Air conditioning equipment controller and self-diagnosis method thereof

Abstract

The invention provides an air conditioning equipment controller and a self-diagnosis method thereof, and relates to the technical field of air conditioning equipment. According to the method, multi-source real-time operation data of the air conditioning equipment is obtained, and operation mode sequences corresponding to different sensor nodes are generated through a data-driven working condition mode recognition and segmentation processing algorithm; constructing a cross-node mode evolution sequence diagram based on the operation mode sequence of each node, and analyzing a propagation path of an abnormal mode; further constructing a structural causal model and identifying key cause nodes in the abnormal mode propagation path by using causal intervention analysis; and a fault propagation topology network is established and a core propagation path is determined, so that an accurate active maintenance decision scheme is output, and active diagnosis and accurate positioning of the fault of the air conditioning equipment are realized.

Description

Technical Field

[0001] This invention relates to the field of air conditioning equipment technology, and in particular to an air conditioning equipment controller and its self-diagnosis method. Background Technology

[0002] As air conditioning equipment gradually develops towards intelligence and efficiency, the demand for operational safety and stability is also increasing. Traditional air conditioning equipment diagnosis methods usually rely on post-event maintenance or periodic manual inspections. This passive diagnosis method not only makes it difficult to make timely and accurate predictions of the equipment's operating status, but also increases the difficulty and cost of equipment maintenance.

[0003] Currently, some air conditioning equipment manufacturers are beginning to use sensors to collect real-time operating data and monitor its status. However, most solutions are limited to simple threshold comparisons or analysis of a single data source, ignoring the complex coupling relationships and abnormal mode propagation paths within the air conditioning system. This results in a need to improve the reliability and accuracy of diagnosis. Furthermore, existing technologies generally lack the ability to deeply explore and model the evolution patterns of abnormal modes and fault propagation paths, making it difficult to accurately identify the root cause of faults. This severely restricts the development of intelligent maintenance and early warning technologies for air conditioning equipment.

[0004] Therefore, there is an urgent need to provide a technical solution that can proactively and accurately diagnose abnormal conditions and analyze the propagation path of air conditioning equipment in order to improve the reliability and maintenance efficiency of the equipment. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes an air conditioning equipment controller and its self-diagnosis method.

[0006] To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, the present invention provides an air conditioning equipment controller, comprising: Pattern recognition module: used to acquire multi-source real-time operating data of air conditioning equipment, perform operating condition pattern recognition and segmentation processing on the multi-source real-time operating data, and generate operating mode sequences corresponding to different sensor nodes; The graph analysis module is used to construct a cross-node pattern evolution time sequence graph based on the operating mode sequence of each sensor node, and analyze the spatiotemporal evolution order in the pattern evolution time sequence graph to determine the propagation path of abnormal modes. Causal analysis module: used to construct a structural causal model based on the propagation path of the abnormal pattern; and to identify key triggering nodes in the propagation path of the abnormal pattern by analyzing the intervention effects in the structural causal model. Decision output module: used to construct a fault propagation topology network based on the key inducing nodes, mine the structural transmission rules in the fault propagation topology network to identify the core propagation path, and output a device proactive maintenance decision scheme based on the core propagation path.

[0007] Furthermore, the generation of the operating mode sequence corresponding to different sensor nodes includes: The multi-source real-time operation data is grouped according to the physical affiliation of the sensor nodes to form a node-level operation data set; In the node-level running data set, identify the structural change segments of the running status on the time axis, and use the structural change segments as the running status segment boundaries; Based on the segmentation boundaries of the operating status, the node-level operating data set is reorganized into segments to generate multiple operating mode fragments; Anomaly denoising is performed on the operation mode segments corresponding to the same sensor node, and they are spliced ​​together in the order of occurrence to form an operation mode sequence with state attribute labels, each corresponding to a different sensor node.

[0008] Furthermore, the identification of structural change segments of the operating state on the time axis includes: A state succession description sequence is constructed for the node-level operational data set to characterize the changing relationship of operational states between adjacent time periods; Detect the position in the state succession description sequence where the succession relationship transitions from a stable state to an unstable state; The detected location is identified as a structural change segment in the operating state.

[0009] Furthermore, the construction of a cross-node mode evolution time sequence diagram based on the operating mode sequence of each sensor node includes: Extract mode switching events from the operating mode sequence of each sensor node and record the order in which these events occur on the timeline. Identify combinations of events with sequential dependencies among mode switching events at different sensor nodes; Map events with dependencies to directed connections between nodes; Based on the directed connection relationship, a cross-node pattern evolution time sequence diagram is constructed.

[0010] Furthermore, the process of determining the propagation path of the abnormal mode is as follows: Extract multiple candidate propagation paths from the source node to the end node from the pattern evolution time sequence graph; Check the consistency of the node order in the candidate propagation path and eliminate the candidate propagation path with order conflicts; Propagation candidate paths that pass the sequential consistency check are identified as anomalous mode propagation paths.

[0011] Furthermore, the construction process of the structural causal model includes: Sensor nodes in the abnormal mode propagation path are used as causal variables, and the propagation direction between nodes is used as a causal constraint. Construct the initial causal dependency structure between nodes based on causal constraints; The causal connections with loops in the initial causal dependency structure are decomposed, and the decomposed causal connections are organized into a structural causal model.

[0012] Furthermore, the method for identifying key trigger nodes in the abnormal pattern propagation path includes: In the structural causal model, the do operator is used to apply single-node replacement interventions to different sensor nodes, generating corresponding intervention propagation results. By comparing the structural differences between the propagation results of intervention and those of no intervention in the propagation path of abnormal patterns, sensor nodes that cause the abnormal pattern propagation path to break down are identified as key inducing nodes.

[0013] Furthermore, the step of constructing a fault propagation topology network based on the key inducing nodes includes: Using key trigger nodes as starting nodes and combining the propagation direction relationships in the abnormal mode propagation path, an initial topology structure for fault propagation is constructed. Introduce reachability constraints in the initial topology of fault propagation to eliminate node connections that do not have propagation continuity; The retained node connections are hierarchically organized to form a fault propagation topology network with hierarchical relationships.

[0014] Furthermore, the method for identifying the core propagation path includes: Extract multiple candidate failure propagation paths originating from key trigger nodes in the failure propagation topology network; Analyze the node coverage relationships in the candidate paths for fault propagation to identify redundant paths that are completely contained by other paths; Fault propagation candidate paths that are not fully included by other paths are identified as core propagation paths in the fault propagation topology network.

[0015] Secondly, the present invention provides a self-diagnostic method for air conditioning equipment, which is based on the above-mentioned implementation of an air conditioning equipment controller, comprising: Acquire multi-source real-time operating data of air conditioning equipment, perform operating condition mode recognition and segmentation processing on the multi-source real-time operating data, and generate operating mode sequences corresponding to different sensor nodes respectively; Based on the operating mode sequence of each sensor node, a cross-node mode evolution time sequence diagram is constructed, and the spatiotemporal evolution order in the mode evolution time sequence diagram is analyzed to determine the propagation path of abnormal modes. A structural causal model is constructed based on the aforementioned abnormal pattern propagation path; by analyzing the intervention effects in the structural causal model, key triggering nodes in the abnormal pattern propagation path are identified. Based on the key inducing nodes, a fault propagation topology network is constructed, and the structural transmission patterns in the fault propagation topology network are mined to identify the core propagation path; and based on the core propagation path, a proactive maintenance decision scheme for the equipment is output.

[0016] Compared with the prior art, the beneficial effects of the present invention are: This invention identifies and automatically segments multi-source real-time operating data of air conditioning equipment, which can accurately capture the inherent structural characteristics of the operating state changing over time. This avoids the diagnostic lag caused by manually setting state boundary points in the traditional method, and effectively improves the real-time perception capability of equipment operating abnormalities.

[0017] This invention constructs a cross-node pattern evolution time sequence diagram to achieve temporal consistency analysis of the propagation path of abnormal patterns, accurately depicting the spatiotemporal propagation process of abnormal states between different sensor nodes, thereby improving the accuracy and comprehensiveness of fault diagnosis and reducing the risk of false alarms and missed alarms.

[0018] This invention, by constructing a structural causal model and introducing a quantitative analysis method for single-node replacement intervention, can accurately identify key causal nodes that cause abnormal operation of air conditioning equipment. This guides the construction of targeted proactive equipment maintenance strategies, improves maintenance efficiency, shortens repair response time, and effectively ensures long-term stable operation of the equipment. Attached Figure Description

[0019] Figure 1 This is an architecture diagram of an air conditioning equipment controller according to Example 1.

[0020] Figure 2 This is a flowchart of a self-diagnosis method for an air conditioning device, as shown in Example 2. Detailed Implementation

[0021] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example

[0022] Please see Figure 1The present invention provides an air conditioning equipment controller, comprising: Pattern recognition module: used to acquire multi-source real-time operating data of air conditioning equipment, perform operating condition pattern recognition and segmentation processing on the multi-source real-time operating data, and generate operating mode sequences corresponding to different sensor nodes; In specific implementation, the multi-source real-time operating data refers to continuous operating status parameters collected in real time by multiple sensors deployed at different functional modules and operating locations of the air conditioning equipment. These parameters include, but are not limited to, refrigerant inlet and outlet temperatures, compressor motor current, fan speed, condenser and evaporator temperatures, and inlet and outlet air temperature and humidity. In this embodiment, each sensor node is preferably located in a sensitive area of ​​the air conditioning equipment's operating conditions, including the compressor component area, heat exchange component area, and electronic control component area. The process of identifying and segmenting the multi-source real-time operating data refers to using a data-driven time series segmentation algorithm (such as a Bayesian online segmentation algorithm) to divide continuous time series data into discrete state segments with a single physical meaning or stable statistical characteristics based on the statistical consistency of the data distribution. This allows for the extraction of operating modes reflecting different steady-state or transitional states of the air conditioning system.

[0023] It should be noted that the sensor node uses a temperature sensor with an accuracy higher than ±0.5%, a high-precision Hall current sensor, and a high-sensitivity humidity sensor to continuously collect air conditioner operating parameters in real time. The specific collection frequency is once per second to ensure the real-time and comprehensiveness of the data.

[0024] The generation of the operating mode sequence corresponding to different sensor nodes includes: The multi-source real-time operation data is grouped according to the physical affiliation of the sensor nodes to form a node-level operation data set; In practice, based on the physical affiliation of the sensor nodes in the air conditioning equipment to different functional modules (such as compressor module, heat exchange module, electronic control module, etc.), the real-time collected multi-source operation data is initially classified and grouped into node-level operation data sets.

[0025] For example, for sensor nodes belonging to the compressor module, the collected data such as compressor motor current, compressor outlet pressure, and outlet temperature are classified into the node-level operating data set of the compressor module; similarly, for condenser module, the collected data such as condenser inlet and outlet temperature difference and fan speed are classified into the node-level operating data set of the condenser module.

[0026] In the node-level running data set, identify the structural change segments of the running status on the time axis, and use the structural change segments as the running status segment boundaries; It should be noted that, in order to accurately identify the structural changes in the operating conditions of air conditioning equipment over time and avoid the limitations of manually pre-setting segment intervals in traditional methods, this embodiment discloses a data-driven method for automatic identification of structural change segments.

[0027] Specifically, the identification of structural change segments of the operating state on the time axis includes: A state succession description sequence is constructed for the node-level operational data set to characterize the changing relationship of operational states between adjacent time periods; In specific implementation, the state succession description sequence is constructed by extracting multidimensional features and compressing the data in the node-level operational data set to create a data sequence that characterizes the state transition features between adjacent time periods. The compressed representation uses principal component analysis, linear discriminant analysis, or an autoencoder to perform dimensionality reduction mapping on the multidimensional feature vectors, extracting low-dimensional representations that reflect the essential characteristics of the operational state. Specific feature extraction methods include using a sliding time window statistical method to extract the mean, variance, and trend changes of operational parameters from the real-time data sequence, thereby constructing the state succession description vector.

[0028] Detect the position in the state succession description sequence where the succession relationship transitions from a stable state to an unstable state; In practice, the description vector obtained by sliding window statistics is compared with a preset steady-state threshold to detect the degree of deviation between the description vector and the threshold, so as to identify the transition position from a steady state to an unstable state.

[0029] For example, the preset steady state threshold can be obtained through statistical analysis of historical long-term stable operating data, such as taking the mean ± 2 standard deviations of the long-term stable operating state description vector as the steady state threshold; when the actual description vector is greater than or equal to the preset steady state threshold, it is determined that a structural state change has occurred.

[0030] The detected location is identified as a structural change segment in the operating state.

[0031] Furthermore, the identified state transition locations are used as the start and end boundaries of the structural change segments of the operating state, so that in the next step, these boundaries can be used to accurately divide the node-level operating data set and form operating mode segments with structural characteristics.

[0032] Based on the segmentation boundaries of the operating status, the node-level operating data set is reorganized into segments to generate multiple operating mode fragments; In implementation, the node-level data set is segmented according to the structural change zone boundaries determined above, and each data set is divided into multiple independent operating mode segments. The internal operating state structure of each mode segment is relatively consistent and stable.

[0033] It should be noted that the data format of each running mode segment is a two-dimensional matrix of time series length × feature dimension, where the feature dimension specifically includes multiple running parameters of the node.

[0034] Anomaly denoising is performed on the operation mode segments corresponding to the same sensor node, and they are spliced ​​together in the order of occurrence to form an operation mode sequence with state attribute labels, each corresponding to a different sensor node.

[0035] In specific implementation, after completing the above-mentioned segment reorganization, the first step is to identify and remove non-steady-state transition data at the moment of switching operating states. The specific identification logic is as follows: calculate the variance or slope within a preset window at the boundary of each operating mode segment. If the absolute value of the variance or slope is greater than or equal to the preset steady-state judgment threshold (for example, greater than or equal to 3 times the standard deviation of the normal fluctuation range of the operating parameter), then the segment is determined to be non-steady-state transition data affected by physical inertia and is removed. Subsequently, multiple operating mode segments belonging to the same sensor node and having stable structural characteristics are spliced ​​together according to the original timestamp order of acquisition. Specifically, during the splicing process, this embodiment also includes labeling the status attributes of each operating mode segment. The specific process is as follows: the feature vector of the operating mode segment is compared with a preset benchmark feature library, and its deviation from the benchmark operating condition features is calculated; if the deviation exceeds a preset abnormal threshold, the segment is labeled as an abnormal mode, otherwise it is labeled as a normal mode; finally, the start and end time metadata of each mode segment on the global time axis and the status attribute label are retained to form the operating mode sequence of each sensor node.

[0036] Understandably, this sequence not only preserves the evolutionary logic of the physical process, but also provides a data foundation for subsequent steps to analyze the spatiotemporal evolution of anomalous patterns across nodes by explicitly labeling anomalous pattern fragments.

[0037] The graph analysis module is used to construct a cross-node pattern evolution time sequence graph based on the operating mode sequence of each sensor node, and analyze the spatiotemporal evolution order in the pattern evolution time sequence graph to determine the propagation path of abnormal modes. It should be noted that the operating mode sequence is a sequence of data obtained for different sensor nodes, reflecting the change of the operating state of the node over time; each operating mode sequence consists of multiple operating mode segments arranged in chronological order, and each operating mode segment corresponds to the relatively stable operating state structure of the air conditioning equipment in a certain time period.

[0038] It is particularly important to note that different sensor nodes have different physical locations, functional attributes, and monitored objects, resulting in variations in the rhythm of change and the timing of mode switching in their operating mode sequences over time. This embodiment uses cross-node mode evolution analysis to characterize the propagation and evolution of abnormal operating states between different nodes, providing a temporal structure basis for subsequent causal analysis.

[0039] In implementation, the construction of a cross-node mode evolution time sequence diagram based on the operating mode sequence of each sensor node includes: Extract mode switching events from the operating mode sequence of each sensor node and record the order in which these events occur on the timeline. In specific implementation, the sequence of operating modes corresponding to each sensor node is traversed and analyzed. The mode switching event corresponds to the previously determined operating state segment boundary. Specifically, when two adjacent operating mode segments change in structural features, the change position is identified as a mode switching event. Each mode switching event includes at least the following information elements: the identifier of the sensor node to which it belongs, the identifier of the operating mode before the switch, the identifier of the operating mode after the switch, and the corresponding timestamp.

[0040] Furthermore, to ensure consistency in cross-node mode evolution analysis, mode switching events of different sensor nodes are uniformly mapped to the same global time axis, and all mode switching events are sorted according to the order of timestamps to form a global mode switching event sequence.

[0041] Identify combinations of events with sequential dependencies among mode switching events at different sensor nodes; In implementation, a sliding window analysis is performed on the global mode switching event sequence to identify combinations of mode switching events that are temporally adjacent or close neighbors between different sensor nodes. Furthermore, by combining the physical affiliation, functional association, and historical operation statistical characteristics of the sensor nodes, it is determined whether there is a sequential dependency relationship between the event combinations.

[0042] For example, if the operating mode of a compressor node changes abnormally within a preset time interval, the operating mode of the condenser node or evaporator node will change accordingly. In this case, it is determined that there is a sequential dependency between the mode switching event of the compressor node and the mode switching event of the subsequent node. The preset time interval is determined based on the thermodynamic response lag time of the air conditioning refrigeration cycle. Specifically, it can be obtained by statistical analysis of historical operating data. For example, 0.5 to 1.5 times the typical time required for the refrigerant to circulate once in the system can be used as the observation window to ensure that the identified dependency conforms to the laws of physical propagation.

[0043] Map events with dependencies to directed connections between nodes; It should be noted that for each group of events with sequential dependencies, the sensor node with the earlier event occurrence time is designated as the starting node, and the sensor node with the later event occurrence time is designated as the target node. Based on this, a directed connection is established between the nodes to represent the potential propagation direction of the change in operating mode.

[0044] It should be understood that the directed connection relationship does not require direct physical connection, but is used to characterize the evolutionary relationship of the operating state at the functional and temporal levels.

[0045] Based on the directed connection relationship, a cross-node pattern evolution time sequence diagram is constructed.

[0046] In implementation, each sensor node is treated as a node unit in the graph, and the above-mentioned directed connection relationship is treated as a directed edge in the graph, thus constructing a cross-node mode evolution time sequence graph. This mode evolution time sequence graph uses time sequence as an implicit constraint to reflect the evolution relationship of different sensor node operation modes in time and space dimensions.

[0047] After constructing the pattern evolution timeline, the timeline is further analyzed to determine the propagation path of the anomalous patterns.

[0048] Specifically, the process of determining the propagation path of the abnormal mode is as follows: Extract multiple candidate propagation paths from the source node to the end node from the pattern evolution time sequence graph; In specific implementation, nodes without incoming edges in the mode evolution time sequence graph, or nodes with the earliest mode switching event time and incoming edge weights lower than the preset strength threshold within the preset observation period, are defined as source nodes, and nodes without outgoing edges are defined as end nodes. Based on the graph traversal method, starting from each source node, multiple propagation candidate paths from the source node to each end node are extracted sequentially along the directed edges.

[0049] It should be noted that each propagation candidate path consists of multiple sensor nodes arranged in the direction of propagation, which are used to represent possible propagation links of the operating mode.

[0050] Check the consistency of the node order in the candidate propagation path and eliminate the candidate propagation path with order conflicts; It is understandable that the order of nodes in each propagation candidate path is consistent with the order of occurrence of the corresponding node mode switching events on the global timeline; when there is a discrepancy between the order of nodes in the path and the actual time order, the propagation candidate path is determined to have an order conflict and is eliminated.

[0051] It is worth mentioning that this step can effectively eliminate non-real propagation paths introduced by occasional synchronous fluctuations or statistical noise.

[0052] Propagation candidate paths that pass the sequential consistency check are identified as anomalous mode propagation paths.

[0053] Furthermore, the remaining propagation candidate paths that satisfy temporal consistency are used as anomalous mode propagation paths to characterize the actual propagation process of anomalous operating states between different sensor nodes of the air conditioning equipment, providing a reliable path constraint basis for the subsequent construction of structural causal models.

[0054] Causal analysis module: used to construct a structural causal model based on the propagation path of the abnormal pattern; and to identify key triggering nodes in the propagation path of the abnormal pattern by analyzing the intervention effects in the structural causal model. Specifically, the construction process of the structural causal model includes: Sensor nodes in the abnormal mode propagation path are used as causal variables, and the propagation direction between nodes is used as a causal constraint. In specific implementation, in this embodiment, each sensor node involved in the above-mentioned abnormal mode propagation path is used as the basic causal variable in the structural causal model, and the causal relationship between each causal variable is limited by the propagation direction of the nodes in the propagation path.

[0055] The causal variable represents the actual operating state or mode of the corresponding sensor node in the air conditioning equipment. The causal constraint condition is specifically manifested as the unidirectional propagation relationship of the abnormal mode from the source node to the target node. That is, if the operating mode of node A changes abnormally, it will directly cause the operating mode of node B to change accordingly. Then, a causal constraint relationship is formed between node A and node B, which is specifically manifested as a directed causal connection from node A to node B.

[0056] For example, if an abnormal change in the operating mode of the compressor outlet temperature sensor node directly causes an abnormal change in the operating mode of the condenser inlet temperature sensor node, then the compressor outlet temperature node can be used as a causal variable, and its propagation relationship to the condenser inlet temperature node can be used as a causal constraint, forming a one-way causal connection from the compressor outlet temperature node to the condenser inlet temperature node.

[0057] Construct the initial causal dependency structure between nodes based on causal constraints; In further implementation, based on the above causal variables and causal constraints, this embodiment uses the structural causal graph method to construct the initial causal dependency structure between nodes; specifically, the initial causal dependency structure is presented in the form of a directed acyclic graph, where each node represents the operating mode state of a specific sensor node, and the directed edges between nodes represent causal dependencies or causal influences.

[0058] It should be noted that the initial causal dependency structure constructed in this embodiment may form a loop structure due to the interrelation of different node operation modes in the actual system. Therefore, after the initial causal dependency structure is constructed, further causal loop identification and decomposition are required to ensure the accuracy and operability of subsequent intervention effect analysis.

[0059] The causal connections with loops in the initial causal dependency structure are decomposed, and the decomposed causal connections are organized into a structural causal model.

[0060] In order to avoid interference from the intervention analysis due to the existence of causal loops during the implementation process, this embodiment identifies and dismantles the causal loops that may exist in the initial causal dependency structure constructed above. The specific dismantling method is as follows: First, a depth-first search algorithm is used to traverse the initial causal dependency structure graph and identify all loop structures. Secondly, for each identified loop structure, the contribution intensity of each causal connection in the loop to the formation of the entire loop is further analyzed using the causal strength assessment method. Specifically, the causal strength assessment method uses quantitative analysis methods such as transfer entropy or Granger causality analysis to determine the strength of each connection. Subsequently, based on the chronological order of the mode switching events on the global timeline, the causal edges in the loop are subjected to temporal constraint verification. Specifically, in this embodiment, causal connections in the loop that violate the direction of the time arrow (i.e., pointing to past moments) or the connections with the lowest causal strength scores within the same time step are defined as weak causal connections, and these connections are decomposed. The specific decomposition logic is as follows: while retaining the physical loop causal feedback, the physical loop is transformed into a directed acyclic structure unfolded in the time dimension by introducing a time delay operator, or edges in the loop with contribution values ​​lower than a preset weight threshold are directly removed from the initial causal dependency structure graph until there are no logical loops in the entire causal structure graph. This processing ensures that the model meets the recursive requirements of the structural causal model (SCM), thereby supporting subsequent stable causal effect deduction.

[0061] After completing the above loop decomposition process, the resulting causal connections are reorganized to form the final structural causal model, which is used for subsequent intervention effect analysis.

[0062] To further explain, the final structural causal model in this embodiment is specifically represented as a directed acyclic graph structure, where nodes represent the operating mode states of each sensor node of the air conditioning equipment, and directed connections represent deterministic causal dependencies between nodes. The construction of the structural causal model lays the foundation for the accurate identification of key inducing nodes in the future.

[0063] Furthermore, the method for identifying key trigger nodes in the abnormal pattern propagation path includes: In the structural causal model, the do operator is used to apply single-node replacement interventions to different sensor nodes, generating corresponding intervention propagation results. In practical implementation, to accurately identify key triggering nodes in the propagation path of abnormal patterns, this embodiment simulates maintenance actions on air conditioning equipment components and uses a single-node replacement intervention strategy for quantitative analysis. The specific implementation steps are as follows: First, the do operator is used to atomically intervene in the target node in the structural causal model. Specifically, based on Pearl causal operator theory, the do operator is used to force the value of the target variable in the mathematical expression of the structural causal model. By truncating and deleting all incoming edges pointing to the target node in the computation graph, the functional dependency between the node and its parent node is logically severed, eliminating the causal contribution of upstream variables to the node. Subsequently, the observed value of the target node is replaced by the current abnormal feature vector with the normal operating benchmark mean vector extracted from the benchmark feature library, corresponding to the current environmental conditions.

[0064] By simulating controlled experiments, the node is forced to return to its normal state, which serves as the input condition for calculating the counterfactual outcome of the intervention. This allows the mathematical model to accurately isolate and quantify the marginal contribution of the node as the source of the fault to the downstream abnormal state.

[0065] Secondly, forward effect deduction is performed based on the structural causal model after intervention; that is, keeping the functional dependencies between nodes other than the target node unchanged, the chain effect of the target node's state change on the state distribution of downstream related nodes is calculated step by step according to the causal propagation direction, so as to obtain the abnormal pattern propagation results after intervention of each node.

[0066] Finally, by traversing and comparing, we record and identify the structural changes that occur in the original abnormal pattern propagation path due to intervention, specifically manifested as the disappearance of abnormal features of downstream nodes, interruption of propagation links, or shortening of propagation paths.

[0067] By comparing the structural differences between the propagation results of intervention and those of no intervention in the propagation path of abnormal patterns, sensor nodes that cause the abnormal pattern propagation path to break down are identified as key inducing nodes.

[0068] In practice, the structural difference analysis between the intervention and non-intervention outcomes specifically includes: First, the abnormal pattern propagation path structure under the uninterrupted state is defined as the baseline structure; Secondly, the transmission path structure obtained after the above intervention was compared with the baseline structure node by node and path by path. The differences in the transmission path structure caused by the intervention were recorded. The types of differences included transmission path breakage, transmission path shortening or reduction in the number of transmission nodes. Furthermore, the difference between the post-intervention propagation path structure and the baseline structure is quantified and expressed as a difference score. The specific calculation formula for the intervention effect score corresponding to each node is as follows: ; In the formula, Score(i) represents the intervention effect score of the i-th node on the abnormal propagation, |P base | represents the total number of sensor nodes included in the propagation path of the anomalous pattern under conditions of no intervention (original anomalous state), |P intervention (i)∣ represents the total number of sensor nodes in the propagation path of the i-th node and its downstream nodes that still exhibit abnormal modes after the i-th node applies normal state intervention.

[0069] It should be noted that the higher the intervention effect score (Score(i)), the better the effect of repairing or resetting the node in blocking abnormal propagation.

[0070] Finally, based on the magnitude of the node intervention effect score, the nodes that cause the abnormal mode propagation path to break or the number of propagation nodes to decrease are further identified as key inducing nodes. These key inducing nodes are the core factors leading to the propagation of abnormal modes in air conditioning equipment, and the specific node information serves as an important basis for subsequent proactive maintenance and control decisions.

[0071] It should be noted that, in this embodiment of the invention, the key trigger node represents the original fault source that triggers the chain propagation of abnormal patterns in the air conditioning equipment from a causal logic perspective. The corresponding physical state mutation is defined as the start of the fault occurrence. Furthermore, the abnormal evolution path starting from the key trigger node is physically mapped to the fault propagation topology network described in the subsequent steps.

[0072] Decision output module: used to construct a fault propagation topology network based on the key inducing nodes, mine the structural transmission rules in the fault propagation topology network to identify the core propagation path, and output a device proactive maintenance decision scheme based on the core propagation path.

[0073] In order to transform the identified key trigger nodes into operable maintenance paths, this embodiment performs causal weighting and simplification on the original pattern evolution relationship based on the core position of the key trigger nodes in the system.

[0074] In implementation, constructing the fault propagation topology network based on the key inducing nodes includes: Using key trigger nodes as starting nodes and combining the propagation direction relationships in the abnormal mode propagation path, an initial topology structure for fault propagation is constructed. In the specific implementation process, the aforementioned key inducing nodes are first used as the starting nodes, i.e., the source nodes, for the construction of the fault propagation topology network. Then, based on the propagation direction between nodes in the abnormal mode propagation path, the initial connection relationship between nodes is gradually established to form the initial topology structure.

[0075] Specifically, the initial topology in this embodiment is represented by a directed graph, where each node corresponds to the operating mode state of a specific sensor node of the air conditioning equipment, and the directed connection between nodes reflects the directional relationship of abnormal mode propagation between nodes. For example, if the abnormal change in compressor motor current is the key cause node, then this node is taken as the source node, and combined with the propagation order between each node in the aforementioned abnormal mode propagation path, subsequent related node connections are gradually added until the terminal node of the propagation path is reached, thereby forming an initial topology with a propagation direction relationship.

[0076] Introduce reachability constraints in the initial topology of fault propagation to eliminate node connections that do not have propagation continuity; In further implementation, to improve the accuracy of the fault propagation topology network structure and the effectiveness of the propagation path, this embodiment introduces propagation reachability constraints based on the initial topology structure. Specific implementation details are as follows: First, a propagation reachability analysis is performed on all node connections in the initial topology. Specifically, the propagation reachability analysis is to determine the logical rationality of the propagation of abnormal patterns between any two directly connected nodes and to determine whether the abnormal pattern can actually propagate to adjacent nodes. Secondly, based on historical data of abnormal mode propagation and actual observation records of node state transitions, the propagation reachability index between nodes is calculated. For example, the statistical probability of node B switching within the thermodynamically permissible time window after node A switches modes is calculated. For node connections with a propagation probability lower than a preset threshold (e.g., 0.1) or whose propagation delay logically violates the refrigeration cycle sequence (e.g., the change in terminal return air temperature precedes the change in compressor exhaust temperature), they are determined to lack propagation reachability.

[0077] Furthermore, node connections that do not meet the predetermined propagation reachability threshold, i.e., connections where the actual propagation probability of abnormal patterns between nodes is lower than the threshold or the propagation logic is unreasonable, are removed from the topology to ensure that the node connections retained in the topology network are all valid and continuous propagation path relationships.

[0078] The retained node connections are hierarchically organized to form a fault propagation topology network with hierarchical relationships.

[0079] In further implementation, after completing the above-mentioned propagation reachability constraint processing, this embodiment performs hierarchical organization on the retained valid node connections to clearly express the multi-level propagation relationship of air conditioning equipment fault propagation; the specific hierarchical organization method includes: First, the key trigger node is taken as the first level (source node level), and then the downstream nodes directly propagated from that node are determined as the second level, and so on, extending downwards level by level. Secondly, if a node receives propagation from multiple upstream nodes simultaneously, the minimum level to which the node belongs is determined based on the actual path length (i.e., the number of nodes) or the propagation delay of each upstream node to that node. Furthermore, the final fault propagation topology network is presented in the form of a multi-level directed tree or forest structure, where each node in each level corresponds to the propagation state at different stages in the propagation of the abnormal mode of the air conditioning equipment, thus clearly expressing the overall structural relationship of fault propagation.

[0080] In order to extract the most diagnostically valuable information from the complex fault evolution network, this embodiment performs path pruning and critical path extraction on the fault propagation topology network.

[0081] Specifically, the method for identifying the core propagation path includes: Extract multiple candidate failure propagation paths originating from key trigger nodes in the failure propagation topology network; In specific implementation, this embodiment selects key inducing nodes as starting nodes from the fault propagation topology network structure after the aforementioned hierarchical organization, and uses depth-first search (DFS) or breadth-first search (BFS) algorithms to fully traverse the network structure in order to extract all potential candidate propagation paths. Furthermore, each candidate propagation path records the complete node sequence from the key trigger node to the terminal node, forming a set of candidate paths.

[0082] For example, the specific candidate paths for fault propagation extracted in this embodiment are shown in Table 1.

[0083] Table 1 Candidate Paths for Fault Propagation

[0084] Analyze the node coverage relationships in the candidate paths for fault propagation to identify redundant paths that are completely contained by other paths; In further implementation, to eliminate redundant paths and highlight core propagation paths, this embodiment specifically analyzes the path coverage relationships in the candidate path set one by one. The analysis process is disclosed as follows: First, the node set of each path is recorded and marked, and then the inclusion and contained relationships between the node sets of different paths are checked one by one. Secondly, propagation paths whose node set is completely contained by other paths are marked as redundant paths. Such paths do not have the core value of fault propagation identification. For example, if the node set {node1, node2, node3} of path A is completely contained by the node set {node1, node2, node3, node4} of path B, then path A is identified as a redundant path.

[0085] Fault propagation candidate paths that are not fully included by other paths are identified as core propagation paths in the fault propagation topology network.

[0086] In implementation, all paths marked as redundant are removed from the candidate path set. After the redundant paths are removed, the final set of paths retained in this embodiment is the core propagation path. The core propagation path is the key path that best represents the main law of the propagation of abnormal modes of air conditioning equipment. The specific path information includes the node sequence from the key inducing node to the terminal node.

[0087] For example, the core propagation path determined in this embodiment is shown in Table 2.

[0088] Table 2 Core Propagation Paths

[0089] It should be noted that the determination of the core propagation path provides a precise basis for generating subsequent proactive maintenance decision-making schemes for air conditioning equipment. The proactive maintenance decision-making scheme is a targeted proactive maintenance and repair strategy based on the core propagation path, specifically including: establishing physical component repair priorities according to the causal hierarchy in the core propagation path, that is, prioritizing the repair of physical hardware corresponding to key causal nodes (e.g., compressor inverter current detection circuits), and predicting the expected recovery time of downstream nodes based on the transmission delay of the core propagation path, thereby dynamically adjusting the monitoring alarm thresholds of abnormal nodes; through this sequential maintenance based on causal logic, the accuracy of fault diagnosis and the effectiveness of maintenance measures are ensured.

[0090] Example 2 like Figure 2 As shown, parts not described in detail in this embodiment are described in Embodiment 1. This embodiment provides a self-diagnostic method for air conditioning equipment, including: Acquire multi-source real-time operating data of air conditioning equipment, perform operating condition mode recognition and segmentation processing on the multi-source real-time operating data, and generate operating mode sequences corresponding to different sensor nodes respectively; Based on the operating mode sequence of each sensor node, a cross-node mode evolution time sequence diagram is constructed, and the spatiotemporal evolution order in the mode evolution time sequence diagram is analyzed to determine the propagation path of abnormal modes. A structural causal model is constructed based on the aforementioned abnormal pattern propagation path; by analyzing the intervention effects in the structural causal model, key triggering nodes in the abnormal pattern propagation path are identified. Based on the key inducing nodes, a fault propagation topology network is constructed, and the structural transmission patterns in the fault propagation topology network are mined to identify the core propagation path; and based on the core propagation path, a proactive maintenance decision scheme for the equipment is output.

[0091] Some of the data in the above formula are calculated by removing dimensions and taking their numerical values. The formula is the closest to the real situation obtained by software simulation of a large amount of collected data. The preset parameters and preset thresholds in the formula are set by those skilled in the art according to the actual situation or obtained through simulation of a large amount of data.

[0092] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. An air conditioning equipment controller, characterized in that, The controller includes: Pattern recognition module: used to acquire multi-source real-time operating data of air conditioning equipment, perform operating condition pattern recognition and segmentation processing on the multi-source real-time operating data, and generate operating mode sequences corresponding to different sensor nodes; The graph analysis module is used to construct a cross-node pattern evolution time sequence graph based on the operating mode sequence of each sensor node, and analyze the spatiotemporal evolution order in the pattern evolution time sequence graph to determine the propagation path of abnormal modes. Causal analysis module: used to construct a structural causal model based on the propagation path of the abnormal pattern; and to identify key triggering nodes in the propagation path of the abnormal pattern by analyzing the intervention effects in the structural causal model. Decision output module: used to construct a fault propagation topology network based on the key inducing nodes, mine the structural transmission rules in the fault propagation topology network to identify the core propagation path, and output a device proactive maintenance decision scheme based on the core propagation path.

2. An air conditioning equipment controller according to claim 1, characterized in that, The generation of the operating mode sequence corresponding to different sensor nodes includes: The multi-source real-time operation data is grouped according to the physical affiliation of the sensor nodes to form a node-level operation data set; In the node-level running data set, identify the structural change segments of the running status on the time axis, and use the structural change segments as the running status segment boundaries; Based on the segmentation boundaries of the operating status, the node-level operating data set is reorganized into segments to generate multiple operating mode fragments; Anomaly denoising is performed on the operation mode segments corresponding to the same sensor node, and they are spliced ​​together in the order of occurrence to form an operation mode sequence with state attribute labels, each corresponding to a different sensor node.

3. An air conditioning equipment controller according to claim 2, characterized in that, The identification of structural change segments in the operating state on the time axis includes: A state succession description sequence is constructed for the node-level operational data set to characterize the changing relationship of operational states between adjacent time periods; Detect the position in the state succession description sequence where the succession relationship transitions from a stable state to an unstable state; The detected location is identified as a structural change segment in the operating state.

4. An air conditioning equipment controller according to claim 1, characterized in that, The construction of a cross-node mode evolution time sequence diagram based on the operating mode sequence of each sensor node includes: Extract mode switching events from the operating mode sequence of each sensor node and record the order in which these events occur on the timeline. Identify combinations of events with sequential dependencies among mode switching events at different sensor nodes; Map events with dependencies to directed connections between nodes; Based on the directed connection relationship, a cross-node pattern evolution time sequence diagram is constructed.

5. An air conditioning equipment controller according to claim 4, characterized in that, The process of determining the propagation path of the abnormal mode is as follows: Extract multiple candidate propagation paths from the source node to the end node from the pattern evolution time sequence graph; Check the consistency of the node order in the candidate propagation path and eliminate the candidate propagation path with order conflicts; Propagation candidate paths that pass the sequential consistency check are identified as anomalous mode propagation paths.

6. An air conditioning equipment controller according to claim 1, characterized in that, The construction process of the structural causal model includes: Sensor nodes in the abnormal mode propagation path are used as causal variables, and the propagation direction between nodes is used as a causal constraint. Construct the initial causal dependency structure between nodes based on causal constraints; The causal connections with loops in the initial causal dependency structure are decomposed, and the decomposed causal connections are organized into a structural causal model.

7. An air conditioning equipment controller according to claim 6, characterized in that, The method for identifying key trigger nodes in the abnormal pattern propagation path includes: In the structural causal model, the do operator is used to apply single-node replacement interventions to different sensor nodes, generating corresponding intervention propagation results. By comparing the structural differences between the propagation results of intervention and those of no intervention in the propagation path of abnormal patterns, sensor nodes that cause the abnormal pattern propagation path to break down are identified as key inducing nodes.

8. An air conditioning equipment controller according to claim 1, characterized in that, The construction of the fault propagation topology network based on the key inducing nodes includes: Using key trigger nodes as starting nodes and combining the propagation direction relationships in the abnormal mode propagation path, an initial topology structure for fault propagation is constructed. Introduce reachability constraints in the initial topology of fault propagation to eliminate node connections that do not have propagation continuity; The retained node connections are hierarchically organized to form a fault propagation topology network with hierarchical relationships.

9. An air conditioning equipment controller according to claim 8, characterized in that, The method for identifying the core propagation path includes: Extract multiple candidate failure propagation paths originating from key trigger nodes in the failure propagation topology network; Analyze the node coverage relationships in the candidate paths for fault propagation to identify redundant paths that are completely contained by other paths; Fault propagation candidate paths that are not fully included by other paths are identified as core propagation paths in the fault propagation topology network.

10. A self-diagnostic method for an air conditioning device, based on the implementation of an air conditioning device controller according to any one of claims 1-8, characterized in that, include: Acquire multi-source real-time operating data of air conditioning equipment, perform operating condition mode recognition and segmentation processing on the multi-source real-time operating data, and generate operating mode sequences corresponding to different sensor nodes respectively; Based on the operating mode sequence of each sensor node, a cross-node mode evolution time sequence diagram is constructed, and the spatiotemporal evolution order in the mode evolution time sequence diagram is analyzed to determine the propagation path of abnormal modes. A structural causal model is constructed based on the aforementioned abnormal pattern propagation path; by analyzing the intervention effects in the structural causal model, key triggering nodes in the abnormal pattern propagation path are identified. Based on the key inducing nodes, a fault propagation topology network is constructed, and the structural transmission patterns in the fault propagation topology network are mined to identify the core propagation path; and based on the core propagation path, a proactive maintenance decision scheme for the equipment is output.

Images