Intelligent control method and system for temperature field of pavement asphalt mixture paving, computer device and computer readable storage medium

Abstract

This application relates to a method, system, computer equipment, and computer-readable storage medium for intelligent control of the temperature field in asphalt pavement paving. The method includes: collecting infrared temperature matrix, internal temperature gradient data, and environmental and equipment operating parameters of the paving area; compensating and correcting occluded areas in the infrared temperature matrix based on the temperature gradient data; fusing the corrected surface temperature matrix with the internal temperature gradient data to form a three-dimensional temperature field tensor; inputting the tensor into a prediction model and combining it with environmental and equipment parameters to output predicted future temperature field data; generating a hierarchical control instruction set based on local temperature differences, temperature change slopes, and minimum temperature thresholds in the predicted data, and distributing it to the corresponding execution systems; collecting actual temperature data after instruction execution, calculating errors, and dynamically updating the prediction model parameters. This application improves the uniformity, density, and long-term durability of asphalt pavement paving.

Description

Technical Field

[0001] This application relates to the field of road engineering construction technology, and in particular to a method, system, computer equipment and computer-readable storage medium for intelligent control of the temperature field of asphalt mixture paving. Background Technology

[0002] With the continuous expansion of highway infrastructure construction in my country, the quality of asphalt pavement construction has become a core element concerning road durability, safety, and driving comfort. During the asphalt mixture paving process, temperature is a decisive factor affecting the compactibility of the mixture, the uniformity of aggregate distribution, and the final pavement density and smoothness. In traditional construction practices, temperature monitoring mainly relies on manual handheld infrared thermometers for fixed-point, intermittent measurements, or the use of fixed infrared thermal imagers to perform two-dimensional scanning of the paved surface, combined with the experience of construction personnel to roughly adjust parameters such as the screed heating power and travel speed of the paver. While this method can reflect surface temperature distribution to some extent, it is limited by the singular measurement method and the subjectivity of manual judgment, making it difficult to achieve full-section, continuous, and high-precision temperature field management.

[0003] Currently, when dealing with the complex and ever-changing paving site environment, conventional infrared thermography can only acquire a two-dimensional temperature matrix of the paved layer surface. It cannot penetrate the material surface to perceive the internal temperature gradient and the dynamic process of heat conduction. This results in the understanding of the temperature field remaining at the surface level, making it difficult to reveal early trends of internal heat accumulation or dissipation. At the same time, phenomena such as water vapor, dust, and smoke from the mixture at the site can easily obstruct the infrared sensor's line of sight, causing temperature data loss or severe distortion, creating monitoring blind spots, and thus affecting the timely identification of local low or high temperature areas.

[0004] Secondly, existing control methods are essentially passive "monitor-response" models, making lagging adjustments based on current or short-term historical temperature data. Since the process window for asphalt mixtures from paving to compaction is extremely short, this lagging control often fails to prevent quality defects such as uneven compaction, segregation, or decreased smoothness caused by temperature inconsistencies. This problem is particularly pronounced under abnormal conditions such as strong winds, low temperatures, or interruptions in material supply. Furthermore, the paving temperature field is influenced by multiple factors, including ambient wind speed, air humidity, material thermal properties, and equipment operating parameters. Traditional experience-based control struggles to quantify the interactions of these factors, resulting in coarse control commands, insufficient self-adaptive capabilities, and difficulty in ensuring consistent and stable construction quality. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides a method, system, computer equipment, and computer-readable storage medium for intelligent control of the temperature field in asphalt mixture paving.

[0006] Firstly, this application provides a method for intelligent control of the temperature field of asphalt mixture paving, employing the following technical solution:

[0007] Real-time acquisition of infrared temperature matrix, environmental parameters, equipment operating parameters, and internal temperature gradient data collected by millimeter-wave radar in the asphalt mixture paving area of ​​the road surface, generating a timestamp-aligned multi-source dataset;

[0008] Based on the internal temperature gradient data, the occlusion area in the infrared temperature matrix is ​​compensated and corrected to construct a corrected temperature matrix.

[0009] The corrected temperature matrix is ​​fused with the internal temperature gradient data to construct a three-dimensional temperature field tensor that includes the surface temperature and the rate of temperature change in the depth direction.

[0010] The three-dimensional temperature field tensor is input into a pre-built prediction model, and combined with environmental parameters and equipment operating parameters, the residual correction module constrained by the physical heat transfer equation outputs future temperature field prediction data.

[0011] Based on the local temperature difference, temperature change slope, and minimum temperature threshold of the predicted future temperature field data, a hierarchical control instruction set is generated; wherein, the hierarchical control instruction set includes instructions for adjusting the heating power of the screed zone, instructions for adjusting the paving speed, and instructions for setting the insulation temperature of the material transport vehicle.

[0012] The graded control instruction set is distributed to the paver's zone heating system, paving speed controller, and material transport vehicle insulation system to execute corresponding operations;

[0013] The actual temperature data after the execution of the hierarchical control instruction set is collected, the error index is calculated, and the parameters of the prediction model are dynamically updated.

[0014] By adopting the above technical solutions, the traditional paving temperature management model, which relies on manual experience, suffers from delayed response, and employs extensive control methods, has been fundamentally changed. By integrating infrared and millimeter-wave data, the system gains a three-dimensional understanding of the temperature field that penetrates the surface and reaches the interior. Through a hybrid prediction model combining physical equations and artificial intelligence, it achieves high-precision prediction of temperature evolution trends. Furthermore, the hierarchical preventive control commands generated based on the predicted data can proactively eliminate temperature differences, smooth temperature gradients, and ensure the temperature of the material source, preventing quality defects such as insufficient compaction, segregation, and poor smoothness caused by uneven temperature from the source. Finally, through an online learning mechanism, the system's long-term adaptability and robustness in complex and ever-changing construction environments are ensured.

[0015] Furthermore, based on the internal temperature gradient data, the occlusion areas in the infrared temperature matrix are compensated and corrected. The step of constructing the corrected temperature matrix includes:

[0016] Acquire reflection intensity data collected by millimeter-wave radar;

[0017] The coordinates of the obstructed area in the infrared temperature matrix are identified based on the reflection intensity data; wherein, the obstructed area includes coordinate points where the reflection intensity is lower than a preset intensity threshold;

[0018] Extract the internal temperature gradient data corresponding to the coordinates of the shading area, and extract the temperature values ​​of the unshading area within a preset range adjacent to the coordinates of the shading area;

[0019] Based on the temperature value of the unshaded area and the internal temperature gradient data, the compensation temperature value of the shaded area is calculated.

[0020] The original temperature value of the corresponding occluded area in the infrared temperature matrix is ​​replaced by the compensated temperature value to generate a corrected temperature matrix.

[0021] Furthermore, the step of fusing the corrected temperature matrix with the internal temperature gradient data to construct a three-dimensional temperature field tensor containing the surface temperature and the rate of temperature change in the depth direction includes:

[0022] Based on the spatial resolution of the correction temperature matrix, a three-dimensional mesh coordinate system matching the paving area is established; wherein, the horizontal resolution of the three-dimensional mesh coordinate system is consistent with the pixel spacing of the correction temperature matrix, and layer depth nodes are set in the vertical direction.

[0023] Assign the temperature values ​​of each coordinate point in the corrected temperature matrix to the corresponding surface nodes of the three-dimensional mesh coordinate system;

[0024] Based on the internal temperature gradient data, the first-order temperature change rate of each surface node in the depth direction is calculated.

[0025] Based on the temperature values ​​of the surface nodes and the first-order temperature change rate in the depth direction, the temperature values ​​of each depth node in the vertical direction in the three-dimensional mesh coordinate system are generated by the heat conduction integral algorithm.

[0026] By integrating the temperature values ​​of the surface nodes and the nodes at each depth, a three-dimensional temperature field tensor is constructed.

[0027] Furthermore, the steps of inputting the three-dimensional temperature field tensor into a pre-built prediction model, combining environmental parameters and equipment operating parameters, and outputting future temperature field prediction data through a residual correction module constrained by the physical heat transfer equation include:

[0028] The three-dimensional temperature field tensor is input into the physical equation calculation unit, and the physical heat transfer equation is solved by combining real-time environmental parameters and equipment operating parameters, and the basic temperature evolution sequence is output.

[0029] The basic temperature evolution sequence is input into the pre-trained artificial neural network in the residual correction module to calculate the prediction bias of the physical equation calculation unit.

[0030] The base temperature evolution sequence is superimposed with the prediction deviation to generate a corrected temperature sequence;

[0031] The corrected temperature sequence is input into the time series prediction unit, which outputs future temperature field prediction data.

[0032] Furthermore, the step of generating a hierarchical control instruction set based on the local temperature difference, temperature change slope, and minimum temperature threshold of the predicted future temperature field data includes:

[0033] Acquire future temperature field prediction data, including temperature values ​​at spatial coordinate points and their temporal evolution sequences;

[0034] Based on the predicted future temperature field data, a temperature field analysis operation is performed, and a hierarchical control instruction set is generated based on the analysis results, specifically including:

[0035] The paving area is divided into grid cells, and the difference between the highest and lowest temperatures in each grid cell is calculated to generate a local temperature difference distribution map. When the local temperature difference exceeds the first set threshold, the location and power adjustment amount of the ironing plate heating zone are determined based on the local temperature difference distribution map, and an ironing plate zone heating power adjustment command is generated.

[0036] A temperature profile is captured along the paving direction, the temperature change per unit length is calculated, and a temperature change slope sequence is generated. When the temperature change slope exceeds the second set threshold, a paving speed correction value is calculated based on the slope change trend, and a paving speed adjustment command is generated.

[0037] Scan the future temperature field prediction data, identify the lowest temperature value within a specified time period, and when the lowest temperature value is lower than the insulation threshold, obtain the real-time location information of the material transport vehicle and calculate the insulation temperature compensation amount, and generate the material transport vehicle insulation temperature setting instruction.

[0038] Furthermore, the steps of collecting actual temperature data after the execution of the hierarchical control instruction set, calculating error indicators, and dynamically updating the parameters of the prediction model include:

[0039] After the hierarchical control instruction set is executed, the actual infrared temperature matrix and internal temperature gradient data of the paving area are collected within a specified time window.

[0040] The actual infrared temperature matrix is ​​spatially aligned with the future temperature field prediction data at the corresponding timestamp to generate a surface temperature deviation matrix.

[0041] Based on the difference in the rate of temperature change in the depth direction between the internal temperature gradient data and the future temperature field prediction data, the gradient direction residual of each spatial point is calculated.

[0042] By fusing the surface temperature deviation matrix and the gradient direction residual, a multi-dimensional error tensor is constructed, with dimensions including horizontal spatial coordinates and depth direction.

[0043] Based on the deviation magnitude of each spatial location in the multidimensional error tensor, the update weight coefficients of the prediction model parameters are dynamically calculated.

[0044] Using the backpropagation algorithm, the weighted multidimensional error tensor is used as the input to the loss function to iteratively update the neural network parameters of the residual correction module in the prediction model.

[0045] Furthermore, the intelligent control method also includes:

[0046] Real-time thickness distribution data of asphalt mixture paving layers are collected using a laser thickness gauge to generate a thickness offset matrix.

[0047] Calculate the heat capacity correction factor for the local region based on the thickness offset matrix;

[0048] The attenuation gradient of the rate of temperature change in the depth direction in the three-dimensional temperature field tensor is adjusted according to the heat capacity correction coefficient.

[0049] Obtain the real-time operating frequency of the paver's vibration compaction mechanism, and calculate the change in material porosity based on the real-time operating frequency;

[0050] The equivalent thermal conductivity in the physical heat transfer equation is corrected based on the change in the porosity of the material.

[0051] The corrected attenuation gradient and equivalent thermal conductivity are input into the prediction model for real-time parameter calibration.

[0052] Secondly, this application provides an intelligent temperature field control system for asphalt mixture paving, which adopts the following technical solution:

[0053] The multi-source data acquisition module is used to collect infrared temperature matrix, environmental parameters, equipment operating parameters, and internal temperature gradient data collected by millimeter-wave radar in the asphalt mixture paving area in real time, and generate a timestamp-aligned multi-source dataset.

[0054] The occlusion compensation and correction module is used to compensate and correct the occlusion area in the infrared temperature matrix based on the internal temperature gradient data, and to construct a correction temperature matrix.

[0055] The three-dimensional temperature field reconstruction module is used to fuse the corrected temperature matrix with the internal temperature gradient data to construct a three-dimensional temperature field tensor that includes the surface temperature and the rate of temperature change in the depth direction.

[0056] The physical information fusion prediction module is used to input the three-dimensional temperature field tensor into a pre-built prediction model, combine environmental parameters and equipment operating parameters, and output future temperature field prediction data through the residual correction module constrained by the physical heat transfer equation.

[0057] The multi-level decision-making and instruction generation module is used to generate a hierarchical control instruction set based on the local area temperature difference, temperature change slope, and minimum temperature threshold of the future temperature field prediction data; wherein, the hierarchical control instruction set includes instructions for adjusting the heating power of the screed zone, instructions for adjusting the paving speed, and instructions for setting the insulation temperature of the material transport vehicle.

[0058] The instruction distribution and execution module is used to distribute the hierarchical control instruction set to the paver zone heating system, paving speed controller and material transport vehicle insulation system to perform corresponding operations;

[0059] The model optimization module is used to collect the actual temperature data after the execution of the hierarchical control instruction set, calculate the error index, and dynamically update the parameters of the prediction model.

[0060] Thirdly, this application provides a computer device, which adopts the following technical solution:

[0061] A computer device includes a memory, a processor, and a computer program stored in the memory, the processor executing the computer program to perform the steps of the method as described in the first aspect.

[0062] Fourthly, this application provides a computer-readable storage medium, which adopts the following technical solution:

[0063] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as in any of the methods in the first aspect.

[0064] In summary, this application includes at least one of the following beneficial technical effects: comprehensive perception of the temperature field is achieved through multi-source data fusion; intelligent prediction of the temperature field is carried out using a prediction model constrained by the physical heat transfer equation; and the heating power of the paver, paving speed, and material truck insulation are coordinated and regulated by a hierarchical control strategy, thereby improving the uniformity and stability of the asphalt mixture paving temperature, effectively reducing temperature segregation, improving the paving quality of the road surface, and reducing energy consumption. Attached Figure Description

[0065] Figure 1 This is a schematic diagram of the first process of a method for intelligent control of the temperature field of asphalt mixture paving in one embodiment of this application.

[0066] Figure 2 This is a schematic diagram of the second process of the intelligent control method for the temperature field of asphalt mixture paving in one embodiment of this application.

[0067] Figure 3 This is a schematic diagram of the third process of the intelligent control method for the temperature field of asphalt mixture paving in one embodiment of this application.

[0068] Figure 4 This is a schematic diagram of the fourth process of the intelligent control method for the temperature field of asphalt mixture paving in one embodiment of this application.

[0069] Figure 5 This is a schematic diagram of the fifth process of the intelligent control method for the temperature field of asphalt mixture paving in one embodiment of this application.

[0070] Figure 6 This is a schematic diagram of the sixth process of the intelligent control method for the temperature field of asphalt mixture paving in one embodiment of this application.

[0071] Figure 7 This is a schematic diagram of the seventh process of the intelligent control method for the temperature field of asphalt mixture paving in one embodiment of this application. Detailed Implementation

[0072] To make the purpose, technical solution, and advantages of this application clearer, the following description is provided in conjunction with the appendix. Figures 1-7 The present application will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the application.

[0073] This application discloses a method for intelligent control of the temperature field of asphalt mixture paving.

[0074] Reference Figure 1 A method for intelligent control of the temperature field of asphalt mixture paving, specifically including:

[0075] Step S101: Real-time acquisition of infrared temperature matrix, environmental parameters, equipment operating parameters, and internal temperature gradient data acquired by millimeter-wave radar in the asphalt mixture paving area of ​​the road surface, generating a timestamp-aligned multi-source dataset.

[0076] Specifically, a multi-dimensional, synchronized on-site data acquisition network is constructed. An infrared sensor array is used to rapidly scan the surface of the paving layer in a non-contact manner, forming a two-dimensional infrared temperature matrix. This matrix reflects the spatial distribution of temperature in the transverse and longitudinal directions of the road surface. However, its drawback is that it can only acquire surface information and is easily affected by water vapor, dust, and other obstructions.

[0077] To overcome this limitation, millimeter-wave radar was introduced. Millimeter waves have a certain penetrating power into non-metallic materials (such as asphalt mixtures), enabling the detection of dielectric property changes at specific depths below the material surface. Dielectric properties are closely related to temperature. Through signal inversion, internal temperature gradient data can be obtained, revealing the rate and trend of heat transfer from the interior of the mixture to the surface. This is crucial for determining internal temperature uniformity and the cooling process. In this embodiment, the internal temperature gradient data is collected at a depth of 0-5 cm below the surface of the asphalt mixture.

[0078] Environmental monitoring parameters collected by the environmental monitoring equipment include wind speed, wind direction, and air humidity. Equipment operating parameters include paving speed, heating power of each zone of the screed, and GPS location information of the material transport vehicle. Furthermore, this step emphasizes the spatiotemporal synchronization of data acquisition; for example, each pixel of the infrared matrix should be aligned in time and space with each detection point of the millimeter-wave radar to ensure the effectiveness of subsequent data fusion.

[0079] Step S102: Based on the internal temperature gradient data, compensate and correct the occlusion area in the infrared temperature matrix to construct a correction temperature matrix.

[0080] At the paving site, moisture evaporation or dust may temporarily obstruct the view of some infrared sensors, resulting in "obstructed areas" in the infrared temperature matrix that produce invalid or severely distorted data. In this case, directly using the surface temperature of adjacent unobstructed areas for interpolation is inaccurate because it ignores any abnormal heat dissipation or heat accumulation that may occur at that point due to obstruction.

[0081] In this embodiment, synchronously acquired internal temperature gradient data is used as the calibration benchmark. The calibration algorithm formula for occlusion area compensation is: Compensation temperature value = Calibration coefficient × Millimeter wave temperature gradient value + (1 - Calibration coefficient) × Average temperature of adjacent areas. The calibration coefficient ranges from 0.6 to 0.8.

[0082] Specifically, the heat conduction process within the mixture is relatively stable and less affected by momentary surface shading. For a shaded surface point, the corresponding internal temperature gradient data can reflect the overall thermal trend at that location (e.g., a high internal temperature and a large gradient should indicate a high surface temperature). This correction algorithm essentially weights and fuses the "internal temperature gradient value" (representing the inferred temperature based on physical laws) with the "average temperature of adjacent regions" (representing the statistical temperature of the spatial neighborhood). A higher weight is given to the internal gradient (calibration coefficient 0.6-0.8) because it provides unique physical information about the shaded point, thereby generating a more accurate and complete "corrected temperature matrix," eliminating data blind spots and providing a reliable surface temperature field for subsequent analysis.

[0083] Step S103: The calibration temperature matrix is ​​fused with the internal temperature gradient data to construct a three-dimensional temperature field tensor that includes the surface temperature and the rate of temperature change in the depth direction.

[0084] The calibration temperature matrix provides the temperature value at each point on the surface of the paved layer (XY plane). The internal temperature gradient data is essentially a set of vectors that indicate the rate and direction of temperature change with depth (Z-axis) below each surface point (XY coordinates). ).

[0085] Specifically, the fusion process is not a simple superposition. Instead, it uses each surface spatial location as an anchor point, correlating the surface temperature value (from the calibration matrix) at that point with its corresponding temperature change rate along the depth direction (from millimeter-wave data), thus constructing a three-dimensional data volume. Mathematically, this data volume can be represented as a "three-dimensional temperature field tensor," where each element (x, y) contains not only a temperature scalar (surface temperature T_surface) but also a vector representing the intensity of internal heat conduction (temperature gradient ∇T). This deepens the system's understanding of the temperature field from "a surface temperature image" to "a spatial entity model with internal heat conduction dynamics," laying a solid foundation for accurately predicting how temperature evolves over time.

[0086] Step S104: Input the three-dimensional temperature field tensor into the pre-built prediction model, combine environmental parameters and equipment operating parameters, and output the future temperature field prediction data through the residual correction module constrained by the physical heat transfer equation.

[0087] Traditional data-driven prediction models are prone to failure or producing predictions that violate physical principles when training data is insufficient or when encountering new working conditions. In this embodiment, a physical heat transfer equation, with heat conduction as its core, is used to calculate the input three-dimensional temperature field tensor. This equation includes parameters such as wind speed, material thermal properties, and paver travel speed. Based on the law of conservation of energy, the physical heat transfer equation deduces the theoretical evolution trend of the temperature field in the future, ensuring the basic physical rationality of the prediction results.

[0088] Furthermore, due to the complexity of the actual paving environment, there are many unmodeled factors (such as material inhomogeneity and variations in the bottom foundation temperature). Therefore, the residual correction module (such as a deep learning network) added in this embodiment does not directly predict the temperature, but rather learns the deviation pattern between the physical model's prediction results and actual historical data, i.e., the "residual". This module corrects the physical prediction results and compensates for nonlinear effects that the physical model fails to cover.

[0089] Step S105: Based on the local area temperature difference, temperature change slope, and minimum temperature threshold of the future temperature field prediction data, a graded control instruction set is generated; wherein, the graded control instruction set includes the screed zone heating power adjustment instruction, paving speed adjustment instruction, and material transport vehicle insulation temperature setting instruction.

[0090] Specifically, the system analyzes and predicts three key features in the data:

[0091] (1) When the temperature difference in a local area exceeds the set threshold (e.g., 5°C), it indicates that the area will soon have a difference in compaction due to uneven cooling. Therefore, the heating power adjustment command of the ironing plate is triggered to provide directional heating to the low-temperature area.

[0092] (2) If the slope of temperature change along the paving direction exceeds the critical value (e.g., greater than 2℃ / m), it indicates that the temperature difference of the mixture in front of or behind the paver is too large, and continuous compaction will produce quality fluctuations. Therefore, the paving speed adjustment command is triggered (e.g., slow down the speed to cool the high-temperature material in front, or speed up the speed to reduce the exposure time of the low-temperature material behind) to smooth the temperature front.

[0093] (3) If it is predicted that the minimum temperature at a certain point will be lower than the "insulation threshold" for effective compaction of asphalt (for example, when the temperature is lower than 140°C after 10 minutes), it means that the mixture has been overcooled in the transport vehicle or while waiting for paving. Therefore, the insulation temperature setting command of the transport vehicle is triggered to regulate the insulation temperature of the transport vehicle and increase the insulation setting value of the subsequent transport vehicles in advance.

[0094] In some embodiments, the calculation formula for the insulation temperature setting command of the material transport vehicle is: target temperature = reference temperature + time compensation coefficient × max(0, estimated arrival time - critical time); the reference temperature is 150-155℃, and the time compensation coefficient is 0.8-1.2.

[0095] Step S106: Distribute the graded control instruction set to the paver zone heating system, paving speed controller and material transport vehicle insulation system to perform the corresponding operations;

[0096] The system distributes commands to corresponding execution terminals through different interfaces and communication protocols based on command type. For screed temperature adjustment requiring rapid and precise response, the command is sent to the paver's temperature control system, which employs a semiconductor thermoelectric device, enabling instantaneous compensation for localized temperatures with millisecond-level response speeds. For paving speeds affecting the overall continuity of the process, commands are sent to the paver's travel controller for smooth speed switching. For material transport vehicles requiring remote control, commands are sent via high-speed wireless communication (such as 5G) to their microwave-assisted heating system, allowing for early intervention in the temperature of the mixture within the hopper.

[0097] Step S107: Collect actual temperature data after the execution of the graded control instruction set, calculate error index and dynamically update the parameters of the prediction model.

[0098] After issuing and executing control commands, the system collects new "actual temperature data" again through the sensor network and compares it with the previously generated "predicted temperature data" to calculate an error index. This error not only measures the accuracy of the prediction but also indirectly reflects whether the actual response of the temperature field under control intervention meets expectations.

[0099] Specifically, when the error continues to exceed the allowable range, it indicates that the current operating condition has exceeded the existing cognitive range of the prediction model (especially the residual correction module). At this point, the system triggers a "dynamic update," using new data to fine-tune the model parameters (incremental learning). During training and updating, the loss function "introduces constraints from the physical heat transfer equation," meaning that the optimization process must not only make the predicted values ​​close to the measured values ​​but also ensure that the corrected model output does not violate the fundamental laws of thermodynamics. This allows the system to continuously accumulate experience and adapt to different materials, environments, and equipment conditions over long-term use, maintaining high prediction accuracy and control effectiveness.

[0100] The above implementation fundamentally changes the traditional paving temperature management model, which relies on manual experience, suffers from delayed response, and employs extensive control methods. By integrating infrared and millimeter-wave data, the system gains a three-dimensional understanding of the temperature field that penetrates the surface and reaches the interior. Through a hybrid prediction model combining physical equations and artificial intelligence, it achieves high-precision prediction of temperature evolution trends. Furthermore, the tiered preventive control commands generated based on the predicted data can proactively eliminate temperature differences, smooth temperature gradients, and ensure the temperature of the material source, preventing quality defects such as insufficient compaction, segregation, and poor smoothness caused by uneven temperature from the source. Finally, through an online learning mechanism, the system's long-term adaptability and robustness in complex and ever-changing construction environments are ensured.

[0101] Reference Figure 2 As one implementation of step S102, the step of compensating and correcting the occlusion area in the infrared temperature matrix based on internal temperature gradient data and constructing the corrected temperature matrix includes:

[0102] Step S201: Obtain the reflection intensity data collected by the millimeter-wave radar;

[0103] The reflection intensity data reflects the energy of the emitted millimeter-wave signal reflected back from the surface and shallow layers of the propagation medium (i.e., asphalt mixture). As a non-metallic composite material, the dielectric constant of asphalt mixture changes with increasing temperature. Millimeter-wave radar emits electromagnetic waves into the paving layer and receives its reflected echoes. The obtained reflection intensity data not only contains surface reflection information but also includes comprehensive information about the interaction between the beam and the material as the beam penetrates to a certain depth (0-5 cm in this embodiment) below the surface.

[0104] Understandably, this reflection intensity data is strongly affected by the surface roughness of the medium, moisture content, and especially surface deposits (such as water films and dust). When the infrared sensor's line of sight is obstructed by water vapor or dust, these obstructions will also significantly attenuate or scatter the millimeter-wave signal, leading to a decrease in its reflection intensity. Therefore, an abnormal decrease in reflection intensity can serve as a reliable indirect indicator for identifying areas of physical line-of-sight obstruction in infrared thermometry.

[0105] Step S202: Identify the coordinates of the occlusion area in the infrared temperature matrix based on the reflection intensity data; wherein, the occlusion area includes coordinate points where the reflection intensity is lower than a preset intensity threshold.

[0106] Specifically, the obstructed area refers to the spatial location where suspended particles such as water vapor and dust are located between the infrared sensor and the paved surface, causing the infrared radiation signal to be absorbed or scattered, thus rendering the infrared temperature measurement value invalid or severely distorted. By setting a "preset intensity threshold," the system can perform binarization analysis on the synchronously acquired millimeter-wave reflection intensity map. Coordinates with abnormally low reflection intensity (below the threshold) may correspond to severely interfered points with extremely high water content or abnormal dust accumulation; but more importantly, through spatial coordinate mapping, the system can accurately locate areas where the signal is normal in the millimeter-wave image but the corresponding data is abnormal in the infrared image. These areas are the coordinates of the "blind spots" formed by the pure infrared obstruction effect. This positioning method overcomes the inherent difficulty of accurately distinguishing between "low-temperature points" and "obstruction points" from the infrared image itself alone.

[0107] Step S203: Extract the internal temperature gradient data corresponding to the coordinates of the shading area, and extract the temperature values ​​of the unshading area within a preset range adjacent to the coordinates of the shading area.

[0108] For a coordinate point that is determined to be shaded, although its actual surface temperature information is missing, it does not exist in isolation. First, the transfer of heat in a continuous medium (such as asphalt mixture) has spatial continuity. The temperature distribution of the unshaded adjacent area around the point (i.e., the "unshaded area") constitutes an important spatial constraint and reference background for inferring the temperature of that point.

[0109] In some embodiments, with the occlusion coordinates as the center, the temperature values ​​of all coordinate points not marked as occlusion areas within a preset range (e.g., a circular area with a radius of 3 pixels) in the infrared temperature matrix are extracted; this is known as "temperature neighborhood environment information." Secondly, although the infrared signal at the occlusion point is invalid, its corresponding millimeter-wave internal temperature gradient data remains a valid physical measurement. This gradient data directly reflects the intensity of heat conduction in the vertical direction at that point and is a key internal state variable for inferring its surface temperature.

[0110] Step S204: Calculate the compensation temperature value of the shaded area based on the temperature value of the unshaded area and the internal temperature gradient data.

[0111] In this process, information from different sensors, reflecting different dimensions, is weighted and fused to calculate the most likely temperature of the occluded point using optimal estimation theory. This step is not a simple interpolation or averaging, but a data fusion decision.

[0112] Specifically, the "temperature value of the unshaded area" represents a statistical inference based on spatial continuity. It is usually the median (rather than the mean) to resist the interference of a few outliers that may exist in the neighborhood, which provides a spatial background benchmark for temperature. The "internal temperature gradient data" provides an inference based on physical laws: a large gradient value means that heat is being transferred rapidly from below the point to the surface, and its surface temperature should be higher than the average trend of the neighborhood; and vice versa.

[0113] In some embodiments, the formula for calculating the compensation temperature value is: Compensation temperature value = weighting coefficient × millimeter wave temperature gradient value + (1 - weighting coefficient) × median temperature of adjacent regions.

[0114] The weighting coefficient (usually between 0.6 and 0.8) gives higher confidence to the physical gradient information from inside the material, because the internal thermal state is less affected by the instantaneous shading of the surface and can better reflect the true thermodynamic state at that point.

[0115] Furthermore, the weighting coefficient is dynamically adjusted based on ambient humidity: in high humidity (≥70%) environments, water vapor is the primary obstruction, causing severe failure of infrared thermometry. However, millimeter waves are relatively less sensitive to water vapor, making their gradient data more reliable; therefore, their weight is increased (e.g., to 0.75). In low humidity environments, dust may be the primary obstruction, making the assumption of continuous neighborhood temperature more robust; therefore, the millimeter wave weight is appropriately reduced (e.g., to 0.65). This dynamic adjustment enables the compensation algorithm to be environmentally adaptive, ensuring that it can output near-realistic compensation values ​​under different operating conditions.

[0116] Step S205: Replace the original temperature value of the corresponding occluded area in the infrared temperature matrix with the compensated temperature value to generate a corrected temperature matrix.

[0117] The system has calculated a theoretically more reasonable compensation temperature value for each identified occlusion coordinate point. By traversing all occlusion area coordinates, the system replaces the potentially distorted or missing "original temperature values" at corresponding positions in the original infrared temperature matrix with the calculated compensation temperature values. The resulting "full-area corrected temperature matrix" eliminates data blind spots caused by on-site occlusion and restores the spatial continuity and physical rationality of the temperature field.

[0118] It should be noted that this corrected temperature matrix is ​​no longer just the raw readings of the infrared sensor, but an enhanced data product that integrates the internal sensing information of the millimeter-wave radar, spatial statistical laws, and physical constraints. Its data quality is significantly higher than that of the original infrared matrix, providing crucial high-quality input for subsequent construction of accurate three-dimensional temperature field models and reliable temperature prediction and control decisions.

[0119] In the above implementation, the reflection intensity of millimeter-wave radar is used as a "diagnostic tool" for infrared occlusion, achieving accurate and adaptive identification of occluded areas. By fusing the spatial temperature statistics of the unoccluded neighborhood with the deep physical information of the internal temperature gradient of the occluded point, and introducing an environmentally adaptive intelligent weighting mechanism, high-precision compensation calculation for the temperature of the occluded point is achieved. Ultimately, this technical solution can dynamically correct the problem of missing or distorted infrared temperature measurement data caused by uncontrollable factors such as moisture and dust on site, outputting a complete and reliable full-frame corrected temperature field image. This technical solution fundamentally improves the quality and reliability of the front-end sensing data of the entire intelligent control system, and is a key pre-requisite technical guarantee to ensure the accuracy and effectiveness of subsequent temperature field modeling, prediction, and control decisions, enhancing its practicality and stability in actual complex and harsh construction environments.

[0120] Reference Figure 3 As one implementation of step S103, the step of fusing the corrected temperature matrix with the internal temperature gradient data to construct a three-dimensional temperature field tensor containing the surface temperature and the rate of temperature change in the depth direction includes:

[0121] Step S301: Based on the spatial resolution of the calibration temperature matrix, establish a three-dimensional mesh coordinate system that matches the paving area; wherein, the horizontal resolution of the three-dimensional mesh coordinate system is consistent with the pixel spacing of the calibration temperature matrix, and layer depth nodes are set in the vertical direction.

[0122] Specifically, a discretized computational framework, strictly corresponding to physical space, is created to hold and correlate all temperature data. This framework is a three-dimensional grid coordinate system, and its core design follows a balance between physical realism and computational efficiency. In the horizontal direction (XY plane), the grid resolution (i.e., the distance between adjacent grid points) must be exactly consistent with the pixel spacing of the "correction temperature matrix." This design is necessary to ensure that every temperature value from the two-dimensional image (correction temperature matrix) can find a unique and precise corresponding surface node coordinate in the three-dimensional grid, achieving seamless mapping and avoiding information distortion caused by coordinate misalignment or scale inconsistency.

[0123] In the vertical direction (Z-axis), a series of layered depth nodes need to be set up. These nodes are not uniformly spaced, but typically distributed exponentially, with a higher node density in the surface to 5cm depth range than in deeper areas. The physical principle behind this is that after asphalt mixture is laid, the surface area experiences rapid heat loss due to contact with air, resulting in very rapid and significant temperature changes within the shallowest few centimeters. As depth increases, heat transfer is influenced by the underlying foundation and the changes become more gradual. Therefore, arranging high-density nodes within the 0-5cm range near the surface can accurately capture this abrupt temperature gradient. In deeper areas, the node spacing can be appropriately increased to reduce model complexity and computational burden, while still ensuring a reasonable description of the deeper temperature trends.

[0124] Step S302: Assign the temperature values ​​of each coordinate point in the calibration temperature matrix to the corresponding surface nodes of the three-dimensional mesh coordinate system.

[0125] In this process, the temperature value stored at each pixel location in the corrected temperature matrix is ​​assigned one-to-one to the "surface node" in the mesh that perfectly coincides with that pixel's coordinates. This is a computationally simple mapping operation, meaning that the temperature state at the top layer of the model is no longer an estimate or interpolation result, but rather derived from corrected real measurement data. The temperature distribution of the surface nodes after assignment constitutes the Dirichlet boundary condition of the three-dimensional temperature field at the top boundary (Z=0 plane). In subsequent algorithms, any calculation of the internal node temperature must start from this solid and accurate surface reference.

[0126] Step S303: Based on the internal temperature gradient data, calculate the first-order temperature change rate of each surface node in the depth direction;

[0127] The internal temperature gradient data measured by millimeter-wave radar is a field distribution data that reflects the rate of temperature change along the radar beam path (vertical direction). This step requires resolving this field data to each of the established surface grid nodes to calculate the rate of temperature change per unit depth directly below each node, i.e., the "first-order temperature change rate in the depth direction" (mathematically expressed as...). Specifically, the first-order rate of temperature change is a direct indicator of the intensity of heat conduction: a positive value indicates that the temperature increases with depth, which may mean that the temperature of the bottom mixture is higher or that the surface dissipates heat faster; a negative value indicates that the temperature decreases with depth.

[0128] However, raw millimeter-wave data may contain high-frequency noise or minute fluctuations due to local material inhomogeneities. Therefore, it is usually necessary to perform Gaussian smoothing filtering on the calculated temperature change rates of adjacent nodes, with the filter window radius related to the thermal conductivity of the paving material. The purpose of this filtering is to suppress non-physical, localized, drastic fluctuations, as the heat conduction process is continuous and the temperature gradient is typically smooth in space. The filter window size is set according to the thermal diffusivity of the material, which ensures that the smoothing process itself conforms to physical laws, resulting in a depth-direction temperature change rate sequence that reflects the overall trend while filtering out measurement noise.

[0129] Step S304: Based on the temperature values ​​of the surface nodes and the first-order temperature change rate in the depth direction, the temperature values ​​of each depth node in the vertical direction in the three-dimensional mesh coordinate system are generated by the heat conduction integral algorithm.

[0130] This method employs integral calculus to reconstruct the continuous temperature distribution across the entire depth dimension, starting from the known surface state and vertical variation patterns. Specifically, for each vertical line in the 3D mesh (i.e., a column of nodes with the same X and Y coordinates but different Z coordinates), the surface node temperature value is used as the initial value (starting point) of the vertical line temperature, and the calculated "depth-direction temperature change rate" from that surface node downwards is used as the slope function to guide how the temperature changes.

[0131] At this point, the heat conduction integral algorithm proceeds step by step along the depth direction (starting from the surface node and moving downwards to each preset depth node). Its basic logic is: the temperature of the next layer node is equal to the temperature of the previous layer node plus the temperature increment between the two layers determined by the rate of temperature change. The specific calculation formula is: Temperature of layer k = Surface temperature + Σ(Temperature change rate of layer i × Spacing between nodes of layer i); where k represents the target layer number; during the accumulation process, i represents the currently calculated depth layer number, which iterates from 1 to k.

[0132] Through this integral calculation, the system no longer needs to directly measure the temperature of each layer (which is difficult to implement in engineering). Instead, it utilizes readily available surface temperature and vertical gradient information, employing rigorous mathematical and physical methods to calculate the temperature value of each node from the surface down to a predetermined depth. This technical solution converts the indirect gradient information provided by millimeter-wave radar into directly usable three-dimensional absolute temperature lattice data, efficiently and cost-effectively constructing a complete three-dimensional temperature field.

[0133] Step S305: Integrate the temperature values ​​of the surface nodes and each depth node to construct a three-dimensional temperature field tensor.

[0134] The tensor dimension is the number of pixels in the width direction × the number of pixels in the length direction × the number of depth layers.

[0135] Specifically, the first dimension is the "number of pixels in the width direction," corresponding to the number of grid points in the horizontal (X-axis) direction of the tiling; the second dimension is the "number of pixels in the length direction," corresponding to the number of grid points in the vertical (Y-axis) direction of the tiling; and the third dimension is the "number of depth layers," corresponding to the number of node layers in the vertical direction (Z-axis). Any element T[x,y,z] in the tensor uniquely represents the temperature value of the point with coordinates (x,y,z) in space.

[0136] In addition, during the construction process, post-processing such as bilinear interpolation can be performed at the interface where material properties change to eliminate numerical abrupt changes that may be caused by discretization and differences in the thermal properties of different materials, making the tensor data field physically smoother and more reasonable.

[0137] In the above embodiments, non-contact gradient measurement using millimeter-wave radar is used as a bridge, with precise surface temperature as an anchor point. Through a heat conduction integral algorithm that conforms to the physical laws of heat conduction, the temperature distribution at any depth within the material is efficiently and economically derived. The resulting three-dimensional temperature field tensor not only boasts high spatial resolution and complete data dimensions, but its generation process also deeply integrates physical constraints. This allows the model to not only characterize the static distribution of temperature in space but also implicitly reflect the dynamic trends of heat transfer, providing a refined and realistic data foundation for subsequent temperature evolution prediction, construction quality analysis, and intelligent control decisions.

[0138] Reference Figure 4 As one implementation of step S104, the step of inputting the three-dimensional temperature field tensor into a pre-built prediction model, combining environmental parameters and equipment operating parameters, and outputting future temperature field prediction data through a residual correction module constrained by the physical heat transfer equation includes:

[0139] Step S401: Input the three-dimensional temperature field tensor into the physical equation calculation unit, combine real-time environmental parameters and equipment operating parameters to solve the physical heat transfer equation, and output the basic temperature evolution sequence.

[0140] This process utilizes classical thermodynamic laws to numerically simulate the natural evolution of the temperature field under given initial and boundary conditions, establishing a predictive baseline that conforms to physical laws. The obtained three-dimensional temperature field tensor (containing the complete initial temperature distribution and internal gradient) is input into the "Physical Equation Calculation Module" to solve the following physical heat transfer equation. This equation is a modified heat conduction physical model used to describe and predict the dynamic changes in the paving layer's temperature field. The calculation formula is as follows:

[0141] ;

[0142] In the above formula, ρ is the density of the asphalt mixture, and C p Let be the specific heat capacity of the asphalt mixture; T be the temperature; is the spatial and time function T(x,y,z,t) to be solved; t be the time; ∇ be the gradient operator; and k be the equivalent thermal conductivity of the asphalt mixture. This is the velocity vector of the paver, representing the macroscopic velocity of the asphalt mixture. This is the thermal diffusivity term, which describes the process of heat conduction from high-temperature regions to low-temperature regions within a material. The convection term describes the impact of material flow caused by paver movement on heat transport; β is the comprehensive environmental impact coefficient, v wind For real-time wind speed, This is an environmental factor, which mainly describes the effect of wind cooling on heat loss from the paved surface. The exponential relationship of wind speed (0.8) is often used to approximate actual convective heat dissipation.

[0143] Specifically, the equation is a partial differential equation. The left side of the equation represents the rate of change of the material's internal energy over time. The first term on the right side describes the spatial diffusion of heat caused by the material's own thermal conductivity (conduction). The second term describes the material advection transport effect caused by the movement of the paver (velocity vector v). The third term introduces the influence of wind speed, a key environmental parameter, on surface heat dissipation.

[0144] In this embodiment, the equation is solved using numerical methods (such as the finite difference method or the finite element method). This allows for a computer simulation of how the temperature field will change step-by-step over a future period, starting from the current moment and governed by purely physical laws, without considering any unknown disturbances or model errors. The simulation result is output as a "basic temperature evolution sequence," a set of chronological snapshots of the temperature field at future moments, representing the theoretically predicted trajectory under the most ideal conditions and in accordance with known physical laws.

[0145] Step S402: Input the basic temperature evolution sequence into the pre-trained artificial neural network in the residual correction module to calculate the prediction bias of the physical equation calculation unit.

[0146] In this embodiment, the pre-trained artificial neural network adopts a spatiotemporal convolutional architecture. Its input layer includes: the gradient tensor of the basic temperature evolution sequence, the encoding of the paving material type, and the mean of historical prediction deviations. The output is a residual tensor with the same dimension as the basic sequence.

[0147] Understandably, in complex real-world paving environments, there are still many disturbances that are difficult to model or measure accurately, such as local material inhomogeneities, base layer temperature variations, minute sensor errors, and spatiotemporal fluctuations in coefficients (e.g., k, β) in the equations. These factors can all lead to a repeatable but mechanistically complex deviation pattern between the physical model's predictions and the actual situation. The residual correction module is designed to capture and learn this pattern.

[0148] Specifically, the input to the artificial neural network is not only the base temperature sequence itself, but may also include the gradient tensor of the sequence (reflecting the change pattern), the paving material type encoding (due to the different thermal properties of different asphalt formulations), and the historical prediction bias mean (capturing the statistical characteristics of systematic errors). This network is trained on a large amount of historical data, and its learning objective is: given the prediction sequence of the physical model and the historical context, to output a prediction bias tensor of the same dimension as the physical prediction sequence. This prediction bias tensor is the "error" that the physical model might make in the current context, inferred by the neural network based on the complex nonlinear mapping relationships it has learned.

[0149] Step S403: Overlay the basic temperature evolution sequence with the prediction bias to generate the corrected temperature sequence;

[0150] This step is a key operation to enable the physical model and the data-driven model to work together. Its logical principle is to integrate the baseline prediction, which represents the theoretical law, with the deviation estimate, which represents the empirical correction, so as to generate an optimized prediction result that both follows the physical laws and is close to the laws of reality.

[0151] Specifically, the basic temperature evolution sequence output by the physics module (representing the theoretical prediction) is added to the prediction bias tensor output by the residual correction module (representing the correction amount inferred by the AI). The physical model provides the main framework and evolution trend of the prediction, ensuring that the results macroscopically conform to the basic laws of thermodynamics and avoiding physical inconsistencies (such as energy non-conservation) that may occur in pure data models. The AI ​​model, based on historical experience, fine-tunes the physical model for systematic overestimation or underestimation under specific complex conditions, compensating for errors caused by the simplification of the physical model. The resulting corrected temperature sequence, under ideal conditions, maintains the rationality of the physical prediction while further improving the accuracy of data-driven predictions. Theoretically, its prediction accuracy will surpass any scheme that uses only a physical model or a pure data model.

[0152] Step S404: Input the corrected temperature sequence into the time series prediction unit and output the future temperature field prediction data.

[0153] This method utilizes time series analysis to extrapolate recent temperature evolution trends that have already undergone physical-data fusion correction, thereby obtaining more distant and continuous future predictions. The corrected temperature series typically covers high-confidence temperature field evolution from the current moment to a relatively short future time window.

[0154] However, for the intelligent control of this application embodiment, it may be necessary to predict the temperature state over a longer period (such as the next few minutes). Therefore, the time series prediction module can employ a recurrent neural network structure such as a gated recurrent unit (GRU) to learn and extrapolate the temporal dynamic patterns inherent in the corrected sequence. Networks such as GRU have memory capabilities and can capture the dependencies and evolutionary inertia of the temperature field in the time dimension. This module uses the corrected temperature sequence as the initial input and historical context, and through its internal cyclic calculations, gradually deduce the temperature field state of multiple frames in the future, exceeding the length of the initial sequence.

[0155] Specifically, the formula for initializing the hidden layer state of the time series prediction module is:

[0156] h_0 = f(W_h·[T_t,∇T_z] + b_h); where W_h represents the weight matrix, T_t represents the 3D temperature field tensor at the current time (time t), ∇T_z represents the spatial mean of the temperature gradient along the depth direction (Z-axis), and b_h represents the bias vector. Optimizing the initial hidden state h_0 (e.g., fusing the current temperature field T_t and the depth gradient mean ∇T_z) helps inject important physical state information into the starting point of the time series prediction, improving the physical consistency of the extrapolation.

[0157] In the above implementation, a physical model based on the law of thermal conduction is used to deduce the theoretical baseline of temperature field evolution in three-dimensional space, ensuring the fundamental physical rationality and interpretability of the prediction results. Then, through a pre-trained artificial intelligence residual correction module, the systemic biases of the physical model caused by simplification assumptions and complex environmental factors are learned and compensated, significantly improving the accuracy of the prediction and its adaptability to the complexity of reality. Finally, the high-precision recent predictions after fusion are extrapolated using a time series model to generate a complete temperature field evolution sequence covering a specified future period.

[0158] The multi-stage, hybrid modeling strategy in this embodiment effectively overcomes the shortcomings of insufficient accuracy in pure physical models and poor generalization and potential violation of physical laws in pure data models, achieving a balance between prediction accuracy, physical rationality, and forward-looking perspective. The resulting multi-frame future temperature field data provides crucial decision-making information support for real-time, advanced intelligent temperature control based on accurate predictions during pavement paving, serving as a core technological guarantee for improving the uniformity of paving quality and the controllability of the construction process.

[0159] Reference Figure 5 As one implementation of step S105, the step of generating a hierarchical control instruction set based on the local temperature difference, temperature change slope, and minimum temperature threshold of the future temperature field prediction data includes:

[0160] Step S501: Obtain future temperature field prediction data, including temperature values ​​and time evolution sequences of spatial coordinate points;

[0161] The future temperature field prediction data is a collection of temperature field data arranged chronologically over a future period. Specifically, the temperature field data for each future moment includes the temperature values ​​of all spatial coordinate points within the paving area at that moment, constituting the spatial distribution information of the temperature field. Connecting these temporally arranged instantaneous temperature fields forms the "evolution sequence" of each spatial point in the time dimension, revealing the dynamic trend of temperature change over time. This spatiotemporal integrated data structure enables the decision-making system to move beyond passively responding to the current state and instead proactively intervene and plan based on the simulation of future evolution trajectories.

[0162] Step S502 involves performing a temperature field analysis based on future temperature field prediction data, and generating a hierarchical control instruction set based on the analysis results. This specifically includes the following three steps for generating control instructions:

[0163] (1) Divide the paving area into grid cells, calculate the difference between the highest and lowest temperatures in each grid cell, and generate a local temperature difference distribution map; when the local temperature difference exceeds the first set threshold, determine the heating zone location and power adjustment amount of the ironing plate according to the local temperature difference distribution map, and generate an ironing plate zone heating power adjustment command.

[0164] This method discretizes the macroscopic temperature field into microscopic management units to identify potential areas of localized quality defects. The compaction quality of asphalt mixtures is directly related to temperature uniformity; excessive local temperature differences can lead to uneven compaction, which in turn can cause a decrease in pavement smoothness or premature damage.

[0165] In this embodiment, the continuous physical space is managed by meshing. The mesh size (e.g., 0.5m × 0.5m) is typically matched to the physical size of the heating zones on the screed, allowing the analysis results to be directly mapped onto the physical structure of the actuator. For each mesh cell, the "difference between the highest and lowest temperatures" (i.e., the range) among all predicted temperature points within it is calculated. This difference is the core indicator for measuring the temperature dispersion, or uniformity, within that cell. This calculation is performed on all mesh cells, ultimately generating a "local temperature difference distribution map" covering the entire paving area. This map visually reveals which areas have highly consistent temperatures (small temperature differences) and which areas exhibit significant alternations of "cold spots" and "hot spots" (large temperature differences) during a future predicted period.

[0166] Furthermore, the first set threshold is an upper limit for temperature difference tolerance set according to the compaction process requirements. When the calculated local temperature difference of one or more grid cells exceeds this threshold, the system determines that intervention is needed in that area. At this time, based on the temperature difference distribution map, the system can accurately locate the specific grid cell with the excessive temperature difference, and then map it to the location of the heating zone of the screed covering that area. Each heating zone can independently control its heating power. To eliminate the temperature difference, the strategy is usually to supplement heat to the relatively low-temperature area within the zone, i.e., adjust the heating power.

[0167] Specifically, the formula for calculating the power adjustment is: ΔP = K_p × (ΔT_actual - ΔT_threshold). This formula embodies the proportional control concept: the larger the actual temperature difference T_actual exceeds the threshold ΔT_threshold, the larger the power increment ΔP that needs to be compensated. The coefficient K_p converts the temperature difference dimension into the power dimension. Finally, the system generates a "zone heating power adjustment instruction" for each zone requiring intervention, containing a "zone location code" and a "target power value or power increment". This step transforms predictive analysis into specific and differentiated control actions for the end effector of the screed, aiming to spatially re-homogenize the temperature field from the last stage of paving.

[0168] (2) Take a temperature profile along the paving direction, calculate the temperature change per unit length, and generate a temperature change slope sequence; when the temperature change slope exceeds the second set threshold, calculate the paving speed correction value according to the slope change trend and generate a paving speed adjustment command.

[0169] Among these methods, analyzing the temperature gradient along the machine's travel direction helps predict risks to the continuity of the construction process. Ideally, paving operations require a smooth transition in the temperature of the mixture along the paving direction. Drastic temperature fluctuations indicate unstable incoming material temperatures or a mismatch between the paver's travel speed and cooling rate. This directly affects the timing and effectiveness of the roller's follow-up compaction, resulting in inconsistent longitudinal compaction.

[0170] In this embodiment, "extracting a temperature profile along the paving direction" refers to extracting a continuous curve composed of temperature values ​​at various spatial points along the trajectory of the paver in the predicted temperature field. Differential or difference calculations are performed on this temperature curve to obtain the temperature change per unit length, i.e., the slope (dT / dL) of temperature change with spatial location. The magnitude and sign of this slope directly reflect the severity and trend of temperature change (e.g., a positive slope indicates that the temperature increases as paving progresses). Applying this calculation to a series of consecutive moments or locations within the predicted time period generates a temperature change slope sequence. This sequence acts as a continuous monitoring signal of the construction process's health; a stable, small slope value represents a stable temperature field, while a sudden increase in the slope value or the appearance of high-frequency oscillations warns of a potential risk of construction continuity interruption.

[0171] Furthermore, the second threshold is a safety limit set for the degree of temperature change along the paving direction. When the slope value in the temperature change slope sequence continuously or instantaneously exceeds this threshold, it indicates that the temperature is changing too rapidly in the longitudinal direction, posing a risk of forming a significant temperature gradient. In this case, the key parameter of paving speed needs to be adjusted to exert influence: if the slope is continuously positive and too large, it indicates that the temperature of the incoming material may be too high or cooling too slowly. Appropriately increasing the paving speed can reduce the exposure and heat dissipation time of the high-temperature material at the current location, allowing the high-temperature zone to pass through more quickly; conversely, if the slope is too large and negative, it indicates that the temperature of the incoming material is too low or cooling too quickly. Appropriately slowing down the speed can give it more heat dissipation time, preventing the low-temperature material from accumulating too quickly.

[0172] In this embodiment, the paving speed correction value is determined by a fuzzy control rule. This rule comprehensively considers two input variables: slope deviation (the difference between the current slope and the ideal slope) and deviation change rate (the speed at which the slope changes). It simulates the experience of a human operator and outputs a reasonable speed correction value. The final generated paving speed adjustment command aims to ensure the continuity and homogeneity of the construction process by actively suppressing longitudinal temperature fluctuations.

[0173] (3) Scan the future temperature field prediction data, identify the lowest temperature value within the specified time period, and when the lowest temperature value is lower than the insulation threshold, obtain the real-time location information of the material transport vehicle and calculate the insulation temperature compensation amount, and generate the material transport vehicle insulation temperature setting instruction.

[0174] Asphalt's bonding properties are extremely sensitive to temperature. If the temperature is too low, the mixture hardens and loses its plasticity, making it impossible to achieve the required density regardless of compaction, resulting in "cold joints" or loose areas – a serious quality hazard. Therefore, a global scan of the entire "future temperature field prediction data" to search for the "lowest temperature value" among all temperature data appearing within a "specified time period" (i.e., the future decision-making time window of interest, such as a few minutes below) is a crucial safety verification task.

[0175] Specifically, the minimum temperature value represents the lower limit of potential future temperature risks. Identifying it is not merely about finding a minimum value, but rather an assessment of the most unfavorable operating conditions. By comparing this predicted minimum value with a preset "insulation threshold" determined based on materials science and construction techniques, the system can determine in advance whether there is a systemic risk of globally excessively low temperatures.

[0176] Furthermore, when the identified predicted minimum temperature value is lower than the insulation threshold required by the process, it indicates that local heating or speed adjustment at the paving site alone cannot prevent low-temperature damage. It is necessary to address the issue at the material source, namely the material transport vehicle. The key to generating the insulation temperature setting instruction for the material transport vehicle is to determine a new, higher insulation target temperature. The specific calculation formula is: T_set = T_min + σ × (T_threshold - T_min) + λ × t_delay.

[0177] Understandably, this temperature calculation requires considering the location information of the material transport vehicles. After the command is issued, the vehicles need to travel for a period of time before reaching the paver, during which time the mixture temperature will continue to decrease (time decay term λ × time required to reach the paver t_delay). Therefore, the new set temperature (T_set) must not only compensate for the currently predicted temperature difference (insulation threshold T_threshold - minimum temperature value T_min), but also allow for a margin for temperature drop during transit (achieved through coefficients σ and time decay coefficient λ), ensuring that when this batch of material arrives at the paver, its temperature is still above the safe threshold. Here, σ is taken as 1.2-1.5, and λ is the time decay coefficient (0.5-0.8℃ / min). After the calculation is completed, this command is sent wirelessly to the insulation system of the relevant material transport vehicles to increase their heating or insulation power.

[0178] After the conditional judgments and calculations in the preceding steps, the system may generate zero, one, or more of the aforementioned instructions. These instructions represent the system's comprehensive response to current and future predicted risks. However, if these instructions are issued sporadically, it may cause conflicts or chaos at the execution end (for example, simultaneously and significantly adjusting speed and power in multiple partitions may trigger oscillations).

[0179] Therefore, the aggregation operation for generating a hierarchical control instruction set is not simply packaging; it also involves potential logical coordination and priority ranking. For example, when both material cart insulation instructions (source control) and screed heating instructions (end-point compensation) exist simultaneously, the system may annotate or adjust the instruction set to reflect the primary and secondary relationship. The final "hierarchical control instruction set" is a structured data object that clearly includes "what to do" (instruction type), "to whom to do it" (equipment identifier), "how much to do" (control quantity), and possibly "when to do it" (timing suggestions). This instruction set is the final output of the entire intelligent control method. Through subsequent execution modules, it will precisely and collaboratively drive equipment such as pavers and material carts, thereby truly transforming predictive intelligence into precise control of the physical world, completing a closed-loop intelligent control chain from perception and prediction to decision-making and execution.

[0180] In the above implementation, three types of instructions are precisely generated based on different types of risks: spatially precise heating instructions for localized temperature differences, process speed adjustment instructions for longitudinal fluctuations, and source material vehicle insulation instructions for overall low temperatures. Ultimately, these instructions are aggregated to form a unified control instruction set, ensuring that multiple actuators can operate in an orderly and coordinated manner. This technical solution achieves precise, automated, and systematic control based on forward-looking prediction, and is a core decision-making technology for ensuring high uniformity, high density, and excellent long-term performance in asphalt pavement paving.

[0181] Reference Figure 6 As one implementation of step S107, the steps of collecting actual temperature data after the execution of the graded control instruction set, calculating error indicators, and dynamically updating the parameters of the prediction model include:

[0182] Step S601: After the hierarchical control instruction set is executed, the actual infrared temperature matrix and internal temperature gradient data of the paving area within a specified time window are collected.

[0183] After the hierarchical control command set is executed, it means that the system has moved from the prediction and decision-making stage to the execution stage. The field equipment (such as screeds, pavers, and material transport vehicles) has been adjusted according to the commands, and the temperature field of the paving layer has been actively intervened. The actual infrared temperature matrix and internal temperature gradient data collected at this time are the real results of this intervention state and the combined effect of external natural cooling.

[0184] The specified time window starts at the command execution time plus the equipment response delay, and ends at the initial setting time of the paving material. Understandably, the start point takes into account the equipment response delay to avoid collecting invalid data before the equipment action takes effect; the end point is usually set before the initial setting time of the paving material because the influence of temperature on compaction quality decreases after initial setting. The data within this window records the entire process from the control action taking effect to the point where the impact becomes basically stable. It is the golden period for verifying the performance of the predictive model under the new condition of "active control" and capturing its deviation patterns.

[0185] Step S602: Spatially align the actual infrared temperature matrix with the future temperature field prediction data corresponding to the timestamp to generate a surface temperature deviation matrix.

[0186] Since the predicted data is gridded data generated by the computer model in digital space, while the actual images captured by the infrared camera have their own pixel coordinate system, viewpoint, and distortion, the two are not naturally one-to-one in space. Spatial alignment operations (including temporal interpolation and affine transformation) are precisely designed to solve this problem.

[0187] Specifically, firstly, time-dimensional interpolation is performed, interpolating the predicted data at discrete time points onto a continuous time series with the exact same data acquisition frequency, ensuring that each actual data point can find a predicted value at the same time for comparison. Next, an affine transformation is used to map the grid coordinates of the predicted data to the image coordinate system of the actual infrared camera. After achieving high-precision spatiotemporal alignment, at each identical time point and physical location, the actual measured temperature value is subtracted from the model's predicted temperature value; the difference constitutes the surface temperature deviation matrix. Each element of this matrix precisely quantifies the model's prediction deviation at that time and location, eliminating spurious errors caused by coordinate misalignment and providing the most direct and cleanest surface temperature error signal for subsequent model correction.

[0188] Step S603: Calculate the gradient direction residual for each spatial point based on the difference between the internal temperature gradient data and the predicted temperature field data in the depth direction.

[0189] Temperature prediction involves not only surface values ​​but also the distribution and trend of internal temperature, characterized by the rate of temperature change along the depth direction (i.e., the temperature gradient). The temperature gradient data obtained from actual measurements by millimeter-wave radar reflects the actual rate of heat transfer from the inside to the outside of the real mixture. The "rate of temperature change along the depth direction" in the predicted data is a theoretical gradient value derived by the model based on its built-in physical equations and parameters (such as the thermal conductivity k).

[0190] By comparing these two sets of data and calculating their vector difference or the magnitude of the difference, the gradient direction residual is obtained. If the surface temperature prediction is accurate but the gradient residual is large, it indicates that although the model correctly predicted the surface result, its simulation of the internal heat conduction process is inaccurate (e.g., underestimating the material's insulation performance or overestimating the heat dissipation rate). This mechanistic error will inevitably lead to a rapid loss of accuracy in surface prediction when operating conditions change. Therefore, the gradient direction residual is a key indicator for evaluating the physical consistency and generalization ability of a model.

[0191] Step S604: Fuse the surface temperature deviation matrix and gradient direction residuals to construct a multi-dimensional error tensor, with dimensions including horizontal spatial coordinates and depth direction;

[0192] The temperature deviation matrix quantifies the model's direct error on a two-dimensional surface in space, specifically on scalar temperature values. The gradient direction residual quantifies the model's physical mechanism error in the third dimension (depth direction) of space, specifically on the vector rate of change. These two reflect the model's performance across different dimensions and physical quantities, and may not always be consistent.

[0193] In this embodiment of the application, the calculation formula for constructing the multidimensional error tensor is as follows:

[0194] ;

[0195] In the above formula, This represents the absolute error point corresponding to the point in the surface temperature deviation matrix. Let α be the absolute value of the gradient direction residual, and let α and β be the weighting coefficients of the temperature deviation term and the gradient residual term, respectively. Let α + β = 1, and let (x, y) be the plane position, t be the time, and z be the depth direction.

[0196] Specifically, these two types of errors are combined into a unified multidimensional error tensor according to their importance (adjusted by weighting coefficients α and β). This tensor contains a composite error value at each location in the spatial (x, y) and temporal (t) dimensions, consisting of the absolute deviation of surface temperature and the difference in depth gradient.

[0197] Step S605: Dynamically calculate the updated weight coefficients of the prediction model parameters based on the deviation magnitudes at each spatial location in the multidimensional error tensor.

[0198] The constructed multi-dimensional error tensor provides a panoramic view of errors across the entire region and time period. However, not all errors are equally important. Small, uniformly distributed errors may be random noise or measurement fluctuations, while locally concentrated, large-amplitude errors often reveal cognitive blind spots or systematic biases in the model under specific operating conditions (such as extreme wind speeds or material inhomogeneity).

[0199] In this embodiment, a piecewise function is pre-defined: for regions with small deviation amplitudes (e.g., |ΔT|≤5℃), a low or standard weight coefficient (e.g., W=0.8) is given to perform mild and conservative updates, avoiding excessive reactions and oscillations in the model to noise; for regions with medium deviation amplitudes (5℃<|ΔT|≤15℃), a higher weight (e.g., W=1.2) is given to encourage the model to strengthen its learning; for regions with large deviation amplitudes (|ΔT|>15℃), the highest weight (e.g., W=1.5) is given to force the model to focus its "attention" on quickly correcting these seriously erroneous predictions.

[0200] This "dynamic adjustment" mechanism makes the model training process no longer a "one-size-fits-all" uniform learning process, but rather targeted and adaptive. It ensures that valuable online learning opportunities and computing resources are prioritized for correcting the most prominent problems, thereby significantly accelerating the model's performance improvement process in complex real-world environments and enhancing its predictive robustness to extreme or abnormal conditions.

[0201] Step S606: Using the error backpropagation algorithm, the weighted multi-dimensional error tensor is used as the input of the loss function to iteratively update the neural network parameters of the residual correction module in the prediction model.

[0202] The final composite error (the loss function value calculated from the weighted multi-dimensional error tensor) is propagated back along the network's computational graph from the output layer to each layer, and the "contribution" (gradient) of each adjustable parameter (i.e., "network parameter") to the total error is calculated. Then, following the direction of gradient descent, these parameters are iteratively updated in small increments at a certain learning rate.

[0203] In this embodiment, the object of iterative updating specifically refers to the network parameters of the residual correction module in the prediction model. This is because in a hybrid prediction architecture, the parameters of the physical equation module are usually based on material properties and are relatively fixed; while the neural network of the residual correction module is designed to learn and compensate for unknown errors in the physical model, and its parameters need to and should also be flexible. Through this algorithm, the network parameters are continuously adjusted, with the goal of minimizing the weighted multidimensional error tensor generated by the final prediction result of the entire model after the residual correction amount output by the network is superimposed.

[0204] Furthermore, the iterative update step should employ a moving average mechanism to ensure that each update partially retains information from historical parameters, preventing drastic fluctuations in model performance or the forgetting of learned effective knowledge due to the randomness of a single update. Through this step, the model can learn from each "prediction-execution-validation" cycle, continuously and steadily optimizing its internal representation, thereby achieving online evolution and lifelong learning of predictive capabilities.

[0205] In the above implementation, after the control command is executed, actual effect data is actively collected and compared with previous predictions using refined spatiotemporal alignment and multi-dimensional (surface temperature and internal gradient) error analysis to construct an error map that comprehensively reflects model defects. Then, a differentiated weighted learning strategy is implemented based on the severity of the error, guiding the model to focus on correcting significant prediction biases. Finally, the parameters of the residual correction neural network responsible for compensating for unknown factors are stably and progressively optimized using an error backpropagation algorithm combined with a moving average mechanism. This technical solution effectively solves the challenges faced by artificial intelligence models in industrial scenarios, such as variable working conditions and the difficulty of collecting complete training data offline. Through online, incremental learning, it continuously improves its prediction accuracy, robustness, and generalization ability, thereby ensuring that the entire intelligent control system can maintain high reliability over a long period.

[0206] Reference Figure 7 As a further implementation of the intelligent control method, the intelligent control method also includes:

[0207] Step S701: Collect asphalt mixture paving layer thickness distribution data in real time using a laser thickness gauge and generate a thickness offset matrix;

[0208] Traditional temperature prediction models typically simulate a uniform design thickness, which is significantly different from the actual spatial distribution of thickness at the construction site due to variations in paver operation, material flowability, and roadbed smoothness. This thickness deviation is not a uniform, minute fluctuation, but may exhibit distinct regional patterns of "overthickness" or "underthickness."

[0209] To this end, this step deploys a laser thickness gauge to rapidly and continuously scan the paved asphalt mixture surface in a non-contact manner, obtaining dense point cloud distance data. By spatially registering and calculating the difference between these real-time measurements and preset design thickness benchmark values, the system generates a thickness offset matrix covering the entire paved area. This matrix, in the form of a regular grid (e.g., 0.5m × 0.5m), accurately records the millimeter-level deviation (Δh) between the actual thickness and the design value of each local unit.

[0210] Step S702: Calculate the heat capacity correction coefficient of the local area based on the thickness offset matrix;

[0211] As a composite material, the heat capacity (i.e., the amount of heat absorbed to raise a unit temperature) of asphalt mixture depends on its mass and specific heat capacity. When the material composition (which determines specific heat capacity) and density are relatively stable, an increase in local thickness is directly equivalent to an increase in the mass of the mixture per unit area in that region, resulting in a corresponding increase in its heat capacity. This strengthens the region's ability to store heat energy, and under the same environmental heat dissipation conditions, the rate of temperature decrease slows down, meaning it has greater thermal inertia.

[0212] In this embodiment, based on the generated thickness offset matrix, a heat capacity correction coefficient is calculated for each mesh element using a physical model. The specific calculation formula is as follows:

[0213] ;

[0214] Among them, C adj The adjusted heat capacity is given by ρ, where ρ is the density of the asphalt mixture, and c is the density of the asphalt mixture. p The specific heat capacity at constant pressure of asphalt mixture, kJ h Δh is the thickness-heat capacity influence coefficient, and Δh is the thickness deviation value.

[0215] Specifically, basic heat capacity It is a theoretical value calculated based on nominal thickness and ideal density. The formula introduces the term (1+k) h The coefficient k corrects for this. When the paving thickness is greater than the design value (Δh>0), it means that there is more material volume at that point, thus increasing the total heat storage capacity (heat capacity); conversely, it decreases, and the coefficient k corrects for this. h The sensitivity to this effect was controlled. The heat capacity correction factor is a dimensionless number representing the ratio between the actual heat capacity at that point and the heat capacity calculated based on the nominal design thickness. For example, for an extra-thick 4 mm area, the correction factor may be greater than 1.1, meaning that its actual heat capacity is more than 10% higher than expected.

[0216] Step S703: Adjust the attenuation gradient of the temperature change rate in the depth direction in the three-dimensional temperature field tensor according to the heat capacity correction coefficient.

[0217] In the three-dimensional temperature field model, the rate of temperature change along the depth direction is the core vector describing the rate of heat conduction from the high-temperature surface to the low-temperature bottom layer, and its value usually exhibits a certain "attenuation" trend along the depth direction. However, this attenuation law is significantly modulated by the local heat capacity: in regions with large heat capacity (i.e., ultra-thick regions), the thermal inertia of the material is large, which can better maintain the uniformity of temperature distribution, making the temperature gradient (∇T_z) decrease more slowly with depth, or in other words, at the same depth, its temperature gradient value is relatively larger.

[0218] To accurately represent this physical effect in the model, the calculated "heat capacity correction factor" (C) is used. adj The algorithm dynamically adjusts the standard "attenuation gradient" preset in the model or calculated in the previous cycle. The adjustment algorithm typically introduces an exponential response function, which uses the deviation of the heat capacity correction coefficient from the nominal value (|C0|0). adj - C nom |) is taken as input, and the output is a scaling factor for the current depth gradient (∇T_z). The specific formula for calculating the corrected decay gradient is:

[0219] ;

[0220] Where μ is the attenuation coefficient, ranging from 0.12 to 0.15, which determines the strength of the effect of heat capacity deviation on temperature gradient attenuation. When C adj Greater than C nom When the region is extremely thick, the scaling factor approaches 1 (or even slightly greater than 1), which means that the gradient decay is slowed down or reversed, making the corrected gradient ∇T_z' better reflect the gentle temperature distribution within the region; conversely, it accelerates the decay. This step ensures that the evolution of the three-dimensional temperature field tensor in the depth dimension can match the real-time, spatially differentiated thickness distribution.

[0221] Step S704: Obtain the real-time operating frequency of the paver's vibration compaction mechanism, and calculate the change in material porosity based on the real-time operating frequency;

[0222] The vibratory compaction mechanism behind the paver uses high-frequency vibration to rearrange the aggregates in the asphalt mixture, reducing voids and thus increasing density. Its real-time operating frequency (f) is a direct representation of the compaction energy input intensity; the higher the frequency, the stronger the impact-vibration effect on the mixture per unit time, making it easier for the aggregates to achieve dense packing and resulting in a decrease in internal air porosity. However, porosity is not a fixed value; it dynamically evolves with the number of compaction passes and changes in real-time vibration parameters.

[0223] In this embodiment, vibration frequency data is collected in real time by installing sensors (such as Hall sensors) on the vibration shaft. Subsequently, based on a pre-established empirical "frequency-porosity" model or mapping relationship developed through experiments, the real-time frequency value is converted into an estimated value of the "material porosity change" currently being experienced by the paving layer. This model reflects the causal relationship between compaction process parameters and the final physical state of the material. Through this step, the system can sense the dynamic changes in the material composition caused by the construction process itself, linking the invisible microstructural evolution (porosity reduction) with the visible equipment operating parameters (vibration frequency), laying the foundation for the next step of correcting the macroscopic thermophysical properties of the material.

[0224] Step S705: Correct the equivalent thermal conductivity in the physical heat transfer equation based on the change in material porosity.

[0225] The equivalent thermal conductivity (k_eff) of asphalt mixtures is a comprehensive reflection of the combined effect of its solid skeleton (aggregate and asphalt) and the air in its internal pores. Since air has extremely low thermal conductivity (approximately 0.026 W / m·K), far lower than that of solid materials (approximately 1.2-1.5 W / m·K), the presence of pores significantly hinders heat conduction. However, as vibration compaction progresses, porosity decreases, the solid contact area increases, and the heat transfer path becomes more unobstructed, leading to an increase in the overall "equivalent thermal conductivity." Traditional models typically use a fixed k value, which completely ignores the dynamic fact that the material's thermal conductivity continuously increases during the compaction process.

[0226] In this embodiment, based on the calculated real-time porosity change, the "equivalent thermal conductivity" under the current time and operating conditions is recalculated using a multiphase composite material thermal conductivity theoretical model (such as the Maxwell-Eucken model or its simplified form). The corrected equivalent thermal conductivity k_eff value more accurately reflects the improved thermal conductivity of the material due to compaction. The specific calculation formula is as follows:

[0227] ;

[0228] In the above formula, k solid The original equivalent thermal conductivity in the physical heat transfer equation is... denoted as the material porosity, and m as the empirical shape factor, ranging from 2.3 to 2.8.

[0229] It should be noted that this dynamic correction is crucial, as it means that the heat transfer rate simulated by the system is no longer based on the assumption of a loose state, but closely follows the compaction progress, and can more realistically predict that as compaction proceeds, the overall rate at which heat is lost from the interior to the surface and the environment will change, thus directly affecting the shape of the cooling curve and the prediction of the final temperature.

[0230] Step S706: Input the corrected attenuation gradient and equivalent thermal conductivity into the prediction model for real-time parameter calibration.

[0231] These corrected parameters are then input into the prediction model in either a "hard constraint" or "soft guide" manner. Specifically, the corrected decay gradient may be used to reinitialize or constrain the representation of the depth dimension in the model's internal state; the corrected equivalent thermal conductivity directly replaces the corresponding coefficient in the physics calculation module.

[0232] Simultaneously, the system uses the actual infrared temperature data collected within the most recent short time window (e.g., the past 30 seconds) as the true value. Employing lightweight optimization algorithms (such as the sliding window least squares method), and starting with the corrected parameters, it fine-tunes the adjustable parts of the model (especially the parameters of the data-driven module) to minimize the error between the model's short-term prediction output and the actual measured value. This step is called real-time parameter calibration, ensuring that the prediction model is no longer a rigid, offline program, but an intelligent agent capable of sensing current thickness variations, understanding the current compaction state, and adjusting its internal cognition accordingly. This allows it to maintain high-precision prediction capabilities and reliable control even when facing unavoidable disturbances during construction.

[0233] In the above embodiments, high-precision laser thickness measurement is used to capture the thickness distribution in real time and establish a physical conversion model from thickness to heat capacity, thereby adjusting the gradient attenuation law within the three-dimensional temperature field. Simultaneously, by monitoring the vibration compaction frequency, the dynamic reduction process of material porosity is inverted, and the equivalent thermal conductivity in the heat transfer equation is updated accordingly. Finally, these real-time corrected physical parameters are used as forced inputs to the prediction model, driving it to perform rapid online calibration. This technical solution enables the prediction model to overcome the limitations of traditional static models, closely following the actual progress of paving operations, dynamically adapting to the spatial non-uniformity and time-varying nature of material properties and geometric states introduced by the construction process itself. By compensating for prediction distortions caused by local over-thickness, under-thickness, and compaction variations, the prediction error in relevant areas is compressed to an extremely low level, enhancing the accuracy and reliability of the intelligent control system in complex, dynamic, and non-ideal real construction environments.

[0234] This application also discloses an intelligent temperature field control system for asphalt mixture paving.

[0235] A smart temperature field control system for asphalt pavement paving specifically includes:

[0236] The multi-source data acquisition module is used to collect infrared temperature matrix, environmental parameters, equipment operating parameters, and internal temperature gradient data collected by millimeter-wave radar in the asphalt mixture paving area in real time, and generate a timestamp-aligned multi-source dataset.

[0237] The occlusion compensation and correction module is used to compensate and correct the occlusion area in the infrared temperature matrix based on the internal temperature gradient data, and to construct the correction temperature matrix.

[0238] The three-dimensional temperature field reconstruction module is used to fuse the corrected temperature matrix with the internal temperature gradient data to construct a three-dimensional temperature field tensor that includes the surface temperature and the rate of temperature change in the depth direction.

[0239] The physical information fusion prediction module is used to input the three-dimensional temperature field tensor into the pre-built prediction model, combine environmental parameters and equipment operating parameters, and output future temperature field prediction data through the residual correction module constrained by the physical heat transfer equation.

[0240] The multi-level decision-making and instruction generation module is used to generate a hierarchical control instruction set based on the local area temperature difference, temperature change slope and minimum temperature threshold of the future temperature field prediction data. The hierarchical control instruction set includes instructions for adjusting the heating power of the screed zone, instructions for adjusting the paving speed and instructions for setting the insulation temperature of the material transport vehicle.

[0241] The instruction distribution and execution module is used to distribute the hierarchical control instruction set to the paver's zone heating system, paving speed controller, and material transport vehicle insulation system to perform corresponding operations;

[0242] The model optimization module is used to collect actual temperature data after the execution of the hierarchical control instruction set, calculate error indicators, and dynamically update the parameters of the prediction model.

[0243] The intelligent temperature field control system for asphalt mixture paving according to an embodiment of this application can implement any of the above methods, and the specific working process of each module in the system can refer to the corresponding process in the above method embodiments.

[0244] In the several embodiments provided in this application, it should be understood that the provided methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for example, the division of a certain module is merely a logical functional division, and in actual implementation there may be other division methods, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0245] This application also discloses a computer device.

[0246] Computer equipment, including memory, processor, and computer program stored in memory and executable on the processor, wherein the processor executes the computer program to implement a method for intelligent control of the temperature field of asphalt mixture paving as described above.

[0247] This application also discloses a computer-readable storage medium.

[0248] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described above in any of the methods for intelligent control of the temperature field of asphalt mixture paving.

[0249] The computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device; the program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0250] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A method for intelligent control of the temperature field of asphalt mixture paving, characterized in that, Intelligent control methods include: Real-time acquisition of infrared temperature matrix, environmental parameters, equipment operating parameters, and internal temperature gradient data collected by millimeter-wave radar in the asphalt mixture paving area of ​​the road surface, generating a timestamp-aligned multi-source dataset; Based on the internal temperature gradient data, the occlusion area in the infrared temperature matrix is ​​compensated and corrected to construct a corrected temperature matrix. The corrected temperature matrix is ​​fused with the internal temperature gradient data to construct a three-dimensional temperature field tensor that includes the surface temperature and the rate of temperature change in the depth direction. The three-dimensional temperature field tensor is input into a pre-built prediction model, and combined with environmental parameters and equipment operating parameters, the residual correction module constrained by the physical heat transfer equation outputs future temperature field prediction data. Based on the local temperature difference, temperature change slope, and minimum temperature threshold of the predicted future temperature field data, a hierarchical control instruction set is generated; wherein, the hierarchical control instruction set includes instructions for adjusting the heating power of the screed zone, instructions for adjusting the paving speed, and instructions for setting the insulation temperature of the material transport vehicle. The graded control instruction set is distributed to the paver's zone heating system, paving speed controller, and material transport vehicle insulation system to execute corresponding operations; Collect actual temperature data after the execution of the hierarchical control instruction set, calculate error index, and dynamically update the parameters of the prediction model; The steps for generating a hierarchical control instruction set based on the local temperature difference, temperature change slope, and minimum temperature threshold of the predicted future temperature field data include: Acquire future temperature field prediction data, including temperature values ​​at spatial coordinate points and their temporal evolution sequences; Based on the predicted future temperature field data, a temperature field analysis operation is performed, and a hierarchical control instruction set is generated based on the analysis results, specifically including: The paving area is divided into grid cells, and the difference between the highest and lowest temperatures in each grid cell is calculated to generate a local temperature difference distribution map. When the local temperature difference exceeds the first set threshold, the location and power adjustment amount of the ironing plate heating zone are determined based on the local temperature difference distribution map, and an ironing plate zone heating power adjustment command is generated. A temperature profile is captured along the paving direction, the temperature change per unit length is calculated, and a temperature change slope sequence is generated. When the temperature change slope exceeds the second set threshold, a paving speed correction value is calculated based on the slope change trend, and a paving speed adjustment command is generated. Scan the future temperature field prediction data, identify the lowest temperature value within a specified time period, and when the lowest temperature value is lower than the insulation threshold, obtain the real-time location information of the material transport vehicle and calculate the insulation temperature compensation amount, and generate the material transport vehicle insulation temperature setting instruction.

2. The intelligent control method for the temperature field of asphalt mixture paving according to claim 1, characterized in that, Based on the internal temperature gradient data, the steps for compensating and correcting the occlusion areas in the infrared temperature matrix and constructing the corrected temperature matrix include: Acquire reflection intensity data collected by millimeter-wave radar; The coordinates of the obstructed area in the infrared temperature matrix are identified based on the reflection intensity data; wherein, the obstructed area includes coordinate points where the reflection intensity is lower than a preset intensity threshold; Extract the internal temperature gradient data corresponding to the coordinates of the shading area, and extract the temperature values ​​of the unshading area within a preset range adjacent to the coordinates of the shading area; Based on the temperature value of the unshaded area and the internal temperature gradient data, the compensation temperature value of the shaded area is calculated. The original temperature value of the corresponding occluded area in the infrared temperature matrix is ​​replaced by the compensated temperature value to generate a corrected temperature matrix.

3. The intelligent control method for the temperature field of asphalt mixture paving according to claim 2, characterized in that, The steps of fusing the corrected temperature matrix with the internal temperature gradient data to construct a three-dimensional temperature field tensor containing the surface temperature and the rate of temperature change in the depth direction include: Based on the spatial resolution of the correction temperature matrix, a three-dimensional mesh coordinate system matching the paving area is established; wherein, the horizontal resolution of the three-dimensional mesh coordinate system is consistent with the pixel spacing of the correction temperature matrix, and layer depth nodes are set in the vertical direction. Assign the temperature values ​​of each coordinate point in the corrected temperature matrix to the corresponding surface nodes of the three-dimensional mesh coordinate system; Based on the internal temperature gradient data, the first-order temperature change rate of each surface node in the depth direction is calculated. Based on the temperature values ​​of the surface nodes and the first-order temperature change rate in the depth direction, the temperature values ​​of each depth node in the vertical direction in the three-dimensional mesh coordinate system are generated by the heat conduction integral algorithm. By integrating the temperature values ​​of the surface nodes and the nodes at each depth, a three-dimensional temperature field tensor is constructed.

4. The intelligent control method for the temperature field of asphalt mixture paving according to claim 1, characterized in that, The steps of inputting the three-dimensional temperature field tensor into a pre-built prediction model, combining environmental parameters and equipment operating parameters, and outputting future temperature field prediction data through a residual correction module constrained by the physical heat transfer equation include: The three-dimensional temperature field tensor is input into the physical equation calculation unit, and the physical heat transfer equation is solved by combining real-time environmental parameters and equipment operating parameters, and the basic temperature evolution sequence is output. The basic temperature evolution sequence is input into the pre-trained artificial neural network in the residual correction module to calculate the prediction bias of the physical equation calculation unit. The base temperature evolution sequence is superimposed with the prediction deviation to generate a corrected temperature sequence; The corrected temperature sequence is input into the time series prediction unit, which outputs future temperature field prediction data.

5. The intelligent control method for the temperature field of asphalt mixture paving according to claim 1, characterized in that, The steps of collecting actual temperature data after the execution of the hierarchical control instruction set, calculating error indicators, and dynamically updating the parameters of the prediction model include: After the hierarchical control instruction set is executed, the actual infrared temperature matrix and internal temperature gradient data of the paving area are collected within a specified time window. The actual infrared temperature matrix is ​​spatially aligned with the future temperature field prediction data at the corresponding timestamp to generate a surface temperature deviation matrix. Based on the difference in the rate of temperature change in the depth direction between the internal temperature gradient data and the future temperature field prediction data, the gradient direction residual of each spatial point is calculated. By fusing the surface temperature deviation matrix and the gradient direction residual, a multi-dimensional error tensor is constructed, with dimensions including horizontal spatial coordinates and depth direction. Based on the deviation magnitude of each spatial location in the multidimensional error tensor, the update weight coefficients of the prediction model parameters are dynamically calculated. Using the backpropagation algorithm, the weighted multidimensional error tensor is used as the input to the loss function to iteratively update the neural network parameters of the residual correction module in the prediction model.

6. A method for intelligent control of the temperature field of asphalt mixture paving according to any one of claims 1 to 5, characterized in that, The intelligent control method also includes: Real-time thickness distribution data of asphalt mixture paving layers are collected using a laser thickness gauge to generate a thickness offset matrix. Calculate the heat capacity correction factor for the local region based on the thickness offset matrix; The attenuation gradient of the rate of temperature change in the depth direction in the three-dimensional temperature field tensor is adjusted according to the heat capacity correction coefficient. Obtain the real-time operating frequency of the paver's vibration compaction mechanism, and calculate the change in material porosity based on the real-time operating frequency; The equivalent thermal conductivity in the physical heat transfer equation is corrected based on the change in the porosity of the material. The corrected attenuation gradient and equivalent thermal conductivity are input into the prediction model for real-time parameter calibration.

7. A smart temperature field control system for asphalt pavement paving, characterized in that, The intelligent control system includes: The multi-source data acquisition module is used to collect infrared temperature matrix, environmental parameters, equipment operating parameters, and internal temperature gradient data collected by millimeter-wave radar in the asphalt mixture paving area in real time, and generate a timestamp-aligned multi-source dataset. The occlusion compensation and correction module is used to compensate and correct the occlusion area in the infrared temperature matrix based on the internal temperature gradient data, and to construct a correction temperature matrix. The three-dimensional temperature field reconstruction module is used to fuse the corrected temperature matrix with the internal temperature gradient data to construct a three-dimensional temperature field tensor that includes the surface temperature and the rate of temperature change in the depth direction. The physical information fusion prediction module is used to input the three-dimensional temperature field tensor into a pre-built prediction model, combine environmental parameters and equipment operating parameters, and output future temperature field prediction data through the residual correction module constrained by the physical heat transfer equation. The multi-level decision-making and instruction generation module is used to generate a hierarchical control instruction set based on the local area temperature difference, temperature change slope, and minimum temperature threshold of the future temperature field prediction data; wherein, the hierarchical control instruction set includes instructions for adjusting the heating power of the screed zone, instructions for adjusting the paving speed, and instructions for setting the insulation temperature of the material transport vehicle. The instruction distribution and execution module is used to distribute the hierarchical control instruction set to the paver zone heating system, paving speed controller and material transport vehicle insulation system to perform corresponding operations; The model optimization module is used to collect the actual temperature data after the execution of the hierarchical control instruction set, calculate the error index, and dynamically update the parameters of the prediction model. The multi-level decision-making and instruction generation module is configured as follows: Acquire future temperature field prediction data, including temperature values ​​at spatial coordinate points and their temporal evolution sequences; Based on the predicted future temperature field data, a temperature field analysis operation is performed, and a hierarchical control instruction set is generated based on the analysis results, specifically including: The paving area is divided into grid cells, and the difference between the highest and lowest temperatures in each grid cell is calculated to generate a local temperature difference distribution map. When the local temperature difference exceeds the first set threshold, the location and power adjustment amount of the ironing plate heating zone are determined based on the local temperature difference distribution map, and an ironing plate zone heating power adjustment command is generated. A temperature profile is captured along the paving direction, the temperature change per unit length is calculated, and a temperature change slope sequence is generated. When the temperature change slope exceeds the second set threshold, a paving speed correction value is calculated based on the slope change trend, and a paving speed adjustment command is generated. Scan the future temperature field prediction data, identify the lowest temperature value within a specified time period, and when the lowest temperature value is lower than the insulation threshold, obtain the real-time location information of the material transport vehicle and calculate the insulation temperature compensation amount, and generate the material transport vehicle insulation temperature setting instruction.

8. A computer device, characterized in that: The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method as claimed in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer program is stored that can be loaded by a processor and executed as described in any one of claims 1 to 6.

Images