Building equipment fault prediction method and system

Abstract

The invention discloses a building equipment fault prediction method and system, and belongs to the technical field of building equipment fault prediction, and the method specifically comprises the steps: collecting operation parameters, environment parameters and operation and maintenance data of building equipment, forming a multi-source heterogeneous data set, carrying out the cross-modal fusion of the multi-source heterogeneous data set, and carrying out the cross-modal fusion of the multi-source heterogeneous data set. Extracting key features of the building equipment by combining space-time constraints of building space topology and building equipment distribution, identifying a causal chain of a building equipment fault by using a causal reasoning model based on the key features, and predicting a fault state of single building equipment at a local level through a hierarchical decision prediction model. According to the method, the operation trend of the cross-building equipment group is predicted on the global level, and the operation adjustment scheme of the single-building equipment is generated on the local level according to the fault prediction result, so that improvement from single-point monitoring to global prediction is realized, and group equipment fault prediction requirements in a complex scene are met.

Description

Technical Field

[0001] This invention belongs to the field of building equipment failure prediction technology, specifically a building equipment failure prediction method and system. Background Technology

[0002] Building equipment, as an important component of modern buildings, includes air conditioning units, elevator systems, water supply and drainage systems, power distribution equipment, and fire protection facilities. The stability and reliability of its operation are crucial to the normal functioning of the building as a whole and the safety of personnel. With the continuous increase in building scale and intelligence, the number and types of building equipment are becoming increasingly complex, significantly increasing the difficulty of operation and maintenance management.

[0003] Currently, the management of building equipment mainly relies on the following methods: regular maintenance mode, which involves inspecting and maintaining equipment according to a preset cycle; threshold-based monitoring mode, which involves deploying sensors on the equipment to collect parameters such as temperature, current, and vibration in real time, and triggering alarms when the parameters exceed the threshold; and single-equipment prediction technology exploration, which uses machine learning or deep learning models to predict the failure of individual building equipment.

[0004] Existing methods applied to the field of building equipment generally suffer from the following problems: the operation of building equipment is constrained by the topology and regional distribution of building space, and existing methods are unable to reflect the complex coupling relationship brought about by spatial proximity and functional dependence, making it difficult to adapt to the needs of group equipment failure prediction in complex scenarios. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention proposes a method and system for predicting building equipment failures. This system can integrate multi-source data, possess causal reasoning capabilities, and predict building equipment failures at both local and global levels, thereby enabling intelligent management and efficient operation and maintenance of building equipment.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for predicting building equipment failures, comprising:

[0008] Collect operating parameters, environmental parameters, and maintenance data of building equipment to form a multi-source heterogeneous dataset;

[0009] Cross-modal fusion is performed on the multi-source heterogeneous dataset, and key features of building equipment are extracted by combining the spatiotemporal constraints of building spatial topology and building equipment distribution. These key features characterize the coupling relationship and fault evolution trend among building equipment.

[0010] Based on the aforementioned key features, a causal reasoning model is used to identify the causal chain of building equipment failures. A hierarchical decision prediction model is used to predict the failure status of a single building equipment at the local level and the operating trend of a group of building equipment at the global level.

[0011] Based on the fault prediction results, operation adjustment plans for individual building equipment are generated at the local level, and scheduling and optimization strategies for cross-building equipment groups are generated at the global level.

[0012] Specifically, cross-modal fusion is performed on the multi-source heterogeneous dataset, and key features of the building equipment are extracted by combining the spatiotemporal constraints of building spatial topology and building equipment distribution, including:

[0013] The multi-source heterogeneous datasets are initially grouped according to the data source type;

[0014] Based on the initial grouping, a unified time series index is established to align data from different sources to the same time axis and generate cross-modal time series data sequences with timestamps;

[0015] Based on building spatial topology information, the cross-modal time series data sequence is mapped with the spatial distribution information of building equipment to form a multi-dimensional spatiotemporal coupling matrix;

[0016] Based on the spatial proximity and functional dependence of building equipment rooms, key features of the spatiotemporal coupling matrix are extracted.

[0017] Specifically, based on the spatial proximity and functional dependence between building equipment rooms, the key features of the spatiotemporal coupling matrix are extracted, including:

[0018] In the spatiotemporal coupling matrix, local associated units are determined based on the spatial proximity of building equipment, and data nodes in each local associated unit are aggregated. The data nodes are the basic elements in the spatiotemporal coupling matrix.

[0019] Based on the aforementioned local association units, and combined with the functional dependencies of building equipment, the interaction information between nodes is recursively traversed to form a multi-level functional dependency chain.

[0020] In the functional dependency chain, an interaction path spanning multiple local associated units is identified, and the features of each node in the path are weighted and combined.

[0021] The weighted features are integrated into a candidate feature set, and the candidate feature set is filtered by a preset screening criterion to obtain the key features of the spatiotemporal coupling matrix.

[0022] Specifically, based on the aforementioned key features, a causal reasoning model is used to identify the causal chain of building equipment failures. A hierarchical decision prediction model is then used to predict the failure state of individual building equipment at a local level and the operational trend of cross-building equipment groups at a global level, including:

[0023] Based on the aforementioned key features, a causal relationship graph is established, and potential causal chains between various building equipment are determined through a layer-by-layer recursive approach.

[0024] The potential causal chain is decomposed into a structure, and the direct and indirect causal relationships between nodes are labeled to form a multi-level causal dependency sequence;

[0025] Based on the multi-level causal dependency sequence, the hierarchical decision prediction model is used to extrapolate the operating status of a single building equipment at the local level, and generate the fault prediction result of the single building equipment.

[0026] By aggregating local-level prediction results to the global level and combining key characteristics of cross-building equipment groups, a prediction model for the global operational status is constructed.

[0027] Based on the prediction model of the global operational status, the potential failure trends and operational evolution paths of cross-building equipment groups are output.

[0028] Specifically, based on the multi-level causal dependency sequence, a hierarchical decision prediction model is used to extrapolate the operating status of a single building device at a local level, generating fault prediction results for that single building device, including:

[0029] The causal nodes related to single building equipment in the multi-level causal dependency sequence are filtered, and a local causal subgraph of single building equipment is established.

[0030] In the local causal subgraph, the historical state sequence of the input node is progressively deduced to generate the multi-time-period state evolution trajectory of a single building equipment.

[0031] Based on the multi-time period state evolution trajectory, the conditional events that trigger key causal nodes are identified, and the corresponding running state transition sequence is generated based on the conditional events.

[0032] The operational state transition sequence is matched with a preset local prediction rule to output the fault prediction result for a single building device.

[0033] Specifically, the process of aggregating local-level prediction results to the global level and combining key characteristics across building equipment groups to construct a global operational status prediction model includes:

[0034] The fault prediction results of each individual building equipment are synchronized and integrated according to spatial location index and timestamp information to form an initial prediction set for the whole system;

[0035] Based on the initial prediction set, and combined with the cross-regional topological relationship of the building equipment group, the prediction results of different regions are hierarchically aggregated to generate regional prediction units.

[0036] Cross-regional correlation analysis is performed on the regional-level prediction units to identify key interaction paths that trigger linkage effects, and these key interaction paths are integrated into the global situation model.

[0037] In the global situation model, each regional prediction unit and key interaction path are uniformly modeled to form a prediction model of the global operational situation.

[0038] Specifically, based on the prediction model of the global operational status, the potential failure trends and operational evolution paths of cross-building equipment groups are output, including:

[0039] In the prediction model of the global operational situation, time series extrapolation is performed on each regional prediction unit to generate an operational evolution sequence across time periods;

[0040] Pattern recognition is performed on the operational evolution sequence, and sequences with the same change characteristics are grouped into a candidate fault trend set;

[0041] In the candidate fault trend set, key trend paths are selected based on the correlation strength across building equipment groups, and a multi-level operation evolution network is established.

[0042] The operational evolution network is structured and output to form potential failure trends and operational evolution paths across building equipment groups.

[0043] Specifically, based on the fault prediction results, the process of generating operation adjustment plans for individual building equipment at the local level and scheduling and optimization strategies for cross-building equipment groups at the global level includes:

[0044] Based on the fault prediction results of single building equipment, the conditions for its operational state transition are determined, and the corresponding local operation adjustment instruction sequence is generated.

[0045] The local operation adjustment command sequence is compared with the status of other building equipment in the area where the single building equipment is located to identify the conflicting operating conditions;

[0046] Based on the identified conflict conditions, a coordination constraint model for cross-building equipment groups is constructed, and a global scheduling scheme is generated based on the coordination constraint model.

[0047] In the global scheduling scheme, resource allocation paths and operating load distribution are modeled in a unified manner to form an optimization strategy across building equipment groups.

[0048] A building equipment failure prediction system, used to implement the building equipment failure prediction method, includes: a data acquisition module, a feature extraction module, a failure prediction module, and a management optimization module;

[0049] The data acquisition module is used to collect the operating parameters, environmental parameters and maintenance data of building equipment to form a multi-source heterogeneous dataset;

[0050] The feature extraction module is used to perform cross-modal fusion on the multi-source heterogeneous dataset, and extract key features of the building equipment by combining the spatiotemporal constraints of building spatial topology and building equipment distribution. The key features characterize the coupling relationship and fault evolution trend between building equipment.

[0051] The fault prediction module is used to identify the causal chain of building equipment faults based on the key features using a causal reasoning model, and to predict the fault state of a single building equipment at the local level and the operating trend of a cross-building equipment group at the global level through a hierarchical decision prediction model.

[0052] The management optimization module is used to generate operation adjustment plans for single building equipment at the local level and scheduling and optimization strategies for cross-building equipment groups at the global level, based on the fault prediction results.

[0053] Specifically, the fault prediction module includes: a causal analysis unit, a single-device fault prediction unit, and a full-device fault prediction unit;

[0054] The causal analysis unit is used to establish a causal relationship graph, analyze the direct and indirect causal relationships between nodes, and form a multi-level causal dependency sequence.

[0055] The single-equipment fault prediction unit, based on the multi-level causal dependency sequence, uses a hierarchical decision prediction model to extrapolate the operating status of a single building equipment at a local level, and generates fault prediction results for the single building equipment.

[0056] The overall equipment fault prediction unit, based on a prediction model of the global operating status, outputs the potential fault trends and operating evolution paths of cross-building equipment groups.

[0057] Compared with the prior art, the beneficial effects of the present invention are:

[0058] This invention proposes a method and system for predicting building equipment failures. Under spatiotemporal constraints, it extracts key features between devices, identifies potential failure chains by combining a causal reasoning model, and uses hierarchical prediction and global scheduling methods to form a dynamic prediction and management closed loop for local devices and group devices. At the same time, the prediction and adjustment results are optimized and managed accordingly. This invention breaks through the limitations of existing technologies that rely solely on single-device data or threshold monitoring, achieving an improvement from single-point monitoring to global prediction. It adapts to the failure prediction needs of group devices in complex scenarios, thereby effectively reducing operation and maintenance costs and improving resource allocation efficiency while ensuring the safe and stable operation of building equipment. Attached Figure Description

[0059] Figure 1 A flowchart of a building equipment fault prediction method provided by the present invention;

[0060] Figure 2 The flowchart for key feature extraction provided for this invention;

[0061] Figure 3 The fault prediction flowchart provided by this invention;

[0062] Figure 4 The present invention provides a structural diagram of a building equipment fault prediction system. Detailed Implementation

[0063] The present application will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present application, but do not limit the present application in any way. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present application. These all fall within the protection scope of the present application.

[0064] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0065] It should be noted that, unless there is a conflict, the various features in the embodiments of this application can be combined with each other, all of which are within the protection scope of this application. Furthermore, although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than the module division in the device or the order in the flowchart. In addition, the terms "first," "second," and "third" used in this application do not limit the data or execution order, but only distinguish identical or similar items with essentially the same function and effect.

[0066] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items.

[0067] Example 1:

[0068] Please see Figures 1-3 The present invention provides an embodiment of a method for predicting building equipment failures, comprising the following specific steps:

[0069] Step S1: Collect the operating parameters, environmental parameters and maintenance data of building equipment to form a multi-source heterogeneous dataset.

[0070] In this embodiment, operating parameters include sensor signals reflecting the internal working conditions of the equipment, such as current, voltage, vibration frequency, and temperature; environmental parameters include data reflecting external conditions, such as temperature and humidity, air quality, and noise levels outside the building and inside the computer room; and maintenance data includes records reflecting human intervention factors, such as equipment maintenance records, manual operation logs, and maintenance cycles. By introducing a unified time index in the data acquisition stage, data from different sources are synchronously mapped to the same timeline. Combining equipment identification, spatial location labeling, and functional tags, the collected operating parameters, environmental parameters, and maintenance data are mapped to the corresponding equipment objects and building topology relationships. Finally, the synchronized and mapped data are combined into an integrated heterogeneous data structure to form a multi-source heterogeneous dataset.

[0071] Step S2: Perform cross-modal fusion on the multi-source heterogeneous dataset, and extract key features of the building equipment by combining the spatiotemporal constraints of building spatial topology and building equipment distribution. The key features characterize the coupling relationship and fault evolution trend among building equipment.

[0072] like Figure 2 As shown, the specific steps of step S2 are as follows:

[0073] Step S201: The multi-source heterogeneous dataset is initially grouped according to the data source type.

[0074] In this embodiment, a basic data classification structure is established through grouping operations. First, the collected raw data is tagged, with the tag content including the data source category, acquisition channel, and corresponding device identifier. Second, data with the same category tag are aggregated to form independent data subsets, preserving their own temporal characteristics and numerical distribution features. Third, relatively independent storage and indexing relationships are formed between the various data subsets, providing clear boundary conditions. Finally, through preliminary grouping, the original multi-source heterogeneous dataset can be organized into three major categories: runtime parameter subset, environmental parameter subset, and operation and maintenance data subset.

[0075] Step S202: Based on the initial grouping, establish a unified time series index, align data from different sources to the same time axis, and generate cross-modal time series data sequences with timestamps.

[0076] In this embodiment, a global unified timeline is determined, and a time-series index is established according to a preset sampling interval. Subsequently, the operating parameters are mapped to the index node through interpolation or downsampling, and the environmental parameters are filled to the corresponding time points according to the principle of temporal proximity. The operation and maintenance data is transformed into a discrete event sequence with timestamps. Through the above steps, data from different sources can have consistent index markings on the unified timeline. A unified time-series index is then constructed as a time reference to ensure that all data are comparable at the same time granularity. During the alignment process, missing or sparse numerical sequences are supplemented by matching timestamps and interpolation methods, so that the operating parameters, environmental parameters, and operation and maintenance data can form a corresponding relationship at the same time point. By adding explicit timestamps to the alignment results, the dynamic characteristics of the original data are not obscured. Finally, a cross-modal time-series data sequence is obtained.

[0077] Step S203: Based on the building space topology information, the cross-modal time series data sequence is mapped with the spatial distribution information of building equipment to form a multi-dimensional spatiotemporal coupling matrix.

[0078] In this embodiment, the cross-modal time-series data sequence has been synchronized in the time dimension; in the spatial dimension, the building topology is transformed into a spatial relationship matrix, and the matrix elements are used to characterize the spatial proximity and connection relationship between building equipment; by fusing and mapping the cross-modal time-series data sequence with the spatial relationship matrix, a multi-dimensional spatiotemporal coupling matrix is ​​obtained. This matrix not only retains the temporal evolution characteristics of various types of data, but also embeds the spatial dependency relationship between building equipment.

[0079] Step S204: Extract the key features of the spatiotemporal coupling matrix based on the spatial proximity and functional dependence of the building equipment rooms.

[0080] The specific steps of step S204 are as follows:

[0081] Step S2041: In the spatiotemporal coupling matrix, local associated units are determined based on the spatial proximity of building equipment, and data nodes in each local associated unit are aggregated, wherein the data nodes are the basic elements in the spatiotemporal coupling matrix.

[0082] In this embodiment, the spatial coordinates and relative distances between devices are extracted using building spatial topology information. A preset proximity threshold is used to determine whether there is a direct spatial connection between devices. Device nodes that meet the spatial proximity condition are grouped into a local association unit, so that each unit represents a relatively close group of devices within a building. In each local association unit, the aggregation process of data nodes is reflected in the aggregated processing of operating parameters, environmental parameters, and maintenance data within the same spatial range. This aggregation method can maintain continuity in the time dimension while strengthening the correlation in the spatial dimension, so that each local association unit can exist independently and can also maintain connections with other units through adjacency. Finally, the resulting local association units form a hierarchical matrix structure.

[0083] Step S2042: Based on the local association unit and combined with the functional dependency relationship of building equipment, recursively traverse the interaction information between nodes to form a multi-level functional dependency chain.

[0084] In this embodiment, a functional label is established for each device node within the local association unit, such as functional attributes like power supply, load bearing, heat dissipation, and transmission. Subsequently, the edges between node pairs that have input-output or control dependencies between devices are identified as functional dependency edges, and a local functional dependency graph is constructed accordingly. Based on this, the nodes and edges in the graph are traversed recursively, expanding the causal chain layer by layer to identify direct and indirect dependencies, resulting in a multi-level functional association structure. The recursive traversal allows the functional relationships between device nodes to extend beyond first-order adjacency to long chains across levels. The formation of multi-level dependency chains allows the interaction information between local nodes to be expanded step by step, enabling the functional logic of operation parameters, environmental parameters, and maintenance data to be associated. Finally, the functional dependency chain is obtained through this traversal.

[0085] Step S2043: In the functional dependency chain, identify the interaction path that spans multiple local associated units, and perform a weighted combination of the features of each node in the path.

[0086] In this embodiment, boundary scanning is performed on the functional dependency chains of all locally associated units to identify device nodes that share functions or have upstream and downstream dependencies between different units. Subsequently, cross-unit connection channels are constructed using these boundary nodes as anchor points, and the transmission path of interactive information is traced on the connection channels to form a set of functional interaction paths that can span multiple locally associated units. By weighting and combining the features of each node in the interaction path, the importance of the node, connection strength, and time sequence features are comprehensively considered to generate a feature vector that is representative of the entire path. Secondly, this weighting method can avoid the dominance of single node features in the prediction results, enabling cross-unit paths to more comprehensively represent the collaborative relationship between devices.

[0087] Step S2044: Integrate the weighted features into a candidate feature set, and filter the candidate feature set according to a preset screening criterion to obtain the key features of the spatiotemporal coupling matrix.

[0088] In this embodiment, the weighted feature results of different nodes in the path are uniformly encoded and indexed and labeled according to device category, spatial location, and functional dependency attributes. The feature results corresponding to each path are aggregated to form a candidate feature set. The candidate feature set is filtered by preset screening criteria to obtain key features representing the overall coupling relationship. The screening criteria include screening based on temporal stability, screening based on spatial redundancy, and screening based on functional dependency strength. First, features with excessive temporal fluctuations or lack of continuity are removed from the candidate feature set. Second, redundant features that are highly repetitive in space and cannot provide incremental information are deleted. Third, features that are in key positions in the functional chain or have high dependency weights are retained. Finally, the key features of the spatiotemporal coupling matrix are obtained.

[0089] Step S3: Based on the key features, use a causal reasoning model to identify the causal chain of building equipment failures, and use a hierarchical decision prediction model to predict the failure status of a single building equipment at the local level and the operating trend of a group of building equipment at the global level.

[0090] like Figure 3 As shown, the specific steps of step S3 are as follows:

[0091] Step S301: Based on the key features, establish a causal relationship graph and determine the potential causal chains between various building equipment by recursively layer by layer.

[0092] In this embodiment, key features are node-based, with each feature defined as a causal node. By analyzing the temporal sequence, spatial coupling, and functional dependency attributes between causal nodes, potentially causal pairs are identified, and directed connections are established between them. Multiple node pairs are then combined to form a causal relationship graph. Nodes with high leadership are selected as starting points in the graph, and a first-level recursion is performed on their downstream nodes. Based on the first-level results, the process is further extended to indirectly dependent nodes, forming a multi-level recursive process. During the recursive process, paths are filtered based on dependency strength to avoid redundant or spurious causal relationships. Finally, potential causal chains are obtained.

[0093] Step S302: The potential causal chain is decomposed into a structured form, and the direct and indirect causal relationships between nodes are labeled to form a multi-level causal dependency sequence.

[0094] In this embodiment, first-order connections between adjacent nodes are extracted from the causal relationship graph and labeled as direct causal relationships; multi-order connections spanning multiple intermediate nodes are traced recursively and labeled as indirect causal relationships; indirect causal relationships are used as an extension layer to fully characterize potential transmission effects across devices and regions; furthermore, multi-level causal units are arranged sequentially according to time sequence and dependency strength to form a hierarchical causal dependency sequence; finally, a multi-level causal dependency sequence is obtained.

[0095] Step S303: Based on the multi-level causal dependency sequence, the hierarchical decision prediction model is used to extrapolate the operating status of a single building equipment at the local level, and generate the fault prediction result of the single building equipment.

[0096] The specific steps of step S303 are as follows:

[0097] Step S3031: Filter the causal nodes related to single building equipment in the multi-level causal dependency sequence and establish a local causal subgraph for the single building equipment.

[0098] In this embodiment, all nodes containing the target device identifier are retrieved from the multi-level causal dependency sequence, and upstream and downstream nodes with causal relationships to these nodes are traced. These related nodes and their connecting edges are combined to form a causal subset closely related to the target device. The causal structure is localized through node filtering and subgraph construction. The filtering operation ensures that only causal nodes related to the target device are retained, avoiding irrelevant information from interfering with local predictions. Secondly, by extracting these nodes and their causal edges, a relatively independent local causal subgraph is formed. This subgraph retains the directionality of the original causal relationship and highlights the connection between the device and the environment, functional dependencies, and maintenance behaviors in its structure. Finally, the local causal subgraph is obtained.

[0099] Step S3032: In the local causal subgraph, the historical state sequence of the input node is progressively deduced to generate the multi-time-period state evolution trajectory of the single building equipment.

[0100] In this embodiment, the historical operating parameters, environmental records, and maintenance data sequences of the input nodes are processed into time segments to form time windows that reflect the trend of state changes. These historical states are input to the starting node of the local causal subgraph and propagated to downstream nodes through causal edges. The transmission effect of the state is calculated at each level of propagation, gradually forming the predicted state sequence of the target device in different time periods. A multi-time period state evolution trajectory is generated through a time-series progressive mechanism, with the historical state of the input nodes as the initial condition to ensure that the prediction process starts from the actual operating situation. The recursive propagation mechanism in the causal subgraph ensures that the dependencies between nodes are dynamically reflected, so that the state evolution not only reflects local trends but also includes the effects of external intervention and environmental conditions. The results of the progressive deduction are integrated into a state trajectory covering multiple future time periods, with each time point in the trajectory representing the possible operating state of the device at that moment. Finally, a multi-time period state evolution trajectory is obtained.

[0101] Step S3033: Based on the multi-time period state evolution trajectory, identify the conditional events that trigger key causal nodes, and generate the corresponding running state transition sequence based on the conditional events.

[0102] In this embodiment, nodes whose state values ​​undergo abrupt changes, exceed thresholds, or conform to specific patterns are searched in the multi-time-period state evolution trajectory, and these moments are marked as conditional event trigger points. These trigger points are mapped to key nodes in the causal subgraph to determine which conditional events can cause significant changes in the state of downstream nodes. Through this process, the originally continuous state evolution trajectory is discretized into several triggering conditional events. These conditional events are transformed into operable operational state transition sequences. Starting with each conditional event, its propagation path in the causal subgraph is traced, and the state transition process between different nodes is recorded. These propagation paths are arranged chronologically to form operational state transition chains. By merging similar paths and eliminating redundant branches, an operational state transition sequence that can completely characterize the equipment's operational evolution process is obtained. Finally, this operational state transition sequence serves as the structured output of local prediction.

[0103] Step S3034: Match the operating state transition sequence with the preset local prediction rules and output the fault prediction result of the single building equipment.

[0104] In this embodiment, the operating state transition sequence is segmented and parsed, and the continuous state change patterns are transformed into identifiable logical segments. These logical segments are compared one by one with local prediction rules, the rules of which include abnormal transition patterns of the equipment in electrical, mechanical, thermal, and environmental dimensions. When a segment in the transition sequence matches the rule conditions, the sequence is determined to correspond to a certain type of fault. Fault prediction results are generated through rule matching. During the comparison process, not only are completely matching transition sequences retrieved, but also partially matching abnormal patterns are identified using a fuzzy matching mechanism to improve coverage. Different rules are assigned weights in the matching results, and multiple possible results are sorted according to priority. Based on the sorting, prediction labels corresponding to the operating state transition sequence are output to form a structured fault prediction result. Finally, the prediction result is bound to the target device.

[0105] Step S304: Aggregate the prediction results at the local level to the global level, and combine the key characteristics of cross-building equipment groups to construct a prediction model of the global operating status.

[0106] The specific steps of step S304 are as follows:

[0107] Step S3041: Synchronously integrate the fault prediction results of each individual building equipment according to the spatial location index and timestamp information to form an initial prediction set for the whole system.

[0108] In this embodiment, spatial index information is bound to the fault prediction result of each building equipment. This information comes from the building topology and can identify the floor, machine room and adjacent equipment of the equipment. At the same time, a unified timestamp is attached to the prediction result to ensure the temporal correspondence of different prediction results. Subsequently, the prediction results with both spatial and temporal identifiers are aggregated in the global dimension to generate an initial prediction set covering the entire building.

[0109] Step S3042: Based on the initial prediction set, and combined with the cross-regional topological relationship of the building equipment group, the prediction results of different regions are hierarchically aggregated to generate regional prediction units.

[0110] In this embodiment, the boundaries of each region and its equipment set are extracted using building topology information, and the relevant results in the initial prediction set are initially classified according to the region where the equipment is located. Subsequently, the cross-regional topological relationships between regions are further identified, including the interconnection dependencies of energy transmission, airflow circulation, water supply and drainage channels or control networks. Based on these cross-regional connection relationships, the prediction results after the initial classification are aggregated hierarchically according to the topological level to obtain prediction data covering both the internal and cross-regional relationships within the region.

[0111] It should be noted that the principle behind this step is to form regional-level prediction units through hierarchical aggregation, enabling the prediction model to be extrapolated at the meso-level. First, aggregation within a region ensures the integration of prediction results from devices within the same region, avoiding local distortions caused by isolated device predictions. Second, cross-regional topological dependencies are introduced into the aggregation process, so that regional-level units not only reflect internal characteristics but also include the dynamic influence of external relationships. Third, the hierarchical aggregation process unifies the modeling of internal and cross-regional characteristics, allowing them to participate in global situation prediction as independent units and interact with other regions as nodes. Ultimately, regional-level prediction units are formed.

[0112] Step S3043: Perform cross-regional correlation analysis on the regional prediction unit, identify the key interaction paths that trigger the linkage effect, and integrate the key interaction paths into the global situation model.

[0113] In this embodiment, connection channels between regions are extracted based on building topology, such as energy pipelines, air circulation paths, water supply and drainage networks, or control signal links, and mapped onto the interaction boundaries of regional prediction units. By comparing the timestamps and state change trends of the prediction results for each region, synchronous fluctuation patterns or lag effects between different regions are analyzed to determine whether there are potential causal transmission paths. Through the above process, key interaction paths that trigger linkages can be screened from the inter-regional connections. These key interaction paths are introduced as constraints into the global situation model, enabling the model to capture the mutual influence between regions, rather than being limited to intra-regional predictions. During the integration process, the interaction paths are weighted to reflect the influence intensity of different paths in global transmission. The integrated global situation model is transformed into a dynamic structure that can simultaneously characterize intra-regional evolution and cross-regional linkages. Finally, this structure provides a complete causal network framework for global prediction.

[0114] Step S3044: In the global situation model, each regional prediction unit and key interaction path are uniformly modeled to form a prediction model of the global operational situation.

[0115] In this embodiment, each regional prediction unit is used as the basic node of the global situation model, while maintaining the integrity of its internal aggregation features. Key interaction paths identified in the previous step are introduced as connecting edges between nodes, forming a topological network covering the entire building. Based on this, by defining unified temporal indexes and state variables, intra-regional evolution and cross-regional transmission can dynamically interact on the same time axis. A unified modeling process is used to form a prediction model that can be used for global extrapolation. During the modeling process, the prediction results of different regional units are normalized to eliminate interference caused by differences in data scale. Secondly, the weights of key interaction paths are integrated to ensure that the importance of the paths is reflected in the global model while maintaining the stability of the overall structure. Thirdly, the prediction algorithm is run on the global situation model through iterative extrapolation, enabling different regional units and interaction paths to form an interactive evolutionary process. Finally, a prediction model of the global operational situation is obtained.

[0116] Step S305: Based on the prediction model of the global operating situation, output the potential failure trends and operating evolution paths of cross-building equipment groups.

[0117] The specific steps of step S305 are as follows:

[0118] Step S3051: In the prediction model of the global operating situation, time series extrapolation is performed on each regional prediction unit to generate an operating evolution sequence across time periods.

[0119] In this embodiment, the initial prediction results of each regional prediction unit are aligned according to a unified timestamp to ensure that each region is comparable at the same time point. Subsequently, a progressive calculation mechanism is introduced into the prediction model of the global operational situation, so that the prediction results of each region can evolve backward based on the state of the previous time slice and update the state of the next time slice through cross-regional interaction paths. Through step-by-step recursion, a regional operational evolution sequence covering multiple future time periods is obtained.

[0120] Step S3052: Perform pattern recognition on the running evolution sequence and group sequences with the same change characteristics into a candidate fault trend set.

[0121] In this embodiment, the evolutionary sequence is transformed into a multi-dimensional feature vector, including features such as the magnitude of change, rate of change, periodicity, and anomalous fluctuation points. Subsequently, clustering or hierarchical classification is used to measure the similarity of the feature vectors between sequences, identifying groups of sequences with similar trends. Through this process, multiple dispersed evolutionary sequences are summarized into several representative pattern categories.

[0122] A candidate fault trend set is formed through a merging process. First, the operational evolution sequences aggregated in the same category reflect the common fault symptoms exhibited by building equipment in different regions or time periods. Second, by comparing the change patterns of different categories, representative candidate trend paths are extracted. These paths cover both typical single-equipment abnormal evolution and group trends that spread across regions. Finally, a candidate fault trend set is formed.

[0123] Step S3053: In the candidate fault trend set, key trend paths are selected based on the correlation strength across building equipment groups, and a multi-level operation evolution network is established.

[0124] In this embodiment, the correlation strength of each path in the candidate fault trend set is calculated in three dimensions: spatial proximity, functional dependence, and temporal synchronization. Then, the results of each dimension are weighted and fused to form a comprehensive score for the entire path. Based on this, paths with comprehensive scores exceeding a preset threshold are marked as key trend paths. In this way, core evolution directions with significant impact on the overall operational status are selected from the candidate trends. The selected key trend paths serve as the backbone links, forming the main layer of the evolutionary network. Secondary paths that were not selected but have certain reference value serve as a supplementary layer to enhance the network's integrity. By establishing cross-layer connections between the main layer and the supplementary layer, the operational evolutionary network can simultaneously reflect the rapidly propagating main trends and potential secondary diffusion paths, ultimately resulting in a multi-layered operational evolutionary network.

[0125] Step S3054: The operational evolution network is output in a structured manner to form potential failure trends and operational evolution paths across building equipment groups.

[0126] In this embodiment, the key trend paths and secondary supplementary paths in the operational evolution network are hierarchically labeled to clearly distinguish their primary and secondary importance in structure. The nodes in the paths are arranged in chronological order, and the transition relationships between nodes are assigned clear directional and weight information, forming a structured expression with temporal, spatial, and causal attributes. The structured output yields potential fault trends and operational evolution paths across building equipment groups. The output trend paths reveal the potential abnormal development direction of linkages between different equipment groups. The serialized representation of the operational evolution paths allows the state changes of each node to be tracked and verified along the causal chain. The output results not only retain the complexity of the global network but also present it in a hierarchical and parsable manner, enabling it to be directly used as input for subsequent adjustment strategies and management decisions. Ultimately, potential fault trends and operational evolution paths are formed.

[0127] Step S4: Based on the fault prediction results, generate an operation adjustment plan for a single building equipment at the local level, and generate a scheduling and optimization strategy for cross-building equipment groups at the global level.

[0128] The specific steps of step S4 are as follows:

[0129] Step S401: Based on the fault prediction results of a single building equipment, determine its operating state transition conditions and generate the corresponding local operation adjustment instruction sequence.

[0130] In this embodiment, the predicted potential fault points are mapped to the adjustable parameters of the equipment to identify the constraints required for the transition from the current operating state to the target safe state. Specifically, the key nodes involved in the fault prediction results are first analyzed to extract the operating variables that may cause anomalies and their threshold ranges. Then, these variables are mapped to the adjustable parameters in the equipment control logic to form a set of conditions on which the state transition depends. Finally, a local operation adjustment instruction sequence is generated based on the condition set. The instruction content includes the adjustment range, execution order, and action time, realizing the active intervention of a single device based on prediction. The result of this process is to transform the abstract prediction results into directly applicable operation instructions.

[0131] Step S402: Compare the local operation adjustment instruction sequence with the status of other building equipment in the area where the single building equipment is located to identify the conflicting operating conditions.

[0132] In this embodiment, conditions that may lead to resource conflicts or control contradictions are identified by the operational dependencies between devices within the region. Specifically, firstly, the functional dependency links between the device and surrounding devices are extracted based on the regional topology, including energy supply, environmental control, and load distribution relationships. Subsequently, the control parameters involved in the local operation adjustment instruction sequence are compared item by item with the current or predicted operating status of adjacent devices to determine whether there are overlapping parameter values, execution order conflicts, or uneven load distribution. Finally, the identified conflict conditions are labeled and recorded in a structured form.

[0133] Step S403: Based on the identified conflict conditions, construct a coordination constraint model for cross-building equipment groups, and generate a global scheduling scheme based on the coordination constraint model.

[0134] In this embodiment, operational conflicts at the local level are abstracted into constraints in global scheduling, thereby driving joint optimization of multiple devices. Specifically, conflict conditions are first classified according to resource type, timing requirements, and functional dependencies to form a multi-dimensional constraint set. Subsequently, this constraint set is mapped to the device network at the regional and global levels, so that each constraint can correspond to a specific device group and operating parameters. On this basis, coordination rules are introduced to combine and prioritize the constraints, establishing a coordination constraint model covering cross-device groups. Finally, the scheduling algorithm is run using this model as input to generate a global scheduling scheme. The scheme includes the running sequence of each device, load distribution path, and resource sharing strategy, achieving conflict resolution and coordinated operation on a global scale.

[0135] Step S404: In the global scheduling scheme, the resource allocation path and the distribution of operating load are modeled in a unified manner to form an optimization strategy across building equipment groups.

[0136] In this embodiment, the energy flow, material flow, and task load on a global scale are expressed holistically to achieve optimal coordination across a group of devices. Specifically, firstly, various resource paths, such as energy supply, environmental control, and information interaction, are transformed into a network structure, and the operating constraints and capacity boundaries of each device are embedded within it. Subsequently, the load demands and resource supplies of different devices are mapped to each other, and a multi-objective optimization function is constructed to balance efficiency, stability, and resource balance. Finally, through joint modeling and iterative calculation of resource paths and load distribution, an optimization strategy covering the entire group of devices is generated. This strategy can coordinate the operating rhythm and balance resource usage on a global scale, and provide a decision-making basis for real-time adjustments in dynamic environments.

[0137] Example 2:

[0138] Please see Figure 4 Another embodiment of the present invention provides: a building equipment fault prediction system, comprising: a data acquisition module, a feature extraction module, a fault prediction module, and a management optimization module;

[0139] The data acquisition module is used to collect the operating parameters, environmental parameters and maintenance data of building equipment to form a multi-source heterogeneous dataset;

[0140] The feature extraction module is used to perform cross-modal fusion on the multi-source heterogeneous dataset, and extract key features of the building equipment by combining the spatiotemporal constraints of building spatial topology and building equipment distribution. The key features characterize the coupling relationship and fault evolution trend between building equipment.

[0141] The fault prediction module is used to identify the causal chain of building equipment faults based on the key features using a causal reasoning model, and to predict the fault state of a single building equipment at the local level and the operating trend of a cross-building equipment group at the global level through a hierarchical decision prediction model.

[0142] The management optimization module is used to generate operation adjustment plans for single building equipment at the local level and scheduling and optimization strategies for cross-building equipment groups at the global level, based on the fault prediction results.

[0143] The fault prediction module includes: a cause-effect analysis unit, a single-device fault prediction unit, and a full-device fault prediction unit;

[0144] The causal analysis unit is used to establish a causal relationship graph, analyze the direct and indirect causal relationships between nodes, and form a multi-level causal dependency sequence.

[0145] The single-equipment fault prediction unit, based on the multi-level causal dependency sequence, uses a hierarchical decision prediction model to extrapolate the operating status of a single building equipment at a local level, and generates fault prediction results for the single building equipment.

[0146] The overall equipment fault prediction unit, based on a prediction model of the global operating status, outputs the potential fault trends and operating evolution paths of cross-building equipment groups.

[0147] In addition, the parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of the corresponding technical solutions in the prior art have not been described in detail, so as to avoid excessive elaboration.

[0148] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting building equipment failures, characterized in that, include: Collect operating parameters, environmental parameters, and maintenance data of building equipment to form a multi-source heterogeneous dataset; Cross-modal fusion is performed on the multi-source heterogeneous dataset, and key features of building equipment are extracted by combining the spatiotemporal constraints of building spatial topology and building equipment distribution. These key features characterize the coupling relationship and fault evolution trend among building equipment. Based on the aforementioned key features, a causal reasoning model is used to identify the causal chain of building equipment failures. A hierarchical decision prediction model is used to predict the failure status of a single building equipment at the local level and the operating trend of a group of building equipment at the global level. Based on the fault prediction results, operation adjustment plans for individual building equipment are generated at the local level, and scheduling and optimization strategies for cross-building equipment groups are generated at the global level.

2. The method for predicting building equipment failure as described in claim 1, characterized in that, Cross-modal fusion is performed on the aforementioned multi-source heterogeneous datasets, and key features of building equipment are extracted by combining the spatiotemporal constraints of building spatial topology and building equipment distribution, including: The multi-source heterogeneous datasets are initially grouped according to the data source type; Based on the initial grouping, a unified time series index is established to align data from different sources to the same time axis and generate cross-modal time series data sequences with timestamps; Based on building spatial topology information, the cross-modal time series data sequence is mapped with the spatial distribution information of building equipment to form a multi-dimensional spatiotemporal coupling matrix; Based on the spatial proximity and functional dependence of building equipment rooms, key features of the spatiotemporal coupling matrix are extracted.

3. The method for predicting building equipment failure as described in claim 2, characterized in that, Based on the spatial proximity and functional dependence between building equipment rooms, the key features of the spatiotemporal coupling matrix are extracted, including: In the spatiotemporal coupling matrix, local associated units are determined based on the spatial proximity of building equipment, and data nodes in each local associated unit are aggregated. The data nodes are the basic elements in the spatiotemporal coupling matrix. Based on the aforementioned local association units, and combined with the functional dependencies of building equipment, the interaction information between nodes is recursively traversed to form a multi-level functional dependency chain. In the functional dependency chain, an interaction path spanning multiple local associated units is identified, and the features of each node in the path are weighted and combined. The weighted features are integrated into a candidate feature set, and the candidate feature set is filtered by a preset screening criterion to obtain the key features of the spatiotemporal coupling matrix.

4. The method for predicting building equipment failure as described in claim 1, characterized in that, Based on the aforementioned key features, a causal inference model is used to identify the causal chain of building equipment failures. A hierarchical decision prediction model is then used to predict the failure state of individual building equipment at a local level and the operational trend of cross-building equipment groups at a global level, including: Based on the aforementioned key features, a causal relationship graph is established, and potential causal chains between various building equipment are determined through a layer-by-layer recursive approach. The potential causal chain is decomposed into a structure, and the direct and indirect causal relationships between nodes are labeled to form a multi-level causal dependency sequence; Based on the multi-level causal dependency sequence, the hierarchical decision prediction model is used to extrapolate the operating status of a single building equipment at the local level, and generate the fault prediction result of the single building equipment. By aggregating local-level prediction results to the global level and combining key characteristics of cross-building equipment groups, a prediction model for the global operational status is constructed. Based on the prediction model of the global operational status, the potential failure trends and operational evolution paths of cross-building equipment groups are output.

5. The method for predicting building equipment failure as described in claim 4, characterized in that, Based on the aforementioned multi-level causal dependency sequence, a hierarchical decision prediction model is used to extrapolate the operating status of a single building device at a local level, generating fault prediction results for that single building device, including: The causal nodes related to single building equipment in the multi-level causal dependency sequence are filtered, and a local causal subgraph of single building equipment is established. In the local causal subgraph, the historical state sequence of the input node is progressively deduced to generate the multi-time-period state evolution trajectory of a single building equipment. Based on the multi-time period state evolution trajectory, the conditional events that trigger key causal nodes are identified, and the corresponding running state transition sequence is generated based on the conditional events. The operational state transition sequence is matched with a preset local prediction rule to output the fault prediction result for a single building device.

6. The method for predicting building equipment failure as described in claim 5, characterized in that, The process of aggregating local-level prediction results to the global level and combining key characteristics of cross-building equipment groups to construct a global operational status prediction model includes: The fault prediction results of each individual building equipment are synchronized and integrated according to spatial location index and timestamp information to form an initial prediction set for the whole system; Based on the initial prediction set, and combined with the cross-regional topological relationship of the building equipment group, the prediction results of different regions are hierarchically aggregated to generate regional prediction units. Cross-regional correlation analysis is performed on the regional-level prediction units to identify key interaction paths that trigger linkage effects, and these key interaction paths are integrated into the global situation model. In the global situation model, each regional prediction unit and key interaction path are uniformly modeled to form a prediction model of the global operational situation.

7. The method for predicting building equipment failure as described in claim 6, characterized in that, Based on the prediction model of the global operational status, the potential failure trends and operational evolution paths of cross-building equipment groups are output, including: In the prediction model of the global operational situation, time series extrapolation is performed on each regional prediction unit to generate an operational evolution sequence across time periods; Pattern recognition is performed on the operational evolution sequence, and sequences with the same change characteristics are grouped into a candidate fault trend set; In the candidate fault trend set, key trend paths are selected based on the correlation strength across building equipment groups, and a multi-level operation evolution network is established. The operational evolution network is structured and output to form potential failure trends and operational evolution paths across building equipment groups.

8. The method for predicting building equipment failure as described in claim 1, characterized in that, Based on the fault prediction results, the process of generating operation adjustment plans for individual building equipment at the local level and scheduling and optimization strategies for cross-building equipment groups at the global level includes: Based on the fault prediction results of single building equipment, the conditions for its operational state transition are determined, and the corresponding local operation adjustment instruction sequence is generated. The local operation adjustment command sequence is compared with the status of other building equipment in the area where the single building equipment is located to identify the conflicting operating conditions; Based on the identified conflict conditions, a coordination constraint model for cross-building equipment groups is constructed, and a global scheduling scheme is generated based on the coordination constraint model. In the global scheduling scheme, resource allocation paths and operating load distribution are modeled in a unified manner to form an optimization strategy across building equipment groups.

9. A building equipment failure prediction system, used to implement the building equipment failure prediction method according to any one of claims 1-8, characterized in that, include: Data acquisition module, feature extraction module, fault prediction module, and management optimization module; The data acquisition module is used to collect the operating parameters, environmental parameters and maintenance data of building equipment to form a multi-source heterogeneous dataset; The feature extraction module is used to perform cross-modal fusion on the multi-source heterogeneous dataset, and extract key features of the building equipment by combining the spatiotemporal constraints of building spatial topology and building equipment distribution. The key features characterize the coupling relationship and fault evolution trend between building equipment. The fault prediction module is used to identify the causal chain of building equipment faults based on the key features using a causal reasoning model, and to predict the fault state of a single building equipment at the local level and the operating trend of a group of building equipment at the global level through a hierarchical decision prediction model. The management optimization module is used to generate operation adjustment plans for single building equipment at the local level and scheduling and optimization strategies for cross-building equipment groups at the global level, based on the fault prediction results.

10. A building equipment failure prediction system as described in claim 9, characterized in that, The fault prediction module includes: a cause-effect analysis unit, a single-device fault prediction unit, and a full-device fault prediction unit; The causal analysis unit is used to establish a causal relationship graph, analyze the direct and indirect causal relationships between nodes, and form a multi-level causal dependency sequence. The single-equipment fault prediction unit, based on the multi-level causal dependency sequence, uses a hierarchical decision prediction model to extrapolate the operating status of a single building equipment at a local level, and generates fault prediction results for the single building equipment. The overall equipment fault prediction unit, based on a prediction model of the global operating status, outputs the potential fault trends and operating evolution paths of cross-building equipment groups.

Images