Automatic modeling method for operation mechanism of building electromechanical system under few-shot labeling scenario

Abstract

The invention discloses a small sample labeling scene-oriented automatic modeling method for an operation mechanism of a building electromechanical system. The method comprises the following steps of: extracting static topological data and dynamic operation data of the building electromechanical system; constructing a topological operation data graph of the building electromechanical system; node topology feature vectors in the topology running diagram are extracted; performing small sample labeling on the operation mode in the topology operation diagram and constructing a meta-learning task through small sample learning; the operation modes of the brand new digital twinning model diagram of the building electromechanical system are automatically classified and labeled, and interpretable system operation rules, mathematical expressions or a causal relationship network are automatically generated on the digital twinning model diagram of the building electromechanical system to serve as a digital twinning operation mechanism model of the building electromechanical system. According to the method, the number of manually labeled samples is reduced, the dependency on experts and the labeling cost are reduced, and efficient, automatic and understandable modeling of the operation mechanism of the building electromechanical system is realized under the condition of a small number of labeled samples.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of building operation and maintenance, and in particular to a building mechanical and electrical system operation mechanism automatic modeling method for a small sample labeling scenario. BACKGROUND

[0002] In the field of building operation and maintenance, the building mechanical and electrical system is usually managed by a building mechanical and electrical system digital twin model. There are two technical paths for modeling the current building mechanical and electrical system digital twin model: physical mechanism model and data-driven model. The physical mechanism model relies on accurate equipment parameters and system architecture knowledge, but in actual installation engineering, the model precision is insufficient due to incomplete information or running deviation, and the model is complex to build and difficult to adapt to actual running deviation. The data-driven modeling method can adapt to complex systems, but requires a large amount of manually labeled samples to identify running conditions and equipment states. However, in new or renovation projects, due to the scarcity of expert resources and high labeling cost, there is a serious data cold start problem, making modeling difficult, and the existing technology lacks deep application of the topological relationship of the building mechanical and electrical system digital twin model, making it difficult to automatically generate an interpretable digital twin model with running mechanism from a small amount of labeled samples, which restricts the popularization and application of intelligent operation and maintenance technology. SUMMARY

[0003] The purpose of the present application is to provide a building mechanical and electrical system operation mechanism automatic modeling method for a small sample labeling scenario, to solve the problems of existing building mechanical and electrical system modeling methods that require a large amount of manually labeled sample data to identify running modes and equipment states, which have strong expert dependence, large sample data volume, high modeling cost, and poor interpretability.

[0004] To solve the above technical problems, the technical solution provided by the present application is: a building mechanical and electrical system operation mechanism automatic modeling method for a small sample labeling scenario, comprising: Step 1, extracting static topological data and dynamic running data of the building mechanical and electrical system; Step 2, constructing a topological running data graph of the building mechanical and electrical system; Step 3, extracting a node topological feature vector in the topological running data graph of the building mechanical and electrical system; Step 4, small sample labeling of the running mode in the topological running data graph of the building mechanical and electrical system and constructing a meta-learning task through small sample learning, comprising: Step 4.1, formulating a running mode label system for the dynamic running data of the building mechanical and electrical system; dividing the running mode of the building mechanical and electrical system into several categories to form a running mode classification system , wherein For the first to the nth category running mode, the running mode after the labeled category is taken as a sample; Step 4.2, organizing an episode training structure, training the sample through the episode training structure in small sample learning in machine learning, in the training process, randomly extracting running modes of the category each time, and selecting samples of the labeled category as a support set in a certain time period, and the rest of the labeled category samples as a query set, to form a meta-learning task, outputting the category of the running mode of the unlabeled time period in the query set through meta-learning and labeling to form a meta-learning model, so that small sample learning has the ability to quickly adapt to the running mode category labeling; Step 5, automatically classifying the running mode of the brand new building mechanical and electrical system digital twin model graph to build a building mechanical and electrical system digital twin running mechanism model; Combining the extracted node topology feature vector with the extracted dynamic running data to construct a “topology-running” joint feature vector, inputting into the meta-learning model to realize automatic classification and labeling of the category of the running mode of the brand new building mechanical and electrical system digital twin model graph under the condition of small samples; Step 6, according to the automatic classification and labeling result of step 5, automatically generating an interpretable system running rule, mathematical expression or causal relationship network on the building mechanical and electrical system digital twin model graph as a building mechanical and electrical system digital twin running mechanism model.

[0005] Further, the building mechanical and electrical system running mechanism automatic modeling method for small sample labeling scenarios provided by the present application further comprises: Step 7, verification and iterative optimization of the meta-learning model: through expert review, simulation comparison and online feedback mechanism, continuously improving the accuracy and reliability of the meta-learning model in identifying the category of the unlabeled running mode, verifying and iteratively updating the meta-learning model.

[0006] Further, the building mechanical and electrical system running mechanism automatic modeling method for small sample labeling scenarios provided by the present application, in step 1, the method for extracting static topology data and dynamic running data of the building mechanical and electrical system comprises: Step 1.1, parsing the static topology data of the building mechanical and electrical system from the building mechanical and electrical system digital twin model graph; wherein the static topology data mainly includes components of the building mechanical and electrical system and connection relationship information between the components, and the static topology data further includes component attribute information; Step 1.2, collecting dynamic operation data of the building mechanical and electrical system from the building automation system and the corresponding sensors deployed on the nodes of the building automation system corresponding to the building mechanical and electrical system digital twin model graph; wherein the dynamic operation data includes operation parameter information and operation state information; wherein the building automation system is a building mechanical and electrical system deployed in a building; Step 1.3, identifying and mapping the components of the building mechanical and electrical system digital twin model graph with the components in the operation data of the building automation system, realizing the corresponding relationship between the static topology data and the dynamic operation data.

[0007] Further, the building mechanical and electrical system operation mechanism automatic modeling method for small sample labeling scenarios provided by the application comprises the following steps: According to the mapping relationship between the extracted static topology data and the dynamic operation data, a heterogeneous graph is converted As a topology running data graph of the building mechanical and electrical system, the node set represents all components, each type of component is assigned a unique type code, and the edge set E represents the physical connection relationship or control relationship between components.

[0008] Further, the building mechanical and electrical system operation mechanism automatic modeling method for small sample labeling scenarios provided by the application comprises the following steps: Step 2.1, initializing the node feature vector: initializing the node feature vector for each node Constructing an initial feature vector ; wherein the node initial feature vector includes component type code, rated parameter, control mode, sensor identifier, system affiliation number, and node three-dimensional coordinate information, and all node feature vectors are collected into a node set; Step 2.2, defining the semantic type of the edge: the connection property between components is physical connection relationship or control relationship, and the edge between components is divided into different types; wherein the edge type includes fluid connection edge, electrical signal edge and control instruction edge, each type of edge is assigned a corresponding weight initial value, and all edges are collected into an edge set.

[0009] Further, the building mechanical and electrical system operation mechanism automatic modeling method for small sample labeling scenarios provided by the application comprises the following steps:

[0010] Further, the building mechanical and electrical system operation mechanism automatic modeling method for small sample labeling scenarios provided by the present application comprises the following steps: Step 3.1, performing multi-layer neighbor sampling and feature aggregation: using the GraphSAGE algorithm, for each node, a fixed number of neighbor nodes are randomly sampled in the 1-hop and 2-hop neighborhoods, the mean value of the neighbor features is calculated using the mean aggregator, and it is spliced with the own features, and after linear transformation and nonlinear activation function update to the node for node embedding, wherein the node embedding formula is: (1); In formula (1), is the node In the layer embedding vector, is an activation function, is a learnable weight matrix, is the node In the first layer embedding vector, is vector splicing, is a mean aggregation function, is the embedding set of all neighbor nodes of the node , denotes the neighbor node set of the node ; Step 3.2, generating the final node embedding: using the GraphSAGE algorithm, after L-layer propagation, the final low-dimensional feature vector of each node is obtained , wherein is a d-dimensional real number vector space, wherein the final low-dimensional feature vector f combines the position, functional angle context description information of the node in the building mechanical and electrical system and the association strength description information between nodes.

[0011] Further, the building mechanical and electrical system operation mechanism automatic modeling method for small sample labeling scenarios provided by the present application comprises the following steps: Step 5.1, constructing a joint feature vector: for each node In the time observation is spliced on the low-dimensional feature vector embedded in the topological operation data graph of the building mechanical and electrical system to obtain a “topology-operation” joint feature vector; Step 5.2, based on the prototype network in small sample learning, automatically classifying and labeling the small sample operation mode of the new building mechanical and electrical system digital twin model graph: on the basis of constructing the N-way K-shot episode in small sample learning, learning the typical mode of each operation mode by using the sample data labeled in the support set, assuming For the first The support set for class-based operation modes, where each sample data is a labeled pair. Encode the feature vectors of all sample data and average them to calculate the prototype vector of this type of operating mode: (2); In equation (2), C K For the prototype vector of the running mode category, For trainable feature encoding networks, For the first The class supports a certain number of sample data sets. Let i be the feature vector of the corresponding node i within a certain time period. The table shows the operating mode category to which the feature vector of the corresponding node i belongs; For any unlabeled time period feature vector in the query set Calculate its relationship with various prototype vectors Euclidean distance It assigns the operating mode to the nearest one and outputs the classification probability of the operating mode through the softmax function, thereby completing the automatic classification and labeling of the operating modes of the digital twin model of the building electromechanical system with small sample data, and automatically constructing the operating mechanism model of the digital twin model of the building electromechanical system.

[0012] Furthermore, in the automatic modeling method for the operating mechanism of building electromechanical systems in small-sample labeled scenarios provided by the present invention, step 6 includes: Step 6.1: Perform correlation analysis on the automatic classification and labeling results of the topology operation data diagram of the building electromechanical system and its operation mode categories to determine the key variables affecting the behavior of the building electromechanical system and their action paths; Step 6.2, Extraction and Parametric Modeling of Typical Operating Characteristic Curves: For each identified operating mode, extract the steady-state or dynamic response relationship between key variables from historical data, and construct a parametric curve model that reflects the operating characteristics of the building electromechanical system. Step 6.3, Constructing an Explainable Operating Rule Extraction and Knowledge Base: Combining cluster analysis and threshold segmentation techniques, key variables under various operating modes are identified to identify operating mode categories, extract typical operating logic, and generate a set of structured rules expressed in IF-THEN form. The rule condition part consists of variable thresholds, trend changes, or state combinations, and the conclusion part describes the system behavior response or mode determination result. The causal relationship network, mathematical expressions, and operating rules are integrated to construct a unified operating mechanism knowledge base and map it onto the digital twin model diagram of the building electromechanical system to automatically construct the operating mechanism and form a digital twin operating mechanism model of the building electromechanical system.

[0013] Compared with the prior art, the beneficial effects of the present application are as follows: The building mechanical and electrical system operation mechanism automatic modeling method for a small sample labeling scene provided by the present application extracts static topology data and dynamic operation data of the building mechanical and electrical system, constructs a topology operation data graph of the building mechanical and electrical system, extracts a node topology feature vector in the topology operation data graph of the building mechanical and electrical system, performs small sample labeling on an operation mode in the topology operation data graph of the building mechanical and electrical system, and constructs a meta-learning task to form a meta-learning model through small sample learning. The un-labeled operation mode is automatically classified through the meta-learning model. The recognized and labeled operation mode category automatically generates an interpretable system operation rule, mathematical expression or causal relationship network on a building mechanical and electrical system digital twin model graph as a building mechanical and electrical system digital twin operation mechanism model. Compared with the modeling method of the traditional data-driven model, the sample amount of the artificial labeled sample is reduced, the dependence on expert labeling and the labeling cost are reduced. Through small sample learning, the meta-learning task is constructed to automatically identify the category of the un-labeled operation mode with a small amount of labeled samples, and the combination of the graph neural network generates an interpretable operation rule, digital expression equation or causal relationship network, thereby realizing efficient, automatic and understandable modeling of the building mechanical and electrical system operation mechanism. The method is suitable for intelligent operation and maintenance and digital twin application in a low-data environment such as cold start of a newly deployed building mechanical and electrical system and renovation of an old building mechanical and electrical system. Under the condition of a small amount of labeled samples, the topology structure of the building information model and the operation data are fused to realize automatic modeling of the building mechanical and electrical system operation mechanism. BRIEF DESCRIPTION OF DRAWINGS

[0014] Fig. 1 is a step flow chart of the building mechanical and electrical system operation mechanism automatic modeling method for a small sample labeling scene; Fig. 2 is a framework flow chart of the building mechanical and electrical system operation mechanism automatic modeling method for a small sample labeling scene. DETAILED DESCRIPTION

[0015] The present application will be described in detail below with reference to the accompanying drawings. The advantages and features of the present application will be more apparent according to the following description. It should be noted that the drawings are very simplified and use non-precise proportions, only to facilitate and clarify the purpose of assisting the description of the embodiments of the present application.

[0016] Please refer to Figs. 1-2 The embodiment of the present application provides a building mechanical and electrical system operation mechanism automatic modeling method for a small sample labeling scene, which comprises: Step 1, extract the static topology data and dynamic operation data of the building mechanical and electrical system, that is, integrate the BIM model, sensor time series data and a small amount of manual annotation, construct a multi-modal data input basis to support subsequent structured modeling and semantic learning. Specifically, it includes: Step 1.1, parse the static topology data of the building mechanical and electrical system from the building mechanical and electrical system digital twin model diagram. The static topology data mainly includes the components of the building mechanical and electrical system such as chillers, pumps, fans, valves, pipes, sensors, and the connection relationship information between components. The static topology data can also include the rated power, flow, control logic description and system category of the corresponding components. Thus, the static topology data with component list and topology structure is formed. The building mechanical and electrical system digital twin model diagram is a BIM diagram with an IFC format. The topology structure here mainly refers to the connection relationship between components and the control relationship that may exist in the BIM diagram.

[0017] Step 1.2, collect the dynamic operation data of the building mechanical and electrical system from the building automation system corresponding to the building mechanical and electrical system digital twin model diagram and the sensors deployed on the corresponding nodes. The dynamic operation data includes temperature, humidity, pressure, current, voltage, frequency, flow, flow rate and other operating parameter information, as well as device start-stop, alarm signal, control mode and other operating state information. The dynamic operation data is also referred to as sensor time series data. The building automation system is the building mechanical and electrical system deployed in the building. The sampling granularity can be 1 minute to 5 minutes, and the sampling period is at least one week to truly reflect the operating state of the building mechanical and electrical system under different operating modes. In order to ensure the accuracy of the collected dynamic operation data, the collected dynamic operation data can be processed for missing value completion, outlier removal and time axis alignment to form a standardized data set with a unified format.

[0018] Step 1.3, identify and map the components of the building mechanical and electrical system digital twin model diagram with the components in the operation data of the building automation system, realizing the correspondence between the static topology data and the dynamic operation data.

[0019] For example, three types of data of the building mechanical and electrical system are collected: (1) Export from the BIM diagram in IFC format to extract static topology data containing component type, connection relationship and spatial coordinates; (2) Obtain dynamic operation data from the building automation system corresponding to the BIM diagram, collect 10 key measurement points, and the time span is more than 7 days, including the number.

[0020] (3) The dynamic operation data of 4 typical time periods is annotated by operation experts as different categories of operating modes, with only one sample for each operating mode, respectively: Sample S1 (02:00-02:30): labeled as "energy-saving operation mode".

[0021] Sample S2 (08:15-08:45): labeled as "transitional adjustment mode", also known as "high-power mode".

[0022] Sample S3 (14:00-14:30): labeled as "steady-state high-load mode".

[0023] Sample S4 (17:30-18:00): labeled as "abnormal fluctuation mode".

[0024] Among them, the operation mode is also called the working condition. Fig. 2 In this embodiment, "energy-saving" means that the "energy-saving operation mode" corresponds to the working condition 1, "high-power" means that the "transitional adjustment mode" corresponds to the working condition 2, "steady-state" means that the "steady-state high-load mode" corresponds to the working condition 3, and "abnormal" means that the "abnormal fluctuation mode" corresponds to the working condition 4.

[0025] Step 2, construct the topological operation data graph of the building mechanical and electrical system, that is, parse the components and their connection relationships in the BIM graph, construct a heterogeneous graph with components as nodes and physical paths or control paths as edges, and express the system structure topological operation graph. The physical path is also called the physical connection relationship, and the control path is also called the control relationship.

[0026] According to the mapping relationship between the extracted static topological data and dynamic operation data, a heterogeneous graph is transformed As a topological operation data graph of the building mechanical and electrical system, the node set represents all components, each type of component is assigned a unique type code, and the edge set E represents the physical connection relationship or control relationship between components. The processing steps of the nodes and edges include: Step 2.1, initialize the node feature vector.

[0027] For each node , an initial feature vector is constructed . The initial feature vector of the node includes the component type code , the rated parameter, the control mode , the sensor identifier , the system belonging number , and the three-dimensional coordinates of the node and other information. The rated parameter includes but is not limited to the rated power . All node feature vectors are collected into a node set. These information form structured input data for training of the graph neural network.

[0028] Step 2.2, define the semantic type of the edge.

[0029] The inter-component connection property is a physical connection relationship or a control relationship acting on the edge between components, and the edge is divided into different types. Among them, the edge types include fluid connection edge, electrical signal edge and control instruction edge, each type of edge is assigned a corresponding weight initial value, and all edges are collected into an edge set for the message passing mechanism of graph neural network training.

[0030] The processing procedures of steps 2.1 to 2.2 correspond to the graph neural network encoding in Fig. 2 .

[0031] For example: parsing the IFC format BIM graph, extracting 12 main components (such as cold source equipment, conveying equipment, end equipment, etc.) in static topology data and dynamic operation data, identifying the connection property of the components (such as the fluid passage as the connection property of the physical connection relationship, and the control link as the connection property of the control relationship), and constructing a heterogeneous graph G=(V,E). For example, there is a fluid connection between component A and component B, and there is a control signal link between component B and component C, forming a path structure of “A→B→C”, realizing the construction of the topology operation data graph of building mechanical and electrical system.

[0032] Step 3, extract the node topology feature vector in the topology operation data graph of building mechanical and electrical system, that is, use GraphSAGE algorithm to perform embedding learning on the topology graph, generate device feature vector fused with structure information, and input as prior knowledge.

[0033] Through the GraphSAGE algorithm in the graph neural network, the nodes of the topology operation data graph of building mechanical and electrical system are embedded and learned, the neighbor information is aggregated to generate a low-dimensional feature vector with context awareness capability, and the position, functional role and correlation strength of the node in the building mechanical and electrical system are obtained. That is, the graph neural network is trained through the GraphSAGE algorithm. Specifically, it includes: Step 3.1, perform multi-layer neighbor sampling and feature aggregation.

[0034] Using the GraphSAGE algorithm, for each node, a fixed number of neighbor nodes are randomly sampled within its 1-hop and 2-hop neighborhood, the mean value of neighbor features is calculated using the mean aggregator, and it is spliced with its own features, and updated to the node through linear transformation and nonlinear activation function for node embedding, where the node embedding formula is: (1); In formula (1), is the node In layer embedding vector, is an activation function, is a learnable weight matrix, for a node In the first layer of embedding vectors, for vector concatenation, for mean aggregation function, for a node embedding set of all neighbor nodes of the node, denotes the neighbor node set of the node .

[0035] Step 3.2, generate the final node embedding.

[0036] After L layers of propagation by the GraphSAGE algorithm, the final low-dimensional feature vector of each node is obtained . Wherein is a d-dimensional real number vector space, in which the final low-dimensional feature vector integrates the position, functional context description information of the node in the building mechanical and electrical system and the association strength description information between nodes. The final low-dimensional feature vector can be used as prior knowledge for subsequent operation mechanism automatic modeling. That is, by learning and training the weight matrix through the node embedding formula, the trained graph neural network is obtained.

[0037] For example: using the GraphSAGE algorithm, setting the embedding dimension d=64, the aggregation layer number L=2, and encoding the topological operation data graph of the building mechanical and electrical system in step 2. For example, for component A, sample its one-hop neighbor components B and C, calculate the neighbor feature mean and concatenate it with its own feature, and after two layers of propagation, output the topological embedding vector of component A ∈ . Wherein is the topological embedding vector of component A, denotes a 64-dimensional real number vector space.

[0038] Step 4, small sample labeling of the operation mode in the topological operation data graph of the building mechanical and electrical system and constructing a meta-learning task through small sample learning, that is, constructing N-way K-shot episode based on labeled samples, integrating sensor time series data and topological structure vector features, and realizing rapid identification of operation mode under the condition of few samples.

[0039] Step 4.1, develop an operation mode label system for dynamic operation data of the building mechanical and electrical system.

[0040] The operation mode of the building mechanical and electrical system is divided into several categories to form an operation mode classification system . Wherein For the first to the nth category running mode, the running mode after the labeled category is taken as a sample. The categories of the running mode include but are not limited to "cooling mode", "dehumidification mode", "night energy-saving mode", "cold machine high-pressure protection mode", etc.

[0041] Step 4.2, organize the episode training structure.

[0042] Train the sample through the episode training structure in small sample learning in machine learning. In the training process, randomly extract a category of running mode each time, and select labeled category samples in each category as a support set in a certain time period, and the rest of the labeled category samples as a query set to form a meta-learning task. The category of the running mode of the unlabeled time period in the query set is output through meta-learning and labeled to form a meta-learning model, so that the small sample learning has the ability to quickly adapt to the running mode category labeling.

[0043] Step 5, automatically classify the running mode of the brand new building mechanical and electrical system digital twin model graph to build a building mechanical and electrical system digital twin running mechanism model, that is, extract the key variable response relationship under various working conditions, build a standardized running characteristic curve and a parameterized model, and represent the dynamic behavior law of the building mechanical and electrical system.

[0044] Combine the extracted node topology feature vector with the extracted dynamic running data to construct a "topology-running" joint feature vector, and input it into the meta-learning model to realize automatic classification and labeling of the running mode of the brand new building mechanical and electrical system digital twin model graph under the condition of small sample. Specifically, it includes: Step 5.1, construct a joint feature vector.

[0045] For each node , the observation value at time is spliced on the low-dimensional feature vector embedded in the topology running data graph of the building mechanical and electrical system to obtain a "topology-running" joint feature vector . Among them is the observation value at time t, which corresponds to the time series data of the sensor, which is simplified as time series data.

[0046] Step 5.2, realize automatic classification and labeling of the small sample running mode of the brand new building mechanical and electrical system digital twin model graph based on the prototype network in small sample learning.

[0047] On the basis of constructing N-way K-shot episode in small sample learning, learn the typical mode of each category of running mode by using the labeled sample data in the support set. Let be the a support set of class running modes, each sample data being a labeled pair Encode and average the feature vectors of all sample data to calculate the prototype vector of the class running mode: (2); In formula (2), C K prototype vector of the running mode class, trainable feature encoding network, the number of sample data in the support set of the class, feature vector of the corresponding node i in a certain period, table the running mode class to which the feature vector of the corresponding node i belongs; For any unlabeled period feature vector in the query set , calculate its Euclidean distance with each prototype vector , assign it to the nearest running mode, and output the classification probability of the running mode through the softmax function, thereby completing the automatic classification and labeling of the running mode of the building mechanical and electrical system digital twin model graph in the case of small sample data, and realizing the automatic construction of the building mechanical and electrical system digital twin running mechanism model.

[0048] For example: construct a 4-way 1-shot episode: the support set contains the above 4 labeled samples (S1-S4), and the query set selects the 20:00-20:30 unlabeled period. For each sample, extract its joint feature sequence, encode it into a fixed length vector through LSTM = () for each prototype vector, calculate the Euclidean distance of the query sample with each prototype vector, identify the nearest “energy-saving running mode”, and complete the automatic classification and labeling of the running mode of the unlabeled period in the query set.

[0049] For the “energy-saving running mode” class, extract the “load rate-unit energy consumption” relationship from all time periods identified as this class. For example, in the load rate 20%-40% interval, the unit energy consumption is maintained at 0.8-1.0 kW / ton, forming the left branch of the U-shaped curve; when the load rate is <20%, the unit energy consumption rises sharply and is identified as the “inefficient zone”. Extract the inflection point threshold (such as load rate = 20%), and construct a parameterized model: when the load rate >20%, the energy efficiency is optimal.

[0050] ​Step 6, based on the automatic classification and labeling results of step 5, automatically generate interpretable system operation rules, mathematical expressions or causal relationship networks on the building mechanical and electrical system digital twin model diagram as the building mechanical and electrical system digital twin operation mechanism model, that is, the combined action path and characteristic model, generate IF-THEN form operation rules, and build readable, verifiable and callable mechanism knowledge base. Among them, the operation rules, mathematical expressions or causal relationship networks are three expression ways of the operation mechanism of the building mechanical and electrical system, that is, the automatic construction of the building mechanical and electrical system digital twin operation mechanism model corresponds to Fig. 2 the construction of building operation mechanism in Step 6.1, correlate the automatic classification and labeling results of the building mechanical and electrical system topology operation data diagram and its operation mode, and determine the key variables and their action paths that affect the behavior of the building mechanical and electrical system. Specifically: Use the component connection relationship provided by BIM diagram as analysis constraint, only carry out correlation and response relationship test between variables with physical connection relationship or control link, and exclude false association without connection basis.

[0051] Through calculating the time series correlation, mutual information and dynamic time warping distance between variables, the variable pairs with leading-lag relationship in time are screened out.

[0052] Using sliding window regression and residual analysis method, identify the input variables that have significant influence on the output of the building mechanical and electrical system under a certain operation mode, and analyze their response strength and delay characteristics.

[0053] Combine the sequence of variable changes and the topology direction of the building mechanical and electrical system to construct the action network reflecting the energy, signal or medium transmission path, and describe the functional dependence relationship and dominant action mechanism of each component in the operation process.

[0054] Step 6.2, extraction and parameterized modeling of typical operation characteristic curve: for each identified operation mode, extract the steady-state or dynamic response relationship between key variables from historical data to construct parameterized curve model reflecting the operation characteristics of the building mechanical and electrical system.

[0055] Through clustering alignment and time normalization, align and average multiple operation trajectories under the same operation mode to generate standardized operation characteristic curves, including input-output response curve, load-energy consumption relationship curve and transition process dynamic curve, etc.

[0056] Segment linearization or basis function expansion is performed on each curve to extract key feature parameters such as response delay, rise time, steady-state gain, slope inflection point and energy efficiency inflection point.

[0057] Establish the correlation model between these characteristic parameters and the system boundary conditions (such as environmental parameters, set values), forming an adaptive operating characteristic description that can be adjusted according to the operating mode.

[0058] The generated characteristic curve and its parameter set together constitute a typical behavior template of the building mechanical and electrical system under this operating mode, which is used for subsequent energy efficiency evaluation, anomaly detection, and control strategy comparison.

[0059] Step 6.3, extraction of interpretable operating rules and knowledge base: combined with clustering analysis and threshold segmentation technology, key variables under various operating modes are identified by operating mode classification, typical operating logic is extracted, and a structured rule set expressed in IF-THEN form is generated. The rule condition part is composed of variable threshold, trend change or state combination, and the conclusion part describes the system behavior response or mode determination result. The causal relationship network, mathematical expression and operating rule are integrated to build a unified operating mechanism knowledge base and mapped to the building mechanical and electrical system digital twin model graph to automatically build the operating mechanism and form the building mechanical and electrical system digital twin operating mechanism model.

[0060] Among them, the operating mechanism knowledge base is stored in a readable format and supports visual display and external system calling, serving as an interpretable decision basis for building mechanical and electrical system fault diagnosis, energy efficiency evaluation and operation optimization.

[0061] For example: based on the curve of step 5 and the action path of step 6, automatically generate IF-THEN rules.

[0062] IF system load rate <20% AND unit energy consumption >1.2 kW / ton, THEN determine as "low efficiency operation, suggest to improve load or start-stop optimization".

[0063] IF key parameter change rate is less than 0.1% / min for 5 consecutive sampling points, THEN determine as "system is in steady state, no need to adjust". All operating rules are stored by operating mode, with data source and confidence score.

[0064] Through the building mechanical and electrical system digital twin operating mechanism model, fault analysis of building mechanical and electrical system operation and maintenance can be carried out, and corresponding maintenance strategies can be output.

[0065] Step 7, verification and iterative optimization of meta-learning model.

[0066] Through expert review, simulation comparison and online feedback mechanism, the accuracy and credibility of the meta-learning model in classifying unlabeled operating modes are evaluated, and the meta-learning model is verified and iteratively updated. Support manual correction and incremental learning, realize system continuous evolution, support closed-loop update and long-term operation adaptation. Specifically including: Step 7.1, machine learning rationality verification is performed.

[0067] The automatically generated operation mode category label is compared with the expert-labeled operation mode category to evaluate the accuracy of the machine learning operation mode category judgment recognition; and the building mechanical and electrical system digital twin operation mechanism model is cross-verified with the operation mode in the historical maintenance record to ensure the effectiveness of the machine learning output.

[0068] Step 7.2, expert feedback and model updating are supported.

[0069] The operation personnel are allowed to correct the wrong label or rule, and the corrected small sample data is re-included in the training set to drive the machine learning fine-tuning of the meta-learning task or re-training of the graph neural network, so that the continuous evolution and adaptability enhancement of the machine learning are realized.

[0070] The generated operation rule is compared with the historical maintenance record, it is found that the "inefficient operation" rule coincides with 3 real work orders, and the accuracy is 92%. The expert corrects 1 rule: "load rate < 15% triggers low-efficiency alarm", the system adds the corrected sample to the training set and re-fine-tunes the meta-learning model. After updating, the recognition accuracy of the meta-learning model on the test set is improved to 95%, and the closed-loop optimization is completed.

[0071] The building mechanical and electrical system operation mechanism automatic modeling method for small sample labeling scenarios provided by the embodiment of the application realizes efficient, automatic and interpretable modeling of the building mechanical and electrical system operation mechanism on the building mechanical and electrical system digital twin model graph by fusing the topological structure information of the building mechanical and electrical system digital twin model graph and the category data of a small amount of manually labeled operation modes, thereby significantly reducing the dependence on large-scale labeled operation mode data, and under the condition that only 3-5 labeled samples per operation mode are required, accurate identification of the operation mode of the building mechanical and electrical system and automatic classification of unlabeled data can be completed; the graph neural network of the GraphSAGE algorithm is introduced to extract the structural correlation features between components, and an episode training mechanism is constructed in combination with a meta-learning framework, so that the generalization ability and operation mode recognition accuracy of the meta-learning model in the cold start scenario are improved; further, based on the role path analysis and typical operation characteristic curve extraction under the topological constraint, a parameterized model and readable rule reflecting the dynamic response law of the system are generated, and the physical rationality and operation understanding of the modeling result are enhanced; the constructed operation mechanism knowledge base can be directly used for fault diagnosis, energy efficiency evaluation and control optimization, and is suitable for building mechanical and electrical system installation and debugging, old building mechanical and electrical system modification and other actual construction engineering scenarios, and has the technical advantages of short modeling period, less expert participation and flexible deployment.

[0072] The building mechanical and electrical system operation mechanism automatic modeling method for small sample labeling scenarios provided by the embodiment of the application has the following advantages: Low data dependency: only 3-5 labeled samples per type of working condition are needed to construct an effective operating mechanism model, solving the cold start problem.

[0073] Knowledge fusion advantage: make full use of the existing static topological data in BIM map, reduce the dependence on dynamic operation data.

[0074] Strong interpretability: the generated operation rules, numerical expression equations and causal relationship networks are easy to understand and verify.

[0075] High generalization ability: knowledge is transferred between different operating mode tasks through a meta-learning framework.

[0076] Engineering practicability: especially suitable for deploying building mechanical and electrical systems in newly built buildings at the initial stage and for building mechanical and electrical system transformation scenes in old buildings, accelerating the landing of intelligent operation and maintenance platforms for building mechanical and electrical systems.

[0077] The present application is not limited to the above-mentioned specific embodiments, and obviously, the above-described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the described embodiments of the present application, all other embodiments obtained by those skilled in the art belong to the scope of protection of the present application. Those skilled in the art can make other levels of modifications and changes to the present application. Thus, if these modifications and changes of the present application belong to the scope of the claims of the present application, the present application also intends to include these modifications and changes.

Claims

1. A method for automatically modeling the operation mechanism of a building mechanical and electrical system in a small sample labeling scenario, characterized in that, Comprising: Step 1, extracting static topology data and dynamic operation data of building mechanical and electrical system; Step 2, constructing topology operation data graph of building mechanical and electrical system; Step 3, extracting node topology feature vector in topology operation data graph of building mechanical and electrical system; Step 4, small sample labeling of operation mode in topology operation data graph of building mechanical and electrical system and constructing meta-learning task through small sample learning, comprising: Step 4.1, establishing a running mode label system of dynamic running data of building mechanical and electrical system; dividing the running mode of building mechanical and electrical system into several categories to form a running mode classification system wherein for the 1st to nth category running mode, the labeled running mode is taken as a sample; Step 4.2, organize episode training structure, train samples through episode training structure in small sample learning in machine learning, in the training process, randomly extract one category of operation mode each time, and select samples of the labeled category in a certain time period as a support set, and the rest of the labeled category samples as a query set to form a meta learning task, output the category of the operation mode of the unlabeled time period in the query set through meta learning and label to form a meta learning model, so that small sample learning has the ability to quickly adapt to the operation mode category labeling; Step 5, automatic classification and labeling of operation mode of new building mechanical and electrical system digital twin model graph; Combine the extracted node topology feature vector with the extracted dynamic operation data to construct a "topology-operation" joint feature vector, which is input into the meta-learning model to automatically classify and label the class of the operation mode of the new building mechanical and electrical system digital twin model graph under the condition of small sample; Step 6, according to the automatic classification and labeling result of step 5, automatically generating interpretable system operation rules, mathematical expressions or causal relationship networks on the building mechanical and electrical system digital twin model graph as the building mechanical and electrical system digital twin operation mechanism model.

2. The method of claim 1, wherein the method is applied to a small sample labeling scenario of a building mechanical and electrical system. Further comprising: Step 7, verification and iterative optimization of meta-learning model: through expert review, simulation comparison and online feedback mechanism, the accuracy and reliability of the meta-learning model in classifying unlabeled operation mode are evaluated, and the meta-learning model is verified and iteratively updated.

3. The method of claim 1, wherein the method is applied to a small sample annotation scenario of a building mechanical and electrical system. In step 1, the method for extracting static topology data and dynamic operation data of building mechanical and electrical system comprises: Step 1.1, parsing the static topology data of building mechanical and electrical system from the building mechanical and electrical system digital twin model graph; wherein the static topology data mainly includes the components of the building mechanical and electrical system and the connection relationship information between the components, wherein the static topology data also includes component attribute information; Step 1.2, collecting dynamic operation data of building mechanical and electrical system from building automation system and corresponding sensors deployed on corresponding nodes; wherein the dynamic operation data includes operation parameter information and operation state information; wherein the building automation system is a building mechanical and electrical system deployed in a building; Step 1.3, mapping the components of the building mechanical and electrical system digital twin model graph with the components in the operation data of the building automation system, realizing the correspondence between the static topology data and the dynamic operation data.

4. The method of claim 1, wherein the method is a small sample oriented automatic modeling method for operation mechanism of building mechanical and electrical system. In step 2, the method for constructing topology operation data graph of building mechanical and electrical system comprises: According to the mapping relationship of the extracted static topology data and dynamic operation data, a heterogeneous graph is converted As a topology operation data graph of building mechanical and electrical systems, wherein a node set represents all components, each type of component is assigned a unique type code, and an edge set E represents the physical connection relationship or control relationship between components.

5. The method of claim 4, wherein the method is applied to a small sample labeling scenario of a building mechanical and electrical system. In step 2, the processing steps of nodes and edges include: Step 2.1, initializing node feature vector: for each node constructing initial feature vector ; wherein the node initial feature vector includes information such as component type code, rated parameter, control mode, sensor identification, system belonging number, and node three-dimensional coordinates, and all node feature vectors are collected into a node set; Step 2.2, define the semantic type of edge: the connection property between components is physical connection relationship or control relationship as the edge between components, and the edge is divided into different types; wherein the edge type includes fluid connection edge, electrical signal edge and control instruction edge, each type of edge is assigned with corresponding weight initial value, and all edges are collected into edge set.

6. The method of claim 1, wherein the method is a small sample oriented automatic modeling method for operation mechanism of building mechanical and electrical system. In step 3, the method for extracting the node topology feature vector in the topology operation data graph of the building mechanical and electrical system includes: embedding learning of the nodes in the topology operation data graph of the building mechanical and electrical system through a GraphSAGE algorithm in a graph neural network, aggregating neighbor information to generate a low-dimensional feature vector with context awareness capability, and obtaining the position, functional role and correlation strength of the node in the building mechanical and electrical system.

7. The method of claim 6, wherein the method is directed to a small sample labeling scenario of a building mechanical and electrical system. The step 3 includes: Step 3.1, performing multi-layer neighbor sampling and feature aggregation: using the GraphSAGE algorithm, for each node, a fixed number of neighbor nodes are randomly sampled in its 1-hop and 2-hop neighborhood, the mean value of neighbor features is calculated using the mean aggregator, and it is spliced with its own features, and after linear transformation and nonlinear activation function update to the node for node embedding, wherein the node embedding formula is: （1）； In formula (1), is a node In the embedding vector of the layer, is an activation function, is a learnable weight matrix, is a node In the first the embedding vector of the layer, is a vector concatenation, is a mean aggregation function, is a node a set of embedding of all neighbor nodes of the node denotes a set of neighbor nodes of the node . Step 3.2, generating the final node embedding: after L-layer propagation using the GraphSAGE algorithm, the final low-dimensional feature vector of each node is obtained wherein is a d-dimensional real number vector space, wherein the final low-dimensional feature vector integrates the position of the node in the building mechanical and electrical system, the functional angle context description information of the node, and the association strength description information between nodes.

8. The method of claim 1, wherein the method is a small sample oriented automatic modeling method for operation mechanism of building mechanical and electrical system. The step 5 includes: Step 5.1, Constructing joint feature vector: for each node At time of observation The low-dimensional feature vector embedded in the topology-operation data graph of the building mechanical and electrical system is spliced to obtain a "topology-operation" joint feature vector; Step 5.2: Automatic classification and labeling of small sample operation modes for the newly built building mechanical and electrical system digital twin model graph based on the prototype network in small sample learning: Based on the construction of N-way K-shot episodes in small sample learning, learn the typical mode of each operation mode using the labeled sample data in the support set, set the first operation mode of the support set, wherein each sample data is a labeled pair Encode and average the feature vectors of all sample data to calculate the prototype vector of the operation mode. （2）； In formula (2), C K a prototype vector of an operation mode category, a trainable feature encoding network, a first a number of sample data in the class support, a feature vector of a corresponding node i within a certain period of time, a table indicating an operation mode category to which the feature vector of the corresponding node i belongs. For any unlabeled time period feature vector in the query set , calculate its Euclidean distance with each class prototype vector , assign it to the operating mode closest to it, and output the classification probability of the operating mode through the softmax function, thereby completing the automatic classification and labeling of the operating mode of the building mechanical and electrical system digital twin model graph in the case of small sample data, and realizing the automatic construction of the building mechanical and electrical system digital twin operating mechanism model.

9. The method of claim 1, wherein the method is a small sample oriented automatic modeling method for operation mechanism of building mechanical and electrical system. The step 6 includes: Step 6.1, correlation analysis of the automatic classification and labeling results of the topology operation data graph of the building mechanical and electrical system and its operation mode categories, to determine the key variables and their action paths that affect the behavior of the building mechanical and electrical system; Step 6.2, extraction and parameterized modeling of typical operating characteristic curves: for each identified operation mode, the steady-state or dynamic response relationship between key variables is extracted from historical data to build a parameterized curve model reflecting the operating characteristics of the building mechanical and electrical system; Step 6.3, extraction and knowledge base of interpretable operation rules: combining clustering analysis and threshold segmentation technology, the key variables under each operation mode are identified to extract typical operation logic and generate a structured rule set expressed in IF-THEN form, the rule condition part is composed of variable threshold, trend change or state combination, and the conclusion part describes the system behavior response or mode determination result; the causal relationship network, mathematical expression and operation rule are fused to build a unified operation mechanism knowledge base and mapped to the building mechanical and electrical system digital twin model graph to automatically build an operation mechanism forming a building mechanical and electrical system digital twin operation mechanism model.

Images