Intelligent evaluation method for application value of green building BIM model

Abstract

The application relates to the technical field of value evaluation, and discloses an application value intelligent evaluation method for a green building BIM model, which comprises the following steps: constructing an observation sequence and identifying the observation sequence by taking an indoor-outdoor temperature difference as an excitation and a heat flow mean value as a response to obtain a component heat response feature vector, and constructing a heterogeneous graph by taking a component as a node, a topology as an edge, a cosine similarity as an edge weight, and a thermal degradation index as a node attribute; weighting and aggregating initial feature values of adjacent nodes to obtain a physical estimated value, and multiplying the physical estimated value by an amplitude ratio to obtain a drift estimated value; superimposing a static semantic vector to obtain a correction embedded vector, generating a dynamic mapping table according to minimum standard Euclidean distance matching target mapping labels, and taking the minimum standard Euclidean distance as a mapping residual error; obtaining a comprehensive credibility score according to a consistency ratio, a cumulative distribution and an amplitude weight, updating the mapping table according to the score, and outputting; and the application can improve the dynamic characterization capability and mapping credibility of the green building BIM model for the real-time thermal performance of components.

Description

Technical Field

[0001] This invention relates to the field of value assessment technology, and in particular to an intelligent assessment method for the application value of a green building BIM model. Background Technology

[0002] In the evaluation of the application value of green building BIM models during the operation phase, the semantic tags of components are derived from static attribute data in the design phase and remain unchanged during subsequent operation and maintenance. However, the thermal performance of the building envelope degrades with the increase of service life, resulting in a discrepancy between the evaluation results based on the semantic tags in the design phase and the actual operating performance of the building. To address this discrepancy, some existing methods add monitoring data fields to the BIM model to record changes in component performance. However, these monitoring data fields are independent of the original semantic tag system, and there is a lack of dynamic mapping between the measured physical parameters during the operation period and the BIM semantic space. The updating of evaluation information relies on manual offline processing, which limits the timeliness of BIM models in supporting decision-making during the building operation phase.

[0003] Furthermore, existing methods typically employ uniform conversion rules when transforming measured physical properties of components into evaluation indicators, failing to adjust the conversion results based on differences in the thermal inertia of the components. Lightweight and heavyweight components exhibit different thermal response amplitudes under the same environmental excitation, and the same measured temperature difference corresponds to different degrees of degradation. Using uniform conversion rules leads to the reduction or smoothing out of the actual degradation differences between the two types of components in the evaluation results, reducing the discriminative power of the evaluation results. Simultaneously, after generating the semantic mapping table, existing methods directly input the mapping results into the evaluation process, lacking a multi-dimensional quality verification step. When sensor data is abnormal or identification parameters deviate, erroneous mapping results cannot be automatically identified and corrected during the process, resulting in a lack of traceable quality evidence for the reliability of the evaluation results. Therefore, there is an urgent need to develop an intelligent evaluation method for the application value of green building BIM models that can adaptively update the semantics of BIM components based on measured physical data during the operational period and possess a self-checking mechanism for the reliability of mapping results. This would improve the consistency between the evaluation results and the actual operating status of the building, as well as the robustness of the evaluation system. Summary of the Invention

[0004] This invention provides an intelligent evaluation method for the application value of green building BIM models to solve the problems mentioned in the background.

[0005] To achieve the above objectives, this invention provides an intelligent evaluation method for the application value of green building BIM models, comprising:

[0006] S1. Construct an observation sequence with the real-time indoor and outdoor temperature difference of the building as the excitation and the average heat flow as the response. Perform parameter identification on the observation sequence to obtain the thermal response feature vector of the component in BIM. With the component as the node, the topological connection relationship of the component as the edge, the cosine similarity between the corresponding thermal response feature vectors of the component as the edge weight, and the thermal degradation index of the component as the node attribute, construct a heterogeneous graph.

[0007] S2. The initial features of the nodes adjacent to each node in the heterogeneous graph are weighted and aggregated to obtain the physical prediction value of the adjacent nodes. The physical prediction value is multiplied by the amplitude ratio in the thermal response feature vector to obtain the drift prediction value.

[0008] S3. The drift prediction value is superimposed with the static semantic vector of the component as a correction embedding vector. The label corresponding to the minimum standard Euclidean distance between the correction embedding vector and the pre-stored ordered continuous semantic label vector set of the target platform is the target mapping label. A dynamic mapping table is generated, and the minimum standard Euclidean distance is used as the mapping residual.

[0009] S4. The comprehensive credibility score is obtained by weighted summation of the consistency ratio between the target mapping label and the target platform historical label, the cumulative distribution of the mapping residual, and the magnitude of the drift estimate relative to the physical estimate.

[0010] S5. Based on the comprehensive credibility score, generate an updated dynamic mapping table and output it to the target platform.

[0011] Compared with the prior art, the present invention has the following beneficial effects:

[0012] 1. In the existing methods for assessing the application value of green building BIM models, the semantic labels of components remain unchanged after being set in the subsequent operation phase. However, the thermal performance of the building envelope continues to degrade with the increase of service life. Static semantic labels cannot reflect the current real physical state of the components, resulting in a systematic deviation between the assessment results and the actual operating performance of the building. Even if some existing methods attempt to add monitoring data fields to the BIM model, the added data is independent of the original semantic label system. There is a lack of an adaptive update mechanism that dynamically maps the measured physical performance to the semantic space, making it difficult to achieve an automated closed loop from physical monitoring during the operation period to BIM semantic updates. This invention uses a one-dimensional transient heat conduction analytical solution to identify parameters of the measured temperature difference and heat flow observation sequence during the operation period, extracts personalized thermal response feature vectors for each component containing time constants and amplitude ratios, and uses the cosine similarity between the feature vectors as the edge weight of the heterogeneous graph and the amplitude ratio as the physical modulation coefficient of the weighted aggregation result. This makes the semantic drift amplitude of the component directly constrained by its own thermal inertia characteristics. Different thermal inert components exhibit differentiated semantic update rates due to differences in amplitude ratios, thereby realizing the adaptive dynamic representation of BIM component semantics for the degradation of thermal performance during the operation period, and solving the problem that static semantic labels cannot be updated synchronously with the degradation of physical performance.

[0013] 2. Existing methods typically output the semantic mapping table directly to the target platform after generation, lacking a multi-dimensional quality verification step for the mapping results. When sensor data is abnormal or parameter identification deviates, erroneous mapping results cannot be automatically identified and corrected, but instead directly participate in the subsequent value assessment process, resulting in a lack of traceable quality basis for the credibility of the assessment conclusions. This invention records the minimum Euclidean distance between the corrected embedding vector and the semantic label vector of the target platform as the mapping residual, and integrates three dimensions—the consistency ratio between the target mapping label and historical labels, the cumulative distribution of the mapping residual, and the magnitude of the drift estimate relative to the physical estimate—for weighted summation to obtain a comprehensive credibility score. The score quantifies the quality boundary of the mapping results from three aspects: temporal stability, matching confidence, and physical modulation intensity. Based on the score, qualified entries are dynamically selected to generate an updated mapping table, realizing closed-loop quality control from one-way mapping output to quantitative assessment of mapping credibility and anomaly self-checking, solving the problem that erroneous mapping results cannot be effectively identified and corrected. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating an intelligent evaluation method for the application value of a green building BIM model, provided as an embodiment of the present invention.

[0015] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0016] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0017] This application provides an intelligent assessment method for the application value of a green building BIM model. The executing entity of this intelligent assessment method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the intelligent assessment method for the application value of a green building BIM model can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0018] Reference Figure 1 The diagram shown is a flowchart illustrating an intelligent evaluation method for the application value of a green building BIM model according to an embodiment of the present invention. In this embodiment, the intelligent evaluation method for the application value of a green building BIM model includes:

[0019] S1. Construct an observation sequence with the real-time indoor and outdoor temperature difference of the building as the excitation and the average heat flow as the response. Perform parameter identification on the observation sequence to obtain the thermal response feature vector of the component in BIM. With the component as the node, the topological connection relationship of the component as the edge, the cosine similarity between the corresponding thermal response feature vectors of the component as the edge weight, and the thermal degradation index of the component as the node attribute, construct a heterogeneous graph.

[0020] In this embodiment of the invention, surface temperature data is collected by an array of temperature sensors arranged on both the inner and outer sides of the building's exterior wall. Multiple temperature sensors are deployed on each side of the surface, and the temperature values ​​of multiple sensors on the same side at the same time are arithmetically averaged to obtain a representative temperature value for that side surface. The difference between the representative temperature value on the outdoor side and the representative temperature value on the indoor side is calculated, and the result is used as the real-time temperature difference at that time. Simultaneously, a heat flow sensor attached to the central area of ​​the inner surface of the exterior wall collects heat flow signals. After signal conditioning and analog-to-digital conversion, a digital form of heat flow density value is obtained as real-time heat flow data. The synchronous sampling frequency of the temperature sensor and the heat flow sensor is once every 10 minutes, and continuous collection is carried out for no less than 72 hours to ensure that the observation sequence covers the complete diurnal temperature fluctuation cycle.

[0021] Wavelet transform denoising was performed on the real-time temperature difference data and real-time heat flow data respectively. The db4 wavelet from the Daubechies wavelet family was selected as the mother wavelet, and the original data sequence was decomposed into five layers of wavelets to obtain the approximation coefficients and detail coefficients of each layer. The detail coefficients were processed using a soft thresholding method, with the threshold being the product of the median absolute value of the detail coefficients of the layer and a preset constant of 0.6745. The thresholded detail coefficients and approximation coefficients were then reconstructed using wavelets to obtain the denoised effective temperature difference data and effective heat flow data.

[0022] An observation sequence is constructed based on the denoised effective temperature difference data and effective heat flux data; the cross-correlation function between the effective temperature difference data sequence and the effective heat flux data sequence is calculated, and the shift corresponding to the maximum value of the cross-correlation function is identified as the thermal response lag time. The effective heat flux data sequence is shifted backward on the time axis by the thermal response lag time so that the phase of the shifted effective heat flux data is aligned with the phase of the effective temperature difference data; the phase-aligned effective temperature difference data and effective heat flux data are paired and arranged in ascending order of time, and the paired data at each moment is taken as an observation sample. All observation samples constitute the observation sequence of the component.

[0023] The observation sequence is parameter-identified based on the one-dimensional transient heat conduction analytical solution; a theoretical heat transfer model is established with thermal diffusivity and thermal resistance as the parameters to be identified, and the real-time temperature difference in the observation sequence is used as the excitation input to the theoretical heat transfer model to obtain a theoretical heat flux data sequence; the difference between the theoretical heat flux data and the real-time heat flux data at the same moment in the observation sequence is calculated, and the difference is used as the residual at that moment. After traversing all moments, a residual sequence is obtained; when the absolute value of the real-time temperature difference at a certain moment in the observation sequence is less than a preset environmental noise threshold of 0.5°C, the observation sample at that moment is marked as a low-excitation sample, and the result is processed according to the formula... Calculate the weight coefficients of the sample, where For real-time temperature difference, Pick C, the weight coefficient of normal samples is always 1.0.

[0024] A recursive least squares algorithm with a forgetting factor is used to iteratively optimize the thermal diffusivity and thermal resistance. The forgetting factor is set to 0.98 to ensure that the algorithm has a moderate dependence on historical data, taking into account both the slow time-varying characteristics of thermal parameters and the response speed to recent changes. In each iteration, the weight coefficient of each sample is multiplied by the corresponding weighting coefficient in the recursive least squares algorithm, so that the contribution of low-excitation samples to parameter correction is reduced. The iteration is repeated until the relative change of the weighted sum of squares of the residual sequence in multiple consecutive iterations is less than the preset convergence threshold of 0.1%, at which point the iteration is terminated. The thermal diffusivity and thermal resistance in the converged state are confirmed as the thermal parameters of the component. The geometric parameters of the component, including thickness, area and volume, are read from the BIM database, and the geometric parameters and thermal parameters are vector-concatenated to obtain the thermal response feature vector of the component.

[0025] Using the components as nodes and the topological connections between components as edges, a heterogeneous graph is constructed. The spatial bounding box coordinates and physical connection port data of each component in the BIM database are read. Based on the spatial bounding box coordinates, it is determined whether there is a coplanar contact area between two components. Based on the physical connection port data, it is determined whether there is an energy transfer path between two components. When both conditions are met, the connection between the two components is determined to be a topological connection. The direction of the topological connection edge is determined by the gradient direction of the real-time indoor-outdoor temperature difference: when the indoor temperature is higher than the outdoor temperature, the direction of the topological connection edge is from the indoor component to the outdoor component; when the outdoor temperature is higher than the indoor temperature, the direction of the topological connection edge is from the outdoor component to the indoor component. Each pair of components determined to have a topological connection is set as an edge in the heterogeneous graph. For each edge, the cosine similarity between the thermal response feature vectors of the two components connected by the edge is calculated, and the cosine similarity is used as the edge weight.

[0026] The service life of a component is calculated based on its construction year, using the difference between the current year and the construction year as the service life. The material aging coefficient is read from the material attribute field of the BIM database. An aging degradation coefficient is obtained by exponentially multiplying the negative first power of the natural constant e by the product of the service life and the preset annual degradation rate. The material aging coefficient and the aging degradation coefficient are multiplied to obtain a comprehensive degradation factor. The comprehensive degradation factor is subtracted from one to obtain the thermal degradation index of the component. The thermal degradation index is confirmed as a node attribute of the component. The component is at least divided into an outer enclosure component and an inner enclosure component. The outer enclosure component connects to outdoor environment nodes and indoor space nodes on both sides, respectively. The inner enclosure component connects to different indoor space nodes on both sides. The value range of node attributes corresponding to different component types is normalized to different numerical intervals according to the component type. Thus, a heterogeneous graph is constructed using all components as the node set, all topological connection edges as the edge set, cosine similarity as the edge weight, and the thermal degradation index as the node attribute.

[0027] S2. The initial features of the nodes adjacent to each node in the heterogeneous graph are weighted and aggregated to obtain the physical prediction value of the adjacent nodes. The physical prediction value is multiplied by the amplitude ratio in the thermal response feature vector to obtain the drift prediction value.

[0028] In this embodiment of the invention, each node in the heterogeneous graph is traversed, and the node currently being processed is recorded as the current node; all adjacent nodes directly connected to the current node in the heterogeneous graph through topological connection edges are retrieved; the thermal parameter components in the thermal response feature vectors corresponding to the current node and the adjacent nodes are read, and the thermal parameter components specifically include the thermal diffusivity coefficient and thermal resistance; the thermal parameter components are normalized, and the normalized thermal parameter components are used as the initial features of the node.

[0029] In the weighted aggregation operation, a multi-head graph attention mechanism is used to calculate the aggregation weight coefficients. Specifically, for each neighboring node of the current node, the initial features of the current node and the initial features of the neighboring node are concatenated. The concatenated vector is mapped to a scalar space through a linear transformation layer. The output value of the scalar space is processed by a nonlinear activation function and used as the attention coefficient of the current node to the neighboring node. The above calculation is performed in parallel using eight attention heads, and each attention head independently outputs a set of attention coefficients. The arithmetic mean of the attention coefficients output by the eight attention heads corresponding to the same neighboring node is taken, and the arithmetic mean is used as the aggregation weight coefficient of the neighboring node relative to the current node.

[0030] The initial features of each neighboring node of the current node are multiplied by their respective aggregation weight coefficients to obtain the weighted initial features of each neighboring node. The weighted initial features of each neighboring node are then summed element-wise according to the corresponding dimension of the vector to obtain an aggregation vector. The mean and variance regularization of each dimension component in the aggregation vector are performed, and the mean and standard deviation of the aggregation vector under the condition that all neighboring nodes participate in the aggregation are calculated. The mean is subtracted from each dimension component of the aggregation vector and then divided by the standard deviation to obtain the aggregation feature vector after mean and variance regularization. The aggregation feature vector eliminates the dimensional shift introduced by the difference in the number of neighboring nodes of different nodes. The dimension components corresponding to thermal resistance and thermal diffusivity in the aggregation feature vector are extracted as the physical estimate of the neighboring nodes of the current node. The physical estimate represents the theoretical thermal response benchmark value of the neighboring nodes of the current node in the non-degenerate state.

[0031] Based on the thermal response feature vector of the current node, the amplitude amplification factor is read as the current amplitude component. The amplitude amplification factor characterizes the steady-state heat flow output capability of the component under unit temperature difference excitation and is derived from the transfer function gain term obtained in the aforementioned parameter identification process. The reference amplitude amplification factor of the component corresponding to the current node in the initial state of completion is read from the BIM database as the reference amplitude component. The reference amplitude is derived from the marked value in the component's factory thermal performance test report. If the reference amplitude component is missing in the BIM database, a cold start initialization operation is performed to replace the reference amplitude component with the typical value of the component's material type in the national standard building materials thermal database.

[0032] Traverse each neighboring node of the current node, read the amplitude amplification coefficient in the thermal response feature vector of the neighboring node as the neighboring amplitude component; divide the current amplitude component by each neighboring amplitude component, and use the result of the division operation as the dynamic amplitude scaling factor corresponding to the neighboring node; multiply each dimension component in the physical prediction value by the dynamic amplitude scaling factor of the corresponding neighboring node dimension by dimension, and use the result of the dimension-by-dimensional multiplication as the initial drift prediction value.

[0033] Perform a physical boundary truncation operation on the initial drift estimate; read the preset physical boundary values ​​of thermal diffusivity and thermal resistance, wherein the physical boundary value of thermal diffusivity includes the lower limit value of thermal diffusivity. m² / s and upper limit of thermal diffusivity The thermal resistance physical boundary values ​​include a lower limit of 0.1 (m²·K) / W and an upper limit of 10 (m²·K) / W. These boundary values ​​are derived from the limit ranges of commonly used building materials in the "Code for Thermal Design of Civil Buildings". The thermal diffusivity component in the initial drift estimate is compared with the lower and upper limits of the thermal diffusivity. If the thermal diffusivity component is less than the lower limit, then the component is assigned the lower limit value; if the thermal diffusivity component is greater than the upper limit value, then the component is assigned the lower limit value. The component is assigned the upper limit value of the thermal diffusivity coefficient; the thermal resistance component in the initial drift estimate is compared with the lower limit value and the upper limit value of thermal resistance. If the thermal resistance component is less than the lower limit value, the component is assigned the lower limit value of thermal resistance; if the thermal resistance component is greater than the upper limit value of thermal resistance, the component is assigned the upper limit value of thermal resistance. Through the above physical boundary truncation operation, the drift estimate of the current node is obtained. The drift estimate retains the differentiated drift rate caused by the difference in amplitude of different thermally inert components, and maintains the stability of the value within a physically reasonable range.

[0034] S3. The drift prediction value is superimposed with the static semantic vector of the component as a correction embedding vector. The label corresponding to the minimum standard Euclidean distance between the correction embedding vector and the pre-stored ordered continuous semantic label vector set of the target platform is the target mapping label. A dynamic mapping table is generated, and the minimum standard Euclidean distance is used as the mapping residual.

[0035] In this embodiment of the invention, the static semantic attributes of the current component are read from the BIM database. The static semantic attributes include component type, material category, thermal insulation level, and fire resistance level. Each attribute category in the static semantic attributes is thermally encoded, and the thermally encoded vectors corresponding to each attribute category are concatenated according to a preset attribute order to obtain the static semantic vector of the component.

[0036] The system checks whether the initial as-built baseline amplitude amplification factor for the current component exists in the BIM database. If the baseline amplitude amplification factor is missing from the BIM database, the system enters the cold start initialization process. In the cold start initialization process, the thermal degradation index of the current component is reset to zero, and the baseline amplitude component is replaced with the standard amplitude amplification factor corresponding to the material type of the component. The standard amplitude amplification factor is read from a preset material standard parameter library. At the same time, the mapping residual corresponding to the cold start initialization operation is multiplied by a penalty coefficient of 2.0 to characterize the additional uncertainty of the mapping result in the cold start state. This artificially lowers the overall credibility score of the component in the subsequent credibility assessment, preventing misjudgment due to the use of standard values ​​to replace real BIM data.

[0037] The drift prediction value and the static semantic vector are superimposed to achieve the fusion of physical drift information and semantic category information. When the dimension of the drift prediction value is the same as the dimension of the static semantic vector, the superposition is performed by element-wise addition. When the dimension of the drift prediction value is different from the dimension of the static semantic vector, the drift prediction value is first input into a linear projection layer. The linear projection layer maps the dimension of the drift prediction value to the same dimension as the static semantic vector, and then the elements are added together to obtain the corrected embedding vector. The corrected embedding vector contains both the dynamic thermal drift information and the static semantic category information of the component, providing a joint representation for subsequent label matching.

[0038] Read a pre-stored ordered continuous semantic tag vector set from the target platform. The ordered continuous semantic tag vector set contains multiple semantic tag vectors, each of which corresponds to an energy efficiency level. The semantic tag vectors are arranged continuously from high to low according to the energy efficiency level. The energy efficiency levels include ten levels from G1 to G10, and each level corresponds to a unique platform-side status tag. Calculate the standard Euclidean distance between the corrected embedding vector and each semantic tag vector in the ordered continuous semantic tag vector set. In the calculation of the standard Euclidean distance, divide each dimension component of the corrected embedding vector and the semantic tag vector by the global standard deviation corresponding to that dimension to eliminate the influence of differences in the dimensions on the distance measurement.

[0039] The minimum value is selected from all calculated standard Euclidean distances, and the energy efficiency level label corresponding to the minimum value is identified as the target mapping label of the current component. The target mapping label is the energy efficiency level label of the component in the target platform. The mapping relationship between the target mapping label and the globally unique identifier of the current component is recorded and written into the dynamic mapping table. The minimum standard Euclidean distance is used as the mapping residual of the current component. The mapping residual is used for subsequent confidence evaluation. The smaller the value, the higher the matching degree between the corrected embedding vector and the target label vector.

[0040] S4. The comprehensive credibility score is obtained by weighted summation of the consistency ratio between the target mapping label and the target platform historical label, the cumulative distribution of the mapping residual, and the magnitude of the drift estimate relative to the physical estimate.

[0041] In this embodiment of the invention, based on the generated dynamic mapping table, the historical mapping record library stored in the target platform is retrieved; the historical mapping record library contains every mapping record of the current component in the past several consecutive mapping cycles, and each mapping record contains the target mapping label and mapping residual corresponding to the mapping cycle.

[0042] In calculating the consistency ratio, the number of mappings with the target mapping tag within the past N consecutive time windows is retrieved from the target platform's historical mapping record library, and the total number of mapping records within the past N consecutive time windows is also retrieved. The number of mappings with the target mapping tag is divided by the total number of mapping records, and the result of the division is used as the consistency ratio of the component. Here, N is the number of mapping periods contained in the sliding time window, each mapping period corresponds to one complete data acquisition and mapping update, and the period duration is 24 hours by default. For example, the number of mappings with the target mapping tag within the past 20 consecutive time windows is retrieved from the target platform's historical mapping record library, and the total number of mapping records within the past 20 consecutive time windows is also retrieved. The number of mappings with the target mapping tag is divided by the total number of mapping records, and the result of the division is used as the consistency ratio of the component. The consistency ratio reflects the stability of the current mapping tag in the historical time series; a higher consistency ratio indicates a higher frequency of recurrence of the current target mapping tag in historical mappings.

[0043] In the calculation of the cumulative distribution, all mapping residuals of the current component over multiple consecutive mapping periods are read, and all mapping residuals are arranged in ascending order of value to form an ordered residual sequence. The count corresponding to the position index of each residual value in the ordered residual sequence is taken as the number of residuals less than or equal to the residual value. The number of residuals is divided by the total number of residuals in the ordered residual sequence to obtain the empirical cumulative probability corresponding to each residual value. The mapping residual obtained in the current mapping period is compared with the residual values ​​in the ordered residual sequence. If the current mapping residual is equal to... The cumulative distribution function value of the current mapped residual is taken as the empirical cumulative probability corresponding to the position of the residual value at a certain position in the ordered residual sequence. If the current mapped residual is between two adjacent residual values ​​in the ordered residual sequence, the empirical cumulative probability corresponding to the smaller adjacent residual value is taken as the cumulative distribution function value of the current mapped residual. The cumulative distribution function value represents the relative position of the current mapped residual in the historical residual distribution. The smaller the current mapped residual is relative to the historical residual, the lower the cumulative distribution function value, indicating that the matching confidence of the current mapping is higher.

[0044] In the amplitude calculation, the components of the drift prediction and the physical prediction of the current node are read in each dimension; the sum of squares of the components of the drift prediction is calculated, and the square root of the sum of squares is taken to obtain the vector norm of the drift prediction; the sum of squares of the components of the physical prediction is calculated, and the square root of the sum of squares is taken to obtain the vector norm of the physical prediction; the vector norm of the drift prediction is divided by the vector norm of the physical prediction, and the result of the division operation is taken as the amplitude of the current node; the amplitude reflects the relative modulation intensity of the drift prediction relative to the physical prediction. An amplitude greater than 1 indicates that the drift prediction is amplified relative to the physical prediction, and an amplitude less than 1 indicates that the drift prediction is compressed relative to the physical prediction.

[0045] In determining the weighting coefficients, the entropy weighting method is used to dynamically calculate the consistency ratio weighting coefficient, cumulative distribution weighting coefficient, and amplitude weighting coefficient of the current component. Specifically, based on the historical data of the current component within the past 20 consecutive time windows, a consistency ratio sequence, a cumulative distribution sequence, and an amplitude sequence are constructed respectively. The information entropy of each sequence is calculated, and the degree of variation of each sequence is determined according to the magnitude of the information entropy. The smaller the information entropy, the more drastic the change of the indicator within the time window, and the greater the weight should be assigned. After normalizing the information entropy of each indicator, 1 minus the normalized information entropy is used as the weighting coefficient of the indicator, thereby obtaining the consistency ratio weighting coefficient, cumulative distribution weighting coefficient, and amplitude weighting coefficient of the current component in this mapping period, and the sum of the three is equal to 1.

[0046] Multiply the consistency ratio by the consistency ratio weight coefficient to obtain the consistency ratio weighted term; multiply the cumulative distribution function value by the cumulative distribution weight coefficient to obtain the cumulative distribution weighted term; multiply the amplitude by the amplitude weight coefficient to obtain the amplitude weighted term; sum the consistency ratio weighted term, the cumulative distribution weighted term, and the amplitude weighted term, and use the summation result as the comprehensive credibility score of the component.

[0047] S5. Based on the comprehensive credibility score, generate an updated dynamic mapping table and output it to the target platform.

[0048] In this embodiment of the invention, the preset confidence threshold is 0.75, and the threshold value ranges from 0 to 1. It can be dynamically adjusted according to the importance of the building, and can be increased to 0.85 for important buildings.

[0049] The comprehensive credibility score of the current component calculated by S4 is compared numerically with the preset credibility threshold. If the comprehensive credibility score is greater than or equal to the preset credibility threshold, the current dynamic mapping table is marked as a trusted mapping table, and the trusted mapping table is transmitted to the target platform in an incremental update manner. In the incremental update, the current trusted mapping table is compared line by line with the historical mapping table transmitted to the target platform in the previous output cycle, and the mapping entries in the current trusted mapping table that have changed relative to the historical mapping table are extracted. The changed mapping entries include newly added mapping relationships and modified mapping relationships. Only the extracted changed entries are transmitted to the target platform. After receiving the partial entries, the target platform adds or replaces the corresponding entries in the locally stored mapping table, thereby reducing communication overhead while maintaining the integrity of the mapping table.

[0050] If the overall credibility score is less than the preset credibility threshold, the current component's correction embedding vector, target mapping label, and mapping residual are combined into a record to be verified. The record to be verified is stored in a preset verification queue, which is used to temporarily store all mapping records whose overall credibility scores have not reached the threshold. At the same time, the output of the current dynamic mapping table to the target platform is paused, and manual or automatic verification is awaited.

[0051] After the trusted mapping table is transmitted to the target platform, the verification queue is cleared. All records to be verified in the verification queue are marked as processed and removed from the verification queue to obtain an empty verification queue. The empty verification queue is used as the initial state of the verification queue for the next mapping cycle to receive new records to be verified. The trusted mapping table transmitted to the target platform this time is used as the historical mapping record for the next mapping cycle and stored in the historical mapping record library of the target platform for consistency ratio calculation and cumulative distribution calculation in subsequent mapping cycles.

[0052] When the system is in a cold start state, the weight coefficient sequence required for the entropy weight method calculation has not yet accumulated enough sample length. At this time, the default weight values ​​of 0.4, 0.3, and 0.3 are temporarily used. After the number of samples in subsequent mapping cycles reaches the threshold, the system will switch to dynamic calculation using the entropy weight method. Other preset parameters used in each step are summarized as follows: the synchronous sampling frequency of the temperature sensor and the heat flow sensor is once every 10 minutes, and continuous sampling is not less than 72 hours; wavelet denoising uses db4 wavelet, 5-level decomposition, and a soft threshold constant of 0.6745; the low-excitation sample judgment threshold is 0.5°C, and the low-excitation sample weight coefficients are calculated according to... Calculation, where The temperature is set to 0.3°C; the forgetting factor for the recursive least squares algorithm is set to 0.98; the convergence criterion is that the relative change of the weighted residual sum of squares in three consecutive iterations is less than 0.1%; the lower limit of the thermal diffusivity truncation at the physical boundary is... / s, upper limit is / s, lower limit of thermal resistance is 0.1 ( K) / W, with a maximum limit of 10 ( K) / W; cold start penalty coefficient is 2.0; the number of time windows N for consistency ratio calculation is 20, and each mapping cycle is 24 hours; the preset credibility threshold is 0.75.

[0053] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0054] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0055] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for intelligently evaluating the application value of a green building BIM model, characterized in that, The method includes: S1. Construct an observation sequence with the real-time indoor and outdoor temperature difference of the building as the excitation and the average heat flow as the response. Perform parameter identification on the observation sequence to obtain the thermal response feature vector of the component in BIM. With the component as the node, the component topological connection relationship as the edge, the cosine similarity between the corresponding thermal response feature vectors of the component as the edge weight, and the thermal degradation index of the component as the node attribute, construct a heterogeneous graph. S2. The initial features of the nodes adjacent to each node in the heterogeneous graph are weighted and aggregated to obtain the physical prediction value of the adjacent nodes. The physical prediction value is multiplied by the amplitude ratio in the thermal response feature vector to obtain the drift prediction value. S3. The drift prediction value is superimposed with the static semantic vector of the component as a correction embedding vector. The label corresponding to the minimum standard Euclidean distance between the correction embedding vector and the pre-stored ordered continuous semantic label vector set of the target platform is the target mapping label. A dynamic mapping table is generated, and the minimum standard Euclidean distance is used as the mapping residual. S4. The comprehensive credibility score is obtained by weighted summation of the consistency ratio between the target mapping label and the target platform historical label, the cumulative distribution of the mapping residual, and the magnitude of the drift estimate relative to the physical estimate. S5. Based on the comprehensive credibility score, generate an updated dynamic mapping table and output it to the target platform.

2. The intelligent evaluation method for the application value of a green building BIM model as described in claim 1, characterized in that, The observation sequence, constructed using the real-time indoor-outdoor temperature difference of a building as the excitation and the average heat flux as the response, includes: The temperature data collected by temperature sensors on both the inside and outside of the building's exterior walls are used to perform a difference calculation, and the result of the difference calculation is used as the real-time temperature difference. Real-time heat flow data is collected simultaneously from heat flow sensors attached to the inner surface of the building's exterior walls; Wavelet transform denoising is performed on the temperature data and the real-time heat flow data respectively to filter out high-frequency noise components in the sensor acquisition link, and the denoised real-time temperature difference and denoised real-time heat flow data are used as the effective excitation and effective response of the component. Calculate the cross-correlation function between the effective excitation and the effective response, identify the time delay corresponding to the peak value in the cross-correlation function as the thermal response lag time, and perform reverse shift compensation on the timestamp of the effective response based on the thermal response lag time so that the phase of the compensated effective response is aligned with the phase of the effective excitation. Based on the consistency of the timestamps of the effective excitation and the effective response after phase alignment, the effective excitation and the effective response under the same timestamp are paired and combined. By arranging the pairing and combination results in ascending order of time, the observation sequence of the components in BIM is obtained.

3. The intelligent evaluation method for the application value of a green building BIM model as described in claim 1, characterized in that, The parameter identification of the observation sequence to obtain the thermal response feature vector of the component in the BIM includes: A theoretical transfer function is constructed with thermal diffusivity and thermal resistance as parameters to be identified. The real-time temperature difference in the observation sequence is used as the excitation input of the theoretical transfer function, and a theoretical heat flow sequence is output. Calculate the residual sequence between the theoretical heat flux sequence and the real-time heat flux data in the observed sequence; When the real-time temperature difference of the effective excitation in the component is lower than the preset environmental noise threshold, the pairing combination result in the corresponding time period is marked as a low excitation sample, and the low excitation sample is assigned a weight coefficient lower than that of the normal sample. The thermal diffusivity coefficient and thermal resistance are iteratively optimized using a recursive least squares algorithm with a forgetting factor. The weight coefficients corresponding to the low-excitation samples are used as the weighting coefficients of each sample in the recursive least squares algorithm until the weighted sum of squares of the residual sequence reaches the convergence criterion. The thermal diffusivity and thermal resistance corresponding to the convergence state are identified as the thermal parameters of the component. The geometric parameters of the component in the BIM database are concatenated with the thermal parameters to obtain the thermal response feature vector of the component.

4. The intelligent evaluation method for the application value of a green building BIM model as described in claim 3, characterized in that, The process of constructing a heterogeneous graph using the components as nodes, the topological connections between components as edges, the cosine similarity between the corresponding thermal response feature vectors of the components as edge weights, and the thermal degradation index of the components as node attributes includes: Read the spatial bounding box coordinate data and physical connection port data of the component, determine whether there is a coplanar contact area between the two components based on the spatial bounding box coordinate data, and determine whether there is an energy transfer path based on the physical connection port data; When both the existence of a coplanar contact area and the existence of an energy transfer path are satisfied, the connection relationship between the two components is determined to be a topological connection relationship. The direction of the topology connection edge is determined by the gradient direction of the real-time temperature difference between indoor and outdoor at the current moment. When the indoor temperature is higher than the outdoor temperature, the direction of the topology connection edge is from the indoor side component to the outdoor side component, and when the outdoor temperature is higher than the indoor temperature, the direction of the topology connection edge is from the outdoor side component to the indoor side component. Using the component as a node in the heterogeneous graph, all pairs of components that are determined to have the topological connection relationship are set as edges between nodes; The cosine similarity between the thermal response feature vectors of the two components connected by the edge is used as the edge weight. Based on the year the component was constructed, the difference between the current year and the year of construction is taken as the service life of the component; The material aging coefficient and the service life of the material attribute field in the BIM database are used as parameters to construct a degradation factor; Using the negative first power of the natural constant e as the base, and the product of the service life and the preset annual attenuation rate as the exponent, the result of the exponentiation is used as the aging attenuation coefficient. Multiply the material aging coefficient by the aging decay coefficient to obtain the comprehensive decay factor; The difference between 1 and the comprehensive attenuation factor is taken as the thermal degradation index of the component; The thermal degradation index is identified as a node attribute of the component. The component is used as a node set, the topological connection edge is used as an edge set, and the edge weight of the topological connection edge and the attribute of the node are used to form a heterogeneous graph.

5. The intelligent evaluation method for the application value of a green building BIM model as described in claim 4, characterized in that, The step of weighted aggregation of the initial features of the adjacent nodes of each node in the heterogeneous graph to obtain the physical estimated value of the adjacent nodes includes: Traverse the nodes in the heterogeneous graph and retrieve the initial feature vectors of the adjacent nodes corresponding to the nodes, which are used as the component set of the heat conduction physical dimension in the thermal response feature vector; Based on the edge weights of the topological connection edges between the current node and the neighboring nodes, the attention coefficients of the neighboring nodes relative to the current node are calculated according to the multi-head graph attention mechanism. The average value of the attention coefficients is then used as the aggregation weight coefficient. The aggregated feature vector is obtained by applying the aggregated weight coefficients to the initial feature vectors of the adjacent nodes and then weighting and summing the results. The mean-variance regularization is applied to each dimension component of the aggregated feature vector to obtain the physical estimate of the adjacent node. The dimensional shift introduced by the difference in the number of adjacent nodes is eliminated, and the regularized aggregated feature vector is confirmed as the physical estimate of the adjacent node.

6. The intelligent evaluation method for the application value of a green building BIM model as described in claim 5, characterized in that, The step of multiplying the physical estimate by the amplitude ratio in the thermal response feature vector to obtain the drift estimate includes: The amplitude amplification factor in the thermal response feature vector of the current node is taken as the current amplitude component; Traverse the adjacent nodes of the current node, take the amplitude amplification coefficient in the thermal response feature vector of the adjacent node as the adjacent amplitude component, divide the current amplitude component by each of the adjacent amplitude components to obtain the dynamic amplitude ratio coefficient corresponding to the adjacent node. The components of each dimension in the physical estimate are multiplied one by one with the dynamic amplitude scaling factor of the corresponding adjacent node, and the result of the one-dimensional multiplication is used as the initial drift estimate. The initial drift estimate is truncated by physical boundaries to obtain the drift estimate of the current node.

7. The intelligent evaluation method for the application value of a green building BIM model as described in claim 1, characterized in that, The step of superimposing the drift prediction value with the static semantic vector of the component as a correction embedding vector, generating a dynamic mapping table based on the label corresponding to the minimum standard Euclidean distance between the correction embedding vector and the pre-stored ordered continuous semantic label vector set of the target platform as the target mapping label, and using the minimum standard Euclidean distance as the mapping residual, includes: The static semantic attributes of the component are one-hot encoded to obtain the static semantic vector of the component; When the initial as-built reference amplitude amplification factor of the component is missing in the BIM database, cold start initialization is performed; The reference amplitude amplification factor stored in the BIM database at the current node in the initial state of completion is used as the reference amplitude component; The cold start initialization includes setting the thermal degradation index of the component to zero, replacing the reference amplitude component with the standard value of the material type to which the component belongs, and multiplying the mapping residual corresponding to the cold start initialization operation by a penalty coefficient greater than one. The aforementioned penalty coefficient serves as a mechanism to prevent misjudgment of uncertainties introduced by replacing real BIM data with standard values; The drift prediction value is superimposed with the static semantic vector to obtain the corrected embedding vector of the component; Based on the corrected embedding vector and the pre-stored ordered continuous semantic tag vector set of the target platform, the standard Euclidean distance between the corrected embedding vector and each tag vector in the ordered continuous semantic tag vector set is calculated one by one. The minimum value is selected from all standard Euclidean distances, and the label corresponding to the minimum value is identified as the target mapping label. The target mapping label is the energy efficiency level label of the component in the target platform. Record the mapping relationship between the target mapping label and the globally unique identifier of the component, write the mapping relationship into a dynamic mapping table, and use the minimum standard Euclidean distance as the mapping residual.

8. The intelligent evaluation method for the application value of a green building BIM model as described in claim 1, characterized in that, The step of basing the data on the consistency ratio between the target mapping label and the target platform's historical labels, the cumulative distribution of the mapping residuals, and the magnitude of the drift estimate relative to the physical estimate includes: The number of mappings with the target mapping tag within the past N time windows in the historical mapping records of the target platform is retrieved, the total number of mapping records within the past N time windows is retrieved, and the number of mappings is divided by the total number of mappings to obtain the consistency ratio of the component. Arrange the residual values ​​in the mapping residual sequence within a continuous mapping period in ascending order of numerical value to form an ordered residual sequence. The corresponding count is the number of residuals less than or equal to the residual value. The number of residuals is divided by the total number of residuals in the ordered residual sequence to obtain the empirical cumulative probability corresponding to each residual value. The empirical cumulative probability corresponding to the residual value in the ordered residual sequence that is equal to or closest to the current mapped residual is taken as the cumulative distribution function value of the mapped residual, and the cumulative distribution function value is taken as the cumulative distribution of the mapped residual. Calculate the vector norm of the drift estimate and the vector norm of the physical estimate, divide the vector norm of the drift estimate by the vector norm of the physical estimate, and use the result of the division operation as the amplitude of the current node.

9. The intelligent evaluation method for the application value of a green building BIM model as described in claim 1, characterized in that, The weighted sum, used as the overall credibility score, includes: The overall credibility score is calculated according to the following formula: ； in, Assess the overall reliability of the component. The consistency ratio weighting coefficient of the component is... The cumulative distribution weighting coefficient of the component is... The amplitude weighting coefficient of the component is... The consistency ratio of the components. The cumulative distribution function value of the mapped residual is dimensionless and ranges from 0 to 1. This represents the magnitude of the current node. in, satisfy The consistency ratio of the component is dimensionless and ranges from 0 to 1; the cumulative distribution function of the mapping residual is dimensionless and ranges from 0 to 1; the amplitude of the current node is dimensionless and is calculated by the ratio of the vector norm of the drift estimate to the vector norm of the physical estimate, and its value is greater than or equal to zero.

10. The intelligent evaluation method for the application value of a green building BIM model as described in claim 9, characterized in that, The step of generating and updating the dynamic mapping table based on the comprehensive credibility score and outputting it to the target platform includes: The overall credibility score is compared with a preset credibility threshold. If the overall credibility score is greater than or equal to the preset credibility threshold, the current dynamic mapping table is marked as a trusted mapping table, and the trusted mapping table is transmitted to the target platform only with the changed data in an incremental update manner. If the overall credibility score is less than the preset credibility threshold, the corrected embedding vector, the target mapping label and the mapping residual are combined into a record to be verified, the record to be verified is stored in the verification queue, and the output of the current dynamic mapping table to the target platform is paused. After the trusted mapping table is transmitted to the target platform, the verification queue is cleared to obtain an empty verification queue, and the trusted mapping table is stored as a historical mapping record in the historical mapping record library of the target platform for the next mapping cycle.

Images