Method for predicting distance between radiant heating coils of building with net zero energy consumption

Abstract

The invention provides a net zero energy consumption building radiation heating coil pipe spacing prediction method, and relates to the field of coil pipe spacing prediction. The process comprises the steps of constructing a net zero energy consumption building radiant heating coil spacing data set, constructing each piece of net zero energy consumption building data into a multi-dimensional input sequence, generating a high-dimensional feature sequence by adopting a double-layer nonlinear transformation network, generating a building thermal disturbance feature sequence in combination with a mean value and a standard deviation of the high-dimensional feature sequence, and calculating the distance between the building radiant heating coils according to the building thermal disturbance feature sequence. Introducing a thermal response significance weight value to carry out weighted fusion and layer normalization to obtain a fusion feature, extracting a thermal potential energy value and a thermal slope, inputting the thermal potential energy value and the thermal slope into a nonlinear gating structure together with the fusion feature to obtain a main heat flux feature and a waste heat feature, and generating a fusion response weight by utilizing an element-by-element multiplication method; and performing weighted summation on the main heat flux characteristic and the waste heat characteristic to obtain a weighted thermal response characteristic, finally inputting the weighted thermal response characteristic into a nonlinear network, and outputting a predicted value of the distance between the radiant heating coils. According to the method, intelligent prediction of coil pipe spacing regulation and control parameters can be realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of coil spacing prediction, and particularly relates to a radiation heating coil spacing prediction method for net zero energy consumption buildings. BACKGROUND

[0002] With the development of net zero energy consumption buildings, radiation heating systems are widely used in residential buildings and office buildings, and become the key technology for realizing low energy consumption operation and thermal comfort. Coil spacing is an important parameter affecting system efficiency and energy consumption level, and its rationality is directly related to system stability and energy saving effect. However, due to the complex and variable building structure and space layout, existing design often relies on experience or fixed calculation, which is difficult to meet the precise configuration in multiple scenarios, resulting in frequent adjustment, low efficiency and affecting the engineering landing effect. Therefore, a radiation heating coil spacing prediction method for net zero energy consumption buildings is needed to improve the design quality and implementation efficiency of net zero energy consumption building projects.

[0003] At present, the setting of coil spacing in the radiation heating system mainly depends on design specifications or engineering experience, and lacks comprehensive consideration of building thermal performance and space structure, making it difficult to realize precise matching and dynamic optimization of parameters. Although some methods introduce simulation analysis or static modeling means, they generally have the problems of single feature extraction dimension, lagging response mechanism and poor model adaptability, which cannot effectively cope with diversified building scenarios, resulting in insufficient accuracy of prediction results.

[0004] The reasonable configuration of coil spacing in net zero energy consumption buildings is jointly influenced by multiple types of building structure information and operation condition data. Different building structures have obvious differences in heat conduction path and local thermal response mechanism during the construction stage, operation stage and seasonal transition process, resulting in significant differences in the sensitivity of coil layout to thermal environment changes. Therefore, a prediction method that integrates multi-source structure features and dynamic state evolution mode can adapt to feature recognition and response mechanism in multiple scenarios, and realize precise prediction of coil spacing in complex building environment. SUMMARY

[0005] The application provides a net zero energy consumption building radiant heating coil spacing prediction method, each net zero energy consumption building data is constructed as a multi-dimensional input sequence, is mapped through a double-layer nonlinear transformation network, a high-dimensional feature sequence with high distinguishability is obtained, a building thermal disturbance feature sequence is further constructed by combining the overall distribution mean and standard deviation thereof, a thermal response significance weight is introduced, the building thermal disturbance feature sequence is weighted and fused, and is subjected to normalization processing, a fusion feature is obtained, a thermal potential energy value and a thermal slope are extracted, and are embedded in a nonlinear gate structure together with the fusion feature, a main heat flux feature and a residual heat feature are branched out, element-by-element interaction of the main heat flux feature and the residual heat feature generates a fusion response weight, and the main heat flux feature and the residual heat feature are weighted and fused based on the weight to form a weighted thermal response feature, finally, the weighted thermal response feature is input into a nonlinear network, and a radiant heating coil spacing prediction value is output, so that the radiant heating coil spacing of the net zero energy consumption building is accurately predicted.

[0006] The technical method adopted by the application to achieve the above-mentioned purpose specifically includes the following steps: S1, collecting data affecting the radiant heating coil spacing of the net zero energy consumption building, and constructing a net zero energy consumption building radiant heating coil spacing dataset; S2, constructing each net zero energy consumption building data in the dataset as a multi-dimensional input sequence, using a double-layer nonlinear transformation network to project the multi-dimensional input sequence in a high-dimensional distinguishable manner, and obtaining a high-dimensional feature sequence after processing by ReLU and GELU activation functions in turn; S3, constructing a building thermal disturbance perception module, generating a building thermal disturbance feature sequence by combining the high-dimensional feature sequence with the overall distribution mean and standard deviation thereof, introducing a thermal response significance weight value, weighting and fusing the building thermal disturbance feature sequence, and obtaining a fusion feature after layer normalization processing; S4, constructing a thermal potential shunt module, extracting a thermal potential energy value and a thermal slope based on the fusion feature, embedding the fusion feature, the thermal potential energy value and the thermal slope in a nonlinear gate structure together, extracting a main heat flux feature, and calculating a residual heat feature based on the main heat flux feature; S5, generating a fusion response weight by using element-by-element multiplication to combine the main heat flux feature and the residual heat feature, weighting and summing the two types of features, and obtaining a weighted thermal response feature; S6, inputting the weighted thermal response feature into a nonlinear network, and outputting a net zero energy consumption building radiant heating coil spacing prediction value.

[0007] Preferably, in S1, the data affecting the radiant heating coil spacing of the net zero energy consumption building includes building type, building area, building height, indoor target temperature set value, water flow, water pressure, construction material, and wall thermal conductivity, and the net zero energy consumption building radiant heating coil spacing dataset is constructed.

[0008] Preferably, the traditional low-order feature processing method is difficult to effectively capture the nonlinear coupling relationship between the radiant heating coil spacing data and the coil spacing of the net zero energy building, resulting in insufficient ability of the model to distinguish the demand for coil spacing adjustment under different building types and working conditions. The present application enhances the feature representation by projecting the input data through the high-dimensional distinguishability projection of the nested activation function and the layer-by-layer weight adjustment mechanism of the double-layer nonlinear transformation network, so that the model has stronger structural adaptability and prediction stability under different building scenarios.

[0009] Preferably, in S2, each net zero energy building data in the net zero energy building radiant heating coil spacing data set is constructed as a multi-dimensional input sequence , and the double-layer nonlinear transformation network is used for high-dimensional distinguishability projection, so that different data type features have stronger distinguishability. The specific mathematical model is: ; In the formula, is the high-dimensional feature sequence obtained through the nonlinear transformation network, , is a learnable weight matrix, , is a learnable bias.

[0010] Preferably, the double-layer nonlinear transformation network has stronger feature expression ability and spatial mapping ability. The first layer introduces the ReLU activation function, which can effectively suppress the propagation of input noise and low information content features, enhance the response strength of the model to local significant features, and the second layer adopts the GELU activation function, which has continuity and high-order derivable characteristics, which is conducive to maintaining the stability of gradient propagation and has higher sensitivity to small fluctuations of input features, thereby improving the discrimination ability of the model to boundary samples, so that different building data have clearer distinguishability in high-dimensional space, providing a more stable and discriminable feature basis for the model.

[0011] Preferably, in the net zero energy building, the radiant heating system needs to maximize thermal efficiency and minimize energy consumption, and the coil spacing as a key control factor needs to be predicted in combination with the heat flux distribution and the thermal potential disturbance response. The traditional method is difficult to capture the nonlinear disturbance trend of local thermal features, and it is difficult to accurately reflect the time-varying thermal heterogeneity caused by temperature and construction materials. Therefore, by constructing the building thermal disturbance feature sequence and introducing the thermal response significance weight value, the identification ability of the model to high thermal sensitive areas can be strengthened, providing a basis for subsequent radiant heating coil spacing prediction.

[0012] Preferably, in S3, a building thermal disturbance perception module is constructed, which specifically includes the following steps: S31, projecting the high-dimensional feature sequence and The mean and standard deviation of the overall distribution are combined to form a building thermal disturbance characteristic sequence to capture the abnormal deviation of the local thermal field under the change of coil spacing. The specific mathematical model is: ; Where, is the characteristic sequence of building thermal disturbance, is a sign function that maps positive numbers to 1, negative numbers to -1, and 0 to 0. is element-wise multiplication, for The mean of for The standard deviation of is a very small positive number to prevent division by zero. To take the absolute value operation; S32. Normalize each feature dimension in the building thermal disturbance feature sequence, square enhance, and average the values ​​to generate the thermal response significance weight value, which measures the importance of the thermal disturbance feature in the overall heat flux regulation. The specific mathematical model is: ; Where, For the The significance weight of the thermal response of the radiant heating coil spacing data of a net zero energy building, 、 They are building thermal disturbance characteristic sequences Middle Data on Radiant Heating Coil Spacing for Net Zero Energy Buildings Hedi The eigenvalues ​​of the dimensions, is the characteristic sequence of building thermal disturbance characteristic dimensions; S33. Use the thermal response significance weight value to perform weighted fusion on the building thermal disturbance feature sequence. After layer normalization, a fusion feature is formed, which highlights the main features while smoothing the influence of abnormal features. The specific mathematical model is: ; Where, To fusion features, is layer normalization, The number of types of data for radiant heating coil spacing in net zero energy buildings, is the characteristic sequence of building thermal disturbance Middle Characterization of radiant heating coil spacing data for a net-zero energy building.

[0013] Preferably, by constructing the building thermal disturbance feature sequence and introducing the thermal response significance weight value, the non-uniform thermal response caused by the change of the coil spacing and the dominant disturbance feature in the local thermal field can be captured at the same time, which is helpful to represent the sensitive thermal zone and the lag thermal zone of different areas of the building under the action of the radiant heating, the thermal response significance weight value is weighted and fused with the original building thermal disturbance feature sequence element by element, the expression strength of each building thermal disturbance feature can be dynamically adjusted, the main heat potential direction which has a key influence on the thermal load regulation is highlighted, the interference of local abnormal disturbance or measurement noise on the feature expression is suppressed, the model obtains more stable and structured thermal response input, and further through the normalization fusion operation, the global thermal field feature alignment is realized, which not only improves the expression clarity of the main heat feature, but also enhances the adaptability of the subsequent prediction model to the structural differences of the building, so that the reliability and precision of the coil spacing prediction are improved.

[0014] Preferably, in the radiant heating system of the net zero energy consumption building, heat is conducted through the radiant structure such as the ground or the wall, and different building thermal response behaviors are significantly different, especially in the coil layout, the heat accumulation area and the heat flow lag area need to be identified to determine whether to encrypt or disperse the coil spacing, the traditional modeling method only relies on the global average response or the boundary condition input, and it is difficult to capture the fine-grained heat slope change between regions, and it is easy to appear the regulation failure problems such as local overheating or insufficient heating, therefore, the heat potential expression, heat slope modeling and heat flow path shunting processing are provided in the present application, so as to provide high-resolution and multi-scale thermal behavior understanding for subsequent coil layout.

[0015] Preferably, in S4, the heat potential shunting module is constructed, including the following steps: S41, the fusion feature is combined with a double norm function and a logarithmic modulation term to form a heat potential value, so as to identify the high heat flow area in the heating structure and highlight the response ability to the local heat accumulation effect, and the specific mathematical model is: ; In the formula, the heat potential value is L2 energy attenuation weight, L1 distribution response modulation weight, natural exponential function, logarithmic function with e as the base, L2 norm, L1 norm; S42, the mean vector of the building thermal disturbance feature sequence is constructed, and the square of the L2 norm of the difference between the heat potential value and the mean vector is taken as the numerator, and the ratio formed by combining the L2 norm of the heat potential value itself is taken as the heat slope, and the specific mathematical model is: ; ; wherein, is a sequence of building thermal disturbance features is a mean vector of is a thermal slope, is a very small positive number; S43, embedding the fusion features into a nonlinear gate control structure with the thermal potential energy value and the thermal slope, extracting the main heat flux features, further obtaining the residual heat features by using the main heat flux features, realizing 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 splicing operation, is a residual heat feature, is an element-wise multiplication.

[0016] Preferably, the thermal potential energy expression is constructed by combining the L1 norm and the L2 norm, and the logarithmic modulation is introduced to enhance the nonlinear modeling capability of the high response area, which can effectively reveal the local heat concentration trend of the building, further quantize the regional heat offset amplitude by the thermal slope, dynamically adjust the contribution of different regional features in the heat field modeling, highlight the dominant area which has directional control significance for the heat diffusion process, and finally embed the thermal potential energy value and the thermal slope into the gate shunt structure, realize directional decomposition of the double branch on the basis of maintaining global information, and make the model can capture the main heat flux and the residual heat dynamic effect, provide more physically reasonable and spatiotemporal robust features for the layout of the coil spacing.

[0017] Preferably, in the net zero energy building, the net zero energy building radiant heating coil spacing is coupled by multiple factors, resulting in complex nonlinear interaction and heat lag phenomenon between the main heat flux path and the residual heat channel, if only a single channel feature is used for prediction modeling, it is often difficult to fully describe the equilibrium state and non-equilibrium fluctuations in the heat diffusion process, therefore, the present application introduces a gate control mechanism, dynamically generates a response weight through the interaction intensity between the heat flow paths, and effectively expresses the regional heterogeneity and regulatory adaptability of the net zero energy building radiant heating coil spacing.

[0018] Preferably, in S5, the main heat flux feature and the residual heat feature are element-wise multiplied, and are compressed by a Sigmoid activation function to reflect the fusion response weight of the current net zero energy building under the heat flux path, and the specific mathematical model is as follows: ; wherein, is a fusion response weight generated by a Sigmoid activation function, is an element-wise multiplication, 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, and the specific mathematical model is as follows: ; wherein, is the final output weighted heat response feature.

[0019] Preferably, by constructing a fusion response weight based on element-level interaction and Sigmoid regulation, the fusion rule between the main heat flux feature and the residual heat feature can be effectively captured, which helps to represent the heat response weight deviation, dynamically adjust the expression strength of the main heat flux feature and the residual heat feature, highlight the heat structure response which has a key impact on the heating performance, and suppress the interference brought by non-dominant heat disturbance.

[0020] Preferably, in the design of net-zero energy buildings, the spacing of the radiant heating system coil directly determines the uniformity of heat distribution and the response efficiency of the heating system, so it is necessary to map the weighted heat response feature to a specific physical control quantity, i.e., to predict the optimal radiant heating system coil spacing that should be set.

[0021] Preferably, in S6, the weighted heat response feature is input into a nonlinear network to obtain a radiant heating coil spacing prediction value, so as to fit the nonlinear mapping relationship between the heat response feature and the coil spacing, and the specific mathematical model is as follows: ; wherein, is the net-zero energy building radiant heating coil spacing prediction value, is a learnable weight matrix, is a learnable bias term.

[0022] Compared with the prior art, the present application has the beneficial effects that: the present application constructs each net zero energy consumption building data as a multi-dimensional input sequence, uses a double-layer nonlinear transformation network for high-dimensional distinguishability projection to form a high-dimensional feature sequence, generates a building thermal disturbance feature sequence in combination with the overall distribution mean and standard deviation of the high-dimensional feature sequence, and introduces a thermal response significance weight value to realize weighted fusion and layer normalization processing of the building thermal disturbance feature sequence, obtain more representative fusion features, extract thermal potential energy values and thermal slopes based on the fusion features, realize effective separation of main heat flux features and residual heat features by embedding a nonlinear gate structure, comprehensively depict the dominant path and lag diffusion process of heat in the structure, and then generate fusion response weights by element-by-element interaction of the main heat flux features and the residual heat features, and weighted sum of the two types of features to form a unified weighted thermal response feature, thereby comprehensively expressing the multi-source heterogeneous characteristics in the building heat conduction process, and finally input the weighted thermal response feature into a nonlinear network to output a net zero energy consumption building radiant heating coil spacing prediction value, realizing intelligent prediction of coil spacing regulation parameters. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 It is a net zero energy consumption building radiant heating coil spacing prediction method steps diagram.

[0024] Figure 2 It is a building thermal disturbance perception module diagram.

[0025] Figure 3 It is a thermal potential shunt module diagram.

[0026] Figure 4 It is a model training process loss curve diagram.

[0027] Figure 5 It is a net zero energy consumption building radiant heating coil spacing prediction effect diagram. DETAILED DESCRIPTION

[0028] The present application provides a kind of net zero energy consumption building radiant heating coil spacing prediction method, this method utilizes double-layer nonlinear transformation network to each net zero energy consumption building data multidimensional input sequence High-dimensional distinguishability projection is carried out, obtains high-dimensional characteristic sequence, combines the mean and standard deviation of high-dimensional characteristics Building thermal disturbance characteristic sequence is generated, heat response significant weight value is introduced to building thermal disturbance characteristic sequence Weighted fusion is carried out, and the stability of feature is improved through layer normalization Formed fusion feature, further extract heat potential value and heat slope, and embed in nonlinear gate structure together with fusion feature Main heat flux feature and residual heat feature are separated out, based on the element-by-element interaction of main heat flux feature and residual heat feature Fusion response weight is generated, and main heat flux feature and residual heat feature are weighted and summed, to obtain weighted heat response feature, finally the weighted heat response feature is input into nonlinear network, and the net zero energy consumption building radiant heating coil spacing prediction value is output, realize the fusion perception and accurate modeling of multi-source information, improve the prediction accuracy of coil spacing in different building scenes, the technical solutions in the embodiments of the present application will be described in detail and completely below, specifically including the following steps, as shown in Figure 1 .

[0029] S1, collect the data affecting the net zero energy consumption building radiant heating coil spacing, and construct the net zero energy consumption building radiant heating coil spacing dataset.

[0030] Further, in S1, 150 data affecting the radiant heating coil spacing of net zero energy consumption building are collected, and the data types include building type, building area, building height, indoor target temperature set value, water flow, water pressure, construction material, wall thermal conductivity, and the net zero energy consumption building radiant heating coil spacing dataset is constructed, and the dataset is divided into training set and validation set according to the ratio of 7:3, for subsequent model training and evaluation.

[0031] S2, each net zero energy consumption building data in the dataset is constructed as a multidimensional input sequence, and double-layer nonlinear transformation network is used to project the multidimensional input sequence High-dimensional distinguishability, and high-dimensional characteristic sequence is obtained after being processed by ReLU and GELU activation function in turn.

[0032] Further, in S2, each net zero energy consumption building data in the net zero energy consumption building radiant heating coil spacing dataset is constructed as a multidimensional input sequence , , wherein is the collection value of the first type data in the net zero energy consumption building data, is the total number of collected net zero energy consumption building data types, and double-layer nonlinear transformation network is used to project High-dimensional distinguishability, so that different data type characteristics have stronger distinguishability, and the specific mathematical model is as follows: ; Where, is a high-dimensional feature sequence obtained through a nonlinear transformation network, ,in It is the first net zero energy building in the world High-dimensional features of type data, is the data dimension of the high-dimensional feature sequence, 、 is the learnable weight matrix, 、 is a learnable bias.

[0033] 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 the thermal response significance weight value, perform weighted fusion on the building thermal disturbance feature sequence, and obtain the fusion feature after layer normalization processing.

[0034] Furthermore, in S3, a building thermal disturbance perception module is constructed, such as Figure 2 As shown, the specific steps include: S31, high-dimensional feature sequence and The mean and standard deviation in the overall distribution are combined to form the characteristic sequence of building thermal disturbance. The specific mathematical model is: ; Where, is the characteristic sequence of building thermal disturbance, is a sign function that maps positive numbers to 1, negative numbers to -1, and 0 to 0. is element-wise multiplication, for The mean of , for The standard deviation of , It is a very small positive number to prevent division by zero. The initial value is set to 0.00001 during implementation. To take the absolute value operation; S32. Normalize each feature dimension in the building thermal disturbance feature sequence, calculate the average value after square enhancement, and generate the thermal response significance weight value. The specific mathematical model is: ; Where, For the The significance weight of the thermal response of the radiant heating coil spacing data of a net zero energy building, 、 They are building thermal disturbance characteristic sequences A method for determining a feature of a net-zero energy building radiant heating coil spacing data A method for determining a feature of a net-zero energy building radiant heating coil spacing data And the characteristic value of the first dimension, The feature dimension of the building thermal disturbance feature sequence ; S33, the building thermal disturbance feature sequence is weighted and fused by using the thermal response significance weight value, and after layer normalization processing, the fusion feature is formed, the main feature is highlighted, and the influence of abnormal features is smoothed, and the specific mathematical model is: ; In the formula, The fusion feature, Layer normalization, The total number of collected net-zero energy building data types, The feature of the building thermal disturbance feature sequence A method for determining a feature of a net-zero energy building radiant heating coil spacing data A method for determining a feature of a net-zero energy building radiant heating coil spacing data

[0035] S4, a thermal potential shunt module is constructed, the thermal potential value and the thermal slope are extracted based on the fusion feature, the fusion feature, the thermal potential value and the thermal slope are embedded in the nonlinear gate structure, the main heat flow feature is extracted, and the residual heat feature is calculated based on the main heat flow feature.

[0036] Further, in S4, the thermal potential shunt module is constructed, as shown in Figure 3 The following steps are included: S41, the fusion feature is combined with a double norm function and a logarithmic modulation term to form a thermal potential value, so as to identify a high heat flow area in the heating structure and highlight the response ability to local heat aggregation effect, and the specific mathematical model is: ; In the formula, The thermal potential value, The L2 energy attenuation weight, the initial value is set to 0.05 in the implementation process, The L1 distribution response modulation weight, the initial value is set to 0.1 in the implementation process, Natural exponential function, Logarithmic function with base e, L2 norm, L1 norm; S42, the mean vector of the building thermal disturbance feature sequence The square of the L2 norm of the difference between the thermal potential value And the mean vector is the numerator, and the thermal potential value The ratio of the L2 norm of itself forms the heat slope, and the specific mathematical model is: ; ; In the formula, is the mean vector of the building heat disturbance feature sequence is the heat slope, is a very small positive number, and the initial value is set to 0.00001 in the implementation process; S43, embed the fusion features, heat potential value and heat slope into the nonlinear gate structure, extract the main heat flux feature, further utilize the main heat flux feature to obtain the residual heat feature, realize the directed separation modeling of the heat flow path, and the specific mathematical model is: ; ; In the formula, is the main heat flux feature, is a learnable weight matrix, is a learnable bias, is a splicing operation, is the residual heat feature, is an element-wise multiplication.

[0037] S5, generate a fusion response weight by using element-wise multiplication to combine the main heat flux feature and the residual heat feature, and perform weighted summation on the two types of features to obtain a weighted heat response feature.

[0038] Further, in S5, the main heat flux feature and the residual heat feature are multiplied element by element, and are compressed through a Sigmoid activation function to reflect the fusion response weight of the current net zero energy consumption building under the heat flux path. The specific mathematical model is as follows: ; In the formula, is the fusion response weight generated through the Sigmoid activation function, is an element-wise multiplication, 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. The specific mathematical model is as follows: ; In the formula, is the final output weighted heat response feature.

[0039] S6, input the weighted heat response feature into a nonlinear network to output a net zero energy consumption building radiant heating coil spacing prediction value.

[0040] ​Further, in S6, the weighted thermal response features are input into a nonlinear network to obtain a radiant heating coil spacing prediction value to fit the nonlinear mapping relationship between the thermal response features and the coil spacing, and the specific mathematical model is: ; In the formula, is a radiant heating coil spacing prediction value of a net zero energy consumption building, is a learnable weight matrix, is a learnable bias term.

[0041] Further, the model is implemented by using Python 3.10 programming language, is constructed and trained based on a PyTorch framework, and is deployed in a GPU acceleration environment supporting CUDA 12.1. In the model training process, an NVIDIA 3090 24GB GPU card is selected for parallel computing. In terms of optimization strategy, a Lookahead combined optimizer is adopted, and a learning rate preheating mechanism is introduced at the early stage of training. The basic learning rate is 0.0005, a CosineAnnealing learning rate scheduler is used for dynamic adjustment to avoid local optimal trap, the batch size is 8, and a weighted smooth L1 loss function is used.

[0042] Further, the loss curve of the model training process of the method is shown in Figure 4 , wherein the abscissa represents the training round, and the ordinate represents the loss value. It can be observed from the figure that as the training round increases, the loss value shows a continuous downward trend and gradually tends to be stable, indicating that the model has good convergence and training stability. The prediction effect of the method is shown in Figure 5 , wherein the abscissa is the number of net zero energy consumption buildings, and the ordinate is the corresponding radiant heating coil spacing value (millimeter). The gray dotted line in the figure represents the actual evaluation value, and the black solid line represents the prediction value output by the method. As can be seen from Figure 5 , the prediction value and the actual evaluation value are highly consistent in the overall trend, with small fluctuation error, fully verifying the accuracy and reliability of the method in the radiant heating coil spacing prediction task of the net zero energy consumption building.

[0043] The above is only a preferred embodiment of the present application, and it should be noted that for ordinary skilled persons in the art, without departing from the inventive concept, several modifications and improvements can be made, which are all within the protection scope of the present application.

Claims

1. A method for predicting the spacing of radiant heating coils in a net zero energy building, characterized in that: The following steps are involved: S1. Collect data that affects the spacing of radiant heating coils in net-zero energy buildings and construct a dataset of spacing of radiant heating coils in net-zero energy buildings. S2. Each net-zero energy building data in the dataset is constructed as a multi-dimensional input sequence, a two-layer nonlinear transformation network is used to perform high-dimensional distinguishable projection on the multi-dimensional input sequence, and a high-dimensional feature sequence is obtained after sequentially processing it through ReLU and GELU activation functions; S3. Constructing a building thermal disturbance perception module, combining the high-dimensional feature sequence with its overall distribution mean and standard deviation to generate a building thermal disturbance feature sequence, introducing a thermal response significance weight value, performing weighted fusion on the building thermal disturbance feature sequence, and obtaining a fusion feature after layer normalization processing; S4. Constructing a thermal potential diversion module, extracting thermal potential energy values ​​and thermal slopes based on the fusion features, embedding the fusion features, thermal potential energy values, and thermal slopes into a nonlinear gating structure, extracting main heat flux features, and calculating residual heat features based on the main heat flux features; S5. Use element-by-element multiplication to combine the main heat flux feature and the residual heat feature to generate a fusion response weight, perform a weighted summation on the two types of features, and obtain a weighted thermal response feature; S6. Input the weighted thermal response characteristics into a nonlinear network, and output a predicted value of the radiant heating coil spacing in a net zero energy building.

2. A method for predicting the spacing of radiant heating coils in a net zero energy building according to claim 1, characterized in that: Data affecting the radiant heating coil spacing in net-zero energy buildings were collected, including building type, building area, building height, indoor target temperature setpoint, water flow, water pressure, construction materials, and wall thermal conductivity, to construct a dataset of radiant heating coil spacing in net-zero energy buildings.

3. A method for predicting the spacing between radiant heating coils in a net zero energy building according to claim 2, characterized in that: Each net-zero energy building data in the net-zero energy building radiant heating coil spacing dataset is constructed as a multidimensional input sequence , using a two-layer nonlinear transformation network for high-dimensional discriminable projection, specifically, using the weight matrix and bias term Perform linear transformation, use ReLU activation function to perform first layer nonlinear activation, then linearly map the activated result with weight matrix and bias term, and process it with GELU activation function to obtain high-dimensional feature sequence .

4. A method for predicting the spacing between radiant heating coils in a net zero energy building according to claim 3, characterized in that: The high-dimensional feature sequence and The mean and standard deviation in the overall distribution are combined to form the characteristic sequence of building thermal disturbance. The specific mathematical model is: ； Where, is the characteristic sequence of building thermal disturbance, is a symbolic function, is element-wise multiplication, for The mean of for The standard deviation of is a very small positive number, It is an absolute value operation.

5. A method for predicting the spacing between radiant heating coils in a net zero energy building according to claim 4, characterized in that: Each feature dimension in the building thermal disturbance feature sequence is normalized, and the mean value is calculated after square enhancement to generate the thermal response significance weight value. The specific mathematical model is: ； Where, For the The significance weight of the thermal response of the radiant heating coil spacing data of a net zero energy building, 、 They are building thermal disturbance characteristic sequences Middle Data on Radiant Heating Coil Spacing for Net Zero Energy Buildings Hedi The eigenvalues ​​of the dimensions, is the characteristic sequence of building thermal disturbance characteristic dimensions; The thermal response significance weight value is used to perform weighted fusion on the building thermal disturbance feature sequence, and after layer normalization, a fusion feature is formed. .

6. A method for predicting the spacing of radiant heating coils in a net zero energy building according to claim 5, characterized in that: The fusion features are combined with the dual norm function and the logarithmic modulation term to form the thermal potential energy value. The specific mathematical model is: ； Where, is the thermal potential energy value, is the L2 energy decay weight, is the L1 distribution response modulation weight, is the natural exponential function, is the logarithmic function with base e, is the L2 norm, is the L1 norm; Constructing a characteristic sequence of building thermal disturbances The mean vector of thermal potential energy is The square of the L2 norm of the difference between the vector and the mean is the numerator, combined with the thermal potential energy value The L2 norm itself constitutes a ratio, forming a thermal slope ; The fusion features, thermal potential energy values ​​and thermal slopes are spliced ​​and embedded into the nonlinear gating structure to generate the main heat flux features. , using the main heat flux feature in The complementary values ​​within the value range are multiplied element-by-element with the fusion features to generate residual heat features. .

7. A method for predicting the spacing of radiant heating coils in a net zero energy building according to claim 6, characterized in that: The main heat flux feature and the residual heat feature are multiplied element by element, and compressed into a fusion response weight through the Sigmoid activation function. Based on the fusion response weight, the main heat flux feature and the residual heat feature are weighted and summed to form a weighted thermal response feature. .

8. A method for predicting the spacing of radiant heating coils in a net zero energy building according to claim 7, characterized in that: The weighted thermal response characteristics are fed into a nonlinear network to predict the spacing between radiant heating coils. .

Images