Wall double-unknown thermal parameter inversion method and system based on physical neural network

Abstract

The application provides a wall double-unknown thermal parameter inversion method and system based on a physical neural network, and belongs to the technical field of building energy saving, and comprises the following steps: obtaining relevant environmental parameters and infrared image data of a wall to be measured at an inversion moment; inputting the infrared image data into a convolutional neural network model to obtain temperature field information of an outer wall surface of the wall to be measured; setting an initial curve of an inner wall surface temperature based on the smooth characteristics and moderate lagging effect of indoor air temperature, and giving an initial value of a heat conductivity coefficient; jointly constructing a physical neural network model according to the Fourier heat conduction law and the boundary conditions of inner and outer surface convective heat exchange; inputting the relevant environmental parameters and the temperature field information of the outer wall surface into the physical neural network model to perform forward solving to obtain the heat transfer coefficient of the wall to be measured and the inner wall surface temperature. The application can obtain stable and reliable wall heat conductivity coefficients and inner wall surface temperatures.

Description

Technical Field

[0001] This invention belongs to the field of building energy conservation technology, and in particular relates to a method and system for inverting two unknown thermal parameters of walls based on physical neural networks. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Given the massive existing building stock and the rapid pace of new construction each year, the thermal performance of building envelopes has become a key factor influencing building energy consumption and indoor environmental quality. It directly determines the magnitude of heating and cooling loads and the stability of the indoor thermal and humidity environment, serving as an indispensable fundamental parameter in building energy-saving design, operational optimization, and retrofit evaluation. Therefore, conducting thermal performance testing of building envelopes is essential. On one hand, thermal performance testing can accurately identify the degradation of building insulation performance and deviations during construction. Based on the test results, the potential for energy-saving retrofits of existing buildings can be quantified, enabling the development of scientifically sound energy-saving retrofit plans, thereby improving building energy efficiency. On the other hand, thermal performance testing data can serve as an important benchmark for the green evaluation of existing buildings, providing quantitative support for green building certification.

[0004] Currently, thermal testing of building envelopes mainly encompasses various methods such as the heat box method, heat flow meter method, and planar heat source method. While these methods can obtain relevant information such as thermal parameters of building envelopes, they generally suffer from a series of problems: First, they require stringent environmental conditions, often necessitating long-term steady-state environments or specific operating conditions, resulting in long testing cycles and repeatability easily affected by external disturbances such as solar radiation, wind speed, and humidity. Second, the deployment of sensors or sampling through openings during the testing process is complex and can cause some damage to the building. Third, most methods involve point or small-area measurements, making it difficult to comprehensively and accurately reflect large-area heterogeneous structures and thermal bridging effects. Fourth, they rely heavily on prior information such as material layers and thicknesses; when on-site parameters are uncertain, the identification accuracy decreases significantly. In summary, these problems make it difficult for existing testing methods to meet the needs of efficient assessment of massive numbers of existing buildings. Summary of the Invention

[0005] To overcome the shortcomings of the existing technologies, this invention proposes a method and system for inverting dual unknown thermal parameters of walls based on physical neural networks. Under the conditions of unsteady-state operation, limited prior information, and stringent non-destructive testing, it achieves robust and accurate identification of the thermal conductivity of walls, while simultaneously acquiring the hourly change results of the inner wall surface temperature. This provides solid and reliable technical support for the energy efficiency diagnosis and energy-saving renovation assessment of existing buildings.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0007] In a first aspect, this invention discloses a method for inverting two unknown thermal parameters of a wall based on a physical neural network, comprising:

[0008] Acquire relevant environmental parameters and infrared image data of the wall under test at the time of inversion;

[0009] The infrared image data is input into a convolutional neural network model to obtain the temperature field information of the outer wall surface of the wall to be tested;

[0010] An initial curve for the inner wall temperature is set based on the smooth characteristics and moderate hysteresis effect of indoor air temperature, and an initial value for the thermal conductivity is given. A physical neural network model is constructed based on Fourier's law of thermal conduction and the convection heat transfer boundary conditions of the inner and outer surfaces. The physical neural network model is trained using a phased two-parameter inversion strategy within a unified evaluation function framework.

[0011] The relevant environmental parameters and the temperature field information of the outer wall are input into the physical neural network model, and the heat transfer coefficient and inner wall temperature of the wall under test are obtained by forward solving.

[0012] Secondly, this invention discloses a wall dual-unknown thermal parameter inversion system based on a physical neural network, comprising:

[0013] The data acquisition module is used to acquire relevant environmental parameters and infrared image data of the wall under test at the time of inversion.

[0014] The outer wall temperature extraction module is used to input the infrared image data into a convolutional neural network model to obtain the temperature field information of the outer wall surface of the wall to be tested;

[0015] The model building module is used to set the initial curve of the inner wall surface temperature based on the smooth characteristics and moderate hysteresis effect of indoor air temperature, and to give the initial value of thermal conductivity; a physical neural network model is jointly constructed according to Fourier's law of thermal conduction and the convection heat transfer boundary conditions of the inner and outer surfaces; wherein, the physical neural network model is trained in a phased two-parameter inversion strategy within a unified evaluation function framework.

[0016] The inversion module is used to input the relevant environmental parameters and the temperature field information of the outer wall into the physical neural network model, and perform forward solving to obtain the heat transfer coefficient and inner wall temperature of the wall under test.

[0017] Thirdly, the present invention discloses an electronic device, including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when run by the processor, complete the steps of the above-mentioned method for inverting dual unknown thermal parameters of a wall based on a physical neural network.

[0018] Fourthly, the present invention discloses a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps of the above-mentioned method for inverting dual unknown thermal parameters of a wall based on a physical neural network.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0020] This invention provides a non-invasive monitoring method that uses only external infrared images and environmental data. It eliminates the need to drill holes in buildings to install measuring points and avoids the need to wait for a long time for steady-state formation. It is perfectly suited to the non-steady-state working environment of in-service buildings and meets the needs of short-term data collection.

[0021] This invention couples solar radiation, wind speed, indoor and outdoor temperatures with the external wall temperature extracted by CNN in real time. The external wall temperature output by CNN in real time is used as the boundary condition and constraint input PINN. The model can be automatically updated with changes in climate and indoor and outdoor temperatures, accurately describing the heat transfer process of the wall under complex climate conditions, and has non-steady-state adaptability.

[0022] This invention employs a piecewise, two-parameter rapid inversion method, using measured time-series temperature of the external wall surface as the core constraint. It integrates meteorological and equipment calibration information to construct a unified evaluation function within the PINN framework, encompassing control relationship residuals, internal and external boundary consistency, and data consistency. This method eliminates the need for indoor sampling and long-term steady-state conditions, providing reliable thermal conductivity and internal wall surface temperature through short-term testing. It is suitable for on-site applications such as external insulation acceptance and energy-saving retrofit assessments, simplifying operation, shortening cycles, reducing costs, and meeting the requirements for rapid, batch, and traceable engineering projects.

[0023] This invention offers high measurement efficiency and low cost: it significantly reduces the number of sensors required in the field and the workload of manual sampling, thus greatly shortening the overall measurement and calculation cycle. While ensuring high measurement accuracy, it effectively reduces equipment procurement and labor costs.

[0024] This invention employs minimal prior knowledge and possesses transferability: it has a low dependence on prior information such as material layers and thickness, and boundary conditions and physical constraints can automatically limit the parameter range. Combined with positive constraints, it ensures that the parameters conform to physical reality, making it easy to promote and apply among buildings in different climate regions, with various exterior wall types and orientations.

[0025] This invention has wide applicability: it is applicable to multiple stages such as energy efficiency inspection, energy-saving renovation acceptance, green certification review and operation and maintenance assessment of existing buildings, and is especially beneficial for rapid and large-scale testing in cold and frigid regions and building complexes with centralized heating.

[0026] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0027] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0028] Figure 1 This is a flowchart of the wall dual-unknown thermal parameter inversion method based on physical neural network described in Embodiment 1 of the present invention.

[0029] Figure 2 This is a structural diagram of the convolutional neural network model described in Embodiment 1 of the present invention.

[0030] Figure 3 This is a structural diagram of the physical neural network model described in Embodiment 1 of the present invention.

[0031] Figure 4 This is a schematic diagram of the structure of the wall dual-unknown thermal parameter inversion method based on physical neural network described in Embodiment 1 of the present invention.

[0032] Figure 5 This is a schematic diagram of the wall heat transfer mechanism described in Embodiment 1 of the present invention. Detailed Implementation

[0033] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0034] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0035] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0036] Example 1

[0037] Traditional steady-state or contact-based testing methods are highly dependent on environmental and operating conditions, have long testing cycles, and may cause some damage to buildings. They are difficult to meet the requirements of large-scale, rapid, and low-cost assessment of existing buildings, making it difficult to identify thermal parameters stably and reliably.

[0038] With the increasing maturity of non-contact measurement technologies such as deep learning, computer vision, infrared thermography, and multispectral imaging, thermal performance testing of building envelopes based on image-physical coupling has become practically feasible. Specifically, models such as convolutional neural networks can automatically and accurately identify thermal defects and temperature field characteristics from infrared image sequences, effectively suppressing noise interference caused by environmental disturbances such as solar radiation and wind speed. Simultaneously, by combining physical constraints with parameter inversion methods, even with limited prior information, the thermal parameters of building envelopes under unsteady conditions can be quickly and non-destructively estimated. This type of method offers significant advantages such as non-destructive testing, large-area coverage, speed, and low cost, enabling thermal parameter testing and online evaluation in existing buildings, thus providing reliable data support for energy-saving retrofit decisions in large-scale existing and new buildings.

[0039] However, in actual existing building scenarios, the heat transfer process of the walls, such as Figure 5 As shown, traditional methods typically only obtain limited known information. Key boundary conditions such as indoor wall temperature, indoor air temperature, and convective heat transfer coefficients are constrained by factors like wall damage and user privacy, making real-time and accurate acquisition difficult. When both the indoor wall temperature and thermal parameters such as wall thermal conductivity and heat transfer coefficient are unknown, traditional methods based on external surface temperature for parameter inversion are prone to ill-posedness and non-uniqueness, making it difficult to obtain reliable results consistently. Therefore, it is necessary to develop a superior method that can simultaneously determine indoor wall temperature and wall thermal parameters with limited known information, without compromising building integrity or relying on long-term steady-state conditions. This method must possess robustness, repeatability, and the ability to be rapidly deployed in engineering projects.

[0040] In one or more embodiments, a method for inverting dual unknown thermal parameters of a wall based on infrared images is disclosed. The dual unknown parameters refer to the unknown thermal conductivity and inner wall surface temperature of the wall. Figure 1 and Figure 4 As shown, it specifically includes:

[0041] Step S1: Obtain relevant environmental parameters and infrared image data of the wall to be tested at the time of inversion.

[0042] Step S1-1: Obtain the relevant environmental parameters at the time of inversion of the enclosure structure of the target to be tested.

[0043] Sensors are used to collect real-time environmental parameters of the building walls, such as indoor air temperature, outdoor air temperature, solar radiation intensity, wind speed, wall thickness, material density, and specific heat capacity.

[0044] Furthermore, the relevant environmental parameters collected from dynamic measurements are cleaned, and Lagrange interpolation is used to fill in any missing values. Timestamp alignment is also performed to ensure that all parameters are synchronized in time, thus preparing them for model input.

[0045] Step S1-2: Without adding any new indoor sensors, determine the indoor air temperature time series T using the sampling average method based on the dynamically measured indoor air temperature values. a,in (t). Specifically, to avoid adding new fixed indoor sensors, an average sampling method is used to construct an indoor air temperature time series, ensuring that it remains synchronized with meteorological and external surface temperature identification results on the time axis. During the detection period, several representative measurement points are selected in the room to be tested. Priority is given to main usage areas near the exterior walls, while avoiding direct sunlight, obvious heat sources, and locations directly exposed to air vents. Indoor air temperature samples are collected at fixed time intervals (5–15 minutes is recommended in this embodiment) to ensure that the sampling period covers the entire detection process. Furthermore, these samples are first denoised, and outliers are removed. Then, the time average is calculated within each time window. Finally, the processed data is aligned with the time axis of the infrared image to form a complete indoor air temperature time series, which serves as the input data for the inner boundary conditions in the subsequent inversion process.

[0046] Step S1-3: Obtain infrared images of the wall enclosure structure to be tested.

[0047] Capture real-time infrared images of the exterior wall surface of the wall to be tested; specifically, use an infrared camera with radiation thermometry capabilities.

[0048] When setting the camera's viewing angle, ensure that the angle between the camera's optical axis and the normal of the wall area to be measured is no greater than 45°. The shooting distance should allow the entire wall area to be measured to enter the camera's field of view, and the image pixel resolution should not exceed 5cm / PX. Avoid shooting from angles with strong specular reflections and areas with glass windows (related interference areas will be removed in the subsequent image segmentation step).

[0049] Calibration basis and setting parameters: Surface emissivity is determined based on the properties of the wall material or by on-site measurement using black tape; apparent reflectance temperature is measured using aluminum foil or a low emissivity sphere; relative humidity and ambient temperature are used for atmospheric transmission correction built into the infrared camera; at the same time, the frame rate and corresponding time step of image acquisition should be recorded.

[0050] Data Acquisition and Time Synchronization: The sequence of continuously captured infrared images of the exterior wall surface of the wall under test should be synchronized with outdoor meteorological parameters in time to ensure the use of subsequent physical neural network models.

[0051] An environment-driven data set with a unified timeline was constructed, continuously and accurately recording key physical quantities such as outdoor air temperature, relative humidity, wind speed, and solar radiation. During this process, it was ensured that this data was strictly aligned with the acquisition time of the infrared images. Interpolation methods were used to fill any potential data gaps, while outliers were eliminated to prevent interference with data accuracy. The result is an environment-driven dataset closely corresponding to the time variables, which will serve as a crucial basis for the external boundary conditions and driving inputs of the subsequent physical information neural network.

[0052] Step S2: Input the infrared image data into the convolutional neural network model to obtain the temperature field information of the outer wall surface of the wall to be tested.

[0053] The trained CNN model is used to analyze infrared images to obtain the continuous time-series temperature of the exterior wall surface. The specific process includes:

[0054] Using the infrared image of the exterior wall surface of the wall to be tested and the shooting parameters as input, a feature extraction network consisting of multi-layer convolution, nonlinear activation functions, downsampling and upsampling operations is used, such as... Figure 2 As shown, the convolutional neural network model consists of an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a first fully connected layer, an activation function layer, a second fully connected layer, and an output layer connected in sequence, which extracts the feature map of the input image.

[0055] Using a region-based statistical method, pixel-by-pixel surface temperature values ​​are output within the effective area through temperature regression and correction. The pixel-level temperature results are then aggregated and processed in chronological order to obtain representative time-series data T of the outer wall surface temperature. w,out (t), which is the temperature field information of the outer wall surface.

[0056] Step S3: Based on the smooth characteristics and moderate hysteresis effect of indoor air temperature, reasonably set the initial curve of the inner wall surface temperature and give an initial value for the thermal conductivity; construct a physical neural network model based on Fourier's law of heat conduction and the convection heat transfer boundary conditions of the inner and outer surfaces, such as... Figure 3 As shown.

[0057] Step S3-1: For the inner wall surface temperature time series, the indoor air temperature time series is obtained using the sampling average method. After noise reduction, time window averaging, and appropriate lag processing, the initial curve T of the inner wall surface temperature is formed. w,in,0 (t), and impose physical feasibility constraints such as amplitude range, rate of change and smoothness to ensure that it is accurately aligned with the meteorological and external surface temperature identification results on the time axis.

[0058] In this embodiment, the indoor air temperature time series is measured in real time. The air convection heat transfer coefficient of each room is regarded as a constant. The inner wall temperature at each moment is derived according to Fourier's law of thermal conductivity to obtain the inner wall temperature time series. The average value of each time point of the inner wall temperature time series is taken to obtain an initial temperature series (that is, it can also be plotted as an initial curve) which is used as the constraint condition for the subsequent physical neural network.

[0059] It should be understood that the applied amplitude range needs to be set according to standards or existing experience.

[0060] Step S3-2: For the thermal conductivity of the wall, based on the wall material composition and existing data, determine the physically feasible range of the wall thermal conductivity, and select the representative value within this range as the initial value k0 of the wall thermal conductivity, while applying positive values ​​and range constraints.

[0061] It should be understood that interval constraints mean that the initial thermal conductivity cannot be arbitrarily assigned a value, but needs to be set to a relatively reasonable but not precise value based on standards or existing experience.

[0062] Step S3-3: Construct a physical neural network model based on Fourier's law of heat conduction and the convection heat transfer boundary conditions of the inner and outer surfaces, specifically as follows:

[0063] A physical neural network model with Physical Information Neural Network (PINN) as its core framework is constructed. This model uses spatial coordinate x and time t as independent variables to characterize the wall temperature field T. w (x,t). Define the wall thermal conductivity k as a material parameter and apply positive values ​​and interval constraints; plot the indoor air temperature time series T. a,in (t) is defined as a time boundary quantity, while imposing physical feasibility constraints such as amplitude range, rate of change, and smoothness.

[0064] Among them, physical feasibility constraints include unsteady-state heat conduction control, as shown in Equation 1, to ensure energy conservation inside the wall; inner boundary, as shown in Equation 2; outer boundary, as shown in Equation 3; initial conditions, as shown in Equation 4; and the above items are combined into a unified evaluation function J according to weights.

[0065] (1)

[0066] (2)

[0067] (3)

[0068] (4)

[0069] In the formula, This is a time series of outdoor air temperatures. The temperature field of the wall is the temperature of the wall at position x and time t, and k is the thermal conductivity of the wall (parameter to be inverted). Let be the density of the wall material, and c be the specific heat capacity of the wall material. h is the heat flux density at the inner wall surface. in The indoor convective heat transfer coefficient is... Let T be the temperature of the inner wall at time t. a,in (t) represents the time series of indoor air temperature. h is the heat flux density at the outer wall surface. out (v) represents the outdoor convective heat transfer coefficient, and v represents the outdoor wind speed. I represents the solar radiation absorption rate of the wall. sum (t) represents the solar radiation intensity. Let x be the wall temperature at time 0.

[0070] In PINN, the outer surface temperature is generated in a forward manner and the evaluation function J is calculated, which serves as a unified evaluation and constraint framework for subsequent phased iterations.

[0071] Based on the Fourier partial differential equation and the convective heat transfer boundary conditions, the following loss function is constructed:

[0072] 1. Fourier equation residual loss:

[0073] (5)

[0074] in, The thermal conductivity of the wall is denoted as .

[0075] 2. Inner wall boundary residuals:

[0076] (6)

[0077] in, Let be the inner wall temperature in the nth iteration. The temperature of the inner wall surface is the temperature of the (n-1)th iteration, which is considered as the actual temperature of the inner wall surface of the room during the iteration process.

[0078] 3. Data constraint residuals:

[0079] (7)

[0080] in, This refers to the outer wall surface temperature calculated after iteration. This represents the actual temperature of the outdoor wall surface.

[0081] 4. Boundary condition loss on the outer wall surface:

[0082] (8)

[0083] in, This is a time series of outdoor air temperatures. The thickness of the wall; The outdoor convective heat transfer coefficient; The solar radiation absorption rate of the wall; This represents the intensity of solar radiation.

[0084] 5. Smoothing constraints include first-order smoothing terms and second-order smoothing terms;

[0085] First-order smoothing term:

[0086] (9)

[0087] in, , As weight, Let be the thermal conductivity of the wall in the nth iteration. Let be the inner wall temperature in the nth iteration. This is the total number of iterations; this term is used to suppress irregular and large fluctuations in unknown parameters.

[0088] Second-order smoothing term:

[0089] (10)

[0090] in, , This is for weighting. This term is used to smooth the data and suppress jagged edges during the process of handling unknown parameters;

[0091] (11)

[0092] in, Assign weights (constant or adaptive to operating conditions) to ensure that parameters are positive, bounded, and smooth over time.

[0093] Preferably, temperature gradient Second-order spatial derivative and time derivative Both utilize the automatic differentiation technique in physical information neural networks and the wall temperature field output by the network. The time derivative is calculated to be... It can also be combined with historical time series outputs to use sliding window difference fitting.

[0094] Furthermore, the joint evaluation function is designed as follows:

[0095] (12)

[0096] in, , , , , This is the weighting coefficient, with a default weight of 1.

[0097] During training, within the framework of the constructed unified evaluation function J, iterative optimization is carried out using the AdamW first-order adaptive optimizer. A phased two-parameter inversion strategy is employed for the physical neural network model to ensure stable convergence. The phased optimization iterative strategy is as follows: First, the initial values ​​of the inner wall temperature and the initial estimated value of the thermal conductivity are reasonably set according to physical laws. Then, under strict physical constraints, forward solving is performed to generate the predicted value of the outer surface temperature, which is then compared with the actual measured outer surface temperature hourly. While keeping the inner wall temperature constant, only the thermal conductivity is optimized and adjusted until it reaches a convergent state. Next, the converged thermal conductivity is fixed, and the inner wall temperature curve is iteratively corrected in reverse, and verified in independent time periods to ensure the accuracy of the results.

[0098] In the first stage, the inner wall temperature is fixed, and the thermal conductivity is optimized.

[0099] Maintain the inner wall temperature T during the nth iteration w,in,n (t) remains constant, and only the wall thermal conductivity k in the nth iteration is considered. n Optimization work will be carried out.

[0100] Using k n ,T w,in,n (t) is used to perform forward solving to obtain the predicted outer surface temperature. This temperature is then compared and analyzed hourly with the measured outer wall temperature obtained through CNN recognition. The evaluation function for the nth iteration is then calculated. Based on this, for k n Perform small-step monotonic updates or interval shrinking updates. When both conditions (13) and (14) are met, record the thermal conductivity at this point as the convergent solution. .

[0101] (13)

[0102] (14)

[0103] in, The thermal conductivity is obtained in the nth iteration. The thermal conductivity is obtained in the (n+1)th iteration. J is the preset relative convergence threshold. (n) J is the value of the joint loss function in the nth iteration. (n+1) This is the value of the joint loss function in the (n+1)th iteration. This is the preset loss convergence threshold.

[0104] In certain necessary situations, the external surface temperature error can be checked during independent periods when the system is not involved in training, thereby ensuring the convergence of the thermal conductivity solution. The robustness.

[0105] In the second stage, the thermal conductivity is fixed, and the inner wall temperature is optimized through reverse iteration.

[0106] Set the thermal conductivity as follows and keep it constant.

[0107] (15)

[0108] In the formula, k n Let be the thermal conductivity of the wall in the nth iteration. This is the convergent solution for thermal conductivity.

[0109] Under the conditions of satisfying smoothness requirements, amplitude range limitations, and rate of change constraints, the inner wall temperature T at time t is... w,in (t) Perform piecewise or spline-based small-amplitude correction operations on R pde R in R data The sub-objective of the main body is minimized, and the inner wall temperature T at time t is updated in the parameter network output. w,in (t). For T w,in (t), using a bounded mapping, is scaled to [T] by the Sigmoid function. min ,T max Simultaneously, time smoothing and rate of change constraints are applied. Iteration continues until the convergence conditions related to formulas 16-17 are met. This modification causes the objective function J to continuously decrease. The curve can be relaxed to effectively suppress potential oscillations. The above operations are continued until the conditions of the following formula are met.

[0110] (16)

[0111] (17)

[0112] In the formula, T w,in,n (t) represents the inner wall temperature in the nth iteration, T w,in,n+1 (t) represents the inner wall temperature in the (n+1)th iteration. This is the preset inner wall temperature threshold.

[0113] Step S4: Input the relevant environmental parameters and the temperature field information of the outer wall into the physical neural network model, and perform forward solving to obtain the heat transfer coefficient and inner wall temperature of the wall to be tested.

[0114] To ensure the inner wall temperature T during the nth iteration w,in,n (t) Under specific conditions where prior knowledge remains unchanged, with R pde R out R data To minimize the main sub-objectives, the thermal conductivity k of the wall in the nth iteration is calculated. n After optimization and updating, the thermal conductivity k of the wall in the (n+1)th iteration is obtained. n+1 For k n+1 Positive constraints are applied using the Softplus function, and irregular jumps during the iterative update process are suppressed by smoothing constraint terms.

[0115] With a fixed thermal conductivity k=k Under the premise of R pde R in R data The sub-objective of the main body is minimized, and the inner wall temperature T at time t is updated in the parameter network output. w,in (t).

[0116] The relevant environmental parameters of the wall under test and the temperature field information of the outer wall surface are input into the trained physical neural network model, which outputs the time-by-time thermal conductivity of the wall under test. With the inner wall temperature T at time t w,in (t) sequence.

[0117] Further calculation of the overall heat transfer coefficient U of the wall:

[0118] (18)

[0119] The final output includes a timestamp, environmental parameters, the inverted thermal conductivity k, and the inner wall temperature T at time t. w,in The report includes the test results for (t) and the final heat transfer coefficient U. Based on the above model, taking the exterior wall of an office building in a certain city as the target enclosure structure, the heat transfer coefficient of the building over a certain period of time was inverted. The entire inversion process is short and can be completed within 2 hours, realizing rapid and non-destructive testing under unsteady conditions.

[0120] This embodiment fully leverages the advantages of integrating neural network model data-driven approaches with physical mechanisms. The design process is simple and easy to implement, and it can be effectively applied to the thermal performance evaluation of the building envelope of existing buildings, providing a scientific basis for building energy-saving renovation.

[0121] This embodiment combines infrared imaging, convolutional neural networks (CNN), and physical information neural networks (PINN) to invert the dual unknown thermal parameters of walls. It can dynamically identify thermal conductivity and the temperature of the inner wall surface in unsteady environments, belonging to the field of building thermal and intelligent detection technology. This method integrates thermal imaging perception with heat transfer mechanism expression to construct a deep fusion framework. It uses CNN to process continuous temperature data from infrared images of the outer wall surface as boundary constraints and establishes a database of infrared images and temperatures. A PINN model is constructed based on the heat transfer effect of the outer boundary to simultaneously invert the thermal conductivity and the inner wall surface temperature. It requires no indoor sensors, can handle actual working conditions, and provides parameter settings for situations without indoor measuring points. It is suitable for scenarios such as building energy efficiency testing and non-invasive testing under unsteady climates, and can be applied to fields such as green building.

[0122] Specifically, firstly, without adding additional indoor sensors, indoor air temperature is determined by selective sampling to form a reasonable value selection strategy; at the same time, outdoor environmental parameters are collected and basic calibration is carried out, and all data are synchronized in time.

[0123] Secondly, by continuously acquiring infrared image sequences of the exterior wall surface, a convolutional neural network (CNN) is used to accurately extract the exterior wall temperature at each moment, and this temperature is input as a data constraint into a physical information neural network (PINN). Based on the strategy of integrating infrared observation of the exterior surface with physical constraints, an initial curve for the interior wall temperature is reasonably set based on the smoothness characteristics and moderate hysteresis effect of indoor air temperature, and an initial value for the thermal conductivity is given. Subsequently, according to Fourier's law of heat conduction and the convective heat transfer boundary conditions of the interior and exterior surfaces, the physical neural network constraint model (PINN) is constructed, using the aforementioned initial curve for the interior wall temperature and the initial value for the thermal conductivity as parameters, and performing forward solving to generate a predicted value for the exterior surface temperature. This predicted value is then compared with the actual exterior surface temperature hourly, thereby constructing an evaluation function covering multiple dimensions such as "physical residual, boundary consistency, and data consistency". Next, a phased optimization method is adopted: In the first phase, the inner wall temperature is kept constant, and only the thermal conductivity is adjusted to minimize the evaluation function until the thermal conductivity reaches a convergent state; in the second phase, based on the convergence of the thermal conductivity, the inner wall temperature curve is iteratively corrected in reverse. During this process, smoothing constraints and physical feasibility constraints are applied to ensure the rationality and reliability of the results. At the same time, the outer surface temperature is verified at independent time periods to check the accuracy of the model. Through the above iterative process of "setting-prediction-comparison-correction", a stable and reliable wall thermal conductivity can be obtained, and the hourly change of the inner wall temperature can be accurately determined simultaneously.

[0124] This method significantly reduces the cost of manual sampling and equipment deployment while ensuring high measurement accuracy. It can be used for energy efficiency inspection of existing buildings, acceptance of energy-saving renovations, and operation and maintenance assessments. It is especially suitable for rapid batch testing of building complexes in cold or frigid regions.

[0125] Example 2

[0126] In one or more embodiments, a wall dual-unknown thermal parameter inversion system based on a physical neural network is disclosed, specifically including:

[0127] The data acquisition module is used to acquire relevant environmental parameters and infrared image data of the wall under test at the time of inversion.

[0128] The outer wall temperature extraction module is used to input the infrared image data into a convolutional neural network model to obtain the temperature field information of the outer wall surface of the wall to be tested;

[0129] The model building module is used to set the initial curve of the inner wall surface temperature based on the smooth characteristics and moderate hysteresis effect of indoor air temperature, and to give the initial value of thermal conductivity; a physical neural network model is jointly constructed according to Fourier's law of thermal conduction and the convection heat transfer boundary conditions of the inner and outer surfaces; wherein, the physical neural network model is trained in a phased two-parameter inversion strategy within a unified evaluation function framework.

[0130] The inversion module is used to input the relevant environmental parameters and the temperature field information of the outer wall into the physical neural network model, and perform forward solving to obtain the heat transfer coefficient and inner wall temperature of the wall under test.

[0131] Example 3

[0132] This embodiment provides a computer-readable storage medium for storing a computer program. When the computer program is executed by a processor, it completes the steps of the above-described method for inverting dual unknown thermal parameters of a wall based on a physical neural network.

[0133] Example 4

[0134] This embodiment provides an electronic device, including a memory and a processor, as well as a computer program stored in the memory and running on the processor. When the computer program is run by the processor, it completes the steps of the above-described method for inverting the dual unknown thermal parameters of a wall based on a physical neural network.

[0135] Specifically, the device includes: a memory: non-temporarily storing the computer program described in Embodiment 3; a processor: coupled to the memory and configured to execute instructions in the memory to implement the steps of the method as described in Embodiment 1; a communication interface: for connecting to a sensor network, receiving collected data, and uploading final results; and a human-machine interface (optional): including a touchscreen for displaying real-time data, optimization processes, and final reports. This electronic device can be a dedicated portable testing instrument, or an industrial tablet PC or ruggedized laptop with the corresponding software installed, suitable for various complex field testing environments.

[0136] When the processor executes the computer program, the electronic device is configured to perform all the steps described in Embodiment 1: first, it receives a series of parameters such as the temperature of the inner and outer walls, indoor and outdoor temperatures, and solar radiation intensity through a data interface; then, it runs the internally stored convolutional neural network and physical neural network models for iterative optimization calculations; and finally, it outputs the inverted heat transfer coefficient value through a display unit or transmits it to other devices through a data interface. This electronic device constitutes the core calculation and control unit of the entire inversion system.

[0137] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0138] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0139] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment, whereby a series of operational steps are performed to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0140] The descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0141] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for inverting two unknown thermal parameters of a wall based on a physical neural network, characterized in that, include: Acquire relevant environmental parameters and infrared image data of the wall under test at the time of inversion; The infrared image data is input into a convolutional neural network model to obtain the temperature field information of the outer wall surface of the wall to be tested; An initial curve for the inner wall temperature is set based on the smooth characteristics and moderate hysteresis effect of indoor air temperature, and an initial value for the thermal conductivity is given. A physical neural network model is constructed based on Fourier's law of thermal conduction and the convection heat transfer boundary conditions of the inner and outer surfaces. The physical neural network model is trained using a phased two-parameter inversion strategy within a unified evaluation function framework. The relevant environmental parameters and the temperature field information of the outer wall surface are input into the physical neural network model, and the heat transfer coefficient and inner wall surface temperature of the wall under test are obtained by forward solving. The unified evaluation function is: in, , , , , These are the weighting coefficients. For the residual loss of the Fourier equation, For the inner wall boundary residual, For data-constrained residuals, For the boundary condition loss of the outer wall surface, For the smoothing constraint term; the phased two-parameter inversion strategy includes: In the first stage, the inner wall temperature is fixed, and the thermal conductivity is optimized. Specifically, under the condition that the inner wall temperature sequence remains unchanged a priori, the sub-objectives consisting of Fourier equation residual loss, outer wall boundary condition loss, and data constraint residual are minimized, and then the wall thermal conductivity output by the parameter network is updated. In the second stage, the thermal conductivity is fixed, and the inner wall temperature is optimized in reverse iteratively. Specifically, the thermal conductivity is set and kept fixed. Smoothness requirements, amplitude range limits, and rate of change constraints are adopted to perform piecewise or spline-style small-scale correction operations on the inner wall temperature curve. The sub-objectives, mainly the Fourier equation residual loss, inner wall boundary residual, and data constraint residual, are minimized to update the inner wall temperature output by the parameter network.

2. The method for inverting dual unknown thermal parameters of a wall based on a physical neural network as described in claim 1, characterized in that, The relevant environmental parameters include the indoor air temperature, outdoor air temperature, solar radiation intensity, wind speed, wall thickness, material density, and specific heat capacity of the wall.

3. The method for inverting dual unknown thermal parameters of a wall based on a physical neural network as described in claim 1, characterized in that, The convolutional neural network model includes an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a first fully connected layer, an activation function layer, a second fully connected layer, and an output layer connected in sequence, to extract feature maps from the input image; Using a surface domain statistical method, the surface temperature values ​​of each pixel are output through temperature regression and correction. The pixel-level temperature results are then summarized and processed in chronological order to obtain the temperature field information of the outer wall surface.

4. The method for inverting dual unknown thermal parameters of a wall based on a physical neural network as described in claim 1, characterized in that, A physical neural network model is constructed based on Fourier's law of heat conduction and the convection heat transfer boundary conditions of the inner and outer surfaces, specifically as follows: A physical neural network model with a physical information neural network as its framework is constructed. This model uses spatial coordinates x and time t as independent variables to characterize the temperature field information of the outer wall. The thermal conductivity k is defined as a material parameter and positive value and interval constraints are applied. The indoor air temperature time series is defined as a time boundary quantity, and physical feasibility constraints of amplitude range, rate of change and smoothness are applied.

5. The method for inverting dual unknown thermal parameters of a wall based on a physical neural network as described in claim 1, characterized in that, The boundary condition loss of the outer wall surface is: in, The thermal conductivity of the wall; Let represent the wall temperature field, indicating the temperature of the wall at position x and time t. The thickness of the wall; v is the outdoor convective heat transfer coefficient; v is the outdoor wind speed; This refers to the outer wall surface temperature calculated after iteration. This is a time series of outdoor air temperatures. The solar radiation absorption rate of the wall; This represents the intensity of solar radiation.

6. A wall dual-unknown thermal parameter inversion system based on physical neural networks, employing the wall dual-unknown thermal parameter inversion method based on physical neural networks as described in any one of claims 1-5, characterized in that, include: The data acquisition module is used to acquire relevant environmental parameters and infrared image data of the wall under test at the time of inversion. The outer wall temperature extraction module is used to input the infrared image data into a convolutional neural network model to obtain the temperature field information of the outer wall surface of the wall to be tested; The model building module is used to set the initial curve of the inner wall surface temperature based on the smooth characteristics and moderate hysteresis effect of indoor air temperature, and to give the initial value of thermal conductivity; a physical neural network model is jointly constructed according to Fourier's law of thermal conduction and the convection heat transfer boundary conditions of the inner and outer surfaces; wherein, the physical neural network model is trained in a phased two-parameter inversion strategy within a unified evaluation function framework. The inversion module is used to input the relevant environmental parameters and the temperature field information of the outer wall into the physical neural network model, and perform forward solving to obtain the heat transfer coefficient and inner wall temperature of the wall under test.

7. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, complete the wall dual-unknown thermal parameter inversion method based on physical neural networks as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, complete the wall dual-unknown thermal parameter inversion method based on physical neural networks as described in any one of claims 1-5.

Images