Building envelope heat flow prediction method based on KAN-Film fusion model

Abstract

The invention relates to a building envelope heat flow prediction method based on a KAN-Film fusion model, and the method comprises the following steps: 1), obtaining the historical indoor and outdoor air temperature time sequence data of a building envelope, and the thermal resistance and thermal capacity of the envelope, and forming a training data set; 2) constructing a model, wherein the model comprises a KAN network layer and a Film modulation layer, and the KAN network layer and the Film modulation layer cooperate to form a heat flow prediction model; 3) model training: inputting training data into a KAN-Film model, calculating a difference value between a predicted value and an actual monitoring heat flow by using a mean square error as a loss function, and iteratively optimizing model parameters; and 4) performing heat flow prediction by using the trained model: calculating a corresponding heat flow prediction value of the enclosure structure according to the new outdoor environment condition and the enclosure structure parameter input. The accuracy, generalization and real-time performance of heat flow prediction of the building enclosure structure can be improved, and the demand of intelligent building energy consumption optimization control can be met.

Description

Technical Field

[0001] The present invention relates to a heat flux prediction method, specifically to a heat flux prediction method for building envelopes based on a KAN-Film fusion model, which is applicable to building energy-saving design and intelligent control, and belongs to the field of building energy consumption prediction integrating artificial intelligence technology. Background Technique

[0002] As the main barrier for indoor-outdoor heat exchange, the dynamic heat transfer process of building envelopes (such as walls) has a significant impact on the overall building energy consumption. Accurately predicting the heat flux change of the envelope is not only the basis for building energy efficiency assessment but also the key to realizing intelligent control and energy-saving operation of air-conditioning systems. However, due to the combined action of multiple factors in the heat transfer process of the envelope, including outdoor air temperature, solar radiation, indoor temperature, and the thermal performance of wall materials, the dynamic response of heat flux is complex and variable, making heat flux prediction a long-term technical difficulty in the fields of building physics and energy management.

[0003] Traditional methods mainly establish analytical models based on heat transfer mechanisms, such as the response factor method, the thermal resistance-capacitance equivalent circuit model, etc. Although they have certain physical interpretability, they usually rely on the acquisition of complete thermal parameters and complex mathematical derivations, making it difficult to meet the requirements of real-time and adaptability of intelligent control systems.

[0004] In recent years, data-driven machine learning methods (such as artificial neural network ANN, support vector machine SVM, etc.) have gradually been applied to the field of building energy consumption prediction to learn the heat flux law through historical data. However, these models generally have problems such as high dependence on data volume, lack of physical constraints, and poor interpretability. Once applied to scenarios outside the training data (such as different climate conditions or wall structures), the prediction accuracy often drops significantly, and the extrapolation ability is seriously insufficient. Considering the high diversity of building structures and usage environments, it is actually difficult to provide sufficient training samples for each building, which greatly limits the popularization and application of pure data-driven methods in engineering practice.

[0005] In response to the above problems, some studies in recent years have attempted to combine physical prior knowledge with data-driven models. For example, Reference 1 ("Physics-Informed Neural Networks for Heat Transfer Problems," S. Cai, Z. Wang, et al., Journal of Heat Transfer, Vol. 143, No. 6: 060801, 202106) discloses the PINN method of embedding physical boundary conditions into neural networks, which improves the physical consistency of applying machine learning methods to heat transfer problems. However, this method has a complex structure and is difficult to effectively introduce high-order thermal parameters, and has not yet been able to achieve general prediction for any enclosure structure. Reference 2 ("A data-driven energy consumption prediction model using multilayer perceptron for building thermal performance," K. Zhang, Y. Li, et al., Energy, Vol. 272: 127078, 202303) proposes an energy consumption prediction method based on the Multilayer Perceptron (MLP) model. Although it performs well in the same climate region, the prediction error increases significantly when facing wall structure changes or cross-regional migrations, revealing its lack of generalization ability.

[0006] In summary, the existing technologies mainly have the following problems in the prediction of heat flow in building envelopes: First, the lack of physical constraints leads to insufficient generalization ability of the model; second, traditional physical models are difficult to quickly adapt to changing working conditions; third, high-precision prediction often relies on a large amount of historical data, restricting its actual application effect.

[0007] Therefore, to solve the technical problems of poor model extrapolation ability, insufficient physical consistency, and high data demand in the existing technologies, it is indeed necessary to provide an innovative heat flow prediction method for building envelopes based on the KAN-Film fusion model to overcome the defects in the existing technologies. Summary of the Invention

[0008] The object of the present invention is to provide a method for predicting the heat flux of building envelopes based on the KAN-Film fusion model, which combines the structural characteristics of the Kolmogorov-Arnold Network (KAN for short) with the Feature-wise Linear Modulation (Film for short) mechanism, and introduces the prior knowledge of the thermal parameters of the building envelope to improve the accuracy, generalization and real-time performance of the heat flux prediction of the building envelope, and meet the actual needs of the intelligent building energy consumption optimization control.

[0009] To achieve the above object, the technical solution adopted by the present invention is: a method for predicting the heat flux of building envelopes based on the KAN-Film fusion model, which includes the following process steps: 1), Data acquisition and preprocessing: Obtain the historical time series data of the outdoor air temperature of the building envelope and the time series data of the indoor air temperature , as well as the thermal resistance R and heat capacity C thermal parameters of the envelope structure, to form the training data set of the prediction model; 2), Construction of the KAN-Film model: Based on the KAN network principle, predefine the structure of the prediction model. This model includes two parts: the KAN network layer and the Film modulation layer. Among them, the KAN network layer is used to extract the mapping relationship in the temperature data and in step 1) to generate the basic prediction value; the Film modulation layer is based on the wall thermal resistance R and heat capacity C, and the output of its activation function is scaled and offset modulated in the feature dimension to achieve dynamic adaptation to the changes in the thermal structure; the heat flux prediction model is jointly constituted by the KAN network layer and the Film modulation layer; 3), Model training: Input the training data obtained in step 1) into the KAN-Film model, use the mean square error as the loss function to calculate the difference between the predicted value and the actual monitored heat flux, and iteratively optimize the model parameters; 4), Use the trained KAN-Film prediction model to predict the heat flux: According to the input of the new outdoor environmental conditions and envelope structure parameters, calculate the corresponding predicted value of the heat flux of the envelope structure, where the Film modulation layer dynamically adjusts the output of the KAN network layer according to the new envelope structure parameters to achieve the adaptation of the model to different envelope structures.

[0010] The method for predicting the heat flux of building envelopes based on the KAN-Film fusion model of the present invention is further: In step 1), the training data set is subjected to normalization processing, and its specific manifestation is:

[0011]

[0012]

[0013] Among them, is the model training data set, and N is the number of training set subsamples; and are the input and output vectors of the i-th training sample; is the outdoor composite temperature time series vector within the i-th training sample; is the indoor air temperature time series vector within the i-th training sample; is the wall heat flux time series vector within the i-th training sample; , are the thermal resistance and heat capacity values within the i-th training sample.

[0014] The building envelope heat flux prediction method based on the KAN-Film fusion model of the present invention is further: in the step 2), the KAN network layer adopts a single hidden layer structure with variable width, including N input nodes and 1 output node, and the input nodes correspond to the outdoor equivalent temperature and indoor temperature data of N pre-selected time steps.

[0015] The building envelope heat flux prediction method based on the KAN-Film fusion model of the present invention is further: in the step 2), the calculation method of the heat flux prediction model for heat flux data is as follows:

[0016] Among them, is the wall heat flux data predicted by the KAN-Film model; and are the modulation factors of the Film layer; represents the output of the KAN network layer for the input temperature time series data .

[0017] The building envelope heat flux prediction method based on the KAN-Film fusion model of the present invention is further: in the step 3), the loss function is specifically:

[0018] Among them, is the loss function; is the number of samples in the training batch; For the model to predict the heat flux value; For the true heat flux value.

[0019] The method for predicting the heat flux of the building envelope based on the KAN-Film fusion model of the present invention further is: in the step 3), the Adam optimization algorithm is used to iteratively update all learnable parameters, including the weight parameters of the KAN network layer and the parameters of the Film modulation network ; in each training iteration step, first calculate the current network output , then calculate the loss function and the gradients of the loss function with respect to each parameter, and update the parameters according to the following algorithm rules:

[0020]

[0021] Where: is the weight parameter of the KAN network layer; is the weight parameter of the Film network layer; is the learning rate of the Adam algorithm; and are the first-order moment estimate and the second-order moment estimate respectively; θ is the gradient; is the update of the gradient θ; is a decimal term to prevent division by zero.

[0022] [[ID=4|6]]The method for predicting the heat flux of the building envelope based on the KAN-Film fusion model of the present invention further is: the parameters to be iteratively optimized in the model include the weight function parameters in the KAN layer and the modulation parameters in the Film layer.

[0023] The method for predicting the heat flux of the building envelope based on the KAN-Film fusion model of the present invention may also be: after completing the model training in step 3), evaluate the performance of the model in the small-sample learning scenario; subsequently, apply the trained model to an independent test data set and external working conditions different from the training conditions to test its generalization performance for unknown inputs. [[ID=%0]]

[0024] Compared with the prior art, the present invention has the following beneficial effects: 1. Structure generalization: It can accurately predict the wall heat flux under different combinations of thermal resistance and heat capacity.

[0025] 2. Physical interpretability: The response function is embedded as a network weight mapping and has a clear physical meaning.

[0026] 3. Strong small-sample learning ability: It can still achieve 80% accuracy when the training data only accounts for 20%.

[0027] 4. High computational efficiency: The single-layer structure model significantly reduces the consumption of computing resources compared with traditional deep networks.

[0028] 5. Greatly improve the model generalization ability and reduce data requirements: By integrating physical parameters into the network structure, the model can adapt to the changes in climate and envelope structures in different regions and still maintain a high prediction accuracy under unseen working conditions. At the same time, due to the integration of physical prior knowledge, the model can also converge and achieve good performance with only a small amount of training data, which is better than traditional black-box models that rely on a large number of samples.

[0029] 6. Ensure physical consistency and interpretability: The present invention pre-embeds the physical model form of the heat transfer of the envelope structure. The KAN network layer uses linear activation to make the network structure equivalent to the first-order response factor model, thereby ensuring that the model output has a physical linear relationship basis with temperature changes and will not produce unreasonable results that violate the heat conduction law. The modulation parameters and respectively reflect the influence of different wall structures on the heat flow amplitude and reference value, making the model have a certain physical interpretability and being more easily understood and trusted by engineering personnel.

[0030] 7. Reduce data requirements and computing costs: The model structure of the present invention is concise, the number of parameters is greatly reduced, and the training and inference computing overheads are lower, which is suitable for real-time control scenarios. Brief Description of the Drawings

[0031] Figure 1 is a schematic diagram of the influencing factors of the heat flow of the envelope structure in step 1) of the present invention.

[0032] Figure 2 is a flowchart of the construction of the KAN-Film model in step 2) of the present invention.

[0033] Figure 3 is the comparison result of the prediction accuracy between the KAN-Film model and the traditional MLP model of the present invention under small-sample training conditions.

[0034] Figure 4 is the performance of the KAN-Film model of the present invention for extrapolation prediction on different heat capacity wall structures.

[0035] Figure 5 is a comparison chart of the heat flow prediction performance of the KAN-Film model, MLP and LSTM models of the present invention for cross-training and testing on different wall thickness data sets.

[0036] Figure 6 is the meteorological data used in the specific implementation case of the present invention.

[0037] Figure 7 is the model output result diagram of the specific implementation case of the present invention. Specific implementation manner

[0038] The present invention is a method for predicting the heat flux of a building envelope structure based on a KAN-Film fusion model, which includes the following process steps: 1), Data acquisition and preprocessing: Since the heat flux of the building envelope structure is affected by factors such as temperature, radiation, and wall thermal parameters, by obtaining the historical outdoor air temperature (the solar radiation has been converted into an equivalent temperature) time series data of the building envelope structure , indoor air temperature time series data , and the thermal resistance R and heat capacity C of the envelope structure, a training data set for the prediction model is formed.

[0039] Please refer to the attached instructions Figure 1 As shown, it shows the process of the heat flux being jointly affected by multiple factors such as wall structure, orientation, radiation, climate, and indoor settings. To construct a heat flux prediction model with structural generalization ability, in this embodiment, a multi-group of building envelope structure heat flux data sets covering different climate conditions, wall structures, azimuth orientations, and usage seasons are systematically collected during the data preparation stage. The specific content of data collection includes: a wide combination of different envelope structure types (such as concrete + EPS insulation layer, aerated concrete, double-layer gypsum board + SPF insulation layer, etc.), corresponding thermal resistance and heat capacity parameters; different azimuth angles (east, south, west, north), corresponding solar radiation incident intensities at different angles; different climate zones and seasons (including the winter heating period and the summer cooling period); different indoor temperature setting values and modes; each sample includes the time series measured equivalent outdoor temperature , indoor temperature (including solar radiation processing). The above factors jointly constitute the comprehensive sample input space required for network training.

[0040] All original temperature and heat flux data are uniformly sampled and time-aligned to form a training sample set for the KAN-Film model: that is, the training data set is normalized, and its specific manifestation is:

[0041]

[0042]

[0043] Among them, is the model training data set, and N is the number of sub-samples in the training set; and is the input and output vector of the i-th training sample; is the outdoor composite temperature time series vector within the i-th training sample; is the indoor air temperature time series vector within the i-th training sample; is the wall heat flux time series vector within the i-th training sample; , are the thermal resistance and heat capacity values within the i-th training sample.

[0044] 2), Construction of the KAN-Film model. Please refer to the appendix of the specification Figure 2 As shown, it demonstrates the entire construction process from obtaining historical meteorological data and indoor temperature, input construction, neural network prediction, Film physical modulation, to finally outputting the predicted heat flux value.

[0045] Specifically, the present invention predefines the prediction model structure based on the KAN network principle. This model includes two parts: the KAN network layer and the Film modulation layer, which are in a parallel structure for receiving and processing different parameters in the model input samples Among them, the KAN network layer is used to extract the temperature data in input step 1) and in the mapping relationship, and generate a basic predicted value. The Film modulation layer is based on the wall thermal resistance R and heat capacity C, and its activation function output performs scaling and offset modulation in the feature dimension to achieve dynamic adaptation to changes in the thermal structure. The KAN network layer and the Film modulation layer cooperate to form a heat flux prediction model with interpretability and extensibility.

[0046] Among them, the prior physical knowledge of the response factor method of wall dynamic heat transfer is embedded in the construction of the KAN network layer. Specifically, the wall heat flux is not an instantaneous response to the indoor-outdoor temperature difference, but a time-delay response process jointly determined by the temperature differences at the current and multiple historical moments in the form of linear convolution. Based on the above physical model, the KAN network is designed as a single hidden layer structure with variable width, including N input nodes and 1 output node. The input nodes correspond to the outdoor equivalent temperature and indoor temperature data of N pre-selected time steps; the activation function of the KAN network is fixed as a linear function, and the time series delay characteristics of the indoor temperature and the outdoor equivalent temperature are used as the dominant variable inputs. Through the mapping of these historical features, the network can achieve a preliminary prediction of the wall heat flux.

[0047] The Film modulation layer receives physical parameters such as the thermal resistance R and heat capacity C of the enclosure structure as modulation inputs, and dynamically modulates the output of the KAN network layer through a shallow auxiliary multi-layer perceptron (MLP) neural network. This auxiliary network takes R and C as inputs and outputs a set of scaling factors and offset factors to modulate the output results of each activation function in the KAN network layer, enabling the model to automatically adjust the output according to the changes in the thermal performance of the enclosure structure. The model output is the predicted dynamic heat flux of the wall.

[0048] Based on the above theoretical basis, the specific calculation method of the heat flux prediction model for heat flux data is as follows:

[0049] where is the predicted wall heat flux data by the KAN-Film model; and are the modulation factors of the Film layer; represents the output of the KAN network layer for the input temperature time series data .

[0050] 3) Model training: Input the training data obtained in step 1) into the KAN-Film model constructed in step 2), use the mean square error as the loss function to calculate the difference between the predicted value and the actual monitored heat flux, and iteratively optimize the model parameters. The specific loss function is:

[0051] where is the loss function; is the number of samples in the training batch; is the predicted heat flux value of the model; is the true heat flux value.

[0052] To ensure the stability of network training and improve the generalization ability of the model, the present invention uses the Z-score normalization method to normalize the temperature time series data in each training sample , and the processing method is as follows:

[0053] where: is the mean value of the variable ; is the variable Standard deviation; Is the normalized value.

[0054] Furthermore, during iterative optimization, the present invention uses the Adam optimization algorithm to iteratively update all learnable parameters, including the weight parameters of the KAN network layer And the parameters of the Film modulation network ; In each training iteration step, first calculate the current network output , then calculate the loss function And the gradients of the loss function with respect to each parameter, and update the parameters according to the following algorithm rules:

[0055]

[0056] Where: Is the weight parameter of the KAN network layer; Is the weight parameter of the Film network layer; Is the learning rate of the Adam algorithm; And Are the first-order moment estimate and the second-order moment estimate respectively; θ is the gradient; Is the update of the gradient θ; Is a small decimal term to prevent division by zero.

[0057] Furthermore, to avoid overfitting, an L1 regularization strategy is introduced in the above training process: imposing a sparsity constraint on the network parameters and using the Early Stopping strategy to continuously monitor the change of the loss on the validation set. If there is no improvement in several consecutive rounds, the training is terminated early to avoid overfitting.

[0058] After completing the model training in step 3), the model can be further verified and adjusted, that is, evaluate the performance of the model in the few-shot learning scenario to address the common problem of insufficient sensing data in the construction field. By restricting the number of training samples, test the learning ability and prediction accuracy of the model with only a small amount of data. Subsequently, apply the trained model to an independent test dataset and external working conditions different from the training conditions to test its generalization performance for unknown inputs, and pay special attention to the adaptability and prediction stability of the model when the wall structure parameters (such as thickness, heat capacity, etc.) change.

[0059] To comprehensively verify the practicability, robustness, and structural extrapolation ability of the method of the present invention, the completed KAN-Film model carried out systematic performance evaluation experiments under various data conditions, covering three aspects: few-shot learning ability, structural extrapolation ability, and cross-structural transfer generalization ability.

[0060] 1. Few-shot training verification In the context of limited training data volume, the prediction accuracy performances of the model of the present invention and the traditional MLP model were compared and analyzed. Figure 3 The attached figure shows the comparison results of the prediction accuracies of the KAN-Film model and the traditional MLP model under few-shot training conditions. The results show that when the training set only accounts for 10% of all the data, KAN-Film can still maintain a low-error output, while the prediction error of the MLP model increases significantly, indicating that the method of the present invention has strong few-shot learning ability and meets the actual application requirements of scarce building thermal monitoring data.

[0061] 2. Structural extrapolation ability test Please refer to the attached figure in the specification Figure 4 As shown, to verify the generalization performance of the model on structural parameter combinations not seen before, a set of heat capacity parameters and wall structure samples not present in the training set were selected for extrapolation prediction. Figure 4 The results show that KAN-Film can accurately capture the changing trend of the wall heat flux, demonstrating its prediction ability under unknown structural conditions, and verifying the extrapolation advantage brought by the physical knowledge embedded in the model. Even if the test data contains combinations of thermal resistance and heat capacity that the model has never seen before, KAN-Film can still accurately predict the changes in the wall heat flux, indicating its good structural extrapolation ability.

[0062] 3. Cross-structural transfer learning ability evaluation Please refer to the attached figure in the specification Figure 5 As shown, cross-training and testing experiments were carried out on datasets of multiple wall thicknesses to compare the generalization performances of KAN-Film with mainstream models such as MLP and LSTM. Figure 5 The attached figure shows the transfer prediction errors between wall structures of different thicknesses (240mm, 250mm, 370mm, 480mm). The figure represents the coefficient of determination R² of different models under each dataset combination in the form of a heat matrix. The results show that KAN-Film is superior to the comparison models under all structural combinations. Especially in the high-heat-capacity thick-wall structure, its prediction error is much lower than that of LSTM and MLP, having stronger structural transfer ability and prediction accuracy, and further verifying its adaptability under complex structural conditions.

[0063] In summary, the KAN-Film model demonstrates superior modeling performance in multi-dimensional tests. Especially in the actual building scenarios with scarce samples and variable structures, it shows extrapolation ability and stability far beyond traditional data-driven methods, fully meeting the "physical knowledge-driven + data fusion" modeling concept proposed by the present invention.

[0064] 4), Use the trained KAN-Film prediction model to predict the heat flux: According to the input of new outdoor environmental conditions and envelope structure parameters, calculate the corresponding predicted value of the envelope heat flux. Among them, the Film modulation layer dynamically adjusts the output of the KAN network layer according to the new envelope structure parameters to enable the model to adapt to different envelope structures.

[0065] Furthermore, the obtained predicted value of the envelope heat flux can be input into the building air-conditioning control system as a reference signal for model predictive control in the future, so as to guide the optimization adjustment of the air-conditioning system, thereby reducing energy consumption fluctuations and improving the indoor temperature control accuracy.

[0066] The KAN-Film prediction model of the present invention is universal for different building structures and climate conditions. Without modifying the model structure, it can complete the prediction of a new envelope structure by inputting the corresponding thermal resistance and heat capacity parameters. For the working conditions of a new envelope structure lacking measured data, the model parameter correction can be completed by adding a small amount of sample data of this working condition to the model training set, so as to control the prediction error within the allowable range.

[0067] To more clearly demonstrate the application process and performance of the method of the present invention in the actual building heat flux prediction task, the following will take the measured data of a small house from August 10th to August 19th, 2024 as an example to detail the implementation steps and application effects of the present invention.

[0068] In this embodiment, the collected outdoor air temperature and solar radiation intensity (the test data is shown in the appendix Figure 6 , in the figure, the red curve is the outdoor air temperature, and the blue curve is the solar radiation intensity) are calculated to obtain the outdoor composite temperature.

[0069] Select the air temperature inside the experimental small house as the input of the indoor thermal boundary condition, and together with the processed outdoor composite temperature constitute the time series input vector , which is used for the feedforward calculation of the heat flux prediction model. In addition, the physical and thermal parameters of the envelope structure of the experimental small house are also known. Among them, the wall thermal resistance is about 1.25 m²K / W, and the heat capacity is 612000 J / m²K. The above thermal resistance and heat capacity are used as modulation factors and input into the Film layer in the KAN-Film model.

[0070] During the model execution phase, the constructed KAN-Film neural network model is used to predict the feature inputs at each time point, and the corresponding instantaneous wall heat flux value q(t) is output. This heat flux prediction represents the heat conduction intensity penetrating the enclosure structure per unit area and is one of the core parameters in building thermal environment regulation. The prediction process is as shown in the aforementioned model structure description. First, the KAN network layer realizes the mapping from the temperature time series to the preliminary heat flux estimation, and then the Film modulation layer receives the R and C parameters and applies scaling and offset transformations, and finally outputs the heat flux prediction value that integrates physical information.

[0071] To verify the effectiveness of the method proposed in the present invention, the heat flux prediction values output by the model are compared point by point with the measured heat flux data in the experiment. The comparison results are as shown in the appendix Figure 7 In the figure, the abscissa represents time (in hours), the ordinate represents the heat flux value per unit area (in W / m²), the blue continuous curve is the measured heat flux data, and the black discontinuous curve is the prediction result of the model of the present invention. This Figure 7 shows the wall heat flux curve predicted using the KAN-Film model and compares it with the measured data, verifying the applicability and accuracy of the model under actual working conditions.

[0072] It can be clearly seen from the figure that the KAN-Film model proposed in the present invention can accurately predict the trend of wall dynamic heat flux changes in most time periods. Especially at the moments when the heat flux changes violently, such as in the morning heating stage and the evening cooling stage, the model responds quickly, has a high degree of trend coincidence, and the peak prediction error is extremely small, reflecting good dynamic tracking ability. Through statistical analysis, the mean absolute error (MAE) of the model of the present invention during the entire test period is less than 0.43 W / m², and the correlation coefficient R 2 reaches 0.97.

[0073] The above specific implementation manners are only preferred embodiments of this creation and are not used to limit this creation. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of this creation shall be included within the protection scope of this creation.

Claims

1. A method for predicting the heat flux of building envelopes based on the KAN-Film fusion model, characterized in that: It includes the following technological steps: 1), Data acquisition and preprocessing: Obtain the historical time series data of the outdoor air temperature of the building envelope , the time series data of the indoor air temperature , and the thermal resistance R and heat capacity C of the envelope, which are thermal engineering parameters, to form a training data set for the prediction model; 2), KAN-Film model construction: Based on the KAN network principle, a prediction model structure is predefined. The model consists of two parts: the KAN network layer and the Film modulation layer. Among them, the KAN network layer is used to extract the temperature data in input step 1) and the mapping relationship in to generate a basic prediction value; The Film modulation layer is based on the wall thermal resistance R and heat capacity C, and the output of its activation function is scaled and offset modulated in the feature dimension to achieve dynamic adaptation to the changes in the thermal structure; The KAN network layer and the Film modulation layer cooperate to form a heat flow prediction model; 3), Model training: Input the training data obtained in step 1) into the KAN-Film model, use the mean square error as the loss function to calculate the difference between the predicted value and the actual monitored heat flux, and iteratively optimize the model parameters; 4), Use the trained KAN-Film prediction model to predict the heat flux: According to the input of new outdoor environmental conditions and enclosure structure parameters, calculate the corresponding predicted value of the enclosure structure heat flux, where the Film modulation layer dynamically adjusts the output of the KAN network layer according to the new enclosure structure parameters to achieve the adaptation of the model to different enclosure structures.

2. The method for predicting heat flow of building envelope structures based on the KAN-Film fusion model according to claim 1, characterized in that: In step 1), the training data set is subjected to normalization processing, and its specific manifestation is: Among them, is the model training data set, and N is the number of training set subsamples; and are the input and output vectors of the i-th training sample; is the outdoor integrated temperature time series vector within the i-th training sample; is the indoor air temperature time series vector within the i-th training sample; is the wall heat flux time series vector within the i-th training sample; , are the thermal resistance and heat capacity values within the i-th training sample.

3. The method for predicting heat flow of building envelope structures based on the KAN-Film fusion model according to claim 1, characterized in that: In step 2), the KAN network layer adopts a single hidden layer structure with a variable width, including N input nodes and 1 output node, and the input nodes correspond to the outdoor equivalent temperature and indoor temperature data of N pre-selected time steps.

4. The method for predicting the heat flux of building envelopes based on the KAN-Film fusion model according to claim 1, wherein: In step 2), the calculation method of the heat flux prediction model for heat flux data is as follows: Among them, is the predicted wall heat flux data by the KAN-Film model; and is the modulation factor of the Film layer; Represents the output of the KAN network layer for the input temperature time-series data of.

5. The method for predicting heat flow of building envelope structures based on the KAN-Film fusion model according to claim 1, characterized in that: In step 3), the loss function is specifically: Among them, is the loss function; The number of samples in the training batch; Predict the numerical value of heat flux for the model; is the true heat flux value.

6. The method for predicting the heat flux of building envelopes based on the KAN-Film fusion model according to claim 5, characterized in that: In step 3), the Adam optimization algorithm is used to iteratively update all learnable parameters, including the weight parameters of the KAN network layer. and Film modulation network parameters ; In each training iteration step, first calculate the current network output , then calculate the loss function and the gradients of the loss function with respect to each parameter, and update the parameters according to the following algorithm rules: Wherein: is the weight parameter of the KAN network layer; The weight parameter for the Film network layer; is the learning rate of the Adam algorithm; and are the first moment estimate and the second moment estimate respectively; θ is the gradient; is the gradient θ update; To prevent fractional terms from being divided by zero.

7. The method for predicting the heat flux of a building envelope based on the KAN-Film fusion model according to claim 5, characterized in that: In step 3), the parameters iteratively optimized in the model include the weight function parameters in the KAN layer and the modulation parameters of the Film layer.

8. The method for predicting the heat flux of building envelopes based on the KAN-Film fusion model according to claim 1, wherein: After completing the model training in step 3), evaluate the performance of the model in the small-sample learning scenario; subsequently, apply the trained model to an independent test data set and external working conditions different from the training conditions to test its generalization performance for unknown inputs.

Images