A method for predicting the spacing of radiant heating coils in net-zero energy buildings

Abstract

The present application provides a kind of net zero energy consumption building radiant heating coil spacing prediction method, it is related to coil spacing prediction field.Process includes building net zero energy consumption building radiant heating coil spacing data set, each net zero energy consumption building data is built as multidimensional input sequence, high-dimensional feature sequence is generated using double-layer nonlinear transformation network, building thermal disturbance feature sequence is generated in combination with the mean and standard deviation of high-dimensional feature sequence, introduce thermal response significance weight value and carry out weighted fusion, obtain fusion characteristics after layer normalization, extract thermal potential value and thermal slope, and input nonlinear gate structure with fusion characteristics, to obtain main heat transfer characteristics and residual heat characteristics, generate fusion response weight using element-by-element multiplication, weighted sum of main heat transfer characteristics and residual heat characteristics to obtain weighted thermal response characteristics, finally, weighted thermal response characteristics are input into linear network, and output radiant heating coil spacing prediction value.This method can realize the intelligent prediction of coil spacing control parameters.

Description

Technical Field

[0001] This invention belongs to the field of coil spacing prediction, specifically relating to a method for predicting the spacing of radiant heating coils in net zero energy buildings. Background Technology

[0002] With the development of net-zero energy buildings, radiant heating systems are widely used in residential and office buildings, becoming a key technology for achieving low-energy operation and thermal comfort. As an important parameter affecting system efficiency and energy consumption, the spacing between coils is directly related to the system's stability and energy-saving effect. However, due to the complexity and variability of building structures and spatial layouts, existing designs often rely on experience or fixed calculations, making it difficult to meet the precise configuration requirements in various scenarios. This leads to frequent adjustments, low efficiency, and affects the project's implementation. Therefore, there is an urgent need for a method to predict the spacing between radiant heating coils in net-zero energy buildings to improve the design quality and implementation efficiency of net-zero energy building projects.

[0003] Currently, the setting of coil spacing in radiant heating systems mainly relies on design specifications or engineering experience, lacking comprehensive consideration of building thermal performance and spatial structure. This makes it difficult to achieve accurate parameter matching and dynamic optimization. Although some methods introduce simulation analysis or static modeling, they generally suffer from problems such as single feature extraction dimensions, lagging response mechanisms, and poor model adaptability, which cannot effectively cope with diverse building scenarios, resulting in insufficient accuracy of prediction results.

[0004] The rational configuration of coil spacing in net-zero energy buildings is influenced by a combination of various building structural information and operational data. Different building structures exhibit significant differences in heat conduction paths and local thermal response mechanisms during construction, operation, and seasonal transitions, resulting in significant differences in the sensitivity of coil layout to changes in the thermal environment. Therefore, a prediction method that integrates multi-source structural characteristics and dynamic state evolution models can adapt to feature recognition and response mechanisms in multiple scenarios, enabling accurate prediction of coil spacing in complex building environments. Summary of the Invention

[0005] This invention provides a method for predicting the spacing of radiant heating coils in net-zero energy buildings. The method constructs a multi-dimensional input sequence from the data of each net-zero energy building, maps it through a two-layer nonlinear transformation network to obtain a highly discriminative high-dimensional feature sequence, and further combines its overall distribution mean and standard deviation to construct a building thermal disturbance feature sequence. A significant weight for thermal response is introduced to weight and fuse the building thermal disturbance feature sequence, followed by normalization to obtain fused features. Thermal potential energy values ​​and thermal slopes are extracted and embedded together with the fused features into a nonlinear gating structure, separating the main heat flow features and residual heat features. The element-wise interaction of the main heat flow features and residual heat features generates a fused response weight, which is then used to weight and fuse the main heat flow features and residual heat features to form a weighted thermal response feature. Finally, the weighted thermal response feature is input into a linear network to output a predicted value for the spacing of radiant heating coils, achieving accurate prediction of the spacing of radiant heating coils in net-zero energy buildings.

[0006] The technical method adopted by the present invention to achieve the above objectives specifically includes the following steps:

[0007] S1. Collect data on the spacing of radiant heating coils in net zero energy buildings and construct a dataset on the spacing of radiant heating coils in net zero energy buildings.

[0008] S2. Construct a multidimensional input sequence from the data of each net zero energy building in the dataset. Use a two-layer nonlinear transformation network to perform high-dimensional discriminative projection on the multidimensional input sequence. After processing with ReLU and GELU activation functions in sequence, obtain a high-dimensional feature sequence.

[0009] S3. Construct a building thermal disturbance sensing module, combine the high-dimensional feature sequence with its overall distribution mean and standard deviation to generate a building thermal disturbance feature sequence, introduce thermal response significance weight value, perform weighted fusion of building thermal disturbance feature sequence, and obtain fused features after layer normalization.

[0010] S4. Construct a heat potential diversion module, extract heat potential energy value and heat slope based on fusion features, embed the fusion features, heat potential energy value and heat slope together into a nonlinear gated structure, extract the main heat flow features, and calculate the waste heat features based on the main heat flow features.

[0011] S5. Using element-wise multiplication to combine the main heat flow characteristics and waste heat characteristics, generate a fusion response weight, and perform a weighted summation of the two types of characteristics to obtain the weighted thermal response characteristics.

[0012] S6. Input the weighted thermal response characteristics into the linear network and output the predicted value of the spacing between radiant heating coils in net zero energy buildings.

[0013] Preferably, in S1, data affecting the spacing of radiant heating coils in net zero energy buildings are collected, including building type, building area, building floor height, indoor target temperature setpoint, water flow rate, water pressure, construction materials, and wall thermal conductivity, to construct a dataset of radiant heating coil spacing for net zero energy buildings.

[0014] Preferably, traditional low-order feature processing methods are difficult to effectively capture the nonlinear coupling relationship between the radiant heating coil spacing data and the coil spacing in net-zero energy buildings. This results in insufficient ability of the model to distinguish the coil spacing adjustment requirements under different building types and operating conditions. This invention uses a nested activation function and a layer-by-layer weight adjustment mechanism of a two-layer nonlinear transformation network to perform high-dimensional discriminative projection on the input data, enhance feature representation, and enable the model to have stronger structural adaptability and predictive stability in different building scenarios.

[0015] Preferably, in S2, each net zero energy building data point in the net zero energy building radiant heating coil spacing dataset is constructed into a multidimensional input sequence. A two-layer nonlinear transformation network is used for high-dimensional discriminative projection, which enhances the distinguishability of features of different data types. The specific mathematical model is as follows:

[0016] ;

[0017] In the formula, The high-dimensional feature sequence is obtained through a nonlinear transformation network. , The weight matrix is ​​a learnable matrix. , This is a learnable bias.

[0018] Preferably, the two-layer nonlinear transformation network has stronger feature representation and spatial mapping capabilities. The first layer introduces the ReLU activation function, which can effectively suppress the propagation of input noise and low-information features, and enhance the model's response strength to local salient features. The second layer adopts the GELU activation function, whose continuity and high-order differentiability are conducive to maintaining the stability of gradient propagation and have higher sensitivity to small fluctuations in input features, thereby improving the model's ability to distinguish boundary samples. Ultimately, this makes different building data have clearer separability in high-dimensional space, providing the model with a more stable and discriminative feature foundation.

[0019] Preferably, in net-zero energy buildings, the radiant heating system needs to maximize thermal efficiency and minimize energy consumption. The coil spacing is a key control factor, and its prediction needs to combine heat flux distribution and thermal potential disturbance response. Traditional methods are difficult to capture the nonlinear disturbance trend of local thermal characteristics and are difficult to accurately reflect the time-varying thermal heterogeneity caused by temperature and construction materials. Therefore, by constructing a building thermal disturbance characteristic sequence and introducing a significant weight value for thermal response, the model's ability to identify high heat-sensitive areas can be enhanced, providing a foundation for subsequent prediction of radiant heating coil spacing.

[0020] Preferably, in S3, constructing the building thermal disturbance sensing module specifically includes the following steps:

[0021] S31, High-dimensional feature sequences and The mean and standard deviation of the overall distribution are combined to form a characteristic sequence of building thermal disturbance, which is used to capture abnormal shifts in the local thermal field under changes in coil spacing. The specific mathematical model is as follows:

[0022] ;

[0023] In the formula, This is a sequence of building thermal disturbance characteristics. This is a sign function that maps positive numbers to 1, negative numbers to -1, and 0 to 0. For element-wise multiplication, for The mean, for standard deviation To ensure that the number is a very small positive number, and to prevent division by zero, This is for taking the absolute value;

[0024] S32. Normalize each feature dimension in the building thermal disturbance feature sequence, and after squared enhancement, calculate the mean to generate a thermal response significance weight value to measure the importance of thermal disturbance features in overall heat flux control. The specific mathematical model is as follows:

[0025] ;

[0026] In the formula, For the first The significance weighting of thermal response data for radiant heating coil spacing in net-zero energy buildings. , These are characteristic sequences of building thermal disturbance. The Middle The first data on the spacing of radiant heating coils in net zero energy buildings and the dimensional eigenvalues Building thermal disturbance characteristic sequence Feature dimensions;

[0027] S33. The building thermal disturbance feature sequence is weighted and fused using the significance weight value of thermal response. After layer normalization, a fused feature is formed, which highlights the main feature while smoothing the influence of anomalous features. The specific mathematical model is as follows:

[0028] ;

[0029] In the formula, As a feature of fusion, For layer normalization, The types and quantities of data related to the spacing of radiant heating coils in net-zero energy buildings. Building thermal disturbance characteristic sequence The Middle Characteristics of the spacing data of radiant heating coils in net zero energy buildings.

[0030] Preferably, by constructing a building thermal disturbance feature sequence and introducing a significant weight value for thermal response, it is possible to simultaneously capture the non-uniform thermal response caused by changes in coil spacing and the dominant disturbance features in the local thermal field. This helps to characterize the sensitive and sluggish thermal zones in different areas of the building under radiant heating. The significant weight value for thermal response is fused element-wise with the original building thermal disturbance feature sequence, which can dynamically adjust the expression intensity of each building thermal disturbance feature, highlight the main thermal potential direction that has a key impact on heat load regulation, suppress the interference of local abnormal disturbances or measurement noise on feature expression, and enable the model to obtain a more stable and structured thermal response input. Furthermore, the global thermal field feature alignment is achieved through normalization fusion operation, which not only improves the clarity of the expression of the main thermal features, but also enhances the adaptability of the subsequent prediction model to differences in building structure, thereby improving the reliability and accuracy of coil spacing prediction.

[0031] Preferably, in a net-zero energy building radiant heating system, heat is conducted spatially through radiant structures such as the ground or walls. Different buildings exhibit significantly different thermal response behaviors, especially in coil layout. It is crucial to identify heat accumulation areas and heat flow lag areas to determine whether to increase or decrease coil spacing. Traditional modeling methods rely solely on global average response or boundary condition inputs, making it difficult to capture fine-grained changes in thermal slope between regions. This can easily lead to control failures such as local overheating or insufficient heating. Therefore, this invention provides a high-resolution, multi-scale understanding of thermal behavior for subsequent coil layout by expressing thermal potential energy, modeling thermal slope, and processing heat flow paths.

[0032] Preferably, in S4, constructing the thermal potential diversion module includes the following steps:

[0033] S41. The fusion characteristics are combined with the dual norm function and logarithmic modulation term to form a thermal potential value, in order to identify high heat flux regions in the heating structure and highlight the response capability to local heat accumulation effects. The specific mathematical model is as follows:

[0034] ;

[0035] In the formula, This is the thermal potential energy value. For L2 energy decay weights, The L1 distributed response modulation weights, It is a natural exponential function. Let e ​​be the base-e logarithmic function. It is the L2 norm. It is an L1 norm;

[0036] S42, Constructing a characteristic sequence of building thermal disturbance The mean vector, using the thermal potential energy value The numerator is the square of the L2 norm of the difference between the value and the mean vector, combined with the thermal potential energy value. The ratio formed by the L2 norm itself creates the thermal slope; the specific mathematical model is as follows:

[0037] ;

[0038] ;

[0039] In the formula, Building thermal disturbance characteristic sequence The mean vector, For thermal slope, It is a very small positive number;

[0040] S43. The fusion features, thermal potential energy value, and thermal slope are embedded into a nonlinear gated structure to extract the main heat flow features. These features are then used to obtain residual heat features, achieving directional separation modeling of the heat flow path. The specific mathematical model is as follows:

[0041] ;

[0042] ;

[0043] In the formula, Main heat transfer characteristics, The weight matrix is ​​a learnable matrix. For learnable bias, For splicing operations, As a residual heat characteristic, This is element-wise multiplication.

[0044] Preferably, by combining the L1 norm and L2 norm to construct a thermal potential energy expression and introducing logarithmic modulation to enhance the nonlinear modeling capability for high-response regions, the local heat concentration trend of buildings can be effectively revealed. Furthermore, by quantifying the regional thermal offset amplitude through thermal slope, the contribution of different regional characteristics in thermal field modeling can be dynamically adjusted, highlighting the dominant regions that have directional control significance for the heat diffusion process. Finally, the thermal potential energy value and thermal slope are jointly embedded into the gated shunt structure, achieving bi-branch directional decomposition while maintaining global information. This enables the model to capture the main heat flux and residual heat dynamic effects separately, providing a more physically reasonable and spatiotemporally robust feature basis for coil spacing layout.

[0045] Preferably, in net-zero energy buildings, the spacing of radiant heating coils is affected by multiple coupled factors, resulting in complex nonlinear interactions and heat lag between the main heat flow path and the waste heat channel. If only a single channel feature is used for prediction and modeling, it is often difficult to fully characterize the equilibrium and non-equilibrium fluctuations in the heat diffusion process. Therefore, this invention introduces a gating control mechanism to dynamically generate response weights through the interaction intensity between heat flow paths, effectively expressing the regional heterogeneity and control adaptability of the spacing of radiant heating coils in net-zero energy buildings.

[0046] Preferably, in S5, the main heat flux characteristics and waste heat characteristics are multiplied element-wise, and then compressed into a fusion response weight reflecting the current net zero energy building under the heat flux path using the Sigmoid activation function. The specific mathematical model is as follows:

[0047] ;

[0048] In the formula, These are the weights of the fused response generated by the Sigmoid activation function. To perform element-wise multiplication, the main heat flux characteristics and residual heat characteristics are weighted and summed separately based on the fused response weights to form a weighted thermal response characteristic. The specific mathematical model is as follows:

[0049] ;

[0050] In the formula, This represents the weighted thermal response characteristics of the final output.

[0051] Preferably, by constructing a fusion response weight based on element-level interaction and Sigmoid regulation, the fusion pattern between the main heat flux characteristics and the waste heat characteristics can be effectively captured. This helps to characterize the thermal response weight shift, dynamically adjust the expression intensity of the main heat flux characteristics and the waste heat characteristics, highlight the thermal structure response that has a key impact on heating performance, and suppress the interference caused by non-dominant thermal disturbances.

[0052] Preferably, in net-zero energy building design, the spacing between the coils of the radiant heating system directly determines the uniformity of heat distribution and the response efficiency of the heating system. Therefore, it is necessary to map the weighted thermal response characteristics into specific physical control quantities, that is, to predict the optimal spacing between the coils of the radiant heating system that should be set.

[0053] Preferably, in step S6, the weighted thermal response features are input into a linear network to obtain the predicted value of the radiant heating coil spacing, so as to fit the mapping relationship between the thermal response features and the coil spacing. The specific mathematical model is as follows:

[0054] ;

[0055] In the formula, This refers to the predicted spacing of radiant heating coils in net-zero energy buildings. The weight matrix is ​​a learnable matrix. This is a learnable bias term.

[0056] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention constructs a multi-dimensional input sequence from the data of each net-zero energy building, uses a two-layer nonlinear transformation network for high-dimensional discriminative projection to form a high-dimensional feature sequence, combines the overall distribution mean and standard deviation of the high-dimensional feature sequence to generate a building thermal disturbance feature sequence, and introduces a significant weight value for thermal response to achieve weighted fusion and layer normalization of the building thermal disturbance feature sequence, obtaining more representative fused features. Based on the fused features, thermal potential energy values ​​and thermal slopes are extracted. By embedding a nonlinear gating structure, the main heat flow features and waste heat features are effectively separated, comprehensively depicting the dominant path and delayed diffusion process of heat in the structure. Then, by performing element-wise interaction between the main heat flow features and waste heat features, a fused response weight is generated, and the two types of features are weighted and summed to form a unified weighted thermal response feature, thereby comprehensively expressing the multi-source heterogeneous characteristics in the building heat conduction process. Finally, the weighted thermal response feature is input into a linear network to output the predicted value of the radiant heating coil spacing of the net-zero energy building, realizing intelligent prediction of the coil spacing control parameter. Attached Figure Description

[0057] Figure 1 A step-by-step diagram illustrating a method for predicting the spacing of radiant heating coils in net-zero energy buildings.

[0058] Figure 2 This is a diagram of a building thermal disturbance sensing module.

[0059] Figure 3 This is a diagram of the thermal potential distribution module.

[0060] Figure 4 This is a graph showing the loss curve during the model training process.

[0061] Figure 5A diagram showing the predicted spacing of radiant heating coils in a net-zero energy building. Detailed Implementation

[0062] This invention proposes a method for predicting the spacing of radiant heating coils in net-zero energy buildings. This method utilizes a two-layer nonlinear transformation network to perform high-dimensional discriminative projection on the multidimensional input sequence of data for each net-zero energy building, obtaining a high-dimensional feature sequence. The mean and standard deviation of the high-dimensional features are combined to generate a building thermal disturbance feature sequence. A significant weight value for thermal response is introduced to weight and fuse the building thermal disturbance feature sequence, and layer normalization is used to improve feature stability, forming a fused feature. Further, thermal potential energy and thermal slope are extracted and embedded together with the fused feature into a nonlinear gating structure, separating the main heat flow feature and the residual heat feature. Based on the element-wise interaction of the main heat flow feature and the residual heat feature, a fused response weight is generated and weighted to sum the main heat flow feature and the residual heat feature, obtaining a weighted thermal response feature. Finally, the weighted thermal response feature is input into a linear network to output the predicted value of the radiant heating coil spacing in net-zero energy buildings. This achieves fusion perception and accurate modeling of multi-source information, improving the prediction accuracy of coil spacing in different building scenarios. The following will describe the technical solution in the embodiments of this invention in detail and completely, specifically including the following steps, such as... Figure 1 As shown.

[0063] S1. Collect data on the spacing of radiant heating coils in net zero energy buildings and construct a dataset of radiant heating coil spacing in net zero energy buildings.

[0064] Furthermore, in S1, data on the spacing of radiant heating coils in 150 net zero energy buildings were collected. The data types included building type, building area, building floor height, indoor target temperature setpoint, water flow rate, water pressure, construction materials, and wall thermal conductivity. A dataset on the spacing of radiant heating coils in net zero energy buildings was constructed. The dataset was divided into a training set and a validation set in a 7:3 ratio for subsequent model training and evaluation.

[0065] S2. Construct a multidimensional input sequence for each net zero energy building in the dataset. Use a two-layer nonlinear transformation network to perform high-dimensional discriminative projection on the multidimensional input sequence. After processing with ReLU and GELU activation functions in sequence, obtain a high-dimensional feature sequence.

[0066] Furthermore, in S2, the data for each net zero-energy building in the net zero-energy building radiant heating coil spacing dataset is constructed as a multidimensional input sequence. , ,in The first in net zero energy building data Collected values ​​of type data, To collect the total number of net zero energy building data types, a two-layer nonlinear transformation network was used to... High-dimensional discriminability projection is performed to enhance the discriminability of features from different data types. The specific mathematical model is as follows:

[0067] ;

[0068] In the formula, The high-dimensional feature sequence is obtained through a nonlinear transformation network. ,in The first in net zero energy building data High-dimensional features of typed data For high-dimensional feature sequences, the data dimension , The weight matrix is ​​a learnable matrix. , This is a learnable bias.

[0069] S3. Construct a building thermal disturbance sensing module, combine the high-dimensional feature sequence with its overall distribution mean and standard deviation to generate a building thermal disturbance feature sequence, introduce thermal response significance weight value, perform weighted fusion of building thermal disturbance feature sequence, and obtain fused features after layer normalization processing.

[0070] Furthermore, in S3, a building thermal disturbance sensing module is constructed, such as... Figure 2 As shown, the specific steps include:

[0071] S31, High-dimensional feature sequences and The mean and standard deviation of the overall distribution are combined to form a characteristic sequence of building thermal disturbance. The specific mathematical model is as follows:

[0072] ;

[0073] In the formula, This is a sequence of building thermal disturbance characteristics. This is a sign function that maps positive numbers to 1, negative numbers to -1, and 0 to 0. For element-wise multiplication, for The mean, , for standard deviation , To prevent division by zero, the initial value is set to 0.00001 during implementation, where the number is an extremely small positive number. This is for taking the absolute value;

[0074] S32. Normalize each feature dimension in the building thermal disturbance feature sequence, and after squared enhancement, calculate the mean to generate the thermal response significance weight value. The specific mathematical model is as follows:

[0075] ;

[0076] In the formula, For the first The significance weighting of thermal response data for radiant heating coil spacing in net-zero energy buildings. , These are characteristic sequences of building thermal disturbance. The Middle The first data on the spacing of radiant heating coils in net zero energy buildings and the dimensional eigenvalues Building thermal disturbance characteristic sequence Feature dimensions;

[0077] S33. The building thermal disturbance feature sequence is weighted and fused using the significance weight value of thermal response. After layer normalization, a fused feature is formed, which highlights the main feature while smoothing the influence of anomalous features. The specific mathematical model is as follows:

[0078] ;

[0079] In the formula, As a feature of fusion, For layer normalization, This represents the total number of net-zero energy building data types collected. Building thermal disturbance characteristic sequence The Middle Characteristics of the spacing data of radiant heating coils in net zero energy buildings.

[0080] S4. Construct a thermal potential diversion module, extract thermal potential energy value and thermal slope based on fusion features, embed the fusion features, thermal potential energy value and thermal slope together into a nonlinear gated structure, extract the main heat flow features, and calculate the waste heat features based on the main heat flow features.

[0081] Furthermore, in S4, a heat potential diversion module is constructed, such as... Figure 3 As shown, it includes the following steps:

[0082] S41. The fusion characteristics are combined with the dual norm function and logarithmic modulation term to form a thermal potential value, in order to identify high heat flux regions in the heating structure and highlight the response capability to local heat accumulation effects. The specific mathematical model is as follows:

[0083] ;

[0084] In the formula, This is the thermal potential energy value. This is the L2 energy decay weight, initially set to 0.05 during implementation. The L1 distributed response modulation weights are initially set to 0.1 during implementation. It is a natural exponential function. Let e ​​be the base-e logarithmic function. It is the L2 norm. It is an L1 norm;

[0085] S42, Constructing a characteristic sequence of building thermal disturbance The mean vector, using the thermal potential energy value The numerator is the square of the L2 norm of the difference between the value and the mean vector, combined with the thermal potential energy value. The ratio formed by the L2 norm itself creates the thermal slope; the specific mathematical model is as follows:

[0086] ;

[0087] ;

[0088] In the formula, Building thermal disturbance characteristic sequence The mean vector, For thermal slope, It is a very small positive number, and its initial value is set to 0.00001 during implementation;

[0089] S43. The fusion features, thermal potential energy value, and thermal slope are embedded into a nonlinear gated structure to extract the main heat flow features. These features are then used to obtain residual heat features, achieving directional separation modeling of the heat flow path. The specific mathematical model is as follows:

[0090] ;

[0091] ;

[0092] In the formula, Main heat transfer characteristics, The weight matrix is ​​a learnable matrix. For learnable bias, For splicing operations, As a residual heat characteristic, This is element-wise multiplication.

[0093] S5. Using element-wise multiplication to combine the main heat flow characteristics and waste heat characteristics, a fusion response weight is generated. The two types of characteristics are then weighted and summed to obtain the weighted thermal response characteristics.

[0094] Furthermore, in S5, the main heat flux characteristics and residual heat characteristics are multiplied element-wise, and then compressed into a fusion response weight reflecting the current net zero energy building under the heat flux path using the Sigmoid activation function. The specific mathematical model is as follows:

[0095] ;

[0096] In the formula, These are the weights of the fused response generated by the Sigmoid activation function. To perform element-wise multiplication, the main heat flux characteristics and residual heat characteristics are weighted and summed separately based on the fused response weights to form a weighted thermal response characteristic. The specific mathematical model is as follows:

[0097] ;

[0098] In the formula, This represents the weighted thermal response characteristics of the final output.

[0099] S6. Input the weighted thermal response characteristics into the linear network and output the predicted value of the spacing between radiant heating coils in net zero energy buildings.

[0100] Furthermore, in S6, the weighted thermal response features are input into a linear network to obtain the predicted value of the radiant heating coil spacing, in order to fit the mapping relationship between the thermal response features and the coil spacing. The specific mathematical model is as follows:

[0101] ;

[0102] In the formula, This refers to the predicted spacing of radiant heating coils in net-zero energy buildings. The weight matrix is ​​a learnable matrix. This is a learnable bias term.

[0103] Furthermore, the model is implemented using Python 3.10, built and trained based on the PyTorch framework, and deployed in a GPU-accelerated environment supporting CUDA 12.1. During model training, an NVIDIA 3090 24GB GPU is used for parallel computation. In terms of optimization strategy, the Lookahead combinatorial optimizer is used, and a learning rate warm-up mechanism is introduced in the early stage of training. The base learning rate is 0.0005, and the CosineAnnealing learning rate scheduler is used for dynamic adjustment to avoid local optima traps. The batch size is 8, and the loss function is weighted smoothed L1 loss.

[0104] Furthermore, the loss curve of the model training process of this method is as follows: Figure 4 As shown in the figure, the horizontal axis represents the number of training epochs, and the vertical axis represents the loss value. It can be observed from the figure that as the number of training epochs increases, the loss value shows a continuous decreasing trend and gradually stabilizes, indicating that the model has good convergence and training stability. The prediction effect of this method is as follows: Figure 5As shown, the horizontal axis represents the building number with net zero energy consumption, and the vertical axis represents the corresponding radiant heating coil spacing value (mm). The gray dashed line in the figure represents the actual assessed value, and the black solid line represents the predicted value output by the method of this invention. Figure 5 As can be seen, the predicted values ​​and the actual assessed values ​​maintain a high degree of consistency in the overall trend, with small fluctuation errors, which fully verifies the accuracy and reliability of the method in the task of predicting the spacing of radiant heating coils in net-zero energy buildings.

[0105] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept of the present invention, and these modifications and improvements all fall within the protection scope of the present invention.

Claims

1. A method for predicting the spacing of radiant heating coils in net-zero energy buildings, characterized in that, The method comprises the following steps: S1, collecting data affecting the radiation heating coil spacing of net zero energy consumption buildings, and constructing a radiation heating coil spacing data set of net zero energy consumption buildings; S2, each net zero energy consumption building data in the data set is constructed as a multi-dimensional input sequence, a double-layer nonlinear transformation network is used to project the multi-dimensional input sequence into a high-dimensional distinguishable space, and a high-dimensional feature sequence is obtained after being processed by ReLU and GELU activation functions in turn; S3, construct a building thermal disturbance perception module, combine the high-dimensional feature sequence with its overall distribution mean and standard deviation to generate a building thermal disturbance feature sequence, introduce a thermal response significance weight value, weight and fuse the building thermal disturbance feature sequence, and obtain a fused feature after layer normalization processing ; S4, a heat potential diversion module is constructed, the fusion feature is used to extract heat potential value and heat slope, the fusion feature, the heat potential value and the heat slope are embedded in a nonlinear gating structure, the main heat flux feature is extracted, and the residual heat feature is calculated based on the main heat flux feature; S5, the main heat flux feature and the residual heat feature are combined by using element-by-element multiplication to generate a fusion response weight, the two types of features are weighted and summed to obtain a weighted heat response feature; S6, the weighted heat response feature is input into a linear network to output a prediction value of the radiation heating coil spacing of the net zero energy consumption building; In S4, the fusion feature, the heat potential value and the heat slope are embedded in a nonlinear gating structure to extract the main heat flux feature, and the residual heat feature is further obtained based on the main heat flux feature to realize directional separation modeling of the heat flow path, and the specific mathematical model is: ； ； wherein, is a main heat flux feature, is a learnable weight matrix, is a learnable bias, is a concatenation operation, is a residual heat feature, is an element-wise multiplication, is a fused feature; is a heat potential value, is a heat slope; The heat potential value is extracted based on the fusion feature, which is combined with a double norm function and a logarithmic modulation term to form the heat potential value, and the specific mathematical model is: ； wherein is a thermal potential value, is an L2 energy decay weight, is an LI distribution response modulation weight, is a natural exponential function, is a natural logarithm function with base e, is an L2 norm, is an LI norm; Constructing a sequence of thermal disturbance features of a building The square of the L2 norm of the difference between the mean vector and the thermal potential vector is the numerator, and the thermal potential value is combined The square of the L2 norm of the difference between the mean vector and the thermal potential vector is the numerator, and the thermal potential value is combined The square of the L2 norm of the difference between the mean vector and the thermal potential vector is the numerator, and the thermal potential value is combined The square of the L2 norm of the difference between the mean vector and the thermal potential vector is the numerator, and the thermal potential value is combined The fusion features, thermal potential energy values, and thermal slope are concatenated and embedded into a nonlinear gated structure to generate the main heat flux feature. Utilizing the main heat flux characteristics in The complementary values ​​and fusion features within the range are multiplied element-wise to generate residual heat features. .

2. A method of predicting the spacing of a radiant heating coil for a net zero energy building according to claim 1, wherein, The data affecting the radiation heating coil spacing of net zero energy consumption buildings is collected, including building type, building area, building height, indoor target temperature setting value, water flow, water pressure, construction material, wall thermal conductivity, and a radiation heating coil spacing data set of net zero energy consumption buildings is constructed.

3. A method of predicting the spacing of a radiant heating coil for a net zero energy building according to claim 2, wherein, Each net-zero energy building data in the net-zero energy building radiant heating coil spacing dataset is constructed as a multi-dimensional input sequence , high-dimensional distinguishability projection is performed by using a double-layer nonlinear transformation network, specifically, linear transformation is performed on the input data by using a weight matrix and a bias term , first layer nonlinear activation is performed through a ReLU activation function, and then the activated result is linearly mapped with a weight matrix and a bias term, and is processed through a GELU activation function, to obtain a high-dimensional feature sequence .

4. A method of predicting the spacing of a radiant heating coil for a net zero energy building according to claim 3, wherein, The high-dimensional feature sequence With The mean and standard deviation in the overall distribution are combined to form a building heat disturbance feature sequence, and the specific mathematical model is: ； wherein is a sequence of building thermal perturbation features, is a sign function, is an element-wise multiplication, is is a mean value of is is a standard deviation of is a small positive number, is an absolute value operation.

5. A method of predicting the spacing of a radiant heating coil for a net zero energy building according to claim 4, wherein, The feature dimensions in the building heat disturbance feature sequence are normalized, and the mean value is calculated after square enhancement to generate a heat response significance weight value, and the specific mathematical model is: ； In the formula, For the first The significance weighting of thermal response data for radiant heating coil spacing in net-zero energy buildings. , These are characteristic sequences of building thermal disturbance. The Middle The first data on the spacing of radiant heating coils in net zero energy buildings and the dimensional eigenvalues Building thermal disturbance characteristic sequence Feature dimensions; The building thermal disturbance feature sequence is weighted and fused by using the thermal response significance weight value, and after layer normalization processing, a fused feature is formed .

6. A method of predicting the spacing of a radiant heating coil for a net zero energy building according to claim 1, wherein, The main heat flux feature and the residual heat feature are multiplied element by element, and compressed into a fusion response weight by a Sigmoid activation function, and the main heat flux feature and the residual heat feature are weighted and summed based on the fusion response weight to form a weighted heat response feature .

7. A method of predicting the spacing of a radiant heating coil for a net zero energy building according to claim 6, wherein, inputting the weighted thermal response features into a linear network to obtain a radiant heating coil spacing prediction value .

Images