A knowledge graph-based heating system inspection system and method

Abstract

The application relates to the technical field of intelligent control, in particular to a knowledge graph-based heating and ventilation system inspection system and method, which comprises a feature extraction module, a graph construction module, an abnormality tracing module, a cause-effect analysis module and a priority module. In the application, water inlet and outlet temperature and flow data of a water chilling unit are collected, deviation features are extracted by combining principal component analysis, the operation state is quantized, a graph is embedded to participate in state judgment, the abnormality detection capability is improved, the initial abnormality is focused on and the analysis range is narrowed by upstream nodes, cause-effect tracing is carried out based on the state influence between devices, a state evolution chain is constructed by means of a Bayesian network, a time sequence and a structural gap are bridged, task priority is evaluated according to the starting point and length of a link, equipment dynamic sequencing and state updating are carried out, the inspection scheduling accuracy and efficiency are improved, and a closed-loop process with high recognition accuracy, small analysis boundary, clear path and optimal resource allocation is formed.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control technology, and in particular to a knowledge graph-based inspection system and method for a heating and ventilation system. Background Art

[0002] The field of intelligent control technology encompasses automated monitoring of building environmental control equipment, multi-source data fusion and analysis, and the integrated application of intelligent decision-making systems. Its core focus is on building a collaborative network of self-learning equipment. This field, based on HVAC systems, encompasses three key technical areas: operational optimization of cooling and heating units, hydraulic balancing of pipe networks, and dynamic terminal load matching. Using distributed sensor arrays to collect pressure, temperature, and flow parameters, combined with a database of equipment performance curves and fuzzy control algorithms, it optimizes energy efficiency under multiple operating conditions. Furthermore, an integrated fault pattern recognition engine extracts features from typical anomalies, such as compressor vibration and heat exchanger scaling.

[0003] The knowledge graph-based HVAC system inspection system and method utilizes ontology modeling technology to construct a semantic network of equipment entities, operating parameters, and environmental variables. This method stores static attributes such as the chiller's energy efficiency coefficient and pump head characteristic curve alongside real-time operation logs in a graph structure. This method employs time-series data slicing technology to extract characteristic waveforms during equipment startup and shutdown phases. A graph convolutional neural network is used to map component degradation patterns to sensor reading fluctuations. The SPARQL query language is used to trace fault propagation paths across subsystems. A rule-based inference engine is then used to compare 120 pre-set inspection indicators, such as valve opening thresholds and condensing temperature gradients, to complete status assessments.

[0004] Existing technologies rely on fuzzy control and distributed sensing systems to coordinate energy efficiency regulation and fault identification. While these technologies achieve certain energy consumption control effects under various operating conditions, they still lack the ability to model operating status for complex system structures. The connections between operating parameters and device entities often rely on rule-based reasoning and threshold settings, lacking the ability to express multidimensional numerical states. This results in low discriminability in operating status assessment. Regarding anomaly analysis, fault propagation path tracing primarily relies on the SPARQL semantic query language. While capable of cross-subsystem access, it is limited by the static properties of the graph structure and cannot dynamically analyze the state evolution chain during causal transmission. This results in jumps in the traceability path and a high rate of false positives. Furthermore, existing technologies lack systematic evaluation criteria for prioritizing equipment anomaly inspections. Instead, they often rely on manual experience or simplified ranking based on the frequency of indicator violations, resulting in insensitive identification of key nodes in complex anomaly networks. For example, compressor anomalies are often indirectly triggered by heat exchanger performance degradation. Without a definitive link determination mechanism, downstream equipment can be easily misidentified as the source, delaying resolution and wasting inspection resources. The above problems have become increasingly prominent in the context of increasing operational complexity, and they urgently need to be supplemented by more accurate and structured analysis mechanisms. Summary of the Invention

[0005] In order to solve the technical problems existing in the prior art, the embodiment of the present invention provides a HVAC system inspection system and method based on a knowledge graph. The technical solution is as follows:

[0006] On the one hand, a knowledge graph-based HVAC system inspection system is provided, which includes:

[0007] The feature extraction module collects the chiller's inlet and outlet temperature and water flow data to calculate the chiller's actual cooling capacity. It then uses the principal component analysis algorithm to extract features based on the deviation between the actual cooling capacity and the nameplate rated cooling capacity. This generates a chiller operation label and passes it to the graph construction module.

[0008] A graph construction module constructs a graph structure with chillers, fan coil units, and heat exchangers as nodes, assigns the chiller operation labels to device attributes, determines node status differences, generates a list of abnormal candidate devices, and passes it to the abnormality tracing module;

[0009] The abnormality tracing module obtains the abnormality occurrence time of the devices in the abnormal candidate device list, identifies the first abnormal device, generates an initial abnormal device set and passes it to the causal analysis module;

[0010] A causal analysis module obtains the state changes of the initial abnormal device set and adjacent devices, performs causal analysis on the state changes through a Bayesian network model, generates an abnormal causal link graph, and transmits it to the priority module;

[0011] The priority module extracts the link starting device and path length based on the abnormal causal chain map, determines the inspection priority, updates the inspection status of the associated devices in the map, and generates an inspection priority result.

[0012] As a further solution of the present invention, the chiller operation label is the operation status category, deviation feature vector, and feature extraction time. The abnormal candidate device list includes the candidate device number, node attribute difference value, and graph position index. The initial abnormal device set includes the first abnormal device number, the time when the abnormality occurred, and the device abnormality type. The abnormal causal link graph refers to the causal path structure, state change dependency, and link node timing information. The inspection priority result is the inspection starting device number, path priority level, and associated device inspection status value.

[0013] As a further solution of the present invention, the feature extraction module includes:

[0014] The temperature acquisition submodule collects the water inlet temperature data, water outlet temperature data and instantaneous water flow data of the chiller, organizes the flow time accumulation and the temperature difference corresponding to the corresponding sampling time interval, and generates a temperature difference flow data table;

[0015] The cooling capacity calculation submodule obtains the water specific heat capacity constant and the water flow rate based on the temperature difference flow rate data table, calculates the cooling capacity value of the time period in combination with the temperature difference, converts and integrates the values, and obtains a cooling capacity value sequence of the time period;

[0016] The deviation analysis submodule normalizes the cooling capacity value sequence of the time period with the rated cooling capacity value on the nameplate of the chiller, calculates the principal component vector characteristics between the two through the principal component analysis algorithm, reconstructs the operating status characteristic index, and generates the chiller operation label.

[0017] As a further solution of the present invention, the principal component vector feature between the two is calculated by the principal component analysis algorithm. , using the formula:

[0018] ;

[0019] in, represents the normalized cooling capacity value of the i-th time point, represents the sliding mean of the cooling capacity series, Represents the dynamic standard deviation of the cooling capacity series, represents the normalized nameplate rated cooling capacity value corresponding to the i-th time point, represents the exponential mean of the rating series, represents the coefficient of variation of the rating series, represents the correction amount of the j-th eigenvalue of the covariance, Represents the time gradient tensor of the eigenvector space, n represents the total number of time series data points, m represents the total number of covariance eigenvalues, and t represents the time variable.

[0020] As a further embodiment of the present invention, the graph construction module includes:

[0021] The node extraction submodule obtains the device identification, connection relationship and deployment information of the chiller, fan coil unit and heat exchanger equipment, as well as the device node group and connection path information, and establishes the device map connection information set;

[0022] The attribute fusion submodule writes the chiller operation tag data into the corresponding node based on the device map connection information set and the chiller operation tag, fills the operation attribute fields of the remaining nodes according to the device type, organizes and marks them as identifiable status fields, and obtains the node operation attribute data group;

[0023] The difference identification submodule calculates the state deviation rate between each node and the chiller node state value based on the node operation attribute data group, determines whether there is an obvious deviation in the operation state, filters the nodes, and obtains a list of abnormal candidate devices.

[0024] As a further solution of the present invention, the abnormality tracing module includes:

[0025] The time extraction submodule obtains the operating status and corresponding timestamp information of the devices in the abnormal candidate device list, extracts the first moment when the device operating status changes as the abnormal time, combines the device number and time into a record entry, and establishes an abnormal occurrence time value set;

[0026] The first-time identification submodule sorts the abnormal time of all devices based on the abnormal occurrence time value set, selects the device corresponding to the earliest time as the starting device, determines whether there are multiple device records and selects the device with the smallest number as the representative device, and generates the first abnormal device number;

[0027] The initial positioning submodule retrieves the corresponding device operating status and attribute information from the abnormal candidate device list according to the first abnormal device number, constructs the starting abnormal path entry and combines the first device attribute field, marks the device as the initial abnormal node and establishes the initial identification tag to obtain the initial abnormal device set.

[0028] As a further solution of the present invention, the causal analysis module includes:

[0029] The state change extraction submodule obtains the device numbers in the initial abnormal device set and the state change data of adjacent devices, extracts the corresponding relationship between the device state change time series and the state change of adjacent devices, organizes and marks the change moments of the devices, and generates a state change time series;

[0030] A causal relationship construction submodule extracts the direct impact relationship between device state changes based on the state change time series, calculates the causal relationship strength of the impact path through the Bayesian network structure, and marks it as a causal relationship chain to obtain the causal relationship path;

[0031] The causal link generation submodule visualizes the device state changes and causal relationships in the graph based on the causal relationship path, constructs a complete abnormal causal link graph, and marks the abnormal causal impact range of each device to generate an abnormal causal link graph.

[0032] As a further solution of the present invention, the causal relationship strength of the impact path is calculated by the Bayesian network structure. , using the formula:

[0033] ;

[0034] in, represents the conditional probability parameter of node A to B, Indicates the time difference between the A / B node status changes. represents the covariance coefficient between node A and the lth associated node, represents the noise suppression parameter of Node B, represents the signal attenuation parameter of Node B, Indicates the total number of associated nodes, including parent nodes and child nodes.

[0035] As a further solution of the present invention, the priority module includes:

[0036] The path extraction submodule, based on the link information in the abnormal causal chain graph, calibrates the starting device and the ending device according to the node sequence, obtains the first device number in the path, and reads the path nodes in sequence according to the connectivity relationship of the edges in the graph to obtain the link node sequence data;

[0037] The length judgment submodule obtains the total number of link nodes in the link node sequence data, marks the priority link when the path length value is equal to the path reference length, and generates a path priority judgment result based on the link starting device number information;

[0038] The status update submodule locates the device nodes that are connected to the link path in the map according to the path priority determination result, updates the value of the device node status label marked as the priority path, writes the corresponding node attribute field, and generates the inspection priority result.

[0039] On the other hand, a knowledge graph-based HVAC system inspection method is provided. The method is applied to a knowledge graph-based HVAC system inspection system. The method includes:

[0040] S1: Collect the chiller's inlet and outlet temperature and water flow data to calculate the chiller's actual cooling capacity. Use the principal component analysis algorithm to extract features based on the deviation between the actual cooling capacity and the nameplate rated cooling capacity, and generate the chiller's operating label.

[0041] S2: Construct a graph structure with chillers, fan coil units, and heat exchangers as nodes, assign the chiller operation labels to device attributes, determine node status differences, and generate a list of abnormal candidate devices;

[0042] S3: Obtain the abnormality occurrence time of the device in the abnormal candidate device list, identify the first abnormal device, and generate an initial abnormal device set;

[0043] S4: Obtaining the state changes of the initial abnormal device set and adjacent devices, performing causal analysis on the state changes using a Bayesian network model, and generating an abnormal causal link graph;

[0044] S5: Based on the abnormal causal chain graph, the link starting device and path length are extracted, the inspection priority is determined, the inspection status of the associated devices in the graph is updated, and the inspection priority result is generated.

[0045] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0046] By jointly collecting the chiller's inlet and outlet water temperatures and water flow rates, dynamic cooling performance data under actual operating conditions is generated. A principal component analysis algorithm is then used to extract deviation characteristics by comparing the nameplate rated cooling capacity. This allows the chiller's operating status to be expressed in the form of quantitative labels, improving the accuracy and standardization of equipment operation identification. This label is embedded in the graph structure as a device attribute and participates in network node status judgment. While ensuring multi-source information collaboration, traceable operating labels are introduced as a basis for state differences, enhancing the discriminative ability of anomaly detection. By extracting the device with the earliest anomaly and its upstream node, the starting range of anomaly analysis is effectively narrowed, and causal relationship analysis is guided to trace the source based on the impact of inter-device status, avoiding information redundancy and path generalization. A Bayesian network is used to construct a causal chain between the initial anomaly and the state evolution of adjacent nodes, allowing the potential impact paths between devices to be structured and bridging the gap between time series fluctuations and structural knowledge. A task priority assessment standard is established based on link starting point and path length, dynamically sorting and updating the status of devices along the fault propagation path, making system inspection and scheduling more targeted. The overall logic presents a step-by-step convergence process from quantitative modeling, structural mapping, anomaly focus, causal identification to priority handling. The participating items work closely together to ensure that the fault identification chain is efficient, closed-loop and scalable. Compared with the original means, it has achieved multi-dimensional gains such as improved state recognition granularity, convergence of anomaly analysis boundaries, clear fault tracing paths and precise configuration of inspection resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0048] Figure 1 is a system flow chart of the present invention;

[0049] Figure 2 is a system block diagram of the present invention;

[0050] Figure 3 Schematic diagram of the steps of the method of the present invention. DETAILED DESCRIPTION

[0051] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0052] In the embodiments of the present invention, words such as "exemplarily" and "including" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0053] In the embodiments of the present invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, when the distinction is not emphasized, the meanings they convey are the same. The terms "of," "corresponding," and "corresponding" may sometimes be used interchangeably. It should be noted that, when the distinction is not emphasized, the meanings they convey are the same.

[0054] In the embodiments of the present invention, sometimes a subscript such as W1 may be written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0055] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0056] The embodiment of the present invention provides a HVAC system inspection system based on knowledge graph, please refer to Figures 1 to 2 The present invention provides a technical solution, a knowledge graph-based HVAC system inspection system comprising:

[0057] The feature extraction module collects the chiller's inlet and outlet temperature and water flow data to calculate the chiller's actual cooling capacity. It then uses the principal component analysis algorithm to extract features based on the deviation between the actual cooling capacity and the nameplate rated cooling capacity. This generates a chiller operation label and passes it to the graph construction module.

[0058] The graph construction module constructs a graph structure with chillers, fan coil units, and heat exchangers as nodes. It assigns chiller operation tags to device attributes and determines node status differences. It then generates a list of candidate abnormal devices and passes it to the abnormality tracing module.

[0059] The anomaly tracing module obtains the anomaly occurrence time of the devices in the abnormal candidate device list, identifies the first abnormal device, generates the initial abnormal device set and passes it to the causal analysis module;

[0060] The causal analysis module obtains the status changes of the initial abnormal device set and adjacent devices, performs causal analysis on the status changes through the Bayesian network model, generates an abnormal causal link map, and passes it to the priority module;

[0061] The priority module extracts the link starting device and path length based on the abnormal causal chain graph, determines the inspection priority, updates the inspection status of the associated devices in the graph, and generates the inspection priority result.

[0062] The chiller operation label is the operation status category, deviation feature vector, and feature extraction time. The abnormal candidate device list includes the candidate device number, node attribute difference value, and graph position index. The initial abnormal device set includes the first abnormal device number, the time when the abnormality occurred, and the device abnormality type. The abnormal causal link graph refers to the causal path structure, state change dependency, and link node timing information. The inspection priority result is the inspection starting device number, path priority level, and associated device inspection status value.

[0063] See also Figure 2 , the feature extraction module includes:

[0064] The temperature acquisition submodule collects the water inlet temperature data, water outlet temperature data and instantaneous water flow data of the chiller, organizes the flow time accumulation and the temperature difference corresponding to the corresponding sampling time interval, and generates a temperature difference flow data table;

[0065] During the operation of the chiller, the PT100 temperature sensor installed at the water inlet collected three sets of water inlet temperature data at 1 second intervals: 25.3°C, 25.1°C, and 24.9°C. The same type of sensor at the water outlet simultaneously recorded water outlet temperature data at 20.8°C, 20.6°C, and 20.5°C. The electromagnetic flowmeter measured instantaneous flow rates of 0.48m³ / s, 0.49m³ / s, and 0.47m³ / s. The temperature difference between the three sets was 4. 5°C, 4.5°C, and 4.4°C are multiplied by the corresponding flow values, respectively, to obtain three product terms: 2.16 m³·°C / s, 2.205 m³·°C / s, and 2.068 m³·°C / s. The accumulated value (2.16 + 2.205 + 2.068) × 600 = 3877.8 m³·°C) is accumulated according to the 10-minute sampling period. This forms a structured data record including the timestamp and the accumulated temperature difference flow value, as shown in Table 1.

[0066] Table 1 Temperature difference flow data table

[0067]

[0068] As shown in Table 1, the product of temperature difference and flow rate is time-integrated in each recording cycle. When the flow rate count value fluctuates by more than ±5%, data verification is automatically triggered, and the data during the abnormal period is marked as pending review. Normal data is converted according to the ISO5167 standard and stored in the database.

[0069] The cooling capacity calculation submodule obtains the water specific heat capacity constant and water flow rate based on the temperature difference flow data table, calculates the cooling capacity value of the period based on the temperature difference, converts and integrates it, and obtains the cooling capacity value sequence of the period;

[0070] The specific heat capacity constant of water is 4.187, and the density is 998 kg / m³. The cumulative value of 3877.8 m³·°C at 10:00 in Table 1 is converted into energy value: 4.187×998×3877.8=16,123,456. Dividing it by the sampling interval of 600 seconds yields 26,872 kW. After conversion, the cooling capacity value for the period is 26,872 kW. The calculation results of three consecutive periods form a sequence of 26872, 27345, and 26418 kW. When it is detected that the fluctuation of the values ​​in adjacent periods exceeds ±8%, the average of the previous five periods is automatically called for smoothing.

[0071] The deviation analysis submodule normalizes the cooling capacity value sequence of each period with the rated cooling capacity value on the chiller nameplate, calculates the principal component vector features between the two using the principal component analysis algorithm, reconstructs the operating status characteristic indicators, and generates the chiller operation label;

[0072] Take the rated cooling capacity of the chiller nameplate As a benchmark, the instance data output by the cooling capacity calculation submodule is Perform normalization to obtain a normalized sequence . Rating sequence taken , exponential mean , the coefficient of variation is set to , according to GB / T18430.1-2007 standard. The covariance eigenvalue correction is taken , obtained through data regression analysis, the temporal gradient tensor norm is calculated as: ,in is the variance of the 24-hour period component, is the covariance term of the periodic component, n represents the total number of time series data points, corresponding to the three sampling points of the actual cooling capacity, m represents the total number of covariance eigenvalues, corresponding to the number of λ correction coefficients, m is 3, t represents the time variable, 0.72 corresponds to the 24-hour cycle load fluctuation variance, and 0.13 reflects the covariance of the peak and valley periods. The principal component vector calculation formula is: , the result Falls into the preset normal operating range , the system automatically generates a regular operation label, when 5 consecutive cycles The performance optimizer is triggered when

[0073] See also Figure 2 , the graph building modules include:

[0074] The node extraction submodule obtains the device identification, connection relationship and deployment information of the chiller, fan coil unit and heat exchanger equipment, as well as the device node group and connection path information, and establishes the device map connection information set;

[0075] Read the device registry from the building automation system and extract the chiller device ID (Example value: CH-2023-08-001), fan coil unit identification (Example value: FCU-5F-12), heat exchanger identification (Example value: HE-3C) parses the fields in the BACnet protocol message and obtains the connection relationship set , where DN150 means the pipe diameter is 150mm, DN200 means the pipe diameter is 200mm, and the adjacency matrix is ​​constructed ( Represents 5 device nodes), elements Representation device With equipment There is a physical connection, and a set of connected paths is generated through a depth-first search algorithm. , the path weight calculation formula is: ,in is the pipe length, is the pipe diameter (mm), is the material coefficient of the stainless steel pipe. In this example, the DN150 pipe is 35m long. Calculate the path weight have to: ;

[0076] Table 2 Equipment connection parameters

[0077]

[0078] As shown in Table 2, when the heat exchanger pressure value is detected Calculate the pressure threshold ratio when ,in , It is the maximum pressure value of DN200 pipe (according to GB / T14976-2012 standard). The early warning mechanism is triggered.

[0079] The attribute fusion submodule connects the information set with the chiller operation tag based on the device map, writes the chiller operation tag data into the corresponding node, fills the operation attribute fields of the remaining nodes according to the device type, organizes and marks them as identifiable status fields, and obtains the node operation attribute data group;

[0080] Set the chiller operation tag From the principal component analysis results, write to the node attribute set , collect the supply air temperature of the fan coil node and return air temperature , calculate the actual temperature difference: ,when And water flow When setting the status label , calculate the thermal efficiency of the heat exchanger node: ,when Time mark , forming a node attribute data set , Represents the heat exchanger identification, Chiller equipment identification, where the values ​​0.76 / 0.68 / 0.72 are normalized status values.

[0081] The difference identification submodule calculates the state deviation rate between each node and the chiller node state value based on the node operation attribute data group, determines whether there is a significant deviation in the operation state, filters the nodes, and obtains a list of abnormal candidate devices;

[0082] in is the reference state value of the chiller (from the principal component vector ), Calculate the fan coil deviation rate for other equipment status values : , set the threshold ,when When included in the exception list, the instance generates , synchronously calculate pipeline pressure ratio : ;when The second level warning is triggered when the pipeline pressure is close to the critical value.

[0083] See also Figure 2 , the exception tracing module includes:

[0084] The time extraction submodule obtains the operating status and corresponding timestamp information of the devices in the abnormal candidate device list, extracts the first moment when the device operating status changes as the abnormal time, combines the device number and time into a record entry, and establishes the abnormal occurrence time value set;

[0085] From the list of abnormal candidate devices Read the device status change record and extract the timestamp data , where device FCU-5F-12 is From normal to efficient heat exchange at any time, HE-3C-B When the efficiency changes from normal to inefficient, the device number and timestamp are combined into a record , establish an exception occurrence time value set When multiple status changes of the same device are detected within 5 minutes, only the first change record is retained.

[0086] Table 3 Abnormal time record table

[0087]

[0088] As shown in Table 3, the timestamp data comes from the NTP synchronized clock built into the device controller, with a time error of less than ±50ms. When the abnormal time difference between two devices is less than 1 second, it is determined to be a concurrent event and a timestamp check mark is added.

[0089] The first-time identification submodule sorts the abnormal time of all devices based on the abnormality occurrence time value set, selects the device corresponding to the earliest time as the starting device, determines whether there are multiple device records, and selects the device with the smallest number as the representative device, and generates the first abnormal device number;

[0090] Sort the time data in the abnormal occurrence time value set to obtain an ordered sequence , select the earliest time The corresponding device FCU-5F-12 is used as the candidate starting device. When the device HE-3C-01 also has an exception at 14:30:02, FCU-5F-12 and HE-3C-01 are compared in alphabetical order according to the device number. The FCU-5F-12 with the earlier alphabetical order is taken as the representative device to generate the first abnormal device number. At the same time, the device is checked for the number of abnormalities in the past 24 hours (2 times in this example). If the number exceeds the threshold by 3 times, a frequent abnormality mark is added.

[0091] The initial positioning submodule retrieves the corresponding device operating status and attribute information from the abnormal candidate device list based on the first abnormal device number, constructs the starting abnormal path entry and combines the first device attribute fields, marks the device as the initial abnormal node and establishes the initial identification tag to obtain the initial abnormal device set;

[0092] according to , Represents the device number determined by the dual rules of time sorting and lexicographic order, retrieving attribute data: supply air temperature , return air temperature , water flow , construct the exception propagation starting path entry , extract the attribute fields of pipe diameter DN150 and pipe length 35m, and mark the initial abnormal level of the node as , retrieve the status data of the three associated downstream nodes (AHU-5F-01, VAV-5F-12, DS-5F-02) and generate the initial abnormal device set Among them, AHU-5F-01 was included because of a 12% drop in air volume at 14:30:06.

[0093] See also Figure 2 , the causal analysis module includes:

[0094] The state change extraction submodule obtains the device numbers in the initial abnormal device set and the state change data of adjacent devices, extracts the corresponding relationship between the device state change time series and the state change of adjacent devices, organizes and marks the change moments of the devices, and generates a state change time series;

[0095] From the initial abnormal device set Extract device status change records, where: : Initial abnormal device set, element is device number, FCU-5F-12 at timestamp The supply air temperature drops (19.2°C → 18.5°C), and AHU-5F-01 The air volume decreases (1200→1056m³ / h). Search the adjacent device VAV-5F-12. The air valve opening changes (85%→70%), DS-5F-02 The pressure difference increases (150→180Pa), and a time series dataset is constructed:

[0096] Table 4 State change time series

[0097]

[0098] As shown in Table 4, the time difference between FCU and AHU is Second, seconds to mark a concurrent event.

[0099] The causal relationship construction submodule extracts the direct impact relationship between device state changes based on the state change time series, calculates the causal relationship strength of the impact path through the Bayesian network structure, and marks it as a causal relationship chain to obtain the causal relationship path;

[0100] Extract the direct impact relationship between device status changes, represents the conditional probability parameter of node A to B, : By counting the data from the previous 30 days, the conditional probability that device B will be abnormal within 5 minutes when device A is abnormal is calculated as 1. (In this example, A is abnormal 12 times, and B follows abnormally 10 times). 2. Indicates the time difference between the A / B node status changes. Seconds: Taken from the difference between AHU abnormal time and FCU abnormal time in Table 4, 3. represents the covariance coefficient between node A and its lth parent node, : represents the covariance coefficient between node A and the lth associated node, obtained by calculating the covariance of 24-hour data (the flow covariance between the chiller and FCU, 4. : Noise suppression parameters, according to the corresponding value of the ambient noise level N25 in the EN12101-6:2005 standard, 5. Second: Signal attenuation parameters, measured by AHU controller step response test, 63.2% attenuation time, 6. : Indicates the total number of associated nodes, including parent nodes and child nodes. In the current scenario, AHU is only directly affected by FCU. Calculation: ;when When it is determined to be a strong causal relationship (the threshold is based on ISO13374-3 standard), the path is generated .

[0101] The causal link generation submodule visualizes device state changes and causal relationships in a graph based on causal relationship paths, constructs a complete abnormal causal link graph, and annotates the abnormal causal impact range of each device to generate an abnormal causal link graph.

[0102] Defining visualization rules in the device topology diagram: Node size (mm), where abnormal level Value: FCU=1.5 (initial anomaly source), AHU=1.2, weight , in the example , the radius of influence (pixels), calculated Pixels, generate causal metadata: Abnormal propagation path: FCU-5F-12 (14:30:02) → AHU-5F-01 (14:30:06) → VAV-5F-12 (14:30:08), key parameter transmission: The supply air temperature drops by 0.7℃, resulting in a 12% decrease in air volume, which in turn causes the air valve opening to be adjusted by 15%.

[0103] See also Figure 2 , the priority module includes:

[0104] The path extraction submodule, based on the link information in the abnormal causal chain graph, calibrates the starting and ending devices according to the node sequence, obtains the first device number in the path, and reads the path nodes in sequence according to the connectivity relationship of the edges in the graph to obtain the link node sequence data;

[0105] Read the path from the abnormal causal chain graph , starting device With termination device , traverse the graph edge set to obtain the intermediate node sequence , extract node attributes: FCU abnormal level , AHU ,VAV , record the total length of the path , path extraction is automatically interrupted when it is detected that the path includes a closed-loop structure (FCU→AHU→FCU).

[0106] Table 5 Path node attribute table

[0107]

[0108] As shown in Table 5, the influence coefficient is the causal strength value of the preceding node. When a node with a coefficient value lower than 0.6 appears in the path, it is marked as a suspicious path segment.

[0109] The length judgment submodule obtains the total number of link nodes in the link node sequence data, marks the priority link when the path length value is equal to the path reference length value, and generates the path priority judgment result based on the link starting device number information;

[0110] Set path base length (Based on the 80% abnormal propagation path length in the data), calculate the instance path length , compared , marked as a high priority path, when there are multiple paths ( Length 2.1), sorted by the following rules: 1. Length descending: 2. Starting device number lexicographical order: FCU-5F-12 generates the priority list before DS-5F-02 , and is the path number. When the path length difference is less than 0.5, the path including more key equipment (chillers) is preferred.

[0111] The status update submodule locates the device nodes that are connected to the link path in the map based on the path priority determination result, updates the value of the device node status label marked as the priority path, writes the corresponding node attribute field, and generates the inspection priority result;

[0112] According to the priority list, positioning Path associated equipment, retrieve its current status tags: FCU-5F-12: "High-efficiency heat exchange", AHU-5F-01: "Insufficient air volume", VAV-5F-12: "Valve abnormality", set the priority coefficient (Path length 3.5 / benchmark 3.0 = 1.17, rounded to 1.5) is written into the node attribute field. The updated status labels become: FCU-5F-12: "High-efficiency heat exchange | Priority 1.5", AHU-5F-01: "Insufficient air volume | Priority 1.5", VAV-5F-12: "Valve abnormality | Priority 1.5". At the same time, the normal device DS-5F-02 directly connected to these nodes is retrieved, its monitoring frequency is increased from 5 minutes / time to 2 minutes / time, and the temperature sampling accuracy is improved from ±0.5°C to ±0.2°C, generating a patrol inspection priority result.

[0113] See also Figure 3 ,The method of the inspection system of the HVAC system based on the knowledge graph includes:

[0114] S1: Collect the chiller's inlet and outlet temperature and water flow data to calculate the chiller's actual cooling capacity. Use the principal component analysis algorithm to extract features based on the deviation between the actual cooling capacity and the nameplate rated cooling capacity, and generate the chiller's operating label.

[0115] S2: Construct a graph structure with chillers, fan coil units, and heat exchangers as nodes. Assign chiller operation labels to device attributes, determine node status differences, and generate a list of candidate abnormal devices.

[0116] S3: Obtain the abnormal occurrence time of the device in the abnormal candidate device list, identify the first abnormal device, and generate the initial abnormal device set;

[0117] S4: Obtain the status changes of the initial abnormal device set and adjacent devices, perform causal analysis on the status changes using a Bayesian network model, and generate an abnormal causal link graph;

[0118] S5: Extract the link starting device and path length based on the abnormal causal chain graph, determine the inspection priority, update the inspection status of the associated devices in the graph, and generate the inspection priority result.

[0119] It should be understood that the term "and / or" as used herein simply describes an association between related objects, indicating three possible relationships: A and / or B. This can mean: A exists alone, A and B exists simultaneously, or B exists alone. A and B can be singular or plural. Furthermore, the character " / " as used herein generally indicates an "or" relationship between the related objects, but it may also indicate an "and / or" relationship. For specific understanding, please refer to the context.

[0120] In this disclosure, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, "at least one of a, b, or c" can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or plural.

[0121] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0122] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be performed using electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to perform the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0123] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described equipment, devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0124] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. Including, the device embodiments described above are merely illustrative, including, the division of units, which is only a logical function division. In actual implementation, there may be other division methods, including multiple units or components that can be combined or integrated into another device, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0125] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0126] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0127] If the function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or the portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the method of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0128] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A knowledge graph-based HVAC system inspection system, characterized in that: The system comprises: The feature extraction module collects the chiller's inlet and outlet temperature and water flow data to calculate the chiller's actual cooling capacity. It uses the principal component analysis algorithm to extract features based on the deviation between the actual cooling capacity and the nameplate rated cooling capacity, generates a chiller operation label, and passes it to the graph construction module. The feature extraction module includes: The temperature acquisition submodule collects the water inlet temperature data, water outlet temperature data and instantaneous water flow data of the chiller, organizes the flow time accumulation and the temperature difference corresponding to the corresponding sampling time interval, and generates a temperature difference flow data table; The cooling capacity calculation submodule obtains the water specific heat capacity constant and the water flow rate based on the temperature difference flow rate data table, calculates the cooling capacity value of the time period in combination with the temperature difference, converts and integrates the values, and obtains a cooling capacity value sequence of the time period; The deviation analysis submodule normalizes the cooling capacity value sequence of the time period with the rated cooling capacity value on the chiller nameplate, calculates the principal component vector features between the two using the principal component analysis algorithm, reconstructs the operating status characteristic index, and generates the chiller operation label; A graph construction module constructs a graph structure with chillers, fan coil units, and heat exchangers as nodes, assigns the chiller operation labels to device attributes, determines node status differences, generates a list of abnormal candidate devices, and passes it to the abnormality tracing module; The abnormality tracing module obtains the abnormality occurrence time of the devices in the abnormal candidate device list, identifies the first abnormal device, generates an initial abnormal device set and passes it to the causal analysis module; A causal analysis module obtains the state changes of the initial abnormal device set and adjacent devices, performs causal analysis on the state changes through a Bayesian network model, generates an abnormal causal link graph, and transmits it to the priority module; The priority module extracts the link starting device and path length based on the abnormal causal chain map, determines the inspection priority, updates the inspection status of the associated devices in the map, and generates an inspection priority result.

2. The inspection system for HVAC system based on knowledge graph according to claim 1 is characterized in that: The chiller operation label is the operation status category, deviation feature vector, and feature extraction time. The abnormal candidate device list includes the candidate device number, node attribute difference value, and graph position index. The initial abnormal device set includes the first abnormal device number, the time when the abnormality occurred, and the device abnormality type. The abnormal causal link graph refers to the causal path structure, state change dependency, and link node timing information. The inspection priority result is the inspection starting device number, path priority level, and associated device inspection status value.

3. The inspection system for HVAC system based on knowledge graph according to claim 1 is characterized in that: The principal component vector feature between the two is calculated by the principal component analysis algorithm , using the formula: ； in, represents the normalized cooling capacity value of the i-th time point, represents the sliding mean of the cooling capacity series, Represents the dynamic standard deviation of the cooling capacity series, represents the normalized nameplate rated cooling capacity value corresponding to the i-th time point, represents the exponential mean of the rating series, represents the coefficient of variation of the rating series, represents the correction amount of the j-th eigenvalue of the covariance, Represents the time gradient tensor of the eigenvector space, n represents the total number of time series data points, m represents the total number of covariance eigenvalues, and t represents the time variable.

4. The inspection system for HVAC system based on knowledge graph according to claim 1 is characterized in that: The graph construction module includes: The node extraction submodule obtains the device identification, connection relationship and deployment information of the chiller, fan coil unit and heat exchanger equipment, as well as the device node group and connection path information, and establishes the device map connection information set; The attribute fusion submodule writes the chiller operation tag data into the corresponding node based on the device map connection information set and the chiller operation tag, fills the operation attribute fields of the remaining nodes according to the device type, organizes and marks them as identifiable status fields, and obtains the node operation attribute data group; The difference identification submodule calculates the state deviation rate between each node and the chiller node state value based on the node operation attribute data group, determines whether there is an obvious deviation in the operation state, filters the nodes, and obtains a list of abnormal candidate devices.

5. The inspection system for HVAC system based on knowledge graph according to claim 1 is characterized in that: The exception tracing module includes: The time extraction submodule obtains the operating status and corresponding timestamp information of the devices in the abnormal candidate device list, extracts the first moment when the device operating status changes as the abnormal time, combines the device number and time into a record entry, and establishes an abnormal occurrence time value set; The first-time identification submodule sorts the abnormal time of all devices based on the abnormal occurrence time value set, selects the device corresponding to the earliest time as the starting device, determines whether there are multiple device records and selects the device with the smallest number as the representative device, and generates the first abnormal device number; The initial positioning submodule retrieves the corresponding device operating status and attribute information from the abnormal candidate device list according to the first abnormal device number, constructs the starting abnormal path entry and combines the first device attribute field, marks the device as the initial abnormal node and establishes the initial identification tag to obtain the initial abnormal device set.

6. The inspection system for HVAC system based on knowledge graph according to claim 1 is characterized in that: The causal analysis module includes: The state change extraction submodule obtains the device numbers in the initial abnormal device set and the state change data of adjacent devices, extracts the corresponding relationship between the device state change time series and the state change of adjacent devices, organizes and marks the change moments of the devices, and generates a state change time series; A causal relationship construction submodule extracts the direct impact relationship between device state changes based on the state change time series, calculates the causal relationship strength of the impact path through the Bayesian network structure, and marks it as a causal relationship chain to obtain the causal relationship path; The causal link generation submodule visualizes the device state changes and causal relationships in the graph based on the causal relationship path, constructs a complete abnormal causal link graph, and marks the abnormal causal impact range of each device to generate an abnormal causal link graph.

7. The inspection system for HVAC system based on knowledge graph according to claim 6 is characterized in that: The causal relationship strength of the impact path is calculated by the Bayesian network structure , using the formula: ； in, represents the conditional probability parameter of node A to B, Indicates the time difference between the A / B node status changes. represents the covariance coefficient between node A and the lth associated node, represents the noise suppression parameter of Node B, represents the signal attenuation parameter of Node B, Indicates the total number of associated nodes, including parent nodes and child nodes.

8. The inspection system for HVAC system based on knowledge graph according to claim 1 is characterized in that: The priority module includes: The path extraction submodule, based on the link information in the abnormal causal chain graph, calibrates the starting device and the ending device according to the node sequence, obtains the first device number in the path, and reads the path nodes in sequence according to the connectivity relationship of the edges in the graph to obtain the link node sequence data; A length judgment submodule obtains the total number of link nodes in the link node sequence data, marks a priority link when the path length value is equal to the path reference length value, and generates a path priority judgment result based on the link starting device number information; The status update submodule locates the device nodes that are connected to the link path in the map according to the path priority determination result, updates the value of the device node status label marked as the priority path, writes the corresponding node attribute field, and generates the inspection priority result.

9. A knowledge graph-based HVAC system inspection method, characterized in that: The method is used to implement the inspection system of the HVAC system based on the knowledge graph according to any one of claims 1 to 8, and the method includes: S1: Collect the chiller's inlet and outlet temperature and water flow data to calculate the chiller's actual cooling capacity. Use the principal component analysis algorithm to extract features based on the deviation between the actual cooling capacity and the nameplate rated cooling capacity, and generate the chiller's operating label. S2: Construct a graph structure with chillers, fan coil units, and heat exchangers as nodes, assign the chiller operation labels to device attributes, determine node status differences, and generate a list of abnormal candidate devices; S3: Obtain the abnormality occurrence time of the device in the abnormal candidate device list, identify the first abnormal device, and generate an initial abnormal device set; S4: Obtaining the state changes of the initial abnormal device set and adjacent devices, performing causal analysis on the state changes using a Bayesian network model, and generating an abnormal causal link graph; S5: Based on the abnormal causal chain graph, the link starting device and path length are extracted, the inspection priority is determined, the inspection status of the associated devices in the graph is updated, and the inspection priority result is generated.

Images