperature cracks. Summary of the Invention

[0011] To address the aforementioned technical challenges, this application provides a method and system for intelligent temperature control of large-volume concrete based on a three-dimensional temperature field. This approach constructs a high-precision, dynamically updated three-dimensional temperature field model, combining deep learning-driven temperature prediction with a reinforcement learning-driven adaptive decision-making mechanism. This achieves a paradigm shift from passive response to proactive intervention, significantly improving the accuracy, foresight, and intelligence of temperature control for large-volume concrete structures.

[0012] The first objective of this application is to provide a smart temperature control method for large-volume concrete based on a three-dimensional temperature field.

[0013] The aforementioned objective of this application is achieved through the following technical solution:

[0014] A method for intelligent temperature control of large-volume concrete based on a three-dimensional temperature field, the method comprising the following steps:

[0015] Collect multimodal image data of the surface of the target large-volume concrete structure and construct a continuous three-dimensional temperature field;

[0016] Based on the three-dimensional temperature field, temperature anomaly regions in the target large-volume concrete structure are identified, and the temperature change trend in the future preset period is predicted by combining the historical monitoring data of the target large-volume concrete structure, construction period environmental parameters and concrete mix proportion parameters.

[0017] Based on the predicted temperature change trend and current temperature state of the target large-volume concrete structure, a cooling control instruction set is generated to regulate the zoned cooling system within the target large-volume concrete structure.

[0018] Cooling control is executed based on the cooling control instruction set, and closed-loop adjustment is performed based on the real-time feedback data of the target large-volume concrete structure until the internal temperature of the target large-volume concrete structure reaches the target range.

[0019] Preferably, the step of acquiring multimodal image data of the surface of the target large-volume concrete structure and constructing a continuous three-dimensional temperature field includes:

[0020] A mobile platform equipped with an infrared thermal imager and a visible light camera is used to acquire multi-view images of the target large-volume concrete structure according to preset movement parameters, thereby obtaining visible light images and infrared thermal images.

[0021] The visible light image is subjected to geometric and radiometric correction, and feature points are extracted for three-dimensional reconstruction. The infrared thermal image is subjected to temperature correction based on a radiometric thermometry model.

[0022] Based on motion recovery structure and multi-view stereo vision algorithm, a high-precision three-dimensional point cloud model of the target large-volume concrete structure is constructed using the corrected visible light image.

[0023] The temperature data of the corrected infrared thermal image is projected onto the three-dimensional point cloud model through coordinate mapping to establish the correspondence between temperature data and geometric position;

[0024] A spatial interpolation algorithm is used to reconstruct the temperature field of the unmeasured region based on the mapped temperature data, generating a continuous three-dimensional temperature field.

[0025] The three-dimensional temperature field is displayed in the form of a color cloud map, and temperature isosurfaces, temperature slices and temperature gradient distribution maps are generated. The maximum temperature, minimum temperature, average temperature, surface-to-internal temperature difference and cooling rate and temperature statistical parameters are calculated and output.

[0026] Preferably, when constructing the three-dimensional temperature field, control points are set up on the surface of the concrete structure, and a temperature data fusion algorithm based on spatiotemporal consistency constraints is adopted. By introducing the temporal continuity constraint of the temperature field at adjacent moments, the temperature jump caused by the movement of the mobile platform is eliminated.

[0027] Preferably, identifying temperature anomaly regions in the target large-volume concrete structure based on the three-dimensional temperature field includes:

[0028] Multi-scale analysis was performed on the three-dimensional temperature field to extract statistical characteristic parameters, including temperature peak, temperature gradient, surface-to-interior temperature difference, and cooling rate.

[0029] The multi-channel feature map, which integrates temperature value, temperature gradient, temporal evolution features, and spatial location information, is input into the deep learning segmentation model.

[0030] The deep learning segmentation model automatically identifies temperature anomaly regions and generates a spatial distribution map of these regions.

[0031] Preferably, the step of predicting the temperature change trend within a preset time period by combining historical monitoring data of the target large-volume concrete structure, construction period environmental parameters, and concrete mix proportion parameters includes:

[0032] Acquire historical three-dimensional temperature field data, construction period environmental parameters, and concrete mix proportion parameters from the historical monitoring data;

[0033] The historical three-dimensional temperature field data, construction period environmental parameters, and concrete mix proportion parameters are input into a dual-network temperature prediction model composed of a long short-term memory network that integrates physical constraints of heat conduction and a deep Q network. The dual-network temperature prediction model achieves collaborative optimization of temperature prediction and temperature control strategies by sharing the underlying feature representation and through joint training.

[0034] The long short-term memory network that integrates the physical constraints of heat conduction predicts the temperature evolution curve and the time of temperature peak occurrence within a preset period in the future. The temperature evolution curve is then used as the state input to the deep Q network to evaluate the long-term value of different cooling control actions and output the corresponding value function.

[0035] Preferably, the step of generating a cooling control instruction set for regulating the zoned cooling system within the target large-volume concrete structure based on the predicted temperature change trend and current temperature state includes:

[0036] Construct a multi-objective reward function with temperature control deviation, energy consumption cost, and temperature distribution uniformity as optimization objectives;

[0037] The predicted temperature change trend and the current temperature state are input as environmental states into a near-end policy optimization algorithm based on the Actor-Critic architecture. The near-end policy optimization algorithm aims to maximize the long-term cumulative return of the multi-objective reward function and decides the cooling water flow rate and target temperature of each zone's cooling pipes.

[0038] The cooling water flow rate and target temperature output by the near-end strategy optimization algorithm are combined with the corresponding cooling pipe number and preset execution time to generate a control instruction set containing the cooling pipe number, flow rate set value and execution time.

[0039] Preferably, when performing cooling control based on the cooling control instruction set, an adaptive parameter-tuned PID control algorithm is used to adjust the cooling water flow rate. The PID parameters of the PID control algorithm are dynamically adjusted according to the temperature deviation and the rate of change of the temperature deviation.

[0040] When the temperature deviation exceeds a preset large deviation threshold, the proportional coefficient in the PID parameter is increased to speed up the response.

[0041] When the temperature deviation is less than the preset small deviation threshold, the proportional coefficient in the PID parameters is reduced and the integral coefficient is increased to eliminate the steady-state error.

[0042] Preferably, before performing cooling control based on the cooling control instruction set, the method further includes:

[0043] The cooling control command set is input into the digital twin model for simulation and pre-run, the effects of different control strategies are evaluated, and the optimal strategy is selected and executed.

[0044] Preferably, the method further includes:

[0045] During or after the cooling control is executed, the actual temperature change data of the target large-volume concrete structure is acquired and compared with the predicted temperature change trend. The deviation is calculated to evaluate the temperature control effect.

[0046] Historical temperature control interaction data is stored in an experience playback buffer, and samples are periodically taken from it to fine-tune the model, enabling online updates of the temperature prediction model and control strategy;

[0047] The knowledge distillation technique is used to transfer the temperature evolution laws and control strategies learned by the trained and optimized complex teacher model to the lightweight student model to support real-time inference on edge devices.

[0048] Based on the temperature control effect evaluation results and the optimized model performance, a temperature control effect evaluation report is generated to provide optimization suggestions for temperature control schemes in subsequent large-volume concrete projects.

[0049] The second objective of this application is to provide an intelligent temperature control system for large-volume concrete based on a three-dimensional temperature field.

[0050] The second objective of this application is achieved through the following technical solution:

[0051] A smart temperature control system for large-volume concrete based on a three-dimensional temperature field, the system comprising:

[0052] The three-dimensional temperature field construction module is used to acquire multimodal image data of the surface of a target large-volume concrete structure and construct a continuous three-dimensional temperature field.

[0053] The anomaly identification and temperature prediction module is used to identify temperature anomaly areas in the target large-volume concrete structure based on the three-dimensional temperature field, and to predict the temperature change trend in the future preset period by combining the historical monitoring data of the target large-volume concrete structure, construction period environmental parameters and concrete mix proportion parameters.

[0054] The cooling control command generation module is used to generate a set of cooling control commands for regulating the zoned cooling system within the target large-volume concrete structure based on the predicted temperature change trend and current temperature state of the target large-volume concrete structure.

[0055] The adaptive execution and closed-loop adjustment module is used to execute cooling control based on the cooling control instruction set and to perform closed-loop adjustment based on the real-time feedback data of the target large-volume concrete structure until the internal temperature of the target large-volume concrete structure reaches the target range.

[0056] Preferably, the three-dimensional temperature field construction module is specifically used for:

[0057] A mobile platform equipped with an infrared thermal imager and a visible light camera is used to acquire multi-view images of the target large-volume concrete structure according to preset movement parameters, thereby obtaining visible light images and infrared thermal images.

[0058] The visible light image is subjected to geometric and radiometric correction, and feature points are extracted for three-dimensional reconstruction. The infrared thermal image is subjected to temperature correction based on a radiometric thermometry model.

[0059] Based on motion recovery structure and multi-view stereo vision algorithm, a high-precision three-dimensional point cloud model of the target large-volume concrete structure is constructed using the corrected visible light image.

[0060] The temperature data of the corrected infrared thermal image is projected onto the three-dimensional point cloud model through coordinate mapping to establish the correspondence between temperature data and geometric position;

[0061] A spatial interpolation algorithm is used to reconstruct the temperature field of the unmeasured region based on the mapped temperature data, generating a continuous three-dimensional temperature field.

[0062] The three-dimensional temperature field is displayed in the form of a color cloud map, and temperature isosurfaces, temperature slices and temperature gradient distribution maps are generated. The maximum temperature, minimum temperature, average temperature, surface-to-internal temperature difference and cooling rate and temperature statistical parameters are calculated and output.

[0063] Preferably, when constructing the three-dimensional temperature field, the three-dimensional temperature field construction module sets up control points on the surface of the concrete structure and adopts a temperature data fusion algorithm based on spatiotemporal consistency constraints. By introducing the temporal continuity constraint of the temperature field at adjacent moments, the temperature jump caused by the movement of the mobile platform is eliminated.

[0064] Preferably, when the anomaly identification and temperature prediction module performs the task of identifying temperature anomaly regions in the target large-volume concrete structure based on the three-dimensional temperature field, it is specifically used for:

[0065] Multi-scale analysis was performed on the three-dimensional temperature field to extract statistical characteristic parameters, including temperature peak, temperature gradient, surface-to-interior temperature difference, and cooling rate.

[0066] The multi-channel feature map, which integrates temperature value, temperature gradient, temporal evolution features, and spatial location information, is input into the deep learning segmentation model.

[0067] The deep learning segmentation model automatically identifies temperature anomaly regions and generates a spatial distribution map of these regions.

[0068] Preferably, when the anomaly identification and temperature prediction module performs the step of predicting the temperature change trend within a preset time period by combining historical monitoring data of the target large-volume concrete structure, construction period environmental parameters, and concrete mix proportion parameters, it is specifically used for:

[0069] Acquire historical three-dimensional temperature field data, construction period environmental parameters, and concrete mix proportion parameters from the historical monitoring data;

[0070] The historical three-dimensional temperature field data, construction period environmental parameters, and concrete mix proportion parameters are input into a dual-network temperature prediction model composed of a long short-term memory network that integrates physical constraints of heat conduction and a deep Q network. The dual-network temperature prediction model achieves collaborative optimization of temperature prediction and temperature control strategies by sharing the underlying feature representation and through joint training.

[0071] The long short-term memory network that integrates the physical constraints of heat conduction predicts the temperature evolution curve and the time of temperature peak occurrence within a preset period in the future. The temperature evolution curve is then used as the state input to the deep Q network to evaluate the long-term value of different cooling control actions and output the corresponding value function.

[0072] Preferably, the cooling control command generation module is specifically used for:

[0073] Construct a multi-objective reward function with temperature control deviation, energy consumption cost, and temperature distribution uniformity as optimization objectives;

[0074] The predicted temperature change trend and the current temperature state are input as environmental states into a near-end policy optimization algorithm based on the Actor-Critic architecture. The near-end policy optimization algorithm aims to maximize the long-term cumulative return of the multi-objective reward function and decides the cooling water flow rate and target temperature of each zone's cooling pipes.

[0075] The cooling water flow rate and target temperature output by the near-end strategy optimization algorithm are combined with the corresponding cooling pipe number and preset execution time to generate a control instruction set containing the cooling pipe number, flow rate set value and execution time.

[0076] Preferably, when the adaptive execution and closed-loop adjustment module executes cooling control based on the cooling control instruction set, it uses an adaptive parameter-tuned PID control algorithm to adjust the cooling water flow rate. The PID parameters of the PID control algorithm are dynamically adjusted according to the temperature deviation and the rate of change of the temperature deviation.

[0077] When the temperature deviation exceeds a preset large deviation threshold, the proportional coefficient in the PID parameter is increased to speed up the response.

[0078] When the temperature deviation is less than the preset small deviation threshold, the proportional coefficient in the PID parameters is reduced and the integral coefficient is increased to eliminate the steady-state error.

[0079] Preferably, the system further includes:

[0080] The digital twin simulation module is used to input the cooling control instruction set into the digital twin model for simulation and pre-run before executing cooling control based on the cooling control instruction set, to evaluate the effects of different control strategies, and to select the optimal strategy for execution.

[0081] Preferably, the system further includes an intelligent optimization and evaluation module, specifically used for:

[0082] During or after the cooling control is executed, the actual temperature change data of the target large-volume concrete structure is acquired and compared with the predicted temperature change trend. The deviation is calculated to evaluate the temperature control effect.

[0083] Historical temperature control interaction data is stored in an experience playback buffer, and samples are periodically taken from it to fine-tune the model, enabling online updates of the temperature prediction model and control strategy;

[0084] The knowledge distillation technique is used to transfer the temperature evolution laws and control strategies learned by the trained and optimized complex teacher model to the lightweight student model to support real-time inference on edge devices.

[0085] Based on the temperature control effect evaluation results and the optimized model performance, a temperature control effect evaluation report is generated to provide optimization suggestions for temperature control schemes in subsequent large-volume concrete projects.

[0086] Compared with the prior art, this application has the following beneficial effects:

[0087] 1. By using a long short-term memory network constrained by thermal conduction, the temperature evolution within a preset time period can be predicted with high accuracy. Combined with deep Q-network reinforcement learning, the optimal cooling strategy can be generated in advance. This transforms the passive mode of responding after the occurrence of an anomaly into an active intervention paradigm of predicting risks and actively regulating, effectively curbing temperature peaks and preventing cracks from forming.

[0088] 2. Utilize a mobile platform equipped with infrared and visible light cameras to perform fully automatic oblique photography, combine motion recovery structure and multi-view stereo vision algorithms to reconstruct centimeter-level geometric models, and generate a continuous and dynamically updated three-dimensional temperature field through radiation thermometry correction and spatial interpolation. This overcomes the shortcomings of traditional point sensors, such as low deployment density, easy damage, and discontinuous data, and provides complete and accurate spatial information support for global temperature control decisions.

[0089] 3. Based on multi-scale features (temperature gradient, surface-to-internal temperature difference, cooling rate, etc.) and deep learning segmentation models, it automatically identifies potential temperature anomaly areas and visualizes their spatial distribution, achieving a leap from local point monitoring to full-field intelligent diagnosis, significantly improving the ability to detect early risks.

[0090] 4. The cooling command generation adopts a multi-objective near-end strategy optimization reinforcement learning algorithm to comprehensively optimize temperature deviation, energy consumption and uniformity. The execution end adopts a deviation self-tuning PID controller, which accelerates the response when the deviation is large and suppresses overshoot and eliminates steady-state error when the deviation is small, ensuring that the control process is both efficient and stable.

[0091] 5. Before issuing control commands, simulate and rehearse using a digital twin model to select the best strategy, reduce trial and error costs, and automatically evaluate the effect after control is completed. Use experience playback and knowledge distillation technology to achieve online fine-tuning and lightweight deployment of the model, enabling the system to have continuous evolution capabilities.

[0092] 6. The three-dimensional temperature field is displayed intuitively in the form of cloud maps, isosurfaces, and slices, which facilitates on-site management. Temperature control effect evaluation reports are automatically generated, accumulating data assets and optimization experience for subsequent projects, and promoting the temperature control of large-volume concrete from experience-driven to data-intelligent driven.

[0093] In summary, this application systematically solves the core pain points of existing technologies, such as response lag, monitoring blind spots, rigid control, and lack of foresight, through a five-in-one technical system integrating high-precision sensing, intelligent prediction, autonomous decision-making, adaptive execution, and closed-loop optimization. It significantly improves the accuracy, intelligence level, and engineering reliability of temperature control for large-volume concrete. Attached Figure Description

[0094] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0095] Figure 1 This is a flowchart of a method for intelligent temperature control of large-volume concrete based on a three-dimensional temperature field in one embodiment of this application.

[0096] Figure 2 This is a schematic diagram of the structure of a large-volume concrete intelligent temperature control system based on a three-dimensional temperature field in one embodiment of this application;

[0097] Figure 3 This is a schematic diagram of the structure of a large-volume concrete intelligent temperature control system based on a three-dimensional temperature field in another embodiment of this application;

[0098] Figure 4 This is a network architecture diagram of a dual-network temperature prediction model in one embodiment of this application;

[0099] Figure 5 This is a flowchart of a digital twin simulation pre-run in one embodiment of this application. Detailed Implementation

[0100] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0101] In the embodiments provided in this application, it should be understood that the disclosed methods and systems can be implemented in other ways. The system embodiments described below are merely illustrative. For example, the division of units and modules is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or modules can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, and can be electrical, mechanical, or other forms.

[0102] like Figure 1 As shown in the figure, this application provides an intelligent temperature control method for large-volume concrete based on a three-dimensional temperature field. The method may include the following steps:

[0103] S1, acquire multimodal image data of the surface of the target large-volume concrete structure and construct a continuous three-dimensional temperature field;

[0104] S2 identifies temperature anomaly areas in the target large-volume concrete structure based on a three-dimensional temperature field, and predicts the temperature change trend within a preset time period by combining historical monitoring data of the target large-volume concrete structure, environmental parameters during construction, and concrete mix proportion parameters.

[0105] S3, Based on the predicted temperature change trend and current temperature state of the target large-volume concrete structure, generate a set of cooling control instructions for regulating the zoned cooling system within the target large-volume concrete structure.

[0106] S4 executes cooling control based on the cooling control instruction set and performs closed-loop adjustment based on the real-time feedback data of the target large-volume concrete structure until the internal temperature of the target large-volume concrete structure reaches the target range.

[0107] This embodiment provides an end-to-end intelligent temperature control method for large-volume concrete. Its core working principle lies in constructing a closed-loop intelligent control system encompassing perception, prediction, decision-making, execution, and feedback. First, multimodal image data of the surface of the target large-volume concrete structure (such as a bridge abutment) is acquired using non-contact methods, and a continuous three-dimensional temperature field model that fully reflects the spatial distribution of temperature on the structure's surface is constructed based on this data. Second, this three-dimensional temperature field model is used to identify potential temperature anomaly areas, and historical monitoring data of the structure, current environmental parameters during construction (such as air temperature, humidity, and wind speed), and the concrete's own mix proportion parameters (such as cement content and water-cement ratio) are integrated as input to predict the temperature change trend inside the concrete over a future period (e.g., 12-24 hours). Next, based on the predicted trend and the current real-time temperature status, a precise cooling control command set is automatically generated to regulate the zoned cooling system embedded within the concrete. Finally, this command set is executed, and real-time feedback data from inside the concrete is continuously collected, forming a closed-loop regulation that dynamically adjusts the cooling strategy until the internal temperature of the concrete is stably controlled within a preset safety target range.

[0108] This embodiment fundamentally changes the traditional passive response mode of temperature control. By introducing a high-precision three-dimensional temperature field as the basis for global perception and combining it with multi-source information such as historical data, environmental data, and material data for proactive prediction, it achieves a paradigm shift from remedial measures after temperature exceedance is detected to intervention measures before temperature exceedance is predicted. This method solves the decision-making blind spot problem caused by sparse monitoring points in traditional technologies, as well as the technical challenge of effectively curbing temperature peaks due to response lag. Through a closed-loop feedback mechanism, it ensures the accuracy and stability of the control process, significantly improves the overall efficiency of temperature control for large-volume concrete, effectively prevents the generation of temperature cracks, and ensures the durability and safety of the structure.

[0109] In one embodiment, acquiring multimodal image data of the surface of a target large-volume concrete structure and constructing a continuous three-dimensional temperature field includes:

[0110] By using a mobile platform equipped with an infrared thermal imager and a visible light camera, multi-view images of the target large-volume concrete structure are acquired according to preset movement parameters, and visible light images and infrared thermal images are obtained.

[0111] Geometric and radiometric corrections are performed on visible light images, and feature points are extracted for 3D reconstruction. Temperature corrections based on a radiometric thermometry model are performed on infrared thermal images.

[0112] Based on motion recovery structure and multi-view stereo vision algorithm, a high-precision three-dimensional point cloud model of the target large-volume concrete structure is constructed using the corrected visible light image.

[0113] The temperature data of the corrected infrared thermal image is projected onto a 3D point cloud model through coordinate mapping to establish the correspondence between temperature data and geometric position.

[0114] A spatial interpolation algorithm is used to reconstruct the temperature field of the unmeasured region based on the mapped temperature data, generating a continuous three-dimensional temperature field.

[0115] The system displays the three-dimensional temperature field in the form of a color cloud map, and generates temperature isosurfaces, temperature slices, and temperature gradient distribution maps. It calculates and outputs parameters including the highest temperature, lowest temperature, average temperature, surface-to-internal temperature difference, cooling rate, and temperature statistics.

[0116] In constructing a continuous three-dimensional temperature field, this embodiment first employs a mobile platform (such as a multi-rotor UAV) equipped with an infrared thermal imager and a visible light camera. Following a pre-planned flight path (setting flight altitude, speed, image overlap rate, and sampling frequency), it performs multi-view oblique photography of the target concrete structure, simultaneously acquiring visible light and infrared thermal images. To ensure data accuracy, high-contrast control points are deployed on the concrete structure surface for subsequent point cloud registration and accuracy correction. Subsequently, the acquired visible light images undergo distortion correction, illumination homogenization, and noise filtering, and SIFT feature points are extracted. Simultaneously, the infrared thermal images undergo non-uniformity correction, temperature calibration, and emissivity correction. The radiation thermometry model is crucial, and its calculation formula is as follows:

[0117]

[0118] In the formula, This is the corrected concrete surface temperature. The temperature value is measured by a thermal imager, and ε is the emissivity of the concrete surface. Here, α is the atmospheric temperature, α is the atmospheric attenuation coefficient, and R is the measurement distance.

[0119] Next, based on the Structure from Motion (SfM) algorithm and the Multiple View Stereo (MVS) algorithm, a high-precision 3D point cloud model is constructed using the corrected visible light image. Then, the temperature data from the corrected infrared thermal image is projected onto the 3D point cloud model through a precise coordinate mapping relationship, establishing the association between the geometric position of each point cloud and its corresponding temperature value.

[0120] Finally, a spatial interpolation algorithm based on the Radial Basis Function (RBF) is used to interpolate the temperature of regions in the point cloud where the temperature was not directly measured, thereby generating a continuous and smooth three-dimensional temperature field. The RBF interpolation function can be expressed as:

[0121]

[0122] in, For Gaussian radial basis functions, These are undetermined coefficients, which are determined by solving a system of linear equations.

[0123] This embodiment overcomes the limitations of traditional fixed or handheld infrared thermal imagers, which have limited monitoring range and are greatly affected by viewing angle and distance. Through fully automated, high-coverage oblique photography using a drone, surface data of large structures (such as bridge abutments) can be acquired efficiently and comprehensively. The introduction of control points and spatiotemporal consistency constraints further improves reconstruction accuracy. Three-dimensional reconstruction technology based on SfM and MVS ensures the accuracy of the geometric model, while rigorous radiation thermometry correction guarantees the authenticity of the temperature data. The resulting continuous three-dimensional temperature field not only eliminates the monitoring blind spots of traditional point sensors but also provides a complete, high-resolution global data foundation for subsequent anomaly identification, trend prediction, and precise control, serving as a prerequisite and key to achieving intelligent temperature control.

[0124] In one embodiment, when constructing a three-dimensional temperature field, control points are set up on the surface of the concrete structure, and a temperature data fusion algorithm based on spatiotemporal consistency constraints is adopted. By introducing the temporal continuity constraint of the temperature field at adjacent moments, the temperature jump caused by the movement of the mobile platform is eliminated.

[0125] During the construction of a three-dimensional temperature field, the movement of the mobile platform (such as a drone) can cause non-physical jumps in temperature data between adjacent moments, even with continuous observation of the same area, due to minute changes in shooting angle and distance. To address this issue, this embodiment introduces a temperature data fusion algorithm based on spatiotemporal consistency constraints during the three-dimensional temperature field construction stage. The core idea of ​​this algorithm is that the temperature field of a concrete structure has physical continuity in the time dimension, meaning that temperature changes between adjacent moments should be smooth. Therefore, when fusing temperature data from different time points, the temperature data fusion algorithm based on spatiotemporal consistency constraints explicitly introduces temporal continuity as a constraint. By minimizing the difference between the measured value and the optimized value, as well as the difference between the optimized values ​​between adjacent moments, it smooths and corrects temperature data noise or jumps introduced by platform movement, ensuring that the final generated three-dimensional temperature field maintains a high degree of consistency in both the spatiotemporal dimensions.

[0126] This embodiment effectively eliminates temperature data artifacts and jumps caused by the movement of the mobile observation platform itself, significantly improving the spatiotemporal consistency and reconstruction accuracy of the three-dimensional temperature field model, and providing key data quality assurance for temperature trend prediction and anomaly detection.

[0127] In one embodiment, identifying temperature anomaly regions in a target large-volume concrete structure based on a three-dimensional temperature field includes:

[0128] Multi-scale analysis of the three-dimensional temperature field was performed to extract statistical characteristic parameters, including temperature peak, temperature gradient, surface-to-interior temperature difference, and cooling rate.

[0129] The multi-channel feature map, which integrates temperature value, temperature gradient, temporal evolution features, and spatial location information, is input into the deep learning segmentation model.

[0130] The system automatically identifies temperature anomaly regions using a deep learning segmentation model and generates a spatial distribution map of these regions.

[0131] In this implementation, when automatically identifying risk areas from a constructed 3D temperature field, the first step is to perform multi-scale analysis on the 3D temperature field to calculate a series of key statistical characteristic parameters, including but not limited to temperature peak, temperature gradient, surface-to-internal temperature difference, and cooling rate. These parameters characterize the thermal state of concrete from different dimensions. Next, these features are fused with the original temperature values, time-series evolution features, and spatial location encoding information to form a multi-channel feature map. This feature map is input into a deep learning segmentation model (this embodiment uses an improved U-Net semantic segmentation network model). Through its encoder-decoder structure and skip connections, this model can effectively fuse multi-scale features, ultimately achieving pixel-level accurate segmentation of temperature anomaly areas (such as areas with excessively rapid temperature rise, excessively large gradients, and excessive surface-to-internal temperature differences), and outputting their specific distribution map in 3D space.

[0132] This embodiment abandons the traditional method of relying on manual experience to set thresholds for anomaly judgment, and achieves intelligent and automated identification of temperature anomalies. The improved U-Net network can comprehensively consider multi-dimensional information such as the absolute value of temperature, spatial gradient, temporal evolution trend, and geometric location, which greatly improves the accuracy and robustness of anomaly identification. The generated spatial distribution map of temperature anomaly areas provides clear target guidance for subsequent precise local control, making temperature control measures more targeted and avoiding energy waste caused by global cooling.

[0133] In one embodiment, predicting the temperature change trend within a preset time period by combining historical monitoring data of the target large-volume concrete structure, construction-period environmental parameters, and concrete mix proportion parameters includes:

[0134] Acquire historical three-dimensional temperature field data, construction period environmental parameters, and concrete mix proportion parameters from historical monitoring data;

[0135] Historical three-dimensional temperature field data, construction period environmental parameters, and concrete mix proportion parameters are input into a dual-network temperature prediction model composed of a long short-term memory network that integrates physical constraints of heat conduction and a deep Q network. The dual-network temperature prediction model achieves synergistic optimization of temperature prediction and temperature control strategies by sharing the underlying feature representation and through joint training.

[0136] By incorporating a long short-term memory network that integrates physical constraints of heat conduction, the temperature evolution curve and the time of temperature peak occurrence within a preset time period are predicted. The temperature evolution curve is then used as the state input to a deep Q-network to evaluate the long-term value of different cooling control actions and output the corresponding value function.

[0137] This embodiment predicts temperature change trends over a preset time period by constructing a model capable of high-precision, forward-looking temperature prediction and collaboratively optimizing temperature control strategies. Figure 4As shown, this model employs an innovative dual-network architecture:

[0138] A Long Short Term Memory (LSTM) network optimized based on physical constraints of thermal conduction is used. This network receives historical three-dimensional temperature field data, construction-period environmental parameters, and concrete mix proportion parameters as input. The physical constraints of thermal conduction are incorporated into the LSTM's loss function to ensure that the prediction results conform to the fundamental physical laws of heat conduction during concrete hydration. This physical information is added as a soft constraint to the loss function, guiding the LSTM to learn a temperature evolution pattern that conforms to physical laws. This network is responsible for learning the time-series characteristics of temperature and outputting the temperature evolution curve and the time of temperature peak occurrence within a preset future time period.

[0139] Deep Q-Network (DQN): This network uses the temperature evolution curve predicted by LSTM as its state input to evaluate the long-term cumulative reward (value function) that different cooling control actions (such as activating a certain partition or setting the flow rate) can obtain in the future. The two networks share the underlying feature representation and are jointly trained, so that the temperature prediction module can provide the policy decision module with the most favorable information to make the optimal decision, thereby achieving synergistic optimization of prediction and decision.

[0140] This embodiment addresses the shortcomings of traditional prediction models, such as lack of physical constraints, poor generalization ability, and the disconnect between prediction and decision-making. The LSTM with integrated physical constraints ensures the physical rationality of the prediction results, while DQN empowers the system to make decisions based on long-term interests. The collaborative optimization mechanism of the two networks ensures that predictions not only pursue accuracy but also usefulness for decision-making, resulting in a more forward-looking and globally optimal temperature control strategy, laying a solid foundation for proactive temperature control.

[0141] In one embodiment, based on the predicted temperature change trend of the target large-volume concrete structure and the current temperature state, a cooling control instruction set for regulating the zoned cooling system within the target large-volume concrete structure is generated, including:

[0142] Construct a multi-objective reward function with temperature control deviation, energy consumption cost, and temperature distribution uniformity as optimization objectives;

[0143] The predicted temperature change trend and the current temperature state are used as environmental state inputs to the near-end policy optimization algorithm based on the Actor-Critic architecture. The near-end policy optimization algorithm aims to maximize the long-term cumulative reward of the multi-objective reward function and decides the cooling water flow rate and target temperature of each zone's cooling pipes.

[0144] The cooling water flow rate and target temperature output by the near-end strategy optimization algorithm are combined with the corresponding cooling pipe number and preset execution time to generate a control instruction set containing the cooling pipe number, flow rate set value and execution time.

[0145] This embodiment transforms the prediction results into specific engineering instructions (i.e., a cooling control instruction set). First, a multi-objective reward function is constructed, which comprehensively quantifies three core optimization objectives: temperature control accuracy (minimizing control deviation), operational economy (minimizing energy consumption costs), and temperature control uniformity (minimizing the standard deviation of temperature distribution). The specific form of this reward function is as follows:

[0146]

[0147] in, To predict temperature, For the target temperature, For energy consumption costs, The standard deviation of temperature distribution (characterizing temperature control uniformity). The weighting coefficients are determined using the analytic hierarchy process (AHP).

[0148] Next, the predicted temperature change trend and the current real-time temperature state are combined as the environmental state and input into a proximal policy optimization (PPO) algorithm based on an Actor-Critic architecture. The training objective of the PPO algorithm is to maximize the long-term cumulative reward of the aforementioned multi-objective reward function. During the inference phase, the PPO Actor network outputs the optimal cooling action based on the current state, namely the cooling water flow rate and target temperature of each cooling pipe zone. Finally, these numerical actions are combined with the specific cooling pipe numbers and preset execution times to generate a complete set of control instructions that can be directly issued and executed.

[0149] This embodiment achieves a seamless transformation from abstract AI decisions to specific engineering instructions. By introducing a multi-objective reward function, it ensures that the generated cooling strategy is optimal in terms of overall performance, rather than optimal in terms of a single metric. The advanced PPO algorithm guarantees the stability and convergence speed of policy learning in complex, high-dimensional state spaces. The generated instruction set contains all the elements required for execution (piping, flow, time), and can directly drive field actuators, achieving a tight integration of intelligent decision-making and automated execution.

[0150] In one embodiment, when performing cooling control based on a cooling control instruction set, an adaptive parameter-tuned PID control algorithm is used to adjust the cooling water flow rate. The PID parameters of the PID control algorithm are dynamically adjusted according to the temperature deviation and the rate of change of the temperature deviation.

[0151] When the temperature deviation exceeds the preset large deviation threshold, increase the proportional coefficient in the PID parameters to speed up the response.

[0152] When the temperature deviation is less than the preset small deviation threshold, the proportional coefficient in the PID parameters is reduced and the integral coefficient is increased to eliminate the steady-state error.

[0153] To ensure the accuracy and robustness of flow control when executing cooling control commands, this embodiment employs an adaptive parameter tuning PID control algorithm. The core of this algorithm lies in the fact that its PID parameters (proportional coefficient Kp, integral coefficient Ki, and derivative coefficient Kd) are not fixed but dynamically adjusted based on the real-time temperature deviation and its rate of change.

[0154] The specific adaptive tuning strategy is as follows: when a large temperature deviation is detected (exceeding the preset large deviation threshold), the proportional coefficient Kp is automatically increased to improve the system's response speed and quickly suppress the temperature rise; while when the temperature deviation is small and close to the target value (below the preset small deviation threshold), the proportional coefficient Kp is decreased to avoid overshoot, and the integral coefficient Ki is increased to effectively eliminate residual steady-state error, ensuring that the temperature can eventually stabilize precisely within the target range.

[0155] This embodiment overcomes the limitations of traditional fixed-parameter PID controllers when dealing with large-scale, nonlinear temperature control processes. The adaptive tuning mechanism enables the controller to automatically adjust its dynamic characteristics according to different operating conditions, pursuing speed under large deviations and accuracy under small deviations, thus balancing the system's response speed and control precision, and effectively ensuring the reliability and efficiency of the cooling execution process.

[0156] In one embodiment, the intelligent temperature control method may further include, prior to executing cooling control based on a cooling control instruction set:

[0157] The cooling control command set is input into the digital twin model for simulation and pre-run, the effects of different control strategies are evaluated, and the optimal strategy is selected and executed.

[0158] Before actually issuing and executing the cooling control command set, this embodiment adds a simulation pre-run step. The command set to be executed is input into a pre-built digital twin model for digital twin simulation pre-run. The digital twin simulation pre-run process is as follows: Figure 5As shown, this digital twin model is a high-fidelity virtual mapping of the target large-volume concrete structure and its cooling system. Its core is solving the three-dimensional unsteady-state heat conduction equations that include cooling boundary conditions. By performing simulations within the digital twin, the expected effects of the strategy can be pre-evaluated, including key indicators such as the anticipated cooling rate, energy consumption level, and temperature field uniformity. Based on the simulation evaluation results, multiple candidate strategies can be compared, and the strategy with the best overall performance can be selected and implemented in the real physical system.

[0159] This embodiment, through the simulation-execution mode, greatly improves the success rate and reliability of temperature control operation. It can verify and optimize control strategies in a virtual environment at low cost and with zero risk, avoiding the possibility of temperature overshoot, energy waste, or even new risks caused by improper strategies in the real structure.

[0160] In one embodiment, the intelligent temperature control method may further include:

[0161] During or after the cooling control is executed, the actual temperature change data of the target large-volume concrete structure is acquired and compared with the predicted temperature change trend. The deviation is calculated to evaluate the temperature control effect.

[0162] Historical temperature control interaction data is stored in an experience playback buffer, and samples are periodically taken from it to fine-tune the model, enabling online updates of the temperature prediction model and control strategy;

[0163] The knowledge distillation technique is used to transfer the temperature evolution laws and control strategies learned by the trained and optimized complex teacher model to the lightweight student model to support real-time inference on edge devices.

[0164] Based on the temperature control effect evaluation results and the optimized model performance, a temperature control effect evaluation report is generated to provide optimization suggestions for temperature control schemes in subsequent large-volume concrete projects.

[0165] During or after the cooling control process, actual temperature change data of the concrete structure is acquired and compared with the predicted results to calculate the deviation, thereby objectively evaluating the effectiveness of the temperature control. The root mean square error (RMSE) can be used as the evaluation metric. All historical temperature control interaction data (including status, actions, rewards, etc.) is stored in an experience replay buffer. Samples are periodically extracted from this buffer to fine-tune the temperature prediction model and control strategy model (i.e., the near-end policy optimization algorithm model based on the Actor-Critic architecture), enabling online learning and continuous optimization. To further improve deployment efficiency, knowledge distillation technology is employed, the core of which is to allow a student model (lightweight) to mimic the output of a teacher model (complex). The student model obtained after knowledge distillation can be deployed on edge computing devices, supporting low-latency real-time inference. Finally, a temperature control effectiveness evaluation report is automatically generated, summarizing the experience and shortcomings of this control measure, and providing data-driven optimization suggestions for the design of temperature control schemes for similar subsequent projects.

[0166] This embodiment, through effect evaluation and online learning, continuously accumulates experience from practice, corrects prediction biases, and optimizes control strategies. Knowledge distillation technology solves the problem of deploying complex models on resource-constrained field equipment, ensuring the engineering practicality of advanced algorithms. The final evaluation report not only summarizes a single project but also helps promote the continuous improvement of temperature control standards across the entire industry.

[0167] like Figure 2 As shown in the figure, this application provides an intelligent temperature control system for large-volume concrete based on a three-dimensional temperature field. This intelligent temperature control system may include:

[0168] The three-dimensional temperature field construction module 201 is used to acquire multimodal image data of the surface of the target large-volume concrete structure and construct a continuous three-dimensional temperature field.

[0169] The anomaly identification and temperature prediction module 202 is used to identify temperature anomaly areas in the target large-volume concrete structure based on the three-dimensional temperature field, and to predict the temperature change trend in the future preset period by combining the historical monitoring data of the target large-volume concrete structure, construction period environmental parameters and concrete mix proportion parameters.

[0170] The cooling control instruction generation module 203 is used to generate a set of cooling control instructions for regulating the zoned cooling system within the target large-volume concrete structure based on the predicted temperature change trend and current temperature state of the target large-volume concrete structure.

[0171] The adaptive execution and closed-loop adjustment module 204 is used to execute cooling control based on the cooling control instruction set and to perform closed-loop adjustment based on the real-time feedback data of the target large-volume concrete structure until the internal temperature of the target large-volume concrete structure reaches the target range.

[0172] In one embodiment, the three-dimensional temperature field construction module 201 is specifically used for:

[0173] By using a mobile platform equipped with an infrared thermal imager and a visible light camera, multi-view images of the target large-volume concrete structure are acquired according to preset movement parameters, and visible light images and infrared thermal images are obtained.

[0174] Geometric and radiometric corrections are performed on visible light images, and feature points are extracted for 3D reconstruction. Temperature corrections based on a radiometric thermometry model are performed on infrared thermal images.

[0175] Based on motion recovery structure and multi-view stereo vision algorithm, a high-precision three-dimensional point cloud model of the target large-volume concrete structure is constructed using the corrected visible light image.

[0176] The temperature data of the corrected infrared thermal image is projected onto a 3D point cloud model through coordinate mapping to establish the correspondence between temperature data and geometric position.

[0177] A spatial interpolation algorithm is used to reconstruct the temperature field of the unmeasured region based on the mapped temperature data, generating a continuous three-dimensional temperature field.

[0178] The system displays the three-dimensional temperature field in the form of a color cloud map, and generates temperature isosurfaces, temperature slices, and temperature gradient distribution maps. It calculates and outputs parameters including the highest temperature, lowest temperature, average temperature, surface-to-internal temperature difference, cooling rate, and temperature statistics.

[0179] In one embodiment, when constructing a three-dimensional temperature field, the three-dimensional temperature field construction module 201 sets up control points on the surface of the concrete structure and adopts a temperature data fusion algorithm based on spatiotemporal consistency constraints. By introducing the temporal continuity constraint of the temperature field at adjacent moments, the temperature jump caused by the movement of the mobile platform is eliminated.

[0180] In one embodiment, when the anomaly identification and temperature prediction module 202 performs temperature anomaly region identification based on a three-dimensional temperature field in a target large-volume concrete structure, it is specifically used for:

[0181] Multi-scale analysis of the three-dimensional temperature field was performed to extract statistical characteristic parameters, including temperature peak, temperature gradient, surface-to-interior temperature difference, and cooling rate.

[0182] The multi-channel feature map, which integrates temperature value, temperature gradient, temporal evolution features, and spatial location information, is input into the deep learning segmentation model.

[0183] The system automatically identifies temperature anomaly regions using a deep learning segmentation model and generates a spatial distribution map of these regions.

[0184] In one embodiment, when the anomaly identification and temperature prediction module 202 performs the step of predicting the temperature change trend within a preset time period by combining historical monitoring data of the target large-volume concrete structure, construction period environmental parameters, and concrete mix proportion parameters, it is specifically used for:

[0185] Acquire historical three-dimensional temperature field data, construction period environmental parameters, and concrete mix proportion parameters from historical monitoring data;

[0186] Historical three-dimensional temperature field data, construction period environmental parameters, and concrete mix proportion parameters are input into a dual-network temperature prediction model composed of a long short-term memory network that integrates physical constraints of heat conduction and a deep Q network. The dual-network temperature prediction model achieves synergistic optimization of temperature prediction and temperature control strategies by sharing the underlying feature representation and through joint training.

[0187] By incorporating a long short-term memory network that integrates physical constraints of heat conduction, the temperature evolution curve and the time of temperature peak occurrence within a preset time period are predicted. The temperature evolution curve is then used as the state input to a deep Q-network to evaluate the long-term value of different cooling control actions and output the corresponding value function.

[0188] In one embodiment, the cooling control command generation module 203 is specifically used for:

[0189] Construct a multi-objective reward function with temperature control deviation, energy consumption cost, and temperature distribution uniformity as optimization objectives;

[0190] The predicted temperature change trend and the current temperature state are used as environmental state inputs to the near-end policy optimization algorithm based on the Actor-Critic architecture. The near-end policy optimization algorithm aims to maximize the long-term cumulative reward of the multi-objective reward function and decides the cooling water flow rate and target temperature of each zone's cooling pipes.

[0191] The cooling water flow rate and target temperature output by the near-end strategy optimization algorithm are combined with the corresponding cooling pipe number and preset execution time to generate a control instruction set containing the cooling pipe number, flow rate set value and execution time.

[0192] In one embodiment, when the adaptive execution and closed-loop adjustment module 204 executes cooling control based on the cooling control instruction set, it uses an adaptive parameter-tuned PID control algorithm to adjust the cooling water flow rate. The PID parameters of the PID control algorithm are dynamically adjusted according to the temperature deviation and the rate of change of the temperature deviation.

[0193] When the temperature deviation exceeds the preset large deviation threshold, increase the proportional coefficient in the PID parameters to speed up the response.

[0194] When the temperature deviation is less than the preset small deviation threshold, the proportional coefficient in the PID parameters is reduced and the integral coefficient is increased to eliminate the steady-state error.

[0195] In one embodiment, the intelligent temperature control system may further include:

[0196] The digital twin simulation module is used to input the cooling control instruction set into the digital twin model for simulation and pre-run before executing cooling control based on the cooling control instruction set, to evaluate the effect of different control strategies, and to select the optimal strategy for execution.

[0197] In one embodiment, the intelligent temperature control system may further include an intelligent optimization and evaluation module, specifically used for:

[0198] During or after the cooling control is executed, the actual temperature change data of the target large-volume concrete structure is acquired and compared with the predicted temperature change trend. The deviation is calculated to evaluate the temperature control effect.

[0199] Historical temperature control interaction data is stored in an experience playback buffer, and samples are periodically taken from it to fine-tune the model, enabling online updates of the temperature prediction model and control strategy;

[0200] The knowledge distillation technique is used to transfer the temperature evolution laws and control strategies learned by the trained and optimized complex teacher model to the lightweight student model to support real-time inference on edge devices.

[0201] Based on the temperature control effect evaluation results and the optimized model performance, a temperature control effect evaluation report is generated to provide optimization suggestions for temperature control schemes in subsequent large-volume concrete projects.

[0202] It should be noted that the working principle and technical effect of the intelligent temperature control system for large-volume concrete based on a three-dimensional temperature field in the above embodiments are the same as those of the intelligent temperature control method for large-volume concrete based on a three-dimensional temperature field in the above embodiments, and will not be repeated here.

[0203] like Figure 3 The diagram shown is a structural schematic of a large-volume concrete intelligent temperature control system based on a three-dimensional temperature field, according to another embodiment of this application. The intelligent temperature control system includes:

[0204] The UAV 3D reconstruction subsystem includes: a multi-rotor UAV platform equipped with an infrared thermal imager and a visible light camera; a 3D reconstruction module for constructing a high-precision 3D point cloud model of the target large-volume concrete structure using corrected visible light images based on motion recovery structure and multi-view stereo vision algorithms; and a temperature field construction module for fusing temperature data with the 3D point cloud.

[0205] The intelligent decision-making subsystem includes: a temperature anomaly identification module, used to identify temperature anomaly regions in the target large-volume concrete structure using a three-dimensional temperature field based on a convolutional neural network; a temperature prediction module, used to predict the temperature change trend within a preset time period based on the physical constraints of heat conduction combined with a dual network architecture of LSTM and DQN, combined with historical monitoring data of the target large-volume concrete structure, construction period environmental parameters and concrete mix proportion parameters; and an intelligent decision-making module, used to evaluate the long-term value of different cooling control actions based on a deep reinforcement learning algorithm and output the corresponding value function.

[0206] The adaptive control subsystem includes: a cooling pipe network embedded in concrete; an intelligent valve control module for regulating the flow rate of each pipe; and a PID controller for achieving precise flow control.

[0207] The closed-loop feedback subsystem includes: a temperature sensor network for real-time monitoring of the internal temperature of the concrete; a flow sensor for monitoring the cooling water flow rate; and a data feedback module for real-time feedback of the temperature data of the target large-volume concrete structure to form a closed-loop control.

[0208] The central control and visualization subsystem includes: a central processing unit, which runs intelligent control algorithms; a 3D visualization module, which displays the temperature field and control effect in real time; and an early warning and report generation module, which generates early warning signals and temperature control effect evaluation reports.

[0209] Specifically, the cooling pipe network adopts a layered and zoned layout, with each layer of cooling pipes divided into multiple independent control zones, each zone being independently controlled by an intelligent solenoid valve; based on the distribution of abnormal temperature zones, the cooling pipes of the corresponding zones are selectively activated to achieve precise local cooling.

[0210] Specifically, the intelligent valve control module adopts a flow regulation strategy based on fuzzy control algorithm, dynamically adjusting the valve opening according to temperature deviation and temperature change rate; when the temperature exceeds the warning threshold, a rapid opening strategy is adopted; when the temperature approaches the target value, a gradual adjustment strategy is adopted to avoid temperature overshoot.

[0211] Specifically, the intelligent temperature control system also includes a simulation pre-operation module based on digital twins. Before the actual temperature control operation is performed, a simulation pre-operation is carried out in the digital twin model to evaluate the effect of different control strategies. The optimal control strategy is selected before the actual operation is performed, thereby improving the success rate and reliability of the control.

[0212] Specifically, the central control and visualization subsystem adopts WebGL-based 3D visualization technology to realize real-time rendering and interactive browsing of the temperature field. It supports dynamic playback of temperature fields at multiple time steps, displays the temperature change process, provides temperature profile analysis function, and allows viewing of temperature distribution in any direction.

[0213] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0214] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for intelligent temperature control of large-volume concrete based on a three-dimensional temperature field, characterized in that, The method includes the following steps: Collect multimodal image data of the surface of the target large-volume concrete structure and construct a continuous three-dimensional temperature field; Based on the three-dimensional temperature field, temperature anomaly regions in the target large-volume concrete structure are identified, and the temperature change trend in the future preset period is predicted by combining the historical monitoring data of the target large-volume concrete structure, construction period environmental parameters and concrete mix proportion parameters. Based on the predicted temperature change trend and current temperature state of the target large-volume concrete structure, a cooling control instruction set is generated to regulate the zoned cooling system within the target large-volume concrete structure. Cooling control is executed based on the cooling control instruction set, and closed-loop adjustment is performed based on the real-time feedback data of the target large-volume concrete structure until the internal temperature of the target large-volume concrete structure reaches the target range.

2. The intelligent temperature control method for large-volume concrete based on a three-dimensional temperature field according to claim 1, characterized in that, The process of acquiring multimodal image data of the surface of the target large-volume concrete structure and constructing a continuous three-dimensional temperature field includes: A mobile platform equipped with an infrared thermal imager and a visible light camera is used to acquire multi-view images of the target large-volume concrete structure according to preset movement parameters, thereby obtaining visible light images and infrared thermal images. The visible light image is subjected to geometric and radiometric correction, and feature points are extracted for three-dimensional reconstruction. The infrared thermal image is subjected to temperature correction based on a radiometric thermometry model. Based on motion recovery structure and multi-view stereo vision algorithm, a high-precision three-dimensional point cloud model of the target large-volume concrete structure is constructed using the corrected visible light image. The temperature data of the corrected infrared thermal image is projected onto the three-dimensional point cloud model through coordinate mapping to establish the correspondence between temperature data and geometric position; A spatial interpolation algorithm is used to reconstruct the temperature field of the unmeasured region based on the mapped temperature data, generating a continuous three-dimensional temperature field. The three-dimensional temperature field is displayed in the form of a color cloud map, and temperature isosurfaces, temperature slices and temperature gradient distribution maps are generated. The maximum temperature, minimum temperature, average temperature, surface-to-internal temperature difference and cooling rate and temperature statistical parameters are calculated and output.

3. The intelligent temperature control method for large-volume concrete based on a three-dimensional temperature field according to claim 1 or 2, characterized in that, When constructing the three-dimensional temperature field, control points are set up on the surface of the concrete structure, and a temperature data fusion algorithm based on spatiotemporal consistency constraints is adopted. By introducing the temporal continuity constraint of the temperature field at adjacent moments, the temperature jump caused by the movement of the moving platform is eliminated.

4. The intelligent temperature control method for large-volume concrete based on a three-dimensional temperature field according to claim 1, characterized in that, The identification of temperature anomaly regions in the target large-volume concrete structure based on the three-dimensional temperature field includes: Multi-scale analysis was performed on the three-dimensional temperature field to extract statistical characteristic parameters, including temperature peak, temperature gradient, surface-to-interior temperature difference, and cooling rate. The multi-channel feature map, which integrates temperature value, temperature gradient, temporal evolution features, and spatial location information, is input into the deep learning segmentation model. The deep learning segmentation model automatically identifies temperature anomaly regions and generates a spatial distribution map of these regions.

5. The intelligent temperature control method for large-volume concrete based on a three-dimensional temperature field according to claim 1, characterized in that, The prediction of temperature change trends within a preset time period, based on historical monitoring data of the target large-volume concrete structure, construction-period environmental parameters, and concrete mix proportion parameters, includes: Acquire historical three-dimensional temperature field data, construction period environmental parameters, and concrete mix proportion parameters from the historical monitoring data; The historical three-dimensional temperature field data, construction period environmental parameters, and concrete mix proportion parameters are input into a dual-network temperature prediction model composed of a long short-term memory network that integrates physical constraints of heat conduction and a deep Q network. The dual-network temperature prediction model achieves collaborative optimization of temperature prediction and temperature control strategies by sharing the underlying feature representation and through joint training. The long short-term memory network that integrates the physical constraints of heat conduction predicts the temperature evolution curve and the time of temperature peak occurrence within a preset period in the future. The temperature evolution curve is then used as the state input to the deep Q network to evaluate the long-term value of different cooling control actions and output the corresponding value function.

6. The intelligent temperature control method for large-volume concrete based on a three-dimensional temperature field according to claim 1, characterized in that, The step of generating a cooling control instruction set for regulating the zoned cooling system within the target large-volume concrete structure based on the predicted temperature change trend and current temperature state includes: Construct a multi-objective reward function with temperature control deviation, energy consumption cost, and temperature distribution uniformity as optimization objectives; The predicted temperature change trend and the current temperature state are input as environmental states into a near-end policy optimization algorithm based on the Actor-Critic architecture. The near-end policy optimization algorithm aims to maximize the long-term cumulative return of the multi-objective reward function and decides the cooling water flow rate and target temperature of each zone's cooling pipes. The cooling water flow rate and target temperature output by the near-end strategy optimization algorithm are combined with the corresponding cooling pipe number and preset execution time to generate a control instruction set containing the cooling pipe number, flow rate set value and execution time.

7. The intelligent temperature control method for large-volume concrete based on a three-dimensional temperature field according to claim 1, characterized in that, When executing cooling control based on the aforementioned cooling control command set, an adaptive parameter-tuned PID control algorithm is used to adjust the cooling water flow rate. The PID parameters of the PID control algorithm are dynamically adjusted according to the temperature deviation and the rate of change of the temperature deviation. When the temperature deviation exceeds a preset large deviation threshold, the proportional coefficient in the PID parameter is increased to speed up the response. When the temperature deviation is less than the preset small deviation threshold, the proportional coefficient in the PID parameters is reduced and the integral coefficient is increased to eliminate the steady-state error.

8. The intelligent temperature control method for large-volume concrete based on a three-dimensional temperature field according to claim 1, characterized in that, Before performing cooling control based on the cooling control instruction set, the method further includes: The cooling control command set is input into the digital twin model for simulation and pre-run, the effects of different control strategies are evaluated, and the optimal strategy is selected and executed.

9. The intelligent temperature control method for large-volume concrete based on a three-dimensional temperature field according to any one of claims 1-8, characterized in that, The method further includes: During or after the cooling control is executed, the actual temperature change data of the target large-volume concrete structure is acquired and compared with the predicted temperature change trend. The deviation is calculated to evaluate the temperature control effect. Historical temperature control interaction data is stored in an experience playback buffer, and samples are periodically taken from it to fine-tune the model, enabling online updates of the temperature prediction model and control strategy; The knowledge distillation technique is used to transfer the temperature evolution laws and control strategies learned by the trained and optimized complex teacher model to the lightweight student model to support real-time inference on edge devices. Based on the temperature control effect evaluation results and the optimized model performance, a temperature control effect evaluation report is generated to provide optimization suggestions for temperature control schemes in subsequent large-volume concrete projects.

10. A smart temperature control system for large-volume concrete based on a three-dimensional temperature field, characterized in that, The system includes: The three-dimensional temperature field construction module is used to acquire multimodal image data of the surface of a target large-volume concrete structure and construct a continuous three-dimensional temperature field. The anomaly identification and temperature prediction module is used to identify temperature anomaly areas in the target large-volume concrete structure based on the three-dimensional temperature field, and to predict the temperature change trend in the future preset period by combining the historical monitoring data of the target large-volume concrete structure, construction period environmental parameters and concrete mix proportion parameters. The cooling control command generation module is used to generate a set of cooling control commands for regulating the zoned cooling system within the target large-volume concrete structure based on the predicted temperature change trend and current temperature state of the target large-volume concrete structure. The adaptive execution and closed-loop adjustment module is used to execute cooling control based on the cooling control instruction set and to perform closed-loop adjustment based on the real-time feedback data of the target large-volume concrete structure until the internal temperature of the target large-volume concrete structure reaches the target range.

Images